* gimple-ssa-store-merging.c (struct store_immediate_info): Add
[official-gcc.git] / gcc / lra-constraints.c
blob3758409bb9c520423a286ff18815e93bfe3c0b9b
1 /* Code for RTL transformations to satisfy insn constraints.
2 Copyright (C) 2010-2017 Free Software Foundation, Inc.
3 Contributed by Vladimir Makarov <vmakarov@redhat.com>.
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
22 /* This file contains code for 3 passes: constraint pass,
23 inheritance/split pass, and pass for undoing failed inheritance and
24 split.
26 The major goal of constraint pass is to transform RTL to satisfy
27 insn and address constraints by:
28 o choosing insn alternatives;
29 o generating *reload insns* (or reloads in brief) and *reload
30 pseudos* which will get necessary hard registers later;
31 o substituting pseudos with equivalent values and removing the
32 instructions that initialized those pseudos.
34 The constraint pass has biggest and most complicated code in LRA.
35 There are a lot of important details like:
36 o reuse of input reload pseudos to simplify reload pseudo
37 allocations;
38 o some heuristics to choose insn alternative to improve the
39 inheritance;
40 o early clobbers etc.
42 The pass is mimicking former reload pass in alternative choosing
43 because the reload pass is oriented to current machine description
44 model. It might be changed if the machine description model is
45 changed.
47 There is special code for preventing all LRA and this pass cycling
48 in case of bugs.
50 On the first iteration of the pass we process every instruction and
51 choose an alternative for each one. On subsequent iterations we try
52 to avoid reprocessing instructions if we can be sure that the old
53 choice is still valid.
55 The inheritance/spilt pass is to transform code to achieve
56 ineheritance and live range splitting. It is done on backward
57 traversal of EBBs.
59 The inheritance optimization goal is to reuse values in hard
60 registers. There is analogous optimization in old reload pass. The
61 inheritance is achieved by following transformation:
63 reload_p1 <- p reload_p1 <- p
64 ... new_p <- reload_p1
65 ... => ...
66 reload_p2 <- p reload_p2 <- new_p
68 where p is spilled and not changed between the insns. Reload_p1 is
69 also called *original pseudo* and new_p is called *inheritance
70 pseudo*.
72 The subsequent assignment pass will try to assign the same (or
73 another if it is not possible) hard register to new_p as to
74 reload_p1 or reload_p2.
76 If the assignment pass fails to assign a hard register to new_p,
77 this file will undo the inheritance and restore the original code.
78 This is because implementing the above sequence with a spilled
79 new_p would make the code much worse. The inheritance is done in
80 EBB scope. The above is just a simplified example to get an idea
81 of the inheritance as the inheritance is also done for non-reload
82 insns.
84 Splitting (transformation) is also done in EBB scope on the same
85 pass as the inheritance:
87 r <- ... or ... <- r r <- ... or ... <- r
88 ... s <- r (new insn -- save)
89 ... =>
90 ... r <- s (new insn -- restore)
91 ... <- r ... <- r
93 The *split pseudo* s is assigned to the hard register of the
94 original pseudo or hard register r.
96 Splitting is done:
97 o In EBBs with high register pressure for global pseudos (living
98 in at least 2 BBs) and assigned to hard registers when there
99 are more one reloads needing the hard registers;
100 o for pseudos needing save/restore code around calls.
102 If the split pseudo still has the same hard register as the
103 original pseudo after the subsequent assignment pass or the
104 original pseudo was split, the opposite transformation is done on
105 the same pass for undoing inheritance. */
107 #undef REG_OK_STRICT
109 #include "config.h"
110 #include "system.h"
111 #include "coretypes.h"
112 #include "backend.h"
113 #include "target.h"
114 #include "rtl.h"
115 #include "tree.h"
116 #include "predict.h"
117 #include "df.h"
118 #include "memmodel.h"
119 #include "tm_p.h"
120 #include "expmed.h"
121 #include "optabs.h"
122 #include "regs.h"
123 #include "ira.h"
124 #include "recog.h"
125 #include "output.h"
126 #include "addresses.h"
127 #include "expr.h"
128 #include "cfgrtl.h"
129 #include "rtl-error.h"
130 #include "params.h"
131 #include "lra.h"
132 #include "lra-int.h"
133 #include "print-rtl.h"
135 /* Value of LRA_CURR_RELOAD_NUM at the beginning of BB of the current
136 insn. Remember that LRA_CURR_RELOAD_NUM is the number of emitted
137 reload insns. */
138 static int bb_reload_num;
140 /* The current insn being processed and corresponding its single set
141 (NULL otherwise), its data (basic block, the insn data, the insn
142 static data, and the mode of each operand). */
143 static rtx_insn *curr_insn;
144 static rtx curr_insn_set;
145 static basic_block curr_bb;
146 static lra_insn_recog_data_t curr_id;
147 static struct lra_static_insn_data *curr_static_id;
148 static machine_mode curr_operand_mode[MAX_RECOG_OPERANDS];
149 /* Mode of the register substituted by its equivalence with VOIDmode
150 (e.g. constant) and whose subreg is given operand of the current
151 insn. VOIDmode in all other cases. */
152 static machine_mode original_subreg_reg_mode[MAX_RECOG_OPERANDS];
156 /* Start numbers for new registers and insns at the current constraints
157 pass start. */
158 static int new_regno_start;
159 static int new_insn_uid_start;
161 /* If LOC is nonnull, strip any outer subreg from it. */
162 static inline rtx *
163 strip_subreg (rtx *loc)
165 return loc && GET_CODE (*loc) == SUBREG ? &SUBREG_REG (*loc) : loc;
168 /* Return hard regno of REGNO or if it is was not assigned to a hard
169 register, use a hard register from its allocno class. */
170 static int
171 get_try_hard_regno (int regno)
173 int hard_regno;
174 enum reg_class rclass;
176 if ((hard_regno = regno) >= FIRST_PSEUDO_REGISTER)
177 hard_regno = lra_get_regno_hard_regno (regno);
178 if (hard_regno >= 0)
179 return hard_regno;
180 rclass = lra_get_allocno_class (regno);
181 if (rclass == NO_REGS)
182 return -1;
183 return ira_class_hard_regs[rclass][0];
186 /* Return the hard regno of X after removing its subreg. If X is not
187 a register or a subreg of a register, return -1. If X is a pseudo,
188 use its assignment. If FINAL_P return the final hard regno which will
189 be after elimination. */
190 static int
191 get_hard_regno (rtx x, bool final_p)
193 rtx reg;
194 int hard_regno;
196 reg = x;
197 if (SUBREG_P (x))
198 reg = SUBREG_REG (x);
199 if (! REG_P (reg))
200 return -1;
201 if (! HARD_REGISTER_NUM_P (hard_regno = REGNO (reg)))
202 hard_regno = lra_get_regno_hard_regno (hard_regno);
203 if (hard_regno < 0)
204 return -1;
205 if (final_p)
206 hard_regno = lra_get_elimination_hard_regno (hard_regno);
207 if (SUBREG_P (x))
208 hard_regno += subreg_regno_offset (hard_regno, GET_MODE (reg),
209 SUBREG_BYTE (x), GET_MODE (x));
210 return hard_regno;
213 /* If REGNO is a hard register or has been allocated a hard register,
214 return the class of that register. If REGNO is a reload pseudo
215 created by the current constraints pass, return its allocno class.
216 Return NO_REGS otherwise. */
217 static enum reg_class
218 get_reg_class (int regno)
220 int hard_regno;
222 if (! HARD_REGISTER_NUM_P (hard_regno = regno))
223 hard_regno = lra_get_regno_hard_regno (regno);
224 if (hard_regno >= 0)
226 hard_regno = lra_get_elimination_hard_regno (hard_regno);
227 return REGNO_REG_CLASS (hard_regno);
229 if (regno >= new_regno_start)
230 return lra_get_allocno_class (regno);
231 return NO_REGS;
234 /* Return true if REG satisfies (or will satisfy) reg class constraint
235 CL. Use elimination first if REG is a hard register. If REG is a
236 reload pseudo created by this constraints pass, assume that it will
237 be allocated a hard register from its allocno class, but allow that
238 class to be narrowed to CL if it is currently a superset of CL.
240 If NEW_CLASS is nonnull, set *NEW_CLASS to the new allocno class of
241 REGNO (reg), or NO_REGS if no change in its class was needed. */
242 static bool
243 in_class_p (rtx reg, enum reg_class cl, enum reg_class *new_class)
245 enum reg_class rclass, common_class;
246 machine_mode reg_mode;
247 int class_size, hard_regno, nregs, i, j;
248 int regno = REGNO (reg);
250 if (new_class != NULL)
251 *new_class = NO_REGS;
252 if (regno < FIRST_PSEUDO_REGISTER)
254 rtx final_reg = reg;
255 rtx *final_loc = &final_reg;
257 lra_eliminate_reg_if_possible (final_loc);
258 return TEST_HARD_REG_BIT (reg_class_contents[cl], REGNO (*final_loc));
260 reg_mode = GET_MODE (reg);
261 rclass = get_reg_class (regno);
262 if (regno < new_regno_start
263 /* Do not allow the constraints for reload instructions to
264 influence the classes of new pseudos. These reloads are
265 typically moves that have many alternatives, and restricting
266 reload pseudos for one alternative may lead to situations
267 where other reload pseudos are no longer allocatable. */
268 || (INSN_UID (curr_insn) >= new_insn_uid_start
269 && curr_insn_set != NULL
270 && ((OBJECT_P (SET_SRC (curr_insn_set))
271 && ! CONSTANT_P (SET_SRC (curr_insn_set)))
272 || (GET_CODE (SET_SRC (curr_insn_set)) == SUBREG
273 && OBJECT_P (SUBREG_REG (SET_SRC (curr_insn_set)))
274 && ! CONSTANT_P (SUBREG_REG (SET_SRC (curr_insn_set)))))))
275 /* When we don't know what class will be used finally for reload
276 pseudos, we use ALL_REGS. */
277 return ((regno >= new_regno_start && rclass == ALL_REGS)
278 || (rclass != NO_REGS && ira_class_subset_p[rclass][cl]
279 && ! hard_reg_set_subset_p (reg_class_contents[cl],
280 lra_no_alloc_regs)));
281 else
283 common_class = ira_reg_class_subset[rclass][cl];
284 if (new_class != NULL)
285 *new_class = common_class;
286 if (hard_reg_set_subset_p (reg_class_contents[common_class],
287 lra_no_alloc_regs))
288 return false;
289 /* Check that there are enough allocatable regs. */
290 class_size = ira_class_hard_regs_num[common_class];
291 for (i = 0; i < class_size; i++)
293 hard_regno = ira_class_hard_regs[common_class][i];
294 nregs = hard_regno_nregs (hard_regno, reg_mode);
295 if (nregs == 1)
296 return true;
297 for (j = 0; j < nregs; j++)
298 if (TEST_HARD_REG_BIT (lra_no_alloc_regs, hard_regno + j)
299 || ! TEST_HARD_REG_BIT (reg_class_contents[common_class],
300 hard_regno + j))
301 break;
302 if (j >= nregs)
303 return true;
305 return false;
309 /* Return true if REGNO satisfies a memory constraint. */
310 static bool
311 in_mem_p (int regno)
313 return get_reg_class (regno) == NO_REGS;
316 /* Return 1 if ADDR is a valid memory address for mode MODE in address
317 space AS, and check that each pseudo has the proper kind of hard
318 reg. */
319 static int
320 valid_address_p (machine_mode mode ATTRIBUTE_UNUSED,
321 rtx addr, addr_space_t as)
323 #ifdef GO_IF_LEGITIMATE_ADDRESS
324 lra_assert (ADDR_SPACE_GENERIC_P (as));
325 GO_IF_LEGITIMATE_ADDRESS (mode, addr, win);
326 return 0;
328 win:
329 return 1;
330 #else
331 return targetm.addr_space.legitimate_address_p (mode, addr, 0, as);
332 #endif
335 namespace {
336 /* Temporarily eliminates registers in an address (for the lifetime of
337 the object). */
338 class address_eliminator {
339 public:
340 address_eliminator (struct address_info *ad);
341 ~address_eliminator ();
343 private:
344 struct address_info *m_ad;
345 rtx *m_base_loc;
346 rtx m_base_reg;
347 rtx *m_index_loc;
348 rtx m_index_reg;
352 address_eliminator::address_eliminator (struct address_info *ad)
353 : m_ad (ad),
354 m_base_loc (strip_subreg (ad->base_term)),
355 m_base_reg (NULL_RTX),
356 m_index_loc (strip_subreg (ad->index_term)),
357 m_index_reg (NULL_RTX)
359 if (m_base_loc != NULL)
361 m_base_reg = *m_base_loc;
362 lra_eliminate_reg_if_possible (m_base_loc);
363 if (m_ad->base_term2 != NULL)
364 *m_ad->base_term2 = *m_ad->base_term;
366 if (m_index_loc != NULL)
368 m_index_reg = *m_index_loc;
369 lra_eliminate_reg_if_possible (m_index_loc);
373 address_eliminator::~address_eliminator ()
375 if (m_base_loc && *m_base_loc != m_base_reg)
377 *m_base_loc = m_base_reg;
378 if (m_ad->base_term2 != NULL)
379 *m_ad->base_term2 = *m_ad->base_term;
381 if (m_index_loc && *m_index_loc != m_index_reg)
382 *m_index_loc = m_index_reg;
385 /* Return true if the eliminated form of AD is a legitimate target address. */
386 static bool
387 valid_address_p (struct address_info *ad)
389 address_eliminator eliminator (ad);
390 return valid_address_p (ad->mode, *ad->outer, ad->as);
393 /* Return true if the eliminated form of memory reference OP satisfies
394 extra (special) memory constraint CONSTRAINT. */
395 static bool
396 satisfies_memory_constraint_p (rtx op, enum constraint_num constraint)
398 struct address_info ad;
400 decompose_mem_address (&ad, op);
401 address_eliminator eliminator (&ad);
402 return constraint_satisfied_p (op, constraint);
405 /* Return true if the eliminated form of address AD satisfies extra
406 address constraint CONSTRAINT. */
407 static bool
408 satisfies_address_constraint_p (struct address_info *ad,
409 enum constraint_num constraint)
411 address_eliminator eliminator (ad);
412 return constraint_satisfied_p (*ad->outer, constraint);
415 /* Return true if the eliminated form of address OP satisfies extra
416 address constraint CONSTRAINT. */
417 static bool
418 satisfies_address_constraint_p (rtx op, enum constraint_num constraint)
420 struct address_info ad;
422 decompose_lea_address (&ad, &op);
423 return satisfies_address_constraint_p (&ad, constraint);
426 /* Initiate equivalences for LRA. As we keep original equivalences
427 before any elimination, we need to make copies otherwise any change
428 in insns might change the equivalences. */
429 void
430 lra_init_equiv (void)
432 ira_expand_reg_equiv ();
433 for (int i = FIRST_PSEUDO_REGISTER; i < max_reg_num (); i++)
435 rtx res;
437 if ((res = ira_reg_equiv[i].memory) != NULL_RTX)
438 ira_reg_equiv[i].memory = copy_rtx (res);
439 if ((res = ira_reg_equiv[i].invariant) != NULL_RTX)
440 ira_reg_equiv[i].invariant = copy_rtx (res);
444 static rtx loc_equivalence_callback (rtx, const_rtx, void *);
446 /* Update equivalence for REGNO. We need to this as the equivalence
447 might contain other pseudos which are changed by their
448 equivalences. */
449 static void
450 update_equiv (int regno)
452 rtx x;
454 if ((x = ira_reg_equiv[regno].memory) != NULL_RTX)
455 ira_reg_equiv[regno].memory
456 = simplify_replace_fn_rtx (x, NULL_RTX, loc_equivalence_callback,
457 NULL_RTX);
458 if ((x = ira_reg_equiv[regno].invariant) != NULL_RTX)
459 ira_reg_equiv[regno].invariant
460 = simplify_replace_fn_rtx (x, NULL_RTX, loc_equivalence_callback,
461 NULL_RTX);
464 /* If we have decided to substitute X with another value, return that
465 value, otherwise return X. */
466 static rtx
467 get_equiv (rtx x)
469 int regno;
470 rtx res;
472 if (! REG_P (x) || (regno = REGNO (x)) < FIRST_PSEUDO_REGISTER
473 || ! ira_reg_equiv[regno].defined_p
474 || ! ira_reg_equiv[regno].profitable_p
475 || lra_get_regno_hard_regno (regno) >= 0)
476 return x;
477 if ((res = ira_reg_equiv[regno].memory) != NULL_RTX)
479 if (targetm.cannot_substitute_mem_equiv_p (res))
480 return x;
481 return res;
483 if ((res = ira_reg_equiv[regno].constant) != NULL_RTX)
484 return res;
485 if ((res = ira_reg_equiv[regno].invariant) != NULL_RTX)
486 return res;
487 gcc_unreachable ();
490 /* If we have decided to substitute X with the equivalent value,
491 return that value after elimination for INSN, otherwise return
492 X. */
493 static rtx
494 get_equiv_with_elimination (rtx x, rtx_insn *insn)
496 rtx res = get_equiv (x);
498 if (x == res || CONSTANT_P (res))
499 return res;
500 return lra_eliminate_regs_1 (insn, res, GET_MODE (res),
501 false, false, 0, true);
504 /* Set up curr_operand_mode. */
505 static void
506 init_curr_operand_mode (void)
508 int nop = curr_static_id->n_operands;
509 for (int i = 0; i < nop; i++)
511 machine_mode mode = GET_MODE (*curr_id->operand_loc[i]);
512 if (mode == VOIDmode)
514 /* The .md mode for address operands is the mode of the
515 addressed value rather than the mode of the address itself. */
516 if (curr_id->icode >= 0 && curr_static_id->operand[i].is_address)
517 mode = Pmode;
518 else
519 mode = curr_static_id->operand[i].mode;
521 curr_operand_mode[i] = mode;
527 /* The page contains code to reuse input reloads. */
529 /* Structure describes input reload of the current insns. */
530 struct input_reload
532 /* True for input reload of matched operands. */
533 bool match_p;
534 /* Reloaded value. */
535 rtx input;
536 /* Reload pseudo used. */
537 rtx reg;
540 /* The number of elements in the following array. */
541 static int curr_insn_input_reloads_num;
542 /* Array containing info about input reloads. It is used to find the
543 same input reload and reuse the reload pseudo in this case. */
544 static struct input_reload curr_insn_input_reloads[LRA_MAX_INSN_RELOADS];
546 /* Initiate data concerning reuse of input reloads for the current
547 insn. */
548 static void
549 init_curr_insn_input_reloads (void)
551 curr_insn_input_reloads_num = 0;
554 /* Create a new pseudo using MODE, RCLASS, ORIGINAL or reuse already
555 created input reload pseudo (only if TYPE is not OP_OUT). Don't
556 reuse pseudo if IN_SUBREG_P is true and the reused pseudo should be
557 wrapped up in SUBREG. The result pseudo is returned through
558 RESULT_REG. Return TRUE if we created a new pseudo, FALSE if we
559 reused the already created input reload pseudo. Use TITLE to
560 describe new registers for debug purposes. */
561 static bool
562 get_reload_reg (enum op_type type, machine_mode mode, rtx original,
563 enum reg_class rclass, bool in_subreg_p,
564 const char *title, rtx *result_reg)
566 int i, regno;
567 enum reg_class new_class;
568 bool unique_p = false;
570 if (type == OP_OUT)
572 *result_reg
573 = lra_create_new_reg_with_unique_value (mode, original, rclass, title);
574 return true;
576 /* Prevent reuse value of expression with side effects,
577 e.g. volatile memory. */
578 if (! side_effects_p (original))
579 for (i = 0; i < curr_insn_input_reloads_num; i++)
581 if (! curr_insn_input_reloads[i].match_p
582 && rtx_equal_p (curr_insn_input_reloads[i].input, original)
583 && in_class_p (curr_insn_input_reloads[i].reg, rclass, &new_class))
585 rtx reg = curr_insn_input_reloads[i].reg;
586 regno = REGNO (reg);
587 /* If input is equal to original and both are VOIDmode,
588 GET_MODE (reg) might be still different from mode.
589 Ensure we don't return *result_reg with wrong mode. */
590 if (GET_MODE (reg) != mode)
592 if (in_subreg_p)
593 continue;
594 if (GET_MODE_SIZE (GET_MODE (reg)) < GET_MODE_SIZE (mode))
595 continue;
596 reg = lowpart_subreg (mode, reg, GET_MODE (reg));
597 if (reg == NULL_RTX || GET_CODE (reg) != SUBREG)
598 continue;
600 *result_reg = reg;
601 if (lra_dump_file != NULL)
603 fprintf (lra_dump_file, " Reuse r%d for reload ", regno);
604 dump_value_slim (lra_dump_file, original, 1);
606 if (new_class != lra_get_allocno_class (regno))
607 lra_change_class (regno, new_class, ", change to", false);
608 if (lra_dump_file != NULL)
609 fprintf (lra_dump_file, "\n");
610 return false;
612 /* If we have an input reload with a different mode, make sure it
613 will get a different hard reg. */
614 else if (REG_P (original)
615 && REG_P (curr_insn_input_reloads[i].input)
616 && REGNO (original) == REGNO (curr_insn_input_reloads[i].input)
617 && (GET_MODE (original)
618 != GET_MODE (curr_insn_input_reloads[i].input)))
619 unique_p = true;
621 *result_reg = (unique_p
622 ? lra_create_new_reg_with_unique_value
623 : lra_create_new_reg) (mode, original, rclass, title);
624 lra_assert (curr_insn_input_reloads_num < LRA_MAX_INSN_RELOADS);
625 curr_insn_input_reloads[curr_insn_input_reloads_num].input = original;
626 curr_insn_input_reloads[curr_insn_input_reloads_num].match_p = false;
627 curr_insn_input_reloads[curr_insn_input_reloads_num++].reg = *result_reg;
628 return true;
633 /* The page contains code to extract memory address parts. */
635 /* Wrapper around REGNO_OK_FOR_INDEX_P, to allow pseudos. */
636 static inline bool
637 ok_for_index_p_nonstrict (rtx reg)
639 unsigned regno = REGNO (reg);
641 return regno >= FIRST_PSEUDO_REGISTER || REGNO_OK_FOR_INDEX_P (regno);
644 /* A version of regno_ok_for_base_p for use here, when all pseudos
645 should count as OK. Arguments as for regno_ok_for_base_p. */
646 static inline bool
647 ok_for_base_p_nonstrict (rtx reg, machine_mode mode, addr_space_t as,
648 enum rtx_code outer_code, enum rtx_code index_code)
650 unsigned regno = REGNO (reg);
652 if (regno >= FIRST_PSEUDO_REGISTER)
653 return true;
654 return ok_for_base_p_1 (regno, mode, as, outer_code, index_code);
659 /* The page contains major code to choose the current insn alternative
660 and generate reloads for it. */
662 /* Return the offset from REGNO of the least significant register
663 in (reg:MODE REGNO).
665 This function is used to tell whether two registers satisfy
666 a matching constraint. (reg:MODE1 REGNO1) matches (reg:MODE2 REGNO2) if:
668 REGNO1 + lra_constraint_offset (REGNO1, MODE1)
669 == REGNO2 + lra_constraint_offset (REGNO2, MODE2) */
671 lra_constraint_offset (int regno, machine_mode mode)
673 lra_assert (regno < FIRST_PSEUDO_REGISTER);
675 scalar_int_mode int_mode;
676 if (WORDS_BIG_ENDIAN
677 && is_a <scalar_int_mode> (mode, &int_mode)
678 && GET_MODE_SIZE (int_mode) > UNITS_PER_WORD)
679 return hard_regno_nregs (regno, mode) - 1;
680 return 0;
683 /* Like rtx_equal_p except that it allows a REG and a SUBREG to match
684 if they are the same hard reg, and has special hacks for
685 auto-increment and auto-decrement. This is specifically intended for
686 process_alt_operands to use in determining whether two operands
687 match. X is the operand whose number is the lower of the two.
689 It is supposed that X is the output operand and Y is the input
690 operand. Y_HARD_REGNO is the final hard regno of register Y or
691 register in subreg Y as we know it now. Otherwise, it is a
692 negative value. */
693 static bool
694 operands_match_p (rtx x, rtx y, int y_hard_regno)
696 int i;
697 RTX_CODE code = GET_CODE (x);
698 const char *fmt;
700 if (x == y)
701 return true;
702 if ((code == REG || (code == SUBREG && REG_P (SUBREG_REG (x))))
703 && (REG_P (y) || (GET_CODE (y) == SUBREG && REG_P (SUBREG_REG (y)))))
705 int j;
707 i = get_hard_regno (x, false);
708 if (i < 0)
709 goto slow;
711 if ((j = y_hard_regno) < 0)
712 goto slow;
714 i += lra_constraint_offset (i, GET_MODE (x));
715 j += lra_constraint_offset (j, GET_MODE (y));
717 return i == j;
720 /* If two operands must match, because they are really a single
721 operand of an assembler insn, then two post-increments are invalid
722 because the assembler insn would increment only once. On the
723 other hand, a post-increment matches ordinary indexing if the
724 post-increment is the output operand. */
725 if (code == POST_DEC || code == POST_INC || code == POST_MODIFY)
726 return operands_match_p (XEXP (x, 0), y, y_hard_regno);
728 /* Two pre-increments are invalid because the assembler insn would
729 increment only once. On the other hand, a pre-increment matches
730 ordinary indexing if the pre-increment is the input operand. */
731 if (GET_CODE (y) == PRE_DEC || GET_CODE (y) == PRE_INC
732 || GET_CODE (y) == PRE_MODIFY)
733 return operands_match_p (x, XEXP (y, 0), -1);
735 slow:
737 if (code == REG && REG_P (y))
738 return REGNO (x) == REGNO (y);
740 if (code == REG && GET_CODE (y) == SUBREG && REG_P (SUBREG_REG (y))
741 && x == SUBREG_REG (y))
742 return true;
743 if (GET_CODE (y) == REG && code == SUBREG && REG_P (SUBREG_REG (x))
744 && SUBREG_REG (x) == y)
745 return true;
747 /* Now we have disposed of all the cases in which different rtx
748 codes can match. */
749 if (code != GET_CODE (y))
750 return false;
752 /* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent. */
753 if (GET_MODE (x) != GET_MODE (y))
754 return false;
756 switch (code)
758 CASE_CONST_UNIQUE:
759 return false;
761 case LABEL_REF:
762 return label_ref_label (x) == label_ref_label (y);
763 case SYMBOL_REF:
764 return XSTR (x, 0) == XSTR (y, 0);
766 default:
767 break;
770 /* Compare the elements. If any pair of corresponding elements fail
771 to match, return false for the whole things. */
773 fmt = GET_RTX_FORMAT (code);
774 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
776 int val, j;
777 switch (fmt[i])
779 case 'w':
780 if (XWINT (x, i) != XWINT (y, i))
781 return false;
782 break;
784 case 'i':
785 if (XINT (x, i) != XINT (y, i))
786 return false;
787 break;
789 case 'e':
790 val = operands_match_p (XEXP (x, i), XEXP (y, i), -1);
791 if (val == 0)
792 return false;
793 break;
795 case '0':
796 break;
798 case 'E':
799 if (XVECLEN (x, i) != XVECLEN (y, i))
800 return false;
801 for (j = XVECLEN (x, i) - 1; j >= 0; --j)
803 val = operands_match_p (XVECEXP (x, i, j), XVECEXP (y, i, j), -1);
804 if (val == 0)
805 return false;
807 break;
809 /* It is believed that rtx's at this level will never
810 contain anything but integers and other rtx's, except for
811 within LABEL_REFs and SYMBOL_REFs. */
812 default:
813 gcc_unreachable ();
816 return true;
819 /* True if X is a constant that can be forced into the constant pool.
820 MODE is the mode of the operand, or VOIDmode if not known. */
821 #define CONST_POOL_OK_P(MODE, X) \
822 ((MODE) != VOIDmode \
823 && CONSTANT_P (X) \
824 && GET_CODE (X) != HIGH \
825 && !targetm.cannot_force_const_mem (MODE, X))
827 /* True if C is a non-empty register class that has too few registers
828 to be safely used as a reload target class. */
829 #define SMALL_REGISTER_CLASS_P(C) \
830 (ira_class_hard_regs_num [(C)] == 1 \
831 || (ira_class_hard_regs_num [(C)] >= 1 \
832 && targetm.class_likely_spilled_p (C)))
834 /* If REG is a reload pseudo, try to make its class satisfying CL. */
835 static void
836 narrow_reload_pseudo_class (rtx reg, enum reg_class cl)
838 enum reg_class rclass;
840 /* Do not make more accurate class from reloads generated. They are
841 mostly moves with a lot of constraints. Making more accurate
842 class may results in very narrow class and impossibility of find
843 registers for several reloads of one insn. */
844 if (INSN_UID (curr_insn) >= new_insn_uid_start)
845 return;
846 if (GET_CODE (reg) == SUBREG)
847 reg = SUBREG_REG (reg);
848 if (! REG_P (reg) || (int) REGNO (reg) < new_regno_start)
849 return;
850 if (in_class_p (reg, cl, &rclass) && rclass != cl)
851 lra_change_class (REGNO (reg), rclass, " Change to", true);
854 /* Searches X for any reference to a reg with the same value as REGNO,
855 returning the rtx of the reference found if any. Otherwise,
856 returns NULL_RTX. */
857 static rtx
858 regno_val_use_in (unsigned int regno, rtx x)
860 const char *fmt;
861 int i, j;
862 rtx tem;
864 if (REG_P (x) && lra_reg_info[REGNO (x)].val == lra_reg_info[regno].val)
865 return x;
867 fmt = GET_RTX_FORMAT (GET_CODE (x));
868 for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
870 if (fmt[i] == 'e')
872 if ((tem = regno_val_use_in (regno, XEXP (x, i))))
873 return tem;
875 else if (fmt[i] == 'E')
876 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
877 if ((tem = regno_val_use_in (regno , XVECEXP (x, i, j))))
878 return tem;
881 return NULL_RTX;
884 /* Return true if all current insn non-output operands except INS (it
885 has a negaitve end marker) do not use pseudos with the same value
886 as REGNO. */
887 static bool
888 check_conflict_input_operands (int regno, signed char *ins)
890 int in;
891 int n_operands = curr_static_id->n_operands;
893 for (int nop = 0; nop < n_operands; nop++)
894 if (! curr_static_id->operand[nop].is_operator
895 && curr_static_id->operand[nop].type != OP_OUT)
897 for (int i = 0; (in = ins[i]) >= 0; i++)
898 if (in == nop)
899 break;
900 if (in < 0
901 && regno_val_use_in (regno, *curr_id->operand_loc[nop]) != NULL_RTX)
902 return false;
904 return true;
907 /* Generate reloads for matching OUT and INS (array of input operand
908 numbers with end marker -1) with reg class GOAL_CLASS, considering
909 output operands OUTS (similar array to INS) needing to be in different
910 registers. Add input and output reloads correspondingly to the lists
911 *BEFORE and *AFTER. OUT might be negative. In this case we generate
912 input reloads for matched input operands INS. EARLY_CLOBBER_P is a flag
913 that the output operand is early clobbered for chosen alternative. */
914 static void
915 match_reload (signed char out, signed char *ins, signed char *outs,
916 enum reg_class goal_class, rtx_insn **before,
917 rtx_insn **after, bool early_clobber_p)
919 bool out_conflict;
920 int i, in;
921 rtx new_in_reg, new_out_reg, reg;
922 machine_mode inmode, outmode;
923 rtx in_rtx = *curr_id->operand_loc[ins[0]];
924 rtx out_rtx = out < 0 ? in_rtx : *curr_id->operand_loc[out];
926 inmode = curr_operand_mode[ins[0]];
927 outmode = out < 0 ? inmode : curr_operand_mode[out];
928 push_to_sequence (*before);
929 if (inmode != outmode)
931 if (partial_subreg_p (outmode, inmode))
933 reg = new_in_reg
934 = lra_create_new_reg_with_unique_value (inmode, in_rtx,
935 goal_class, "");
936 if (SCALAR_INT_MODE_P (inmode))
937 new_out_reg = gen_lowpart_SUBREG (outmode, reg);
938 else
939 new_out_reg = gen_rtx_SUBREG (outmode, reg, 0);
940 LRA_SUBREG_P (new_out_reg) = 1;
941 /* If the input reg is dying here, we can use the same hard
942 register for REG and IN_RTX. We do it only for original
943 pseudos as reload pseudos can die although original
944 pseudos still live where reload pseudos dies. */
945 if (REG_P (in_rtx) && (int) REGNO (in_rtx) < lra_new_regno_start
946 && find_regno_note (curr_insn, REG_DEAD, REGNO (in_rtx))
947 && (!early_clobber_p
948 || check_conflict_input_operands(REGNO (in_rtx), ins)))
949 lra_assign_reg_val (REGNO (in_rtx), REGNO (reg));
951 else
953 reg = new_out_reg
954 = lra_create_new_reg_with_unique_value (outmode, out_rtx,
955 goal_class, "");
956 if (SCALAR_INT_MODE_P (outmode))
957 new_in_reg = gen_lowpart_SUBREG (inmode, reg);
958 else
959 new_in_reg = gen_rtx_SUBREG (inmode, reg, 0);
960 /* NEW_IN_REG is non-paradoxical subreg. We don't want
961 NEW_OUT_REG living above. We add clobber clause for
962 this. This is just a temporary clobber. We can remove
963 it at the end of LRA work. */
964 rtx_insn *clobber = emit_clobber (new_out_reg);
965 LRA_TEMP_CLOBBER_P (PATTERN (clobber)) = 1;
966 LRA_SUBREG_P (new_in_reg) = 1;
967 if (GET_CODE (in_rtx) == SUBREG)
969 rtx subreg_reg = SUBREG_REG (in_rtx);
971 /* If SUBREG_REG is dying here and sub-registers IN_RTX
972 and NEW_IN_REG are similar, we can use the same hard
973 register for REG and SUBREG_REG. */
974 if (REG_P (subreg_reg)
975 && (int) REGNO (subreg_reg) < lra_new_regno_start
976 && GET_MODE (subreg_reg) == outmode
977 && SUBREG_BYTE (in_rtx) == SUBREG_BYTE (new_in_reg)
978 && find_regno_note (curr_insn, REG_DEAD, REGNO (subreg_reg))
979 && (! early_clobber_p
980 || check_conflict_input_operands (REGNO (subreg_reg),
981 ins)))
982 lra_assign_reg_val (REGNO (subreg_reg), REGNO (reg));
986 else
988 /* Pseudos have values -- see comments for lra_reg_info.
989 Different pseudos with the same value do not conflict even if
990 they live in the same place. When we create a pseudo we
991 assign value of original pseudo (if any) from which we
992 created the new pseudo. If we create the pseudo from the
993 input pseudo, the new pseudo will have no conflict with the
994 input pseudo which is wrong when the input pseudo lives after
995 the insn and as the new pseudo value is changed by the insn
996 output. Therefore we create the new pseudo from the output
997 except the case when we have single matched dying input
998 pseudo.
1000 We cannot reuse the current output register because we might
1001 have a situation like "a <- a op b", where the constraints
1002 force the second input operand ("b") to match the output
1003 operand ("a"). "b" must then be copied into a new register
1004 so that it doesn't clobber the current value of "a".
1006 We can not use the same value if the output pseudo is
1007 early clobbered or the input pseudo is mentioned in the
1008 output, e.g. as an address part in memory, because
1009 output reload will actually extend the pseudo liveness.
1010 We don't care about eliminable hard regs here as we are
1011 interesting only in pseudos. */
1013 /* Matching input's register value is the same as one of the other
1014 output operand. Output operands in a parallel insn must be in
1015 different registers. */
1016 out_conflict = false;
1017 if (REG_P (in_rtx))
1019 for (i = 0; outs[i] >= 0; i++)
1021 rtx other_out_rtx = *curr_id->operand_loc[outs[i]];
1022 if (REG_P (other_out_rtx)
1023 && (regno_val_use_in (REGNO (in_rtx), other_out_rtx)
1024 != NULL_RTX))
1026 out_conflict = true;
1027 break;
1032 new_in_reg = new_out_reg
1033 = (! early_clobber_p && ins[1] < 0 && REG_P (in_rtx)
1034 && (int) REGNO (in_rtx) < lra_new_regno_start
1035 && find_regno_note (curr_insn, REG_DEAD, REGNO (in_rtx))
1036 && (! early_clobber_p
1037 || check_conflict_input_operands (REGNO (in_rtx), ins))
1038 && (out < 0
1039 || regno_val_use_in (REGNO (in_rtx), out_rtx) == NULL_RTX)
1040 && !out_conflict
1041 ? lra_create_new_reg (inmode, in_rtx, goal_class, "")
1042 : lra_create_new_reg_with_unique_value (outmode, out_rtx,
1043 goal_class, ""));
1045 /* In operand can be got from transformations before processing insn
1046 constraints. One example of such transformations is subreg
1047 reloading (see function simplify_operand_subreg). The new
1048 pseudos created by the transformations might have inaccurate
1049 class (ALL_REGS) and we should make their classes more
1050 accurate. */
1051 narrow_reload_pseudo_class (in_rtx, goal_class);
1052 lra_emit_move (copy_rtx (new_in_reg), in_rtx);
1053 *before = get_insns ();
1054 end_sequence ();
1055 /* Add the new pseudo to consider values of subsequent input reload
1056 pseudos. */
1057 lra_assert (curr_insn_input_reloads_num < LRA_MAX_INSN_RELOADS);
1058 curr_insn_input_reloads[curr_insn_input_reloads_num].input = in_rtx;
1059 curr_insn_input_reloads[curr_insn_input_reloads_num].match_p = true;
1060 curr_insn_input_reloads[curr_insn_input_reloads_num++].reg = new_in_reg;
1061 for (i = 0; (in = ins[i]) >= 0; i++)
1063 lra_assert
1064 (GET_MODE (*curr_id->operand_loc[in]) == VOIDmode
1065 || GET_MODE (new_in_reg) == GET_MODE (*curr_id->operand_loc[in]));
1066 *curr_id->operand_loc[in] = new_in_reg;
1068 lra_update_dups (curr_id, ins);
1069 if (out < 0)
1070 return;
1071 /* See a comment for the input operand above. */
1072 narrow_reload_pseudo_class (out_rtx, goal_class);
1073 if (find_reg_note (curr_insn, REG_UNUSED, out_rtx) == NULL_RTX)
1075 start_sequence ();
1076 lra_emit_move (out_rtx, copy_rtx (new_out_reg));
1077 emit_insn (*after);
1078 *after = get_insns ();
1079 end_sequence ();
1081 *curr_id->operand_loc[out] = new_out_reg;
1082 lra_update_dup (curr_id, out);
1085 /* Return register class which is union of all reg classes in insn
1086 constraint alternative string starting with P. */
1087 static enum reg_class
1088 reg_class_from_constraints (const char *p)
1090 int c, len;
1091 enum reg_class op_class = NO_REGS;
1094 switch ((c = *p, len = CONSTRAINT_LEN (c, p)), c)
1096 case '#':
1097 case ',':
1098 return op_class;
1100 case 'g':
1101 op_class = reg_class_subunion[op_class][GENERAL_REGS];
1102 break;
1104 default:
1105 enum constraint_num cn = lookup_constraint (p);
1106 enum reg_class cl = reg_class_for_constraint (cn);
1107 if (cl == NO_REGS)
1109 if (insn_extra_address_constraint (cn))
1110 op_class
1111 = (reg_class_subunion
1112 [op_class][base_reg_class (VOIDmode, ADDR_SPACE_GENERIC,
1113 ADDRESS, SCRATCH)]);
1114 break;
1117 op_class = reg_class_subunion[op_class][cl];
1118 break;
1120 while ((p += len), c);
1121 return op_class;
1124 /* If OP is a register, return the class of the register as per
1125 get_reg_class, otherwise return NO_REGS. */
1126 static inline enum reg_class
1127 get_op_class (rtx op)
1129 return REG_P (op) ? get_reg_class (REGNO (op)) : NO_REGS;
1132 /* Return generated insn mem_pseudo:=val if TO_P or val:=mem_pseudo
1133 otherwise. If modes of MEM_PSEUDO and VAL are different, use
1134 SUBREG for VAL to make them equal. */
1135 static rtx_insn *
1136 emit_spill_move (bool to_p, rtx mem_pseudo, rtx val)
1138 if (GET_MODE (mem_pseudo) != GET_MODE (val))
1140 /* Usually size of mem_pseudo is greater than val size but in
1141 rare cases it can be less as it can be defined by target
1142 dependent macro HARD_REGNO_CALLER_SAVE_MODE. */
1143 if (! MEM_P (val))
1145 val = gen_lowpart_SUBREG (GET_MODE (mem_pseudo),
1146 GET_CODE (val) == SUBREG
1147 ? SUBREG_REG (val) : val);
1148 LRA_SUBREG_P (val) = 1;
1150 else
1152 mem_pseudo = gen_lowpart_SUBREG (GET_MODE (val), mem_pseudo);
1153 LRA_SUBREG_P (mem_pseudo) = 1;
1156 return to_p ? gen_move_insn (mem_pseudo, val)
1157 : gen_move_insn (val, mem_pseudo);
1160 /* Process a special case insn (register move), return true if we
1161 don't need to process it anymore. INSN should be a single set
1162 insn. Set up that RTL was changed through CHANGE_P and that hook
1163 TARGET_SECONDARY_MEMORY_NEEDED says to use secondary memory through
1164 SEC_MEM_P. */
1165 static bool
1166 check_and_process_move (bool *change_p, bool *sec_mem_p ATTRIBUTE_UNUSED)
1168 int sregno, dregno;
1169 rtx dest, src, dreg, sreg, new_reg, scratch_reg;
1170 rtx_insn *before;
1171 enum reg_class dclass, sclass, secondary_class;
1172 secondary_reload_info sri;
1174 lra_assert (curr_insn_set != NULL_RTX);
1175 dreg = dest = SET_DEST (curr_insn_set);
1176 sreg = src = SET_SRC (curr_insn_set);
1177 if (GET_CODE (dest) == SUBREG)
1178 dreg = SUBREG_REG (dest);
1179 if (GET_CODE (src) == SUBREG)
1180 sreg = SUBREG_REG (src);
1181 if (! (REG_P (dreg) || MEM_P (dreg)) || ! (REG_P (sreg) || MEM_P (sreg)))
1182 return false;
1183 sclass = dclass = NO_REGS;
1184 if (REG_P (dreg))
1185 dclass = get_reg_class (REGNO (dreg));
1186 gcc_assert (dclass < LIM_REG_CLASSES);
1187 if (dclass == ALL_REGS)
1188 /* ALL_REGS is used for new pseudos created by transformations
1189 like reload of SUBREG_REG (see function
1190 simplify_operand_subreg). We don't know their class yet. We
1191 should figure out the class from processing the insn
1192 constraints not in this fast path function. Even if ALL_REGS
1193 were a right class for the pseudo, secondary_... hooks usually
1194 are not define for ALL_REGS. */
1195 return false;
1196 if (REG_P (sreg))
1197 sclass = get_reg_class (REGNO (sreg));
1198 gcc_assert (sclass < LIM_REG_CLASSES);
1199 if (sclass == ALL_REGS)
1200 /* See comments above. */
1201 return false;
1202 if (sclass == NO_REGS && dclass == NO_REGS)
1203 return false;
1204 if (targetm.secondary_memory_needed (GET_MODE (src), sclass, dclass)
1205 && ((sclass != NO_REGS && dclass != NO_REGS)
1206 || (GET_MODE (src)
1207 != targetm.secondary_memory_needed_mode (GET_MODE (src)))))
1209 *sec_mem_p = true;
1210 return false;
1212 if (! REG_P (dreg) || ! REG_P (sreg))
1213 return false;
1214 sri.prev_sri = NULL;
1215 sri.icode = CODE_FOR_nothing;
1216 sri.extra_cost = 0;
1217 secondary_class = NO_REGS;
1218 /* Set up hard register for a reload pseudo for hook
1219 secondary_reload because some targets just ignore unassigned
1220 pseudos in the hook. */
1221 if (dclass != NO_REGS && lra_get_regno_hard_regno (REGNO (dreg)) < 0)
1223 dregno = REGNO (dreg);
1224 reg_renumber[dregno] = ira_class_hard_regs[dclass][0];
1226 else
1227 dregno = -1;
1228 if (sclass != NO_REGS && lra_get_regno_hard_regno (REGNO (sreg)) < 0)
1230 sregno = REGNO (sreg);
1231 reg_renumber[sregno] = ira_class_hard_regs[sclass][0];
1233 else
1234 sregno = -1;
1235 if (sclass != NO_REGS)
1236 secondary_class
1237 = (enum reg_class) targetm.secondary_reload (false, dest,
1238 (reg_class_t) sclass,
1239 GET_MODE (src), &sri);
1240 if (sclass == NO_REGS
1241 || ((secondary_class != NO_REGS || sri.icode != CODE_FOR_nothing)
1242 && dclass != NO_REGS))
1244 enum reg_class old_sclass = secondary_class;
1245 secondary_reload_info old_sri = sri;
1247 sri.prev_sri = NULL;
1248 sri.icode = CODE_FOR_nothing;
1249 sri.extra_cost = 0;
1250 secondary_class
1251 = (enum reg_class) targetm.secondary_reload (true, src,
1252 (reg_class_t) dclass,
1253 GET_MODE (src), &sri);
1254 /* Check the target hook consistency. */
1255 lra_assert
1256 ((secondary_class == NO_REGS && sri.icode == CODE_FOR_nothing)
1257 || (old_sclass == NO_REGS && old_sri.icode == CODE_FOR_nothing)
1258 || (secondary_class == old_sclass && sri.icode == old_sri.icode));
1260 if (sregno >= 0)
1261 reg_renumber [sregno] = -1;
1262 if (dregno >= 0)
1263 reg_renumber [dregno] = -1;
1264 if (secondary_class == NO_REGS && sri.icode == CODE_FOR_nothing)
1265 return false;
1266 *change_p = true;
1267 new_reg = NULL_RTX;
1268 if (secondary_class != NO_REGS)
1269 new_reg = lra_create_new_reg_with_unique_value (GET_MODE (src), NULL_RTX,
1270 secondary_class,
1271 "secondary");
1272 start_sequence ();
1273 if (sri.icode == CODE_FOR_nothing)
1274 lra_emit_move (new_reg, src);
1275 else
1277 enum reg_class scratch_class;
1279 scratch_class = (reg_class_from_constraints
1280 (insn_data[sri.icode].operand[2].constraint));
1281 scratch_reg = (lra_create_new_reg_with_unique_value
1282 (insn_data[sri.icode].operand[2].mode, NULL_RTX,
1283 scratch_class, "scratch"));
1284 emit_insn (GEN_FCN (sri.icode) (new_reg != NULL_RTX ? new_reg : dest,
1285 src, scratch_reg));
1287 before = get_insns ();
1288 end_sequence ();
1289 lra_process_new_insns (curr_insn, before, NULL, "Inserting the move");
1290 if (new_reg != NULL_RTX)
1291 SET_SRC (curr_insn_set) = new_reg;
1292 else
1294 if (lra_dump_file != NULL)
1296 fprintf (lra_dump_file, "Deleting move %u\n", INSN_UID (curr_insn));
1297 dump_insn_slim (lra_dump_file, curr_insn);
1299 lra_set_insn_deleted (curr_insn);
1300 return true;
1302 return false;
1305 /* The following data describe the result of process_alt_operands.
1306 The data are used in curr_insn_transform to generate reloads. */
1308 /* The chosen reg classes which should be used for the corresponding
1309 operands. */
1310 static enum reg_class goal_alt[MAX_RECOG_OPERANDS];
1311 /* True if the operand should be the same as another operand and that
1312 other operand does not need a reload. */
1313 static bool goal_alt_match_win[MAX_RECOG_OPERANDS];
1314 /* True if the operand does not need a reload. */
1315 static bool goal_alt_win[MAX_RECOG_OPERANDS];
1316 /* True if the operand can be offsetable memory. */
1317 static bool goal_alt_offmemok[MAX_RECOG_OPERANDS];
1318 /* The number of an operand to which given operand can be matched to. */
1319 static int goal_alt_matches[MAX_RECOG_OPERANDS];
1320 /* The number of elements in the following array. */
1321 static int goal_alt_dont_inherit_ops_num;
1322 /* Numbers of operands whose reload pseudos should not be inherited. */
1323 static int goal_alt_dont_inherit_ops[MAX_RECOG_OPERANDS];
1324 /* True if the insn commutative operands should be swapped. */
1325 static bool goal_alt_swapped;
1326 /* The chosen insn alternative. */
1327 static int goal_alt_number;
1329 /* True if the corresponding operand is the result of an equivalence
1330 substitution. */
1331 static bool equiv_substition_p[MAX_RECOG_OPERANDS];
1333 /* The following five variables are used to choose the best insn
1334 alternative. They reflect final characteristics of the best
1335 alternative. */
1337 /* Number of necessary reloads and overall cost reflecting the
1338 previous value and other unpleasantness of the best alternative. */
1339 static int best_losers, best_overall;
1340 /* Overall number hard registers used for reloads. For example, on
1341 some targets we need 2 general registers to reload DFmode and only
1342 one floating point register. */
1343 static int best_reload_nregs;
1344 /* Overall number reflecting distances of previous reloading the same
1345 value. The distances are counted from the current BB start. It is
1346 used to improve inheritance chances. */
1347 static int best_reload_sum;
1349 /* True if the current insn should have no correspondingly input or
1350 output reloads. */
1351 static bool no_input_reloads_p, no_output_reloads_p;
1353 /* True if we swapped the commutative operands in the current
1354 insn. */
1355 static int curr_swapped;
1357 /* if CHECK_ONLY_P is false, arrange for address element *LOC to be a
1358 register of class CL. Add any input reloads to list BEFORE. AFTER
1359 is nonnull if *LOC is an automodified value; handle that case by
1360 adding the required output reloads to list AFTER. Return true if
1361 the RTL was changed.
1363 if CHECK_ONLY_P is true, check that the *LOC is a correct address
1364 register. Return false if the address register is correct. */
1365 static bool
1366 process_addr_reg (rtx *loc, bool check_only_p, rtx_insn **before, rtx_insn **after,
1367 enum reg_class cl)
1369 int regno;
1370 enum reg_class rclass, new_class;
1371 rtx reg;
1372 rtx new_reg;
1373 machine_mode mode;
1374 bool subreg_p, before_p = false;
1376 subreg_p = GET_CODE (*loc) == SUBREG;
1377 if (subreg_p)
1379 reg = SUBREG_REG (*loc);
1380 mode = GET_MODE (reg);
1382 /* For mode with size bigger than ptr_mode, there unlikely to be "mov"
1383 between two registers with different classes, but there normally will
1384 be "mov" which transfers element of vector register into the general
1385 register, and this normally will be a subreg which should be reloaded
1386 as a whole. This is particularly likely to be triggered when
1387 -fno-split-wide-types specified. */
1388 if (!REG_P (reg)
1389 || in_class_p (reg, cl, &new_class)
1390 || GET_MODE_SIZE (mode) <= GET_MODE_SIZE (ptr_mode))
1391 loc = &SUBREG_REG (*loc);
1394 reg = *loc;
1395 mode = GET_MODE (reg);
1396 if (! REG_P (reg))
1398 if (check_only_p)
1399 return true;
1400 /* Always reload memory in an address even if the target supports
1401 such addresses. */
1402 new_reg = lra_create_new_reg_with_unique_value (mode, reg, cl, "address");
1403 before_p = true;
1405 else
1407 regno = REGNO (reg);
1408 rclass = get_reg_class (regno);
1409 if (! check_only_p
1410 && (*loc = get_equiv_with_elimination (reg, curr_insn)) != reg)
1412 if (lra_dump_file != NULL)
1414 fprintf (lra_dump_file,
1415 "Changing pseudo %d in address of insn %u on equiv ",
1416 REGNO (reg), INSN_UID (curr_insn));
1417 dump_value_slim (lra_dump_file, *loc, 1);
1418 fprintf (lra_dump_file, "\n");
1420 *loc = copy_rtx (*loc);
1422 if (*loc != reg || ! in_class_p (reg, cl, &new_class))
1424 if (check_only_p)
1425 return true;
1426 reg = *loc;
1427 if (get_reload_reg (after == NULL ? OP_IN : OP_INOUT,
1428 mode, reg, cl, subreg_p, "address", &new_reg))
1429 before_p = true;
1431 else if (new_class != NO_REGS && rclass != new_class)
1433 if (check_only_p)
1434 return true;
1435 lra_change_class (regno, new_class, " Change to", true);
1436 return false;
1438 else
1439 return false;
1441 if (before_p)
1443 push_to_sequence (*before);
1444 lra_emit_move (new_reg, reg);
1445 *before = get_insns ();
1446 end_sequence ();
1448 *loc = new_reg;
1449 if (after != NULL)
1451 start_sequence ();
1452 lra_emit_move (before_p ? copy_rtx (reg) : reg, new_reg);
1453 emit_insn (*after);
1454 *after = get_insns ();
1455 end_sequence ();
1457 return true;
1460 /* Insert move insn in simplify_operand_subreg. BEFORE returns
1461 the insn to be inserted before curr insn. AFTER returns the
1462 the insn to be inserted after curr insn. ORIGREG and NEWREG
1463 are the original reg and new reg for reload. */
1464 static void
1465 insert_move_for_subreg (rtx_insn **before, rtx_insn **after, rtx origreg,
1466 rtx newreg)
1468 if (before)
1470 push_to_sequence (*before);
1471 lra_emit_move (newreg, origreg);
1472 *before = get_insns ();
1473 end_sequence ();
1475 if (after)
1477 start_sequence ();
1478 lra_emit_move (origreg, newreg);
1479 emit_insn (*after);
1480 *after = get_insns ();
1481 end_sequence ();
1485 static int valid_address_p (machine_mode mode, rtx addr, addr_space_t as);
1486 static bool process_address (int, bool, rtx_insn **, rtx_insn **);
1488 /* Make reloads for subreg in operand NOP with internal subreg mode
1489 REG_MODE, add new reloads for further processing. Return true if
1490 any change was done. */
1491 static bool
1492 simplify_operand_subreg (int nop, machine_mode reg_mode)
1494 int hard_regno;
1495 rtx_insn *before, *after;
1496 machine_mode mode, innermode;
1497 rtx reg, new_reg;
1498 rtx operand = *curr_id->operand_loc[nop];
1499 enum reg_class regclass;
1500 enum op_type type;
1502 before = after = NULL;
1504 if (GET_CODE (operand) != SUBREG)
1505 return false;
1507 mode = GET_MODE (operand);
1508 reg = SUBREG_REG (operand);
1509 innermode = GET_MODE (reg);
1510 type = curr_static_id->operand[nop].type;
1511 if (MEM_P (reg))
1513 const bool addr_was_valid
1514 = valid_address_p (innermode, XEXP (reg, 0), MEM_ADDR_SPACE (reg));
1515 alter_subreg (curr_id->operand_loc[nop], false);
1516 rtx subst = *curr_id->operand_loc[nop];
1517 lra_assert (MEM_P (subst));
1519 if (!addr_was_valid
1520 || valid_address_p (GET_MODE (subst), XEXP (subst, 0),
1521 MEM_ADDR_SPACE (subst))
1522 || ((get_constraint_type (lookup_constraint
1523 (curr_static_id->operand[nop].constraint))
1524 != CT_SPECIAL_MEMORY)
1525 /* We still can reload address and if the address is
1526 valid, we can remove subreg without reloading its
1527 inner memory. */
1528 && valid_address_p (GET_MODE (subst),
1529 regno_reg_rtx
1530 [ira_class_hard_regs
1531 [base_reg_class (GET_MODE (subst),
1532 MEM_ADDR_SPACE (subst),
1533 ADDRESS, SCRATCH)][0]],
1534 MEM_ADDR_SPACE (subst))))
1536 /* If we change the address for a paradoxical subreg of memory, the
1537 new address might violate the necessary alignment or the access
1538 might be slow; take this into consideration. We need not worry
1539 about accesses beyond allocated memory for paradoxical memory
1540 subregs as we don't substitute such equiv memory (see processing
1541 equivalences in function lra_constraints) and because for spilled
1542 pseudos we allocate stack memory enough for the biggest
1543 corresponding paradoxical subreg.
1545 However, do not blindly simplify a (subreg (mem ...)) for
1546 WORD_REGISTER_OPERATIONS targets as this may lead to loading junk
1547 data into a register when the inner is narrower than outer or
1548 missing important data from memory when the inner is wider than
1549 outer. This rule only applies to modes that are no wider than
1550 a word. */
1551 if (!(GET_MODE_PRECISION (mode) != GET_MODE_PRECISION (innermode)
1552 && GET_MODE_SIZE (mode) <= UNITS_PER_WORD
1553 && GET_MODE_SIZE (innermode) <= UNITS_PER_WORD
1554 && WORD_REGISTER_OPERATIONS)
1555 && (!(MEM_ALIGN (subst) < GET_MODE_ALIGNMENT (mode)
1556 && targetm.slow_unaligned_access (mode, MEM_ALIGN (subst)))
1557 || (MEM_ALIGN (reg) < GET_MODE_ALIGNMENT (innermode)
1558 && targetm.slow_unaligned_access (innermode,
1559 MEM_ALIGN (reg)))))
1560 return true;
1562 *curr_id->operand_loc[nop] = operand;
1564 /* But if the address was not valid, we cannot reload the MEM without
1565 reloading the address first. */
1566 if (!addr_was_valid)
1567 process_address (nop, false, &before, &after);
1569 /* INNERMODE is fast, MODE slow. Reload the mem in INNERMODE. */
1570 enum reg_class rclass
1571 = (enum reg_class) targetm.preferred_reload_class (reg, ALL_REGS);
1572 if (get_reload_reg (curr_static_id->operand[nop].type, innermode,
1573 reg, rclass, TRUE, "slow mem", &new_reg))
1575 bool insert_before, insert_after;
1576 bitmap_set_bit (&lra_subreg_reload_pseudos, REGNO (new_reg));
1578 insert_before = (type != OP_OUT
1579 || partial_subreg_p (mode, innermode));
1580 insert_after = type != OP_IN;
1581 insert_move_for_subreg (insert_before ? &before : NULL,
1582 insert_after ? &after : NULL,
1583 reg, new_reg);
1585 SUBREG_REG (operand) = new_reg;
1587 /* Convert to MODE. */
1588 reg = operand;
1589 rclass
1590 = (enum reg_class) targetm.preferred_reload_class (reg, ALL_REGS);
1591 if (get_reload_reg (curr_static_id->operand[nop].type, mode, reg,
1592 rclass, TRUE, "slow mem", &new_reg))
1594 bool insert_before, insert_after;
1595 bitmap_set_bit (&lra_subreg_reload_pseudos, REGNO (new_reg));
1597 insert_before = type != OP_OUT;
1598 insert_after = type != OP_IN;
1599 insert_move_for_subreg (insert_before ? &before : NULL,
1600 insert_after ? &after : NULL,
1601 reg, new_reg);
1603 *curr_id->operand_loc[nop] = new_reg;
1604 lra_process_new_insns (curr_insn, before, after,
1605 "Inserting slow mem reload");
1606 return true;
1609 /* If the address was valid and became invalid, prefer to reload
1610 the memory. Typical case is when the index scale should
1611 correspond the memory. */
1612 *curr_id->operand_loc[nop] = operand;
1613 /* Do not return false here as the MEM_P (reg) will be processed
1614 later in this function. */
1616 else if (REG_P (reg) && REGNO (reg) < FIRST_PSEUDO_REGISTER)
1618 alter_subreg (curr_id->operand_loc[nop], false);
1619 return true;
1621 else if (CONSTANT_P (reg))
1623 /* Try to simplify subreg of constant. It is usually result of
1624 equivalence substitution. */
1625 if (innermode == VOIDmode
1626 && (innermode = original_subreg_reg_mode[nop]) == VOIDmode)
1627 innermode = curr_static_id->operand[nop].mode;
1628 if ((new_reg = simplify_subreg (mode, reg, innermode,
1629 SUBREG_BYTE (operand))) != NULL_RTX)
1631 *curr_id->operand_loc[nop] = new_reg;
1632 return true;
1635 /* Put constant into memory when we have mixed modes. It generates
1636 a better code in most cases as it does not need a secondary
1637 reload memory. It also prevents LRA looping when LRA is using
1638 secondary reload memory again and again. */
1639 if (CONSTANT_P (reg) && CONST_POOL_OK_P (reg_mode, reg)
1640 && SCALAR_INT_MODE_P (reg_mode) != SCALAR_INT_MODE_P (mode))
1642 SUBREG_REG (operand) = force_const_mem (reg_mode, reg);
1643 alter_subreg (curr_id->operand_loc[nop], false);
1644 return true;
1646 /* Force a reload of the SUBREG_REG if this is a constant or PLUS or
1647 if there may be a problem accessing OPERAND in the outer
1648 mode. */
1649 if ((REG_P (reg)
1650 && REGNO (reg) >= FIRST_PSEUDO_REGISTER
1651 && (hard_regno = lra_get_regno_hard_regno (REGNO (reg))) >= 0
1652 /* Don't reload paradoxical subregs because we could be looping
1653 having repeatedly final regno out of hard regs range. */
1654 && (hard_regno_nregs (hard_regno, innermode)
1655 >= hard_regno_nregs (hard_regno, mode))
1656 && simplify_subreg_regno (hard_regno, innermode,
1657 SUBREG_BYTE (operand), mode) < 0
1658 /* Don't reload subreg for matching reload. It is actually
1659 valid subreg in LRA. */
1660 && ! LRA_SUBREG_P (operand))
1661 || CONSTANT_P (reg) || GET_CODE (reg) == PLUS || MEM_P (reg))
1663 enum reg_class rclass;
1665 if (REG_P (reg))
1666 /* There is a big probability that we will get the same class
1667 for the new pseudo and we will get the same insn which
1668 means infinite looping. So spill the new pseudo. */
1669 rclass = NO_REGS;
1670 else
1671 /* The class will be defined later in curr_insn_transform. */
1672 rclass
1673 = (enum reg_class) targetm.preferred_reload_class (reg, ALL_REGS);
1675 if (get_reload_reg (curr_static_id->operand[nop].type, reg_mode, reg,
1676 rclass, TRUE, "subreg reg", &new_reg))
1678 bool insert_before, insert_after;
1679 bitmap_set_bit (&lra_subreg_reload_pseudos, REGNO (new_reg));
1681 insert_before = (type != OP_OUT
1682 || read_modify_subreg_p (operand));
1683 insert_after = (type != OP_IN);
1684 insert_move_for_subreg (insert_before ? &before : NULL,
1685 insert_after ? &after : NULL,
1686 reg, new_reg);
1688 SUBREG_REG (operand) = new_reg;
1689 lra_process_new_insns (curr_insn, before, after,
1690 "Inserting subreg reload");
1691 return true;
1693 /* Force a reload for a paradoxical subreg. For paradoxical subreg,
1694 IRA allocates hardreg to the inner pseudo reg according to its mode
1695 instead of the outermode, so the size of the hardreg may not be enough
1696 to contain the outermode operand, in that case we may need to insert
1697 reload for the reg. For the following two types of paradoxical subreg,
1698 we need to insert reload:
1699 1. If the op_type is OP_IN, and the hardreg could not be paired with
1700 other hardreg to contain the outermode operand
1701 (checked by in_hard_reg_set_p), we need to insert the reload.
1702 2. If the op_type is OP_OUT or OP_INOUT.
1704 Here is a paradoxical subreg example showing how the reload is generated:
1706 (insn 5 4 7 2 (set (reg:TI 106 [ __comp ])
1707 (subreg:TI (reg:DI 107 [ __comp ]) 0)) {*movti_internal_rex64}
1709 In IRA, reg107 is allocated to a DImode hardreg. We use x86-64 as example
1710 here, if reg107 is assigned to hardreg R15, because R15 is the last
1711 hardreg, compiler cannot find another hardreg to pair with R15 to
1712 contain TImode data. So we insert a TImode reload reg180 for it.
1713 After reload is inserted:
1715 (insn 283 0 0 (set (subreg:DI (reg:TI 180 [orig:107 __comp ] [107]) 0)
1716 (reg:DI 107 [ __comp ])) -1
1717 (insn 5 4 7 2 (set (reg:TI 106 [ __comp ])
1718 (subreg:TI (reg:TI 180 [orig:107 __comp ] [107]) 0)) {*movti_internal_rex64}
1720 Two reload hard registers will be allocated to reg180 to save TImode data
1721 in LRA_assign. */
1722 else if (REG_P (reg)
1723 && REGNO (reg) >= FIRST_PSEUDO_REGISTER
1724 && (hard_regno = lra_get_regno_hard_regno (REGNO (reg))) >= 0
1725 && (hard_regno_nregs (hard_regno, innermode)
1726 < hard_regno_nregs (hard_regno, mode))
1727 && (regclass = lra_get_allocno_class (REGNO (reg)))
1728 && (type != OP_IN
1729 || !in_hard_reg_set_p (reg_class_contents[regclass],
1730 mode, hard_regno)))
1732 /* The class will be defined later in curr_insn_transform. */
1733 enum reg_class rclass
1734 = (enum reg_class) targetm.preferred_reload_class (reg, ALL_REGS);
1736 if (get_reload_reg (curr_static_id->operand[nop].type, mode, reg,
1737 rclass, TRUE, "paradoxical subreg", &new_reg))
1739 rtx subreg;
1740 bool insert_before, insert_after;
1742 PUT_MODE (new_reg, mode);
1743 subreg = gen_lowpart_SUBREG (innermode, new_reg);
1744 bitmap_set_bit (&lra_subreg_reload_pseudos, REGNO (new_reg));
1746 insert_before = (type != OP_OUT);
1747 insert_after = (type != OP_IN);
1748 insert_move_for_subreg (insert_before ? &before : NULL,
1749 insert_after ? &after : NULL,
1750 reg, subreg);
1752 SUBREG_REG (operand) = new_reg;
1753 lra_process_new_insns (curr_insn, before, after,
1754 "Inserting paradoxical subreg reload");
1755 return true;
1757 return false;
1760 /* Return TRUE if X refers for a hard register from SET. */
1761 static bool
1762 uses_hard_regs_p (rtx x, HARD_REG_SET set)
1764 int i, j, x_hard_regno;
1765 machine_mode mode;
1766 const char *fmt;
1767 enum rtx_code code;
1769 if (x == NULL_RTX)
1770 return false;
1771 code = GET_CODE (x);
1772 mode = GET_MODE (x);
1773 if (code == SUBREG)
1775 mode = wider_subreg_mode (x);
1776 x = SUBREG_REG (x);
1777 code = GET_CODE (x);
1780 if (REG_P (x))
1782 x_hard_regno = get_hard_regno (x, true);
1783 return (x_hard_regno >= 0
1784 && overlaps_hard_reg_set_p (set, mode, x_hard_regno));
1786 if (MEM_P (x))
1788 struct address_info ad;
1790 decompose_mem_address (&ad, x);
1791 if (ad.base_term != NULL && uses_hard_regs_p (*ad.base_term, set))
1792 return true;
1793 if (ad.index_term != NULL && uses_hard_regs_p (*ad.index_term, set))
1794 return true;
1796 fmt = GET_RTX_FORMAT (code);
1797 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
1799 if (fmt[i] == 'e')
1801 if (uses_hard_regs_p (XEXP (x, i), set))
1802 return true;
1804 else if (fmt[i] == 'E')
1806 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
1807 if (uses_hard_regs_p (XVECEXP (x, i, j), set))
1808 return true;
1811 return false;
1814 /* Return true if OP is a spilled pseudo. */
1815 static inline bool
1816 spilled_pseudo_p (rtx op)
1818 return (REG_P (op)
1819 && REGNO (op) >= FIRST_PSEUDO_REGISTER && in_mem_p (REGNO (op)));
1822 /* Return true if X is a general constant. */
1823 static inline bool
1824 general_constant_p (rtx x)
1826 return CONSTANT_P (x) && (! flag_pic || LEGITIMATE_PIC_OPERAND_P (x));
1829 static bool
1830 reg_in_class_p (rtx reg, enum reg_class cl)
1832 if (cl == NO_REGS)
1833 return get_reg_class (REGNO (reg)) == NO_REGS;
1834 return in_class_p (reg, cl, NULL);
1837 /* Return true if SET of RCLASS contains no hard regs which can be
1838 used in MODE. */
1839 static bool
1840 prohibited_class_reg_set_mode_p (enum reg_class rclass,
1841 HARD_REG_SET &set,
1842 machine_mode mode)
1844 HARD_REG_SET temp;
1846 lra_assert (hard_reg_set_subset_p (reg_class_contents[rclass], set));
1847 COPY_HARD_REG_SET (temp, set);
1848 AND_COMPL_HARD_REG_SET (temp, lra_no_alloc_regs);
1849 return (hard_reg_set_subset_p
1850 (temp, ira_prohibited_class_mode_regs[rclass][mode]));
1854 /* Used to check validity info about small class input operands. It
1855 should be incremented at start of processing an insn
1856 alternative. */
1857 static unsigned int curr_small_class_check = 0;
1859 /* Update number of used inputs of class OP_CLASS for operand NOP.
1860 Return true if we have more such class operands than the number of
1861 available regs. */
1862 static bool
1863 update_and_check_small_class_inputs (int nop, enum reg_class op_class)
1865 static unsigned int small_class_check[LIM_REG_CLASSES];
1866 static int small_class_input_nums[LIM_REG_CLASSES];
1868 if (SMALL_REGISTER_CLASS_P (op_class)
1869 /* We are interesting in classes became small because of fixing
1870 some hard regs, e.g. by an user through GCC options. */
1871 && hard_reg_set_intersect_p (reg_class_contents[op_class],
1872 ira_no_alloc_regs)
1873 && (curr_static_id->operand[nop].type != OP_OUT
1874 || curr_static_id->operand[nop].early_clobber))
1876 if (small_class_check[op_class] == curr_small_class_check)
1877 small_class_input_nums[op_class]++;
1878 else
1880 small_class_check[op_class] = curr_small_class_check;
1881 small_class_input_nums[op_class] = 1;
1883 if (small_class_input_nums[op_class] > ira_class_hard_regs_num[op_class])
1884 return true;
1886 return false;
1889 /* Major function to choose the current insn alternative and what
1890 operands should be reloaded and how. If ONLY_ALTERNATIVE is not
1891 negative we should consider only this alternative. Return false if
1892 we can not choose the alternative or find how to reload the
1893 operands. */
1894 static bool
1895 process_alt_operands (int only_alternative)
1897 bool ok_p = false;
1898 int nop, overall, nalt;
1899 int n_alternatives = curr_static_id->n_alternatives;
1900 int n_operands = curr_static_id->n_operands;
1901 /* LOSERS counts the operands that don't fit this alternative and
1902 would require loading. */
1903 int losers;
1904 int addr_losers;
1905 /* REJECT is a count of how undesirable this alternative says it is
1906 if any reloading is required. If the alternative matches exactly
1907 then REJECT is ignored, but otherwise it gets this much counted
1908 against it in addition to the reloading needed. */
1909 int reject;
1910 /* This is defined by '!' or '?' alternative constraint and added to
1911 reject. But in some cases it can be ignored. */
1912 int static_reject;
1913 int op_reject;
1914 /* The number of elements in the following array. */
1915 int early_clobbered_regs_num;
1916 /* Numbers of operands which are early clobber registers. */
1917 int early_clobbered_nops[MAX_RECOG_OPERANDS];
1918 enum reg_class curr_alt[MAX_RECOG_OPERANDS];
1919 HARD_REG_SET curr_alt_set[MAX_RECOG_OPERANDS];
1920 bool curr_alt_match_win[MAX_RECOG_OPERANDS];
1921 bool curr_alt_win[MAX_RECOG_OPERANDS];
1922 bool curr_alt_offmemok[MAX_RECOG_OPERANDS];
1923 int curr_alt_matches[MAX_RECOG_OPERANDS];
1924 /* The number of elements in the following array. */
1925 int curr_alt_dont_inherit_ops_num;
1926 /* Numbers of operands whose reload pseudos should not be inherited. */
1927 int curr_alt_dont_inherit_ops[MAX_RECOG_OPERANDS];
1928 rtx op;
1929 /* The register when the operand is a subreg of register, otherwise the
1930 operand itself. */
1931 rtx no_subreg_reg_operand[MAX_RECOG_OPERANDS];
1932 /* The register if the operand is a register or subreg of register,
1933 otherwise NULL. */
1934 rtx operand_reg[MAX_RECOG_OPERANDS];
1935 int hard_regno[MAX_RECOG_OPERANDS];
1936 machine_mode biggest_mode[MAX_RECOG_OPERANDS];
1937 int reload_nregs, reload_sum;
1938 bool costly_p;
1939 enum reg_class cl;
1941 /* Calculate some data common for all alternatives to speed up the
1942 function. */
1943 for (nop = 0; nop < n_operands; nop++)
1945 rtx reg;
1947 op = no_subreg_reg_operand[nop] = *curr_id->operand_loc[nop];
1948 /* The real hard regno of the operand after the allocation. */
1949 hard_regno[nop] = get_hard_regno (op, true);
1951 operand_reg[nop] = reg = op;
1952 biggest_mode[nop] = GET_MODE (op);
1953 if (GET_CODE (op) == SUBREG)
1955 biggest_mode[nop] = wider_subreg_mode (op);
1956 operand_reg[nop] = reg = SUBREG_REG (op);
1958 if (! REG_P (reg))
1959 operand_reg[nop] = NULL_RTX;
1960 else if (REGNO (reg) >= FIRST_PSEUDO_REGISTER
1961 || ((int) REGNO (reg)
1962 == lra_get_elimination_hard_regno (REGNO (reg))))
1963 no_subreg_reg_operand[nop] = reg;
1964 else
1965 operand_reg[nop] = no_subreg_reg_operand[nop]
1966 /* Just use natural mode for elimination result. It should
1967 be enough for extra constraints hooks. */
1968 = regno_reg_rtx[hard_regno[nop]];
1971 /* The constraints are made of several alternatives. Each operand's
1972 constraint looks like foo,bar,... with commas separating the
1973 alternatives. The first alternatives for all operands go
1974 together, the second alternatives go together, etc.
1976 First loop over alternatives. */
1977 alternative_mask preferred = curr_id->preferred_alternatives;
1978 if (only_alternative >= 0)
1979 preferred &= ALTERNATIVE_BIT (only_alternative);
1981 for (nalt = 0; nalt < n_alternatives; nalt++)
1983 /* Loop over operands for one constraint alternative. */
1984 if (!TEST_BIT (preferred, nalt))
1985 continue;
1987 curr_small_class_check++;
1988 overall = losers = addr_losers = 0;
1989 static_reject = reject = reload_nregs = reload_sum = 0;
1990 for (nop = 0; nop < n_operands; nop++)
1992 int inc = (curr_static_id
1993 ->operand_alternative[nalt * n_operands + nop].reject);
1994 if (lra_dump_file != NULL && inc != 0)
1995 fprintf (lra_dump_file,
1996 " Staticly defined alt reject+=%d\n", inc);
1997 static_reject += inc;
1999 reject += static_reject;
2000 early_clobbered_regs_num = 0;
2002 for (nop = 0; nop < n_operands; nop++)
2004 const char *p;
2005 char *end;
2006 int len, c, m, i, opalt_num, this_alternative_matches;
2007 bool win, did_match, offmemok, early_clobber_p;
2008 /* false => this operand can be reloaded somehow for this
2009 alternative. */
2010 bool badop;
2011 /* true => this operand can be reloaded if the alternative
2012 allows regs. */
2013 bool winreg;
2014 /* True if a constant forced into memory would be OK for
2015 this operand. */
2016 bool constmemok;
2017 enum reg_class this_alternative, this_costly_alternative;
2018 HARD_REG_SET this_alternative_set, this_costly_alternative_set;
2019 bool this_alternative_match_win, this_alternative_win;
2020 bool this_alternative_offmemok;
2021 bool scratch_p;
2022 machine_mode mode;
2023 enum constraint_num cn;
2025 opalt_num = nalt * n_operands + nop;
2026 if (curr_static_id->operand_alternative[opalt_num].anything_ok)
2028 /* Fast track for no constraints at all. */
2029 curr_alt[nop] = NO_REGS;
2030 CLEAR_HARD_REG_SET (curr_alt_set[nop]);
2031 curr_alt_win[nop] = true;
2032 curr_alt_match_win[nop] = false;
2033 curr_alt_offmemok[nop] = false;
2034 curr_alt_matches[nop] = -1;
2035 continue;
2038 op = no_subreg_reg_operand[nop];
2039 mode = curr_operand_mode[nop];
2041 win = did_match = winreg = offmemok = constmemok = false;
2042 badop = true;
2044 early_clobber_p = false;
2045 p = curr_static_id->operand_alternative[opalt_num].constraint;
2047 this_costly_alternative = this_alternative = NO_REGS;
2048 /* We update set of possible hard regs besides its class
2049 because reg class might be inaccurate. For example,
2050 union of LO_REGS (l), HI_REGS(h), and STACK_REG(k) in ARM
2051 is translated in HI_REGS because classes are merged by
2052 pairs and there is no accurate intermediate class. */
2053 CLEAR_HARD_REG_SET (this_alternative_set);
2054 CLEAR_HARD_REG_SET (this_costly_alternative_set);
2055 this_alternative_win = false;
2056 this_alternative_match_win = false;
2057 this_alternative_offmemok = false;
2058 this_alternative_matches = -1;
2060 /* An empty constraint should be excluded by the fast
2061 track. */
2062 lra_assert (*p != 0 && *p != ',');
2064 op_reject = 0;
2065 /* Scan this alternative's specs for this operand; set WIN
2066 if the operand fits any letter in this alternative.
2067 Otherwise, clear BADOP if this operand could fit some
2068 letter after reloads, or set WINREG if this operand could
2069 fit after reloads provided the constraint allows some
2070 registers. */
2071 costly_p = false;
2074 switch ((c = *p, len = CONSTRAINT_LEN (c, p)), c)
2076 case '\0':
2077 len = 0;
2078 break;
2079 case ',':
2080 c = '\0';
2081 break;
2083 case '&':
2084 early_clobber_p = true;
2085 break;
2087 case '$':
2088 op_reject += LRA_MAX_REJECT;
2089 break;
2090 case '^':
2091 op_reject += LRA_LOSER_COST_FACTOR;
2092 break;
2094 case '#':
2095 /* Ignore rest of this alternative. */
2096 c = '\0';
2097 break;
2099 case '0': case '1': case '2': case '3': case '4':
2100 case '5': case '6': case '7': case '8': case '9':
2102 int m_hregno;
2103 bool match_p;
2105 m = strtoul (p, &end, 10);
2106 p = end;
2107 len = 0;
2108 lra_assert (nop > m);
2110 this_alternative_matches = m;
2111 m_hregno = get_hard_regno (*curr_id->operand_loc[m], false);
2112 /* We are supposed to match a previous operand.
2113 If we do, we win if that one did. If we do
2114 not, count both of the operands as losers.
2115 (This is too conservative, since most of the
2116 time only a single reload insn will be needed
2117 to make the two operands win. As a result,
2118 this alternative may be rejected when it is
2119 actually desirable.) */
2120 match_p = false;
2121 if (operands_match_p (*curr_id->operand_loc[nop],
2122 *curr_id->operand_loc[m], m_hregno))
2124 /* We should reject matching of an early
2125 clobber operand if the matching operand is
2126 not dying in the insn. */
2127 if (! curr_static_id->operand[m].early_clobber
2128 || operand_reg[nop] == NULL_RTX
2129 || (find_regno_note (curr_insn, REG_DEAD,
2130 REGNO (op))
2131 || REGNO (op) == REGNO (operand_reg[m])))
2132 match_p = true;
2134 if (match_p)
2136 /* If we are matching a non-offsettable
2137 address where an offsettable address was
2138 expected, then we must reject this
2139 combination, because we can't reload
2140 it. */
2141 if (curr_alt_offmemok[m]
2142 && MEM_P (*curr_id->operand_loc[m])
2143 && curr_alt[m] == NO_REGS && ! curr_alt_win[m])
2144 continue;
2146 else
2148 /* Operands don't match. Both operands must
2149 allow a reload register, otherwise we
2150 cannot make them match. */
2151 if (curr_alt[m] == NO_REGS)
2152 break;
2153 /* Retroactively mark the operand we had to
2154 match as a loser, if it wasn't already and
2155 it wasn't matched to a register constraint
2156 (e.g it might be matched by memory). */
2157 if (curr_alt_win[m]
2158 && (operand_reg[m] == NULL_RTX
2159 || hard_regno[m] < 0))
2161 losers++;
2162 reload_nregs
2163 += (ira_reg_class_max_nregs[curr_alt[m]]
2164 [GET_MODE (*curr_id->operand_loc[m])]);
2167 /* Prefer matching earlyclobber alternative as
2168 it results in less hard regs required for
2169 the insn than a non-matching earlyclobber
2170 alternative. */
2171 if (curr_static_id->operand[m].early_clobber)
2173 if (lra_dump_file != NULL)
2174 fprintf
2175 (lra_dump_file,
2176 " %d Matching earlyclobber alt:"
2177 " reject--\n",
2178 nop);
2179 reject--;
2181 /* Otherwise we prefer no matching
2182 alternatives because it gives more freedom
2183 in RA. */
2184 else if (operand_reg[nop] == NULL_RTX
2185 || (find_regno_note (curr_insn, REG_DEAD,
2186 REGNO (operand_reg[nop]))
2187 == NULL_RTX))
2189 if (lra_dump_file != NULL)
2190 fprintf
2191 (lra_dump_file,
2192 " %d Matching alt: reject+=2\n",
2193 nop);
2194 reject += 2;
2197 /* If we have to reload this operand and some
2198 previous operand also had to match the same
2199 thing as this operand, we don't know how to do
2200 that. */
2201 if (!match_p || !curr_alt_win[m])
2203 for (i = 0; i < nop; i++)
2204 if (curr_alt_matches[i] == m)
2205 break;
2206 if (i < nop)
2207 break;
2209 else
2210 did_match = true;
2212 /* This can be fixed with reloads if the operand
2213 we are supposed to match can be fixed with
2214 reloads. */
2215 badop = false;
2216 this_alternative = curr_alt[m];
2217 COPY_HARD_REG_SET (this_alternative_set, curr_alt_set[m]);
2218 winreg = this_alternative != NO_REGS;
2219 break;
2222 case 'g':
2223 if (MEM_P (op)
2224 || general_constant_p (op)
2225 || spilled_pseudo_p (op))
2226 win = true;
2227 cl = GENERAL_REGS;
2228 goto reg;
2230 default:
2231 cn = lookup_constraint (p);
2232 switch (get_constraint_type (cn))
2234 case CT_REGISTER:
2235 cl = reg_class_for_constraint (cn);
2236 if (cl != NO_REGS)
2237 goto reg;
2238 break;
2240 case CT_CONST_INT:
2241 if (CONST_INT_P (op)
2242 && insn_const_int_ok_for_constraint (INTVAL (op), cn))
2243 win = true;
2244 break;
2246 case CT_MEMORY:
2247 if (MEM_P (op)
2248 && satisfies_memory_constraint_p (op, cn))
2249 win = true;
2250 else if (spilled_pseudo_p (op))
2251 win = true;
2253 /* If we didn't already win, we can reload constants
2254 via force_const_mem or put the pseudo value into
2255 memory, or make other memory by reloading the
2256 address like for 'o'. */
2257 if (CONST_POOL_OK_P (mode, op)
2258 || MEM_P (op) || REG_P (op)
2259 /* We can restore the equiv insn by a
2260 reload. */
2261 || equiv_substition_p[nop])
2262 badop = false;
2263 constmemok = true;
2264 offmemok = true;
2265 break;
2267 case CT_ADDRESS:
2268 /* If we didn't already win, we can reload the address
2269 into a base register. */
2270 if (satisfies_address_constraint_p (op, cn))
2271 win = true;
2272 cl = base_reg_class (VOIDmode, ADDR_SPACE_GENERIC,
2273 ADDRESS, SCRATCH);
2274 badop = false;
2275 goto reg;
2277 case CT_FIXED_FORM:
2278 if (constraint_satisfied_p (op, cn))
2279 win = true;
2280 break;
2282 case CT_SPECIAL_MEMORY:
2283 if (MEM_P (op)
2284 && satisfies_memory_constraint_p (op, cn))
2285 win = true;
2286 else if (spilled_pseudo_p (op))
2287 win = true;
2288 break;
2290 break;
2292 reg:
2293 this_alternative = reg_class_subunion[this_alternative][cl];
2294 IOR_HARD_REG_SET (this_alternative_set,
2295 reg_class_contents[cl]);
2296 if (costly_p)
2298 this_costly_alternative
2299 = reg_class_subunion[this_costly_alternative][cl];
2300 IOR_HARD_REG_SET (this_costly_alternative_set,
2301 reg_class_contents[cl]);
2303 if (mode == BLKmode)
2304 break;
2305 winreg = true;
2306 if (REG_P (op))
2308 if (hard_regno[nop] >= 0
2309 && in_hard_reg_set_p (this_alternative_set,
2310 mode, hard_regno[nop]))
2311 win = true;
2312 else if (hard_regno[nop] < 0
2313 && in_class_p (op, this_alternative, NULL))
2314 win = true;
2316 break;
2318 if (c != ' ' && c != '\t')
2319 costly_p = c == '*';
2321 while ((p += len), c);
2323 scratch_p = (operand_reg[nop] != NULL_RTX
2324 && lra_former_scratch_p (REGNO (operand_reg[nop])));
2325 /* Record which operands fit this alternative. */
2326 if (win)
2328 this_alternative_win = true;
2329 if (operand_reg[nop] != NULL_RTX)
2331 if (hard_regno[nop] >= 0)
2333 if (in_hard_reg_set_p (this_costly_alternative_set,
2334 mode, hard_regno[nop]))
2336 if (lra_dump_file != NULL)
2337 fprintf (lra_dump_file,
2338 " %d Costly set: reject++\n",
2339 nop);
2340 reject++;
2343 else
2345 /* Prefer won reg to spilled pseudo under other
2346 equal conditions for possibe inheritance. */
2347 if (! scratch_p)
2349 if (lra_dump_file != NULL)
2350 fprintf
2351 (lra_dump_file,
2352 " %d Non pseudo reload: reject++\n",
2353 nop);
2354 reject++;
2356 if (in_class_p (operand_reg[nop],
2357 this_costly_alternative, NULL))
2359 if (lra_dump_file != NULL)
2360 fprintf
2361 (lra_dump_file,
2362 " %d Non pseudo costly reload:"
2363 " reject++\n",
2364 nop);
2365 reject++;
2368 /* We simulate the behavior of old reload here.
2369 Although scratches need hard registers and it
2370 might result in spilling other pseudos, no reload
2371 insns are generated for the scratches. So it
2372 might cost something but probably less than old
2373 reload pass believes. */
2374 if (scratch_p)
2376 if (lra_dump_file != NULL)
2377 fprintf (lra_dump_file,
2378 " %d Scratch win: reject+=2\n",
2379 nop);
2380 reject += 2;
2384 else if (did_match)
2385 this_alternative_match_win = true;
2386 else
2388 int const_to_mem = 0;
2389 bool no_regs_p;
2391 reject += op_reject;
2392 /* Never do output reload of stack pointer. It makes
2393 impossible to do elimination when SP is changed in
2394 RTL. */
2395 if (op == stack_pointer_rtx && ! frame_pointer_needed
2396 && curr_static_id->operand[nop].type != OP_IN)
2397 goto fail;
2399 /* If this alternative asks for a specific reg class, see if there
2400 is at least one allocatable register in that class. */
2401 no_regs_p
2402 = (this_alternative == NO_REGS
2403 || (hard_reg_set_subset_p
2404 (reg_class_contents[this_alternative],
2405 lra_no_alloc_regs)));
2407 /* For asms, verify that the class for this alternative is possible
2408 for the mode that is specified. */
2409 if (!no_regs_p && INSN_CODE (curr_insn) < 0)
2411 int i;
2412 for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
2413 if (targetm.hard_regno_mode_ok (i, mode)
2414 && in_hard_reg_set_p (reg_class_contents[this_alternative],
2415 mode, i))
2416 break;
2417 if (i == FIRST_PSEUDO_REGISTER)
2418 winreg = false;
2421 /* If this operand accepts a register, and if the
2422 register class has at least one allocatable register,
2423 then this operand can be reloaded. */
2424 if (winreg && !no_regs_p)
2425 badop = false;
2427 if (badop)
2429 if (lra_dump_file != NULL)
2430 fprintf (lra_dump_file,
2431 " alt=%d: Bad operand -- refuse\n",
2432 nalt);
2433 goto fail;
2436 if (this_alternative != NO_REGS)
2438 HARD_REG_SET available_regs;
2440 COPY_HARD_REG_SET (available_regs,
2441 reg_class_contents[this_alternative]);
2442 AND_COMPL_HARD_REG_SET
2443 (available_regs,
2444 ira_prohibited_class_mode_regs[this_alternative][mode]);
2445 AND_COMPL_HARD_REG_SET (available_regs, lra_no_alloc_regs);
2446 if (hard_reg_set_empty_p (available_regs))
2448 /* There are no hard regs holding a value of given
2449 mode. */
2450 if (offmemok)
2452 this_alternative = NO_REGS;
2453 if (lra_dump_file != NULL)
2454 fprintf (lra_dump_file,
2455 " %d Using memory because of"
2456 " a bad mode: reject+=2\n",
2457 nop);
2458 reject += 2;
2460 else
2462 if (lra_dump_file != NULL)
2463 fprintf (lra_dump_file,
2464 " alt=%d: Wrong mode -- refuse\n",
2465 nalt);
2466 goto fail;
2471 /* If not assigned pseudo has a class which a subset of
2472 required reg class, it is a less costly alternative
2473 as the pseudo still can get a hard reg of necessary
2474 class. */
2475 if (! no_regs_p && REG_P (op) && hard_regno[nop] < 0
2476 && (cl = get_reg_class (REGNO (op))) != NO_REGS
2477 && ira_class_subset_p[this_alternative][cl])
2479 if (lra_dump_file != NULL)
2480 fprintf
2481 (lra_dump_file,
2482 " %d Super set class reg: reject-=3\n", nop);
2483 reject -= 3;
2486 this_alternative_offmemok = offmemok;
2487 if (this_costly_alternative != NO_REGS)
2489 if (lra_dump_file != NULL)
2490 fprintf (lra_dump_file,
2491 " %d Costly loser: reject++\n", nop);
2492 reject++;
2494 /* If the operand is dying, has a matching constraint,
2495 and satisfies constraints of the matched operand
2496 which failed to satisfy the own constraints, most probably
2497 the reload for this operand will be gone. */
2498 if (this_alternative_matches >= 0
2499 && !curr_alt_win[this_alternative_matches]
2500 && REG_P (op)
2501 && find_regno_note (curr_insn, REG_DEAD, REGNO (op))
2502 && (hard_regno[nop] >= 0
2503 ? in_hard_reg_set_p (this_alternative_set,
2504 mode, hard_regno[nop])
2505 : in_class_p (op, this_alternative, NULL)))
2507 if (lra_dump_file != NULL)
2508 fprintf
2509 (lra_dump_file,
2510 " %d Dying matched operand reload: reject++\n",
2511 nop);
2512 reject++;
2514 else
2516 /* Strict_low_part requires to reload the register
2517 not the sub-register. In this case we should
2518 check that a final reload hard reg can hold the
2519 value mode. */
2520 if (curr_static_id->operand[nop].strict_low
2521 && REG_P (op)
2522 && hard_regno[nop] < 0
2523 && GET_CODE (*curr_id->operand_loc[nop]) == SUBREG
2524 && ira_class_hard_regs_num[this_alternative] > 0
2525 && (!targetm.hard_regno_mode_ok
2526 (ira_class_hard_regs[this_alternative][0],
2527 GET_MODE (*curr_id->operand_loc[nop]))))
2529 if (lra_dump_file != NULL)
2530 fprintf
2531 (lra_dump_file,
2532 " alt=%d: Strict low subreg reload -- refuse\n",
2533 nalt);
2534 goto fail;
2536 losers++;
2538 if (operand_reg[nop] != NULL_RTX
2539 /* Output operands and matched input operands are
2540 not inherited. The following conditions do not
2541 exactly describe the previous statement but they
2542 are pretty close. */
2543 && curr_static_id->operand[nop].type != OP_OUT
2544 && (this_alternative_matches < 0
2545 || curr_static_id->operand[nop].type != OP_IN))
2547 int last_reload = (lra_reg_info[ORIGINAL_REGNO
2548 (operand_reg[nop])]
2549 .last_reload);
2551 /* The value of reload_sum has sense only if we
2552 process insns in their order. It happens only on
2553 the first constraints sub-pass when we do most of
2554 reload work. */
2555 if (lra_constraint_iter == 1 && last_reload > bb_reload_num)
2556 reload_sum += last_reload - bb_reload_num;
2558 /* If this is a constant that is reloaded into the
2559 desired class by copying it to memory first, count
2560 that as another reload. This is consistent with
2561 other code and is required to avoid choosing another
2562 alternative when the constant is moved into memory.
2563 Note that the test here is precisely the same as in
2564 the code below that calls force_const_mem. */
2565 if (CONST_POOL_OK_P (mode, op)
2566 && ((targetm.preferred_reload_class
2567 (op, this_alternative) == NO_REGS)
2568 || no_input_reloads_p))
2570 const_to_mem = 1;
2571 if (! no_regs_p)
2572 losers++;
2575 /* Alternative loses if it requires a type of reload not
2576 permitted for this insn. We can always reload
2577 objects with a REG_UNUSED note. */
2578 if ((curr_static_id->operand[nop].type != OP_IN
2579 && no_output_reloads_p
2580 && ! find_reg_note (curr_insn, REG_UNUSED, op))
2581 || (curr_static_id->operand[nop].type != OP_OUT
2582 && no_input_reloads_p && ! const_to_mem)
2583 || (this_alternative_matches >= 0
2584 && (no_input_reloads_p
2585 || (no_output_reloads_p
2586 && (curr_static_id->operand
2587 [this_alternative_matches].type != OP_IN)
2588 && ! find_reg_note (curr_insn, REG_UNUSED,
2589 no_subreg_reg_operand
2590 [this_alternative_matches])))))
2592 if (lra_dump_file != NULL)
2593 fprintf
2594 (lra_dump_file,
2595 " alt=%d: No input/otput reload -- refuse\n",
2596 nalt);
2597 goto fail;
2600 /* Alternative loses if it required class pseudo can not
2601 hold value of required mode. Such insns can be
2602 described by insn definitions with mode iterators. */
2603 if (GET_MODE (*curr_id->operand_loc[nop]) != VOIDmode
2604 && ! hard_reg_set_empty_p (this_alternative_set)
2605 /* It is common practice for constraints to use a
2606 class which does not have actually enough regs to
2607 hold the value (e.g. x86 AREG for mode requiring
2608 more one general reg). Therefore we have 2
2609 conditions to check that the reload pseudo can
2610 not hold the mode value. */
2611 && (!targetm.hard_regno_mode_ok
2612 (ira_class_hard_regs[this_alternative][0],
2613 GET_MODE (*curr_id->operand_loc[nop])))
2614 /* The above condition is not enough as the first
2615 reg in ira_class_hard_regs can be not aligned for
2616 multi-words mode values. */
2617 && (prohibited_class_reg_set_mode_p
2618 (this_alternative, this_alternative_set,
2619 GET_MODE (*curr_id->operand_loc[nop]))))
2621 if (lra_dump_file != NULL)
2622 fprintf (lra_dump_file,
2623 " alt=%d: reload pseudo for op %d "
2624 " can not hold the mode value -- refuse\n",
2625 nalt, nop);
2626 goto fail;
2629 /* Check strong discouragement of reload of non-constant
2630 into class THIS_ALTERNATIVE. */
2631 if (! CONSTANT_P (op) && ! no_regs_p
2632 && (targetm.preferred_reload_class
2633 (op, this_alternative) == NO_REGS
2634 || (curr_static_id->operand[nop].type == OP_OUT
2635 && (targetm.preferred_output_reload_class
2636 (op, this_alternative) == NO_REGS))))
2638 if (lra_dump_file != NULL)
2639 fprintf (lra_dump_file,
2640 " %d Non-prefered reload: reject+=%d\n",
2641 nop, LRA_MAX_REJECT);
2642 reject += LRA_MAX_REJECT;
2645 if (! (MEM_P (op) && offmemok)
2646 && ! (const_to_mem && constmemok))
2648 /* We prefer to reload pseudos over reloading other
2649 things, since such reloads may be able to be
2650 eliminated later. So bump REJECT in other cases.
2651 Don't do this in the case where we are forcing a
2652 constant into memory and it will then win since
2653 we don't want to have a different alternative
2654 match then. */
2655 if (! (REG_P (op) && REGNO (op) >= FIRST_PSEUDO_REGISTER))
2657 if (lra_dump_file != NULL)
2658 fprintf
2659 (lra_dump_file,
2660 " %d Non-pseudo reload: reject+=2\n",
2661 nop);
2662 reject += 2;
2665 if (! no_regs_p)
2666 reload_nregs
2667 += ira_reg_class_max_nregs[this_alternative][mode];
2669 if (SMALL_REGISTER_CLASS_P (this_alternative))
2671 if (lra_dump_file != NULL)
2672 fprintf
2673 (lra_dump_file,
2674 " %d Small class reload: reject+=%d\n",
2675 nop, LRA_LOSER_COST_FACTOR / 2);
2676 reject += LRA_LOSER_COST_FACTOR / 2;
2680 /* We are trying to spill pseudo into memory. It is
2681 usually more costly than moving to a hard register
2682 although it might takes the same number of
2683 reloads.
2685 Non-pseudo spill may happen also. Suppose a target allows both
2686 register and memory in the operand constraint alternatives,
2687 then it's typical that an eliminable register has a substition
2688 of "base + offset" which can either be reloaded by a simple
2689 "new_reg <= base + offset" which will match the register
2690 constraint, or a similar reg addition followed by further spill
2691 to and reload from memory which will match the memory
2692 constraint, but this memory spill will be much more costly
2693 usually.
2695 Code below increases the reject for both pseudo and non-pseudo
2696 spill. */
2697 if (no_regs_p
2698 && !(MEM_P (op) && offmemok)
2699 && !(REG_P (op) && hard_regno[nop] < 0))
2701 if (lra_dump_file != NULL)
2702 fprintf
2703 (lra_dump_file,
2704 " %d Spill %spseudo into memory: reject+=3\n",
2705 nop, REG_P (op) ? "" : "Non-");
2706 reject += 3;
2707 if (VECTOR_MODE_P (mode))
2709 /* Spilling vectors into memory is usually more
2710 costly as they contain big values. */
2711 if (lra_dump_file != NULL)
2712 fprintf
2713 (lra_dump_file,
2714 " %d Spill vector pseudo: reject+=2\n",
2715 nop);
2716 reject += 2;
2720 /* When we use an operand requiring memory in given
2721 alternative, the insn should write *and* read the
2722 value to/from memory it is costly in comparison with
2723 an insn alternative which does not use memory
2724 (e.g. register or immediate operand). We exclude
2725 memory operand for such case as we can satisfy the
2726 memory constraints by reloading address. */
2727 if (no_regs_p && offmemok && !MEM_P (op))
2729 if (lra_dump_file != NULL)
2730 fprintf
2731 (lra_dump_file,
2732 " Using memory insn operand %d: reject+=3\n",
2733 nop);
2734 reject += 3;
2737 /* If reload requires moving value through secondary
2738 memory, it will need one more insn at least. */
2739 if (this_alternative != NO_REGS
2740 && REG_P (op) && (cl = get_reg_class (REGNO (op))) != NO_REGS
2741 && ((curr_static_id->operand[nop].type != OP_OUT
2742 && targetm.secondary_memory_needed (GET_MODE (op), cl,
2743 this_alternative))
2744 || (curr_static_id->operand[nop].type != OP_IN
2745 && (targetm.secondary_memory_needed
2746 (GET_MODE (op), this_alternative, cl)))))
2747 losers++;
2749 /* Input reloads can be inherited more often than output
2750 reloads can be removed, so penalize output
2751 reloads. */
2752 if (!REG_P (op) || curr_static_id->operand[nop].type != OP_IN)
2754 if (lra_dump_file != NULL)
2755 fprintf
2756 (lra_dump_file,
2757 " %d Non input pseudo reload: reject++\n",
2758 nop);
2759 reject++;
2762 if (MEM_P (op) && offmemok)
2763 addr_losers++;
2764 else if (curr_static_id->operand[nop].type == OP_INOUT)
2766 if (lra_dump_file != NULL)
2767 fprintf
2768 (lra_dump_file,
2769 " %d Input/Output reload: reject+=%d\n",
2770 nop, LRA_LOSER_COST_FACTOR);
2771 reject += LRA_LOSER_COST_FACTOR;
2775 if (early_clobber_p && ! scratch_p)
2777 if (lra_dump_file != NULL)
2778 fprintf (lra_dump_file,
2779 " %d Early clobber: reject++\n", nop);
2780 reject++;
2782 /* ??? We check early clobbers after processing all operands
2783 (see loop below) and there we update the costs more.
2784 Should we update the cost (may be approximately) here
2785 because of early clobber register reloads or it is a rare
2786 or non-important thing to be worth to do it. */
2787 overall = (losers * LRA_LOSER_COST_FACTOR + reject
2788 - (addr_losers == losers ? static_reject : 0));
2789 if ((best_losers == 0 || losers != 0) && best_overall < overall)
2791 if (lra_dump_file != NULL)
2792 fprintf (lra_dump_file,
2793 " alt=%d,overall=%d,losers=%d -- refuse\n",
2794 nalt, overall, losers);
2795 goto fail;
2798 if (update_and_check_small_class_inputs (nop, this_alternative))
2800 if (lra_dump_file != NULL)
2801 fprintf (lra_dump_file,
2802 " alt=%d, not enough small class regs -- refuse\n",
2803 nalt);
2804 goto fail;
2806 curr_alt[nop] = this_alternative;
2807 COPY_HARD_REG_SET (curr_alt_set[nop], this_alternative_set);
2808 curr_alt_win[nop] = this_alternative_win;
2809 curr_alt_match_win[nop] = this_alternative_match_win;
2810 curr_alt_offmemok[nop] = this_alternative_offmemok;
2811 curr_alt_matches[nop] = this_alternative_matches;
2813 if (this_alternative_matches >= 0
2814 && !did_match && !this_alternative_win)
2815 curr_alt_win[this_alternative_matches] = false;
2817 if (early_clobber_p && operand_reg[nop] != NULL_RTX)
2818 early_clobbered_nops[early_clobbered_regs_num++] = nop;
2821 if (curr_insn_set != NULL_RTX && n_operands == 2
2822 /* Prevent processing non-move insns. */
2823 && (GET_CODE (SET_SRC (curr_insn_set)) == SUBREG
2824 || SET_SRC (curr_insn_set) == no_subreg_reg_operand[1])
2825 && ((! curr_alt_win[0] && ! curr_alt_win[1]
2826 && REG_P (no_subreg_reg_operand[0])
2827 && REG_P (no_subreg_reg_operand[1])
2828 && (reg_in_class_p (no_subreg_reg_operand[0], curr_alt[1])
2829 || reg_in_class_p (no_subreg_reg_operand[1], curr_alt[0])))
2830 || (! curr_alt_win[0] && curr_alt_win[1]
2831 && REG_P (no_subreg_reg_operand[1])
2832 /* Check that we reload memory not the memory
2833 address. */
2834 && ! (curr_alt_offmemok[0]
2835 && MEM_P (no_subreg_reg_operand[0]))
2836 && reg_in_class_p (no_subreg_reg_operand[1], curr_alt[0]))
2837 || (curr_alt_win[0] && ! curr_alt_win[1]
2838 && REG_P (no_subreg_reg_operand[0])
2839 /* Check that we reload memory not the memory
2840 address. */
2841 && ! (curr_alt_offmemok[1]
2842 && MEM_P (no_subreg_reg_operand[1]))
2843 && reg_in_class_p (no_subreg_reg_operand[0], curr_alt[1])
2844 && (! CONST_POOL_OK_P (curr_operand_mode[1],
2845 no_subreg_reg_operand[1])
2846 || (targetm.preferred_reload_class
2847 (no_subreg_reg_operand[1],
2848 (enum reg_class) curr_alt[1]) != NO_REGS))
2849 /* If it is a result of recent elimination in move
2850 insn we can transform it into an add still by
2851 using this alternative. */
2852 && GET_CODE (no_subreg_reg_operand[1]) != PLUS)))
2854 /* We have a move insn and a new reload insn will be similar
2855 to the current insn. We should avoid such situation as
2856 it results in LRA cycling. */
2857 if (lra_dump_file != NULL)
2858 fprintf (lra_dump_file,
2859 " Cycle danger: overall += LRA_MAX_REJECT\n");
2860 overall += LRA_MAX_REJECT;
2862 ok_p = true;
2863 curr_alt_dont_inherit_ops_num = 0;
2864 for (nop = 0; nop < early_clobbered_regs_num; nop++)
2866 int i, j, clobbered_hard_regno, first_conflict_j, last_conflict_j;
2867 HARD_REG_SET temp_set;
2869 i = early_clobbered_nops[nop];
2870 if ((! curr_alt_win[i] && ! curr_alt_match_win[i])
2871 || hard_regno[i] < 0)
2872 continue;
2873 lra_assert (operand_reg[i] != NULL_RTX);
2874 clobbered_hard_regno = hard_regno[i];
2875 CLEAR_HARD_REG_SET (temp_set);
2876 add_to_hard_reg_set (&temp_set, biggest_mode[i], clobbered_hard_regno);
2877 first_conflict_j = last_conflict_j = -1;
2878 for (j = 0; j < n_operands; j++)
2879 if (j == i
2880 /* We don't want process insides of match_operator and
2881 match_parallel because otherwise we would process
2882 their operands once again generating a wrong
2883 code. */
2884 || curr_static_id->operand[j].is_operator)
2885 continue;
2886 else if ((curr_alt_matches[j] == i && curr_alt_match_win[j])
2887 || (curr_alt_matches[i] == j && curr_alt_match_win[i]))
2888 continue;
2889 /* If we don't reload j-th operand, check conflicts. */
2890 else if ((curr_alt_win[j] || curr_alt_match_win[j])
2891 && uses_hard_regs_p (*curr_id->operand_loc[j], temp_set))
2893 if (first_conflict_j < 0)
2894 first_conflict_j = j;
2895 last_conflict_j = j;
2897 if (last_conflict_j < 0)
2898 continue;
2899 /* If earlyclobber operand conflicts with another
2900 non-matching operand which is actually the same register
2901 as the earlyclobber operand, it is better to reload the
2902 another operand as an operand matching the earlyclobber
2903 operand can be also the same. */
2904 if (first_conflict_j == last_conflict_j
2905 && operand_reg[last_conflict_j] != NULL_RTX
2906 && ! curr_alt_match_win[last_conflict_j]
2907 && REGNO (operand_reg[i]) == REGNO (operand_reg[last_conflict_j]))
2909 curr_alt_win[last_conflict_j] = false;
2910 curr_alt_dont_inherit_ops[curr_alt_dont_inherit_ops_num++]
2911 = last_conflict_j;
2912 losers++;
2913 /* Early clobber was already reflected in REJECT. */
2914 lra_assert (reject > 0);
2915 if (lra_dump_file != NULL)
2916 fprintf
2917 (lra_dump_file,
2918 " %d Conflict early clobber reload: reject--\n",
2920 reject--;
2921 overall += LRA_LOSER_COST_FACTOR - 1;
2923 else
2925 /* We need to reload early clobbered register and the
2926 matched registers. */
2927 for (j = 0; j < n_operands; j++)
2928 if (curr_alt_matches[j] == i)
2930 curr_alt_match_win[j] = false;
2931 losers++;
2932 overall += LRA_LOSER_COST_FACTOR;
2934 if (! curr_alt_match_win[i])
2935 curr_alt_dont_inherit_ops[curr_alt_dont_inherit_ops_num++] = i;
2936 else
2938 /* Remember pseudos used for match reloads are never
2939 inherited. */
2940 lra_assert (curr_alt_matches[i] >= 0);
2941 curr_alt_win[curr_alt_matches[i]] = false;
2943 curr_alt_win[i] = curr_alt_match_win[i] = false;
2944 losers++;
2945 /* Early clobber was already reflected in REJECT. */
2946 lra_assert (reject > 0);
2947 if (lra_dump_file != NULL)
2948 fprintf
2949 (lra_dump_file,
2950 " %d Matched conflict early clobber reloads: "
2951 "reject--\n",
2953 reject--;
2954 overall += LRA_LOSER_COST_FACTOR - 1;
2957 if (lra_dump_file != NULL)
2958 fprintf (lra_dump_file, " alt=%d,overall=%d,losers=%d,rld_nregs=%d\n",
2959 nalt, overall, losers, reload_nregs);
2961 /* If this alternative can be made to work by reloading, and it
2962 needs less reloading than the others checked so far, record
2963 it as the chosen goal for reloading. */
2964 if ((best_losers != 0 && losers == 0)
2965 || (((best_losers == 0 && losers == 0)
2966 || (best_losers != 0 && losers != 0))
2967 && (best_overall > overall
2968 || (best_overall == overall
2969 /* If the cost of the reloads is the same,
2970 prefer alternative which requires minimal
2971 number of reload regs. */
2972 && (reload_nregs < best_reload_nregs
2973 || (reload_nregs == best_reload_nregs
2974 && (best_reload_sum < reload_sum
2975 || (best_reload_sum == reload_sum
2976 && nalt < goal_alt_number))))))))
2978 for (nop = 0; nop < n_operands; nop++)
2980 goal_alt_win[nop] = curr_alt_win[nop];
2981 goal_alt_match_win[nop] = curr_alt_match_win[nop];
2982 goal_alt_matches[nop] = curr_alt_matches[nop];
2983 goal_alt[nop] = curr_alt[nop];
2984 goal_alt_offmemok[nop] = curr_alt_offmemok[nop];
2986 goal_alt_dont_inherit_ops_num = curr_alt_dont_inherit_ops_num;
2987 for (nop = 0; nop < curr_alt_dont_inherit_ops_num; nop++)
2988 goal_alt_dont_inherit_ops[nop] = curr_alt_dont_inherit_ops[nop];
2989 goal_alt_swapped = curr_swapped;
2990 best_overall = overall;
2991 best_losers = losers;
2992 best_reload_nregs = reload_nregs;
2993 best_reload_sum = reload_sum;
2994 goal_alt_number = nalt;
2996 if (losers == 0)
2997 /* Everything is satisfied. Do not process alternatives
2998 anymore. */
2999 break;
3000 fail:
3003 return ok_p;
3006 /* Make reload base reg from address AD. */
3007 static rtx
3008 base_to_reg (struct address_info *ad)
3010 enum reg_class cl;
3011 int code = -1;
3012 rtx new_inner = NULL_RTX;
3013 rtx new_reg = NULL_RTX;
3014 rtx_insn *insn;
3015 rtx_insn *last_insn = get_last_insn();
3017 lra_assert (ad->disp == ad->disp_term);
3018 cl = base_reg_class (ad->mode, ad->as, ad->base_outer_code,
3019 get_index_code (ad));
3020 new_reg = lra_create_new_reg (GET_MODE (*ad->base), NULL_RTX,
3021 cl, "base");
3022 new_inner = simplify_gen_binary (PLUS, GET_MODE (new_reg), new_reg,
3023 ad->disp_term == NULL
3024 ? const0_rtx
3025 : *ad->disp_term);
3026 if (!valid_address_p (ad->mode, new_inner, ad->as))
3027 return NULL_RTX;
3028 insn = emit_insn (gen_rtx_SET (new_reg, *ad->base));
3029 code = recog_memoized (insn);
3030 if (code < 0)
3032 delete_insns_since (last_insn);
3033 return NULL_RTX;
3036 return new_inner;
3039 /* Make reload base reg + disp from address AD. Return the new pseudo. */
3040 static rtx
3041 base_plus_disp_to_reg (struct address_info *ad)
3043 enum reg_class cl;
3044 rtx new_reg;
3046 lra_assert (ad->base == ad->base_term && ad->disp == ad->disp_term);
3047 cl = base_reg_class (ad->mode, ad->as, ad->base_outer_code,
3048 get_index_code (ad));
3049 new_reg = lra_create_new_reg (GET_MODE (*ad->base_term), NULL_RTX,
3050 cl, "base + disp");
3051 lra_emit_add (new_reg, *ad->base_term, *ad->disp_term);
3052 return new_reg;
3055 /* Make reload of index part of address AD. Return the new
3056 pseudo. */
3057 static rtx
3058 index_part_to_reg (struct address_info *ad)
3060 rtx new_reg;
3062 new_reg = lra_create_new_reg (GET_MODE (*ad->index), NULL_RTX,
3063 INDEX_REG_CLASS, "index term");
3064 expand_mult (GET_MODE (*ad->index), *ad->index_term,
3065 GEN_INT (get_index_scale (ad)), new_reg, 1);
3066 return new_reg;
3069 /* Return true if we can add a displacement to address AD, even if that
3070 makes the address invalid. The fix-up code requires any new address
3071 to be the sum of the BASE_TERM, INDEX and DISP_TERM fields. */
3072 static bool
3073 can_add_disp_p (struct address_info *ad)
3075 return (!ad->autoinc_p
3076 && ad->segment == NULL
3077 && ad->base == ad->base_term
3078 && ad->disp == ad->disp_term);
3081 /* Make equiv substitution in address AD. Return true if a substitution
3082 was made. */
3083 static bool
3084 equiv_address_substitution (struct address_info *ad)
3086 rtx base_reg, new_base_reg, index_reg, new_index_reg, *base_term, *index_term;
3087 HOST_WIDE_INT disp, scale;
3088 bool change_p;
3090 base_term = strip_subreg (ad->base_term);
3091 if (base_term == NULL)
3092 base_reg = new_base_reg = NULL_RTX;
3093 else
3095 base_reg = *base_term;
3096 new_base_reg = get_equiv_with_elimination (base_reg, curr_insn);
3098 index_term = strip_subreg (ad->index_term);
3099 if (index_term == NULL)
3100 index_reg = new_index_reg = NULL_RTX;
3101 else
3103 index_reg = *index_term;
3104 new_index_reg = get_equiv_with_elimination (index_reg, curr_insn);
3106 if (base_reg == new_base_reg && index_reg == new_index_reg)
3107 return false;
3108 disp = 0;
3109 change_p = false;
3110 if (lra_dump_file != NULL)
3112 fprintf (lra_dump_file, "Changing address in insn %d ",
3113 INSN_UID (curr_insn));
3114 dump_value_slim (lra_dump_file, *ad->outer, 1);
3116 if (base_reg != new_base_reg)
3118 if (REG_P (new_base_reg))
3120 *base_term = new_base_reg;
3121 change_p = true;
3123 else if (GET_CODE (new_base_reg) == PLUS
3124 && REG_P (XEXP (new_base_reg, 0))
3125 && CONST_INT_P (XEXP (new_base_reg, 1))
3126 && can_add_disp_p (ad))
3128 disp += INTVAL (XEXP (new_base_reg, 1));
3129 *base_term = XEXP (new_base_reg, 0);
3130 change_p = true;
3132 if (ad->base_term2 != NULL)
3133 *ad->base_term2 = *ad->base_term;
3135 if (index_reg != new_index_reg)
3137 if (REG_P (new_index_reg))
3139 *index_term = new_index_reg;
3140 change_p = true;
3142 else if (GET_CODE (new_index_reg) == PLUS
3143 && REG_P (XEXP (new_index_reg, 0))
3144 && CONST_INT_P (XEXP (new_index_reg, 1))
3145 && can_add_disp_p (ad)
3146 && (scale = get_index_scale (ad)))
3148 disp += INTVAL (XEXP (new_index_reg, 1)) * scale;
3149 *index_term = XEXP (new_index_reg, 0);
3150 change_p = true;
3153 if (disp != 0)
3155 if (ad->disp != NULL)
3156 *ad->disp = plus_constant (GET_MODE (*ad->inner), *ad->disp, disp);
3157 else
3159 *ad->inner = plus_constant (GET_MODE (*ad->inner), *ad->inner, disp);
3160 update_address (ad);
3162 change_p = true;
3164 if (lra_dump_file != NULL)
3166 if (! change_p)
3167 fprintf (lra_dump_file, " -- no change\n");
3168 else
3170 fprintf (lra_dump_file, " on equiv ");
3171 dump_value_slim (lra_dump_file, *ad->outer, 1);
3172 fprintf (lra_dump_file, "\n");
3175 return change_p;
3178 /* Major function to make reloads for an address in operand NOP or
3179 check its correctness (If CHECK_ONLY_P is true). The supported
3180 cases are:
3182 1) an address that existed before LRA started, at which point it
3183 must have been valid. These addresses are subject to elimination
3184 and may have become invalid due to the elimination offset being out
3185 of range.
3187 2) an address created by forcing a constant to memory
3188 (force_const_to_mem). The initial form of these addresses might
3189 not be valid, and it is this function's job to make them valid.
3191 3) a frame address formed from a register and a (possibly zero)
3192 constant offset. As above, these addresses might not be valid and
3193 this function must make them so.
3195 Add reloads to the lists *BEFORE and *AFTER. We might need to add
3196 reloads to *AFTER because of inc/dec, {pre, post} modify in the
3197 address. Return true for any RTL change.
3199 The function is a helper function which does not produce all
3200 transformations (when CHECK_ONLY_P is false) which can be
3201 necessary. It does just basic steps. To do all necessary
3202 transformations use function process_address. */
3203 static bool
3204 process_address_1 (int nop, bool check_only_p,
3205 rtx_insn **before, rtx_insn **after)
3207 struct address_info ad;
3208 rtx new_reg;
3209 HOST_WIDE_INT scale;
3210 rtx op = *curr_id->operand_loc[nop];
3211 const char *constraint = curr_static_id->operand[nop].constraint;
3212 enum constraint_num cn = lookup_constraint (constraint);
3213 bool change_p = false;
3215 if (MEM_P (op)
3216 && GET_MODE (op) == BLKmode
3217 && GET_CODE (XEXP (op, 0)) == SCRATCH)
3218 return false;
3220 if (insn_extra_address_constraint (cn))
3221 decompose_lea_address (&ad, curr_id->operand_loc[nop]);
3222 /* Do not attempt to decompose arbitrary addresses generated by combine
3223 for asm operands with loose constraints, e.g 'X'. */
3224 else if (MEM_P (op)
3225 && !(get_constraint_type (cn) == CT_FIXED_FORM
3226 && constraint_satisfied_p (op, cn)))
3227 decompose_mem_address (&ad, op);
3228 else if (GET_CODE (op) == SUBREG
3229 && MEM_P (SUBREG_REG (op)))
3230 decompose_mem_address (&ad, SUBREG_REG (op));
3231 else
3232 return false;
3233 /* If INDEX_REG_CLASS is assigned to base_term already and isn't to
3234 index_term, swap them so to avoid assigning INDEX_REG_CLASS to both
3235 when INDEX_REG_CLASS is a single register class. */
3236 if (ad.base_term != NULL
3237 && ad.index_term != NULL
3238 && ira_class_hard_regs_num[INDEX_REG_CLASS] == 1
3239 && REG_P (*ad.base_term)
3240 && REG_P (*ad.index_term)
3241 && in_class_p (*ad.base_term, INDEX_REG_CLASS, NULL)
3242 && ! in_class_p (*ad.index_term, INDEX_REG_CLASS, NULL))
3244 std::swap (ad.base, ad.index);
3245 std::swap (ad.base_term, ad.index_term);
3247 if (! check_only_p)
3248 change_p = equiv_address_substitution (&ad);
3249 if (ad.base_term != NULL
3250 && (process_addr_reg
3251 (ad.base_term, check_only_p, before,
3252 (ad.autoinc_p
3253 && !(REG_P (*ad.base_term)
3254 && find_regno_note (curr_insn, REG_DEAD,
3255 REGNO (*ad.base_term)) != NULL_RTX)
3256 ? after : NULL),
3257 base_reg_class (ad.mode, ad.as, ad.base_outer_code,
3258 get_index_code (&ad)))))
3260 change_p = true;
3261 if (ad.base_term2 != NULL)
3262 *ad.base_term2 = *ad.base_term;
3264 if (ad.index_term != NULL
3265 && process_addr_reg (ad.index_term, check_only_p,
3266 before, NULL, INDEX_REG_CLASS))
3267 change_p = true;
3269 /* Target hooks sometimes don't treat extra-constraint addresses as
3270 legitimate address_operands, so handle them specially. */
3271 if (insn_extra_address_constraint (cn)
3272 && satisfies_address_constraint_p (&ad, cn))
3273 return change_p;
3275 if (check_only_p)
3276 return change_p;
3278 /* There are three cases where the shape of *AD.INNER may now be invalid:
3280 1) the original address was valid, but either elimination or
3281 equiv_address_substitution was applied and that made
3282 the address invalid.
3284 2) the address is an invalid symbolic address created by
3285 force_const_to_mem.
3287 3) the address is a frame address with an invalid offset.
3289 4) the address is a frame address with an invalid base.
3291 All these cases involve a non-autoinc address, so there is no
3292 point revalidating other types. */
3293 if (ad.autoinc_p || valid_address_p (&ad))
3294 return change_p;
3296 /* Any index existed before LRA started, so we can assume that the
3297 presence and shape of the index is valid. */
3298 push_to_sequence (*before);
3299 lra_assert (ad.disp == ad.disp_term);
3300 if (ad.base == NULL)
3302 if (ad.index == NULL)
3304 rtx_insn *insn;
3305 rtx_insn *last = get_last_insn ();
3306 int code = -1;
3307 enum reg_class cl = base_reg_class (ad.mode, ad.as,
3308 SCRATCH, SCRATCH);
3309 rtx addr = *ad.inner;
3311 new_reg = lra_create_new_reg (Pmode, NULL_RTX, cl, "addr");
3312 if (HAVE_lo_sum)
3314 /* addr => lo_sum (new_base, addr), case (2) above. */
3315 insn = emit_insn (gen_rtx_SET
3316 (new_reg,
3317 gen_rtx_HIGH (Pmode, copy_rtx (addr))));
3318 code = recog_memoized (insn);
3319 if (code >= 0)
3321 *ad.inner = gen_rtx_LO_SUM (Pmode, new_reg, addr);
3322 if (! valid_address_p (ad.mode, *ad.outer, ad.as))
3324 /* Try to put lo_sum into register. */
3325 insn = emit_insn (gen_rtx_SET
3326 (new_reg,
3327 gen_rtx_LO_SUM (Pmode, new_reg, addr)));
3328 code = recog_memoized (insn);
3329 if (code >= 0)
3331 *ad.inner = new_reg;
3332 if (! valid_address_p (ad.mode, *ad.outer, ad.as))
3334 *ad.inner = addr;
3335 code = -1;
3341 if (code < 0)
3342 delete_insns_since (last);
3345 if (code < 0)
3347 /* addr => new_base, case (2) above. */
3348 lra_emit_move (new_reg, addr);
3350 for (insn = last == NULL_RTX ? get_insns () : NEXT_INSN (last);
3351 insn != NULL_RTX;
3352 insn = NEXT_INSN (insn))
3353 if (recog_memoized (insn) < 0)
3354 break;
3355 if (insn != NULL_RTX)
3357 /* Do nothing if we cannot generate right insns.
3358 This is analogous to reload pass behavior. */
3359 delete_insns_since (last);
3360 end_sequence ();
3361 return false;
3363 *ad.inner = new_reg;
3366 else
3368 /* index * scale + disp => new base + index * scale,
3369 case (1) above. */
3370 enum reg_class cl = base_reg_class (ad.mode, ad.as, PLUS,
3371 GET_CODE (*ad.index));
3373 lra_assert (INDEX_REG_CLASS != NO_REGS);
3374 new_reg = lra_create_new_reg (Pmode, NULL_RTX, cl, "disp");
3375 lra_emit_move (new_reg, *ad.disp);
3376 *ad.inner = simplify_gen_binary (PLUS, GET_MODE (new_reg),
3377 new_reg, *ad.index);
3380 else if (ad.index == NULL)
3382 int regno;
3383 enum reg_class cl;
3384 rtx set;
3385 rtx_insn *insns, *last_insn;
3386 /* Try to reload base into register only if the base is invalid
3387 for the address but with valid offset, case (4) above. */
3388 start_sequence ();
3389 new_reg = base_to_reg (&ad);
3391 /* base + disp => new base, cases (1) and (3) above. */
3392 /* Another option would be to reload the displacement into an
3393 index register. However, postreload has code to optimize
3394 address reloads that have the same base and different
3395 displacements, so reloading into an index register would
3396 not necessarily be a win. */
3397 if (new_reg == NULL_RTX)
3398 new_reg = base_plus_disp_to_reg (&ad);
3399 insns = get_insns ();
3400 last_insn = get_last_insn ();
3401 /* If we generated at least two insns, try last insn source as
3402 an address. If we succeed, we generate one less insn. */
3403 if (last_insn != insns && (set = single_set (last_insn)) != NULL_RTX
3404 && GET_CODE (SET_SRC (set)) == PLUS
3405 && REG_P (XEXP (SET_SRC (set), 0))
3406 && CONSTANT_P (XEXP (SET_SRC (set), 1)))
3408 *ad.inner = SET_SRC (set);
3409 if (valid_address_p (ad.mode, *ad.outer, ad.as))
3411 *ad.base_term = XEXP (SET_SRC (set), 0);
3412 *ad.disp_term = XEXP (SET_SRC (set), 1);
3413 cl = base_reg_class (ad.mode, ad.as, ad.base_outer_code,
3414 get_index_code (&ad));
3415 regno = REGNO (*ad.base_term);
3416 if (regno >= FIRST_PSEUDO_REGISTER
3417 && cl != lra_get_allocno_class (regno))
3418 lra_change_class (regno, cl, " Change to", true);
3419 new_reg = SET_SRC (set);
3420 delete_insns_since (PREV_INSN (last_insn));
3423 /* Try if target can split displacement into legitimite new disp
3424 and offset. If it's the case, we replace the last insn with
3425 insns for base + offset => new_reg and set new_reg + new disp
3426 to *ad.inner. */
3427 last_insn = get_last_insn ();
3428 if ((set = single_set (last_insn)) != NULL_RTX
3429 && GET_CODE (SET_SRC (set)) == PLUS
3430 && REG_P (XEXP (SET_SRC (set), 0))
3431 && REGNO (XEXP (SET_SRC (set), 0)) < FIRST_PSEUDO_REGISTER
3432 && CONST_INT_P (XEXP (SET_SRC (set), 1)))
3434 rtx addend, disp = XEXP (SET_SRC (set), 1);
3435 if (targetm.legitimize_address_displacement (&disp, &addend,
3436 ad.mode))
3438 rtx_insn *new_insns;
3439 start_sequence ();
3440 lra_emit_add (new_reg, XEXP (SET_SRC (set), 0), addend);
3441 new_insns = get_insns ();
3442 end_sequence ();
3443 new_reg = gen_rtx_PLUS (Pmode, new_reg, disp);
3444 delete_insns_since (PREV_INSN (last_insn));
3445 add_insn (new_insns);
3446 insns = get_insns ();
3449 end_sequence ();
3450 emit_insn (insns);
3451 *ad.inner = new_reg;
3453 else if (ad.disp_term != NULL)
3455 /* base + scale * index + disp => new base + scale * index,
3456 case (1) above. */
3457 new_reg = base_plus_disp_to_reg (&ad);
3458 *ad.inner = simplify_gen_binary (PLUS, GET_MODE (new_reg),
3459 new_reg, *ad.index);
3461 else if ((scale = get_index_scale (&ad)) == 1)
3463 /* The last transformation to one reg will be made in
3464 curr_insn_transform function. */
3465 end_sequence ();
3466 return false;
3468 else if (scale != 0)
3470 /* base + scale * index => base + new_reg,
3471 case (1) above.
3472 Index part of address may become invalid. For example, we
3473 changed pseudo on the equivalent memory and a subreg of the
3474 pseudo onto the memory of different mode for which the scale is
3475 prohibitted. */
3476 new_reg = index_part_to_reg (&ad);
3477 *ad.inner = simplify_gen_binary (PLUS, GET_MODE (new_reg),
3478 *ad.base_term, new_reg);
3480 else
3482 enum reg_class cl = base_reg_class (ad.mode, ad.as,
3483 SCRATCH, SCRATCH);
3484 rtx addr = *ad.inner;
3486 new_reg = lra_create_new_reg (Pmode, NULL_RTX, cl, "addr");
3487 /* addr => new_base. */
3488 lra_emit_move (new_reg, addr);
3489 *ad.inner = new_reg;
3491 *before = get_insns ();
3492 end_sequence ();
3493 return true;
3496 /* If CHECK_ONLY_P is false, do address reloads until it is necessary.
3497 Use process_address_1 as a helper function. Return true for any
3498 RTL changes.
3500 If CHECK_ONLY_P is true, just check address correctness. Return
3501 false if the address correct. */
3502 static bool
3503 process_address (int nop, bool check_only_p,
3504 rtx_insn **before, rtx_insn **after)
3506 bool res = false;
3508 while (process_address_1 (nop, check_only_p, before, after))
3510 if (check_only_p)
3511 return true;
3512 res = true;
3514 return res;
3517 /* Emit insns to reload VALUE into a new register. VALUE is an
3518 auto-increment or auto-decrement RTX whose operand is a register or
3519 memory location; so reloading involves incrementing that location.
3520 IN is either identical to VALUE, or some cheaper place to reload
3521 value being incremented/decremented from.
3523 INC_AMOUNT is the number to increment or decrement by (always
3524 positive and ignored for POST_MODIFY/PRE_MODIFY).
3526 Return pseudo containing the result. */
3527 static rtx
3528 emit_inc (enum reg_class new_rclass, rtx in, rtx value, int inc_amount)
3530 /* REG or MEM to be copied and incremented. */
3531 rtx incloc = XEXP (value, 0);
3532 /* Nonzero if increment after copying. */
3533 int post = (GET_CODE (value) == POST_DEC || GET_CODE (value) == POST_INC
3534 || GET_CODE (value) == POST_MODIFY);
3535 rtx_insn *last;
3536 rtx inc;
3537 rtx_insn *add_insn;
3538 int code;
3539 rtx real_in = in == value ? incloc : in;
3540 rtx result;
3541 bool plus_p = true;
3543 if (GET_CODE (value) == PRE_MODIFY || GET_CODE (value) == POST_MODIFY)
3545 lra_assert (GET_CODE (XEXP (value, 1)) == PLUS
3546 || GET_CODE (XEXP (value, 1)) == MINUS);
3547 lra_assert (rtx_equal_p (XEXP (XEXP (value, 1), 0), XEXP (value, 0)));
3548 plus_p = GET_CODE (XEXP (value, 1)) == PLUS;
3549 inc = XEXP (XEXP (value, 1), 1);
3551 else
3553 if (GET_CODE (value) == PRE_DEC || GET_CODE (value) == POST_DEC)
3554 inc_amount = -inc_amount;
3556 inc = GEN_INT (inc_amount);
3559 if (! post && REG_P (incloc))
3560 result = incloc;
3561 else
3562 result = lra_create_new_reg (GET_MODE (value), value, new_rclass,
3563 "INC/DEC result");
3565 if (real_in != result)
3567 /* First copy the location to the result register. */
3568 lra_assert (REG_P (result));
3569 emit_insn (gen_move_insn (result, real_in));
3572 /* We suppose that there are insns to add/sub with the constant
3573 increment permitted in {PRE/POST)_{DEC/INC/MODIFY}. At least the
3574 old reload worked with this assumption. If the assumption
3575 becomes wrong, we should use approach in function
3576 base_plus_disp_to_reg. */
3577 if (in == value)
3579 /* See if we can directly increment INCLOC. */
3580 last = get_last_insn ();
3581 add_insn = emit_insn (plus_p
3582 ? gen_add2_insn (incloc, inc)
3583 : gen_sub2_insn (incloc, inc));
3585 code = recog_memoized (add_insn);
3586 if (code >= 0)
3588 if (! post && result != incloc)
3589 emit_insn (gen_move_insn (result, incloc));
3590 return result;
3592 delete_insns_since (last);
3595 /* If couldn't do the increment directly, must increment in RESULT.
3596 The way we do this depends on whether this is pre- or
3597 post-increment. For pre-increment, copy INCLOC to the reload
3598 register, increment it there, then save back. */
3599 if (! post)
3601 if (real_in != result)
3602 emit_insn (gen_move_insn (result, real_in));
3603 if (plus_p)
3604 emit_insn (gen_add2_insn (result, inc));
3605 else
3606 emit_insn (gen_sub2_insn (result, inc));
3607 if (result != incloc)
3608 emit_insn (gen_move_insn (incloc, result));
3610 else
3612 /* Post-increment.
3614 Because this might be a jump insn or a compare, and because
3615 RESULT may not be available after the insn in an input
3616 reload, we must do the incrementing before the insn being
3617 reloaded for.
3619 We have already copied IN to RESULT. Increment the copy in
3620 RESULT, save that back, then decrement RESULT so it has
3621 the original value. */
3622 if (plus_p)
3623 emit_insn (gen_add2_insn (result, inc));
3624 else
3625 emit_insn (gen_sub2_insn (result, inc));
3626 emit_insn (gen_move_insn (incloc, result));
3627 /* Restore non-modified value for the result. We prefer this
3628 way because it does not require an additional hard
3629 register. */
3630 if (plus_p)
3632 if (CONST_INT_P (inc))
3633 emit_insn (gen_add2_insn (result,
3634 gen_int_mode (-INTVAL (inc),
3635 GET_MODE (result))));
3636 else
3637 emit_insn (gen_sub2_insn (result, inc));
3639 else
3640 emit_insn (gen_add2_insn (result, inc));
3642 return result;
3645 /* Return true if the current move insn does not need processing as we
3646 already know that it satisfies its constraints. */
3647 static bool
3648 simple_move_p (void)
3650 rtx dest, src;
3651 enum reg_class dclass, sclass;
3653 lra_assert (curr_insn_set != NULL_RTX);
3654 dest = SET_DEST (curr_insn_set);
3655 src = SET_SRC (curr_insn_set);
3657 /* If the instruction has multiple sets we need to process it even if it
3658 is single_set. This can happen if one or more of the SETs are dead.
3659 See PR73650. */
3660 if (multiple_sets (curr_insn))
3661 return false;
3663 return ((dclass = get_op_class (dest)) != NO_REGS
3664 && (sclass = get_op_class (src)) != NO_REGS
3665 /* The backend guarantees that register moves of cost 2
3666 never need reloads. */
3667 && targetm.register_move_cost (GET_MODE (src), sclass, dclass) == 2);
3670 /* Swap operands NOP and NOP + 1. */
3671 static inline void
3672 swap_operands (int nop)
3674 std::swap (curr_operand_mode[nop], curr_operand_mode[nop + 1]);
3675 std::swap (original_subreg_reg_mode[nop], original_subreg_reg_mode[nop + 1]);
3676 std::swap (*curr_id->operand_loc[nop], *curr_id->operand_loc[nop + 1]);
3677 std::swap (equiv_substition_p[nop], equiv_substition_p[nop + 1]);
3678 /* Swap the duplicates too. */
3679 lra_update_dup (curr_id, nop);
3680 lra_update_dup (curr_id, nop + 1);
3683 /* Main entry point of the constraint code: search the body of the
3684 current insn to choose the best alternative. It is mimicking insn
3685 alternative cost calculation model of former reload pass. That is
3686 because machine descriptions were written to use this model. This
3687 model can be changed in future. Make commutative operand exchange
3688 if it is chosen.
3690 if CHECK_ONLY_P is false, do RTL changes to satisfy the
3691 constraints. Return true if any change happened during function
3692 call.
3694 If CHECK_ONLY_P is true then don't do any transformation. Just
3695 check that the insn satisfies all constraints. If the insn does
3696 not satisfy any constraint, return true. */
3697 static bool
3698 curr_insn_transform (bool check_only_p)
3700 int i, j, k;
3701 int n_operands;
3702 int n_alternatives;
3703 int n_outputs;
3704 int commutative;
3705 signed char goal_alt_matched[MAX_RECOG_OPERANDS][MAX_RECOG_OPERANDS];
3706 signed char match_inputs[MAX_RECOG_OPERANDS + 1];
3707 signed char outputs[MAX_RECOG_OPERANDS + 1];
3708 rtx_insn *before, *after;
3709 bool alt_p = false;
3710 /* Flag that the insn has been changed through a transformation. */
3711 bool change_p;
3712 bool sec_mem_p;
3713 bool use_sec_mem_p;
3714 int max_regno_before;
3715 int reused_alternative_num;
3717 curr_insn_set = single_set (curr_insn);
3718 if (curr_insn_set != NULL_RTX && simple_move_p ())
3719 return false;
3721 no_input_reloads_p = no_output_reloads_p = false;
3722 goal_alt_number = -1;
3723 change_p = sec_mem_p = false;
3724 /* JUMP_INSNs and CALL_INSNs are not allowed to have any output
3725 reloads; neither are insns that SET cc0. Insns that use CC0 are
3726 not allowed to have any input reloads. */
3727 if (JUMP_P (curr_insn) || CALL_P (curr_insn))
3728 no_output_reloads_p = true;
3730 if (HAVE_cc0 && reg_referenced_p (cc0_rtx, PATTERN (curr_insn)))
3731 no_input_reloads_p = true;
3732 if (HAVE_cc0 && reg_set_p (cc0_rtx, PATTERN (curr_insn)))
3733 no_output_reloads_p = true;
3735 n_operands = curr_static_id->n_operands;
3736 n_alternatives = curr_static_id->n_alternatives;
3738 /* Just return "no reloads" if insn has no operands with
3739 constraints. */
3740 if (n_operands == 0 || n_alternatives == 0)
3741 return false;
3743 max_regno_before = max_reg_num ();
3745 for (i = 0; i < n_operands; i++)
3747 goal_alt_matched[i][0] = -1;
3748 goal_alt_matches[i] = -1;
3751 commutative = curr_static_id->commutative;
3753 /* Now see what we need for pseudos that didn't get hard regs or got
3754 the wrong kind of hard reg. For this, we must consider all the
3755 operands together against the register constraints. */
3757 best_losers = best_overall = INT_MAX;
3758 best_reload_sum = 0;
3760 curr_swapped = false;
3761 goal_alt_swapped = false;
3763 if (! check_only_p)
3764 /* Make equivalence substitution and memory subreg elimination
3765 before address processing because an address legitimacy can
3766 depend on memory mode. */
3767 for (i = 0; i < n_operands; i++)
3769 rtx op, subst, old;
3770 bool op_change_p = false;
3772 if (curr_static_id->operand[i].is_operator)
3773 continue;
3775 old = op = *curr_id->operand_loc[i];
3776 if (GET_CODE (old) == SUBREG)
3777 old = SUBREG_REG (old);
3778 subst = get_equiv_with_elimination (old, curr_insn);
3779 original_subreg_reg_mode[i] = VOIDmode;
3780 equiv_substition_p[i] = false;
3781 if (subst != old)
3783 equiv_substition_p[i] = true;
3784 subst = copy_rtx (subst);
3785 lra_assert (REG_P (old));
3786 if (GET_CODE (op) != SUBREG)
3787 *curr_id->operand_loc[i] = subst;
3788 else
3790 SUBREG_REG (op) = subst;
3791 if (GET_MODE (subst) == VOIDmode)
3792 original_subreg_reg_mode[i] = GET_MODE (old);
3794 if (lra_dump_file != NULL)
3796 fprintf (lra_dump_file,
3797 "Changing pseudo %d in operand %i of insn %u on equiv ",
3798 REGNO (old), i, INSN_UID (curr_insn));
3799 dump_value_slim (lra_dump_file, subst, 1);
3800 fprintf (lra_dump_file, "\n");
3802 op_change_p = change_p = true;
3804 if (simplify_operand_subreg (i, GET_MODE (old)) || op_change_p)
3806 change_p = true;
3807 lra_update_dup (curr_id, i);
3811 /* Reload address registers and displacements. We do it before
3812 finding an alternative because of memory constraints. */
3813 before = after = NULL;
3814 for (i = 0; i < n_operands; i++)
3815 if (! curr_static_id->operand[i].is_operator
3816 && process_address (i, check_only_p, &before, &after))
3818 if (check_only_p)
3819 return true;
3820 change_p = true;
3821 lra_update_dup (curr_id, i);
3824 if (change_p)
3825 /* If we've changed the instruction then any alternative that
3826 we chose previously may no longer be valid. */
3827 lra_set_used_insn_alternative (curr_insn, -1);
3829 if (! check_only_p && curr_insn_set != NULL_RTX
3830 && check_and_process_move (&change_p, &sec_mem_p))
3831 return change_p;
3833 try_swapped:
3835 reused_alternative_num = check_only_p ? -1 : curr_id->used_insn_alternative;
3836 if (lra_dump_file != NULL && reused_alternative_num >= 0)
3837 fprintf (lra_dump_file, "Reusing alternative %d for insn #%u\n",
3838 reused_alternative_num, INSN_UID (curr_insn));
3840 if (process_alt_operands (reused_alternative_num))
3841 alt_p = true;
3843 if (check_only_p)
3844 return ! alt_p || best_losers != 0;
3846 /* If insn is commutative (it's safe to exchange a certain pair of
3847 operands) then we need to try each alternative twice, the second
3848 time matching those two operands as if we had exchanged them. To
3849 do this, really exchange them in operands.
3851 If we have just tried the alternatives the second time, return
3852 operands to normal and drop through. */
3854 if (reused_alternative_num < 0 && commutative >= 0)
3856 curr_swapped = !curr_swapped;
3857 if (curr_swapped)
3859 swap_operands (commutative);
3860 goto try_swapped;
3862 else
3863 swap_operands (commutative);
3866 if (! alt_p && ! sec_mem_p)
3868 /* No alternative works with reloads?? */
3869 if (INSN_CODE (curr_insn) >= 0)
3870 fatal_insn ("unable to generate reloads for:", curr_insn);
3871 error_for_asm (curr_insn,
3872 "inconsistent operand constraints in an %<asm%>");
3873 /* Avoid further trouble with this insn. Don't generate use
3874 pattern here as we could use the insn SP offset. */
3875 lra_set_insn_deleted (curr_insn);
3876 return true;
3879 /* If the best alternative is with operands 1 and 2 swapped, swap
3880 them. Update the operand numbers of any reloads already
3881 pushed. */
3883 if (goal_alt_swapped)
3885 if (lra_dump_file != NULL)
3886 fprintf (lra_dump_file, " Commutative operand exchange in insn %u\n",
3887 INSN_UID (curr_insn));
3889 /* Swap the duplicates too. */
3890 swap_operands (commutative);
3891 change_p = true;
3894 /* Some targets' TARGET_SECONDARY_MEMORY_NEEDED (e.g. x86) are defined
3895 too conservatively. So we use the secondary memory only if there
3896 is no any alternative without reloads. */
3897 use_sec_mem_p = false;
3898 if (! alt_p)
3899 use_sec_mem_p = true;
3900 else if (sec_mem_p)
3902 for (i = 0; i < n_operands; i++)
3903 if (! goal_alt_win[i] && ! goal_alt_match_win[i])
3904 break;
3905 use_sec_mem_p = i < n_operands;
3908 if (use_sec_mem_p)
3910 int in = -1, out = -1;
3911 rtx new_reg, src, dest, rld;
3912 machine_mode sec_mode, rld_mode;
3914 lra_assert (curr_insn_set != NULL_RTX && sec_mem_p);
3915 dest = SET_DEST (curr_insn_set);
3916 src = SET_SRC (curr_insn_set);
3917 for (i = 0; i < n_operands; i++)
3918 if (*curr_id->operand_loc[i] == dest)
3919 out = i;
3920 else if (*curr_id->operand_loc[i] == src)
3921 in = i;
3922 for (i = 0; i < curr_static_id->n_dups; i++)
3923 if (out < 0 && *curr_id->dup_loc[i] == dest)
3924 out = curr_static_id->dup_num[i];
3925 else if (in < 0 && *curr_id->dup_loc[i] == src)
3926 in = curr_static_id->dup_num[i];
3927 lra_assert (out >= 0 && in >= 0
3928 && curr_static_id->operand[out].type == OP_OUT
3929 && curr_static_id->operand[in].type == OP_IN);
3930 rld = partial_subreg_p (GET_MODE (src), GET_MODE (dest)) ? src : dest;
3931 rld_mode = GET_MODE (rld);
3932 sec_mode = targetm.secondary_memory_needed_mode (rld_mode);
3933 new_reg = lra_create_new_reg (sec_mode, NULL_RTX,
3934 NO_REGS, "secondary");
3935 /* If the mode is changed, it should be wider. */
3936 lra_assert (!partial_subreg_p (sec_mode, rld_mode));
3937 if (sec_mode != rld_mode)
3939 /* If the target says specifically to use another mode for
3940 secondary memory moves we can not reuse the original
3941 insn. */
3942 after = emit_spill_move (false, new_reg, dest);
3943 lra_process_new_insns (curr_insn, NULL, after,
3944 "Inserting the sec. move");
3945 /* We may have non null BEFORE here (e.g. after address
3946 processing. */
3947 push_to_sequence (before);
3948 before = emit_spill_move (true, new_reg, src);
3949 emit_insn (before);
3950 before = get_insns ();
3951 end_sequence ();
3952 lra_process_new_insns (curr_insn, before, NULL, "Changing on");
3953 lra_set_insn_deleted (curr_insn);
3955 else if (dest == rld)
3957 *curr_id->operand_loc[out] = new_reg;
3958 lra_update_dup (curr_id, out);
3959 after = emit_spill_move (false, new_reg, dest);
3960 lra_process_new_insns (curr_insn, NULL, after,
3961 "Inserting the sec. move");
3963 else
3965 *curr_id->operand_loc[in] = new_reg;
3966 lra_update_dup (curr_id, in);
3967 /* See comments above. */
3968 push_to_sequence (before);
3969 before = emit_spill_move (true, new_reg, src);
3970 emit_insn (before);
3971 before = get_insns ();
3972 end_sequence ();
3973 lra_process_new_insns (curr_insn, before, NULL,
3974 "Inserting the sec. move");
3976 lra_update_insn_regno_info (curr_insn);
3977 return true;
3980 lra_assert (goal_alt_number >= 0);
3981 lra_set_used_insn_alternative (curr_insn, goal_alt_number);
3983 if (lra_dump_file != NULL)
3985 const char *p;
3987 fprintf (lra_dump_file, " Choosing alt %d in insn %u:",
3988 goal_alt_number, INSN_UID (curr_insn));
3989 for (i = 0; i < n_operands; i++)
3991 p = (curr_static_id->operand_alternative
3992 [goal_alt_number * n_operands + i].constraint);
3993 if (*p == '\0')
3994 continue;
3995 fprintf (lra_dump_file, " (%d) ", i);
3996 for (; *p != '\0' && *p != ',' && *p != '#'; p++)
3997 fputc (*p, lra_dump_file);
3999 if (INSN_CODE (curr_insn) >= 0
4000 && (p = get_insn_name (INSN_CODE (curr_insn))) != NULL)
4001 fprintf (lra_dump_file, " {%s}", p);
4002 if (curr_id->sp_offset != 0)
4003 fprintf (lra_dump_file, " (sp_off=%" HOST_WIDE_INT_PRINT "d)",
4004 curr_id->sp_offset);
4005 fprintf (lra_dump_file, "\n");
4008 /* Right now, for any pair of operands I and J that are required to
4009 match, with J < I, goal_alt_matches[I] is J. Add I to
4010 goal_alt_matched[J]. */
4012 for (i = 0; i < n_operands; i++)
4013 if ((j = goal_alt_matches[i]) >= 0)
4015 for (k = 0; goal_alt_matched[j][k] >= 0; k++)
4017 /* We allow matching one output operand and several input
4018 operands. */
4019 lra_assert (k == 0
4020 || (curr_static_id->operand[j].type == OP_OUT
4021 && curr_static_id->operand[i].type == OP_IN
4022 && (curr_static_id->operand
4023 [goal_alt_matched[j][0]].type == OP_IN)));
4024 goal_alt_matched[j][k] = i;
4025 goal_alt_matched[j][k + 1] = -1;
4028 for (i = 0; i < n_operands; i++)
4029 goal_alt_win[i] |= goal_alt_match_win[i];
4031 /* Any constants that aren't allowed and can't be reloaded into
4032 registers are here changed into memory references. */
4033 for (i = 0; i < n_operands; i++)
4034 if (goal_alt_win[i])
4036 int regno;
4037 enum reg_class new_class;
4038 rtx reg = *curr_id->operand_loc[i];
4040 if (GET_CODE (reg) == SUBREG)
4041 reg = SUBREG_REG (reg);
4043 if (REG_P (reg) && (regno = REGNO (reg)) >= FIRST_PSEUDO_REGISTER)
4045 bool ok_p = in_class_p (reg, goal_alt[i], &new_class);
4047 if (new_class != NO_REGS && get_reg_class (regno) != new_class)
4049 lra_assert (ok_p);
4050 lra_change_class (regno, new_class, " Change to", true);
4054 else
4056 const char *constraint;
4057 char c;
4058 rtx op = *curr_id->operand_loc[i];
4059 rtx subreg = NULL_RTX;
4060 machine_mode mode = curr_operand_mode[i];
4062 if (GET_CODE (op) == SUBREG)
4064 subreg = op;
4065 op = SUBREG_REG (op);
4066 mode = GET_MODE (op);
4069 if (CONST_POOL_OK_P (mode, op)
4070 && ((targetm.preferred_reload_class
4071 (op, (enum reg_class) goal_alt[i]) == NO_REGS)
4072 || no_input_reloads_p))
4074 rtx tem = force_const_mem (mode, op);
4076 change_p = true;
4077 if (subreg != NULL_RTX)
4078 tem = gen_rtx_SUBREG (mode, tem, SUBREG_BYTE (subreg));
4080 *curr_id->operand_loc[i] = tem;
4081 lra_update_dup (curr_id, i);
4082 process_address (i, false, &before, &after);
4084 /* If the alternative accepts constant pool refs directly
4085 there will be no reload needed at all. */
4086 if (subreg != NULL_RTX)
4087 continue;
4088 /* Skip alternatives before the one requested. */
4089 constraint = (curr_static_id->operand_alternative
4090 [goal_alt_number * n_operands + i].constraint);
4091 for (;
4092 (c = *constraint) && c != ',' && c != '#';
4093 constraint += CONSTRAINT_LEN (c, constraint))
4095 enum constraint_num cn = lookup_constraint (constraint);
4096 if ((insn_extra_memory_constraint (cn)
4097 || insn_extra_special_memory_constraint (cn))
4098 && satisfies_memory_constraint_p (tem, cn))
4099 break;
4101 if (c == '\0' || c == ',' || c == '#')
4102 continue;
4104 goal_alt_win[i] = true;
4108 n_outputs = 0;
4109 outputs[0] = -1;
4110 for (i = 0; i < n_operands; i++)
4112 int regno;
4113 bool optional_p = false;
4114 rtx old, new_reg;
4115 rtx op = *curr_id->operand_loc[i];
4117 if (goal_alt_win[i])
4119 if (goal_alt[i] == NO_REGS
4120 && REG_P (op)
4121 /* When we assign NO_REGS it means that we will not
4122 assign a hard register to the scratch pseudo by
4123 assigment pass and the scratch pseudo will be
4124 spilled. Spilled scratch pseudos are transformed
4125 back to scratches at the LRA end. */
4126 && lra_former_scratch_operand_p (curr_insn, i)
4127 && lra_former_scratch_p (REGNO (op)))
4129 int regno = REGNO (op);
4130 lra_change_class (regno, NO_REGS, " Change to", true);
4131 if (lra_get_regno_hard_regno (regno) >= 0)
4132 /* We don't have to mark all insn affected by the
4133 spilled pseudo as there is only one such insn, the
4134 current one. */
4135 reg_renumber[regno] = -1;
4136 lra_assert (bitmap_single_bit_set_p
4137 (&lra_reg_info[REGNO (op)].insn_bitmap));
4139 /* We can do an optional reload. If the pseudo got a hard
4140 reg, we might improve the code through inheritance. If
4141 it does not get a hard register we coalesce memory/memory
4142 moves later. Ignore move insns to avoid cycling. */
4143 if (! lra_simple_p
4144 && lra_undo_inheritance_iter < LRA_MAX_INHERITANCE_PASSES
4145 && goal_alt[i] != NO_REGS && REG_P (op)
4146 && (regno = REGNO (op)) >= FIRST_PSEUDO_REGISTER
4147 && regno < new_regno_start
4148 && ! lra_former_scratch_p (regno)
4149 && reg_renumber[regno] < 0
4150 /* Check that the optional reload pseudo will be able to
4151 hold given mode value. */
4152 && ! (prohibited_class_reg_set_mode_p
4153 (goal_alt[i], reg_class_contents[goal_alt[i]],
4154 PSEUDO_REGNO_MODE (regno)))
4155 && (curr_insn_set == NULL_RTX
4156 || !((REG_P (SET_SRC (curr_insn_set))
4157 || MEM_P (SET_SRC (curr_insn_set))
4158 || GET_CODE (SET_SRC (curr_insn_set)) == SUBREG)
4159 && (REG_P (SET_DEST (curr_insn_set))
4160 || MEM_P (SET_DEST (curr_insn_set))
4161 || GET_CODE (SET_DEST (curr_insn_set)) == SUBREG))))
4162 optional_p = true;
4163 else
4164 continue;
4167 /* Operands that match previous ones have already been handled. */
4168 if (goal_alt_matches[i] >= 0)
4169 continue;
4171 /* We should not have an operand with a non-offsettable address
4172 appearing where an offsettable address will do. It also may
4173 be a case when the address should be special in other words
4174 not a general one (e.g. it needs no index reg). */
4175 if (goal_alt_matched[i][0] == -1 && goal_alt_offmemok[i] && MEM_P (op))
4177 enum reg_class rclass;
4178 rtx *loc = &XEXP (op, 0);
4179 enum rtx_code code = GET_CODE (*loc);
4181 push_to_sequence (before);
4182 rclass = base_reg_class (GET_MODE (op), MEM_ADDR_SPACE (op),
4183 MEM, SCRATCH);
4184 if (GET_RTX_CLASS (code) == RTX_AUTOINC)
4185 new_reg = emit_inc (rclass, *loc, *loc,
4186 /* This value does not matter for MODIFY. */
4187 GET_MODE_SIZE (GET_MODE (op)));
4188 else if (get_reload_reg (OP_IN, Pmode, *loc, rclass, FALSE,
4189 "offsetable address", &new_reg))
4190 lra_emit_move (new_reg, *loc);
4191 before = get_insns ();
4192 end_sequence ();
4193 *loc = new_reg;
4194 lra_update_dup (curr_id, i);
4196 else if (goal_alt_matched[i][0] == -1)
4198 machine_mode mode;
4199 rtx reg, *loc;
4200 int hard_regno, byte;
4201 enum op_type type = curr_static_id->operand[i].type;
4203 loc = curr_id->operand_loc[i];
4204 mode = curr_operand_mode[i];
4205 if (GET_CODE (*loc) == SUBREG)
4207 reg = SUBREG_REG (*loc);
4208 byte = SUBREG_BYTE (*loc);
4209 if (REG_P (reg)
4210 /* Strict_low_part requires reloading the register and not
4211 just the subreg. Likewise for a strict subreg no wider
4212 than a word for WORD_REGISTER_OPERATIONS targets. */
4213 && (curr_static_id->operand[i].strict_low
4214 || (!paradoxical_subreg_p (mode, GET_MODE (reg))
4215 && (hard_regno
4216 = get_try_hard_regno (REGNO (reg))) >= 0
4217 && (simplify_subreg_regno
4218 (hard_regno,
4219 GET_MODE (reg), byte, mode) < 0)
4220 && (goal_alt[i] == NO_REGS
4221 || (simplify_subreg_regno
4222 (ira_class_hard_regs[goal_alt[i]][0],
4223 GET_MODE (reg), byte, mode) >= 0)))
4224 || (GET_MODE_PRECISION (mode)
4225 < GET_MODE_PRECISION (GET_MODE (reg))
4226 && GET_MODE_SIZE (GET_MODE (reg)) <= UNITS_PER_WORD
4227 && WORD_REGISTER_OPERATIONS)))
4229 /* An OP_INOUT is required when reloading a subreg of a
4230 mode wider than a word to ensure that data beyond the
4231 word being reloaded is preserved. Also automatically
4232 ensure that strict_low_part reloads are made into
4233 OP_INOUT which should already be true from the backend
4234 constraints. */
4235 if (type == OP_OUT
4236 && (curr_static_id->operand[i].strict_low
4237 || read_modify_subreg_p (*loc)))
4238 type = OP_INOUT;
4239 loc = &SUBREG_REG (*loc);
4240 mode = GET_MODE (*loc);
4243 old = *loc;
4244 if (get_reload_reg (type, mode, old, goal_alt[i],
4245 loc != curr_id->operand_loc[i], "", &new_reg)
4246 && type != OP_OUT)
4248 push_to_sequence (before);
4249 lra_emit_move (new_reg, old);
4250 before = get_insns ();
4251 end_sequence ();
4253 *loc = new_reg;
4254 if (type != OP_IN
4255 && find_reg_note (curr_insn, REG_UNUSED, old) == NULL_RTX)
4257 start_sequence ();
4258 lra_emit_move (type == OP_INOUT ? copy_rtx (old) : old, new_reg);
4259 emit_insn (after);
4260 after = get_insns ();
4261 end_sequence ();
4262 *loc = new_reg;
4264 for (j = 0; j < goal_alt_dont_inherit_ops_num; j++)
4265 if (goal_alt_dont_inherit_ops[j] == i)
4267 lra_set_regno_unique_value (REGNO (new_reg));
4268 break;
4270 lra_update_dup (curr_id, i);
4272 else if (curr_static_id->operand[i].type == OP_IN
4273 && (curr_static_id->operand[goal_alt_matched[i][0]].type
4274 == OP_OUT
4275 || (curr_static_id->operand[goal_alt_matched[i][0]].type
4276 == OP_INOUT
4277 && (operands_match_p
4278 (*curr_id->operand_loc[i],
4279 *curr_id->operand_loc[goal_alt_matched[i][0]],
4280 -1)))))
4282 /* generate reloads for input and matched outputs. */
4283 match_inputs[0] = i;
4284 match_inputs[1] = -1;
4285 match_reload (goal_alt_matched[i][0], match_inputs, outputs,
4286 goal_alt[i], &before, &after,
4287 curr_static_id->operand_alternative
4288 [goal_alt_number * n_operands + goal_alt_matched[i][0]]
4289 .earlyclobber);
4291 else if ((curr_static_id->operand[i].type == OP_OUT
4292 || (curr_static_id->operand[i].type == OP_INOUT
4293 && (operands_match_p
4294 (*curr_id->operand_loc[i],
4295 *curr_id->operand_loc[goal_alt_matched[i][0]],
4296 -1))))
4297 && (curr_static_id->operand[goal_alt_matched[i][0]].type
4298 == OP_IN))
4299 /* Generate reloads for output and matched inputs. */
4300 match_reload (i, goal_alt_matched[i], outputs, goal_alt[i], &before,
4301 &after, curr_static_id->operand_alternative
4302 [goal_alt_number * n_operands + i].earlyclobber);
4303 else if (curr_static_id->operand[i].type == OP_IN
4304 && (curr_static_id->operand[goal_alt_matched[i][0]].type
4305 == OP_IN))
4307 /* Generate reloads for matched inputs. */
4308 match_inputs[0] = i;
4309 for (j = 0; (k = goal_alt_matched[i][j]) >= 0; j++)
4310 match_inputs[j + 1] = k;
4311 match_inputs[j + 1] = -1;
4312 match_reload (-1, match_inputs, outputs, goal_alt[i], &before,
4313 &after, false);
4315 else
4316 /* We must generate code in any case when function
4317 process_alt_operands decides that it is possible. */
4318 gcc_unreachable ();
4320 /* Memorise processed outputs so that output remaining to be processed
4321 can avoid using the same register value (see match_reload). */
4322 if (curr_static_id->operand[i].type == OP_OUT)
4324 outputs[n_outputs++] = i;
4325 outputs[n_outputs] = -1;
4328 if (optional_p)
4330 rtx reg = op;
4332 lra_assert (REG_P (reg));
4333 regno = REGNO (reg);
4334 op = *curr_id->operand_loc[i]; /* Substitution. */
4335 if (GET_CODE (op) == SUBREG)
4336 op = SUBREG_REG (op);
4337 gcc_assert (REG_P (op) && (int) REGNO (op) >= new_regno_start);
4338 bitmap_set_bit (&lra_optional_reload_pseudos, REGNO (op));
4339 lra_reg_info[REGNO (op)].restore_rtx = reg;
4340 if (lra_dump_file != NULL)
4341 fprintf (lra_dump_file,
4342 " Making reload reg %d for reg %d optional\n",
4343 REGNO (op), regno);
4346 if (before != NULL_RTX || after != NULL_RTX
4347 || max_regno_before != max_reg_num ())
4348 change_p = true;
4349 if (change_p)
4351 lra_update_operator_dups (curr_id);
4352 /* Something changes -- process the insn. */
4353 lra_update_insn_regno_info (curr_insn);
4355 lra_process_new_insns (curr_insn, before, after, "Inserting insn reload");
4356 return change_p;
4359 /* Return true if INSN satisfies all constraints. In other words, no
4360 reload insns are needed. */
4361 bool
4362 lra_constrain_insn (rtx_insn *insn)
4364 int saved_new_regno_start = new_regno_start;
4365 int saved_new_insn_uid_start = new_insn_uid_start;
4366 bool change_p;
4368 curr_insn = insn;
4369 curr_id = lra_get_insn_recog_data (curr_insn);
4370 curr_static_id = curr_id->insn_static_data;
4371 new_insn_uid_start = get_max_uid ();
4372 new_regno_start = max_reg_num ();
4373 change_p = curr_insn_transform (true);
4374 new_regno_start = saved_new_regno_start;
4375 new_insn_uid_start = saved_new_insn_uid_start;
4376 return ! change_p;
4379 /* Return true if X is in LIST. */
4380 static bool
4381 in_list_p (rtx x, rtx list)
4383 for (; list != NULL_RTX; list = XEXP (list, 1))
4384 if (XEXP (list, 0) == x)
4385 return true;
4386 return false;
4389 /* Return true if X contains an allocatable hard register (if
4390 HARD_REG_P) or a (spilled if SPILLED_P) pseudo. */
4391 static bool
4392 contains_reg_p (rtx x, bool hard_reg_p, bool spilled_p)
4394 int i, j;
4395 const char *fmt;
4396 enum rtx_code code;
4398 code = GET_CODE (x);
4399 if (REG_P (x))
4401 int regno = REGNO (x);
4402 HARD_REG_SET alloc_regs;
4404 if (hard_reg_p)
4406 if (regno >= FIRST_PSEUDO_REGISTER)
4407 regno = lra_get_regno_hard_regno (regno);
4408 if (regno < 0)
4409 return false;
4410 COMPL_HARD_REG_SET (alloc_regs, lra_no_alloc_regs);
4411 return overlaps_hard_reg_set_p (alloc_regs, GET_MODE (x), regno);
4413 else
4415 if (regno < FIRST_PSEUDO_REGISTER)
4416 return false;
4417 if (! spilled_p)
4418 return true;
4419 return lra_get_regno_hard_regno (regno) < 0;
4422 fmt = GET_RTX_FORMAT (code);
4423 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
4425 if (fmt[i] == 'e')
4427 if (contains_reg_p (XEXP (x, i), hard_reg_p, spilled_p))
4428 return true;
4430 else if (fmt[i] == 'E')
4432 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
4433 if (contains_reg_p (XVECEXP (x, i, j), hard_reg_p, spilled_p))
4434 return true;
4437 return false;
4440 /* Process all regs in location *LOC and change them on equivalent
4441 substitution. Return true if any change was done. */
4442 static bool
4443 loc_equivalence_change_p (rtx *loc)
4445 rtx subst, reg, x = *loc;
4446 bool result = false;
4447 enum rtx_code code = GET_CODE (x);
4448 const char *fmt;
4449 int i, j;
4451 if (code == SUBREG)
4453 reg = SUBREG_REG (x);
4454 if ((subst = get_equiv_with_elimination (reg, curr_insn)) != reg
4455 && GET_MODE (subst) == VOIDmode)
4457 /* We cannot reload debug location. Simplify subreg here
4458 while we know the inner mode. */
4459 *loc = simplify_gen_subreg (GET_MODE (x), subst,
4460 GET_MODE (reg), SUBREG_BYTE (x));
4461 return true;
4464 if (code == REG && (subst = get_equiv_with_elimination (x, curr_insn)) != x)
4466 *loc = subst;
4467 return true;
4470 /* Scan all the operand sub-expressions. */
4471 fmt = GET_RTX_FORMAT (code);
4472 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
4474 if (fmt[i] == 'e')
4475 result = loc_equivalence_change_p (&XEXP (x, i)) || result;
4476 else if (fmt[i] == 'E')
4477 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
4478 result
4479 = loc_equivalence_change_p (&XVECEXP (x, i, j)) || result;
4481 return result;
4484 /* Similar to loc_equivalence_change_p, but for use as
4485 simplify_replace_fn_rtx callback. DATA is insn for which the
4486 elimination is done. If it null we don't do the elimination. */
4487 static rtx
4488 loc_equivalence_callback (rtx loc, const_rtx, void *data)
4490 if (!REG_P (loc))
4491 return NULL_RTX;
4493 rtx subst = (data == NULL
4494 ? get_equiv (loc) : get_equiv_with_elimination (loc, (rtx_insn *) data));
4495 if (subst != loc)
4496 return subst;
4498 return NULL_RTX;
4501 /* Maximum number of generated reload insns per an insn. It is for
4502 preventing this pass cycling in a bug case. */
4503 #define MAX_RELOAD_INSNS_NUMBER LRA_MAX_INSN_RELOADS
4505 /* The current iteration number of this LRA pass. */
4506 int lra_constraint_iter;
4508 /* True if we substituted equiv which needs checking register
4509 allocation correctness because the equivalent value contains
4510 allocatable hard registers or when we restore multi-register
4511 pseudo. */
4512 bool lra_risky_transformations_p;
4514 /* Return true if REGNO is referenced in more than one block. */
4515 static bool
4516 multi_block_pseudo_p (int regno)
4518 basic_block bb = NULL;
4519 unsigned int uid;
4520 bitmap_iterator bi;
4522 if (regno < FIRST_PSEUDO_REGISTER)
4523 return false;
4525 EXECUTE_IF_SET_IN_BITMAP (&lra_reg_info[regno].insn_bitmap, 0, uid, bi)
4526 if (bb == NULL)
4527 bb = BLOCK_FOR_INSN (lra_insn_recog_data[uid]->insn);
4528 else if (BLOCK_FOR_INSN (lra_insn_recog_data[uid]->insn) != bb)
4529 return true;
4530 return false;
4533 /* Return true if LIST contains a deleted insn. */
4534 static bool
4535 contains_deleted_insn_p (rtx_insn_list *list)
4537 for (; list != NULL_RTX; list = list->next ())
4538 if (NOTE_P (list->insn ())
4539 && NOTE_KIND (list->insn ()) == NOTE_INSN_DELETED)
4540 return true;
4541 return false;
4544 /* Return true if X contains a pseudo dying in INSN. */
4545 static bool
4546 dead_pseudo_p (rtx x, rtx_insn *insn)
4548 int i, j;
4549 const char *fmt;
4550 enum rtx_code code;
4552 if (REG_P (x))
4553 return (insn != NULL_RTX
4554 && find_regno_note (insn, REG_DEAD, REGNO (x)) != NULL_RTX);
4555 code = GET_CODE (x);
4556 fmt = GET_RTX_FORMAT (code);
4557 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
4559 if (fmt[i] == 'e')
4561 if (dead_pseudo_p (XEXP (x, i), insn))
4562 return true;
4564 else if (fmt[i] == 'E')
4566 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
4567 if (dead_pseudo_p (XVECEXP (x, i, j), insn))
4568 return true;
4571 return false;
4574 /* Return true if INSN contains a dying pseudo in INSN right hand
4575 side. */
4576 static bool
4577 insn_rhs_dead_pseudo_p (rtx_insn *insn)
4579 rtx set = single_set (insn);
4581 gcc_assert (set != NULL);
4582 return dead_pseudo_p (SET_SRC (set), insn);
4585 /* Return true if any init insn of REGNO contains a dying pseudo in
4586 insn right hand side. */
4587 static bool
4588 init_insn_rhs_dead_pseudo_p (int regno)
4590 rtx_insn_list *insns = ira_reg_equiv[regno].init_insns;
4592 if (insns == NULL)
4593 return false;
4594 for (; insns != NULL_RTX; insns = insns->next ())
4595 if (insn_rhs_dead_pseudo_p (insns->insn ()))
4596 return true;
4597 return false;
4600 /* Return TRUE if REGNO has a reverse equivalence. The equivalence is
4601 reverse only if we have one init insn with given REGNO as a
4602 source. */
4603 static bool
4604 reverse_equiv_p (int regno)
4606 rtx_insn_list *insns = ira_reg_equiv[regno].init_insns;
4607 rtx set;
4609 if (insns == NULL)
4610 return false;
4611 if (! INSN_P (insns->insn ())
4612 || insns->next () != NULL)
4613 return false;
4614 if ((set = single_set (insns->insn ())) == NULL_RTX)
4615 return false;
4616 return REG_P (SET_SRC (set)) && (int) REGNO (SET_SRC (set)) == regno;
4619 /* Return TRUE if REGNO was reloaded in an equivalence init insn. We
4620 call this function only for non-reverse equivalence. */
4621 static bool
4622 contains_reloaded_insn_p (int regno)
4624 rtx set;
4625 rtx_insn_list *list = ira_reg_equiv[regno].init_insns;
4627 for (; list != NULL; list = list->next ())
4628 if ((set = single_set (list->insn ())) == NULL_RTX
4629 || ! REG_P (SET_DEST (set))
4630 || (int) REGNO (SET_DEST (set)) != regno)
4631 return true;
4632 return false;
4635 /* Entry function of LRA constraint pass. Return true if the
4636 constraint pass did change the code. */
4637 bool
4638 lra_constraints (bool first_p)
4640 bool changed_p;
4641 int i, hard_regno, new_insns_num;
4642 unsigned int min_len, new_min_len, uid;
4643 rtx set, x, reg, dest_reg;
4644 basic_block last_bb;
4645 bitmap_iterator bi;
4647 lra_constraint_iter++;
4648 if (lra_dump_file != NULL)
4649 fprintf (lra_dump_file, "\n********** Local #%d: **********\n\n",
4650 lra_constraint_iter);
4651 changed_p = false;
4652 if (pic_offset_table_rtx
4653 && REGNO (pic_offset_table_rtx) >= FIRST_PSEUDO_REGISTER)
4654 lra_risky_transformations_p = true;
4655 else
4656 /* On the first iteration we should check IRA assignment
4657 correctness. In rare cases, the assignments can be wrong as
4658 early clobbers operands are ignored in IRA. */
4659 lra_risky_transformations_p = first_p;
4660 new_insn_uid_start = get_max_uid ();
4661 new_regno_start = first_p ? lra_constraint_new_regno_start : max_reg_num ();
4662 /* Mark used hard regs for target stack size calulations. */
4663 for (i = FIRST_PSEUDO_REGISTER; i < new_regno_start; i++)
4664 if (lra_reg_info[i].nrefs != 0
4665 && (hard_regno = lra_get_regno_hard_regno (i)) >= 0)
4667 int j, nregs;
4669 nregs = hard_regno_nregs (hard_regno, lra_reg_info[i].biggest_mode);
4670 for (j = 0; j < nregs; j++)
4671 df_set_regs_ever_live (hard_regno + j, true);
4673 /* Do elimination before the equivalence processing as we can spill
4674 some pseudos during elimination. */
4675 lra_eliminate (false, first_p);
4676 auto_bitmap equiv_insn_bitmap (&reg_obstack);
4677 for (i = FIRST_PSEUDO_REGISTER; i < new_regno_start; i++)
4678 if (lra_reg_info[i].nrefs != 0)
4680 ira_reg_equiv[i].profitable_p = true;
4681 reg = regno_reg_rtx[i];
4682 if (lra_get_regno_hard_regno (i) < 0 && (x = get_equiv (reg)) != reg)
4684 bool pseudo_p = contains_reg_p (x, false, false);
4686 /* After RTL transformation, we can not guarantee that
4687 pseudo in the substitution was not reloaded which might
4688 make equivalence invalid. For example, in reverse
4689 equiv of p0
4691 p0 <- ...
4693 equiv_mem <- p0
4695 the memory address register was reloaded before the 2nd
4696 insn. */
4697 if ((! first_p && pseudo_p)
4698 /* We don't use DF for compilation speed sake. So it
4699 is problematic to update live info when we use an
4700 equivalence containing pseudos in more than one
4701 BB. */
4702 || (pseudo_p && multi_block_pseudo_p (i))
4703 /* If an init insn was deleted for some reason, cancel
4704 the equiv. We could update the equiv insns after
4705 transformations including an equiv insn deletion
4706 but it is not worthy as such cases are extremely
4707 rare. */
4708 || contains_deleted_insn_p (ira_reg_equiv[i].init_insns)
4709 /* If it is not a reverse equivalence, we check that a
4710 pseudo in rhs of the init insn is not dying in the
4711 insn. Otherwise, the live info at the beginning of
4712 the corresponding BB might be wrong after we
4713 removed the insn. When the equiv can be a
4714 constant, the right hand side of the init insn can
4715 be a pseudo. */
4716 || (! reverse_equiv_p (i)
4717 && (init_insn_rhs_dead_pseudo_p (i)
4718 /* If we reloaded the pseudo in an equivalence
4719 init insn, we can not remove the equiv init
4720 insns and the init insns might write into
4721 const memory in this case. */
4722 || contains_reloaded_insn_p (i)))
4723 /* Prevent access beyond equivalent memory for
4724 paradoxical subregs. */
4725 || (MEM_P (x)
4726 && (GET_MODE_SIZE (lra_reg_info[i].biggest_mode)
4727 > GET_MODE_SIZE (GET_MODE (x))))
4728 || (pic_offset_table_rtx
4729 && ((CONST_POOL_OK_P (PSEUDO_REGNO_MODE (i), x)
4730 && (targetm.preferred_reload_class
4731 (x, lra_get_allocno_class (i)) == NO_REGS))
4732 || contains_symbol_ref_p (x))))
4733 ira_reg_equiv[i].defined_p = false;
4734 if (contains_reg_p (x, false, true))
4735 ira_reg_equiv[i].profitable_p = false;
4736 if (get_equiv (reg) != reg)
4737 bitmap_ior_into (equiv_insn_bitmap, &lra_reg_info[i].insn_bitmap);
4740 for (i = FIRST_PSEUDO_REGISTER; i < new_regno_start; i++)
4741 update_equiv (i);
4742 /* We should add all insns containing pseudos which should be
4743 substituted by their equivalences. */
4744 EXECUTE_IF_SET_IN_BITMAP (equiv_insn_bitmap, 0, uid, bi)
4745 lra_push_insn_by_uid (uid);
4746 min_len = lra_insn_stack_length ();
4747 new_insns_num = 0;
4748 last_bb = NULL;
4749 changed_p = false;
4750 while ((new_min_len = lra_insn_stack_length ()) != 0)
4752 curr_insn = lra_pop_insn ();
4753 --new_min_len;
4754 curr_bb = BLOCK_FOR_INSN (curr_insn);
4755 if (curr_bb != last_bb)
4757 last_bb = curr_bb;
4758 bb_reload_num = lra_curr_reload_num;
4760 if (min_len > new_min_len)
4762 min_len = new_min_len;
4763 new_insns_num = 0;
4765 if (new_insns_num > MAX_RELOAD_INSNS_NUMBER)
4766 internal_error
4767 ("Max. number of generated reload insns per insn is achieved (%d)\n",
4768 MAX_RELOAD_INSNS_NUMBER);
4769 new_insns_num++;
4770 if (DEBUG_INSN_P (curr_insn))
4772 /* We need to check equivalence in debug insn and change
4773 pseudo to the equivalent value if necessary. */
4774 curr_id = lra_get_insn_recog_data (curr_insn);
4775 if (bitmap_bit_p (equiv_insn_bitmap, INSN_UID (curr_insn)))
4777 rtx old = *curr_id->operand_loc[0];
4778 *curr_id->operand_loc[0]
4779 = simplify_replace_fn_rtx (old, NULL_RTX,
4780 loc_equivalence_callback, curr_insn);
4781 if (old != *curr_id->operand_loc[0])
4783 lra_update_insn_regno_info (curr_insn);
4784 changed_p = true;
4788 else if (INSN_P (curr_insn))
4790 if ((set = single_set (curr_insn)) != NULL_RTX)
4792 dest_reg = SET_DEST (set);
4793 /* The equivalence pseudo could be set up as SUBREG in a
4794 case when it is a call restore insn in a mode
4795 different from the pseudo mode. */
4796 if (GET_CODE (dest_reg) == SUBREG)
4797 dest_reg = SUBREG_REG (dest_reg);
4798 if ((REG_P (dest_reg)
4799 && (x = get_equiv (dest_reg)) != dest_reg
4800 /* Remove insns which set up a pseudo whose value
4801 can not be changed. Such insns might be not in
4802 init_insns because we don't update equiv data
4803 during insn transformations.
4805 As an example, let suppose that a pseudo got
4806 hard register and on the 1st pass was not
4807 changed to equivalent constant. We generate an
4808 additional insn setting up the pseudo because of
4809 secondary memory movement. Then the pseudo is
4810 spilled and we use the equiv constant. In this
4811 case we should remove the additional insn and
4812 this insn is not init_insns list. */
4813 && (! MEM_P (x) || MEM_READONLY_P (x)
4814 /* Check that this is actually an insn setting
4815 up the equivalence. */
4816 || in_list_p (curr_insn,
4817 ira_reg_equiv
4818 [REGNO (dest_reg)].init_insns)))
4819 || (((x = get_equiv (SET_SRC (set))) != SET_SRC (set))
4820 && in_list_p (curr_insn,
4821 ira_reg_equiv
4822 [REGNO (SET_SRC (set))].init_insns)))
4824 /* This is equiv init insn of pseudo which did not get a
4825 hard register -- remove the insn. */
4826 if (lra_dump_file != NULL)
4828 fprintf (lra_dump_file,
4829 " Removing equiv init insn %i (freq=%d)\n",
4830 INSN_UID (curr_insn),
4831 REG_FREQ_FROM_BB (BLOCK_FOR_INSN (curr_insn)));
4832 dump_insn_slim (lra_dump_file, curr_insn);
4834 if (contains_reg_p (x, true, false))
4835 lra_risky_transformations_p = true;
4836 lra_set_insn_deleted (curr_insn);
4837 continue;
4840 curr_id = lra_get_insn_recog_data (curr_insn);
4841 curr_static_id = curr_id->insn_static_data;
4842 init_curr_insn_input_reloads ();
4843 init_curr_operand_mode ();
4844 if (curr_insn_transform (false))
4845 changed_p = true;
4846 /* Check non-transformed insns too for equiv change as USE
4847 or CLOBBER don't need reloads but can contain pseudos
4848 being changed on their equivalences. */
4849 else if (bitmap_bit_p (equiv_insn_bitmap, INSN_UID (curr_insn))
4850 && loc_equivalence_change_p (&PATTERN (curr_insn)))
4852 lra_update_insn_regno_info (curr_insn);
4853 changed_p = true;
4858 /* If we used a new hard regno, changed_p should be true because the
4859 hard reg is assigned to a new pseudo. */
4860 if (flag_checking && !changed_p)
4862 for (i = FIRST_PSEUDO_REGISTER; i < new_regno_start; i++)
4863 if (lra_reg_info[i].nrefs != 0
4864 && (hard_regno = lra_get_regno_hard_regno (i)) >= 0)
4866 int j, nregs = hard_regno_nregs (hard_regno,
4867 PSEUDO_REGNO_MODE (i));
4869 for (j = 0; j < nregs; j++)
4870 lra_assert (df_regs_ever_live_p (hard_regno + j));
4873 return changed_p;
4876 static void initiate_invariants (void);
4877 static void finish_invariants (void);
4879 /* Initiate the LRA constraint pass. It is done once per
4880 function. */
4881 void
4882 lra_constraints_init (void)
4884 initiate_invariants ();
4887 /* Finalize the LRA constraint pass. It is done once per
4888 function. */
4889 void
4890 lra_constraints_finish (void)
4892 finish_invariants ();
4897 /* Structure describes invariants for ineheritance. */
4898 struct lra_invariant
4900 /* The order number of the invariant. */
4901 int num;
4902 /* The invariant RTX. */
4903 rtx invariant_rtx;
4904 /* The origin insn of the invariant. */
4905 rtx_insn *insn;
4908 typedef lra_invariant invariant_t;
4909 typedef invariant_t *invariant_ptr_t;
4910 typedef const invariant_t *const_invariant_ptr_t;
4912 /* Pointer to the inheritance invariants. */
4913 static vec<invariant_ptr_t> invariants;
4915 /* Allocation pool for the invariants. */
4916 static object_allocator<lra_invariant> *invariants_pool;
4918 /* Hash table for the invariants. */
4919 static htab_t invariant_table;
4921 /* Hash function for INVARIANT. */
4922 static hashval_t
4923 invariant_hash (const void *invariant)
4925 rtx inv = ((const_invariant_ptr_t) invariant)->invariant_rtx;
4926 return lra_rtx_hash (inv);
4929 /* Equal function for invariants INVARIANT1 and INVARIANT2. */
4930 static int
4931 invariant_eq_p (const void *invariant1, const void *invariant2)
4933 rtx inv1 = ((const_invariant_ptr_t) invariant1)->invariant_rtx;
4934 rtx inv2 = ((const_invariant_ptr_t) invariant2)->invariant_rtx;
4936 return rtx_equal_p (inv1, inv2);
4939 /* Insert INVARIANT_RTX into the table if it is not there yet. Return
4940 invariant which is in the table. */
4941 static invariant_ptr_t
4942 insert_invariant (rtx invariant_rtx)
4944 void **entry_ptr;
4945 invariant_t invariant;
4946 invariant_ptr_t invariant_ptr;
4948 invariant.invariant_rtx = invariant_rtx;
4949 entry_ptr = htab_find_slot (invariant_table, &invariant, INSERT);
4950 if (*entry_ptr == NULL)
4952 invariant_ptr = invariants_pool->allocate ();
4953 invariant_ptr->invariant_rtx = invariant_rtx;
4954 invariant_ptr->insn = NULL;
4955 invariants.safe_push (invariant_ptr);
4956 *entry_ptr = (void *) invariant_ptr;
4958 return (invariant_ptr_t) *entry_ptr;
4961 /* Initiate the invariant table. */
4962 static void
4963 initiate_invariants (void)
4965 invariants.create (100);
4966 invariants_pool
4967 = new object_allocator<lra_invariant> ("Inheritance invariants");
4968 invariant_table = htab_create (100, invariant_hash, invariant_eq_p, NULL);
4971 /* Finish the invariant table. */
4972 static void
4973 finish_invariants (void)
4975 htab_delete (invariant_table);
4976 delete invariants_pool;
4977 invariants.release ();
4980 /* Make the invariant table empty. */
4981 static void
4982 clear_invariants (void)
4984 htab_empty (invariant_table);
4985 invariants_pool->release ();
4986 invariants.truncate (0);
4991 /* This page contains code to do inheritance/split
4992 transformations. */
4994 /* Number of reloads passed so far in current EBB. */
4995 static int reloads_num;
4997 /* Number of calls passed so far in current EBB. */
4998 static int calls_num;
5000 /* Current reload pseudo check for validity of elements in
5001 USAGE_INSNS. */
5002 static int curr_usage_insns_check;
5004 /* Info about last usage of registers in EBB to do inheritance/split
5005 transformation. Inheritance transformation is done from a spilled
5006 pseudo and split transformations from a hard register or a pseudo
5007 assigned to a hard register. */
5008 struct usage_insns
5010 /* If the value is equal to CURR_USAGE_INSNS_CHECK, then the member
5011 value INSNS is valid. The insns is chain of optional debug insns
5012 and a finishing non-debug insn using the corresponding reg. The
5013 value is also used to mark the registers which are set up in the
5014 current insn. The negated insn uid is used for this. */
5015 int check;
5016 /* Value of global reloads_num at the last insn in INSNS. */
5017 int reloads_num;
5018 /* Value of global reloads_nums at the last insn in INSNS. */
5019 int calls_num;
5020 /* It can be true only for splitting. And it means that the restore
5021 insn should be put after insn given by the following member. */
5022 bool after_p;
5023 /* Next insns in the current EBB which use the original reg and the
5024 original reg value is not changed between the current insn and
5025 the next insns. In order words, e.g. for inheritance, if we need
5026 to use the original reg value again in the next insns we can try
5027 to use the value in a hard register from a reload insn of the
5028 current insn. */
5029 rtx insns;
5032 /* Map: regno -> corresponding pseudo usage insns. */
5033 static struct usage_insns *usage_insns;
5035 static void
5036 setup_next_usage_insn (int regno, rtx insn, int reloads_num, bool after_p)
5038 usage_insns[regno].check = curr_usage_insns_check;
5039 usage_insns[regno].insns = insn;
5040 usage_insns[regno].reloads_num = reloads_num;
5041 usage_insns[regno].calls_num = calls_num;
5042 usage_insns[regno].after_p = after_p;
5045 /* The function is used to form list REGNO usages which consists of
5046 optional debug insns finished by a non-debug insn using REGNO.
5047 RELOADS_NUM is current number of reload insns processed so far. */
5048 static void
5049 add_next_usage_insn (int regno, rtx_insn *insn, int reloads_num)
5051 rtx next_usage_insns;
5053 if (usage_insns[regno].check == curr_usage_insns_check
5054 && (next_usage_insns = usage_insns[regno].insns) != NULL_RTX
5055 && DEBUG_INSN_P (insn))
5057 /* Check that we did not add the debug insn yet. */
5058 if (next_usage_insns != insn
5059 && (GET_CODE (next_usage_insns) != INSN_LIST
5060 || XEXP (next_usage_insns, 0) != insn))
5061 usage_insns[regno].insns = gen_rtx_INSN_LIST (VOIDmode, insn,
5062 next_usage_insns);
5064 else if (NONDEBUG_INSN_P (insn))
5065 setup_next_usage_insn (regno, insn, reloads_num, false);
5066 else
5067 usage_insns[regno].check = 0;
5070 /* Return first non-debug insn in list USAGE_INSNS. */
5071 static rtx_insn *
5072 skip_usage_debug_insns (rtx usage_insns)
5074 rtx insn;
5076 /* Skip debug insns. */
5077 for (insn = usage_insns;
5078 insn != NULL_RTX && GET_CODE (insn) == INSN_LIST;
5079 insn = XEXP (insn, 1))
5081 return safe_as_a <rtx_insn *> (insn);
5084 /* Return true if we need secondary memory moves for insn in
5085 USAGE_INSNS after inserting inherited pseudo of class INHER_CL
5086 into the insn. */
5087 static bool
5088 check_secondary_memory_needed_p (enum reg_class inher_cl ATTRIBUTE_UNUSED,
5089 rtx usage_insns ATTRIBUTE_UNUSED)
5091 rtx_insn *insn;
5092 rtx set, dest;
5093 enum reg_class cl;
5095 if (inher_cl == ALL_REGS
5096 || (insn = skip_usage_debug_insns (usage_insns)) == NULL_RTX)
5097 return false;
5098 lra_assert (INSN_P (insn));
5099 if ((set = single_set (insn)) == NULL_RTX || ! REG_P (SET_DEST (set)))
5100 return false;
5101 dest = SET_DEST (set);
5102 if (! REG_P (dest))
5103 return false;
5104 lra_assert (inher_cl != NO_REGS);
5105 cl = get_reg_class (REGNO (dest));
5106 return (cl != NO_REGS && cl != ALL_REGS
5107 && targetm.secondary_memory_needed (GET_MODE (dest), inher_cl, cl));
5110 /* Registers involved in inheritance/split in the current EBB
5111 (inheritance/split pseudos and original registers). */
5112 static bitmap_head check_only_regs;
5114 /* Reload pseudos can not be involded in invariant inheritance in the
5115 current EBB. */
5116 static bitmap_head invalid_invariant_regs;
5118 /* Do inheritance transformations for insn INSN, which defines (if
5119 DEF_P) or uses ORIGINAL_REGNO. NEXT_USAGE_INSNS specifies which
5120 instruction in the EBB next uses ORIGINAL_REGNO; it has the same
5121 form as the "insns" field of usage_insns. Return true if we
5122 succeed in such transformation.
5124 The transformations look like:
5126 p <- ... i <- ...
5127 ... p <- i (new insn)
5128 ... =>
5129 <- ... p ... <- ... i ...
5131 ... i <- p (new insn)
5132 <- ... p ... <- ... i ...
5133 ... =>
5134 <- ... p ... <- ... i ...
5135 where p is a spilled original pseudo and i is a new inheritance pseudo.
5138 The inheritance pseudo has the smallest class of two classes CL and
5139 class of ORIGINAL REGNO. */
5140 static bool
5141 inherit_reload_reg (bool def_p, int original_regno,
5142 enum reg_class cl, rtx_insn *insn, rtx next_usage_insns)
5144 if (optimize_function_for_size_p (cfun))
5145 return false;
5147 enum reg_class rclass = lra_get_allocno_class (original_regno);
5148 rtx original_reg = regno_reg_rtx[original_regno];
5149 rtx new_reg, usage_insn;
5150 rtx_insn *new_insns;
5152 lra_assert (! usage_insns[original_regno].after_p);
5153 if (lra_dump_file != NULL)
5154 fprintf (lra_dump_file,
5155 " <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n");
5156 if (! ira_reg_classes_intersect_p[cl][rclass])
5158 if (lra_dump_file != NULL)
5160 fprintf (lra_dump_file,
5161 " Rejecting inheritance for %d "
5162 "because of disjoint classes %s and %s\n",
5163 original_regno, reg_class_names[cl],
5164 reg_class_names[rclass]);
5165 fprintf (lra_dump_file,
5166 " >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
5168 return false;
5170 if ((ira_class_subset_p[cl][rclass] && cl != rclass)
5171 /* We don't use a subset of two classes because it can be
5172 NO_REGS. This transformation is still profitable in most
5173 cases even if the classes are not intersected as register
5174 move is probably cheaper than a memory load. */
5175 || ira_class_hard_regs_num[cl] < ira_class_hard_regs_num[rclass])
5177 if (lra_dump_file != NULL)
5178 fprintf (lra_dump_file, " Use smallest class of %s and %s\n",
5179 reg_class_names[cl], reg_class_names[rclass]);
5181 rclass = cl;
5183 if (check_secondary_memory_needed_p (rclass, next_usage_insns))
5185 /* Reject inheritance resulting in secondary memory moves.
5186 Otherwise, there is a danger in LRA cycling. Also such
5187 transformation will be unprofitable. */
5188 if (lra_dump_file != NULL)
5190 rtx_insn *insn = skip_usage_debug_insns (next_usage_insns);
5191 rtx set = single_set (insn);
5193 lra_assert (set != NULL_RTX);
5195 rtx dest = SET_DEST (set);
5197 lra_assert (REG_P (dest));
5198 fprintf (lra_dump_file,
5199 " Rejecting inheritance for insn %d(%s)<-%d(%s) "
5200 "as secondary mem is needed\n",
5201 REGNO (dest), reg_class_names[get_reg_class (REGNO (dest))],
5202 original_regno, reg_class_names[rclass]);
5203 fprintf (lra_dump_file,
5204 " >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
5206 return false;
5208 new_reg = lra_create_new_reg (GET_MODE (original_reg), original_reg,
5209 rclass, "inheritance");
5210 start_sequence ();
5211 if (def_p)
5212 lra_emit_move (original_reg, new_reg);
5213 else
5214 lra_emit_move (new_reg, original_reg);
5215 new_insns = get_insns ();
5216 end_sequence ();
5217 if (NEXT_INSN (new_insns) != NULL_RTX)
5219 if (lra_dump_file != NULL)
5221 fprintf (lra_dump_file,
5222 " Rejecting inheritance %d->%d "
5223 "as it results in 2 or more insns:\n",
5224 original_regno, REGNO (new_reg));
5225 dump_rtl_slim (lra_dump_file, new_insns, NULL, -1, 0);
5226 fprintf (lra_dump_file,
5227 " >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
5229 return false;
5231 lra_substitute_pseudo_within_insn (insn, original_regno, new_reg, false);
5232 lra_update_insn_regno_info (insn);
5233 if (! def_p)
5234 /* We now have a new usage insn for original regno. */
5235 setup_next_usage_insn (original_regno, new_insns, reloads_num, false);
5236 if (lra_dump_file != NULL)
5237 fprintf (lra_dump_file, " Original reg change %d->%d (bb%d):\n",
5238 original_regno, REGNO (new_reg), BLOCK_FOR_INSN (insn)->index);
5239 lra_reg_info[REGNO (new_reg)].restore_rtx = regno_reg_rtx[original_regno];
5240 bitmap_set_bit (&check_only_regs, REGNO (new_reg));
5241 bitmap_set_bit (&check_only_regs, original_regno);
5242 bitmap_set_bit (&lra_inheritance_pseudos, REGNO (new_reg));
5243 if (def_p)
5244 lra_process_new_insns (insn, NULL, new_insns,
5245 "Add original<-inheritance");
5246 else
5247 lra_process_new_insns (insn, new_insns, NULL,
5248 "Add inheritance<-original");
5249 while (next_usage_insns != NULL_RTX)
5251 if (GET_CODE (next_usage_insns) != INSN_LIST)
5253 usage_insn = next_usage_insns;
5254 lra_assert (NONDEBUG_INSN_P (usage_insn));
5255 next_usage_insns = NULL;
5257 else
5259 usage_insn = XEXP (next_usage_insns, 0);
5260 lra_assert (DEBUG_INSN_P (usage_insn));
5261 next_usage_insns = XEXP (next_usage_insns, 1);
5263 lra_substitute_pseudo (&usage_insn, original_regno, new_reg, false);
5264 lra_update_insn_regno_info (as_a <rtx_insn *> (usage_insn));
5265 if (lra_dump_file != NULL)
5267 fprintf (lra_dump_file,
5268 " Inheritance reuse change %d->%d (bb%d):\n",
5269 original_regno, REGNO (new_reg),
5270 BLOCK_FOR_INSN (usage_insn)->index);
5271 dump_insn_slim (lra_dump_file, as_a <rtx_insn *> (usage_insn));
5274 if (lra_dump_file != NULL)
5275 fprintf (lra_dump_file,
5276 " >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
5277 return true;
5280 /* Return true if we need a caller save/restore for pseudo REGNO which
5281 was assigned to a hard register. */
5282 static inline bool
5283 need_for_call_save_p (int regno)
5285 lra_assert (regno >= FIRST_PSEUDO_REGISTER && reg_renumber[regno] >= 0);
5286 return (usage_insns[regno].calls_num < calls_num
5287 && (overlaps_hard_reg_set_p
5288 ((flag_ipa_ra &&
5289 ! hard_reg_set_empty_p (lra_reg_info[regno].actual_call_used_reg_set))
5290 ? lra_reg_info[regno].actual_call_used_reg_set
5291 : call_used_reg_set,
5292 PSEUDO_REGNO_MODE (regno), reg_renumber[regno])
5293 || (targetm.hard_regno_call_part_clobbered
5294 (reg_renumber[regno], PSEUDO_REGNO_MODE (regno)))));
5297 /* Global registers occurring in the current EBB. */
5298 static bitmap_head ebb_global_regs;
5300 /* Return true if we need a split for hard register REGNO or pseudo
5301 REGNO which was assigned to a hard register.
5302 POTENTIAL_RELOAD_HARD_REGS contains hard registers which might be
5303 used for reloads since the EBB end. It is an approximation of the
5304 used hard registers in the split range. The exact value would
5305 require expensive calculations. If we were aggressive with
5306 splitting because of the approximation, the split pseudo will save
5307 the same hard register assignment and will be removed in the undo
5308 pass. We still need the approximation because too aggressive
5309 splitting would result in too inaccurate cost calculation in the
5310 assignment pass because of too many generated moves which will be
5311 probably removed in the undo pass. */
5312 static inline bool
5313 need_for_split_p (HARD_REG_SET potential_reload_hard_regs, int regno)
5315 int hard_regno = regno < FIRST_PSEUDO_REGISTER ? regno : reg_renumber[regno];
5317 lra_assert (hard_regno >= 0);
5318 return ((TEST_HARD_REG_BIT (potential_reload_hard_regs, hard_regno)
5319 /* Don't split eliminable hard registers, otherwise we can
5320 split hard registers like hard frame pointer, which
5321 lives on BB start/end according to DF-infrastructure,
5322 when there is a pseudo assigned to the register and
5323 living in the same BB. */
5324 && (regno >= FIRST_PSEUDO_REGISTER
5325 || ! TEST_HARD_REG_BIT (eliminable_regset, hard_regno))
5326 && ! TEST_HARD_REG_BIT (lra_no_alloc_regs, hard_regno)
5327 /* Don't split call clobbered hard regs living through
5328 calls, otherwise we might have a check problem in the
5329 assign sub-pass as in the most cases (exception is a
5330 situation when lra_risky_transformations_p value is
5331 true) the assign pass assumes that all pseudos living
5332 through calls are assigned to call saved hard regs. */
5333 && (regno >= FIRST_PSEUDO_REGISTER
5334 || ! TEST_HARD_REG_BIT (call_used_reg_set, regno)
5335 || usage_insns[regno].calls_num == calls_num)
5336 /* We need at least 2 reloads to make pseudo splitting
5337 profitable. We should provide hard regno splitting in
5338 any case to solve 1st insn scheduling problem when
5339 moving hard register definition up might result in
5340 impossibility to find hard register for reload pseudo of
5341 small register class. */
5342 && (usage_insns[regno].reloads_num
5343 + (regno < FIRST_PSEUDO_REGISTER ? 0 : 3) < reloads_num)
5344 && (regno < FIRST_PSEUDO_REGISTER
5345 /* For short living pseudos, spilling + inheritance can
5346 be considered a substitution for splitting.
5347 Therefore we do not splitting for local pseudos. It
5348 decreases also aggressiveness of splitting. The
5349 minimal number of references is chosen taking into
5350 account that for 2 references splitting has no sense
5351 as we can just spill the pseudo. */
5352 || (regno >= FIRST_PSEUDO_REGISTER
5353 && lra_reg_info[regno].nrefs > 3
5354 && bitmap_bit_p (&ebb_global_regs, regno))))
5355 || (regno >= FIRST_PSEUDO_REGISTER && need_for_call_save_p (regno)));
5358 /* Return class for the split pseudo created from original pseudo with
5359 ALLOCNO_CLASS and MODE which got a hard register HARD_REGNO. We
5360 choose subclass of ALLOCNO_CLASS which contains HARD_REGNO and
5361 results in no secondary memory movements. */
5362 static enum reg_class
5363 choose_split_class (enum reg_class allocno_class,
5364 int hard_regno ATTRIBUTE_UNUSED,
5365 machine_mode mode ATTRIBUTE_UNUSED)
5367 int i;
5368 enum reg_class cl, best_cl = NO_REGS;
5369 enum reg_class hard_reg_class ATTRIBUTE_UNUSED
5370 = REGNO_REG_CLASS (hard_regno);
5372 if (! targetm.secondary_memory_needed (mode, allocno_class, allocno_class)
5373 && TEST_HARD_REG_BIT (reg_class_contents[allocno_class], hard_regno))
5374 return allocno_class;
5375 for (i = 0;
5376 (cl = reg_class_subclasses[allocno_class][i]) != LIM_REG_CLASSES;
5377 i++)
5378 if (! targetm.secondary_memory_needed (mode, cl, hard_reg_class)
5379 && ! targetm.secondary_memory_needed (mode, hard_reg_class, cl)
5380 && TEST_HARD_REG_BIT (reg_class_contents[cl], hard_regno)
5381 && (best_cl == NO_REGS
5382 || ira_class_hard_regs_num[best_cl] < ira_class_hard_regs_num[cl]))
5383 best_cl = cl;
5384 return best_cl;
5387 /* Copy any equivalence information from ORIGINAL_REGNO to NEW_REGNO.
5388 It only makes sense to call this function if NEW_REGNO is always
5389 equal to ORIGINAL_REGNO. */
5391 static void
5392 lra_copy_reg_equiv (unsigned int new_regno, unsigned int original_regno)
5394 if (!ira_reg_equiv[original_regno].defined_p)
5395 return;
5397 ira_expand_reg_equiv ();
5398 ira_reg_equiv[new_regno].defined_p = true;
5399 if (ira_reg_equiv[original_regno].memory)
5400 ira_reg_equiv[new_regno].memory
5401 = copy_rtx (ira_reg_equiv[original_regno].memory);
5402 if (ira_reg_equiv[original_regno].constant)
5403 ira_reg_equiv[new_regno].constant
5404 = copy_rtx (ira_reg_equiv[original_regno].constant);
5405 if (ira_reg_equiv[original_regno].invariant)
5406 ira_reg_equiv[new_regno].invariant
5407 = copy_rtx (ira_reg_equiv[original_regno].invariant);
5410 /* Do split transformations for insn INSN, which defines or uses
5411 ORIGINAL_REGNO. NEXT_USAGE_INSNS specifies which instruction in
5412 the EBB next uses ORIGINAL_REGNO; it has the same form as the
5413 "insns" field of usage_insns.
5415 The transformations look like:
5417 p <- ... p <- ...
5418 ... s <- p (new insn -- save)
5419 ... =>
5420 ... p <- s (new insn -- restore)
5421 <- ... p ... <- ... p ...
5423 <- ... p ... <- ... p ...
5424 ... s <- p (new insn -- save)
5425 ... =>
5426 ... p <- s (new insn -- restore)
5427 <- ... p ... <- ... p ...
5429 where p is an original pseudo got a hard register or a hard
5430 register and s is a new split pseudo. The save is put before INSN
5431 if BEFORE_P is true. Return true if we succeed in such
5432 transformation. */
5433 static bool
5434 split_reg (bool before_p, int original_regno, rtx_insn *insn,
5435 rtx next_usage_insns)
5437 enum reg_class rclass;
5438 rtx original_reg;
5439 int hard_regno, nregs;
5440 rtx new_reg, usage_insn;
5441 rtx_insn *restore, *save;
5442 bool after_p;
5443 bool call_save_p;
5444 machine_mode mode;
5446 if (original_regno < FIRST_PSEUDO_REGISTER)
5448 rclass = ira_allocno_class_translate[REGNO_REG_CLASS (original_regno)];
5449 hard_regno = original_regno;
5450 call_save_p = false;
5451 nregs = 1;
5452 mode = lra_reg_info[hard_regno].biggest_mode;
5453 machine_mode reg_rtx_mode = GET_MODE (regno_reg_rtx[hard_regno]);
5454 /* A reg can have a biggest_mode of VOIDmode if it was only ever seen
5455 as part of a multi-word register. In that case, or if the biggest
5456 mode was larger than a register, just use the reg_rtx. Otherwise,
5457 limit the size to that of the biggest access in the function. */
5458 if (mode == VOIDmode
5459 || paradoxical_subreg_p (mode, reg_rtx_mode))
5461 original_reg = regno_reg_rtx[hard_regno];
5462 mode = reg_rtx_mode;
5464 else
5465 original_reg = gen_rtx_REG (mode, hard_regno);
5467 else
5469 mode = PSEUDO_REGNO_MODE (original_regno);
5470 hard_regno = reg_renumber[original_regno];
5471 nregs = hard_regno_nregs (hard_regno, mode);
5472 rclass = lra_get_allocno_class (original_regno);
5473 original_reg = regno_reg_rtx[original_regno];
5474 call_save_p = need_for_call_save_p (original_regno);
5476 lra_assert (hard_regno >= 0);
5477 if (lra_dump_file != NULL)
5478 fprintf (lra_dump_file,
5479 " ((((((((((((((((((((((((((((((((((((((((((((((((\n");
5481 if (call_save_p)
5483 mode = HARD_REGNO_CALLER_SAVE_MODE (hard_regno,
5484 hard_regno_nregs (hard_regno, mode),
5485 mode);
5486 new_reg = lra_create_new_reg (mode, NULL_RTX, NO_REGS, "save");
5488 else
5490 rclass = choose_split_class (rclass, hard_regno, mode);
5491 if (rclass == NO_REGS)
5493 if (lra_dump_file != NULL)
5495 fprintf (lra_dump_file,
5496 " Rejecting split of %d(%s): "
5497 "no good reg class for %d(%s)\n",
5498 original_regno,
5499 reg_class_names[lra_get_allocno_class (original_regno)],
5500 hard_regno,
5501 reg_class_names[REGNO_REG_CLASS (hard_regno)]);
5502 fprintf
5503 (lra_dump_file,
5504 " ))))))))))))))))))))))))))))))))))))))))))))))))\n");
5506 return false;
5508 /* Split_if_necessary can split hard registers used as part of a
5509 multi-register mode but splits each register individually. The
5510 mode used for each independent register may not be supported
5511 so reject the split. Splitting the wider mode should theoretically
5512 be possible but is not implemented. */
5513 if (!targetm.hard_regno_mode_ok (hard_regno, mode))
5515 if (lra_dump_file != NULL)
5517 fprintf (lra_dump_file,
5518 " Rejecting split of %d(%s): unsuitable mode %s\n",
5519 original_regno,
5520 reg_class_names[lra_get_allocno_class (original_regno)],
5521 GET_MODE_NAME (mode));
5522 fprintf
5523 (lra_dump_file,
5524 " ))))))))))))))))))))))))))))))))))))))))))))))))\n");
5526 return false;
5528 new_reg = lra_create_new_reg (mode, original_reg, rclass, "split");
5529 reg_renumber[REGNO (new_reg)] = hard_regno;
5531 int new_regno = REGNO (new_reg);
5532 save = emit_spill_move (true, new_reg, original_reg);
5533 if (NEXT_INSN (save) != NULL_RTX && !call_save_p)
5535 if (lra_dump_file != NULL)
5537 fprintf
5538 (lra_dump_file,
5539 " Rejecting split %d->%d resulting in > 2 save insns:\n",
5540 original_regno, new_regno);
5541 dump_rtl_slim (lra_dump_file, save, NULL, -1, 0);
5542 fprintf (lra_dump_file,
5543 " ))))))))))))))))))))))))))))))))))))))))))))))))\n");
5545 return false;
5547 restore = emit_spill_move (false, new_reg, original_reg);
5548 if (NEXT_INSN (restore) != NULL_RTX && !call_save_p)
5550 if (lra_dump_file != NULL)
5552 fprintf (lra_dump_file,
5553 " Rejecting split %d->%d "
5554 "resulting in > 2 restore insns:\n",
5555 original_regno, new_regno);
5556 dump_rtl_slim (lra_dump_file, restore, NULL, -1, 0);
5557 fprintf (lra_dump_file,
5558 " ))))))))))))))))))))))))))))))))))))))))))))))))\n");
5560 return false;
5562 /* Transfer equivalence information to the spill register, so that
5563 if we fail to allocate the spill register, we have the option of
5564 rematerializing the original value instead of spilling to the stack. */
5565 if (!HARD_REGISTER_NUM_P (original_regno)
5566 && mode == PSEUDO_REGNO_MODE (original_regno))
5567 lra_copy_reg_equiv (new_regno, original_regno);
5568 after_p = usage_insns[original_regno].after_p;
5569 lra_reg_info[new_regno].restore_rtx = regno_reg_rtx[original_regno];
5570 bitmap_set_bit (&check_only_regs, new_regno);
5571 bitmap_set_bit (&check_only_regs, original_regno);
5572 bitmap_set_bit (&lra_split_regs, new_regno);
5573 for (;;)
5575 if (GET_CODE (next_usage_insns) != INSN_LIST)
5577 usage_insn = next_usage_insns;
5578 break;
5580 usage_insn = XEXP (next_usage_insns, 0);
5581 lra_assert (DEBUG_INSN_P (usage_insn));
5582 next_usage_insns = XEXP (next_usage_insns, 1);
5583 lra_substitute_pseudo (&usage_insn, original_regno, new_reg, false);
5584 lra_update_insn_regno_info (as_a <rtx_insn *> (usage_insn));
5585 if (lra_dump_file != NULL)
5587 fprintf (lra_dump_file, " Split reuse change %d->%d:\n",
5588 original_regno, new_regno);
5589 dump_insn_slim (lra_dump_file, as_a <rtx_insn *> (usage_insn));
5592 lra_assert (NOTE_P (usage_insn) || NONDEBUG_INSN_P (usage_insn));
5593 lra_assert (usage_insn != insn || (after_p && before_p));
5594 lra_process_new_insns (as_a <rtx_insn *> (usage_insn),
5595 after_p ? NULL : restore,
5596 after_p ? restore : NULL,
5597 call_save_p
5598 ? "Add reg<-save" : "Add reg<-split");
5599 lra_process_new_insns (insn, before_p ? save : NULL,
5600 before_p ? NULL : save,
5601 call_save_p
5602 ? "Add save<-reg" : "Add split<-reg");
5603 if (nregs > 1)
5604 /* If we are trying to split multi-register. We should check
5605 conflicts on the next assignment sub-pass. IRA can allocate on
5606 sub-register levels, LRA do this on pseudos level right now and
5607 this discrepancy may create allocation conflicts after
5608 splitting. */
5609 lra_risky_transformations_p = true;
5610 if (lra_dump_file != NULL)
5611 fprintf (lra_dump_file,
5612 " ))))))))))))))))))))))))))))))))))))))))))))))))\n");
5613 return true;
5616 /* Recognize that we need a split transformation for insn INSN, which
5617 defines or uses REGNO in its insn biggest MODE (we use it only if
5618 REGNO is a hard register). POTENTIAL_RELOAD_HARD_REGS contains
5619 hard registers which might be used for reloads since the EBB end.
5620 Put the save before INSN if BEFORE_P is true. MAX_UID is maximla
5621 uid before starting INSN processing. Return true if we succeed in
5622 such transformation. */
5623 static bool
5624 split_if_necessary (int regno, machine_mode mode,
5625 HARD_REG_SET potential_reload_hard_regs,
5626 bool before_p, rtx_insn *insn, int max_uid)
5628 bool res = false;
5629 int i, nregs = 1;
5630 rtx next_usage_insns;
5632 if (regno < FIRST_PSEUDO_REGISTER)
5633 nregs = hard_regno_nregs (regno, mode);
5634 for (i = 0; i < nregs; i++)
5635 if (usage_insns[regno + i].check == curr_usage_insns_check
5636 && (next_usage_insns = usage_insns[regno + i].insns) != NULL_RTX
5637 /* To avoid processing the register twice or more. */
5638 && ((GET_CODE (next_usage_insns) != INSN_LIST
5639 && INSN_UID (next_usage_insns) < max_uid)
5640 || (GET_CODE (next_usage_insns) == INSN_LIST
5641 && (INSN_UID (XEXP (next_usage_insns, 0)) < max_uid)))
5642 && need_for_split_p (potential_reload_hard_regs, regno + i)
5643 && split_reg (before_p, regno + i, insn, next_usage_insns))
5644 res = true;
5645 return res;
5648 /* Return TRUE if rtx X is considered as an invariant for
5649 inheritance. */
5650 static bool
5651 invariant_p (const_rtx x)
5653 machine_mode mode;
5654 const char *fmt;
5655 enum rtx_code code;
5656 int i, j;
5658 code = GET_CODE (x);
5659 mode = GET_MODE (x);
5660 if (code == SUBREG)
5662 x = SUBREG_REG (x);
5663 code = GET_CODE (x);
5664 mode = wider_subreg_mode (mode, GET_MODE (x));
5667 if (MEM_P (x))
5668 return false;
5670 if (REG_P (x))
5672 int i, nregs, regno = REGNO (x);
5674 if (regno >= FIRST_PSEUDO_REGISTER || regno == STACK_POINTER_REGNUM
5675 || TEST_HARD_REG_BIT (eliminable_regset, regno)
5676 || GET_MODE_CLASS (GET_MODE (x)) == MODE_CC)
5677 return false;
5678 nregs = hard_regno_nregs (regno, mode);
5679 for (i = 0; i < nregs; i++)
5680 if (! fixed_regs[regno + i]
5681 /* A hard register may be clobbered in the current insn
5682 but we can ignore this case because if the hard
5683 register is used it should be set somewhere after the
5684 clobber. */
5685 || bitmap_bit_p (&invalid_invariant_regs, regno + i))
5686 return false;
5688 fmt = GET_RTX_FORMAT (code);
5689 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
5691 if (fmt[i] == 'e')
5693 if (! invariant_p (XEXP (x, i)))
5694 return false;
5696 else if (fmt[i] == 'E')
5698 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
5699 if (! invariant_p (XVECEXP (x, i, j)))
5700 return false;
5703 return true;
5706 /* We have 'dest_reg <- invariant'. Let us try to make an invariant
5707 inheritance transformation (using dest_reg instead invariant in a
5708 subsequent insn). */
5709 static bool
5710 process_invariant_for_inheritance (rtx dst_reg, rtx invariant_rtx)
5712 invariant_ptr_t invariant_ptr;
5713 rtx_insn *insn, *new_insns;
5714 rtx insn_set, insn_reg, new_reg;
5715 int insn_regno;
5716 bool succ_p = false;
5717 int dst_regno = REGNO (dst_reg);
5718 machine_mode dst_mode = GET_MODE (dst_reg);
5719 enum reg_class cl = lra_get_allocno_class (dst_regno), insn_reg_cl;
5721 invariant_ptr = insert_invariant (invariant_rtx);
5722 if ((insn = invariant_ptr->insn) != NULL_RTX)
5724 /* We have a subsequent insn using the invariant. */
5725 insn_set = single_set (insn);
5726 lra_assert (insn_set != NULL);
5727 insn_reg = SET_DEST (insn_set);
5728 lra_assert (REG_P (insn_reg));
5729 insn_regno = REGNO (insn_reg);
5730 insn_reg_cl = lra_get_allocno_class (insn_regno);
5732 if (dst_mode == GET_MODE (insn_reg)
5733 /* We should consider only result move reg insns which are
5734 cheap. */
5735 && targetm.register_move_cost (dst_mode, cl, insn_reg_cl) == 2
5736 && targetm.register_move_cost (dst_mode, cl, cl) == 2)
5738 if (lra_dump_file != NULL)
5739 fprintf (lra_dump_file,
5740 " [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[\n");
5741 new_reg = lra_create_new_reg (dst_mode, dst_reg,
5742 cl, "invariant inheritance");
5743 bitmap_set_bit (&lra_inheritance_pseudos, REGNO (new_reg));
5744 bitmap_set_bit (&check_only_regs, REGNO (new_reg));
5745 lra_reg_info[REGNO (new_reg)].restore_rtx = PATTERN (insn);
5746 start_sequence ();
5747 lra_emit_move (new_reg, dst_reg);
5748 new_insns = get_insns ();
5749 end_sequence ();
5750 lra_process_new_insns (curr_insn, NULL, new_insns,
5751 "Add invariant inheritance<-original");
5752 start_sequence ();
5753 lra_emit_move (SET_DEST (insn_set), new_reg);
5754 new_insns = get_insns ();
5755 end_sequence ();
5756 lra_process_new_insns (insn, NULL, new_insns,
5757 "Changing reload<-inheritance");
5758 lra_set_insn_deleted (insn);
5759 succ_p = true;
5760 if (lra_dump_file != NULL)
5762 fprintf (lra_dump_file,
5763 " Invariant inheritance reuse change %d (bb%d):\n",
5764 REGNO (new_reg), BLOCK_FOR_INSN (insn)->index);
5765 dump_insn_slim (lra_dump_file, insn);
5766 fprintf (lra_dump_file,
5767 " ]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]\n");
5771 invariant_ptr->insn = curr_insn;
5772 return succ_p;
5775 /* Check only registers living at the current program point in the
5776 current EBB. */
5777 static bitmap_head live_regs;
5779 /* Update live info in EBB given by its HEAD and TAIL insns after
5780 inheritance/split transformation. The function removes dead moves
5781 too. */
5782 static void
5783 update_ebb_live_info (rtx_insn *head, rtx_insn *tail)
5785 unsigned int j;
5786 int i, regno;
5787 bool live_p;
5788 rtx_insn *prev_insn;
5789 rtx set;
5790 bool remove_p;
5791 basic_block last_bb, prev_bb, curr_bb;
5792 bitmap_iterator bi;
5793 struct lra_insn_reg *reg;
5794 edge e;
5795 edge_iterator ei;
5797 last_bb = BLOCK_FOR_INSN (tail);
5798 prev_bb = NULL;
5799 for (curr_insn = tail;
5800 curr_insn != PREV_INSN (head);
5801 curr_insn = prev_insn)
5803 prev_insn = PREV_INSN (curr_insn);
5804 /* We need to process empty blocks too. They contain
5805 NOTE_INSN_BASIC_BLOCK referring for the basic block. */
5806 if (NOTE_P (curr_insn) && NOTE_KIND (curr_insn) != NOTE_INSN_BASIC_BLOCK)
5807 continue;
5808 curr_bb = BLOCK_FOR_INSN (curr_insn);
5809 if (curr_bb != prev_bb)
5811 if (prev_bb != NULL)
5813 /* Update df_get_live_in (prev_bb): */
5814 EXECUTE_IF_SET_IN_BITMAP (&check_only_regs, 0, j, bi)
5815 if (bitmap_bit_p (&live_regs, j))
5816 bitmap_set_bit (df_get_live_in (prev_bb), j);
5817 else
5818 bitmap_clear_bit (df_get_live_in (prev_bb), j);
5820 if (curr_bb != last_bb)
5822 /* Update df_get_live_out (curr_bb): */
5823 EXECUTE_IF_SET_IN_BITMAP (&check_only_regs, 0, j, bi)
5825 live_p = bitmap_bit_p (&live_regs, j);
5826 if (! live_p)
5827 FOR_EACH_EDGE (e, ei, curr_bb->succs)
5828 if (bitmap_bit_p (df_get_live_in (e->dest), j))
5830 live_p = true;
5831 break;
5833 if (live_p)
5834 bitmap_set_bit (df_get_live_out (curr_bb), j);
5835 else
5836 bitmap_clear_bit (df_get_live_out (curr_bb), j);
5839 prev_bb = curr_bb;
5840 bitmap_and (&live_regs, &check_only_regs, df_get_live_out (curr_bb));
5842 if (! NONDEBUG_INSN_P (curr_insn))
5843 continue;
5844 curr_id = lra_get_insn_recog_data (curr_insn);
5845 curr_static_id = curr_id->insn_static_data;
5846 remove_p = false;
5847 if ((set = single_set (curr_insn)) != NULL_RTX
5848 && REG_P (SET_DEST (set))
5849 && (regno = REGNO (SET_DEST (set))) >= FIRST_PSEUDO_REGISTER
5850 && SET_DEST (set) != pic_offset_table_rtx
5851 && bitmap_bit_p (&check_only_regs, regno)
5852 && ! bitmap_bit_p (&live_regs, regno))
5853 remove_p = true;
5854 /* See which defined values die here. */
5855 for (reg = curr_id->regs; reg != NULL; reg = reg->next)
5856 if (reg->type == OP_OUT && ! reg->subreg_p)
5857 bitmap_clear_bit (&live_regs, reg->regno);
5858 for (reg = curr_static_id->hard_regs; reg != NULL; reg = reg->next)
5859 if (reg->type == OP_OUT && ! reg->subreg_p)
5860 bitmap_clear_bit (&live_regs, reg->regno);
5861 if (curr_id->arg_hard_regs != NULL)
5862 /* Make clobbered argument hard registers die. */
5863 for (i = 0; (regno = curr_id->arg_hard_regs[i]) >= 0; i++)
5864 if (regno >= FIRST_PSEUDO_REGISTER)
5865 bitmap_clear_bit (&live_regs, regno - FIRST_PSEUDO_REGISTER);
5866 /* Mark each used value as live. */
5867 for (reg = curr_id->regs; reg != NULL; reg = reg->next)
5868 if (reg->type != OP_OUT
5869 && bitmap_bit_p (&check_only_regs, reg->regno))
5870 bitmap_set_bit (&live_regs, reg->regno);
5871 for (reg = curr_static_id->hard_regs; reg != NULL; reg = reg->next)
5872 if (reg->type != OP_OUT
5873 && bitmap_bit_p (&check_only_regs, reg->regno))
5874 bitmap_set_bit (&live_regs, reg->regno);
5875 if (curr_id->arg_hard_regs != NULL)
5876 /* Make used argument hard registers live. */
5877 for (i = 0; (regno = curr_id->arg_hard_regs[i]) >= 0; i++)
5878 if (regno < FIRST_PSEUDO_REGISTER
5879 && bitmap_bit_p (&check_only_regs, regno))
5880 bitmap_set_bit (&live_regs, regno);
5881 /* It is quite important to remove dead move insns because it
5882 means removing dead store. We don't need to process them for
5883 constraints. */
5884 if (remove_p)
5886 if (lra_dump_file != NULL)
5888 fprintf (lra_dump_file, " Removing dead insn:\n ");
5889 dump_insn_slim (lra_dump_file, curr_insn);
5891 lra_set_insn_deleted (curr_insn);
5896 /* The structure describes info to do an inheritance for the current
5897 insn. We need to collect such info first before doing the
5898 transformations because the transformations change the insn
5899 internal representation. */
5900 struct to_inherit
5902 /* Original regno. */
5903 int regno;
5904 /* Subsequent insns which can inherit original reg value. */
5905 rtx insns;
5908 /* Array containing all info for doing inheritance from the current
5909 insn. */
5910 static struct to_inherit to_inherit[LRA_MAX_INSN_RELOADS];
5912 /* Number elements in the previous array. */
5913 static int to_inherit_num;
5915 /* Add inheritance info REGNO and INSNS. Their meaning is described in
5916 structure to_inherit. */
5917 static void
5918 add_to_inherit (int regno, rtx insns)
5920 int i;
5922 for (i = 0; i < to_inherit_num; i++)
5923 if (to_inherit[i].regno == regno)
5924 return;
5925 lra_assert (to_inherit_num < LRA_MAX_INSN_RELOADS);
5926 to_inherit[to_inherit_num].regno = regno;
5927 to_inherit[to_inherit_num++].insns = insns;
5930 /* Return the last non-debug insn in basic block BB, or the block begin
5931 note if none. */
5932 static rtx_insn *
5933 get_last_insertion_point (basic_block bb)
5935 rtx_insn *insn;
5937 FOR_BB_INSNS_REVERSE (bb, insn)
5938 if (NONDEBUG_INSN_P (insn) || NOTE_INSN_BASIC_BLOCK_P (insn))
5939 return insn;
5940 gcc_unreachable ();
5943 /* Set up RES by registers living on edges FROM except the edge (FROM,
5944 TO) or by registers set up in a jump insn in BB FROM. */
5945 static void
5946 get_live_on_other_edges (basic_block from, basic_block to, bitmap res)
5948 rtx_insn *last;
5949 struct lra_insn_reg *reg;
5950 edge e;
5951 edge_iterator ei;
5953 lra_assert (to != NULL);
5954 bitmap_clear (res);
5955 FOR_EACH_EDGE (e, ei, from->succs)
5956 if (e->dest != to)
5957 bitmap_ior_into (res, df_get_live_in (e->dest));
5958 last = get_last_insertion_point (from);
5959 if (! JUMP_P (last))
5960 return;
5961 curr_id = lra_get_insn_recog_data (last);
5962 for (reg = curr_id->regs; reg != NULL; reg = reg->next)
5963 if (reg->type != OP_IN)
5964 bitmap_set_bit (res, reg->regno);
5967 /* Used as a temporary results of some bitmap calculations. */
5968 static bitmap_head temp_bitmap;
5970 /* We split for reloads of small class of hard regs. The following
5971 defines how many hard regs the class should have to be qualified as
5972 small. The code is mostly oriented to x86/x86-64 architecture
5973 where some insns need to use only specific register or pair of
5974 registers and these register can live in RTL explicitly, e.g. for
5975 parameter passing. */
5976 static const int max_small_class_regs_num = 2;
5978 /* Do inheritance/split transformations in EBB starting with HEAD and
5979 finishing on TAIL. We process EBB insns in the reverse order.
5980 Return true if we did any inheritance/split transformation in the
5981 EBB.
5983 We should avoid excessive splitting which results in worse code
5984 because of inaccurate cost calculations for spilling new split
5985 pseudos in such case. To achieve this we do splitting only if
5986 register pressure is high in given basic block and there are reload
5987 pseudos requiring hard registers. We could do more register
5988 pressure calculations at any given program point to avoid necessary
5989 splitting even more but it is to expensive and the current approach
5990 works well enough. */
5991 static bool
5992 inherit_in_ebb (rtx_insn *head, rtx_insn *tail)
5994 int i, src_regno, dst_regno, nregs;
5995 bool change_p, succ_p, update_reloads_num_p;
5996 rtx_insn *prev_insn, *last_insn;
5997 rtx next_usage_insns, curr_set;
5998 enum reg_class cl;
5999 struct lra_insn_reg *reg;
6000 basic_block last_processed_bb, curr_bb = NULL;
6001 HARD_REG_SET potential_reload_hard_regs, live_hard_regs;
6002 bitmap to_process;
6003 unsigned int j;
6004 bitmap_iterator bi;
6005 bool head_p, after_p;
6007 change_p = false;
6008 curr_usage_insns_check++;
6009 clear_invariants ();
6010 reloads_num = calls_num = 0;
6011 bitmap_clear (&check_only_regs);
6012 bitmap_clear (&invalid_invariant_regs);
6013 last_processed_bb = NULL;
6014 CLEAR_HARD_REG_SET (potential_reload_hard_regs);
6015 COPY_HARD_REG_SET (live_hard_regs, eliminable_regset);
6016 IOR_HARD_REG_SET (live_hard_regs, lra_no_alloc_regs);
6017 /* We don't process new insns generated in the loop. */
6018 for (curr_insn = tail; curr_insn != PREV_INSN (head); curr_insn = prev_insn)
6020 prev_insn = PREV_INSN (curr_insn);
6021 if (BLOCK_FOR_INSN (curr_insn) != NULL)
6022 curr_bb = BLOCK_FOR_INSN (curr_insn);
6023 if (last_processed_bb != curr_bb)
6025 /* We are at the end of BB. Add qualified living
6026 pseudos for potential splitting. */
6027 to_process = df_get_live_out (curr_bb);
6028 if (last_processed_bb != NULL)
6030 /* We are somewhere in the middle of EBB. */
6031 get_live_on_other_edges (curr_bb, last_processed_bb,
6032 &temp_bitmap);
6033 to_process = &temp_bitmap;
6035 last_processed_bb = curr_bb;
6036 last_insn = get_last_insertion_point (curr_bb);
6037 after_p = (! JUMP_P (last_insn)
6038 && (! CALL_P (last_insn)
6039 || (find_reg_note (last_insn,
6040 REG_NORETURN, NULL_RTX) == NULL_RTX
6041 && ! SIBLING_CALL_P (last_insn))));
6042 CLEAR_HARD_REG_SET (potential_reload_hard_regs);
6043 EXECUTE_IF_SET_IN_BITMAP (to_process, 0, j, bi)
6045 if ((int) j >= lra_constraint_new_regno_start)
6046 break;
6047 if (j < FIRST_PSEUDO_REGISTER || reg_renumber[j] >= 0)
6049 if (j < FIRST_PSEUDO_REGISTER)
6050 SET_HARD_REG_BIT (live_hard_regs, j);
6051 else
6052 add_to_hard_reg_set (&live_hard_regs,
6053 PSEUDO_REGNO_MODE (j),
6054 reg_renumber[j]);
6055 setup_next_usage_insn (j, last_insn, reloads_num, after_p);
6059 src_regno = dst_regno = -1;
6060 curr_set = single_set (curr_insn);
6061 if (curr_set != NULL_RTX && REG_P (SET_DEST (curr_set)))
6062 dst_regno = REGNO (SET_DEST (curr_set));
6063 if (curr_set != NULL_RTX && REG_P (SET_SRC (curr_set)))
6064 src_regno = REGNO (SET_SRC (curr_set));
6065 update_reloads_num_p = true;
6066 if (src_regno < lra_constraint_new_regno_start
6067 && src_regno >= FIRST_PSEUDO_REGISTER
6068 && reg_renumber[src_regno] < 0
6069 && dst_regno >= lra_constraint_new_regno_start
6070 && (cl = lra_get_allocno_class (dst_regno)) != NO_REGS)
6072 /* 'reload_pseudo <- original_pseudo'. */
6073 if (ira_class_hard_regs_num[cl] <= max_small_class_regs_num)
6074 reloads_num++;
6075 update_reloads_num_p = false;
6076 succ_p = false;
6077 if (usage_insns[src_regno].check == curr_usage_insns_check
6078 && (next_usage_insns = usage_insns[src_regno].insns) != NULL_RTX)
6079 succ_p = inherit_reload_reg (false, src_regno, cl,
6080 curr_insn, next_usage_insns);
6081 if (succ_p)
6082 change_p = true;
6083 else
6084 setup_next_usage_insn (src_regno, curr_insn, reloads_num, false);
6085 if (hard_reg_set_subset_p (reg_class_contents[cl], live_hard_regs))
6086 IOR_HARD_REG_SET (potential_reload_hard_regs,
6087 reg_class_contents[cl]);
6089 else if (src_regno < 0
6090 && dst_regno >= lra_constraint_new_regno_start
6091 && invariant_p (SET_SRC (curr_set))
6092 && (cl = lra_get_allocno_class (dst_regno)) != NO_REGS
6093 && ! bitmap_bit_p (&invalid_invariant_regs, dst_regno)
6094 && ! bitmap_bit_p (&invalid_invariant_regs,
6095 ORIGINAL_REGNO(regno_reg_rtx[dst_regno])))
6097 /* 'reload_pseudo <- invariant'. */
6098 if (ira_class_hard_regs_num[cl] <= max_small_class_regs_num)
6099 reloads_num++;
6100 update_reloads_num_p = false;
6101 if (process_invariant_for_inheritance (SET_DEST (curr_set), SET_SRC (curr_set)))
6102 change_p = true;
6103 if (hard_reg_set_subset_p (reg_class_contents[cl], live_hard_regs))
6104 IOR_HARD_REG_SET (potential_reload_hard_regs,
6105 reg_class_contents[cl]);
6107 else if (src_regno >= lra_constraint_new_regno_start
6108 && dst_regno < lra_constraint_new_regno_start
6109 && dst_regno >= FIRST_PSEUDO_REGISTER
6110 && reg_renumber[dst_regno] < 0
6111 && (cl = lra_get_allocno_class (src_regno)) != NO_REGS
6112 && usage_insns[dst_regno].check == curr_usage_insns_check
6113 && (next_usage_insns
6114 = usage_insns[dst_regno].insns) != NULL_RTX)
6116 if (ira_class_hard_regs_num[cl] <= max_small_class_regs_num)
6117 reloads_num++;
6118 update_reloads_num_p = false;
6119 /* 'original_pseudo <- reload_pseudo'. */
6120 if (! JUMP_P (curr_insn)
6121 && inherit_reload_reg (true, dst_regno, cl,
6122 curr_insn, next_usage_insns))
6123 change_p = true;
6124 /* Invalidate. */
6125 usage_insns[dst_regno].check = 0;
6126 if (hard_reg_set_subset_p (reg_class_contents[cl], live_hard_regs))
6127 IOR_HARD_REG_SET (potential_reload_hard_regs,
6128 reg_class_contents[cl]);
6130 else if (INSN_P (curr_insn))
6132 int iter;
6133 int max_uid = get_max_uid ();
6135 curr_id = lra_get_insn_recog_data (curr_insn);
6136 curr_static_id = curr_id->insn_static_data;
6137 to_inherit_num = 0;
6138 /* Process insn definitions. */
6139 for (iter = 0; iter < 2; iter++)
6140 for (reg = iter == 0 ? curr_id->regs : curr_static_id->hard_regs;
6141 reg != NULL;
6142 reg = reg->next)
6143 if (reg->type != OP_IN
6144 && (dst_regno = reg->regno) < lra_constraint_new_regno_start)
6146 if (dst_regno >= FIRST_PSEUDO_REGISTER && reg->type == OP_OUT
6147 && reg_renumber[dst_regno] < 0 && ! reg->subreg_p
6148 && usage_insns[dst_regno].check == curr_usage_insns_check
6149 && (next_usage_insns
6150 = usage_insns[dst_regno].insns) != NULL_RTX)
6152 struct lra_insn_reg *r;
6154 for (r = curr_id->regs; r != NULL; r = r->next)
6155 if (r->type != OP_OUT && r->regno == dst_regno)
6156 break;
6157 /* Don't do inheritance if the pseudo is also
6158 used in the insn. */
6159 if (r == NULL)
6160 /* We can not do inheritance right now
6161 because the current insn reg info (chain
6162 regs) can change after that. */
6163 add_to_inherit (dst_regno, next_usage_insns);
6165 /* We can not process one reg twice here because of
6166 usage_insns invalidation. */
6167 if ((dst_regno < FIRST_PSEUDO_REGISTER
6168 || reg_renumber[dst_regno] >= 0)
6169 && ! reg->subreg_p && reg->type != OP_IN)
6171 HARD_REG_SET s;
6173 if (split_if_necessary (dst_regno, reg->biggest_mode,
6174 potential_reload_hard_regs,
6175 false, curr_insn, max_uid))
6176 change_p = true;
6177 CLEAR_HARD_REG_SET (s);
6178 if (dst_regno < FIRST_PSEUDO_REGISTER)
6179 add_to_hard_reg_set (&s, reg->biggest_mode, dst_regno);
6180 else
6181 add_to_hard_reg_set (&s, PSEUDO_REGNO_MODE (dst_regno),
6182 reg_renumber[dst_regno]);
6183 AND_COMPL_HARD_REG_SET (live_hard_regs, s);
6185 /* We should invalidate potential inheritance or
6186 splitting for the current insn usages to the next
6187 usage insns (see code below) as the output pseudo
6188 prevents this. */
6189 if ((dst_regno >= FIRST_PSEUDO_REGISTER
6190 && reg_renumber[dst_regno] < 0)
6191 || (reg->type == OP_OUT && ! reg->subreg_p
6192 && (dst_regno < FIRST_PSEUDO_REGISTER
6193 || reg_renumber[dst_regno] >= 0)))
6195 /* Invalidate and mark definitions. */
6196 if (dst_regno >= FIRST_PSEUDO_REGISTER)
6197 usage_insns[dst_regno].check = -(int) INSN_UID (curr_insn);
6198 else
6200 nregs = hard_regno_nregs (dst_regno,
6201 reg->biggest_mode);
6202 for (i = 0; i < nregs; i++)
6203 usage_insns[dst_regno + i].check
6204 = -(int) INSN_UID (curr_insn);
6208 /* Process clobbered call regs. */
6209 if (curr_id->arg_hard_regs != NULL)
6210 for (i = 0; (dst_regno = curr_id->arg_hard_regs[i]) >= 0; i++)
6211 if (dst_regno >= FIRST_PSEUDO_REGISTER)
6212 usage_insns[dst_regno - FIRST_PSEUDO_REGISTER].check
6213 = -(int) INSN_UID (curr_insn);
6214 if (! JUMP_P (curr_insn))
6215 for (i = 0; i < to_inherit_num; i++)
6216 if (inherit_reload_reg (true, to_inherit[i].regno,
6217 ALL_REGS, curr_insn,
6218 to_inherit[i].insns))
6219 change_p = true;
6220 if (CALL_P (curr_insn))
6222 rtx cheap, pat, dest;
6223 rtx_insn *restore;
6224 int regno, hard_regno;
6226 calls_num++;
6227 if ((cheap = find_reg_note (curr_insn,
6228 REG_RETURNED, NULL_RTX)) != NULL_RTX
6229 && ((cheap = XEXP (cheap, 0)), true)
6230 && (regno = REGNO (cheap)) >= FIRST_PSEUDO_REGISTER
6231 && (hard_regno = reg_renumber[regno]) >= 0
6232 && usage_insns[regno].check == curr_usage_insns_check
6233 /* If there are pending saves/restores, the
6234 optimization is not worth. */
6235 && usage_insns[regno].calls_num == calls_num - 1
6236 && TEST_HARD_REG_BIT (call_used_reg_set, hard_regno))
6238 /* Restore the pseudo from the call result as
6239 REG_RETURNED note says that the pseudo value is
6240 in the call result and the pseudo is an argument
6241 of the call. */
6242 pat = PATTERN (curr_insn);
6243 if (GET_CODE (pat) == PARALLEL)
6244 pat = XVECEXP (pat, 0, 0);
6245 dest = SET_DEST (pat);
6246 /* For multiple return values dest is PARALLEL.
6247 Currently we handle only single return value case. */
6248 if (REG_P (dest))
6250 start_sequence ();
6251 emit_move_insn (cheap, copy_rtx (dest));
6252 restore = get_insns ();
6253 end_sequence ();
6254 lra_process_new_insns (curr_insn, NULL, restore,
6255 "Inserting call parameter restore");
6256 /* We don't need to save/restore of the pseudo from
6257 this call. */
6258 usage_insns[regno].calls_num = calls_num;
6259 bitmap_set_bit (&check_only_regs, regno);
6263 to_inherit_num = 0;
6264 /* Process insn usages. */
6265 for (iter = 0; iter < 2; iter++)
6266 for (reg = iter == 0 ? curr_id->regs : curr_static_id->hard_regs;
6267 reg != NULL;
6268 reg = reg->next)
6269 if ((reg->type != OP_OUT
6270 || (reg->type == OP_OUT && reg->subreg_p))
6271 && (src_regno = reg->regno) < lra_constraint_new_regno_start)
6273 if (src_regno >= FIRST_PSEUDO_REGISTER
6274 && reg_renumber[src_regno] < 0 && reg->type == OP_IN)
6276 if (usage_insns[src_regno].check == curr_usage_insns_check
6277 && (next_usage_insns
6278 = usage_insns[src_regno].insns) != NULL_RTX
6279 && NONDEBUG_INSN_P (curr_insn))
6280 add_to_inherit (src_regno, next_usage_insns);
6281 else if (usage_insns[src_regno].check
6282 != -(int) INSN_UID (curr_insn))
6283 /* Add usages but only if the reg is not set up
6284 in the same insn. */
6285 add_next_usage_insn (src_regno, curr_insn, reloads_num);
6287 else if (src_regno < FIRST_PSEUDO_REGISTER
6288 || reg_renumber[src_regno] >= 0)
6290 bool before_p;
6291 rtx_insn *use_insn = curr_insn;
6293 before_p = (JUMP_P (curr_insn)
6294 || (CALL_P (curr_insn) && reg->type == OP_IN));
6295 if (NONDEBUG_INSN_P (curr_insn)
6296 && (! JUMP_P (curr_insn) || reg->type == OP_IN)
6297 && split_if_necessary (src_regno, reg->biggest_mode,
6298 potential_reload_hard_regs,
6299 before_p, curr_insn, max_uid))
6301 if (reg->subreg_p)
6302 lra_risky_transformations_p = true;
6303 change_p = true;
6304 /* Invalidate. */
6305 usage_insns[src_regno].check = 0;
6306 if (before_p)
6307 use_insn = PREV_INSN (curr_insn);
6309 if (NONDEBUG_INSN_P (curr_insn))
6311 if (src_regno < FIRST_PSEUDO_REGISTER)
6312 add_to_hard_reg_set (&live_hard_regs,
6313 reg->biggest_mode, src_regno);
6314 else
6315 add_to_hard_reg_set (&live_hard_regs,
6316 PSEUDO_REGNO_MODE (src_regno),
6317 reg_renumber[src_regno]);
6319 add_next_usage_insn (src_regno, use_insn, reloads_num);
6322 /* Process used call regs. */
6323 if (curr_id->arg_hard_regs != NULL)
6324 for (i = 0; (src_regno = curr_id->arg_hard_regs[i]) >= 0; i++)
6325 if (src_regno < FIRST_PSEUDO_REGISTER)
6327 SET_HARD_REG_BIT (live_hard_regs, src_regno);
6328 add_next_usage_insn (src_regno, curr_insn, reloads_num);
6330 for (i = 0; i < to_inherit_num; i++)
6332 src_regno = to_inherit[i].regno;
6333 if (inherit_reload_reg (false, src_regno, ALL_REGS,
6334 curr_insn, to_inherit[i].insns))
6335 change_p = true;
6336 else
6337 setup_next_usage_insn (src_regno, curr_insn, reloads_num, false);
6340 if (update_reloads_num_p
6341 && NONDEBUG_INSN_P (curr_insn) && curr_set != NULL_RTX)
6343 int regno = -1;
6344 if ((REG_P (SET_DEST (curr_set))
6345 && (regno = REGNO (SET_DEST (curr_set))) >= lra_constraint_new_regno_start
6346 && reg_renumber[regno] < 0
6347 && (cl = lra_get_allocno_class (regno)) != NO_REGS)
6348 || (REG_P (SET_SRC (curr_set))
6349 && (regno = REGNO (SET_SRC (curr_set))) >= lra_constraint_new_regno_start
6350 && reg_renumber[regno] < 0
6351 && (cl = lra_get_allocno_class (regno)) != NO_REGS))
6353 if (ira_class_hard_regs_num[cl] <= max_small_class_regs_num)
6354 reloads_num++;
6355 if (hard_reg_set_subset_p (reg_class_contents[cl], live_hard_regs))
6356 IOR_HARD_REG_SET (potential_reload_hard_regs,
6357 reg_class_contents[cl]);
6360 if (NONDEBUG_INSN_P (curr_insn))
6362 int regno;
6364 /* Invalidate invariants with changed regs. */
6365 curr_id = lra_get_insn_recog_data (curr_insn);
6366 for (reg = curr_id->regs; reg != NULL; reg = reg->next)
6367 if (reg->type != OP_IN)
6369 bitmap_set_bit (&invalid_invariant_regs, reg->regno);
6370 bitmap_set_bit (&invalid_invariant_regs,
6371 ORIGINAL_REGNO (regno_reg_rtx[reg->regno]));
6373 curr_static_id = curr_id->insn_static_data;
6374 for (reg = curr_static_id->hard_regs; reg != NULL; reg = reg->next)
6375 if (reg->type != OP_IN)
6376 bitmap_set_bit (&invalid_invariant_regs, reg->regno);
6377 if (curr_id->arg_hard_regs != NULL)
6378 for (i = 0; (regno = curr_id->arg_hard_regs[i]) >= 0; i++)
6379 if (regno >= FIRST_PSEUDO_REGISTER)
6380 bitmap_set_bit (&invalid_invariant_regs,
6381 regno - FIRST_PSEUDO_REGISTER);
6383 /* We reached the start of the current basic block. */
6384 if (prev_insn == NULL_RTX || prev_insn == PREV_INSN (head)
6385 || BLOCK_FOR_INSN (prev_insn) != curr_bb)
6387 /* We reached the beginning of the current block -- do
6388 rest of spliting in the current BB. */
6389 to_process = df_get_live_in (curr_bb);
6390 if (BLOCK_FOR_INSN (head) != curr_bb)
6392 /* We are somewhere in the middle of EBB. */
6393 get_live_on_other_edges (EDGE_PRED (curr_bb, 0)->src,
6394 curr_bb, &temp_bitmap);
6395 to_process = &temp_bitmap;
6397 head_p = true;
6398 EXECUTE_IF_SET_IN_BITMAP (to_process, 0, j, bi)
6400 if ((int) j >= lra_constraint_new_regno_start)
6401 break;
6402 if (((int) j < FIRST_PSEUDO_REGISTER || reg_renumber[j] >= 0)
6403 && usage_insns[j].check == curr_usage_insns_check
6404 && (next_usage_insns = usage_insns[j].insns) != NULL_RTX)
6406 if (need_for_split_p (potential_reload_hard_regs, j))
6408 if (lra_dump_file != NULL && head_p)
6410 fprintf (lra_dump_file,
6411 " ----------------------------------\n");
6412 head_p = false;
6414 if (split_reg (false, j, bb_note (curr_bb),
6415 next_usage_insns))
6416 change_p = true;
6418 usage_insns[j].check = 0;
6423 return change_p;
6426 /* This value affects EBB forming. If probability of edge from EBB to
6427 a BB is not greater than the following value, we don't add the BB
6428 to EBB. */
6429 #define EBB_PROBABILITY_CUTOFF \
6430 ((REG_BR_PROB_BASE * LRA_INHERITANCE_EBB_PROBABILITY_CUTOFF) / 100)
6432 /* Current number of inheritance/split iteration. */
6433 int lra_inheritance_iter;
6435 /* Entry function for inheritance/split pass. */
6436 void
6437 lra_inheritance (void)
6439 int i;
6440 basic_block bb, start_bb;
6441 edge e;
6443 lra_inheritance_iter++;
6444 if (lra_inheritance_iter > LRA_MAX_INHERITANCE_PASSES)
6445 return;
6446 timevar_push (TV_LRA_INHERITANCE);
6447 if (lra_dump_file != NULL)
6448 fprintf (lra_dump_file, "\n********** Inheritance #%d: **********\n\n",
6449 lra_inheritance_iter);
6450 curr_usage_insns_check = 0;
6451 usage_insns = XNEWVEC (struct usage_insns, lra_constraint_new_regno_start);
6452 for (i = 0; i < lra_constraint_new_regno_start; i++)
6453 usage_insns[i].check = 0;
6454 bitmap_initialize (&check_only_regs, &reg_obstack);
6455 bitmap_initialize (&invalid_invariant_regs, &reg_obstack);
6456 bitmap_initialize (&live_regs, &reg_obstack);
6457 bitmap_initialize (&temp_bitmap, &reg_obstack);
6458 bitmap_initialize (&ebb_global_regs, &reg_obstack);
6459 FOR_EACH_BB_FN (bb, cfun)
6461 start_bb = bb;
6462 if (lra_dump_file != NULL)
6463 fprintf (lra_dump_file, "EBB");
6464 /* Form a EBB starting with BB. */
6465 bitmap_clear (&ebb_global_regs);
6466 bitmap_ior_into (&ebb_global_regs, df_get_live_in (bb));
6467 for (;;)
6469 if (lra_dump_file != NULL)
6470 fprintf (lra_dump_file, " %d", bb->index);
6471 if (bb->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)
6472 || LABEL_P (BB_HEAD (bb->next_bb)))
6473 break;
6474 e = find_fallthru_edge (bb->succs);
6475 if (! e)
6476 break;
6477 if (e->probability.initialized_p ()
6478 && e->probability.to_reg_br_prob_base () < EBB_PROBABILITY_CUTOFF)
6479 break;
6480 bb = bb->next_bb;
6482 bitmap_ior_into (&ebb_global_regs, df_get_live_out (bb));
6483 if (lra_dump_file != NULL)
6484 fprintf (lra_dump_file, "\n");
6485 if (inherit_in_ebb (BB_HEAD (start_bb), BB_END (bb)))
6486 /* Remember that the EBB head and tail can change in
6487 inherit_in_ebb. */
6488 update_ebb_live_info (BB_HEAD (start_bb), BB_END (bb));
6490 bitmap_clear (&ebb_global_regs);
6491 bitmap_clear (&temp_bitmap);
6492 bitmap_clear (&live_regs);
6493 bitmap_clear (&invalid_invariant_regs);
6494 bitmap_clear (&check_only_regs);
6495 free (usage_insns);
6497 timevar_pop (TV_LRA_INHERITANCE);
6502 /* This page contains code to undo failed inheritance/split
6503 transformations. */
6505 /* Current number of iteration undoing inheritance/split. */
6506 int lra_undo_inheritance_iter;
6508 /* Fix BB live info LIVE after removing pseudos created on pass doing
6509 inheritance/split which are REMOVED_PSEUDOS. */
6510 static void
6511 fix_bb_live_info (bitmap live, bitmap removed_pseudos)
6513 unsigned int regno;
6514 bitmap_iterator bi;
6516 EXECUTE_IF_SET_IN_BITMAP (removed_pseudos, 0, regno, bi)
6517 if (bitmap_clear_bit (live, regno)
6518 && REG_P (lra_reg_info[regno].restore_rtx))
6519 bitmap_set_bit (live, REGNO (lra_reg_info[regno].restore_rtx));
6522 /* Return regno of the (subreg of) REG. Otherwise, return a negative
6523 number. */
6524 static int
6525 get_regno (rtx reg)
6527 if (GET_CODE (reg) == SUBREG)
6528 reg = SUBREG_REG (reg);
6529 if (REG_P (reg))
6530 return REGNO (reg);
6531 return -1;
6534 /* Delete a move INSN with destination reg DREGNO and a previous
6535 clobber insn with the same regno. The inheritance/split code can
6536 generate moves with preceding clobber and when we delete such moves
6537 we should delete the clobber insn too to keep the correct life
6538 info. */
6539 static void
6540 delete_move_and_clobber (rtx_insn *insn, int dregno)
6542 rtx_insn *prev_insn = PREV_INSN (insn);
6544 lra_set_insn_deleted (insn);
6545 lra_assert (dregno >= 0);
6546 if (prev_insn != NULL && NONDEBUG_INSN_P (prev_insn)
6547 && GET_CODE (PATTERN (prev_insn)) == CLOBBER
6548 && dregno == get_regno (XEXP (PATTERN (prev_insn), 0)))
6549 lra_set_insn_deleted (prev_insn);
6552 /* Remove inheritance/split pseudos which are in REMOVE_PSEUDOS and
6553 return true if we did any change. The undo transformations for
6554 inheritance looks like
6555 i <- i2
6556 p <- i => p <- i2
6557 or removing
6558 p <- i, i <- p, and i <- i3
6559 where p is original pseudo from which inheritance pseudo i was
6560 created, i and i3 are removed inheritance pseudos, i2 is another
6561 not removed inheritance pseudo. All split pseudos or other
6562 occurrences of removed inheritance pseudos are changed on the
6563 corresponding original pseudos.
6565 The function also schedules insns changed and created during
6566 inheritance/split pass for processing by the subsequent constraint
6567 pass. */
6568 static bool
6569 remove_inheritance_pseudos (bitmap remove_pseudos)
6571 basic_block bb;
6572 int regno, sregno, prev_sregno, dregno;
6573 rtx restore_rtx;
6574 rtx set, prev_set;
6575 rtx_insn *prev_insn;
6576 bool change_p, done_p;
6578 change_p = ! bitmap_empty_p (remove_pseudos);
6579 /* We can not finish the function right away if CHANGE_P is true
6580 because we need to marks insns affected by previous
6581 inheritance/split pass for processing by the subsequent
6582 constraint pass. */
6583 FOR_EACH_BB_FN (bb, cfun)
6585 fix_bb_live_info (df_get_live_in (bb), remove_pseudos);
6586 fix_bb_live_info (df_get_live_out (bb), remove_pseudos);
6587 FOR_BB_INSNS_REVERSE (bb, curr_insn)
6589 if (! INSN_P (curr_insn))
6590 continue;
6591 done_p = false;
6592 sregno = dregno = -1;
6593 if (change_p && NONDEBUG_INSN_P (curr_insn)
6594 && (set = single_set (curr_insn)) != NULL_RTX)
6596 dregno = get_regno (SET_DEST (set));
6597 sregno = get_regno (SET_SRC (set));
6600 if (sregno >= 0 && dregno >= 0)
6602 if (bitmap_bit_p (remove_pseudos, dregno)
6603 && ! REG_P (lra_reg_info[dregno].restore_rtx))
6605 /* invariant inheritance pseudo <- original pseudo */
6606 if (lra_dump_file != NULL)
6608 fprintf (lra_dump_file, " Removing invariant inheritance:\n");
6609 dump_insn_slim (lra_dump_file, curr_insn);
6610 fprintf (lra_dump_file, "\n");
6612 delete_move_and_clobber (curr_insn, dregno);
6613 done_p = true;
6615 else if (bitmap_bit_p (remove_pseudos, sregno)
6616 && ! REG_P (lra_reg_info[sregno].restore_rtx))
6618 /* reload pseudo <- invariant inheritance pseudo */
6619 start_sequence ();
6620 /* We can not just change the source. It might be
6621 an insn different from the move. */
6622 emit_insn (lra_reg_info[sregno].restore_rtx);
6623 rtx_insn *new_insns = get_insns ();
6624 end_sequence ();
6625 lra_assert (single_set (new_insns) != NULL
6626 && SET_DEST (set) == SET_DEST (single_set (new_insns)));
6627 lra_process_new_insns (curr_insn, NULL, new_insns,
6628 "Changing reload<-invariant inheritance");
6629 delete_move_and_clobber (curr_insn, dregno);
6630 done_p = true;
6632 else if ((bitmap_bit_p (remove_pseudos, sregno)
6633 && (get_regno (lra_reg_info[sregno].restore_rtx) == dregno
6634 || (bitmap_bit_p (remove_pseudos, dregno)
6635 && get_regno (lra_reg_info[sregno].restore_rtx) >= 0
6636 && (get_regno (lra_reg_info[sregno].restore_rtx)
6637 == get_regno (lra_reg_info[dregno].restore_rtx)))))
6638 || (bitmap_bit_p (remove_pseudos, dregno)
6639 && get_regno (lra_reg_info[dregno].restore_rtx) == sregno))
6640 /* One of the following cases:
6641 original <- removed inheritance pseudo
6642 removed inherit pseudo <- another removed inherit pseudo
6643 removed inherit pseudo <- original pseudo
6645 removed_split_pseudo <- original_reg
6646 original_reg <- removed_split_pseudo */
6648 if (lra_dump_file != NULL)
6650 fprintf (lra_dump_file, " Removing %s:\n",
6651 bitmap_bit_p (&lra_split_regs, sregno)
6652 || bitmap_bit_p (&lra_split_regs, dregno)
6653 ? "split" : "inheritance");
6654 dump_insn_slim (lra_dump_file, curr_insn);
6656 delete_move_and_clobber (curr_insn, dregno);
6657 done_p = true;
6659 else if (bitmap_bit_p (remove_pseudos, sregno)
6660 && bitmap_bit_p (&lra_inheritance_pseudos, sregno))
6662 /* Search the following pattern:
6663 inherit_or_split_pseudo1 <- inherit_or_split_pseudo2
6664 original_pseudo <- inherit_or_split_pseudo1
6665 where the 2nd insn is the current insn and
6666 inherit_or_split_pseudo2 is not removed. If it is found,
6667 change the current insn onto:
6668 original_pseudo <- inherit_or_split_pseudo2. */
6669 for (prev_insn = PREV_INSN (curr_insn);
6670 prev_insn != NULL_RTX && ! NONDEBUG_INSN_P (prev_insn);
6671 prev_insn = PREV_INSN (prev_insn))
6673 if (prev_insn != NULL_RTX && BLOCK_FOR_INSN (prev_insn) == bb
6674 && (prev_set = single_set (prev_insn)) != NULL_RTX
6675 /* There should be no subregs in insn we are
6676 searching because only the original reg might
6677 be in subreg when we changed the mode of
6678 load/store for splitting. */
6679 && REG_P (SET_DEST (prev_set))
6680 && REG_P (SET_SRC (prev_set))
6681 && (int) REGNO (SET_DEST (prev_set)) == sregno
6682 && ((prev_sregno = REGNO (SET_SRC (prev_set)))
6683 >= FIRST_PSEUDO_REGISTER)
6684 && (lra_reg_info[prev_sregno].restore_rtx == NULL_RTX
6686 /* As we consider chain of inheritance or
6687 splitting described in above comment we should
6688 check that sregno and prev_sregno were
6689 inheritance/split pseudos created from the
6690 same original regno. */
6691 (get_regno (lra_reg_info[sregno].restore_rtx) >= 0
6692 && (get_regno (lra_reg_info[sregno].restore_rtx)
6693 == get_regno (lra_reg_info[prev_sregno].restore_rtx))))
6694 && ! bitmap_bit_p (remove_pseudos, prev_sregno))
6696 lra_assert (GET_MODE (SET_SRC (prev_set))
6697 == GET_MODE (regno_reg_rtx[sregno]));
6698 if (GET_CODE (SET_SRC (set)) == SUBREG)
6699 SUBREG_REG (SET_SRC (set)) = SET_SRC (prev_set);
6700 else
6701 SET_SRC (set) = SET_SRC (prev_set);
6702 /* As we are finishing with processing the insn
6703 here, check the destination too as it might
6704 inheritance pseudo for another pseudo. */
6705 if (bitmap_bit_p (remove_pseudos, dregno)
6706 && bitmap_bit_p (&lra_inheritance_pseudos, dregno)
6707 && (restore_rtx
6708 = lra_reg_info[dregno].restore_rtx) != NULL_RTX)
6710 if (GET_CODE (SET_DEST (set)) == SUBREG)
6711 SUBREG_REG (SET_DEST (set)) = restore_rtx;
6712 else
6713 SET_DEST (set) = restore_rtx;
6715 lra_push_insn_and_update_insn_regno_info (curr_insn);
6716 lra_set_used_insn_alternative_by_uid
6717 (INSN_UID (curr_insn), -1);
6718 done_p = true;
6719 if (lra_dump_file != NULL)
6721 fprintf (lra_dump_file, " Change reload insn:\n");
6722 dump_insn_slim (lra_dump_file, curr_insn);
6727 if (! done_p)
6729 struct lra_insn_reg *reg;
6730 bool restored_regs_p = false;
6731 bool kept_regs_p = false;
6733 curr_id = lra_get_insn_recog_data (curr_insn);
6734 for (reg = curr_id->regs; reg != NULL; reg = reg->next)
6736 regno = reg->regno;
6737 restore_rtx = lra_reg_info[regno].restore_rtx;
6738 if (restore_rtx != NULL_RTX)
6740 if (change_p && bitmap_bit_p (remove_pseudos, regno))
6742 lra_substitute_pseudo_within_insn
6743 (curr_insn, regno, restore_rtx, false);
6744 restored_regs_p = true;
6746 else
6747 kept_regs_p = true;
6750 if (NONDEBUG_INSN_P (curr_insn) && kept_regs_p)
6752 /* The instruction has changed since the previous
6753 constraints pass. */
6754 lra_push_insn_and_update_insn_regno_info (curr_insn);
6755 lra_set_used_insn_alternative_by_uid
6756 (INSN_UID (curr_insn), -1);
6758 else if (restored_regs_p)
6759 /* The instruction has been restored to the form that
6760 it had during the previous constraints pass. */
6761 lra_update_insn_regno_info (curr_insn);
6762 if (restored_regs_p && lra_dump_file != NULL)
6764 fprintf (lra_dump_file, " Insn after restoring regs:\n");
6765 dump_insn_slim (lra_dump_file, curr_insn);
6770 return change_p;
6773 /* If optional reload pseudos failed to get a hard register or was not
6774 inherited, it is better to remove optional reloads. We do this
6775 transformation after undoing inheritance to figure out necessity to
6776 remove optional reloads easier. Return true if we do any
6777 change. */
6778 static bool
6779 undo_optional_reloads (void)
6781 bool change_p, keep_p;
6782 unsigned int regno, uid;
6783 bitmap_iterator bi, bi2;
6784 rtx_insn *insn;
6785 rtx set, src, dest;
6786 auto_bitmap removed_optional_reload_pseudos (&reg_obstack);
6788 bitmap_copy (removed_optional_reload_pseudos, &lra_optional_reload_pseudos);
6789 EXECUTE_IF_SET_IN_BITMAP (&lra_optional_reload_pseudos, 0, regno, bi)
6791 keep_p = false;
6792 /* Keep optional reloads from previous subpasses. */
6793 if (lra_reg_info[regno].restore_rtx == NULL_RTX
6794 /* If the original pseudo changed its allocation, just
6795 removing the optional pseudo is dangerous as the original
6796 pseudo will have longer live range. */
6797 || reg_renumber[REGNO (lra_reg_info[regno].restore_rtx)] >= 0)
6798 keep_p = true;
6799 else if (reg_renumber[regno] >= 0)
6800 EXECUTE_IF_SET_IN_BITMAP (&lra_reg_info[regno].insn_bitmap, 0, uid, bi2)
6802 insn = lra_insn_recog_data[uid]->insn;
6803 if ((set = single_set (insn)) == NULL_RTX)
6804 continue;
6805 src = SET_SRC (set);
6806 dest = SET_DEST (set);
6807 if (! REG_P (src) || ! REG_P (dest))
6808 continue;
6809 if (REGNO (dest) == regno
6810 /* Ignore insn for optional reloads itself. */
6811 && REGNO (lra_reg_info[regno].restore_rtx) != REGNO (src)
6812 /* Check only inheritance on last inheritance pass. */
6813 && (int) REGNO (src) >= new_regno_start
6814 /* Check that the optional reload was inherited. */
6815 && bitmap_bit_p (&lra_inheritance_pseudos, REGNO (src)))
6817 keep_p = true;
6818 break;
6821 if (keep_p)
6823 bitmap_clear_bit (removed_optional_reload_pseudos, regno);
6824 if (lra_dump_file != NULL)
6825 fprintf (lra_dump_file, "Keep optional reload reg %d\n", regno);
6828 change_p = ! bitmap_empty_p (removed_optional_reload_pseudos);
6829 auto_bitmap insn_bitmap (&reg_obstack);
6830 EXECUTE_IF_SET_IN_BITMAP (removed_optional_reload_pseudos, 0, regno, bi)
6832 if (lra_dump_file != NULL)
6833 fprintf (lra_dump_file, "Remove optional reload reg %d\n", regno);
6834 bitmap_copy (insn_bitmap, &lra_reg_info[regno].insn_bitmap);
6835 EXECUTE_IF_SET_IN_BITMAP (insn_bitmap, 0, uid, bi2)
6837 insn = lra_insn_recog_data[uid]->insn;
6838 if ((set = single_set (insn)) != NULL_RTX)
6840 src = SET_SRC (set);
6841 dest = SET_DEST (set);
6842 if (REG_P (src) && REG_P (dest)
6843 && ((REGNO (src) == regno
6844 && (REGNO (lra_reg_info[regno].restore_rtx)
6845 == REGNO (dest)))
6846 || (REGNO (dest) == regno
6847 && (REGNO (lra_reg_info[regno].restore_rtx)
6848 == REGNO (src)))))
6850 if (lra_dump_file != NULL)
6852 fprintf (lra_dump_file, " Deleting move %u\n",
6853 INSN_UID (insn));
6854 dump_insn_slim (lra_dump_file, insn);
6856 delete_move_and_clobber (insn, REGNO (dest));
6857 continue;
6859 /* We should not worry about generation memory-memory
6860 moves here as if the corresponding inheritance did
6861 not work (inheritance pseudo did not get a hard reg),
6862 we remove the inheritance pseudo and the optional
6863 reload. */
6865 lra_substitute_pseudo_within_insn
6866 (insn, regno, lra_reg_info[regno].restore_rtx, false);
6867 lra_update_insn_regno_info (insn);
6868 if (lra_dump_file != NULL)
6870 fprintf (lra_dump_file,
6871 " Restoring original insn:\n");
6872 dump_insn_slim (lra_dump_file, insn);
6876 /* Clear restore_regnos. */
6877 EXECUTE_IF_SET_IN_BITMAP (&lra_optional_reload_pseudos, 0, regno, bi)
6878 lra_reg_info[regno].restore_rtx = NULL_RTX;
6879 return change_p;
6882 /* Entry function for undoing inheritance/split transformation. Return true
6883 if we did any RTL change in this pass. */
6884 bool
6885 lra_undo_inheritance (void)
6887 unsigned int regno;
6888 int hard_regno;
6889 int n_all_inherit, n_inherit, n_all_split, n_split;
6890 rtx restore_rtx;
6891 bitmap_iterator bi;
6892 bool change_p;
6894 lra_undo_inheritance_iter++;
6895 if (lra_undo_inheritance_iter > LRA_MAX_INHERITANCE_PASSES)
6896 return false;
6897 if (lra_dump_file != NULL)
6898 fprintf (lra_dump_file,
6899 "\n********** Undoing inheritance #%d: **********\n\n",
6900 lra_undo_inheritance_iter);
6901 auto_bitmap remove_pseudos (&reg_obstack);
6902 n_inherit = n_all_inherit = 0;
6903 EXECUTE_IF_SET_IN_BITMAP (&lra_inheritance_pseudos, 0, regno, bi)
6904 if (lra_reg_info[regno].restore_rtx != NULL_RTX)
6906 n_all_inherit++;
6907 if (reg_renumber[regno] < 0
6908 /* If the original pseudo changed its allocation, just
6909 removing inheritance is dangerous as for changing
6910 allocation we used shorter live-ranges. */
6911 && (! REG_P (lra_reg_info[regno].restore_rtx)
6912 || reg_renumber[REGNO (lra_reg_info[regno].restore_rtx)] < 0))
6913 bitmap_set_bit (remove_pseudos, regno);
6914 else
6915 n_inherit++;
6917 if (lra_dump_file != NULL && n_all_inherit != 0)
6918 fprintf (lra_dump_file, "Inherit %d out of %d (%.2f%%)\n",
6919 n_inherit, n_all_inherit,
6920 (double) n_inherit / n_all_inherit * 100);
6921 n_split = n_all_split = 0;
6922 EXECUTE_IF_SET_IN_BITMAP (&lra_split_regs, 0, regno, bi)
6923 if ((restore_rtx = lra_reg_info[regno].restore_rtx) != NULL_RTX)
6925 int restore_regno = REGNO (restore_rtx);
6927 n_all_split++;
6928 hard_regno = (restore_regno >= FIRST_PSEUDO_REGISTER
6929 ? reg_renumber[restore_regno] : restore_regno);
6930 if (hard_regno < 0 || reg_renumber[regno] == hard_regno)
6931 bitmap_set_bit (remove_pseudos, regno);
6932 else
6934 n_split++;
6935 if (lra_dump_file != NULL)
6936 fprintf (lra_dump_file, " Keep split r%d (orig=r%d)\n",
6937 regno, restore_regno);
6940 if (lra_dump_file != NULL && n_all_split != 0)
6941 fprintf (lra_dump_file, "Split %d out of %d (%.2f%%)\n",
6942 n_split, n_all_split,
6943 (double) n_split / n_all_split * 100);
6944 change_p = remove_inheritance_pseudos (remove_pseudos);
6945 /* Clear restore_regnos. */
6946 EXECUTE_IF_SET_IN_BITMAP (&lra_inheritance_pseudos, 0, regno, bi)
6947 lra_reg_info[regno].restore_rtx = NULL_RTX;
6948 EXECUTE_IF_SET_IN_BITMAP (&lra_split_regs, 0, regno, bi)
6949 lra_reg_info[regno].restore_rtx = NULL_RTX;
6950 change_p = undo_optional_reloads () || change_p;
6951 return change_p;