Line data Source code
1 : /* Optimize by combining instructions for GNU compiler.
2 : Copyright (C) 1987-2026 Free Software Foundation, Inc.
3 :
4 : This file is part of GCC.
5 :
6 : GCC is free software; you can redistribute it and/or modify it under
7 : the terms of the GNU General Public License as published by the Free
8 : Software Foundation; either version 3, or (at your option) any later
9 : version.
10 :
11 : GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 : WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 : FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 : for more details.
15 :
16 : You should have received a copy of the GNU General Public License
17 : along with GCC; see the file COPYING3. If not see
18 : <http://www.gnu.org/licenses/>. */
19 :
20 : /* This module is essentially the "combiner" phase of the U. of Arizona
21 : Portable Optimizer, but redone to work on our list-structured
22 : representation for RTL instead of their string representation.
23 :
24 : The LOG_LINKS of each insn identify the most recent assignment
25 : to each REG used in the insn. It is a list of previous insns,
26 : each of which contains a SET for a REG that is used in this insn
27 : and not used or set in between. LOG_LINKs never cross basic blocks.
28 : They were set up by the preceding pass (lifetime analysis).
29 :
30 : We try to combine each pair of insns joined by a logical link.
31 : We also try to combine triplets of insns A, B and C when C has
32 : a link back to B and B has a link back to A. Likewise for a
33 : small number of quadruplets of insns A, B, C and D for which
34 : there's high likelihood of success.
35 :
36 : We check (with modified_between_p) to avoid combining in such a way
37 : as to move a computation to a place where its value would be different.
38 :
39 : Combination is done by mathematically substituting the previous
40 : insn(s) values for the regs they set into the expressions in
41 : the later insns that refer to these regs. If the result is a valid insn
42 : for our target machine, according to the machine description,
43 : we install it, delete the earlier insns, and update the data flow
44 : information (LOG_LINKS and REG_NOTES) for what we did.
45 :
46 : There are a few exceptions where the dataflow information isn't
47 : completely updated (however this is only a local issue since it is
48 : regenerated before the next pass that uses it):
49 :
50 : - reg_live_length is not updated
51 : - reg_n_refs is not adjusted in the rare case when a register is
52 : no longer required in a computation
53 : - there are extremely rare cases (see distribute_notes) when a
54 : REG_DEAD note is lost
55 : - a LOG_LINKS entry that refers to an insn with multiple SETs may be
56 : removed because there is no way to know which register it was
57 : linking
58 :
59 : To simplify substitution, we combine only when the earlier insn(s)
60 : consist of only a single assignment. To simplify updating afterward,
61 : we never combine when a subroutine call appears in the middle. */
62 :
63 : #include "config.h"
64 : #include "system.h"
65 : #include "coretypes.h"
66 : #include "backend.h"
67 : #include "target.h"
68 : #include "rtl.h"
69 : #include "tree.h"
70 : #include "cfghooks.h"
71 : #include "predict.h"
72 : #include "df.h"
73 : #include "memmodel.h"
74 : #include "tm_p.h"
75 : #include "optabs.h"
76 : #include "regs.h"
77 : #include "emit-rtl.h"
78 : #include "recog.h"
79 : #include "cgraph.h"
80 : #include "stor-layout.h"
81 : #include "cfgrtl.h"
82 : #include "cfgcleanup.h"
83 : /* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
84 : #include "explow.h"
85 : #include "insn-attr.h"
86 : #include "rtlhooks-def.h"
87 : #include "expr.h"
88 : #include "tree-pass.h"
89 : #include "valtrack.h"
90 : #include "rtl-iter.h"
91 : #include "print-rtl.h"
92 : #include "function-abi.h"
93 : #include "rtlanal.h"
94 :
95 : /* Number of attempts to combine instructions in this function. */
96 :
97 : static int combine_attempts;
98 :
99 : /* Number of attempts that got as far as substitution in this function. */
100 :
101 : static int combine_merges;
102 :
103 : /* Number of instructions combined with added SETs in this function. */
104 :
105 : static int combine_extras;
106 :
107 : /* Number of instructions combined in this function. */
108 :
109 : static int combine_successes;
110 :
111 : /* combine_instructions may try to replace the right hand side of the
112 : second instruction with the value of an associated REG_EQUAL note
113 : before throwing it at try_combine. That is problematic when there
114 : is a REG_DEAD note for a register used in the old right hand side
115 : and can cause distribute_notes to do wrong things. This is the
116 : second instruction if it has been so modified, null otherwise. */
117 :
118 : static rtx_insn *i2mod;
119 :
120 : /* When I2MOD is nonnull, this is a copy of the old right hand side. */
121 :
122 : static rtx i2mod_old_rhs;
123 :
124 : /* When I2MOD is nonnull, this is a copy of the new right hand side. */
125 :
126 : static rtx i2mod_new_rhs;
127 :
128 : struct reg_stat_type {
129 : /* Record last point of death of (hard or pseudo) register n. */
130 : rtx_insn *last_death;
131 :
132 : /* Record last point of modification of (hard or pseudo) register n. */
133 : rtx_insn *last_set;
134 :
135 : /* The next group of fields allows the recording of the last value assigned
136 : to (hard or pseudo) register n. We use this information to see if an
137 : operation being processed is redundant given a prior operation performed
138 : on the register. For example, an `and' with a constant is redundant if
139 : all the zero bits are already known to be turned off.
140 :
141 : We use an approach similar to that used by cse, but change it in the
142 : following ways:
143 :
144 : (1) We do not want to reinitialize at each label.
145 : (2) It is useful, but not critical, to know the actual value assigned
146 : to a register. Often just its form is helpful.
147 :
148 : Therefore, we maintain the following fields:
149 :
150 : last_set_value the last value assigned
151 : last_set_label records the value of label_tick when the
152 : register was assigned
153 : last_set_table_tick records the value of label_tick when a
154 : value using the register is assigned
155 : last_set_invalid set to true when it is not valid
156 : to use the value of this register in some
157 : register's value
158 :
159 : To understand the usage of these tables, it is important to understand
160 : the distinction between the value in last_set_value being valid and
161 : the register being validly contained in some other expression in the
162 : table.
163 :
164 : (The next two parameters are out of date).
165 :
166 : reg_stat[i].last_set_value is valid if it is nonzero, and either
167 : reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
168 :
169 : Register I may validly appear in any expression returned for the value
170 : of another register if reg_n_sets[i] is 1. It may also appear in the
171 : value for register J if reg_stat[j].last_set_invalid is zero, or
172 : reg_stat[i].last_set_label < reg_stat[j].last_set_label.
173 :
174 : If an expression is found in the table containing a register which may
175 : not validly appear in an expression, the register is replaced by
176 : something that won't match, (clobber (const_int 0)). */
177 :
178 : /* Record last value assigned to (hard or pseudo) register n. */
179 :
180 : rtx last_set_value;
181 :
182 : /* Record the value of label_tick when an expression involving register n
183 : is placed in last_set_value. */
184 :
185 : int last_set_table_tick;
186 :
187 : /* Record the value of label_tick when the value for register n is placed in
188 : last_set_value. */
189 :
190 : int last_set_label;
191 :
192 : /* These fields are maintained in parallel with last_set_value and are
193 : used to store the mode in which the register was last set, the bits
194 : that were known to be zero when it was last set, and the number of
195 : sign bits copies it was known to have when it was last set. */
196 :
197 : unsigned HOST_WIDE_INT last_set_nonzero_bits;
198 : unsigned short last_set_sign_bit_copies;
199 : ENUM_BITFIELD(machine_mode) last_set_mode : MACHINE_MODE_BITSIZE;
200 :
201 : /* Set to true if references to register n in expressions should not be
202 : used. last_set_invalid is set nonzero when this register is being
203 : assigned to and last_set_table_tick == label_tick. */
204 :
205 : bool last_set_invalid;
206 :
207 : /* Some registers that are set more than once and used in more than one
208 : basic block are nevertheless always set in similar ways. For example,
209 : a QImode register may be loaded from memory in two places on a machine
210 : where byte loads zero extend.
211 :
212 : We record in the following fields if a register has some leading bits
213 : that are always equal to the sign bit, and what we know about the
214 : nonzero bits of a register, specifically which bits are known to be
215 : zero.
216 :
217 : If an entry is zero, it means that we don't know anything special. */
218 :
219 : unsigned short sign_bit_copies;
220 :
221 : unsigned HOST_WIDE_INT nonzero_bits;
222 :
223 : /* Record the value of the label_tick when the last truncation
224 : happened. The field truncated_to_mode is only valid if
225 : truncation_label == label_tick. */
226 :
227 : int truncation_label;
228 :
229 : /* Record the last truncation seen for this register. If truncation
230 : is not a nop to this mode we might be able to save an explicit
231 : truncation if we know that value already contains a truncated
232 : value. */
233 :
234 : ENUM_BITFIELD(machine_mode) truncated_to_mode : MACHINE_MODE_BITSIZE;
235 : };
236 :
237 :
238 : static vec<reg_stat_type> reg_stat;
239 :
240 : /* One plus the highest pseudo for which we track REG_N_SETS.
241 : regstat_init_n_sets_and_refs allocates the array for REG_N_SETS just once,
242 : but during combine_split_insns new pseudos can be created. As we don't have
243 : updated DF information in that case, it is hard to initialize the array
244 : after growing. The combiner only cares about REG_N_SETS (regno) == 1,
245 : so instead of growing the arrays, just assume all newly created pseudos
246 : during combine might be set multiple times. */
247 :
248 : static unsigned int reg_n_sets_max;
249 :
250 : /* Record the luid of the last insn that invalidated memory
251 : (anything that writes memory, and subroutine calls, but not pushes). */
252 :
253 : static int mem_last_set;
254 :
255 : /* Record the luid of the last CALL_INSN
256 : so we can tell whether a potential combination crosses any calls. */
257 :
258 : static int last_call_luid;
259 :
260 : /* When `subst' is called, this is the insn that is being modified
261 : (by combining in a previous insn). The PATTERN of this insn
262 : is still the old pattern partially modified and it should not be
263 : looked at, but this may be used to examine the successors of the insn
264 : to judge whether a simplification is valid. */
265 :
266 : static rtx_insn *subst_insn;
267 :
268 : /* This is the lowest LUID that `subst' is currently dealing with.
269 : get_last_value will not return a value if the register was set at or
270 : after this LUID. If not for this mechanism, we could get confused if
271 : I2 or I1 in try_combine were an insn that used the old value of a register
272 : to obtain a new value. In that case, we might erroneously get the
273 : new value of the register when we wanted the old one. */
274 :
275 : static int subst_low_luid;
276 :
277 : /* This contains any hard registers that are used in newpat; reg_dead_at_p
278 : must consider all these registers to be always live. */
279 :
280 : static HARD_REG_SET newpat_used_regs;
281 :
282 : /* This is an insn to which a LOG_LINKS entry has been added. If this
283 : insn is the earlier than I2 or I3, combine should rescan starting at
284 : that location. */
285 :
286 : static rtx_insn *added_links_insn;
287 :
288 : /* And similarly, for notes. */
289 :
290 : static rtx_insn *added_notes_insn;
291 :
292 : /* Basic block in which we are performing combines. */
293 : static basic_block this_basic_block;
294 : static bool optimize_this_for_speed_p;
295 :
296 :
297 : /* Length of the currently allocated uid_insn_cost array. */
298 :
299 : static int max_uid_known;
300 :
301 : /* The following array records the insn_cost for every insn
302 : in the instruction stream. */
303 :
304 : static int *uid_insn_cost;
305 :
306 : /* The following array records the LOG_LINKS for every insn in the
307 : instruction stream as struct insn_link pointers. */
308 :
309 : struct insn_link {
310 : rtx_insn *insn;
311 : unsigned int regno;
312 : int insn_count;
313 : struct insn_link *next;
314 : };
315 :
316 : static struct insn_link **uid_log_links;
317 :
318 : static inline int
319 754226742 : insn_uid_check (const_rtx insn)
320 : {
321 754226742 : int uid = INSN_UID (insn);
322 754226742 : gcc_checking_assert (uid <= max_uid_known);
323 754226742 : return uid;
324 : }
325 :
326 : #define INSN_COST(INSN) (uid_insn_cost[insn_uid_check (INSN)])
327 : #define LOG_LINKS(INSN) (uid_log_links[insn_uid_check (INSN)])
328 :
329 : #define FOR_EACH_LOG_LINK(L, INSN) \
330 : for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
331 :
332 : /* Links for LOG_LINKS are allocated from this obstack. */
333 :
334 : static struct obstack insn_link_obstack;
335 :
336 : /* Allocate a link. */
337 :
338 : static inline struct insn_link *
339 37910861 : alloc_insn_link (rtx_insn *insn, unsigned int regno, struct insn_link *next)
340 : {
341 37910861 : struct insn_link *l
342 37910861 : = (struct insn_link *) obstack_alloc (&insn_link_obstack,
343 : sizeof (struct insn_link));
344 37910861 : l->insn = insn;
345 37910861 : l->regno = regno;
346 37910861 : l->insn_count = 0;
347 37910861 : l->next = next;
348 37910861 : return l;
349 : }
350 :
351 : /* Incremented for each basic block. */
352 :
353 : static int label_tick;
354 :
355 : /* Reset to label_tick for each extended basic block in scanning order. */
356 :
357 : static int label_tick_ebb_start;
358 :
359 : /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
360 : largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
361 :
362 : static scalar_int_mode nonzero_bits_mode;
363 :
364 : /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
365 : be safely used. It is zero while computing them and after combine has
366 : completed. This former test prevents propagating values based on
367 : previously set values, which can be incorrect if a variable is modified
368 : in a loop. */
369 :
370 : static int nonzero_sign_valid;
371 :
372 :
373 : /* Record one modification to rtl structure
374 : to be undone by storing old_contents into *where. */
375 :
376 : enum undo_kind { UNDO_RTX, UNDO_INT, UNDO_MODE, UNDO_LINKS };
377 :
378 : struct undo
379 : {
380 : struct undo *next;
381 : enum undo_kind kind;
382 : union { rtx r; int i; machine_mode m; struct insn_link *l; } old_contents;
383 : union { rtx *r; int *i; int regno; struct insn_link **l; } where;
384 : };
385 :
386 : /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
387 : num_undo says how many are currently recorded.
388 :
389 : other_insn is nonzero if we have modified some other insn in the process
390 : of working on subst_insn. It must be verified too. */
391 :
392 : struct undobuf
393 : {
394 : struct undo *undos;
395 : struct undo *frees;
396 : rtx_insn *other_insn;
397 : };
398 :
399 : static struct undobuf undobuf;
400 :
401 : /* Number of times the pseudo being substituted for
402 : was found and replaced. */
403 :
404 : static int n_occurrences;
405 :
406 : static rtx reg_nonzero_bits_for_combine (const_rtx, scalar_int_mode,
407 : scalar_int_mode,
408 : unsigned HOST_WIDE_INT *);
409 : static rtx reg_num_sign_bit_copies_for_combine (const_rtx, scalar_int_mode,
410 : scalar_int_mode,
411 : unsigned int *);
412 : static void do_SUBST (rtx *, rtx);
413 : static void do_SUBST_INT (int *, int);
414 : static void init_reg_last (void);
415 : static void setup_incoming_promotions (rtx_insn *);
416 : static void set_nonzero_bits_and_sign_copies (rtx, const_rtx, void *);
417 : static bool cant_combine_insn_p (rtx_insn *);
418 : static bool can_combine_p (rtx_insn *, rtx_insn *, rtx_insn *, rtx_insn *,
419 : rtx_insn *, rtx_insn *, rtx *, rtx *);
420 : static bool combinable_i3pat (rtx_insn *, rtx *, rtx, rtx, rtx,
421 : bool, bool, rtx *);
422 : static bool contains_muldiv (rtx);
423 : static rtx_insn *try_combine (rtx_insn *, rtx_insn *, rtx_insn *, rtx_insn *,
424 : bool *, rtx_insn *);
425 : static void undo_all (void);
426 : static void undo_commit (void);
427 : static rtx *find_split_point (rtx *, rtx_insn *, bool);
428 : static rtx subst (rtx, rtx, rtx, bool, bool, bool);
429 : static rtx combine_simplify_rtx (rtx, machine_mode, bool, bool);
430 : static rtx simplify_if_then_else (rtx);
431 : static rtx simplify_set (rtx);
432 : static rtx simplify_logical (rtx);
433 : static rtx expand_compound_operation (rtx);
434 : static const_rtx expand_field_assignment (const_rtx);
435 : static rtx make_extraction (machine_mode, rtx, HOST_WIDE_INT, rtx,
436 : unsigned HOST_WIDE_INT, bool, bool, bool);
437 : static int get_pos_from_mask (unsigned HOST_WIDE_INT,
438 : unsigned HOST_WIDE_INT *);
439 : static rtx canon_reg_for_combine (rtx, rtx);
440 : static rtx force_int_to_mode (rtx, scalar_int_mode, scalar_int_mode,
441 : scalar_int_mode, unsigned HOST_WIDE_INT, bool);
442 : static rtx force_to_mode (rtx, machine_mode,
443 : unsigned HOST_WIDE_INT, bool);
444 : static rtx if_then_else_cond (rtx, rtx *, rtx *);
445 : static rtx known_cond (rtx, enum rtx_code, rtx, rtx);
446 : static bool rtx_equal_for_field_assignment_p (rtx, rtx, bool = false);
447 : static rtx make_field_assignment (rtx);
448 : static rtx apply_distributive_law (rtx);
449 : static rtx distribute_and_simplify_rtx (rtx, int);
450 : static rtx simplify_and_const_int_1 (scalar_int_mode, rtx,
451 : unsigned HOST_WIDE_INT);
452 : static rtx simplify_and_const_int (rtx, scalar_int_mode, rtx,
453 : unsigned HOST_WIDE_INT);
454 : static bool merge_outer_ops (enum rtx_code *, HOST_WIDE_INT *, enum rtx_code,
455 : HOST_WIDE_INT, machine_mode, bool *);
456 : static rtx simplify_shift_const_1 (enum rtx_code, machine_mode, rtx, int);
457 : static rtx simplify_shift_const (rtx, enum rtx_code, machine_mode, rtx,
458 : int);
459 : static int recog_for_combine (rtx *, rtx_insn *, rtx *, unsigned = 0, unsigned = 0);
460 : static rtx gen_lowpart_for_combine (machine_mode, rtx);
461 : static rtx gen_lowpart_for_combine_no_emit (machine_mode, rtx);
462 : static enum rtx_code simplify_compare_const (enum rtx_code, machine_mode,
463 : rtx *, rtx *);
464 : static enum rtx_code simplify_comparison (enum rtx_code, rtx *, rtx *);
465 : static void update_table_tick (rtx);
466 : static void record_value_for_reg (rtx, rtx_insn *, rtx);
467 : static void check_promoted_subreg (rtx_insn *, rtx);
468 : static void record_dead_and_set_regs_1 (rtx, const_rtx, void *);
469 : static void record_dead_and_set_regs (rtx_insn *);
470 : static bool get_last_value_validate (rtx *, rtx_insn *, int, bool);
471 : static rtx get_last_value (const_rtx);
472 : static void reg_dead_at_p_1 (rtx, const_rtx, void *);
473 : static bool reg_dead_at_p (rtx, rtx_insn *);
474 : static void move_deaths (rtx, rtx, int, rtx_insn *, rtx *);
475 : static bool reg_bitfield_target_p (rtx, rtx);
476 : static void distribute_notes (rtx, rtx_insn *, rtx_insn *, rtx_insn *,
477 : rtx, rtx, rtx);
478 : static void distribute_links (struct insn_link *, rtx_insn * = nullptr,
479 : int limit = INT_MAX);
480 : static void mark_used_regs_combine (rtx);
481 : static void record_promoted_value (rtx_insn *, rtx);
482 : static bool unmentioned_reg_p (rtx, rtx);
483 : static void record_truncated_values (rtx *, void *);
484 : static bool reg_truncated_to_mode (machine_mode, const_rtx);
485 : static rtx gen_lowpart_or_truncate (machine_mode, rtx);
486 :
487 :
488 : /* It is not safe to use ordinary gen_lowpart in combine.
489 : See comments in gen_lowpart_for_combine. */
490 : #undef RTL_HOOKS_GEN_LOWPART
491 : #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
492 :
493 : /* Our implementation of gen_lowpart never emits a new pseudo. */
494 : #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
495 : #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine_no_emit
496 :
497 : #undef RTL_HOOKS_REG_NONZERO_REG_BITS
498 : #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
499 :
500 : #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
501 : #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
502 :
503 : #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
504 : #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
505 :
506 : static const struct rtl_hooks combine_rtl_hooks = RTL_HOOKS_INITIALIZER;
507 :
508 :
509 : /* Convenience wrapper for the canonicalize_comparison target hook.
510 : Target hooks cannot use enum rtx_code. */
511 : static inline void
512 24283785 : target_canonicalize_comparison (enum rtx_code *code, rtx *op0, rtx *op1,
513 : bool op0_preserve_value)
514 : {
515 24283785 : int code_int = (int)*code;
516 24283785 : targetm.canonicalize_comparison (&code_int, op0, op1, op0_preserve_value);
517 24283785 : *code = (enum rtx_code)code_int;
518 816527 : }
519 :
520 : /* Try to split PATTERN found in INSN. This returns NULL_RTX if
521 : PATTERN cannot be split. Otherwise, it returns an insn sequence.
522 : Updates OLD_NREGS with the max number of regs before the split
523 : and NEW_NREGS after the split.
524 : This is a wrapper around split_insns which ensures that the
525 : reg_stat vector is made larger if the splitter creates a new
526 : register. */
527 :
528 : static rtx_insn *
529 11701337 : combine_split_insns (rtx pattern, rtx_insn *insn,
530 : unsigned int *old_nregs,
531 : unsigned int *new_regs)
532 : {
533 11701337 : rtx_insn *ret;
534 11701337 : unsigned int nregs;
535 11701337 : *old_nregs = max_reg_num ();
536 11701337 : ret = split_insns (pattern, insn);
537 11701337 : *new_regs = nregs = max_reg_num ();
538 23402674 : if (nregs > reg_stat.length ())
539 3236 : reg_stat.safe_grow_cleared (nregs, true);
540 11701337 : return ret;
541 : }
542 :
543 : /* This is used by find_single_use to locate an rtx in LOC that
544 : contains exactly one use of DEST, which is typically a REG.
545 : It returns a pointer to the innermost rtx expression
546 : containing DEST. Appearances of DEST that are being used to
547 : totally replace it are not counted. */
548 :
549 : static rtx *
550 32913841 : find_single_use_1 (rtx dest, rtx *loc)
551 : {
552 39813578 : rtx x = *loc;
553 39813578 : enum rtx_code code = GET_CODE (x);
554 39813578 : rtx *result = NULL;
555 39813578 : rtx *this_result;
556 39813578 : int i;
557 39813578 : const char *fmt;
558 :
559 39813578 : switch (code)
560 : {
561 : case CONST:
562 : case LABEL_REF:
563 : case SYMBOL_REF:
564 : CASE_CONST_ANY:
565 : case CLOBBER:
566 : return 0;
567 :
568 6855432 : case SET:
569 : /* If the destination is anything other than PC, a REG or a SUBREG
570 : of a REG that occupies all of the REG, the insn uses DEST if
571 : it is mentioned in the destination or the source. Otherwise, we
572 : need just check the source. */
573 6855432 : if (GET_CODE (SET_DEST (x)) != PC
574 6855432 : && !REG_P (SET_DEST (x))
575 6856736 : && ! (GET_CODE (SET_DEST (x)) == SUBREG
576 1304 : && REG_P (SUBREG_REG (SET_DEST (x)))
577 1304 : && !read_modify_subreg_p (SET_DEST (x))))
578 : break;
579 :
580 6854309 : return find_single_use_1 (dest, &SET_SRC (x));
581 :
582 45428 : case MEM:
583 45428 : case SUBREG:
584 45428 : return find_single_use_1 (dest, &XEXP (x, 0));
585 :
586 : default:
587 : break;
588 : }
589 :
590 : /* If it wasn't one of the common cases above, check each expression and
591 : vector of this code. Look for a unique usage of DEST. */
592 :
593 19826985 : fmt = GET_RTX_FORMAT (code);
594 52980525 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
595 : {
596 33162754 : if (fmt[i] == 'e')
597 : {
598 32814371 : if (dest == XEXP (x, i)
599 32814371 : || (REG_P (dest) && REG_P (XEXP (x, i))
600 819935 : && REGNO (dest) == REGNO (XEXP (x, i))))
601 : this_result = loc;
602 : else
603 25954279 : this_result = find_single_use_1 (dest, &XEXP (x, i));
604 :
605 32814371 : if (result == NULL)
606 : result = this_result;
607 43725 : else if (this_result)
608 : /* Duplicate usage. */
609 : return NULL;
610 : }
611 348383 : else if (fmt[i] == 'E')
612 : {
613 53477 : int j;
614 :
615 162660 : for (j = XVECLEN (x, i) - 1; j >= 0; j--)
616 : {
617 113160 : if (XVECEXP (x, i, j) == dest
618 113160 : || (REG_P (dest)
619 113160 : && REG_P (XVECEXP (x, i, j))
620 4546 : && REGNO (XVECEXP (x, i, j)) == REGNO (dest)))
621 : this_result = loc;
622 : else
623 113160 : this_result = find_single_use_1 (dest, &XVECEXP (x, i, j));
624 :
625 113160 : if (result == NULL)
626 : result = this_result;
627 18596 : else if (this_result)
628 : return NULL;
629 : }
630 : }
631 : }
632 :
633 : return result;
634 : }
635 :
636 :
637 : /* See if DEST, produced in INSN, is used only a single time in the
638 : sequel. If so, return a pointer to the innermost rtx expression in which
639 : it is used.
640 :
641 : If PLOC is nonzero, *PLOC is set to the insn containing the single use.
642 :
643 : Otherwise, we find the single use by finding an insn that has a
644 : LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
645 : only referenced once in that insn, we know that it must be the first
646 : and last insn referencing DEST. */
647 :
648 : static rtx *
649 7429908 : find_single_use (rtx dest, rtx_insn *insn, rtx_insn **ploc)
650 : {
651 7429908 : basic_block bb;
652 7429908 : rtx_insn *next;
653 7429908 : rtx *result;
654 7429908 : struct insn_link *link;
655 :
656 7429908 : if (!REG_P (dest))
657 : return 0;
658 :
659 7429908 : bb = BLOCK_FOR_INSN (insn);
660 10166260 : for (next = NEXT_INSN (insn);
661 10166260 : next && BLOCK_FOR_INSN (next) == bb;
662 2736352 : next = NEXT_INSN (next))
663 9582754 : if (NONDEBUG_INSN_P (next) && dead_or_set_p (next, dest))
664 : {
665 9125649 : FOR_EACH_LOG_LINK (link, next)
666 8116947 : if (link->insn == insn && link->regno == REGNO (dest))
667 : break;
668 :
669 7855104 : if (link)
670 : {
671 6846402 : result = find_single_use_1 (dest, &PATTERN (next));
672 6846402 : if (ploc)
673 6846402 : *ploc = next;
674 6846402 : return result;
675 : }
676 : }
677 :
678 : return 0;
679 : }
680 :
681 : /* Substitute NEWVAL, an rtx expression, into INTO, a place in some
682 : insn. The substitution can be undone by undo_all. If INTO is already
683 : set to NEWVAL, do not record this change. Because computing NEWVAL might
684 : also call SUBST, we have to compute it before we put anything into
685 : the undo table. */
686 :
687 : static void
688 845852835 : do_SUBST (rtx *into, rtx newval)
689 : {
690 845852835 : struct undo *buf;
691 845852835 : rtx oldval = *into;
692 :
693 845852835 : if (oldval == newval)
694 : return;
695 :
696 : /* We'd like to catch as many invalid transformations here as
697 : possible. Unfortunately, there are way too many mode changes
698 : that are perfectly valid, so we'd waste too much effort for
699 : little gain doing the checks here. Focus on catching invalid
700 : transformations involving integer constants. */
701 94916792 : if (GET_MODE_CLASS (GET_MODE (oldval)) == MODE_INT
702 57939108 : && CONST_INT_P (newval))
703 : {
704 : /* Sanity check that we're replacing oldval with a CONST_INT
705 : that is a valid sign-extension for the original mode. */
706 1797795 : gcc_assert (INTVAL (newval)
707 : == trunc_int_for_mode (INTVAL (newval), GET_MODE (oldval)));
708 :
709 : /* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
710 : CONST_INT is not valid, because after the replacement, the
711 : original mode would be gone. Unfortunately, we can't tell
712 : when do_SUBST is called to replace the operand thereof, so we
713 : perform this test on oldval instead, checking whether an
714 : invalid replacement took place before we got here. */
715 1797795 : gcc_assert (!(GET_CODE (oldval) == SUBREG
716 : && CONST_INT_P (SUBREG_REG (oldval))));
717 1797795 : gcc_assert (!(GET_CODE (oldval) == ZERO_EXTEND
718 : && CONST_INT_P (XEXP (oldval, 0))));
719 : }
720 :
721 94916792 : if (undobuf.frees)
722 90856787 : buf = undobuf.frees, undobuf.frees = buf->next;
723 : else
724 4060005 : buf = XNEW (struct undo);
725 :
726 94916792 : buf->kind = UNDO_RTX;
727 94916792 : buf->where.r = into;
728 94916792 : buf->old_contents.r = oldval;
729 94916792 : *into = newval;
730 :
731 94916792 : buf->next = undobuf.undos, undobuf.undos = buf;
732 : }
733 :
734 : #define SUBST(INTO, NEWVAL) do_SUBST (&(INTO), (NEWVAL))
735 :
736 : /* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
737 : for the value of a HOST_WIDE_INT value (including CONST_INT) is
738 : not safe. */
739 :
740 : static void
741 15854526 : do_SUBST_INT (int *into, int newval)
742 : {
743 15854526 : struct undo *buf;
744 15854526 : int oldval = *into;
745 :
746 15854526 : if (oldval == newval)
747 : return;
748 :
749 6742047 : if (undobuf.frees)
750 6233597 : buf = undobuf.frees, undobuf.frees = buf->next;
751 : else
752 508450 : buf = XNEW (struct undo);
753 :
754 6742047 : buf->kind = UNDO_INT;
755 6742047 : buf->where.i = into;
756 6742047 : buf->old_contents.i = oldval;
757 6742047 : *into = newval;
758 :
759 6742047 : buf->next = undobuf.undos, undobuf.undos = buf;
760 : }
761 :
762 : #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT (&(INTO), (NEWVAL))
763 :
764 : /* Similar to SUBST, but just substitute the mode. This is used when
765 : changing the mode of a pseudo-register, so that any other
766 : references to the entry in the regno_reg_rtx array will change as
767 : well. */
768 :
769 : static void
770 1421342 : subst_mode (int regno, machine_mode newval)
771 : {
772 1421342 : struct undo *buf;
773 1421342 : rtx reg = regno_reg_rtx[regno];
774 1421342 : machine_mode oldval = GET_MODE (reg);
775 :
776 1421342 : if (oldval == newval)
777 : return;
778 :
779 1421342 : if (undobuf.frees)
780 1346030 : buf = undobuf.frees, undobuf.frees = buf->next;
781 : else
782 75312 : buf = XNEW (struct undo);
783 :
784 1421342 : buf->kind = UNDO_MODE;
785 1421342 : buf->where.regno = regno;
786 1421342 : buf->old_contents.m = oldval;
787 1421342 : adjust_reg_mode (reg, newval);
788 :
789 1421342 : buf->next = undobuf.undos, undobuf.undos = buf;
790 : }
791 :
792 : /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
793 :
794 : static void
795 70759 : do_SUBST_LINK (struct insn_link **into, struct insn_link *newval)
796 : {
797 70759 : struct undo *buf;
798 70759 : struct insn_link * oldval = *into;
799 :
800 70759 : if (oldval == newval)
801 : return;
802 :
803 70759 : if (undobuf.frees)
804 67734 : buf = undobuf.frees, undobuf.frees = buf->next;
805 : else
806 3025 : buf = XNEW (struct undo);
807 :
808 70759 : buf->kind = UNDO_LINKS;
809 70759 : buf->where.l = into;
810 70759 : buf->old_contents.l = oldval;
811 70759 : *into = newval;
812 :
813 70759 : buf->next = undobuf.undos, undobuf.undos = buf;
814 : }
815 :
816 : #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
817 :
818 : /* Subroutine of try_combine. Determine whether the replacement patterns
819 : NEWPAT, NEWI2PAT and NEWOTHERPAT are more expensive according to insn_cost
820 : than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
821 : that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
822 : undobuf.other_insn may also both be NULL_RTX. Return false if the cost
823 : of all the instructions can be estimated and the replacements are more
824 : expensive than the original sequence. */
825 :
826 : static bool
827 4231112 : combine_validate_cost (rtx_insn *i0, rtx_insn *i1, rtx_insn *i2, rtx_insn *i3,
828 : rtx newpat, rtx newi2pat, rtx newotherpat)
829 : {
830 4231112 : int i0_cost, i1_cost, i2_cost, i3_cost;
831 4231112 : int new_i2_cost, new_i3_cost;
832 4231112 : int old_cost, new_cost;
833 :
834 : /* Lookup the original insn_costs. */
835 4231112 : i2_cost = INSN_COST (i2);
836 4231112 : i3_cost = INSN_COST (i3);
837 :
838 4231112 : if (i1)
839 : {
840 117899 : i1_cost = INSN_COST (i1);
841 117899 : if (i0)
842 : {
843 5184 : i0_cost = INSN_COST (i0);
844 5064 : old_cost = (i0_cost > 0 && i1_cost > 0 && i2_cost > 0 && i3_cost > 0
845 10236 : ? i0_cost + i1_cost + i2_cost + i3_cost : 0);
846 : }
847 : else
848 : {
849 108062 : old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0
850 220775 : ? i1_cost + i2_cost + i3_cost : 0);
851 : i0_cost = 0;
852 : }
853 : }
854 : else
855 : {
856 4113213 : old_cost = (i2_cost > 0 && i3_cost > 0) ? i2_cost + i3_cost : 0;
857 : i1_cost = i0_cost = 0;
858 : }
859 :
860 : /* If we have split a PARALLEL I2 to I1,I2, we have counted its cost twice;
861 : correct that. */
862 4231112 : if (old_cost && i1 && INSN_UID (i1) == INSN_UID (i2))
863 2350 : old_cost -= i1_cost;
864 :
865 :
866 : /* Calculate the replacement insn_costs. */
867 4231112 : rtx tmp = PATTERN (i3);
868 4231112 : PATTERN (i3) = newpat;
869 4231112 : int tmpi = INSN_CODE (i3);
870 4231112 : INSN_CODE (i3) = -1;
871 4231112 : new_i3_cost = insn_cost (i3, optimize_this_for_speed_p);
872 4231112 : PATTERN (i3) = tmp;
873 4231112 : INSN_CODE (i3) = tmpi;
874 4231112 : if (newi2pat)
875 : {
876 211507 : tmp = PATTERN (i2);
877 211507 : PATTERN (i2) = newi2pat;
878 211507 : tmpi = INSN_CODE (i2);
879 211507 : INSN_CODE (i2) = -1;
880 211507 : new_i2_cost = insn_cost (i2, optimize_this_for_speed_p);
881 211507 : PATTERN (i2) = tmp;
882 211507 : INSN_CODE (i2) = tmpi;
883 211507 : new_cost = (new_i2_cost > 0 && new_i3_cost > 0)
884 211507 : ? new_i2_cost + new_i3_cost : 0;
885 : }
886 : else
887 : {
888 : new_cost = new_i3_cost;
889 : new_i2_cost = 0;
890 : }
891 :
892 4231112 : if (undobuf.other_insn)
893 : {
894 194574 : int old_other_cost, new_other_cost;
895 :
896 194574 : old_other_cost = INSN_COST (undobuf.other_insn);
897 194574 : tmp = PATTERN (undobuf.other_insn);
898 194574 : PATTERN (undobuf.other_insn) = newotherpat;
899 194574 : tmpi = INSN_CODE (undobuf.other_insn);
900 194574 : INSN_CODE (undobuf.other_insn) = -1;
901 194574 : new_other_cost = insn_cost (undobuf.other_insn,
902 : optimize_this_for_speed_p);
903 194574 : PATTERN (undobuf.other_insn) = tmp;
904 194574 : INSN_CODE (undobuf.other_insn) = tmpi;
905 194574 : if (old_other_cost > 0 && new_other_cost > 0)
906 : {
907 194574 : old_cost += old_other_cost;
908 194574 : new_cost += new_other_cost;
909 : }
910 : else
911 : old_cost = 0;
912 : }
913 :
914 : /* Disallow this combination if both new_cost and old_cost are greater than
915 : zero, and new_cost is greater than old cost. */
916 4231112 : bool reject = old_cost > 0 && new_cost > old_cost;
917 :
918 4231112 : if (dump_file)
919 : {
920 484 : fprintf (dump_file, "%s combination of insns ",
921 : reject ? "rejecting" : "allowing");
922 244 : if (i0)
923 0 : fprintf (dump_file, "%d, ", INSN_UID (i0));
924 244 : if (i1 && INSN_UID (i1) != INSN_UID (i2))
925 1 : fprintf (dump_file, "%d, ", INSN_UID (i1));
926 244 : fprintf (dump_file, "%d and %d\n", INSN_UID (i2), INSN_UID (i3));
927 :
928 244 : fprintf (dump_file, "original costs ");
929 244 : if (i0)
930 0 : fprintf (dump_file, "%d + ", i0_cost);
931 244 : if (i1 && INSN_UID (i1) != INSN_UID (i2))
932 1 : fprintf (dump_file, "%d + ", i1_cost);
933 244 : fprintf (dump_file, "%d + %d = %d\n", i2_cost, i3_cost, old_cost);
934 :
935 244 : if (newi2pat)
936 19 : fprintf (dump_file, "replacement costs %d + %d = %d\n",
937 : new_i2_cost, new_i3_cost, new_cost);
938 : else
939 225 : fprintf (dump_file, "replacement cost %d\n", new_cost);
940 : }
941 :
942 4231112 : if (reject)
943 : return false;
944 :
945 : /* Update the uid_insn_cost array with the replacement costs. */
946 4022156 : INSN_COST (i2) = new_i2_cost;
947 4022156 : INSN_COST (i3) = new_i3_cost;
948 4022156 : if (i1)
949 : {
950 99960 : INSN_COST (i1) = 0;
951 99960 : if (i0)
952 5058 : INSN_COST (i0) = 0;
953 : }
954 :
955 : return true;
956 : }
957 :
958 :
959 : /* Delete any insns that copy a register to itself.
960 : Return true if the CFG was changed. */
961 :
962 : static bool
963 999512 : delete_noop_moves (void)
964 : {
965 999512 : rtx_insn *insn, *next;
966 999512 : basic_block bb;
967 :
968 999512 : bool edges_deleted = false;
969 :
970 11358052 : FOR_EACH_BB_FN (bb, cfun)
971 : {
972 136957706 : for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
973 : {
974 126599166 : next = NEXT_INSN (insn);
975 126599166 : if (INSN_P (insn) && noop_move_p (insn))
976 : {
977 6859 : if (dump_file)
978 0 : fprintf (dump_file, "deleting noop move %d\n", INSN_UID (insn));
979 :
980 6859 : edges_deleted |= delete_insn_and_edges (insn);
981 : }
982 : }
983 : }
984 :
985 999512 : return edges_deleted;
986 : }
987 :
988 :
989 : /* Return false if we do not want to (or cannot) combine DEF. */
990 : static bool
991 41730659 : can_combine_def_p (df_ref def)
992 : {
993 : /* Do not consider if it is pre/post modification in MEM. */
994 41730659 : if (DF_REF_FLAGS (def) & DF_REF_PRE_POST_MODIFY)
995 : return false;
996 :
997 40063303 : unsigned int regno = DF_REF_REGNO (def);
998 :
999 : /* Do not combine frame pointer adjustments. */
1000 40063303 : if ((regno == FRAME_POINTER_REGNUM
1001 0 : && (!reload_completed || frame_pointer_needed))
1002 2062 : || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
1003 40063303 : && regno == HARD_FRAME_POINTER_REGNUM
1004 : && (!reload_completed || frame_pointer_needed))
1005 40061241 : || (FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
1006 0 : && regno == ARG_POINTER_REGNUM && fixed_regs[regno]))
1007 2062 : return false;
1008 :
1009 : return true;
1010 : }
1011 :
1012 : /* Return false if we do not want to (or cannot) combine USE. */
1013 : static bool
1014 77460385 : can_combine_use_p (df_ref use)
1015 : {
1016 : /* Do not consider the usage of the stack pointer by function call. */
1017 0 : if (DF_REF_FLAGS (use) & DF_REF_CALL_STACK_USAGE)
1018 0 : return false;
1019 :
1020 : return true;
1021 : }
1022 :
1023 : /* Fill in log links field for all insns. */
1024 :
1025 : static void
1026 999512 : create_log_links (void)
1027 : {
1028 999512 : basic_block bb;
1029 999512 : rtx_insn **next_use;
1030 999512 : rtx_insn *insn;
1031 999512 : df_ref def, use;
1032 :
1033 999512 : next_use = XCNEWVEC (rtx_insn *, max_reg_num ());
1034 :
1035 : /* Pass through each block from the end, recording the uses of each
1036 : register and establishing log links when def is encountered.
1037 : Note that we do not clear next_use array in order to save time,
1038 : so we have to test whether the use is in the same basic block as def.
1039 :
1040 : There are a few cases below when we do not consider the definition or
1041 : usage -- these are taken from original flow.c did. Don't ask me why it is
1042 : done this way; I don't know and if it works, I don't want to know. */
1043 :
1044 11358052 : FOR_EACH_BB_FN (bb, cfun)
1045 : {
1046 136941790 : FOR_BB_INSNS_REVERSE (bb, insn)
1047 : {
1048 126583250 : if (!NONDEBUG_INSN_P (insn))
1049 64986276 : continue;
1050 :
1051 : /* Log links are created only once. */
1052 61596974 : gcc_assert (!LOG_LINKS (insn));
1053 :
1054 492882547 : FOR_EACH_INSN_DEF (def, insn)
1055 : {
1056 431285573 : unsigned int regno = DF_REF_REGNO (def);
1057 431285573 : rtx_insn *use_insn;
1058 :
1059 431285573 : if (!next_use[regno])
1060 389554914 : continue;
1061 :
1062 41730659 : if (!can_combine_def_p (def))
1063 1669418 : continue;
1064 :
1065 40061241 : use_insn = next_use[regno];
1066 40061241 : next_use[regno] = NULL;
1067 :
1068 40061241 : if (BLOCK_FOR_INSN (use_insn) != bb)
1069 2238448 : continue;
1070 :
1071 : /* flow.c claimed:
1072 :
1073 : We don't build a LOG_LINK for hard registers contained
1074 : in ASM_OPERANDs. If these registers get replaced,
1075 : we might wind up changing the semantics of the insn,
1076 : even if reload can make what appear to be valid
1077 : assignments later. */
1078 37823645 : if (regno < FIRST_PSEUDO_REGISTER
1079 37822793 : && asm_noperands (PATTERN (use_insn)) >= 0)
1080 852 : continue;
1081 :
1082 : /* Don't add duplicate links between instructions. */
1083 37821941 : struct insn_link *links;
1084 50605183 : FOR_EACH_LOG_LINK (links, use_insn)
1085 12783242 : if (insn == links->insn && regno == links->regno)
1086 : break;
1087 :
1088 37821941 : if (!links)
1089 37821941 : LOG_LINKS (use_insn)
1090 75643882 : = alloc_insn_link (insn, regno, LOG_LINKS (use_insn));
1091 : }
1092 :
1093 139057359 : FOR_EACH_INSN_USE (use, insn)
1094 150292759 : if (can_combine_use_p (use))
1095 72832374 : next_use[DF_REF_REGNO (use)] = insn;
1096 : }
1097 : }
1098 :
1099 999512 : free (next_use);
1100 999512 : }
1101 :
1102 : /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1103 : true if we found a LOG_LINK that proves that A feeds B. This only works
1104 : if there are no instructions between A and B which could have a link
1105 : depending on A, since in that case we would not record a link for B. */
1106 :
1107 : static bool
1108 12807337 : insn_a_feeds_b (rtx_insn *a, rtx_insn *b)
1109 : {
1110 12807337 : struct insn_link *links;
1111 16182939 : FOR_EACH_LOG_LINK (links, b)
1112 13648627 : if (links->insn == a)
1113 : return true;
1114 : return false;
1115 : }
1116 :
1117 : /* Main entry point for combiner. F is the first insn of the function.
1118 : NREGS is the first unused pseudo-reg number.
1119 :
1120 : Return nonzero if the CFG was changed (e.g. if the combiner has
1121 : turned an indirect jump instruction into a direct jump). */
1122 : static bool
1123 1043685 : combine_instructions (rtx_insn *f, unsigned int nregs)
1124 : {
1125 1043685 : rtx_insn *insn, *next;
1126 1043685 : struct insn_link *links, *nextlinks;
1127 1043685 : rtx_insn *first;
1128 1043685 : basic_block last_bb;
1129 :
1130 1043685 : bool new_direct_jump_p = false;
1131 :
1132 3096666 : for (first = f; first && !NONDEBUG_INSN_P (first); )
1133 2052981 : first = NEXT_INSN (first);
1134 1043685 : if (!first)
1135 : return false;
1136 :
1137 999512 : combine_attempts = 0;
1138 999512 : combine_merges = 0;
1139 999512 : combine_extras = 0;
1140 999512 : combine_successes = 0;
1141 :
1142 999512 : rtl_hooks = combine_rtl_hooks;
1143 :
1144 999512 : reg_stat.safe_grow_cleared (nregs, true);
1145 :
1146 999512 : init_recog_no_volatile ();
1147 :
1148 : /* Allocate array for insn info. */
1149 999512 : max_uid_known = get_max_uid ();
1150 999512 : uid_log_links = XCNEWVEC (struct insn_link *, max_uid_known + 1);
1151 999512 : uid_insn_cost = XCNEWVEC (int, max_uid_known + 1);
1152 999512 : gcc_obstack_init (&insn_link_obstack);
1153 :
1154 999512 : nonzero_bits_mode = int_mode_for_size (HOST_BITS_PER_WIDE_INT, 0).require ();
1155 :
1156 : /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1157 : problems when, for example, we have j <<= 1 in a loop. */
1158 :
1159 999512 : nonzero_sign_valid = 0;
1160 999512 : label_tick = label_tick_ebb_start = 1;
1161 :
1162 : /* Scan all SETs and see if we can deduce anything about what
1163 : bits are known to be zero for some registers and how many copies
1164 : of the sign bit are known to exist for those registers.
1165 :
1166 : Also set any known values so that we can use it while searching
1167 : for what bits are known to be set. */
1168 :
1169 999512 : setup_incoming_promotions (first);
1170 : /* Allow the entry block and the first block to fall into the same EBB.
1171 : Conceptually the incoming promotions are assigned to the entry block. */
1172 999512 : last_bb = ENTRY_BLOCK_PTR_FOR_FN (cfun);
1173 :
1174 999512 : create_log_links ();
1175 11358052 : FOR_EACH_BB_FN (this_basic_block, cfun)
1176 : {
1177 10358540 : optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
1178 10358540 : last_call_luid = 0;
1179 10358540 : mem_last_set = -1;
1180 :
1181 10358540 : label_tick++;
1182 10358540 : if (!single_pred_p (this_basic_block)
1183 10358540 : || single_pred (this_basic_block) != last_bb)
1184 4978155 : label_tick_ebb_start = label_tick;
1185 10358540 : last_bb = this_basic_block;
1186 :
1187 136941790 : FOR_BB_INSNS (this_basic_block, insn)
1188 126583250 : if (INSN_P (insn) && BLOCK_FOR_INSN (insn))
1189 : {
1190 110161564 : rtx links;
1191 :
1192 110161564 : subst_low_luid = DF_INSN_LUID (insn);
1193 110161564 : subst_insn = insn;
1194 :
1195 110161564 : note_stores (insn, set_nonzero_bits_and_sign_copies, insn);
1196 110161564 : record_dead_and_set_regs (insn);
1197 :
1198 110161564 : if (AUTO_INC_DEC)
1199 : for (links = REG_NOTES (insn); links; links = XEXP (links, 1))
1200 : if (REG_NOTE_KIND (links) == REG_INC)
1201 : set_nonzero_bits_and_sign_copies (XEXP (links, 0), NULL_RTX,
1202 : insn);
1203 :
1204 : /* Record the current insn_cost of this instruction. */
1205 110161564 : INSN_COST (insn) = insn_cost (insn, optimize_this_for_speed_p);
1206 110161564 : if (dump_file)
1207 : {
1208 1695 : fprintf (dump_file, "insn_cost %d for ", INSN_COST (insn));
1209 1695 : dump_insn_slim (dump_file, insn);
1210 : }
1211 : }
1212 : }
1213 :
1214 999512 : nonzero_sign_valid = 1;
1215 :
1216 : /* Now scan all the insns in forward order. */
1217 999512 : label_tick = label_tick_ebb_start = 1;
1218 999512 : init_reg_last ();
1219 999512 : setup_incoming_promotions (first);
1220 999512 : last_bb = ENTRY_BLOCK_PTR_FOR_FN (cfun);
1221 999512 : int max_combine = param_max_combine_insns;
1222 :
1223 11358052 : FOR_EACH_BB_FN (this_basic_block, cfun)
1224 : {
1225 10358540 : rtx_insn *last_combined_insn = NULL;
1226 :
1227 : /* Ignore instruction combination in basic blocks that are going to
1228 : be removed as unreachable anyway. See PR82386. */
1229 10358540 : if (EDGE_COUNT (this_basic_block->preds) == 0)
1230 1575 : continue;
1231 :
1232 10356965 : optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
1233 10356965 : last_call_luid = 0;
1234 10356965 : mem_last_set = -1;
1235 :
1236 10356965 : label_tick++;
1237 10356965 : if (!single_pred_p (this_basic_block)
1238 10356965 : || single_pred (this_basic_block) != last_bb)
1239 4977815 : label_tick_ebb_start = label_tick;
1240 10356965 : last_bb = this_basic_block;
1241 :
1242 10356965 : rtl_profile_for_bb (this_basic_block);
1243 10356965 : for (insn = BB_HEAD (this_basic_block);
1244 141522228 : insn != NEXT_INSN (BB_END (this_basic_block));
1245 127143107 : insn = next ? next : NEXT_INSN (insn))
1246 : {
1247 131165263 : next = 0;
1248 131165263 : if (!NONDEBUG_INSN_P (insn))
1249 65243674 : continue;
1250 :
1251 : while (last_combined_insn
1252 65923377 : && (!NONDEBUG_INSN_P (last_combined_insn)
1253 55751858 : || last_combined_insn->deleted ()))
1254 1788 : last_combined_insn = PREV_INSN (last_combined_insn);
1255 65921589 : if (last_combined_insn == NULL_RTX
1256 55751292 : || BLOCK_FOR_INSN (last_combined_insn) != this_basic_block
1257 121672682 : || DF_INSN_LUID (last_combined_insn) <= DF_INSN_LUID (insn))
1258 : last_combined_insn = insn;
1259 :
1260 : /* See if we know about function return values before this
1261 : insn based upon SUBREG flags. */
1262 65921589 : check_promoted_subreg (insn, PATTERN (insn));
1263 :
1264 : /* See if we can find hardregs and subreg of pseudos in
1265 : narrower modes. This could help turning TRUNCATEs
1266 : into SUBREGs. */
1267 65921589 : note_uses (&PATTERN (insn), record_truncated_values, NULL);
1268 :
1269 : /* Try this insn with each insn it links back to. */
1270 :
1271 102778315 : FOR_EACH_LOG_LINK (links, insn)
1272 40748301 : if ((next = try_combine (insn, links->insn, NULL,
1273 : NULL, &new_direct_jump_p,
1274 : last_combined_insn)) != 0)
1275 : {
1276 3891575 : statistics_counter_event (cfun, "two-insn combine", 1);
1277 3891575 : goto retry;
1278 : }
1279 :
1280 : /* Try each sequence of three linked insns ending with this one. */
1281 :
1282 62030014 : if (max_combine >= 3)
1283 98307707 : FOR_EACH_LOG_LINK (links, insn)
1284 : {
1285 36463144 : rtx_insn *link = links->insn;
1286 :
1287 : /* If the linked insn has been replaced by a note, then there
1288 : is no point in pursuing this chain any further. */
1289 36463144 : if (NOTE_P (link))
1290 238 : continue;
1291 :
1292 54309263 : FOR_EACH_LOG_LINK (nextlinks, link)
1293 17924252 : if ((next = try_combine (insn, link, nextlinks->insn,
1294 : NULL, &new_direct_jump_p,
1295 : last_combined_insn)) != 0)
1296 : {
1297 77895 : statistics_counter_event (cfun, "three-insn combine", 1);
1298 77895 : goto retry;
1299 : }
1300 : }
1301 :
1302 : /* Try combining an insn with two different insns whose results it
1303 : uses. */
1304 61844563 : if (max_combine >= 3)
1305 98194815 : FOR_EACH_LOG_LINK (links, insn)
1306 48827948 : for (nextlinks = links->next; nextlinks;
1307 12463272 : nextlinks = nextlinks->next)
1308 12477696 : if ((next = try_combine (insn, links->insn,
1309 : nextlinks->insn, NULL,
1310 : &new_direct_jump_p,
1311 : last_combined_insn)) != 0)
1312 :
1313 : {
1314 14424 : statistics_counter_event (cfun, "three-insn combine", 1);
1315 14424 : goto retry;
1316 : }
1317 :
1318 : /* Try four-instruction combinations. */
1319 61830139 : if (max_combine >= 4)
1320 98171569 : FOR_EACH_LOG_LINK (links, insn)
1321 : {
1322 36346444 : struct insn_link *next1;
1323 36346444 : rtx_insn *link = links->insn;
1324 :
1325 : /* If the linked insn has been replaced by a note, then there
1326 : is no point in pursuing this chain any further. */
1327 36346444 : if (NOTE_P (link))
1328 238 : continue;
1329 :
1330 54170618 : FOR_EACH_LOG_LINK (next1, link)
1331 : {
1332 17825723 : rtx_insn *link1 = next1->insn;
1333 17825723 : if (NOTE_P (link1))
1334 77 : continue;
1335 : /* I0 -> I1 -> I2 -> I3. */
1336 29117153 : FOR_EACH_LOG_LINK (nextlinks, link1)
1337 11292595 : if ((next = try_combine (insn, link, link1,
1338 : nextlinks->insn,
1339 : &new_direct_jump_p,
1340 : last_combined_insn)) != 0)
1341 : {
1342 1088 : statistics_counter_event (cfun, "four-insn combine", 1);
1343 1088 : goto retry;
1344 : }
1345 : /* I0, I1 -> I2, I2 -> I3. */
1346 21821760 : for (nextlinks = next1->next; nextlinks;
1347 3997202 : nextlinks = nextlinks->next)
1348 3997425 : if ((next = try_combine (insn, link, link1,
1349 : nextlinks->insn,
1350 : &new_direct_jump_p,
1351 : last_combined_insn)) != 0)
1352 : {
1353 223 : statistics_counter_event (cfun, "four-insn combine", 1);
1354 223 : goto retry;
1355 : }
1356 : }
1357 :
1358 48804107 : for (next1 = links->next; next1; next1 = next1->next)
1359 : {
1360 12462836 : rtx_insn *link1 = next1->insn;
1361 12462836 : if (NOTE_P (link1))
1362 8 : continue;
1363 : /* I0 -> I2; I1, I2 -> I3. */
1364 15756088 : FOR_EACH_LOG_LINK (nextlinks, link)
1365 3296706 : if ((next = try_combine (insn, link, link1,
1366 : nextlinks->insn,
1367 : &new_direct_jump_p,
1368 : last_combined_insn)) != 0)
1369 : {
1370 3446 : statistics_counter_event (cfun, "four-insn combine", 1);
1371 3446 : goto retry;
1372 : }
1373 : /* I0 -> I1; I1, I2 -> I3. */
1374 15944990 : FOR_EACH_LOG_LINK (nextlinks, link1)
1375 3485786 : if ((next = try_combine (insn, link, link1,
1376 : nextlinks->insn,
1377 : &new_direct_jump_p,
1378 : last_combined_insn)) != 0)
1379 : {
1380 178 : statistics_counter_event (cfun, "four-insn combine", 1);
1381 178 : goto retry;
1382 : }
1383 : }
1384 : }
1385 :
1386 : /* Try this insn with each REG_EQUAL note it links back to. */
1387 98296381 : FOR_EACH_LOG_LINK (links, insn)
1388 : {
1389 36396948 : rtx set, note;
1390 36396948 : rtx_insn *temp = links->insn;
1391 36396948 : if ((set = single_set (temp)) != 0
1392 36007892 : && (note = find_reg_equal_equiv_note (temp)) != 0
1393 2545606 : && (note = XEXP (note, 0), GET_CODE (note)) != EXPR_LIST
1394 2545606 : && ! side_effects_p (SET_SRC (set))
1395 : /* Avoid using a register that may already been marked
1396 : dead by an earlier instruction. */
1397 2545606 : && ! unmentioned_reg_p (note, SET_SRC (set))
1398 37628116 : && (GET_MODE (note) == VOIDmode
1399 27439 : ? SCALAR_INT_MODE_P (GET_MODE (SET_DEST (set)))
1400 1203729 : : (GET_MODE (SET_DEST (set)) == GET_MODE (note)
1401 1203702 : && (GET_CODE (SET_DEST (set)) != ZERO_EXTRACT
1402 0 : || (GET_MODE (XEXP (SET_DEST (set), 0))
1403 : == GET_MODE (note))))))
1404 : {
1405 : /* Temporarily replace the set's source with the
1406 : contents of the REG_EQUAL note. The insn will
1407 : be deleted or recognized by try_combine. */
1408 1231124 : rtx orig_src = SET_SRC (set);
1409 1231124 : rtx orig_dest = SET_DEST (set);
1410 1231124 : if (GET_CODE (SET_DEST (set)) == ZERO_EXTRACT)
1411 0 : SET_DEST (set) = XEXP (SET_DEST (set), 0);
1412 1231124 : SET_SRC (set) = note;
1413 1231124 : i2mod = temp;
1414 1231124 : i2mod_old_rhs = copy_rtx (orig_src);
1415 1231124 : i2mod_new_rhs = copy_rtx (note);
1416 1231124 : next = try_combine (insn, i2mod, NULL, NULL,
1417 : &new_direct_jump_p,
1418 : last_combined_insn);
1419 1231124 : i2mod = NULL;
1420 1231124 : if (next)
1421 : {
1422 33327 : statistics_counter_event (cfun, "insn-with-note combine", 1);
1423 33327 : goto retry;
1424 : }
1425 1197797 : INSN_CODE (temp) = -1;
1426 1197797 : SET_SRC (set) = orig_src;
1427 1197797 : SET_DEST (set) = orig_dest;
1428 : }
1429 : }
1430 :
1431 61899433 : if (!NOTE_P (insn))
1432 61899433 : record_dead_and_set_regs (insn);
1433 :
1434 131165263 : retry:
1435 131165263 : ;
1436 : }
1437 : }
1438 :
1439 999512 : default_rtl_profile ();
1440 999512 : clear_bb_flags ();
1441 :
1442 999512 : if (purge_all_dead_edges ())
1443 1436 : new_direct_jump_p = true;
1444 999512 : if (delete_noop_moves ())
1445 0 : new_direct_jump_p = true;
1446 :
1447 : /* Clean up. */
1448 999512 : obstack_free (&insn_link_obstack, NULL);
1449 999512 : free (uid_log_links);
1450 999512 : free (uid_insn_cost);
1451 999512 : reg_stat.release ();
1452 :
1453 999512 : {
1454 999512 : struct undo *undo, *next;
1455 5646304 : for (undo = undobuf.frees; undo; undo = next)
1456 : {
1457 4646792 : next = undo->next;
1458 4646792 : free (undo);
1459 : }
1460 999512 : undobuf.frees = 0;
1461 : }
1462 :
1463 999512 : statistics_counter_event (cfun, "attempts", combine_attempts);
1464 999512 : statistics_counter_event (cfun, "merges", combine_merges);
1465 999512 : statistics_counter_event (cfun, "extras", combine_extras);
1466 999512 : statistics_counter_event (cfun, "successes", combine_successes);
1467 :
1468 999512 : nonzero_sign_valid = 0;
1469 999512 : rtl_hooks = general_rtl_hooks;
1470 :
1471 : /* Make recognizer allow volatile MEMs again. */
1472 999512 : init_recog ();
1473 :
1474 999512 : return new_direct_jump_p;
1475 : }
1476 :
1477 : /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1478 :
1479 : static void
1480 999512 : init_reg_last (void)
1481 : {
1482 999512 : unsigned int i;
1483 999512 : reg_stat_type *p;
1484 :
1485 141964010 : FOR_EACH_VEC_ELT (reg_stat, i, p)
1486 140964498 : memset (p, 0, offsetof (reg_stat_type, sign_bit_copies));
1487 999512 : }
1488 :
1489 : /* Set up any promoted values for incoming argument registers. */
1490 :
1491 : static void
1492 1999024 : setup_incoming_promotions (rtx_insn *first)
1493 : {
1494 1999024 : tree arg;
1495 1999024 : bool strictly_local = false;
1496 :
1497 5416328 : for (arg = DECL_ARGUMENTS (current_function_decl); arg;
1498 3417304 : arg = DECL_CHAIN (arg))
1499 : {
1500 3417304 : rtx x, reg = DECL_INCOMING_RTL (arg);
1501 3417304 : int uns1, uns3;
1502 3417304 : machine_mode mode1, mode2, mode3, mode4;
1503 :
1504 : /* Only continue if the incoming argument is in a register. */
1505 3417304 : if (!REG_P (reg))
1506 3417210 : continue;
1507 :
1508 : /* Determine, if possible, whether all call sites of the current
1509 : function lie within the current compilation unit. (This does
1510 : take into account the exporting of a function via taking its
1511 : address, and so forth.) */
1512 2681302 : strictly_local
1513 2681302 : = cgraph_node::local_info_node (current_function_decl)->local;
1514 :
1515 : /* The mode and signedness of the argument before any promotions happen
1516 : (equal to the mode of the pseudo holding it at that stage). */
1517 2681302 : mode1 = TYPE_MODE (TREE_TYPE (arg));
1518 2681302 : uns1 = TYPE_UNSIGNED (TREE_TYPE (arg));
1519 :
1520 : /* The mode and signedness of the argument after any source language and
1521 : TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1522 2681302 : mode2 = TYPE_MODE (DECL_ARG_TYPE (arg));
1523 2681302 : uns3 = TYPE_UNSIGNED (DECL_ARG_TYPE (arg));
1524 :
1525 : /* The mode and signedness of the argument as it is actually passed,
1526 : see assign_parm_setup_reg in function.cc. */
1527 2681302 : mode3 = promote_function_mode (TREE_TYPE (arg), mode1, &uns3,
1528 2681302 : TREE_TYPE (cfun->decl), 0);
1529 :
1530 : /* The mode of the register in which the argument is being passed. */
1531 2681302 : mode4 = GET_MODE (reg);
1532 :
1533 : /* Eliminate sign extensions in the callee when:
1534 : (a) A mode promotion has occurred; */
1535 2681302 : if (mode1 == mode3)
1536 2681208 : continue;
1537 : /* (b) The mode of the register is the same as the mode of
1538 : the argument as it is passed; */
1539 94 : if (mode3 != mode4)
1540 0 : continue;
1541 : /* (c) There's no language level extension; */
1542 94 : if (mode1 == mode2)
1543 : ;
1544 : /* (c.1) All callers are from the current compilation unit. If that's
1545 : the case we don't have to rely on an ABI, we only have to know
1546 : what we're generating right now, and we know that we will do the
1547 : mode1 to mode2 promotion with the given sign. */
1548 0 : else if (!strictly_local)
1549 0 : continue;
1550 : /* (c.2) The combination of the two promotions is useful. This is
1551 : true when the signs match, or if the first promotion is unsigned.
1552 : In the later case, (sign_extend (zero_extend x)) is the same as
1553 : (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
1554 0 : else if (uns1)
1555 0 : uns3 = true;
1556 0 : else if (uns3)
1557 0 : continue;
1558 :
1559 : /* Record that the value was promoted from mode1 to mode3,
1560 : so that any sign extension at the head of the current
1561 : function may be eliminated. */
1562 94 : x = gen_rtx_CLOBBER (mode1, const0_rtx);
1563 94 : x = gen_rtx_fmt_e ((uns3 ? ZERO_EXTEND : SIGN_EXTEND), mode3, x);
1564 94 : record_value_for_reg (reg, first, x);
1565 : }
1566 1999024 : }
1567 :
1568 : /* If MODE has a precision lower than PREC and SRC is a non-negative constant
1569 : that would appear negative in MODE, sign-extend SRC for use in nonzero_bits
1570 : because some machines (maybe most) will actually do the sign-extension and
1571 : this is the conservative approach.
1572 :
1573 : ??? For 2.5, try to tighten up the MD files in this regard instead of this
1574 : kludge. */
1575 :
1576 : static rtx
1577 0 : sign_extend_short_imm (rtx src, machine_mode mode, unsigned int prec)
1578 : {
1579 0 : scalar_int_mode int_mode;
1580 0 : if (CONST_INT_P (src)
1581 0 : && is_a <scalar_int_mode> (mode, &int_mode)
1582 0 : && GET_MODE_PRECISION (int_mode) < prec
1583 0 : && INTVAL (src) > 0
1584 0 : && val_signbit_known_set_p (int_mode, INTVAL (src)))
1585 0 : src = GEN_INT (INTVAL (src) | ~GET_MODE_MASK (int_mode));
1586 :
1587 0 : return src;
1588 : }
1589 :
1590 : /* Update RSP for pseudo-register X from INSN's REG_EQUAL note (if one exists)
1591 : and SET. */
1592 :
1593 : static void
1594 23303659 : update_rsp_from_reg_equal (reg_stat_type *rsp, rtx_insn *insn, const_rtx set,
1595 : rtx x)
1596 : {
1597 23303659 : rtx reg_equal_note = insn ? find_reg_equal_equiv_note (insn) : NULL_RTX;
1598 23303659 : unsigned HOST_WIDE_INT bits = 0;
1599 23303659 : rtx reg_equal = NULL, src = SET_SRC (set);
1600 23303659 : unsigned int num = 0;
1601 :
1602 23303659 : if (reg_equal_note)
1603 993594 : reg_equal = XEXP (reg_equal_note, 0);
1604 :
1605 23303659 : if (SHORT_IMMEDIATES_SIGN_EXTEND)
1606 : {
1607 : src = sign_extend_short_imm (src, GET_MODE (x), BITS_PER_WORD);
1608 : if (reg_equal)
1609 : reg_equal = sign_extend_short_imm (reg_equal, GET_MODE (x), BITS_PER_WORD);
1610 : }
1611 :
1612 : /* Don't call nonzero_bits if it cannot change anything. */
1613 23303659 : if (rsp->nonzero_bits != HOST_WIDE_INT_M1U)
1614 : {
1615 20122108 : machine_mode mode = GET_MODE (x);
1616 20122108 : if (GET_MODE_CLASS (mode) == MODE_INT
1617 20122108 : && HWI_COMPUTABLE_MODE_P (mode))
1618 20121976 : mode = nonzero_bits_mode;
1619 20122108 : bits = nonzero_bits (src, mode);
1620 20122108 : if (reg_equal && bits)
1621 939654 : bits &= nonzero_bits (reg_equal, mode);
1622 20122108 : rsp->nonzero_bits |= bits;
1623 : }
1624 :
1625 : /* Don't call num_sign_bit_copies if it cannot change anything. */
1626 23303659 : if (rsp->sign_bit_copies != 1)
1627 : {
1628 19972974 : num = num_sign_bit_copies (SET_SRC (set), GET_MODE (x));
1629 19972974 : if (reg_equal && maybe_ne (num, GET_MODE_PRECISION (GET_MODE (x))))
1630 : {
1631 936210 : unsigned int numeq = num_sign_bit_copies (reg_equal, GET_MODE (x));
1632 936210 : if (num == 0 || numeq > num)
1633 19972974 : num = numeq;
1634 : }
1635 19972974 : if (rsp->sign_bit_copies == 0 || num < rsp->sign_bit_copies)
1636 19278987 : rsp->sign_bit_copies = num;
1637 : }
1638 23303659 : }
1639 :
1640 : /* Called via note_stores. If X is a pseudo that is narrower than
1641 : HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1642 :
1643 : If we are setting only a portion of X and we can't figure out what
1644 : portion, assume all bits will be used since we don't know what will
1645 : be happening.
1646 :
1647 : Similarly, set how many bits of X are known to be copies of the sign bit
1648 : at all locations in the function. This is the smallest number implied
1649 : by any set of X. */
1650 :
1651 : static void
1652 72241784 : set_nonzero_bits_and_sign_copies (rtx x, const_rtx set, void *data)
1653 : {
1654 72241784 : rtx_insn *insn = (rtx_insn *) data;
1655 72241784 : scalar_int_mode mode;
1656 :
1657 72241784 : if (REG_P (x)
1658 58173211 : && REGNO (x) >= FIRST_PSEUDO_REGISTER
1659 : /* If this register is undefined at the start of the file, we can't
1660 : say what its contents were. */
1661 57892746 : && ! REGNO_REG_SET_P
1662 : (DF_LR_IN (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb), REGNO (x))
1663 28846896 : && is_a <scalar_int_mode> (GET_MODE (x), &mode)
1664 96456956 : && HWI_COMPUTABLE_MODE_P (mode))
1665 : {
1666 23501811 : reg_stat_type *rsp = ®_stat[REGNO (x)];
1667 :
1668 23501811 : if (set == 0 || GET_CODE (set) == CLOBBER)
1669 : {
1670 21734 : rsp->nonzero_bits = GET_MODE_MASK (mode);
1671 21734 : rsp->sign_bit_copies = 1;
1672 21734 : return;
1673 : }
1674 :
1675 : /* If this register is being initialized using itself, and the
1676 : register is uninitialized in this basic block, and there are
1677 : no LOG_LINKS which set the register, then part of the
1678 : register is uninitialized. In that case we can't assume
1679 : anything about the number of nonzero bits.
1680 :
1681 : ??? We could do better if we checked this in
1682 : reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1683 : could avoid making assumptions about the insn which initially
1684 : sets the register, while still using the information in other
1685 : insns. We would have to be careful to check every insn
1686 : involved in the combination. */
1687 :
1688 23480077 : if (insn
1689 22065189 : && reg_referenced_p (x, PATTERN (insn))
1690 25993135 : && !REGNO_REG_SET_P (DF_LR_IN (BLOCK_FOR_INSN (insn)),
1691 : REGNO (x)))
1692 : {
1693 246789 : struct insn_link *link;
1694 :
1695 365938 : FOR_EACH_LOG_LINK (link, insn)
1696 281802 : if (dead_or_set_p (link->insn, x))
1697 : break;
1698 246789 : if (!link)
1699 : {
1700 84136 : rsp->nonzero_bits = GET_MODE_MASK (mode);
1701 84136 : rsp->sign_bit_copies = 1;
1702 84136 : return;
1703 : }
1704 : }
1705 :
1706 : /* If this is a complex assignment, see if we can convert it into a
1707 : simple assignment. */
1708 23395941 : set = expand_field_assignment (set);
1709 :
1710 : /* If this is a simple assignment, or we have a paradoxical SUBREG,
1711 : set what we know about X. */
1712 :
1713 23395941 : if (SET_DEST (set) == x
1714 23395941 : || (paradoxical_subreg_p (SET_DEST (set))
1715 4303 : && SUBREG_REG (SET_DEST (set)) == x))
1716 23303659 : update_rsp_from_reg_equal (rsp, insn, set, x);
1717 : else
1718 : {
1719 92282 : rsp->nonzero_bits = GET_MODE_MASK (mode);
1720 92282 : rsp->sign_bit_copies = 1;
1721 : }
1722 : }
1723 : }
1724 :
1725 : /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1726 : optionally insns that were previously combined into I3 or that will be
1727 : combined into the merger of INSN and I3. The order is PRED, PRED2,
1728 : INSN, SUCC, SUCC2, I3.
1729 :
1730 : Return false if the combination is not allowed for any reason.
1731 :
1732 : If the combination is allowed, *PDEST will be set to the single
1733 : destination of INSN and *PSRC to the single source, and this function
1734 : will return true. */
1735 :
1736 : static bool
1737 60524818 : can_combine_p (rtx_insn *insn, rtx_insn *i3, rtx_insn *pred ATTRIBUTE_UNUSED,
1738 : rtx_insn *pred2 ATTRIBUTE_UNUSED, rtx_insn *succ, rtx_insn *succ2,
1739 : rtx *pdest, rtx *psrc)
1740 : {
1741 60524818 : int i;
1742 60524818 : const_rtx set = 0;
1743 60524818 : rtx src, dest;
1744 60524818 : rtx_insn *p;
1745 60524818 : rtx link;
1746 60524818 : bool all_adjacent = true;
1747 60524818 : bool (*is_volatile_p) (const_rtx);
1748 :
1749 60524818 : if (succ)
1750 : {
1751 14150703 : if (succ2)
1752 : {
1753 2064834 : if (next_active_insn (succ2) != i3)
1754 193718 : all_adjacent = false;
1755 2064834 : if (next_active_insn (succ) != succ2)
1756 2004749 : all_adjacent = false;
1757 : }
1758 12085869 : else if (next_active_insn (succ) != i3)
1759 2004749 : all_adjacent = false;
1760 14150703 : if (next_active_insn (insn) != succ)
1761 16841434 : all_adjacent = false;
1762 : }
1763 46374115 : else if (next_active_insn (insn) != i3)
1764 16841434 : all_adjacent = false;
1765 :
1766 : /* Can combine only if previous insn is a SET of a REG or a SUBREG,
1767 : or a PARALLEL consisting of such a SET and CLOBBERs.
1768 :
1769 : If INSN has CLOBBER parallel parts, ignore them for our processing.
1770 : By definition, these happen during the execution of the insn. When it
1771 : is merged with another insn, all bets are off. If they are, in fact,
1772 : needed and aren't also supplied in I3, they may be added by
1773 : recog_for_combine. Otherwise, it won't match.
1774 :
1775 : We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1776 : note.
1777 :
1778 : Get the source and destination of INSN. If more than one, can't
1779 : combine. */
1780 :
1781 60524818 : if (GET_CODE (PATTERN (insn)) == SET)
1782 : set = PATTERN (insn);
1783 15930487 : else if (GET_CODE (PATTERN (insn)) == PARALLEL
1784 15930487 : && GET_CODE (XVECEXP (PATTERN (insn), 0, 0)) == SET)
1785 : {
1786 47708124 : for (i = 0; i < XVECLEN (PATTERN (insn), 0); i++)
1787 : {
1788 32302704 : rtx elt = XVECEXP (PATTERN (insn), 0, i);
1789 :
1790 32302704 : switch (GET_CODE (elt))
1791 : {
1792 : /* This is important to combine floating point insns
1793 : for the SH4 port. */
1794 126802 : case USE:
1795 : /* Combining an isolated USE doesn't make sense.
1796 : We depend here on combinable_i3pat to reject them. */
1797 : /* The code below this loop only verifies that the inputs of
1798 : the SET in INSN do not change. We call reg_set_between_p
1799 : to verify that the REG in the USE does not change between
1800 : I3 and INSN.
1801 : If the USE in INSN was for a pseudo register, the matching
1802 : insn pattern will likely match any register; combining this
1803 : with any other USE would only be safe if we knew that the
1804 : used registers have identical values, or if there was
1805 : something to tell them apart, e.g. different modes. For
1806 : now, we forgo such complicated tests and simply disallow
1807 : combining of USES of pseudo registers with any other USE. */
1808 126802 : if (REG_P (XEXP (elt, 0))
1809 126802 : && GET_CODE (PATTERN (i3)) == PARALLEL)
1810 : {
1811 227 : rtx i3pat = PATTERN (i3);
1812 227 : int i = XVECLEN (i3pat, 0) - 1;
1813 227 : unsigned int regno = REGNO (XEXP (elt, 0));
1814 :
1815 465 : do
1816 : {
1817 465 : rtx i3elt = XVECEXP (i3pat, 0, i);
1818 :
1819 465 : if (GET_CODE (i3elt) == USE
1820 209 : && REG_P (XEXP (i3elt, 0))
1821 701 : && (REGNO (XEXP (i3elt, 0)) == regno
1822 182 : ? reg_set_between_p (XEXP (elt, 0),
1823 27 : PREV_INSN (insn), i3)
1824 : : regno >= FIRST_PSEUDO_REGISTER))
1825 182 : return false;
1826 : }
1827 283 : while (--i >= 0);
1828 : }
1829 : break;
1830 :
1831 : /* We can ignore CLOBBERs. */
1832 : case CLOBBER:
1833 : break;
1834 :
1835 16524450 : case SET:
1836 : /* Ignore SETs whose result isn't used but not those that
1837 : have side-effects. */
1838 16524450 : if (find_reg_note (insn, REG_UNUSED, SET_DEST (elt))
1839 186822 : && insn_nothrow_p (insn)
1840 16698111 : && !side_effects_p (elt))
1841 : break;
1842 :
1843 : /* If we have already found a SET, this is a second one and
1844 : so we cannot combine with this insn. */
1845 16433339 : if (set)
1846 : return false;
1847 :
1848 : set = elt;
1849 : break;
1850 :
1851 : default:
1852 : /* Anything else means we can't combine. */
1853 : return false;
1854 : }
1855 : }
1856 :
1857 15405420 : if (set == 0
1858 : /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1859 : so don't do anything with it. */
1860 15405420 : || GET_CODE (SET_SRC (set)) == ASM_OPERANDS)
1861 : return false;
1862 : }
1863 : else
1864 : return false;
1865 :
1866 : if (set == 0)
1867 : return false;
1868 :
1869 : /* The simplification in expand_field_assignment may call back to
1870 : get_last_value, so set safe guard here. */
1871 59980855 : subst_low_luid = DF_INSN_LUID (insn);
1872 :
1873 59980855 : set = expand_field_assignment (set);
1874 59980855 : src = SET_SRC (set), dest = SET_DEST (set);
1875 :
1876 : /* Do not eliminate user-specified register if it is in an
1877 : asm input because we may break the register asm usage defined
1878 : in GCC manual if allow to do so.
1879 : Be aware that this may cover more cases than we expect but this
1880 : should be harmless. */
1881 59474054 : if (REG_P (dest) && REG_USERVAR_P (dest) && HARD_REGISTER_P (dest)
1882 59980858 : && extract_asm_operands (PATTERN (i3)))
1883 : return false;
1884 :
1885 : /* Don't eliminate a store in the stack pointer. */
1886 59980855 : if (dest == stack_pointer_rtx
1887 : /* Don't combine with an insn that sets a register to itself if it has
1888 : a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1889 58083146 : || (rtx_equal_p (src, dest) && find_reg_note (insn, REG_EQUAL, NULL_RTX))
1890 : /* Can't merge an ASM_OPERANDS. */
1891 58083146 : || GET_CODE (src) == ASM_OPERANDS
1892 : /* Can't merge a function call. */
1893 58079532 : || GET_CODE (src) == CALL
1894 : /* Don't eliminate a function call argument. */
1895 58079532 : || (CALL_P (i3)
1896 8729994 : && (find_reg_fusage (i3, USE, dest)
1897 174690 : || (REG_P (dest)
1898 174690 : && REGNO (dest) < FIRST_PSEUDO_REGISTER
1899 283 : && global_regs[REGNO (dest)])))
1900 : /* Don't substitute into an incremented register. */
1901 : || FIND_REG_INC_NOTE (i3, dest)
1902 : || (succ && FIND_REG_INC_NOTE (succ, dest))
1903 58079532 : || (succ2 && FIND_REG_INC_NOTE (succ2, dest))
1904 : /* Don't substitute into a non-local goto, this confuses CFG. */
1905 49524225 : || (JUMP_P (i3) && find_reg_note (i3, REG_NON_LOCAL_GOTO, NULL_RTX))
1906 : /* Make sure that DEST is not used after INSN but before SUCC, or
1907 : after SUCC and before SUCC2, or after SUCC2 but before I3. */
1908 49523504 : || (!all_adjacent
1909 12125887 : && ((succ2
1910 917690 : && (reg_used_between_p (dest, succ2, i3)
1911 897470 : || reg_used_between_p (dest, succ, succ2)))
1912 12073555 : || (!succ2 && succ && reg_used_between_p (dest, succ, i3))
1913 11806995 : || (!succ2 && !succ && reg_used_between_p (dest, insn, i3))
1914 11806995 : || (succ
1915 : /* SUCC and SUCC2 can be split halves from a PARALLEL; in
1916 : that case SUCC is not in the insn stream, so use SUCC2
1917 : instead for this test. */
1918 9754706 : && reg_used_between_p (dest, insn,
1919 : succ2
1920 865358 : && INSN_UID (succ) == INSN_UID (succ2)
1921 : ? succ2 : succ))))
1922 : /* Make sure that the value that is to be substituted for the register
1923 : does not use any registers whose values alter in between. However,
1924 : If the insns are adjacent, a use can't cross a set even though we
1925 : think it might (this can happen for a sequence of insns each setting
1926 : the same destination; last_set of that register might point to
1927 : a NOTE). If INSN has a REG_EQUIV note, the register is always
1928 : equivalent to the memory so the substitution is valid even if there
1929 : are intervening stores. Also, don't move a volatile asm or
1930 : UNSPEC_VOLATILE across any other insns. */
1931 : || (! all_adjacent
1932 11806995 : && (((!MEM_P (src)
1933 3300018 : || ! find_reg_note (insn, REG_EQUIV, src))
1934 11692926 : && modified_between_p (src, insn, i3))
1935 10772426 : || (GET_CODE (src) == ASM_OPERANDS && MEM_VOLATILE_P (src))
1936 10772426 : || GET_CODE (src) == UNSPEC_VOLATILE))
1937 : /* Don't combine across a CALL_INSN, because that would possibly
1938 : change whether the life span of some REGs crosses calls or not,
1939 : and it is a pain to update that information.
1940 : Exception: if source is a constant, moving it later can't hurt.
1941 : Accept that as a special case. */
1942 108140292 : || (DF_INSN_LUID (insn) < last_call_luid && ! CONSTANT_P (src)))
1943 12146830 : return false;
1944 :
1945 : /* DEST must be a REG. */
1946 47834025 : if (REG_P (dest))
1947 : {
1948 : /* If register alignment is being enforced for multi-word items in all
1949 : cases except for parameters, it is possible to have a register copy
1950 : insn referencing a hard register that is not allowed to contain the
1951 : mode being copied and which would not be valid as an operand of most
1952 : insns. Eliminate this problem by not combining with such an insn.
1953 :
1954 : Also, on some machines we don't want to extend the life of a hard
1955 : register. */
1956 :
1957 47332060 : if (REG_P (src)
1958 47332060 : && ((REGNO (dest) < FIRST_PSEUDO_REGISTER
1959 29350 : && !targetm.hard_regno_mode_ok (REGNO (dest), GET_MODE (dest)))
1960 : /* Don't extend the life of a hard register unless it is
1961 : user variable (if we have few registers) or it can't
1962 : fit into the desired register (meaning something special
1963 : is going on).
1964 : Also avoid substituting a return register into I3, because
1965 : reload can't handle a conflict with constraints of other
1966 : inputs. */
1967 2542435 : || (REGNO (src) < FIRST_PSEUDO_REGISTER
1968 37187 : && !targetm.hard_regno_mode_ok (REGNO (src),
1969 37187 : GET_MODE (src)))))
1970 0 : return false;
1971 : }
1972 : else
1973 : return false;
1974 :
1975 :
1976 47332060 : if (GET_CODE (PATTERN (i3)) == PARALLEL)
1977 34826099 : for (i = XVECLEN (PATTERN (i3), 0) - 1; i >= 0; i--)
1978 23460229 : if (GET_CODE (XVECEXP (PATTERN (i3), 0, i)) == CLOBBER)
1979 : {
1980 11048056 : rtx reg = XEXP (XVECEXP (PATTERN (i3), 0, i), 0);
1981 :
1982 : /* If the clobber represents an earlyclobber operand, we must not
1983 : substitute an expression containing the clobbered register.
1984 : As we do not analyze the constraint strings here, we have to
1985 : make the conservative assumption. However, if the register is
1986 : a fixed hard reg, the clobber cannot represent any operand;
1987 : we leave it up to the machine description to either accept or
1988 : reject use-and-clobber patterns. */
1989 11048056 : if (!REG_P (reg)
1990 10702404 : || REGNO (reg) >= FIRST_PSEUDO_REGISTER
1991 21703176 : || !fixed_regs[REGNO (reg)])
1992 431236 : if (reg_overlap_mentioned_p (reg, src))
1993 : return false;
1994 : }
1995 :
1996 : /* If INSN contains anything volatile, or is an `asm' (whether volatile
1997 : or not), reject, unless nothing volatile comes between it and I3 */
1998 :
1999 47331388 : if (GET_CODE (src) == ASM_OPERANDS || volatile_refs_p (src))
2000 : {
2001 : /* Make sure neither succ nor succ2 contains a volatile reference. */
2002 694555 : if (succ2 != 0 && volatile_refs_p (PATTERN (succ2)))
2003 : return false;
2004 694462 : if (succ != 0 && volatile_refs_p (PATTERN (succ)))
2005 : return false;
2006 : /* We'll check insns between INSN and I3 below. */
2007 : }
2008 :
2009 : /* If INSN is an asm, and DEST is a hard register, reject, since it has
2010 : to be an explicit register variable, and was chosen for a reason. */
2011 :
2012 47295953 : if (GET_CODE (src) == ASM_OPERANDS
2013 47295953 : && REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER)
2014 : return false;
2015 :
2016 : /* If INSN contains volatile references (specifically volatile MEMs),
2017 : we cannot combine across any other volatile references.
2018 : Even if INSN doesn't contain volatile references, any intervening
2019 : volatile insn might affect machine state. */
2020 :
2021 93931918 : is_volatile_p = volatile_refs_p (PATTERN (insn))
2022 47295953 : ? volatile_refs_p
2023 : : volatile_insn_p;
2024 :
2025 210456335 : for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
2026 116069001 : if (INSN_P (p) && p != succ && p != succ2 && is_volatile_p (PATTERN (p)))
2027 : return false;
2028 :
2029 : /* If INSN contains an autoincrement or autodecrement, make sure that
2030 : register is not used between there and I3, and not already used in
2031 : I3 either. Neither must it be used in PRED or SUCC, if they exist.
2032 : Also insist that I3 not be a jump if using LRA; if it were one
2033 : and the incremented register were spilled, we would lose.
2034 : Reload handles this correctly. */
2035 :
2036 47091381 : if (AUTO_INC_DEC)
2037 : for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
2038 : if (REG_NOTE_KIND (link) == REG_INC
2039 : && ((JUMP_P (i3) && targetm.lra_p ())
2040 : || reg_used_between_p (XEXP (link, 0), insn, i3)
2041 : || (pred != NULL_RTX
2042 : && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred)))
2043 : || (pred2 != NULL_RTX
2044 : && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred2)))
2045 : || (succ != NULL_RTX
2046 : && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ)))
2047 : || (succ2 != NULL_RTX
2048 : && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ2)))
2049 : || reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i3))))
2050 : return false;
2051 :
2052 : /* If we get here, we have passed all the tests and the combination is
2053 : to be allowed. */
2054 :
2055 47091381 : *pdest = dest;
2056 47091381 : *psrc = src;
2057 :
2058 47091381 : return true;
2059 : }
2060 :
2061 : /* LOC is the location within I3 that contains its pattern or the component
2062 : of a PARALLEL of the pattern. We validate that it is valid for combining.
2063 :
2064 : One problem is if I3 modifies its output, as opposed to replacing it
2065 : entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2066 : doing so would produce an insn that is not equivalent to the original insns.
2067 :
2068 : Consider:
2069 :
2070 : (set (reg:DI 101) (reg:DI 100))
2071 : (set (subreg:SI (reg:DI 101) 0) <foo>)
2072 :
2073 : This is NOT equivalent to:
2074 :
2075 : (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2076 : (set (reg:DI 101) (reg:DI 100))])
2077 :
2078 : Not only does this modify 100 (in which case it might still be valid
2079 : if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2080 :
2081 : We can also run into a problem if I2 sets a register that I1
2082 : uses and I1 gets directly substituted into I3 (not via I2). In that
2083 : case, we would be getting the wrong value of I2DEST into I3, so we
2084 : must reject the combination. This case occurs when I2 and I1 both
2085 : feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2086 : If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2087 : of a SET must prevent combination from occurring. The same situation
2088 : can occur for I0, in which case I0_NOT_IN_SRC is set.
2089 :
2090 : Before doing the above check, we first try to expand a field assignment
2091 : into a set of logical operations.
2092 :
2093 : If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2094 : we place a register that is both set and used within I3. If more than one
2095 : such register is detected, we fail.
2096 :
2097 : Return true if the combination is valid, false otherwise. */
2098 :
2099 : static bool
2100 67217136 : combinable_i3pat (rtx_insn *i3, rtx *loc, rtx i2dest, rtx i1dest, rtx i0dest,
2101 : bool i1_not_in_src, bool i0_not_in_src, rtx *pi3dest_killed)
2102 : {
2103 67217136 : rtx x = *loc;
2104 :
2105 67217136 : if (GET_CODE (x) == SET)
2106 : {
2107 45627290 : rtx set = x ;
2108 45627290 : rtx dest = SET_DEST (set);
2109 45627290 : rtx src = SET_SRC (set);
2110 45627290 : rtx inner_dest = dest;
2111 45627290 : rtx subdest;
2112 :
2113 45627290 : while (GET_CODE (inner_dest) == STRICT_LOW_PART
2114 46120683 : || GET_CODE (inner_dest) == SUBREG
2115 46120683 : || GET_CODE (inner_dest) == ZERO_EXTRACT)
2116 493393 : inner_dest = XEXP (inner_dest, 0);
2117 :
2118 : /* Check for the case where I3 modifies its output, as discussed
2119 : above. We don't want to prevent pseudos from being combined
2120 : into the address of a MEM, so only prevent the combination if
2121 : i1 or i2 set the same MEM. */
2122 473038 : if ((inner_dest != dest &&
2123 : (!MEM_P (inner_dest)
2124 793 : || rtx_equal_p (i2dest, inner_dest)
2125 793 : || (i1dest && rtx_equal_p (i1dest, inner_dest))
2126 793 : || (i0dest && rtx_equal_p (i0dest, inner_dest)))
2127 472245 : && (reg_overlap_mentioned_p (i2dest, inner_dest)
2128 351333 : || (i1dest && reg_overlap_mentioned_p (i1dest, inner_dest))
2129 350076 : || (i0dest && reg_overlap_mentioned_p (i0dest, inner_dest))))
2130 :
2131 : /* This is the same test done in can_combine_p except we can't test
2132 : all_adjacent; we don't have to, since this instruction will stay
2133 : in place, thus we are not considering increasing the lifetime of
2134 : INNER_DEST.
2135 :
2136 : Also, if this insn sets a function argument, combining it with
2137 : something that might need a spill could clobber a previous
2138 : function argument; the all_adjacent test in can_combine_p also
2139 : checks this; here, we do a more specific test for this case. */
2140 :
2141 45505012 : || (REG_P (inner_dest)
2142 28916880 : && REGNO (inner_dest) < FIRST_PSEUDO_REGISTER
2143 7281022 : && !targetm.hard_regno_mode_ok (REGNO (inner_dest),
2144 7281022 : GET_MODE (inner_dest)))
2145 45505012 : || (i1_not_in_src && reg_overlap_mentioned_p (i1dest, src))
2146 91125605 : || (i0_not_in_src && reg_overlap_mentioned_p (i0dest, src)))
2147 153115 : return false;
2148 :
2149 : /* If DEST is used in I3, it is being killed in this insn, so
2150 : record that for later. We have to consider paradoxical
2151 : subregs here, since they kill the whole register, but we
2152 : ignore partial subregs, STRICT_LOW_PART, etc.
2153 : Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2154 : STACK_POINTER_REGNUM, since these are always considered to be
2155 : live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2156 45474175 : subdest = dest;
2157 45474175 : if (GET_CODE (subdest) == SUBREG && !partial_subreg_p (subdest))
2158 250954 : subdest = SUBREG_REG (subdest);
2159 45474175 : if (pi3dest_killed
2160 33119840 : && REG_P (subdest)
2161 20815190 : && reg_referenced_p (subdest, PATTERN (i3))
2162 1143866 : && REGNO (subdest) != FRAME_POINTER_REGNUM
2163 1143866 : && (HARD_FRAME_POINTER_IS_FRAME_POINTER
2164 1143866 : || REGNO (subdest) != HARD_FRAME_POINTER_REGNUM)
2165 1143866 : && (FRAME_POINTER_REGNUM == ARG_POINTER_REGNUM
2166 1143866 : || (REGNO (subdest) != ARG_POINTER_REGNUM
2167 0 : || ! fixed_regs [REGNO (subdest)]))
2168 46618041 : && REGNO (subdest) != STACK_POINTER_REGNUM)
2169 : {
2170 1105234 : if (*pi3dest_killed)
2171 : return false;
2172 :
2173 1087015 : *pi3dest_killed = subdest;
2174 : }
2175 : }
2176 :
2177 21589846 : else if (GET_CODE (x) == PARALLEL)
2178 : {
2179 : int i;
2180 :
2181 32841704 : for (i = 0; i < XVECLEN (x, 0); i++)
2182 22121511 : if (! combinable_i3pat (i3, &XVECEXP (x, 0, i), i2dest, i1dest, i0dest,
2183 : i1_not_in_src, i0_not_in_src, pi3dest_killed))
2184 : return false;
2185 : }
2186 :
2187 : return true;
2188 : }
2189 :
2190 : /* Return true if X is an arithmetic expression that contains a multiplication
2191 : and division. We don't count multiplications by powers of two here. */
2192 :
2193 : static bool
2194 17009004 : contains_muldiv (rtx x)
2195 : {
2196 17675501 : switch (GET_CODE (x))
2197 : {
2198 : case MOD: case DIV: case UMOD: case UDIV:
2199 : return true;
2200 :
2201 474897 : case MULT:
2202 474897 : return ! (CONST_INT_P (XEXP (x, 1))
2203 126097 : && pow2p_hwi (UINTVAL (XEXP (x, 1))));
2204 17041055 : default:
2205 17041055 : if (BINARY_P (x))
2206 5842129 : return contains_muldiv (XEXP (x, 0))
2207 5842129 : || contains_muldiv (XEXP (x, 1));
2208 :
2209 11198926 : if (UNARY_P (x))
2210 666497 : return contains_muldiv (XEXP (x, 0));
2211 :
2212 : return false;
2213 : }
2214 : }
2215 :
2216 : /* Determine whether INSN can be used in a combination. Return true if
2217 : not. This is used in try_combine to detect early some cases where we
2218 : can't perform combinations. */
2219 :
2220 : static bool
2221 163287917 : cant_combine_insn_p (rtx_insn *insn)
2222 : {
2223 163287917 : rtx set;
2224 163287917 : rtx src, dest;
2225 :
2226 : /* If this isn't really an insn, we can't do anything.
2227 : This can occur when flow deletes an insn that it has merged into an
2228 : auto-increment address. */
2229 163287917 : if (!NONDEBUG_INSN_P (insn))
2230 : return true;
2231 :
2232 : /* Never combine loads and stores involving hard regs that are likely
2233 : to be spilled. The register allocator can usually handle such
2234 : reg-reg moves by tying. If we allow the combiner to make
2235 : substitutions of likely-spilled regs, reload might die.
2236 : As an exception, we allow combinations involving fixed regs; these are
2237 : not available to the register allocator so there's no risk involved. */
2238 :
2239 163287503 : set = single_set (insn);
2240 163287503 : if (! set)
2241 : return false;
2242 150425337 : src = SET_SRC (set);
2243 150425337 : dest = SET_DEST (set);
2244 150425337 : if (GET_CODE (src) == SUBREG)
2245 984617 : src = SUBREG_REG (src);
2246 150425337 : if (GET_CODE (dest) == SUBREG)
2247 1570390 : dest = SUBREG_REG (dest);
2248 40500554 : if (REG_P (src) && REG_P (dest)
2249 184167148 : && ((HARD_REGISTER_P (src)
2250 6650223 : && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (src))
2251 : #ifdef LEAF_REGISTERS
2252 : && ! LEAF_REGISTERS [REGNO (src)])
2253 : #else
2254 : )
2255 : #endif
2256 27396970 : || (HARD_REGISTER_P (dest)
2257 19348105 : && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (dest))
2258 19079252 : && targetm.class_likely_spilled_p (REGNO_REG_CLASS (REGNO (dest))))))
2259 23856474 : return true;
2260 :
2261 : return false;
2262 : }
2263 :
2264 : struct likely_spilled_retval_info
2265 : {
2266 : unsigned regno, nregs;
2267 : unsigned mask;
2268 : };
2269 :
2270 : /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2271 : hard registers that are known to be written to / clobbered in full. */
2272 : static void
2273 160801 : likely_spilled_retval_1 (rtx x, const_rtx set, void *data)
2274 : {
2275 160801 : struct likely_spilled_retval_info *const info =
2276 : (struct likely_spilled_retval_info *) data;
2277 160801 : unsigned regno, nregs;
2278 160801 : unsigned new_mask;
2279 :
2280 160801 : if (!REG_P (XEXP (set, 0)))
2281 : return;
2282 160801 : regno = REGNO (x);
2283 160801 : if (regno >= info->regno + info->nregs)
2284 : return;
2285 160801 : nregs = REG_NREGS (x);
2286 160801 : if (regno + nregs <= info->regno)
2287 : return;
2288 160801 : new_mask = (2U << (nregs - 1)) - 1;
2289 160801 : if (regno < info->regno)
2290 0 : new_mask >>= info->regno - regno;
2291 : else
2292 160801 : new_mask <<= regno - info->regno;
2293 160801 : info->mask &= ~new_mask;
2294 : }
2295 :
2296 : /* Return true iff part of the return value is live during INSN, and
2297 : it is likely spilled. This can happen when more than one insn is needed
2298 : to copy the return value, e.g. when we consider to combine into the
2299 : second copy insn for a complex value. */
2300 :
2301 : static bool
2302 46715159 : likely_spilled_retval_p (rtx_insn *insn)
2303 : {
2304 46715159 : rtx_insn *use = BB_END (this_basic_block);
2305 46715159 : rtx reg;
2306 46715159 : rtx_insn *p;
2307 46715159 : unsigned regno, nregs;
2308 : /* We assume here that no machine mode needs more than
2309 : 32 hard registers when the value overlaps with a register
2310 : for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2311 46715159 : unsigned mask;
2312 46715159 : struct likely_spilled_retval_info info;
2313 :
2314 46715159 : if (!NONJUMP_INSN_P (use) || GET_CODE (PATTERN (use)) != USE || insn == use)
2315 : return false;
2316 3079863 : reg = XEXP (PATTERN (use), 0);
2317 3079863 : if (!REG_P (reg) || !targetm.calls.function_value_regno_p (REGNO (reg)))
2318 0 : return false;
2319 3079863 : regno = REGNO (reg);
2320 3079863 : nregs = REG_NREGS (reg);
2321 3079863 : if (nregs == 1)
2322 : return false;
2323 158091 : mask = (2U << (nregs - 1)) - 1;
2324 :
2325 : /* Disregard parts of the return value that are set later. */
2326 158091 : info.regno = regno;
2327 158091 : info.nregs = nregs;
2328 158091 : info.mask = mask;
2329 539967 : for (p = PREV_INSN (use); info.mask && p != insn; p = PREV_INSN (p))
2330 223785 : if (INSN_P (p))
2331 223785 : note_stores (p, likely_spilled_retval_1, &info);
2332 316182 : mask = info.mask;
2333 :
2334 : /* Check if any of the (probably) live return value registers is
2335 : likely spilled. */
2336 : nregs --;
2337 316182 : do
2338 : {
2339 316182 : if ((mask & 1 << nregs)
2340 316182 : && targetm.class_likely_spilled_p (REGNO_REG_CLASS (regno + nregs)))
2341 : return true;
2342 316172 : } while (nregs--);
2343 : return false;
2344 : }
2345 :
2346 : /* Adjust INSN after we made a change to its destination.
2347 :
2348 : Changing the destination can invalidate notes that say something about
2349 : the results of the insn and a LOG_LINK pointing to the insn. */
2350 :
2351 : static void
2352 18161 : adjust_for_new_dest (rtx_insn *insn)
2353 : {
2354 : /* For notes, be conservative and simply remove them. */
2355 18161 : remove_reg_equal_equiv_notes (insn, true);
2356 :
2357 : /* The new insn will have a destination that was previously the destination
2358 : of an insn just above it. Call distribute_links to make a LOG_LINK from
2359 : the next use of that destination. */
2360 :
2361 18161 : rtx set = single_set (insn);
2362 18161 : gcc_assert (set);
2363 :
2364 18161 : rtx reg = SET_DEST (set);
2365 :
2366 18161 : while (GET_CODE (reg) == ZERO_EXTRACT
2367 18161 : || GET_CODE (reg) == STRICT_LOW_PART
2368 36322 : || GET_CODE (reg) == SUBREG)
2369 0 : reg = XEXP (reg, 0);
2370 18161 : gcc_assert (REG_P (reg));
2371 :
2372 18161 : distribute_links (alloc_insn_link (insn, REGNO (reg), NULL));
2373 :
2374 18161 : df_insn_rescan (insn);
2375 18161 : }
2376 :
2377 : /* Return TRUE if combine can reuse reg X in mode MODE.
2378 : ADDED_SETS is trueif the original set is still required. */
2379 : static bool
2380 2653455 : can_change_dest_mode (rtx x, bool added_sets, machine_mode mode)
2381 : {
2382 2653455 : unsigned int regno;
2383 :
2384 2653455 : if (!REG_P (x))
2385 : return false;
2386 :
2387 : /* Don't change between modes with different underlying register sizes,
2388 : since this could lead to invalid subregs. */
2389 2653455 : if (maybe_ne (REGMODE_NATURAL_SIZE (mode),
2390 2653455 : REGMODE_NATURAL_SIZE (GET_MODE (x))))
2391 : return false;
2392 :
2393 2653455 : regno = REGNO (x);
2394 : /* Allow hard registers if the new mode is legal, and occupies no more
2395 : registers than the old mode. */
2396 2653455 : if (regno < FIRST_PSEUDO_REGISTER)
2397 1169187 : return (targetm.hard_regno_mode_ok (regno, mode)
2398 1169187 : && REG_NREGS (x) >= hard_regno_nregs (regno, mode));
2399 :
2400 : /* Or a pseudo that is only used once. */
2401 1484268 : return (regno < reg_n_sets_max
2402 1484266 : && REG_N_SETS (regno) == 1
2403 1432314 : && !added_sets
2404 2916582 : && !REG_USERVAR_P (x));
2405 : }
2406 :
2407 :
2408 : /* Check whether X, the destination of a set, refers to part of
2409 : the register specified by REG. */
2410 :
2411 : static bool
2412 18406 : reg_subword_p (rtx x, rtx reg)
2413 : {
2414 : /* Check that reg is an integer mode register. */
2415 18406 : if (!REG_P (reg) || GET_MODE_CLASS (GET_MODE (reg)) != MODE_INT)
2416 : return false;
2417 :
2418 17936 : if (GET_CODE (x) == STRICT_LOW_PART
2419 16055 : || GET_CODE (x) == ZERO_EXTRACT)
2420 1904 : x = XEXP (x, 0);
2421 :
2422 17936 : return GET_CODE (x) == SUBREG
2423 17729 : && !paradoxical_subreg_p (x)
2424 17729 : && SUBREG_REG (x) == reg
2425 35665 : && GET_MODE_CLASS (GET_MODE (x)) == MODE_INT;
2426 : }
2427 :
2428 : /* Return whether PAT is a PARALLEL of exactly N register SETs followed
2429 : by an arbitrary number of CLOBBERs. */
2430 : static bool
2431 100087555 : is_parallel_of_n_reg_sets (rtx pat, int n)
2432 : {
2433 100087555 : if (GET_CODE (pat) != PARALLEL)
2434 : return false;
2435 :
2436 26809778 : int len = XVECLEN (pat, 0);
2437 26809778 : if (len < n)
2438 : return false;
2439 :
2440 : int i;
2441 53256737 : for (i = 0; i < n; i++)
2442 50515371 : if (GET_CODE (XVECEXP (pat, 0, i)) != SET
2443 29850425 : || !REG_P (SET_DEST (XVECEXP (pat, 0, i))))
2444 : return false;
2445 3108898 : for ( ; i < len; i++)
2446 961205 : switch (GET_CODE (XVECEXP (pat, 0, i)))
2447 : {
2448 367533 : case CLOBBER:
2449 367533 : if (XEXP (XVECEXP (pat, 0, i), 0) == const0_rtx)
2450 : return false;
2451 367532 : break;
2452 : default:
2453 : return false;
2454 : }
2455 : return true;
2456 : }
2457 :
2458 : /* Return whether INSN, a PARALLEL of N register SETs (and maybe some
2459 : CLOBBERs), can be split into individual SETs in that order, without
2460 : changing semantics. */
2461 : static bool
2462 371145 : can_split_parallel_of_n_reg_sets (rtx_insn *insn, int n)
2463 : {
2464 371145 : if (!insn_nothrow_p (insn))
2465 : return false;
2466 :
2467 369636 : rtx pat = PATTERN (insn);
2468 :
2469 369636 : int i, j;
2470 995208 : for (i = 0; i < n; i++)
2471 : {
2472 682422 : if (side_effects_p (SET_SRC (XVECEXP (pat, 0, i))))
2473 : return false;
2474 :
2475 679314 : rtx reg = SET_DEST (XVECEXP (pat, 0, i));
2476 :
2477 992100 : for (j = i + 1; j < n; j++)
2478 366528 : if (reg_referenced_p (reg, XVECEXP (pat, 0, j)))
2479 : return false;
2480 : }
2481 :
2482 : return true;
2483 : }
2484 :
2485 : /* Return whether X is just a single_set, with the source
2486 : a general_operand. */
2487 : static bool
2488 65881012 : is_just_move (rtx_insn *x)
2489 : {
2490 65881012 : rtx set = single_set (x);
2491 65881012 : if (!set)
2492 : return false;
2493 :
2494 65463483 : return general_operand (SET_SRC (set), VOIDmode);
2495 : }
2496 :
2497 : /* Callback function to count autoincs. */
2498 :
2499 : static int
2500 1030774 : count_auto_inc (rtx, rtx, rtx, rtx, rtx, void *arg)
2501 : {
2502 1030774 : (*((int *) arg))++;
2503 :
2504 1030774 : return 0;
2505 : }
2506 :
2507 : /* Try to combine the insns I0, I1 and I2 into I3.
2508 : Here I0, I1 and I2 appear earlier than I3.
2509 : I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2510 : I3.
2511 :
2512 : If we are combining more than two insns and the resulting insn is not
2513 : recognized, try splitting it into two insns. If that happens, I2 and I3
2514 : are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2515 : Otherwise, I0, I1 and I2 are pseudo-deleted.
2516 :
2517 : Return 0 if the combination does not work. Then nothing is changed.
2518 : If we did the combination, return the insn at which combine should
2519 : resume scanning.
2520 :
2521 : Set NEW_DIRECT_JUMP_P to true if try_combine creates a
2522 : new direct jump instruction.
2523 :
2524 : LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2525 : been I3 passed to an earlier try_combine within the same basic
2526 : block. */
2527 :
2528 : static rtx_insn *
2529 94453885 : try_combine (rtx_insn *i3, rtx_insn *i2, rtx_insn *i1, rtx_insn *i0,
2530 : bool *new_direct_jump_p, rtx_insn *last_combined_insn)
2531 : {
2532 : /* New patterns for I3 and I2, respectively. */
2533 94453885 : rtx newpat, newi2pat = 0;
2534 94453885 : rtvec newpat_vec_with_clobbers = 0;
2535 94453885 : bool substed_i2 = false, substed_i1 = false, substed_i0 = false;
2536 : /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2537 : dead. */
2538 94453885 : bool added_sets_0, added_sets_1, added_sets_2;
2539 : /* Total number of SETs to put into I3. */
2540 94453885 : int total_sets;
2541 : /* Nonzero if I2's or I1's body now appears in I3. */
2542 94453885 : int i2_is_used = 0, i1_is_used = 0;
2543 : /* INSN_CODEs for new I3, new I2, and user of condition code. */
2544 94453885 : int insn_code_number, i2_code_number = 0, other_code_number = 0;
2545 : /* Contains I3 if the destination of I3 is used in its source, which means
2546 : that the old life of I3 is being killed. If that usage is placed into
2547 : I2 and not in I3, a REG_DEAD note must be made. */
2548 94453885 : rtx i3dest_killed = 0;
2549 : /* SET_DEST and SET_SRC of I2, I1 and I0. */
2550 94453885 : rtx i2dest = 0, i2src = 0, i1dest = 0, i1src = 0, i0dest = 0, i0src = 0;
2551 : /* Copy of SET_SRC of I1 and I0, if needed. */
2552 94453885 : rtx i1src_copy = 0, i0src_copy = 0, i0src_copy2 = 0;
2553 : /* Set if I2DEST was reused as a scratch register. */
2554 94453885 : bool i2scratch = false;
2555 : /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2556 94453885 : rtx i0pat = 0, i1pat = 0, i2pat = 0;
2557 : /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2558 94453885 : bool i2dest_in_i2src = false, i1dest_in_i1src = false;
2559 94453885 : bool i2dest_in_i1src = false, i0dest_in_i0src = false;
2560 94453885 : bool i1dest_in_i0src = false, i2dest_in_i0src = false;;
2561 94453885 : bool i2dest_killed = false, i1dest_killed = false, i0dest_killed = false;
2562 94453885 : bool i1_feeds_i2_n = false, i0_feeds_i2_n = false, i0_feeds_i1_n = false;
2563 : /* Notes that must be added to REG_NOTES in I3 and I2. */
2564 94453885 : rtx new_i3_notes, new_i2_notes;
2565 : /* Notes that we substituted I3 into I2 instead of the normal case. */
2566 94453885 : bool i3_subst_into_i2 = false;
2567 : /* Notes that I1, I2 or I3 is a MULT operation. */
2568 94453885 : bool have_mult = false;
2569 94453885 : bool swap_i2i3 = false;
2570 94453885 : bool split_i2i3 = false;
2571 94453885 : bool changed_i3_dest = false;
2572 94453885 : bool i2_was_move = false, i3_was_move = false;
2573 94453885 : int n_auto_inc = 0;
2574 :
2575 94453885 : int maxreg;
2576 94453885 : rtx_insn *temp_insn;
2577 94453885 : rtx temp_expr;
2578 94453885 : struct insn_link *link;
2579 94453885 : rtx other_pat = 0;
2580 94453885 : rtx new_other_notes;
2581 94453885 : int i;
2582 94453885 : scalar_int_mode dest_mode, temp_mode;
2583 94453885 : bool has_non_call_exception = false;
2584 :
2585 : /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
2586 : never be). */
2587 94453885 : if (i1 == i2 || i0 == i2 || (i0 && i0 == i1))
2588 : return 0;
2589 :
2590 : /* Only try four-insn combinations when there's high likelihood of
2591 : success. Look for simple insns, such as loads of constants or
2592 : binary operations involving a constant. */
2593 21668631 : if (i0)
2594 : {
2595 21668631 : int i;
2596 21668631 : int ngood = 0;
2597 21668631 : int nshift = 0;
2598 21668631 : rtx set0, set3;
2599 :
2600 21668631 : if (!flag_expensive_optimizations)
2601 : return 0;
2602 :
2603 86303396 : for (i = 0; i < 4; i++)
2604 : {
2605 70550487 : rtx_insn *insn = i == 0 ? i0 : i == 1 ? i1 : i == 2 ? i2 : i3;
2606 70550487 : rtx set = single_set (insn);
2607 70550487 : rtx src;
2608 70550487 : if (!set)
2609 2314621 : continue;
2610 68235866 : src = SET_SRC (set);
2611 68235866 : if (CONSTANT_P (src))
2612 : {
2613 4551043 : ngood += 2;
2614 4551043 : break;
2615 : }
2616 63684823 : else if (BINARY_P (src) && CONSTANT_P (XEXP (src, 1)))
2617 7898538 : ngood++;
2618 55786285 : else if (GET_CODE (src) == IF_THEN_ELSE)
2619 2055849 : ngood++;
2620 53730436 : else if (GET_CODE (src) == ASHIFT || GET_CODE (src) == ASHIFTRT
2621 53645036 : || GET_CODE (src) == LSHIFTRT)
2622 115442 : nshift++;
2623 : }
2624 :
2625 : /* If I0 loads a memory and I3 sets the same memory, then I1 and I2
2626 : are likely manipulating its value. Ideally we'll be able to combine
2627 : all four insns into a bitfield insertion of some kind.
2628 :
2629 : Note the source in I0 might be inside a sign/zero extension and the
2630 : memory modes in I0 and I3 might be different. So extract the address
2631 : from the destination of I3 and search for it in the source of I0.
2632 :
2633 : In the event that there's a match but the source/dest do not actually
2634 : refer to the same memory, the worst that happens is we try some
2635 : combinations that we wouldn't have otherwise. */
2636 20303952 : if ((set0 = single_set (i0))
2637 : /* Ensure the source of SET0 is a MEM, possibly buried inside
2638 : an extension. */
2639 20172720 : && (GET_CODE (SET_SRC (set0)) == MEM
2640 16909510 : || ((GET_CODE (SET_SRC (set0)) == ZERO_EXTEND
2641 16909510 : || GET_CODE (SET_SRC (set0)) == SIGN_EXTEND)
2642 565900 : && GET_CODE (XEXP (SET_SRC (set0), 0)) == MEM))
2643 3375105 : && (set3 = single_set (i3))
2644 : /* Ensure the destination of SET3 is a MEM. */
2645 2926994 : && GET_CODE (SET_DEST (set3)) == MEM
2646 : /* Would it be better to extract the base address for the MEM
2647 : in SET3 and look for that? I don't have cases where it matters
2648 : but I could envision such cases. */
2649 20648966 : && rtx_referenced_p (XEXP (SET_DEST (set3), 0), SET_SRC (set0)))
2650 22804 : ngood += 2;
2651 :
2652 20303952 : if (ngood < 2 && nshift < 2)
2653 : return 0;
2654 : }
2655 :
2656 : /* Exit early if one of the insns involved can't be used for
2657 : combinations. */
2658 79446944 : if (CALL_P (i2)
2659 74429879 : || (i1 && CALL_P (i1))
2660 71032490 : || (i0 && CALL_P (i0))
2661 70572047 : || cant_combine_insn_p (i3)
2662 67320738 : || cant_combine_insn_p (i2)
2663 51821850 : || (i1 && cant_combine_insn_p (i1))
2664 46944784 : || (i0 && cant_combine_insn_p (i0))
2665 126162103 : || likely_spilled_retval_p (i3))
2666 32731795 : return 0;
2667 :
2668 46715149 : combine_attempts++;
2669 46715149 : undobuf.other_insn = 0;
2670 :
2671 : /* Reset the hard register usage information. */
2672 46715149 : CLEAR_HARD_REG_SET (newpat_used_regs);
2673 :
2674 46715149 : if (dump_file && (dump_flags & TDF_DETAILS))
2675 : {
2676 174 : if (i0)
2677 20 : fprintf (dump_file, "\nTrying %d, %d, %d -> %d:\n",
2678 20 : INSN_UID (i0), INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
2679 154 : else if (i1)
2680 26 : fprintf (dump_file, "\nTrying %d, %d -> %d:\n",
2681 26 : INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
2682 : else
2683 128 : fprintf (dump_file, "\nTrying %d -> %d:\n",
2684 128 : INSN_UID (i2), INSN_UID (i3));
2685 :
2686 174 : if (i0)
2687 20 : dump_insn_slim (dump_file, i0);
2688 174 : if (i1)
2689 46 : dump_insn_slim (dump_file, i1);
2690 174 : dump_insn_slim (dump_file, i2);
2691 174 : dump_insn_slim (dump_file, i3);
2692 : }
2693 :
2694 : /* If multiple insns feed into one of I2 or I3, they can be in any
2695 : order. To simplify the code below, reorder them in sequence. */
2696 46715149 : if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i2))
2697 : std::swap (i0, i2);
2698 46715149 : if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i1))
2699 : std::swap (i0, i1);
2700 46715149 : if (i1 && DF_INSN_LUID (i1) > DF_INSN_LUID (i2))
2701 : std::swap (i1, i2);
2702 :
2703 46715149 : added_links_insn = 0;
2704 46715149 : added_notes_insn = 0;
2705 :
2706 : /* First check for one important special case that the code below will
2707 : not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2708 : and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2709 : we may be able to replace that destination with the destination of I3.
2710 : This occurs in the common code where we compute both a quotient and
2711 : remainder into a structure, in which case we want to do the computation
2712 : directly into the structure to avoid register-register copies.
2713 :
2714 : Note that this case handles both multiple sets in I2 and also cases
2715 : where I2 has a number of CLOBBERs inside the PARALLEL.
2716 :
2717 : We make very conservative checks below and only try to handle the
2718 : most common cases of this. For example, we only handle the case
2719 : where I2 and I3 are adjacent to avoid making difficult register
2720 : usage tests. */
2721 :
2722 29203411 : if (i1 == 0 && NONJUMP_INSN_P (i3) && GET_CODE (PATTERN (i3)) == SET
2723 15254765 : && REG_P (SET_SRC (PATTERN (i3)))
2724 5147808 : && REGNO (SET_SRC (PATTERN (i3))) >= FIRST_PSEUDO_REGISTER
2725 4938136 : && find_reg_note (i3, REG_DEAD, SET_SRC (PATTERN (i3)))
2726 4062767 : && GET_CODE (PATTERN (i2)) == PARALLEL
2727 1066056 : && ! side_effects_p (SET_DEST (PATTERN (i3)))
2728 : /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2729 : below would need to check what is inside (and reg_overlap_mentioned_p
2730 : doesn't support those codes anyway). Don't allow those destinations;
2731 : the resulting insn isn't likely to be recognized anyway. */
2732 577343 : && GET_CODE (SET_DEST (PATTERN (i3))) != ZERO_EXTRACT
2733 577315 : && GET_CODE (SET_DEST (PATTERN (i3))) != STRICT_LOW_PART
2734 576127 : && ! reg_overlap_mentioned_p (SET_SRC (PATTERN (i3)),
2735 576127 : SET_DEST (PATTERN (i3)))
2736 47291161 : && next_active_insn (i2) == i3)
2737 : {
2738 356764 : rtx p2 = PATTERN (i2);
2739 :
2740 : /* Make sure that the destination of I3,
2741 : which we are going to substitute into one output of I2,
2742 : is not used within another output of I2. We must avoid making this:
2743 : (parallel [(set (mem (reg 69)) ...)
2744 : (set (reg 69) ...)])
2745 : which is not well-defined as to order of actions.
2746 : (Besides, reload can't handle output reloads for this.)
2747 :
2748 : The problem can also happen if the dest of I3 is a memory ref,
2749 : if another dest in I2 is an indirect memory ref.
2750 :
2751 : Neither can this PARALLEL be an asm. We do not allow combining
2752 : that usually (see can_combine_p), so do not here either. */
2753 356764 : bool ok = true;
2754 1082400 : for (i = 0; ok && i < XVECLEN (p2, 0); i++)
2755 : {
2756 725636 : if ((GET_CODE (XVECEXP (p2, 0, i)) == SET
2757 356292 : || GET_CODE (XVECEXP (p2, 0, i)) == CLOBBER)
2758 1449877 : && reg_overlap_mentioned_p (SET_DEST (PATTERN (i3)),
2759 724241 : SET_DEST (XVECEXP (p2, 0, i))))
2760 : ok = false;
2761 724867 : else if (GET_CODE (XVECEXP (p2, 0, i)) == SET
2762 368577 : && GET_CODE (SET_SRC (XVECEXP (p2, 0, i))) == ASM_OPERANDS)
2763 1894 : ok = false;
2764 : }
2765 :
2766 356764 : if (ok)
2767 421015 : for (i = 0; i < XVECLEN (p2, 0); i++)
2768 389453 : if (GET_CODE (XVECEXP (p2, 0, i)) == SET
2769 389453 : && SET_DEST (XVECEXP (p2, 0, i)) == SET_SRC (PATTERN (i3)))
2770 : {
2771 323308 : combine_merges++;
2772 :
2773 323308 : subst_insn = i3;
2774 323308 : subst_low_luid = DF_INSN_LUID (i2);
2775 :
2776 323308 : added_sets_2 = added_sets_1 = added_sets_0 = false;
2777 323308 : i2src = SET_SRC (XVECEXP (p2, 0, i));
2778 323308 : i2dest = SET_DEST (XVECEXP (p2, 0, i));
2779 323308 : i2dest_killed = dead_or_set_p (i2, i2dest);
2780 :
2781 : /* Replace the dest in I2 with our dest and make the resulting
2782 : insn the new pattern for I3. Then skip to where we validate
2783 : the pattern. Everything was set up above. */
2784 323308 : SUBST (SET_DEST (XVECEXP (p2, 0, i)), SET_DEST (PATTERN (i3)));
2785 323308 : newpat = p2;
2786 323308 : i3_subst_into_i2 = true;
2787 323308 : goto validate_replacement;
2788 : }
2789 : }
2790 :
2791 : /* If I2 is setting a pseudo to a constant and I3 is setting some
2792 : sub-part of it to another constant, merge them by making a new
2793 : constant. */
2794 46391841 : if (i1 == 0
2795 28880103 : && (temp_expr = single_set (i2)) != 0
2796 28606822 : && is_a <scalar_int_mode> (GET_MODE (SET_DEST (temp_expr)), &temp_mode)
2797 18840618 : && CONST_SCALAR_INT_P (SET_SRC (temp_expr))
2798 2785788 : && GET_CODE (PATTERN (i3)) == SET
2799 1375464 : && CONST_SCALAR_INT_P (SET_SRC (PATTERN (i3)))
2800 46410247 : && reg_subword_p (SET_DEST (PATTERN (i3)), SET_DEST (temp_expr)))
2801 : {
2802 17729 : rtx dest = SET_DEST (PATTERN (i3));
2803 17729 : rtx temp_dest = SET_DEST (temp_expr);
2804 17729 : int offset = -1;
2805 17729 : int width = 0;
2806 :
2807 17729 : if (GET_CODE (dest) == ZERO_EXTRACT)
2808 : {
2809 1 : if (CONST_INT_P (XEXP (dest, 1))
2810 1 : && CONST_INT_P (XEXP (dest, 2))
2811 2 : && is_a <scalar_int_mode> (GET_MODE (XEXP (dest, 0)),
2812 : &dest_mode))
2813 : {
2814 1 : width = INTVAL (XEXP (dest, 1));
2815 1 : offset = INTVAL (XEXP (dest, 2));
2816 1 : dest = XEXP (dest, 0);
2817 1 : if (BITS_BIG_ENDIAN)
2818 : offset = GET_MODE_PRECISION (dest_mode) - width - offset;
2819 : }
2820 : }
2821 : else
2822 : {
2823 17728 : if (GET_CODE (dest) == STRICT_LOW_PART)
2824 1881 : dest = XEXP (dest, 0);
2825 17728 : if (is_a <scalar_int_mode> (GET_MODE (dest), &dest_mode))
2826 : {
2827 17728 : width = GET_MODE_PRECISION (dest_mode);
2828 17728 : offset = 0;
2829 : }
2830 : }
2831 :
2832 17729 : if (offset >= 0)
2833 : {
2834 : /* If this is the low part, we're done. */
2835 17729 : if (subreg_lowpart_p (dest))
2836 : ;
2837 : /* Handle the case where inner is twice the size of outer. */
2838 4723 : else if (GET_MODE_PRECISION (temp_mode)
2839 4723 : == 2 * GET_MODE_PRECISION (dest_mode))
2840 4720 : offset += GET_MODE_PRECISION (dest_mode);
2841 : /* Otherwise give up for now. */
2842 : else
2843 : offset = -1;
2844 : }
2845 :
2846 17726 : if (offset >= 0)
2847 : {
2848 17726 : rtx inner = SET_SRC (PATTERN (i3));
2849 17726 : rtx outer = SET_SRC (temp_expr);
2850 :
2851 35452 : wide_int o = wi::insert (rtx_mode_t (outer, temp_mode),
2852 17726 : rtx_mode_t (inner, dest_mode),
2853 35452 : offset, width);
2854 :
2855 17726 : combine_merges++;
2856 17726 : subst_insn = i3;
2857 17726 : subst_low_luid = DF_INSN_LUID (i2);
2858 17726 : added_sets_2 = added_sets_1 = added_sets_0 = false;
2859 17726 : i2dest = temp_dest;
2860 17726 : i2dest_killed = dead_or_set_p (i2, i2dest);
2861 :
2862 : /* Replace the source in I2 with the new constant and make the
2863 : resulting insn the new pattern for I3. Then skip to where we
2864 : validate the pattern. Everything was set up above. */
2865 17726 : SUBST (SET_SRC (temp_expr),
2866 : immed_wide_int_const (o, temp_mode));
2867 :
2868 17726 : newpat = PATTERN (i2);
2869 :
2870 : /* The dest of I3 has been replaced with the dest of I2. */
2871 17726 : changed_i3_dest = true;
2872 17726 : goto validate_replacement;
2873 17726 : }
2874 : }
2875 :
2876 : /* If we have no I1 and I2 looks like:
2877 : (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
2878 : (set Y OP)])
2879 : make up a dummy I1 that is
2880 : (set Y OP)
2881 : and change I2 to be
2882 : (set (reg:CC X) (compare:CC Y (const_int 0)))
2883 :
2884 : (We can ignore any trailing CLOBBERs.)
2885 :
2886 : This undoes a previous combination and allows us to match a branch-and-
2887 : decrement insn. */
2888 :
2889 46374115 : if (i1 == 0
2890 28862377 : && is_parallel_of_n_reg_sets (PATTERN (i2), 2)
2891 225720 : && (GET_MODE_CLASS (GET_MODE (SET_DEST (XVECEXP (PATTERN (i2), 0, 0))))
2892 : == MODE_CC)
2893 143561 : && GET_CODE (SET_SRC (XVECEXP (PATTERN (i2), 0, 0))) == COMPARE
2894 117355 : && XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 1) == const0_rtx
2895 77508 : && rtx_equal_p (XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 0),
2896 77508 : SET_SRC (XVECEXP (PATTERN (i2), 0, 1)))
2897 70977 : && !reg_used_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 0)), i2, i3)
2898 46445092 : && !reg_used_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 1)), i2, i3))
2899 : {
2900 : /* We make I1 with the same INSN_UID as I2. This gives it
2901 : the same DF_INSN_LUID for value tracking. Our fake I1 will
2902 : never appear in the insn stream so giving it the same INSN_UID
2903 : as I2 will not cause a problem. */
2904 :
2905 141518 : i1 = gen_rtx_INSN (VOIDmode, NULL, i2, BLOCK_FOR_INSN (i2),
2906 70759 : XVECEXP (PATTERN (i2), 0, 1), INSN_LOCATION (i2),
2907 : -1, NULL_RTX);
2908 70759 : INSN_UID (i1) = INSN_UID (i2);
2909 :
2910 70759 : SUBST (PATTERN (i2), XVECEXP (PATTERN (i2), 0, 0));
2911 70759 : SUBST (XEXP (SET_SRC (PATTERN (i2)), 0),
2912 : SET_DEST (PATTERN (i1)));
2913 70759 : unsigned int regno = REGNO (SET_DEST (PATTERN (i1)));
2914 70759 : SUBST_LINK (LOG_LINKS (i2),
2915 : alloc_insn_link (i1, regno, LOG_LINKS (i2)));
2916 : }
2917 :
2918 : /* If I2 is a PARALLEL of two SETs of REGs (and perhaps some CLOBBERs),
2919 : make those two SETs separate I1 and I2 insns, and make an I0 that is
2920 : the original I1. */
2921 46374115 : if (i0 == 0
2922 43827041 : && is_parallel_of_n_reg_sets (PATTERN (i2), 2)
2923 371145 : && can_split_parallel_of_n_reg_sets (i2, 2)
2924 312786 : && !reg_used_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 0)), i2, i3)
2925 274436 : && !reg_used_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 1)), i2, i3)
2926 256517 : && !reg_set_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 0)), i2, i3)
2927 46630623 : && !reg_set_between_p (SET_DEST (XVECEXP (PATTERN (i2), 0, 1)), i2, i3))
2928 : {
2929 : /* If there is no I1, there is no I0 either. */
2930 256508 : i0 = i1;
2931 :
2932 : /* We make I1 with the same INSN_UID as I2. This gives it
2933 : the same DF_INSN_LUID for value tracking. Our fake I1 will
2934 : never appear in the insn stream so giving it the same INSN_UID
2935 : as I2 will not cause a problem. */
2936 :
2937 513016 : i1 = gen_rtx_INSN (VOIDmode, NULL, i2, BLOCK_FOR_INSN (i2),
2938 256508 : XVECEXP (PATTERN (i2), 0, 0), INSN_LOCATION (i2),
2939 : -1, NULL_RTX);
2940 256508 : INSN_UID (i1) = INSN_UID (i2);
2941 :
2942 256508 : SUBST (PATTERN (i2), XVECEXP (PATTERN (i2), 0, 1));
2943 : }
2944 :
2945 : /* Verify that I2 and maybe I1 and I0 can be combined into I3. */
2946 46374115 : if (!can_combine_p (i2, i3, i0, i1, NULL, NULL, &i2dest, &i2src))
2947 : {
2948 11843713 : if (dump_file && (dump_flags & TDF_DETAILS))
2949 8 : fprintf (dump_file, "Can't combine i2 into i3\n");
2950 11843713 : undo_all ();
2951 11843713 : return 0;
2952 : }
2953 34530402 : if (i1 && !can_combine_p (i1, i3, i0, NULL, i2, NULL, &i1dest, &i1src))
2954 : {
2955 1362239 : if (dump_file && (dump_flags & TDF_DETAILS))
2956 0 : fprintf (dump_file, "Can't combine i1 into i3\n");
2957 1362239 : undo_all ();
2958 1362239 : return 0;
2959 : }
2960 33168163 : if (i0 && !can_combine_p (i0, i3, NULL, NULL, i1, i2, &i0dest, &i0src))
2961 : {
2962 227485 : if (dump_file && (dump_flags & TDF_DETAILS))
2963 0 : fprintf (dump_file, "Can't combine i0 into i3\n");
2964 227485 : undo_all ();
2965 227485 : return 0;
2966 : }
2967 :
2968 : /* With non-call exceptions we can end up trying to combine multiple
2969 : insns with possible EH side effects. Make sure we can combine
2970 : that to a single insn which means there must be at most one insn
2971 : in the combination with an EH side effect. */
2972 32940678 : if (cfun->can_throw_non_call_exceptions)
2973 : {
2974 6058601 : if (find_reg_note (i3, REG_EH_REGION, NULL_RTX)
2975 6034950 : || find_reg_note (i2, REG_EH_REGION, NULL_RTX)
2976 6034868 : || (i1 && find_reg_note (i1, REG_EH_REGION, NULL_RTX))
2977 12093467 : || (i0 && find_reg_note (i0, REG_EH_REGION, NULL_RTX)))
2978 : {
2979 23735 : has_non_call_exception = true;
2980 23735 : if (insn_could_throw_p (i3)
2981 23735 : + insn_could_throw_p (i2)
2982 23735 : + (i1 ? insn_could_throw_p (i1) : 0)
2983 23735 : + (i0 ? insn_could_throw_p (i0) : 0) > 1)
2984 : {
2985 172 : if (dump_file && (dump_flags & TDF_DETAILS))
2986 0 : fprintf (dump_file, "Can't combine multiple insns with EH "
2987 : "side-effects\n");
2988 172 : undo_all ();
2989 172 : return 0;
2990 : }
2991 : }
2992 : }
2993 :
2994 : /* Record whether i2 and i3 are trivial moves. */
2995 32940506 : i2_was_move = is_just_move (i2);
2996 32940506 : i3_was_move = is_just_move (i3);
2997 :
2998 : /* Record whether I2DEST is used in I2SRC and similarly for the other
2999 : cases. Knowing this will help in register status updating below. */
3000 32940506 : i2dest_in_i2src = reg_overlap_mentioned_p (i2dest, i2src);
3001 32940506 : i1dest_in_i1src = i1 && reg_overlap_mentioned_p (i1dest, i1src);
3002 10496100 : i2dest_in_i1src = i1 && reg_overlap_mentioned_p (i2dest, i1src);
3003 32940506 : i0dest_in_i0src = i0 && reg_overlap_mentioned_p (i0dest, i0src);
3004 1837349 : i1dest_in_i0src = i0 && reg_overlap_mentioned_p (i1dest, i0src);
3005 1837349 : i2dest_in_i0src = i0 && reg_overlap_mentioned_p (i2dest, i0src);
3006 32940506 : i2dest_killed = dead_or_set_p (i2, i2dest);
3007 32940506 : i1dest_killed = i1 && dead_or_set_p (i1, i1dest);
3008 32940506 : i0dest_killed = i0 && dead_or_set_p (i0, i0dest);
3009 :
3010 : /* For the earlier insns, determine which of the subsequent ones they
3011 : feed. */
3012 32940506 : i1_feeds_i2_n = i1 && insn_a_feeds_b (i1, i2);
3013 32940506 : i0_feeds_i1_n = i0 && insn_a_feeds_b (i0, i1);
3014 3200810 : i0_feeds_i2_n = (i0 && (!i0_feeds_i1_n ? insn_a_feeds_b (i0, i2)
3015 1363461 : : (!reg_overlap_mentioned_p (i1dest, i0dest)
3016 1340271 : && reg_overlap_mentioned_p (i0dest, i2src))));
3017 :
3018 : /* Ensure that I3's pattern can be the destination of combines. */
3019 32940506 : if (! combinable_i3pat (i3, &PATTERN (i3), i2dest, i1dest, i0dest,
3020 32940506 : i1 && i2dest_in_i1src && !i1_feeds_i2_n,
3021 1837349 : i0 && ((i2dest_in_i0src && !i0_feeds_i2_n)
3022 1810523 : || (i1dest_in_i0src && !i0_feeds_i1_n)),
3023 : &i3dest_killed))
3024 : {
3025 171294 : undo_all ();
3026 171294 : return 0;
3027 : }
3028 :
3029 : /* See if any of the insns is a MULT operation. Unless one is, we will
3030 : reject a combination that is, since it must be slower. Be conservative
3031 : here. */
3032 32769212 : if (GET_CODE (i2src) == MULT
3033 31973012 : || (i1 != 0 && GET_CODE (i1src) == MULT)
3034 31634790 : || (i0 != 0 && GET_CODE (i0src) == MULT)
3035 64358916 : || (GET_CODE (PATTERN (i3)) == SET
3036 24850307 : && GET_CODE (SET_SRC (PATTERN (i3))) == MULT))
3037 : have_mult = true;
3038 :
3039 : /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
3040 : We used to do this EXCEPT in one case: I3 has a post-inc in an
3041 : output operand. However, that exception can give rise to insns like
3042 : mov r3,(r3)+
3043 : which is a famous insn on the PDP-11 where the value of r3 used as the
3044 : source was model-dependent. Avoid this sort of thing. */
3045 :
3046 : #if 0
3047 : if (!(GET_CODE (PATTERN (i3)) == SET
3048 : && REG_P (SET_SRC (PATTERN (i3)))
3049 : && MEM_P (SET_DEST (PATTERN (i3)))
3050 : && (GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_INC
3051 : || GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_DEC)))
3052 : /* It's not the exception. */
3053 : #endif
3054 32769212 : if (AUTO_INC_DEC)
3055 : {
3056 : rtx link;
3057 : for (link = REG_NOTES (i3); link; link = XEXP (link, 1))
3058 : if (REG_NOTE_KIND (link) == REG_INC
3059 : && (reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i2))
3060 : || (i1 != 0
3061 : && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i1)))))
3062 : {
3063 : undo_all ();
3064 : return 0;
3065 : }
3066 : }
3067 :
3068 : /* See if the SETs in I1 or I2 need to be kept around in the merged
3069 : instruction: whenever the value set there is still needed past I3.
3070 : For the SET in I2, this is easy: we see if I2DEST dies or is set in I3.
3071 :
3072 : For the SET in I1, we have two cases: if I1 and I2 independently feed
3073 : into I3, the set in I1 needs to be kept around unless I1DEST dies
3074 : or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
3075 : in I1 needs to be kept around unless I1DEST dies or is set in either
3076 : I2 or I3. The same considerations apply to I0. */
3077 :
3078 32769212 : added_sets_2 = !dead_or_set_p (i3, i2dest);
3079 :
3080 32769212 : if (i1)
3081 10442792 : added_sets_1 = !(dead_or_set_p (i3, i1dest)
3082 8006557 : || (i1_feeds_i2_n && dead_or_set_p (i2, i1dest)));
3083 : else
3084 : added_sets_1 = false;
3085 :
3086 32769212 : if (i0)
3087 2676346 : added_sets_0 = !(dead_or_set_p (i3, i0dest)
3088 1640618 : || (i0_feeds_i1_n && dead_or_set_p (i1, i0dest))
3089 335483 : || ((i0_feeds_i2_n || (i0_feeds_i1_n && i1_feeds_i2_n))
3090 805828 : && dead_or_set_p (i2, i0dest)));
3091 : else
3092 : added_sets_0 = false;
3093 :
3094 : /* We are about to copy insns for the case where they need to be kept
3095 : around. Check that they can be copied in the merged instruction. */
3096 :
3097 32769212 : if (targetm.cannot_copy_insn_p
3098 32769212 : && ((added_sets_2 && targetm.cannot_copy_insn_p (i2))
3099 0 : || (i1 && added_sets_1 && targetm.cannot_copy_insn_p (i1))
3100 0 : || (i0 && added_sets_0 && targetm.cannot_copy_insn_p (i0))))
3101 : {
3102 0 : undo_all ();
3103 0 : return 0;
3104 : }
3105 :
3106 : /* We cannot safely duplicate volatile references in any case. */
3107 :
3108 7313970 : if ((added_sets_2 && volatile_refs_p (PATTERN (i2)))
3109 32732413 : || (added_sets_1 && volatile_refs_p (PATTERN (i1)))
3110 65473303 : || (added_sets_0 && volatile_refs_p (PATTERN (i0))))
3111 : {
3112 67458 : undo_all ();
3113 67458 : return 0;
3114 : }
3115 :
3116 : /* Count how many auto_inc expressions there were in the original insns;
3117 : we need to have the same number in the resulting patterns. */
3118 :
3119 32701754 : if (i0)
3120 1805186 : for_each_inc_dec (PATTERN (i0), count_auto_inc, &n_auto_inc);
3121 32701754 : if (i1)
3122 10410655 : for_each_inc_dec (PATTERN (i1), count_auto_inc, &n_auto_inc);
3123 32701754 : for_each_inc_dec (PATTERN (i2), count_auto_inc, &n_auto_inc);
3124 32701754 : for_each_inc_dec (PATTERN (i3), count_auto_inc, &n_auto_inc);
3125 :
3126 : /* If the set in I2 needs to be kept around, we must make a copy of
3127 : PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3128 : PATTERN (I2), we are only substituting for the original I1DEST, not into
3129 : an already-substituted copy. This also prevents making self-referential
3130 : rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3131 : I2DEST. */
3132 :
3133 32701754 : if (added_sets_2)
3134 : {
3135 7274287 : if (GET_CODE (PATTERN (i2)) == PARALLEL)
3136 2269301 : i2pat = gen_rtx_SET (i2dest, copy_rtx (i2src));
3137 : else
3138 5004986 : i2pat = copy_rtx (PATTERN (i2));
3139 : }
3140 :
3141 32701754 : if (added_sets_1)
3142 : {
3143 3988138 : if (GET_CODE (PATTERN (i1)) == PARALLEL)
3144 1291763 : i1pat = gen_rtx_SET (i1dest, copy_rtx (i1src));
3145 : else
3146 2696375 : i1pat = copy_rtx (PATTERN (i1));
3147 : }
3148 :
3149 32701754 : if (added_sets_0)
3150 : {
3151 500215 : if (GET_CODE (PATTERN (i0)) == PARALLEL)
3152 182822 : i0pat = gen_rtx_SET (i0dest, copy_rtx (i0src));
3153 : else
3154 317393 : i0pat = copy_rtx (PATTERN (i0));
3155 : }
3156 :
3157 32701754 : combine_merges++;
3158 :
3159 : /* Substitute in the latest insn for the regs set by the earlier ones. */
3160 :
3161 32701754 : maxreg = max_reg_num ();
3162 :
3163 32701754 : subst_insn = i3;
3164 :
3165 : /* Many machines have insns that can both perform an
3166 : arithmetic operation and set the condition code. These operations will
3167 : be represented as a PARALLEL with the first element of the vector
3168 : being a COMPARE of an arithmetic operation with the constant zero.
3169 : The second element of the vector will set some pseudo to the result
3170 : of the same arithmetic operation. If we simplify the COMPARE, we won't
3171 : match such a pattern and so will generate an extra insn. Here we test
3172 : for this case, where both the comparison and the operation result are
3173 : needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3174 : I2SRC. Later we will make the PARALLEL that contains I2. */
3175 :
3176 22291099 : if (i1 == 0 && added_sets_2 && GET_CODE (PATTERN (i3)) == SET
3177 4289810 : && GET_CODE (SET_SRC (PATTERN (i3))) == COMPARE
3178 1836963 : && CONST_INT_P (XEXP (SET_SRC (PATTERN (i3)), 1))
3179 33593992 : && rtx_equal_p (XEXP (SET_SRC (PATTERN (i3)), 0), i2dest))
3180 : {
3181 822797 : rtx newpat_dest;
3182 822797 : rtx *cc_use_loc = NULL;
3183 822797 : rtx_insn *cc_use_insn = NULL;
3184 822797 : rtx op0 = i2src, op1 = XEXP (SET_SRC (PATTERN (i3)), 1);
3185 822797 : machine_mode compare_mode, orig_compare_mode;
3186 822797 : enum rtx_code compare_code = UNKNOWN, orig_compare_code = UNKNOWN;
3187 822797 : scalar_int_mode mode;
3188 :
3189 822797 : newpat = PATTERN (i3);
3190 822797 : newpat_dest = SET_DEST (newpat);
3191 822797 : compare_mode = orig_compare_mode = GET_MODE (newpat_dest);
3192 :
3193 822797 : if (undobuf.other_insn == 0
3194 822797 : && (cc_use_loc = find_single_use (SET_DEST (newpat), i3,
3195 : &cc_use_insn)))
3196 : {
3197 816527 : compare_code = orig_compare_code = GET_CODE (*cc_use_loc);
3198 816527 : if (is_a <scalar_int_mode> (GET_MODE (i2dest), &mode))
3199 816527 : compare_code = simplify_compare_const (compare_code, mode,
3200 : &op0, &op1);
3201 816527 : target_canonicalize_comparison (&compare_code, &op0, &op1, 1);
3202 : }
3203 :
3204 : /* Do the rest only if op1 is const0_rtx, which may be the
3205 : result of simplification. */
3206 822797 : if (op1 == const0_rtx)
3207 : {
3208 : /* If a single use of the CC is found, prepare to modify it
3209 : when SELECT_CC_MODE returns a new CC-class mode, or when
3210 : the above simplify_compare_const() returned a new comparison
3211 : operator. undobuf.other_insn is assigned the CC use insn
3212 : when modifying it. */
3213 508095 : if (cc_use_loc)
3214 : {
3215 : #ifdef SELECT_CC_MODE
3216 505467 : machine_mode new_mode
3217 505467 : = SELECT_CC_MODE (compare_code, op0, op1);
3218 505467 : if (new_mode != orig_compare_mode
3219 505467 : && can_change_dest_mode (SET_DEST (newpat),
3220 : added_sets_2, new_mode))
3221 : {
3222 402 : unsigned int regno = REGNO (newpat_dest);
3223 402 : compare_mode = new_mode;
3224 402 : if (regno < FIRST_PSEUDO_REGISTER)
3225 402 : newpat_dest = gen_rtx_REG (compare_mode, regno);
3226 : else
3227 : {
3228 0 : subst_mode (regno, compare_mode);
3229 0 : newpat_dest = regno_reg_rtx[regno];
3230 : }
3231 : }
3232 : #endif
3233 : /* Cases for modifying the CC-using comparison. */
3234 505467 : if (compare_code != orig_compare_code
3235 416 : && COMPARISON_P (*cc_use_loc))
3236 : {
3237 : /* Replace cc_use_loc with entire new RTX. */
3238 416 : SUBST (*cc_use_loc,
3239 : gen_rtx_fmt_ee (compare_code, GET_MODE (*cc_use_loc),
3240 : newpat_dest, const0_rtx));
3241 416 : undobuf.other_insn = cc_use_insn;
3242 : }
3243 505051 : else if (compare_mode != orig_compare_mode)
3244 : {
3245 1 : subrtx_ptr_iterator::array_type array;
3246 :
3247 : /* Just replace the CC reg with a new mode. */
3248 4 : FOR_EACH_SUBRTX_PTR (iter, array, cc_use_loc, NONCONST)
3249 : {
3250 3 : rtx *loc = *iter;
3251 3 : if (REG_P (*loc)
3252 3 : && REGNO (*loc) == REGNO (newpat_dest))
3253 : {
3254 1 : SUBST (*loc, newpat_dest);
3255 1 : iter.skip_subrtxes ();
3256 : }
3257 : }
3258 1 : undobuf.other_insn = cc_use_insn;
3259 1 : }
3260 : }
3261 :
3262 : /* Now we modify the current newpat:
3263 : First, SET_DEST(newpat) is updated if the CC mode has been
3264 : altered. For targets without SELECT_CC_MODE, this should be
3265 : optimized away. */
3266 508095 : if (compare_mode != orig_compare_mode)
3267 402 : SUBST (SET_DEST (newpat), newpat_dest);
3268 : /* This is always done to propagate i2src into newpat. */
3269 508095 : SUBST (SET_SRC (newpat),
3270 : gen_rtx_COMPARE (compare_mode, op0, op1));
3271 : /* Create new version of i2pat if needed; the below PARALLEL
3272 : creation needs this to work correctly. */
3273 508095 : if (! rtx_equal_p (i2src, op0))
3274 31 : i2pat = gen_rtx_SET (i2dest, op0);
3275 508095 : i2_is_used = 1;
3276 : }
3277 : }
3278 :
3279 822797 : if (i2_is_used == 0)
3280 : {
3281 : /* It is possible that the source of I2 or I1 may be performing
3282 : an unneeded operation, such as a ZERO_EXTEND of something
3283 : that is known to have the high part zero. Handle that case
3284 : by letting subst look at the inner insns.
3285 :
3286 : Another way to do this would be to have a function that tries
3287 : to simplify a single insn instead of merging two or more
3288 : insns. We don't do this because of the potential of infinite
3289 : loops and because of the potential extra memory required.
3290 : However, doing it the way we are is a bit of a kludge and
3291 : doesn't catch all cases.
3292 :
3293 : But only do this if -fexpensive-optimizations since it slows
3294 : things down and doesn't usually win.
3295 :
3296 : This is not done in the COMPARE case above because the
3297 : unmodified I2PAT is used in the PARALLEL and so a pattern
3298 : with a modified I2SRC would not match. */
3299 :
3300 32193659 : if (flag_expensive_optimizations)
3301 : {
3302 : /* Pass pc_rtx so no substitutions are done, just
3303 : simplifications. */
3304 29999807 : if (i1)
3305 : {
3306 9766718 : subst_low_luid = DF_INSN_LUID (i1);
3307 9766718 : i1src = subst (i1src, pc_rtx, pc_rtx, false, false, false);
3308 : }
3309 :
3310 29999807 : subst_low_luid = DF_INSN_LUID (i2);
3311 29999807 : i2src = subst (i2src, pc_rtx, pc_rtx, false, false, false);
3312 : }
3313 :
3314 32193659 : n_occurrences = 0; /* `subst' counts here */
3315 32193659 : subst_low_luid = DF_INSN_LUID (i2);
3316 :
3317 : /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3318 : copy of I2SRC each time we substitute it, in order to avoid creating
3319 : self-referential RTL when we will be substituting I1SRC for I1DEST
3320 : later. Likewise if I0 feeds into I2, either directly or indirectly
3321 : through I1, and I0DEST is in I0SRC. */
3322 32193659 : newpat = subst (PATTERN (i3), i2dest, i2src, false, false,
3323 32193659 : (i1_feeds_i2_n && i1dest_in_i1src)
3324 32193659 : || ((i0_feeds_i2_n || (i0_feeds_i1_n && i1_feeds_i2_n))
3325 : && i0dest_in_i0src));
3326 32193659 : substed_i2 = true;
3327 :
3328 : /* Record whether I2's body now appears within I3's body. */
3329 32193659 : i2_is_used = n_occurrences;
3330 : }
3331 :
3332 : /* If we already got a failure, don't try to do more. Otherwise, try to
3333 : substitute I1 if we have it. */
3334 :
3335 32701754 : if (i1 && GET_CODE (newpat) != CLOBBER)
3336 : {
3337 : /* Before we can do this substitution, we must redo the test done
3338 : above (see detailed comments there) that ensures I1DEST isn't
3339 : mentioned in any SETs in NEWPAT that are field assignments. */
3340 10367725 : if (!combinable_i3pat (NULL, &newpat, i1dest, NULL_RTX, NULL_RTX,
3341 : false, false, 0))
3342 : {
3343 38 : undo_all ();
3344 38 : return 0;
3345 : }
3346 :
3347 10367687 : n_occurrences = 0;
3348 10367687 : subst_low_luid = DF_INSN_LUID (i1);
3349 :
3350 : /* If the following substitution will modify I1SRC, make a copy of it
3351 : for the case where it is substituted for I1DEST in I2PAT later. */
3352 10367687 : if (added_sets_2 && i1_feeds_i2_n)
3353 1427664 : i1src_copy = copy_rtx (i1src);
3354 :
3355 : /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3356 : copy of I1SRC each time we substitute it, in order to avoid creating
3357 : self-referential RTL when we will be substituting I0SRC for I0DEST
3358 : later. */
3359 20735374 : newpat = subst (newpat, i1dest, i1src, false, false,
3360 10367687 : i0_feeds_i1_n && i0dest_in_i0src);
3361 10367687 : substed_i1 = true;
3362 :
3363 : /* Record whether I1's body now appears within I3's body. */
3364 10367687 : i1_is_used = n_occurrences;
3365 : }
3366 :
3367 : /* Likewise for I0 if we have it. */
3368 :
3369 32701716 : if (i0 && GET_CODE (newpat) != CLOBBER)
3370 : {
3371 1787394 : if (!combinable_i3pat (NULL, &newpat, i0dest, NULL_RTX, NULL_RTX,
3372 : false, false, 0))
3373 : {
3374 2 : undo_all ();
3375 2 : return 0;
3376 : }
3377 :
3378 : /* If the following substitution will modify I0SRC, make a copy of it
3379 : for the case where it is substituted for I0DEST in I1PAT later. */
3380 1787392 : if (added_sets_1 && i0_feeds_i1_n)
3381 354359 : i0src_copy = copy_rtx (i0src);
3382 : /* And a copy for I0DEST in I2PAT substitution. */
3383 1787392 : if (added_sets_2 && ((i0_feeds_i1_n && i1_feeds_i2_n)
3384 203161 : || (i0_feeds_i2_n)))
3385 317554 : i0src_copy2 = copy_rtx (i0src);
3386 :
3387 1787392 : n_occurrences = 0;
3388 1787392 : subst_low_luid = DF_INSN_LUID (i0);
3389 1787392 : newpat = subst (newpat, i0dest, i0src, false, false, false);
3390 1787392 : substed_i0 = true;
3391 : }
3392 :
3393 32701714 : if (n_auto_inc)
3394 : {
3395 515948 : int new_n_auto_inc = 0;
3396 515948 : for_each_inc_dec (newpat, count_auto_inc, &new_n_auto_inc);
3397 :
3398 515948 : if (n_auto_inc != new_n_auto_inc)
3399 : {
3400 1126 : if (dump_file && (dump_flags & TDF_DETAILS))
3401 0 : fprintf (dump_file, "Number of auto_inc expressions changed\n");
3402 1126 : undo_all ();
3403 1126 : return 0;
3404 : }
3405 : }
3406 :
3407 : /* Fail if an autoincrement side-effect has been duplicated. Be careful
3408 : to count all the ways that I2SRC and I1SRC can be used. */
3409 32700588 : if ((FIND_REG_INC_NOTE (i2, NULL_RTX) != 0
3410 : && i2_is_used + added_sets_2 > 1)
3411 : || (i1 != 0 && FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
3412 : && (i1_is_used + added_sets_1 + (added_sets_2 && i1_feeds_i2_n) > 1))
3413 : || (i0 != 0 && FIND_REG_INC_NOTE (i0, NULL_RTX) != 0
3414 : && (n_occurrences + added_sets_0
3415 : + (added_sets_1 && i0_feeds_i1_n)
3416 : + (added_sets_2 && i0_feeds_i2_n) > 1))
3417 : /* Fail if we tried to make a new register. */
3418 32700588 : || max_reg_num () != maxreg
3419 : /* Fail if we couldn't do something and have a CLOBBER. */
3420 32700588 : || GET_CODE (newpat) == CLOBBER
3421 : /* Fail if this new pattern is a MULT and we didn't have one before
3422 : at the outer level. */
3423 65089051 : || (GET_CODE (newpat) == SET && GET_CODE (SET_SRC (newpat)) == MULT
3424 271928 : && ! have_mult))
3425 : {
3426 334764 : undo_all ();
3427 334764 : return 0;
3428 : }
3429 :
3430 : /* If the actions of the earlier insns must be kept
3431 : in addition to substituting them into the latest one,
3432 : we must make a new PARALLEL for the latest insn
3433 : to hold additional the SETs. */
3434 :
3435 32365824 : if (added_sets_0 || added_sets_1 || added_sets_2)
3436 : {
3437 10759625 : int extra_sets = added_sets_0 + added_sets_1 + added_sets_2;
3438 10759625 : combine_extras++;
3439 :
3440 10759625 : if (GET_CODE (newpat) == PARALLEL)
3441 : {
3442 2077167 : rtvec old = XVEC (newpat, 0);
3443 2077167 : total_sets = XVECLEN (newpat, 0) + extra_sets;
3444 2077167 : newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
3445 2077167 : memcpy (XVEC (newpat, 0)->elem, &old->elem[0],
3446 2077167 : sizeof (old->elem[0]) * old->num_elem);
3447 : }
3448 : else
3449 : {
3450 8682458 : rtx old = newpat;
3451 8682458 : total_sets = 1 + extra_sets;
3452 8682458 : newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
3453 8682458 : XVECEXP (newpat, 0, 0) = old;
3454 : }
3455 :
3456 10759625 : if (added_sets_0)
3457 485041 : XVECEXP (newpat, 0, --total_sets) = i0pat;
3458 :
3459 10759625 : if (added_sets_1)
3460 : {
3461 3945910 : rtx t = i1pat;
3462 3945910 : if (i0_feeds_i1_n)
3463 352625 : t = subst (t, i0dest, i0src_copy ? i0src_copy : i0src,
3464 : false, false, false);
3465 :
3466 3945910 : XVECEXP (newpat, 0, --total_sets) = t;
3467 : }
3468 10759625 : if (added_sets_2)
3469 : {
3470 7229898 : rtx t = i2pat;
3471 7229898 : if (i1_feeds_i2_n)
3472 1416326 : t = subst (t, i1dest, i1src_copy ? i1src_copy : i1src, false, false,
3473 1416326 : i0_feeds_i1_n && i0dest_in_i0src);
3474 7229898 : if ((i0_feeds_i1_n && i1_feeds_i2_n) || i0_feeds_i2_n)
3475 316558 : t = subst (t, i0dest, i0src_copy2 ? i0src_copy2 : i0src,
3476 : false, false, false);
3477 :
3478 7229898 : XVECEXP (newpat, 0, --total_sets) = t;
3479 : }
3480 : }
3481 :
3482 25135926 : validate_replacement:
3483 :
3484 : /* Note which hard regs this insn has as inputs. */
3485 32706858 : mark_used_regs_combine (newpat);
3486 :
3487 : /* If recog_for_combine fails, it strips existing clobbers. If we'll
3488 : consider splitting this pattern, we might need these clobbers. */
3489 32706858 : if (i1 && GET_CODE (newpat) == PARALLEL
3490 7174564 : && GET_CODE (XVECEXP (newpat, 0, XVECLEN (newpat, 0) - 1)) == CLOBBER)
3491 : {
3492 1658316 : int len = XVECLEN (newpat, 0);
3493 :
3494 1658316 : newpat_vec_with_clobbers = rtvec_alloc (len);
3495 6683688 : for (i = 0; i < len; i++)
3496 3367056 : RTVEC_ELT (newpat_vec_with_clobbers, i) = XVECEXP (newpat, 0, i);
3497 : }
3498 :
3499 : /* We have recognized nothing yet. */
3500 32706858 : insn_code_number = -1;
3501 :
3502 : /* See if this is a PARALLEL of two SETs where one SET's destination is
3503 : a register that is unused and this isn't marked as an instruction that
3504 : might trap in an EH region. In that case, we just need the other SET.
3505 : We prefer this over the PARALLEL.
3506 :
3507 : This can occur when simplifying a divmod insn. We *must* test for this
3508 : case here because the code below that splits two independent SETs doesn't
3509 : handle this case correctly when it updates the register status.
3510 :
3511 : It's pointless doing this if we originally had two sets, one from
3512 : i3, and one from i2. Combining then splitting the parallel results
3513 : in the original i2 again plus an invalid insn (which we delete).
3514 : The net effect is only to move instructions around, which makes
3515 : debug info less accurate.
3516 :
3517 : If the remaining SET came from I2 its destination should not be used
3518 : between I2 and I3. See PR82024. */
3519 :
3520 7229898 : if (!(added_sets_2 && i1 == 0)
3521 27398137 : && is_parallel_of_n_reg_sets (newpat, 2)
3522 34257686 : && asm_noperands (newpat) < 0)
3523 : {
3524 1549941 : rtx set0 = XVECEXP (newpat, 0, 0);
3525 1549941 : rtx set1 = XVECEXP (newpat, 0, 1);
3526 1549941 : rtx oldpat = newpat;
3527 :
3528 1549941 : if (((REG_P (SET_DEST (set1))
3529 1549941 : && find_reg_note (i3, REG_UNUSED, SET_DEST (set1)))
3530 1508887 : || (GET_CODE (SET_DEST (set1)) == SUBREG
3531 0 : && find_reg_note (i3, REG_UNUSED, SUBREG_REG (SET_DEST (set1)))))
3532 41054 : && insn_nothrow_p (i3)
3533 1589746 : && !side_effects_p (SET_SRC (set1)))
3534 : {
3535 39538 : newpat = set0;
3536 39538 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3537 : }
3538 :
3539 1510403 : else if (((REG_P (SET_DEST (set0))
3540 1510403 : && find_reg_note (i3, REG_UNUSED, SET_DEST (set0)))
3541 1486651 : || (GET_CODE (SET_DEST (set0)) == SUBREG
3542 0 : && find_reg_note (i3, REG_UNUSED,
3543 0 : SUBREG_REG (SET_DEST (set0)))))
3544 23752 : && insn_nothrow_p (i3)
3545 1533575 : && !side_effects_p (SET_SRC (set0)))
3546 : {
3547 23129 : rtx dest = SET_DEST (set1);
3548 23129 : if (GET_CODE (dest) == SUBREG)
3549 0 : dest = SUBREG_REG (dest);
3550 23129 : if (!reg_used_between_p (dest, i2, i3))
3551 : {
3552 23128 : newpat = set1;
3553 23128 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3554 :
3555 23128 : if (insn_code_number >= 0)
3556 : changed_i3_dest = true;
3557 : }
3558 : }
3559 :
3560 39538 : if (insn_code_number < 0)
3561 1544405 : newpat = oldpat;
3562 : }
3563 :
3564 : /* Is the result of combination a valid instruction? */
3565 1544405 : if (insn_code_number < 0)
3566 32701322 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3567 :
3568 : /* If we were combining three insns and the result is a simple SET
3569 : with no ASM_OPERANDS that wasn't recognized, try to split it into two
3570 : insns. There are two ways to do this. It can be split using a
3571 : machine-specific method (like when you have an addition of a large
3572 : constant) or by combine in the function find_split_point. */
3573 :
3574 10225926 : if (i1 && insn_code_number < 0 && GET_CODE (newpat) == SET
3575 37329904 : && asm_noperands (newpat) < 0)
3576 : {
3577 4622569 : rtx parallel, *split;
3578 4622569 : rtx_insn *m_split_insn;
3579 4622569 : unsigned int old_nregs, new_nregs;
3580 :
3581 : /* See if the MD file can split NEWPAT. If it can't, see if letting it
3582 : use I2DEST as a scratch register will help. In the latter case,
3583 : convert I2DEST to the mode of the source of NEWPAT if we can. */
3584 :
3585 4622569 : m_split_insn = combine_split_insns (newpat, i3, &old_nregs, &new_nregs);
3586 :
3587 : /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3588 : inputs of NEWPAT. */
3589 :
3590 : /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3591 : possible to try that as a scratch reg. This would require adding
3592 : more code to make it work though. */
3593 :
3594 4622569 : if (m_split_insn == 0 && ! reg_overlap_mentioned_p (i2dest, newpat))
3595 : {
3596 4486950 : machine_mode new_mode = GET_MODE (SET_DEST (newpat));
3597 :
3598 : /* ??? Reusing i2dest without resetting the reg_stat entry for it
3599 : (temporarily, until we are committed to this instruction
3600 : combination) does not work: for example, any call to nonzero_bits
3601 : on the register (from a splitter in the MD file, for example)
3602 : will get the old information, which is invalid.
3603 :
3604 : Since nowadays we can create registers during combine just fine,
3605 : we should just create a new one here, not reuse i2dest. */
3606 :
3607 : /* First try to split using the original register as a
3608 : scratch register. */
3609 4486950 : parallel = gen_rtx_PARALLEL (VOIDmode,
3610 : gen_rtvec (2, newpat,
3611 : gen_rtx_CLOBBER (VOIDmode,
3612 : i2dest)));
3613 4486950 : m_split_insn = combine_split_insns (parallel, i3, &old_nregs, &new_nregs);
3614 :
3615 : /* If that didn't work, try changing the mode of I2DEST if
3616 : we can. */
3617 4486950 : if (m_split_insn == 0
3618 4486950 : && new_mode != GET_MODE (i2dest)
3619 1720830 : && new_mode != VOIDmode
3620 5637173 : && can_change_dest_mode (i2dest, added_sets_2, new_mode))
3621 : {
3622 857692 : machine_mode old_mode = GET_MODE (i2dest);
3623 857692 : rtx ni2dest;
3624 :
3625 857692 : if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
3626 9976 : ni2dest = gen_rtx_REG (new_mode, REGNO (i2dest));
3627 : else
3628 : {
3629 847716 : subst_mode (REGNO (i2dest), new_mode);
3630 847716 : ni2dest = regno_reg_rtx[REGNO (i2dest)];
3631 : }
3632 :
3633 857692 : parallel = (gen_rtx_PARALLEL
3634 : (VOIDmode,
3635 : gen_rtvec (2, newpat,
3636 : gen_rtx_CLOBBER (VOIDmode,
3637 : ni2dest))));
3638 857692 : m_split_insn = combine_split_insns (parallel, i3, &old_nregs, &new_nregs);
3639 :
3640 857692 : if (m_split_insn == 0
3641 857692 : && REGNO (i2dest) >= FIRST_PSEUDO_REGISTER)
3642 : {
3643 847716 : struct undo *buf;
3644 :
3645 847716 : adjust_reg_mode (regno_reg_rtx[REGNO (i2dest)], old_mode);
3646 847716 : buf = undobuf.undos;
3647 847716 : undobuf.undos = buf->next;
3648 847716 : buf->next = undobuf.frees;
3649 847716 : undobuf.frees = buf;
3650 : }
3651 : }
3652 :
3653 4486950 : i2scratch = m_split_insn != 0;
3654 : }
3655 :
3656 : /* If recog_for_combine has discarded clobbers, try to use them
3657 : again for the split. */
3658 4622569 : if (m_split_insn == 0 && newpat_vec_with_clobbers)
3659 : {
3660 1606062 : parallel = gen_rtx_PARALLEL (VOIDmode, newpat_vec_with_clobbers);
3661 1606062 : m_split_insn = combine_split_insns (parallel, i3, &old_nregs, &new_nregs);
3662 : }
3663 :
3664 4634223 : if (m_split_insn && NEXT_INSN (m_split_insn) == NULL_RTX)
3665 : {
3666 1476 : rtx m_split_pat = PATTERN (m_split_insn);
3667 1476 : insn_code_number = recog_for_combine (&m_split_pat, i3, &new_i3_notes,
3668 : old_nregs, new_nregs);
3669 1476 : if (insn_code_number >= 0)
3670 177 : newpat = m_split_pat;
3671 : }
3672 10178 : else if (m_split_insn && NEXT_INSN (NEXT_INSN (m_split_insn)) == NULL_RTX
3673 4631271 : && (next_nonnote_nondebug_insn (i2) == i3
3674 6 : || !modified_between_p (PATTERN (m_split_insn), i2, i3)))
3675 : {
3676 10178 : rtx i2set, i3set;
3677 10178 : rtx newi3pat = PATTERN (NEXT_INSN (m_split_insn));
3678 10178 : newi2pat = PATTERN (m_split_insn);
3679 :
3680 10178 : i3set = single_set (NEXT_INSN (m_split_insn));
3681 10178 : i2set = single_set (m_split_insn);
3682 :
3683 10178 : i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3684 :
3685 : /* If I2 or I3 has multiple SETs, we won't know how to track
3686 : register status, so don't use these insns. If I2's destination
3687 : is used between I2 and I3, we also can't use these insns. */
3688 :
3689 10178 : if (i2_code_number >= 0 && i2set && i3set
3690 20356 : && (next_nonnote_nondebug_insn (i2) == i3
3691 6 : || ! reg_used_between_p (SET_DEST (i2set), i2, i3)))
3692 10178 : insn_code_number = recog_for_combine (&newi3pat, i3,
3693 : &new_i3_notes,
3694 : old_nregs, new_nregs);
3695 10178 : if (insn_code_number >= 0)
3696 10178 : newpat = newi3pat;
3697 :
3698 : /* It is possible that both insns now set the destination of I3.
3699 : If so, we must show an extra use of it. */
3700 :
3701 10178 : if (insn_code_number >= 0)
3702 : {
3703 10178 : rtx new_i3_dest = SET_DEST (i3set);
3704 10178 : rtx new_i2_dest = SET_DEST (i2set);
3705 :
3706 10178 : while (GET_CODE (new_i3_dest) == ZERO_EXTRACT
3707 10218 : || GET_CODE (new_i3_dest) == STRICT_LOW_PART
3708 20418 : || GET_CODE (new_i3_dest) == SUBREG)
3709 40 : new_i3_dest = XEXP (new_i3_dest, 0);
3710 :
3711 10178 : while (GET_CODE (new_i2_dest) == ZERO_EXTRACT
3712 10178 : || GET_CODE (new_i2_dest) == STRICT_LOW_PART
3713 20356 : || GET_CODE (new_i2_dest) == SUBREG)
3714 0 : new_i2_dest = XEXP (new_i2_dest, 0);
3715 :
3716 10178 : if (REG_P (new_i3_dest)
3717 6268 : && REG_P (new_i2_dest)
3718 6268 : && REGNO (new_i3_dest) == REGNO (new_i2_dest)
3719 10178 : && REGNO (new_i2_dest) < reg_n_sets_max)
3720 0 : INC_REG_N_SETS (REGNO (new_i2_dest), 1);
3721 : }
3722 : }
3723 :
3724 : /* If we can split it and use I2DEST, go ahead and see if that
3725 : helps things be recognized. Verify that none of the registers
3726 : are set between I2 and I3. */
3727 1299 : if (insn_code_number < 0
3728 4612214 : && (split = find_split_point (&newpat, i3, false)) != 0
3729 : /* We need I2DEST in the proper mode. If it is a hard register
3730 : or the only use of a pseudo, we can change its mode.
3731 : Make sure we don't change a hard register to have a mode that
3732 : isn't valid for it, or change the number of registers. */
3733 4354495 : && (GET_MODE (*split) == GET_MODE (i2dest)
3734 1640241 : || GET_MODE (*split) == VOIDmode
3735 1288852 : || can_change_dest_mode (i2dest, added_sets_2,
3736 : GET_MODE (*split)))
3737 3642684 : && (next_nonnote_nondebug_insn (i2) == i3
3738 580822 : || !modified_between_p (*split, i2, i3))
3739 : /* We can't overwrite I2DEST if its value is still used by
3740 : NEWPAT. */
3741 3612699 : && ! reg_referenced_p (i2dest, newpat)
3742 : /* We should not split a possibly trapping part when we
3743 : care about non-call EH and have REG_EH_REGION notes
3744 : to distribute. */
3745 8163907 : && ! (cfun->can_throw_non_call_exceptions
3746 378761 : && has_non_call_exception
3747 118 : && may_trap_p (*split)))
3748 : {
3749 3542519 : rtx newdest = i2dest;
3750 3542519 : enum rtx_code split_code = GET_CODE (*split);
3751 3542519 : machine_mode split_mode = GET_MODE (*split);
3752 3542519 : bool subst_done = false;
3753 3542519 : newi2pat = NULL_RTX;
3754 :
3755 3542519 : i2scratch = true;
3756 :
3757 : /* *SPLIT may be part of I2SRC, so make sure we have the
3758 : original expression around for later debug processing.
3759 : We should not need I2SRC any more in other cases. */
3760 3542519 : if (MAY_HAVE_DEBUG_BIND_INSNS)
3761 1720228 : i2src = copy_rtx (i2src);
3762 : else
3763 1822291 : i2src = NULL;
3764 :
3765 : /* Get NEWDEST as a register in the proper mode. We have already
3766 : validated that we can do this. */
3767 3542519 : if (GET_MODE (i2dest) != split_mode && split_mode != VOIDmode)
3768 : {
3769 573626 : if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
3770 0 : newdest = gen_rtx_REG (split_mode, REGNO (i2dest));
3771 : else
3772 : {
3773 573626 : subst_mode (REGNO (i2dest), split_mode);
3774 573626 : newdest = regno_reg_rtx[REGNO (i2dest)];
3775 : }
3776 : }
3777 :
3778 : /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3779 : an ASHIFT. This can occur if it was inside a PLUS and hence
3780 : appeared to be a memory address. This is a kludge. */
3781 3542519 : if (split_code == MULT
3782 204475 : && CONST_INT_P (XEXP (*split, 1))
3783 105732 : && INTVAL (XEXP (*split, 1)) > 0
3784 3645178 : && (i = exact_log2 (UINTVAL (XEXP (*split, 1)))) >= 0)
3785 : {
3786 72573 : rtx i_rtx = gen_int_shift_amount (split_mode, i);
3787 72573 : SUBST (*split, gen_rtx_ASHIFT (split_mode,
3788 : XEXP (*split, 0), i_rtx));
3789 : /* Update split_code because we may not have a multiply
3790 : anymore. */
3791 72573 : split_code = GET_CODE (*split);
3792 : }
3793 :
3794 : /* Similarly for (plus (mult FOO (const_int pow2))). */
3795 3542519 : if (split_code == PLUS
3796 675118 : && GET_CODE (XEXP (*split, 0)) == MULT
3797 108701 : && CONST_INT_P (XEXP (XEXP (*split, 0), 1))
3798 38947 : && INTVAL (XEXP (XEXP (*split, 0), 1)) > 0
3799 3577518 : && (i = exact_log2 (UINTVAL (XEXP (XEXP (*split, 0), 1)))) >= 0)
3800 : {
3801 6790 : rtx nsplit = XEXP (*split, 0);
3802 6790 : rtx i_rtx = gen_int_shift_amount (GET_MODE (nsplit), i);
3803 6790 : SUBST (XEXP (*split, 0), gen_rtx_ASHIFT (GET_MODE (nsplit),
3804 : XEXP (nsplit, 0),
3805 : i_rtx));
3806 : /* Update split_code because we may not have a multiply
3807 : anymore. */
3808 6790 : split_code = GET_CODE (*split);
3809 : }
3810 :
3811 : #ifdef INSN_SCHEDULING
3812 : /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3813 : be written as a ZERO_EXTEND. */
3814 3542519 : if (split_code == SUBREG && MEM_P (SUBREG_REG (*split)))
3815 : {
3816 : /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3817 : what it really is. */
3818 12465 : if (load_extend_op (GET_MODE (SUBREG_REG (*split)))
3819 : == SIGN_EXTEND)
3820 : SUBST (*split, gen_rtx_SIGN_EXTEND (split_mode,
3821 : SUBREG_REG (*split)));
3822 : else
3823 12465 : SUBST (*split, gen_rtx_ZERO_EXTEND (split_mode,
3824 : SUBREG_REG (*split)));
3825 : }
3826 : #endif
3827 :
3828 : /* Attempt to split binary operators using arithmetic identities. */
3829 3542519 : if (BINARY_P (SET_SRC (newpat))
3830 2975238 : && split_mode == GET_MODE (SET_SRC (newpat))
3831 5572766 : && ! side_effects_p (SET_SRC (newpat)))
3832 : {
3833 2015928 : rtx setsrc = SET_SRC (newpat);
3834 2015928 : machine_mode mode = GET_MODE (setsrc);
3835 2015928 : enum rtx_code code = GET_CODE (setsrc);
3836 2015928 : rtx src_op0 = XEXP (setsrc, 0);
3837 2015928 : rtx src_op1 = XEXP (setsrc, 1);
3838 :
3839 : /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3840 2015928 : if (rtx_equal_p (src_op0, src_op1))
3841 : {
3842 1471 : newi2pat = gen_rtx_SET (newdest, src_op0);
3843 1471 : SUBST (XEXP (setsrc, 0), newdest);
3844 1471 : SUBST (XEXP (setsrc, 1), newdest);
3845 1471 : subst_done = true;
3846 : }
3847 : /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3848 2014457 : else if ((code == PLUS || code == MULT)
3849 978238 : && GET_CODE (src_op0) == code
3850 416456 : && GET_CODE (XEXP (src_op0, 0)) == code
3851 173740 : && (INTEGRAL_MODE_P (mode)
3852 : || (FLOAT_MODE_P (mode)
3853 99989 : && flag_unsafe_math_optimizations)))
3854 : {
3855 77704 : rtx p = XEXP (XEXP (src_op0, 0), 0);
3856 77704 : rtx q = XEXP (XEXP (src_op0, 0), 1);
3857 77704 : rtx r = XEXP (src_op0, 1);
3858 77704 : rtx s = src_op1;
3859 :
3860 : /* Split both "((X op Y) op X) op Y" and
3861 : "((X op Y) op Y) op X" as "T op T" where T is
3862 : "X op Y". */
3863 77962 : if ((rtx_equal_p (p,r) && rtx_equal_p (q,s))
3864 77863 : || (rtx_equal_p (p,s) && rtx_equal_p (q,r)))
3865 : {
3866 99 : newi2pat = gen_rtx_SET (newdest, XEXP (src_op0, 0));
3867 99 : SUBST (XEXP (setsrc, 0), newdest);
3868 99 : SUBST (XEXP (setsrc, 1), newdest);
3869 99 : subst_done = true;
3870 : }
3871 : /* Split "((X op X) op Y) op Y)" as "T op T" where
3872 : T is "X op Y". */
3873 77605 : else if (rtx_equal_p (p,q) && rtx_equal_p (r,s))
3874 : {
3875 60 : rtx tmp = simplify_gen_binary (code, mode, p, r);
3876 60 : newi2pat = gen_rtx_SET (newdest, tmp);
3877 60 : SUBST (XEXP (setsrc, 0), newdest);
3878 60 : SUBST (XEXP (setsrc, 1), newdest);
3879 60 : subst_done = true;
3880 : }
3881 : }
3882 : }
3883 :
3884 1630 : if (!subst_done)
3885 : {
3886 3540889 : newi2pat = gen_rtx_SET (newdest, *split);
3887 3540889 : SUBST (*split, newdest);
3888 : }
3889 :
3890 3542519 : i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3891 :
3892 : /* recog_for_combine might have added CLOBBERs to newi2pat.
3893 : Make sure NEWPAT does not depend on the clobbered regs. */
3894 3542519 : if (GET_CODE (newi2pat) == PARALLEL)
3895 2447573 : for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
3896 1642501 : if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
3897 : {
3898 837429 : rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
3899 837429 : if (reg_overlap_mentioned_p (reg, newpat))
3900 : {
3901 18884 : undo_all ();
3902 18884 : return 0;
3903 : }
3904 : }
3905 :
3906 : /* If the split point was a MULT and we didn't have one before,
3907 : don't use one now. */
3908 3523635 : if (i2_code_number >= 0 && ! (split_code == MULT && ! have_mult))
3909 2120372 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3910 : }
3911 : }
3912 :
3913 : /* Check for a case where we loaded from memory in a narrow mode and
3914 : then sign extended it, but we need both registers. In that case,
3915 : we have a PARALLEL with both loads from the same memory location.
3916 : We can split this into a load from memory followed by a register-register
3917 : copy. This saves at least one insn, more if register allocation can
3918 : eliminate the copy.
3919 :
3920 : We cannot do this if the involved modes have more than one elements,
3921 : like for vector or complex modes.
3922 :
3923 : We cannot do this if the destination of the first assignment is a
3924 : condition code register. We eliminate this case by making sure
3925 : the SET_DEST and SET_SRC have the same mode.
3926 :
3927 : We cannot do this if the destination of the second assignment is
3928 : a register that we have already assumed is zero-extended. Similarly
3929 : for a SUBREG of such a register. */
3930 :
3931 5603357 : else if (i1 && insn_code_number < 0 && asm_noperands (newpat) < 0
3932 5546842 : && GET_CODE (newpat) == PARALLEL
3933 5545046 : && XVECLEN (newpat, 0) == 2
3934 4604265 : && GET_CODE (XVECEXP (newpat, 0, 0)) == SET
3935 4604067 : && GET_CODE (SET_SRC (XVECEXP (newpat, 0, 0))) == SIGN_EXTEND
3936 21676 : && (GET_MODE (SET_DEST (XVECEXP (newpat, 0, 0)))
3937 21676 : == GET_MODE (SET_SRC (XVECEXP (newpat, 0, 0))))
3938 21676 : && ! VECTOR_MODE_P (GET_MODE (SET_DEST (XVECEXP (newpat, 0, 0))))
3939 : && ! COMPLEX_MODE_P (GET_MODE (SET_DEST (XVECEXP (newpat, 0, 0))))
3940 20257 : && GET_CODE (XVECEXP (newpat, 0, 1)) == SET
3941 20257 : && rtx_equal_p (SET_SRC (XVECEXP (newpat, 0, 1)),
3942 20257 : XEXP (SET_SRC (XVECEXP (newpat, 0, 0)), 0))
3943 5074 : && !modified_between_p (SET_SRC (XVECEXP (newpat, 0, 1)), i2, i3)
3944 5074 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
3945 5074 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
3946 5074 : && ! (temp_expr = SET_DEST (XVECEXP (newpat, 0, 1)),
3947 : (REG_P (temp_expr)
3948 5074 : && reg_stat[REGNO (temp_expr)].nonzero_bits != 0
3949 5210 : && known_lt (GET_MODE_PRECISION (GET_MODE (temp_expr)),
3950 : BITS_PER_WORD)
3951 4945 : && known_lt (GET_MODE_PRECISION (GET_MODE (temp_expr)),
3952 : HOST_BITS_PER_INT)
3953 1182 : && (reg_stat[REGNO (temp_expr)].nonzero_bits
3954 1182 : != GET_MODE_MASK (word_mode))))
3955 5005 : && ! (GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) == SUBREG
3956 0 : && (temp_expr = SUBREG_REG (SET_DEST (XVECEXP (newpat, 0, 1))),
3957 0 : (REG_P (temp_expr)
3958 0 : && reg_stat[REGNO (temp_expr)].nonzero_bits != 0
3959 0 : && known_lt (GET_MODE_PRECISION (GET_MODE (temp_expr)),
3960 : BITS_PER_WORD)
3961 0 : && known_lt (GET_MODE_PRECISION (GET_MODE (temp_expr)),
3962 : HOST_BITS_PER_INT)
3963 0 : && (reg_stat[REGNO (temp_expr)].nonzero_bits
3964 0 : != GET_MODE_MASK (word_mode)))))
3965 5005 : && ! reg_overlap_mentioned_p (SET_DEST (XVECEXP (newpat, 0, 1)),
3966 5005 : SET_SRC (XVECEXP (newpat, 0, 1)))
3967 28089247 : && ! find_reg_note (i3, REG_UNUSED,
3968 4958 : SET_DEST (XVECEXP (newpat, 0, 0))))
3969 : {
3970 4958 : rtx ni2dest;
3971 :
3972 4958 : newi2pat = XVECEXP (newpat, 0, 0);
3973 4958 : ni2dest = SET_DEST (XVECEXP (newpat, 0, 0));
3974 4958 : newpat = XVECEXP (newpat, 0, 1);
3975 4958 : SUBST (SET_SRC (newpat),
3976 : gen_lowpart (GET_MODE (SET_SRC (newpat)), ni2dest));
3977 4958 : i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3978 :
3979 4958 : if (i2_code_number >= 0)
3980 0 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3981 :
3982 4958 : if (insn_code_number >= 0)
3983 : swap_i2i3 = 1;
3984 : }
3985 :
3986 : /* Similarly, check for a case where we have a PARALLEL of two independent
3987 : SETs but we started with three insns. In this case, we can do the sets
3988 : as two separate insns. This case occurs when some SET allows two
3989 : other insns to combine, but the destination of that SET is still live.
3990 :
3991 : Also do this if we started with two insns and (at least) one of the
3992 : resulting sets is a noop; this noop will be deleted later.
3993 :
3994 : Also do this if we started with two insns neither of which was a simple
3995 : move. */
3996 :
3997 24054434 : else if (insn_code_number < 0 && asm_noperands (newpat) < 0
3998 24036996 : && GET_CODE (newpat) == PARALLEL
3999 10961527 : && XVECLEN (newpat, 0) == 2
4000 9930294 : && GET_CODE (XVECEXP (newpat, 0, 0)) == SET
4001 9825309 : && GET_CODE (XVECEXP (newpat, 0, 1)) == SET
4002 9766990 : && (i1
4003 5187387 : || set_noop_p (XVECEXP (newpat, 0, 0))
4004 5186930 : || set_noop_p (XVECEXP (newpat, 0, 1))
4005 5186928 : || (!i2_was_move && !i3_was_move))
4006 6458962 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != ZERO_EXTRACT
4007 6458245 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != STRICT_LOW_PART
4008 6458065 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
4009 6457484 : && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
4010 6457470 : && ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 1)),
4011 : XVECEXP (newpat, 0, 0))
4012 5359594 : && ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 0)),
4013 5359594 : XVECEXP (newpat, 0, 1))
4014 33714447 : && ! (contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 0)))
4015 405712 : && contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 1)))))
4016 : {
4017 5000670 : rtx set0 = XVECEXP (newpat, 0, 0);
4018 5000670 : rtx set1 = XVECEXP (newpat, 0, 1);
4019 :
4020 : /* Normally, it doesn't matter which of the two is done first, but
4021 : one which uses any regs/memory set or used in between i2 and i3
4022 : can't be first. The PARALLEL might also have been pre-existing
4023 : in i3, so we need to make sure that we won't wrongly hoist a SET
4024 : to i2 that would conflict with a death note present in there, or
4025 : would have its dest modified or used between i2 and i3. */
4026 5000670 : if ((set_noop_p (set1)
4027 5000670 : || (!modified_between_p (SET_SRC (set1), i2, i3)
4028 9958697 : && !(REG_P (SET_DEST (set1))
4029 4966333 : && find_reg_note (i2, REG_DEAD, SET_DEST (set1)))
4030 5018102 : && !(GET_CODE (SET_DEST (set1)) == SUBREG
4031 26031 : && find_reg_note (i2, REG_DEAD,
4032 26031 : SUBREG_REG (SET_DEST (set1))))
4033 4992071 : && !modified_between_p (SET_DEST (set1), i2, i3)
4034 4992071 : && !reg_used_between_p (SET_DEST (set1), i2, i3)))
4035 : /* If I3 is a jump, ensure that set0 is a jump so that
4036 : we do not create invalid RTL. */
4037 9992735 : && (!JUMP_P (i3) || SET_DEST (set0) == pc_rtx))
4038 : {
4039 4992065 : newi2pat = set1;
4040 4992065 : newpat = set0;
4041 : }
4042 8605 : else if ((set_noop_p (set0)
4043 8599 : || (!modified_between_p (SET_SRC (set0), i2, i3)
4044 590 : && !(REG_P (SET_DEST (set0))
4045 295 : && find_reg_note (i2, REG_DEAD, SET_DEST (set0)))
4046 295 : && !(GET_CODE (SET_DEST (set0)) == SUBREG
4047 0 : && find_reg_note (i2, REG_DEAD,
4048 0 : SUBREG_REG (SET_DEST (set0))))
4049 295 : && !modified_between_p (SET_DEST (set0), i2, i3)
4050 294 : && !reg_used_between_p (SET_DEST (set0), i2, i3)))
4051 : /* If I3 is a jump, ensure that set1 is a jump so that
4052 : we do not create invalid RTL. */
4053 8899 : && (!JUMP_P (i3) || SET_DEST (set1) == pc_rtx))
4054 : {
4055 300 : newi2pat = set0;
4056 300 : newpat = set1;
4057 : }
4058 : else
4059 : {
4060 8305 : undo_all ();
4061 8305 : return 0;
4062 : }
4063 :
4064 4992365 : i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
4065 :
4066 4992365 : if (i2_code_number >= 0)
4067 : {
4068 : /* recog_for_combine might have added CLOBBERs to newi2pat.
4069 : Make sure NEWPAT does not depend on the clobbered regs. */
4070 3688000 : if (GET_CODE (newi2pat) == PARALLEL)
4071 : {
4072 1294586 : for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
4073 866315 : if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
4074 : {
4075 438044 : rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
4076 438044 : if (reg_overlap_mentioned_p (reg, newpat))
4077 : {
4078 3447 : undo_all ();
4079 3447 : return 0;
4080 : }
4081 : }
4082 : }
4083 :
4084 3684553 : insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
4085 :
4086 : /* Likewise, recog_for_combine might have added clobbers to NEWPAT.
4087 : Checking that the SET0's SET_DEST and SET1's SET_DEST aren't
4088 : mentioned/clobbered, ensures NEWI2PAT's SET_DEST is live. */
4089 3684553 : if (insn_code_number >= 0 && GET_CODE (newpat) == PARALLEL)
4090 : {
4091 60767 : for (i = XVECLEN (newpat, 0) - 1; i >= 0; i--)
4092 40522 : if (GET_CODE (XVECEXP (newpat, 0, i)) == CLOBBER)
4093 : {
4094 20277 : rtx reg = XEXP (XVECEXP (newpat, 0, i), 0);
4095 20277 : if (reg_overlap_mentioned_p (reg, SET_DEST (set0))
4096 20277 : || reg_overlap_mentioned_p (reg, SET_DEST (set1)))
4097 : {
4098 0 : undo_all ();
4099 0 : return 0;
4100 : }
4101 : }
4102 : }
4103 :
4104 : if (insn_code_number >= 0)
4105 : split_i2i3 = true;
4106 : }
4107 : }
4108 :
4109 : /* If it still isn't recognized, fail and change things back the way they
4110 : were. */
4111 28986711 : if ((insn_code_number < 0
4112 : /* Is the result a reasonable ASM_OPERANDS? */
4113 32516072 : && (! check_asm_operands (newpat) || added_sets_1 || added_sets_2)))
4114 : {
4115 28436932 : undo_all ();
4116 28436932 : return 0;
4117 : }
4118 :
4119 : /* If we had to change another insn, make sure it is valid also. */
4120 4239290 : if (undobuf.other_insn)
4121 : {
4122 202752 : CLEAR_HARD_REG_SET (newpat_used_regs);
4123 :
4124 202752 : other_pat = PATTERN (undobuf.other_insn);
4125 202752 : other_code_number = recog_for_combine (&other_pat, undobuf.other_insn,
4126 : &new_other_notes);
4127 :
4128 202752 : if (other_code_number < 0 && ! check_asm_operands (other_pat))
4129 : {
4130 8178 : undo_all ();
4131 8178 : return 0;
4132 : }
4133 : }
4134 :
4135 : /* Reject this combination if insn_cost reports that the replacement
4136 : instructions are more expensive than the originals. */
4137 4231112 : if (!combine_validate_cost (i0, i1, i2, i3, newpat, newi2pat, other_pat))
4138 : {
4139 208956 : undo_all ();
4140 208956 : return 0;
4141 : }
4142 :
4143 4022156 : if (MAY_HAVE_DEBUG_BIND_INSNS)
4144 : {
4145 2186720 : struct undo *undo;
4146 :
4147 6557743 : for (undo = undobuf.undos; undo; undo = undo->next)
4148 4371023 : if (undo->kind == UNDO_MODE)
4149 : {
4150 1919 : rtx reg = regno_reg_rtx[undo->where.regno];
4151 1919 : machine_mode new_mode = GET_MODE (reg);
4152 1919 : machine_mode old_mode = undo->old_contents.m;
4153 :
4154 : /* Temporarily revert mode back. */
4155 1919 : adjust_reg_mode (reg, old_mode);
4156 :
4157 1919 : if (reg == i2dest && i2scratch)
4158 : {
4159 : /* If we used i2dest as a scratch register with a
4160 : different mode, substitute it for the original
4161 : i2src while its original mode is temporarily
4162 : restored, and then clear i2scratch so that we don't
4163 : do it again later. */
4164 1919 : propagate_for_debug (i2, last_combined_insn, reg, i2src,
4165 : this_basic_block);
4166 1919 : i2scratch = false;
4167 : /* Put back the new mode. */
4168 1919 : adjust_reg_mode (reg, new_mode);
4169 : }
4170 : else
4171 : {
4172 0 : rtx tempreg = gen_raw_REG (old_mode, REGNO (reg));
4173 0 : rtx_insn *first, *last;
4174 :
4175 0 : if (reg == i2dest)
4176 : {
4177 : first = i2;
4178 : last = last_combined_insn;
4179 : }
4180 : else
4181 : {
4182 0 : first = i3;
4183 0 : last = undobuf.other_insn;
4184 0 : gcc_assert (last);
4185 0 : if (DF_INSN_LUID (last)
4186 0 : < DF_INSN_LUID (last_combined_insn))
4187 0 : last = last_combined_insn;
4188 : }
4189 :
4190 : /* We're dealing with a reg that changed mode but not
4191 : meaning, so we want to turn it into a subreg for
4192 : the new mode. However, because of REG sharing and
4193 : because its mode had already changed, we have to do
4194 : it in two steps. First, replace any debug uses of
4195 : reg, with its original mode temporarily restored,
4196 : with this copy we have created; then, replace the
4197 : copy with the SUBREG of the original shared reg,
4198 : once again changed to the new mode. */
4199 0 : propagate_for_debug (first, last, reg, tempreg,
4200 : this_basic_block);
4201 0 : adjust_reg_mode (reg, new_mode);
4202 0 : propagate_for_debug (first, last, tempreg,
4203 : lowpart_subreg (old_mode, reg, new_mode),
4204 : this_basic_block);
4205 : }
4206 : }
4207 : }
4208 :
4209 : /* If we will be able to accept this, we have made a
4210 : change to the destination of I3. This requires us to
4211 : do a few adjustments. */
4212 :
4213 4022156 : if (changed_i3_dest)
4214 : {
4215 18161 : PATTERN (i3) = newpat;
4216 18161 : adjust_for_new_dest (i3);
4217 : }
4218 :
4219 4022156 : bool only_i3_changed = !i0 && !i1 && rtx_equal_p (newi2pat, PATTERN (i2));
4220 :
4221 : /* If only i3 has changed, any split of the combined instruction just
4222 : restored i2 to its original state. No destinations moved from i3
4223 : to i2. */
4224 : if (only_i3_changed)
4225 : split_i2i3 = false;
4226 :
4227 : /* We now know that we can do this combination. Merge the insns and
4228 : update the status of registers and LOG_LINKS. */
4229 :
4230 4022156 : if (undobuf.other_insn)
4231 : {
4232 194438 : rtx note, next;
4233 :
4234 194438 : PATTERN (undobuf.other_insn) = other_pat;
4235 :
4236 : /* If any of the notes in OTHER_INSN were REG_DEAD or REG_UNUSED,
4237 : ensure that they are still valid. Then add any non-duplicate
4238 : notes added by recog_for_combine. */
4239 580379 : for (note = REG_NOTES (undobuf.other_insn); note; note = next)
4240 : {
4241 385941 : next = XEXP (note, 1);
4242 :
4243 385941 : if ((REG_NOTE_KIND (note) == REG_DEAD
4244 197359 : && !reg_referenced_p (XEXP (note, 0),
4245 197359 : PATTERN (undobuf.other_insn)))
4246 381418 : ||(REG_NOTE_KIND (note) == REG_UNUSED
4247 28 : && !reg_set_p (XEXP (note, 0),
4248 28 : PATTERN (undobuf.other_insn)))
4249 : /* Simply drop equal note since it may be no longer valid
4250 : for other_insn. It may be possible to record that CC
4251 : register is changed and only discard those notes, but
4252 : in practice it's unnecessary complication and doesn't
4253 : give any meaningful improvement.
4254 :
4255 : See PR78559. */
4256 381418 : || REG_NOTE_KIND (note) == REG_EQUAL
4257 767221 : || REG_NOTE_KIND (note) == REG_EQUIV)
4258 4661 : remove_note (undobuf.other_insn, note);
4259 : }
4260 :
4261 194438 : distribute_notes (new_other_notes, undobuf.other_insn,
4262 : undobuf.other_insn, NULL, NULL_RTX, NULL_RTX,
4263 : NULL_RTX);
4264 : }
4265 :
4266 4022156 : if (swap_i2i3)
4267 : {
4268 : /* I3 now uses what used to be its destination and which is now
4269 : I2's destination. This requires us to do a few adjustments. */
4270 0 : PATTERN (i3) = newpat;
4271 0 : adjust_for_new_dest (i3);
4272 : }
4273 :
4274 4022156 : if (swap_i2i3 || split_i2i3)
4275 : {
4276 : /* We might need a LOG_LINK from I3 to I2. But then we used to
4277 : have one, so we still will.
4278 :
4279 : However, some later insn might be using I2's dest and have
4280 : a LOG_LINK pointing at I3. We should change it to point at
4281 : I2 instead. */
4282 :
4283 : /* newi2pat is usually a SET here; however, recog_for_combine might
4284 : have added some clobbers. */
4285 30127 : rtx x = newi2pat;
4286 30127 : if (GET_CODE (x) == PARALLEL)
4287 484 : x = XVECEXP (newi2pat, 0, 0);
4288 :
4289 30127 : if (REG_P (SET_DEST (x))
4290 6 : || (GET_CODE (SET_DEST (x)) == SUBREG
4291 0 : && REG_P (SUBREG_REG (SET_DEST (x)))))
4292 : {
4293 30121 : unsigned int regno = reg_or_subregno (SET_DEST (x));
4294 :
4295 30121 : bool done = false;
4296 521566 : for (rtx_insn *insn = NEXT_INSN (i3);
4297 521566 : !done
4298 521566 : && insn
4299 520213 : && INSN_P (insn)
4300 1013011 : && BLOCK_FOR_INSN (insn) == this_basic_block;
4301 491445 : insn = NEXT_INSN (insn))
4302 : {
4303 491445 : if (DEBUG_INSN_P (insn))
4304 176741 : continue;
4305 314704 : struct insn_link *link;
4306 589999 : FOR_EACH_LOG_LINK (link, insn)
4307 275305 : if (link->insn == i3 && link->regno == regno)
4308 : {
4309 10 : link->insn = i2;
4310 10 : done = true;
4311 10 : break;
4312 : }
4313 : }
4314 : }
4315 : }
4316 :
4317 4022156 : {
4318 4022156 : rtx i3notes, i2notes, i1notes = 0, i0notes = 0;
4319 4022156 : struct insn_link *i3links, *i2links, *i1links = 0, *i0links = 0;
4320 4022156 : rtx midnotes = 0;
4321 4022156 : int from_luid;
4322 : /* Compute which registers we expect to eliminate. newi2pat may be setting
4323 : either i3dest or i2dest, so we must check it. */
4324 103911 : rtx elim_i2 = ((newi2pat && reg_set_p (i2dest, newi2pat))
4325 3929784 : || i2dest_in_i2src || i2dest_in_i1src || i2dest_in_i0src
4326 3854209 : || !i2dest_killed
4327 7875285 : ? 0 : i2dest);
4328 : /* For i1, we need to compute both local elimination and global
4329 : elimination information with respect to newi2pat because i1dest
4330 : may be the same as i3dest, in which case newi2pat may be setting
4331 : i1dest. Global information is used when distributing REG_DEAD
4332 : note for i2 and i3, in which case it does matter if newi2pat sets
4333 : i1dest or not.
4334 :
4335 : Local information is used when distributing REG_DEAD note for i1,
4336 : in which case it doesn't matter if newi2pat sets i1dest or not.
4337 : See PR62151, if we have four insns combination:
4338 : i0: r0 <- i0src
4339 : i1: r1 <- i1src (using r0)
4340 : REG_DEAD (r0)
4341 : i2: r0 <- i2src (using r1)
4342 : i3: r3 <- i3src (using r0)
4343 : ix: using r0
4344 : From i1's point of view, r0 is eliminated, no matter if it is set
4345 : by newi2pat or not. In other words, REG_DEAD info for r0 in i1
4346 : should be discarded.
4347 :
4348 : Note local information only affects cases in forms like "I1->I2->I3",
4349 : "I0->I1->I2->I3" or "I0&I1->I2, I2->I3". For other cases like
4350 : "I0->I1, I1&I2->I3" or "I1&I2->I3", newi2pat won't set i1dest or
4351 : i0dest anyway. */
4352 99960 : rtx local_elim_i1 = (i1 == 0 || i1dest_in_i1src || i1dest_in_i0src
4353 99890 : || !i1dest_killed
4354 4022156 : ? 0 : i1dest);
4355 99889 : rtx elim_i1 = (local_elim_i1 == 0
4356 99889 : || (newi2pat && reg_set_p (i1dest, newi2pat))
4357 99889 : ? 0 : i1dest);
4358 : /* Same case as i1. */
4359 5058 : rtx local_elim_i0 = (i0 == 0 || i0dest_in_i0src || !i0dest_killed
4360 4022156 : ? 0 : i0dest);
4361 5040 : rtx elim_i0 = (local_elim_i0 == 0
4362 5040 : || (newi2pat && reg_set_p (i0dest, newi2pat))
4363 5040 : ? 0 : i0dest);
4364 :
4365 : /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4366 : clear them. */
4367 4022156 : i3notes = REG_NOTES (i3), i3links = LOG_LINKS (i3);
4368 4022156 : i2notes = REG_NOTES (i2), i2links = LOG_LINKS (i2);
4369 4022156 : if (i1)
4370 99960 : i1notes = REG_NOTES (i1), i1links = LOG_LINKS (i1);
4371 4022156 : if (i0)
4372 5058 : i0notes = REG_NOTES (i0), i0links = LOG_LINKS (i0);
4373 :
4374 : /* Ensure that we do not have something that should not be shared but
4375 : occurs multiple times in the new insns. Check this by first
4376 : resetting all the `used' flags and then copying anything is shared. */
4377 :
4378 4022156 : reset_used_flags (i3notes);
4379 4022156 : reset_used_flags (i2notes);
4380 4022156 : reset_used_flags (i1notes);
4381 4022156 : reset_used_flags (i0notes);
4382 4022156 : reset_used_flags (newpat);
4383 4022156 : reset_used_flags (newi2pat);
4384 4022156 : if (undobuf.other_insn)
4385 194438 : reset_used_flags (PATTERN (undobuf.other_insn));
4386 :
4387 4022156 : i3notes = copy_rtx_if_shared (i3notes);
4388 4022156 : i2notes = copy_rtx_if_shared (i2notes);
4389 4022156 : i1notes = copy_rtx_if_shared (i1notes);
4390 4022156 : i0notes = copy_rtx_if_shared (i0notes);
4391 4022156 : newpat = copy_rtx_if_shared (newpat);
4392 4022156 : newi2pat = copy_rtx_if_shared (newi2pat);
4393 4022156 : if (undobuf.other_insn)
4394 194438 : reset_used_flags (PATTERN (undobuf.other_insn));
4395 :
4396 4022156 : INSN_CODE (i3) = insn_code_number;
4397 4022156 : PATTERN (i3) = newpat;
4398 :
4399 4022156 : if (CALL_P (i3) && CALL_INSN_FUNCTION_USAGE (i3))
4400 : {
4401 356134 : for (rtx link = CALL_INSN_FUNCTION_USAGE (i3); link;
4402 273922 : link = XEXP (link, 1))
4403 : {
4404 273922 : if (substed_i2)
4405 : {
4406 : /* I2SRC must still be meaningful at this point. Some
4407 : splitting operations can invalidate I2SRC, but those
4408 : operations do not apply to calls. */
4409 273922 : gcc_assert (i2src);
4410 273922 : XEXP (link, 0) = simplify_replace_rtx (XEXP (link, 0),
4411 : i2dest, i2src);
4412 : }
4413 273922 : if (substed_i1)
4414 0 : XEXP (link, 0) = simplify_replace_rtx (XEXP (link, 0),
4415 : i1dest, i1src);
4416 273922 : if (substed_i0)
4417 0 : XEXP (link, 0) = simplify_replace_rtx (XEXP (link, 0),
4418 : i0dest, i0src);
4419 : }
4420 : }
4421 :
4422 4022156 : if (undobuf.other_insn)
4423 194438 : INSN_CODE (undobuf.other_insn) = other_code_number;
4424 :
4425 : /* We had one special case above where I2 had more than one set and
4426 : we replaced a destination of one of those sets with the destination
4427 : of I3. In that case, we have to update LOG_LINKS of insns later
4428 : in this basic block. Note that this (expensive) case is rare.
4429 :
4430 : Also, in this case, we must pretend that all REG_NOTEs for I2
4431 : actually came from I3, so that REG_UNUSED notes from I2 will be
4432 : properly handled. */
4433 :
4434 4022156 : if (i3_subst_into_i2)
4435 : {
4436 206763 : for (i = 0; i < XVECLEN (PATTERN (i2), 0); i++)
4437 141897 : if ((GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == SET
4438 66067 : || GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == CLOBBER)
4439 141077 : && REG_P (SET_DEST (XVECEXP (PATTERN (i2), 0, i)))
4440 126708 : && SET_DEST (XVECEXP (PATTERN (i2), 0, i)) != i2dest
4441 268605 : && ! find_reg_note (i2, REG_UNUSED,
4442 126708 : SET_DEST (XVECEXP (PATTERN (i2), 0, i))))
4443 29433279 : for (temp_insn = NEXT_INSN (i2);
4444 : temp_insn
4445 29433279 : && (this_basic_block->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)
4446 29335994 : || BB_HEAD (this_basic_block) != temp_insn);
4447 29375625 : temp_insn = NEXT_INSN (temp_insn))
4448 29375625 : if (temp_insn != i3 && NONDEBUG_INSN_P (temp_insn))
4449 19075506 : FOR_EACH_LOG_LINK (link, temp_insn)
4450 6932826 : if (link->insn == i2)
4451 438 : link->insn = i3;
4452 :
4453 64866 : if (i3notes)
4454 : {
4455 : rtx link = i3notes;
4456 71821 : while (XEXP (link, 1))
4457 : link = XEXP (link, 1);
4458 64866 : XEXP (link, 1) = i2notes;
4459 : }
4460 : else
4461 : i3notes = i2notes;
4462 : i2notes = 0;
4463 : }
4464 :
4465 4022156 : LOG_LINKS (i3) = NULL;
4466 4022156 : REG_NOTES (i3) = 0;
4467 4022156 : LOG_LINKS (i2) = NULL;
4468 4022156 : REG_NOTES (i2) = 0;
4469 :
4470 4022156 : if (newi2pat)
4471 : {
4472 103911 : if (MAY_HAVE_DEBUG_BIND_INSNS && i2scratch)
4473 11350 : propagate_for_debug (i2, last_combined_insn, i2dest, i2src,
4474 : this_basic_block);
4475 103911 : INSN_CODE (i2) = i2_code_number;
4476 103911 : PATTERN (i2) = newi2pat;
4477 : }
4478 : else
4479 : {
4480 3918245 : if (MAY_HAVE_DEBUG_BIND_INSNS && i2src)
4481 2118869 : propagate_for_debug (i2, last_combined_insn, i2dest, i2src,
4482 : this_basic_block);
4483 3918245 : SET_INSN_DELETED (i2);
4484 : }
4485 :
4486 4022156 : if (i1)
4487 : {
4488 99960 : LOG_LINKS (i1) = NULL;
4489 99960 : REG_NOTES (i1) = 0;
4490 99960 : if (MAY_HAVE_DEBUG_BIND_INSNS)
4491 51998 : propagate_for_debug (i1, last_combined_insn, i1dest, i1src,
4492 : this_basic_block);
4493 99960 : SET_INSN_DELETED (i1);
4494 : }
4495 :
4496 4022156 : if (i0)
4497 : {
4498 5058 : LOG_LINKS (i0) = NULL;
4499 5058 : REG_NOTES (i0) = 0;
4500 5058 : if (MAY_HAVE_DEBUG_BIND_INSNS)
4501 3788 : propagate_for_debug (i0, last_combined_insn, i0dest, i0src,
4502 : this_basic_block);
4503 5058 : SET_INSN_DELETED (i0);
4504 : }
4505 :
4506 : /* Get death notes for everything that is now used in either I3 or
4507 : I2 and used to die in a previous insn. If we built two new
4508 : patterns, move from I1 to I2 then I2 to I3 so that we get the
4509 : proper movement on registers that I2 modifies. */
4510 :
4511 4022156 : if (i0)
4512 5058 : from_luid = DF_INSN_LUID (i0);
4513 4017098 : else if (i1)
4514 94902 : from_luid = DF_INSN_LUID (i1);
4515 : else
4516 3922196 : from_luid = DF_INSN_LUID (i2);
4517 4022156 : if (newi2pat)
4518 103911 : move_deaths (newi2pat, NULL_RTX, from_luid, i2, &midnotes);
4519 4022156 : move_deaths (newpat, newi2pat, from_luid, i3, &midnotes);
4520 :
4521 : /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4522 4022156 : if (i3notes)
4523 7334655 : distribute_notes (i3notes, i3, i3, newi2pat ? i2 : NULL,
4524 : elim_i2, elim_i1, elim_i0);
4525 4022156 : if (i2notes)
4526 5670424 : distribute_notes (i2notes, i2, i3, newi2pat ? i2 : NULL,
4527 : elim_i2, elim_i1, elim_i0);
4528 4022156 : if (i1notes)
4529 60052 : distribute_notes (i1notes, i1, i3, newi2pat ? i2 : NULL,
4530 : elim_i2, local_elim_i1, local_elim_i0);
4531 4022156 : if (i0notes)
4532 4413 : distribute_notes (i0notes, i0, i3, newi2pat ? i2 : NULL,
4533 : elim_i2, elim_i1, local_elim_i0);
4534 4022156 : if (midnotes)
4535 4825470 : distribute_notes (midnotes, NULL, i3, newi2pat ? i2 : NULL,
4536 : elim_i2, elim_i1, elim_i0);
4537 :
4538 : /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4539 : know these are REG_UNUSED and want them to go to the desired insn,
4540 : so we always pass it as i3. */
4541 :
4542 4022156 : if (newi2pat && new_i2_notes)
4543 39864 : distribute_notes (new_i2_notes, i2, i2, NULL, NULL_RTX, NULL_RTX,
4544 : NULL_RTX);
4545 :
4546 4022156 : if (new_i3_notes)
4547 145093 : distribute_notes (new_i3_notes, i3, i3, NULL, NULL_RTX, NULL_RTX,
4548 : NULL_RTX);
4549 :
4550 : /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4551 : put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4552 : I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4553 : in that case, it might delete I2. Similarly for I2 and I1.
4554 : Show an additional death due to the REG_DEAD note we make here. If
4555 : we discard it in distribute_notes, we will decrement it again. */
4556 :
4557 4022156 : if (i3dest_killed)
4558 : {
4559 296287 : rtx new_note = alloc_reg_note (REG_DEAD, i3dest_killed, NULL_RTX);
4560 296287 : if (newi2pat && reg_set_p (i3dest_killed, newi2pat))
4561 1032 : distribute_notes (new_note, NULL, i2, NULL, elim_i2,
4562 : elim_i1, elim_i0);
4563 : else
4564 589114 : distribute_notes (new_note, NULL, i3, newi2pat ? i2 : NULL,
4565 : elim_i2, elim_i1, elim_i0);
4566 : }
4567 :
4568 4022156 : if (i2dest_in_i2src)
4569 : {
4570 74784 : rtx new_note = alloc_reg_note (REG_DEAD, i2dest, NULL_RTX);
4571 74784 : if (newi2pat && reg_set_p (i2dest, newi2pat))
4572 721 : distribute_notes (new_note, NULL, i2, NULL, NULL_RTX,
4573 : NULL_RTX, NULL_RTX);
4574 : else
4575 148088 : distribute_notes (new_note, NULL, i3, newi2pat ? i2 : NULL,
4576 : NULL_RTX, NULL_RTX, NULL_RTX);
4577 : }
4578 :
4579 4022156 : if (i1dest_in_i1src)
4580 : {
4581 68 : rtx new_note = alloc_reg_note (REG_DEAD, i1dest, NULL_RTX);
4582 68 : if (newi2pat && reg_set_p (i1dest, newi2pat))
4583 4 : distribute_notes (new_note, NULL, i2, NULL, NULL_RTX,
4584 : NULL_RTX, NULL_RTX);
4585 : else
4586 110 : distribute_notes (new_note, NULL, i3, newi2pat ? i2 : NULL,
4587 : NULL_RTX, NULL_RTX, NULL_RTX);
4588 : }
4589 :
4590 4022156 : if (i0dest_in_i0src)
4591 : {
4592 18 : rtx new_note = alloc_reg_note (REG_DEAD, i0dest, NULL_RTX);
4593 18 : if (newi2pat && reg_set_p (i0dest, newi2pat))
4594 0 : distribute_notes (new_note, NULL, i2, NULL, NULL_RTX,
4595 : NULL_RTX, NULL_RTX);
4596 : else
4597 36 : distribute_notes (new_note, NULL, i3, newi2pat ? i2 : NULL,
4598 : NULL_RTX, NULL_RTX, NULL_RTX);
4599 : }
4600 :
4601 4022156 : if (only_i3_changed)
4602 30987 : distribute_links (i3links, i3, param_max_combine_search_insns);
4603 : else
4604 : {
4605 3991169 : distribute_links (i3links);
4606 3991169 : distribute_links (i2links, i2);
4607 3991169 : distribute_links (i1links);
4608 3991169 : distribute_links (i0links);
4609 : }
4610 :
4611 4022156 : if (REG_P (i2dest))
4612 : {
4613 4022156 : struct insn_link *link;
4614 4022156 : rtx_insn *i2_insn = 0;
4615 4022156 : rtx i2_val = 0, set;
4616 :
4617 : /* The insn that used to set this register doesn't exist, and
4618 : this life of the register may not exist either. See if one of
4619 : I3's links points to an insn that sets I2DEST. If it does,
4620 : that is now the last known value for I2DEST. If we don't update
4621 : this and I2 set the register to a value that depended on its old
4622 : contents, we will get confused. If this insn is used, thing
4623 : will be set correctly in combine_instructions. */
4624 7358237 : FOR_EACH_LOG_LINK (link, i3)
4625 3336081 : if ((set = single_set (link->insn)) != 0
4626 3336081 : && rtx_equal_p (i2dest, SET_DEST (set)))
4627 45450 : i2_insn = link->insn, i2_val = SET_SRC (set);
4628 :
4629 4022156 : record_value_for_reg (i2dest, i2_insn, i2_val);
4630 :
4631 : /* If the reg formerly set in I2 died only once and that was in I3,
4632 : zero its use count so it won't make `reload' do any work. */
4633 4022156 : if (! added_sets_2
4634 3894142 : && (newi2pat == 0 || ! reg_mentioned_p (i2dest, newi2pat))
4635 3855886 : && ! i2dest_in_i2src
4636 7822974 : && REGNO (i2dest) < reg_n_sets_max)
4637 3800816 : INC_REG_N_SETS (REGNO (i2dest), -1);
4638 : }
4639 :
4640 4022156 : if (i1 && REG_P (i1dest))
4641 : {
4642 99960 : struct insn_link *link;
4643 99960 : rtx_insn *i1_insn = 0;
4644 99960 : rtx i1_val = 0, set;
4645 :
4646 174645 : FOR_EACH_LOG_LINK (link, i3)
4647 74685 : if ((set = single_set (link->insn)) != 0
4648 74685 : && rtx_equal_p (i1dest, SET_DEST (set)))
4649 441 : i1_insn = link->insn, i1_val = SET_SRC (set);
4650 :
4651 99960 : record_value_for_reg (i1dest, i1_insn, i1_val);
4652 :
4653 99960 : if (! added_sets_1
4654 : && ! i1dest_in_i1src
4655 99960 : && REGNO (i1dest) < reg_n_sets_max)
4656 94215 : INC_REG_N_SETS (REGNO (i1dest), -1);
4657 : }
4658 :
4659 4022156 : if (i0 && REG_P (i0dest))
4660 : {
4661 5058 : struct insn_link *link;
4662 5058 : rtx_insn *i0_insn = 0;
4663 5058 : rtx i0_val = 0, set;
4664 :
4665 7127 : FOR_EACH_LOG_LINK (link, i3)
4666 2069 : if ((set = single_set (link->insn)) != 0
4667 2069 : && rtx_equal_p (i0dest, SET_DEST (set)))
4668 0 : i0_insn = link->insn, i0_val = SET_SRC (set);
4669 :
4670 5058 : record_value_for_reg (i0dest, i0_insn, i0_val);
4671 :
4672 5058 : if (! added_sets_0
4673 : && ! i0dest_in_i0src
4674 5058 : && REGNO (i0dest) < reg_n_sets_max)
4675 4999 : INC_REG_N_SETS (REGNO (i0dest), -1);
4676 : }
4677 :
4678 : /* Update reg_stat[].nonzero_bits et al for any changes that may have
4679 : been made to this insn. The order is important, because newi2pat
4680 : can affect nonzero_bits of newpat. */
4681 4022156 : if (newi2pat)
4682 103911 : note_pattern_stores (newi2pat, set_nonzero_bits_and_sign_copies, NULL);
4683 4022156 : note_pattern_stores (newpat, set_nonzero_bits_and_sign_copies, NULL);
4684 : }
4685 :
4686 4022156 : if (undobuf.other_insn != NULL_RTX)
4687 : {
4688 194438 : if (dump_file)
4689 : {
4690 12 : fprintf (dump_file, "modifying other_insn ");
4691 12 : dump_insn_slim (dump_file, undobuf.other_insn);
4692 : }
4693 194438 : df_insn_rescan (undobuf.other_insn);
4694 : }
4695 :
4696 4022156 : if (i0 && !(NOTE_P (i0) && (NOTE_KIND (i0) == NOTE_INSN_DELETED)))
4697 : {
4698 0 : if (dump_file)
4699 : {
4700 0 : fprintf (dump_file, "modifying insn i0 ");
4701 0 : dump_insn_slim (dump_file, i0);
4702 : }
4703 0 : df_insn_rescan (i0);
4704 : }
4705 :
4706 4022156 : if (i1 && !(NOTE_P (i1) && (NOTE_KIND (i1) == NOTE_INSN_DELETED)))
4707 : {
4708 0 : if (dump_file)
4709 : {
4710 0 : fprintf (dump_file, "modifying insn i1 ");
4711 0 : dump_insn_slim (dump_file, i1);
4712 : }
4713 0 : df_insn_rescan (i1);
4714 : }
4715 :
4716 4022156 : if (i2 && !(NOTE_P (i2) && (NOTE_KIND (i2) == NOTE_INSN_DELETED)))
4717 : {
4718 103911 : if (dump_file)
4719 : {
4720 15 : fprintf (dump_file, "modifying insn i2 ");
4721 15 : dump_insn_slim (dump_file, i2);
4722 : }
4723 103911 : df_insn_rescan (i2);
4724 : }
4725 :
4726 4022156 : if (i3 && !(NOTE_P (i3) && (NOTE_KIND (i3) == NOTE_INSN_DELETED)))
4727 : {
4728 4022156 : if (dump_file)
4729 : {
4730 240 : fprintf (dump_file, "modifying insn i3 ");
4731 240 : dump_insn_slim (dump_file, i3);
4732 : }
4733 4022156 : df_insn_rescan (i3);
4734 : }
4735 :
4736 : /* Set new_direct_jump_p if a new return or simple jump instruction
4737 : has been created. Adjust the CFG accordingly. */
4738 4022156 : if (returnjump_p (i3) || any_uncondjump_p (i3))
4739 : {
4740 162 : *new_direct_jump_p = 1;
4741 162 : mark_jump_label (PATTERN (i3), i3, 0);
4742 162 : update_cfg_for_uncondjump (i3);
4743 : }
4744 :
4745 4022156 : if (undobuf.other_insn != NULL_RTX
4746 4022156 : && (returnjump_p (undobuf.other_insn)
4747 194438 : || any_uncondjump_p (undobuf.other_insn)))
4748 : {
4749 1903 : *new_direct_jump_p = 1;
4750 1903 : update_cfg_for_uncondjump (undobuf.other_insn);
4751 : }
4752 :
4753 4022156 : if (GET_CODE (PATTERN (i3)) == TRAP_IF
4754 4022156 : && XEXP (PATTERN (i3), 0) == const1_rtx)
4755 : {
4756 0 : basic_block bb = BLOCK_FOR_INSN (i3);
4757 0 : gcc_assert (bb);
4758 0 : remove_edge (split_block (bb, i3));
4759 0 : emit_barrier_after_bb (bb);
4760 0 : *new_direct_jump_p = 1;
4761 : }
4762 :
4763 4022156 : if (undobuf.other_insn
4764 194438 : && GET_CODE (PATTERN (undobuf.other_insn)) == TRAP_IF
4765 4022156 : && XEXP (PATTERN (undobuf.other_insn), 0) == const1_rtx)
4766 : {
4767 0 : basic_block bb = BLOCK_FOR_INSN (undobuf.other_insn);
4768 0 : gcc_assert (bb);
4769 0 : remove_edge (split_block (bb, undobuf.other_insn));
4770 0 : emit_barrier_after_bb (bb);
4771 0 : *new_direct_jump_p = 1;
4772 : }
4773 :
4774 : /* A noop might also need cleaning up of CFG, if it comes from the
4775 : simplification of a jump. */
4776 4022156 : if (JUMP_P (i3)
4777 46789 : && GET_CODE (newpat) == SET
4778 34133 : && SET_SRC (newpat) == pc_rtx
4779 404 : && SET_DEST (newpat) == pc_rtx)
4780 : {
4781 404 : *new_direct_jump_p = 1;
4782 404 : update_cfg_for_uncondjump (i3);
4783 : }
4784 :
4785 4022156 : if (undobuf.other_insn != NULL_RTX
4786 194438 : && JUMP_P (undobuf.other_insn)
4787 188488 : && GET_CODE (PATTERN (undobuf.other_insn)) == SET
4788 188488 : && SET_SRC (PATTERN (undobuf.other_insn)) == pc_rtx
4789 4024160 : && SET_DEST (PATTERN (undobuf.other_insn)) == pc_rtx)
4790 : {
4791 2004 : *new_direct_jump_p = 1;
4792 2004 : update_cfg_for_uncondjump (undobuf.other_insn);
4793 : }
4794 :
4795 4022156 : combine_successes++;
4796 4022156 : undo_commit ();
4797 :
4798 4022156 : if (only_i3_changed)
4799 : return i3;
4800 :
4801 3991169 : rtx_insn *ret = newi2pat ? i2 : i3;
4802 3991169 : if (added_links_insn && DF_INSN_LUID (added_links_insn) < DF_INSN_LUID (ret))
4803 : ret = added_links_insn;
4804 3991169 : if (added_notes_insn && DF_INSN_LUID (added_notes_insn) < DF_INSN_LUID (ret))
4805 : ret = added_notes_insn;
4806 :
4807 : return ret;
4808 : }
4809 :
4810 : /* Get a marker for undoing to the current state. */
4811 :
4812 : static void *
4813 37205189 : get_undo_marker (void)
4814 : {
4815 37205189 : return undobuf.undos;
4816 : }
4817 :
4818 : /* Undo the modifications up to the marker. */
4819 :
4820 : static void
4821 43784948 : undo_to_marker (void *marker)
4822 : {
4823 43784948 : struct undo *undo, *next;
4824 :
4825 138229099 : for (undo = undobuf.undos; undo != marker; undo = next)
4826 : {
4827 94444151 : gcc_assert (undo);
4828 :
4829 94444151 : next = undo->next;
4830 94444151 : switch (undo->kind)
4831 : {
4832 87464059 : case UNDO_RTX:
4833 87464059 : *undo->where.r = undo->old_contents.r;
4834 87464059 : break;
4835 6339947 : case UNDO_INT:
4836 6339947 : *undo->where.i = undo->old_contents.i;
4837 6339947 : break;
4838 569705 : case UNDO_MODE:
4839 569705 : adjust_reg_mode (regno_reg_rtx[undo->where.regno],
4840 : undo->old_contents.m);
4841 569705 : break;
4842 70440 : case UNDO_LINKS:
4843 70440 : *undo->where.l = undo->old_contents.l;
4844 70440 : break;
4845 0 : default:
4846 0 : gcc_unreachable ();
4847 : }
4848 :
4849 94444151 : undo->next = undobuf.frees;
4850 94444151 : undobuf.frees = undo;
4851 : }
4852 :
4853 43784948 : undobuf.undos = (struct undo *) marker;
4854 43784948 : }
4855 :
4856 : /* Undo all the modifications recorded in undobuf. */
4857 :
4858 : static void
4859 42692993 : undo_all (void)
4860 : {
4861 42692993 : undo_to_marker (0);
4862 0 : }
4863 :
4864 : /* We've committed to accepting the changes we made. Move all
4865 : of the undos to the free list. */
4866 :
4867 : static void
4868 4022156 : undo_commit (void)
4869 : {
4870 4022156 : struct undo *undo, *next;
4871 :
4872 11881229 : for (undo = undobuf.undos; undo; undo = next)
4873 : {
4874 7859073 : next = undo->next;
4875 7859073 : undo->next = undobuf.frees;
4876 7859073 : undobuf.frees = undo;
4877 : }
4878 4022156 : undobuf.undos = 0;
4879 4022156 : }
4880 :
4881 : /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4882 : where we have an arithmetic expression and return that point. LOC will
4883 : be inside INSN.
4884 :
4885 : try_combine will call this function to see if an insn can be split into
4886 : two insns. */
4887 :
4888 : static rtx *
4889 31008341 : find_split_point (rtx *loc, rtx_insn *insn, bool set_src)
4890 : {
4891 31008341 : rtx x = *loc;
4892 31008341 : enum rtx_code code = GET_CODE (x);
4893 31008341 : rtx *split;
4894 31008341 : unsigned HOST_WIDE_INT len = 0;
4895 31008341 : HOST_WIDE_INT pos = 0;
4896 31008341 : bool unsignedp = false;
4897 31008341 : rtx inner = NULL_RTX;
4898 31008341 : scalar_int_mode mode, inner_mode;
4899 :
4900 : /* First special-case some codes. */
4901 31008341 : switch (code)
4902 : {
4903 976330 : case SUBREG:
4904 : #ifdef INSN_SCHEDULING
4905 : /* If we are making a paradoxical SUBREG invalid, it becomes a split
4906 : point. */
4907 976330 : if (MEM_P (SUBREG_REG (x)))
4908 : return loc;
4909 : #endif
4910 960866 : return find_split_point (&SUBREG_REG (x), insn, false);
4911 :
4912 1516234 : case MEM:
4913 : /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4914 : using LO_SUM and HIGH. */
4915 1516234 : if (HAVE_lo_sum && (GET_CODE (XEXP (x, 0)) == CONST
4916 : || GET_CODE (XEXP (x, 0)) == SYMBOL_REF))
4917 : {
4918 : machine_mode address_mode = get_address_mode (x);
4919 :
4920 : SUBST (XEXP (x, 0),
4921 : gen_rtx_LO_SUM (address_mode,
4922 : gen_rtx_HIGH (address_mode, XEXP (x, 0)),
4923 : XEXP (x, 0)));
4924 : return &XEXP (XEXP (x, 0), 0);
4925 : }
4926 :
4927 : /* If we have a PLUS whose second operand is a constant and the
4928 : address is not valid, perhaps we can split it up using
4929 : the machine-specific way to split large constants. We use
4930 : the first pseudo-reg (one of the virtual regs) as a placeholder;
4931 : it will not remain in the result. */
4932 1516234 : if (GET_CODE (XEXP (x, 0)) == PLUS
4933 1000578 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
4934 3227878 : && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
4935 711066 : MEM_ADDR_SPACE (x)))
4936 : {
4937 128064 : rtx reg = regno_reg_rtx[FIRST_PSEUDO_REGISTER];
4938 128064 : unsigned int old_nregs, new_nregs;
4939 128064 : rtx_insn *seq = combine_split_insns (gen_rtx_SET (reg, XEXP (x, 0)),
4940 : subst_insn, &old_nregs, &new_nregs);
4941 :
4942 : /* This should have produced two insns, each of which sets our
4943 : placeholder. If the source of the second is a valid address,
4944 : we can put both sources together and make a split point
4945 : in the middle. */
4946 :
4947 128064 : if (seq
4948 54 : && NEXT_INSN (seq) != NULL_RTX
4949 0 : && NEXT_INSN (NEXT_INSN (seq)) == NULL_RTX
4950 0 : && NONJUMP_INSN_P (seq)
4951 0 : && GET_CODE (PATTERN (seq)) == SET
4952 0 : && SET_DEST (PATTERN (seq)) == reg
4953 0 : && ! reg_mentioned_p (reg,
4954 0 : SET_SRC (PATTERN (seq)))
4955 0 : && NONJUMP_INSN_P (NEXT_INSN (seq))
4956 0 : && GET_CODE (PATTERN (NEXT_INSN (seq))) == SET
4957 0 : && SET_DEST (PATTERN (NEXT_INSN (seq))) == reg
4958 128064 : && memory_address_addr_space_p
4959 128064 : (GET_MODE (x), SET_SRC (PATTERN (NEXT_INSN (seq))),
4960 0 : MEM_ADDR_SPACE (x)))
4961 : {
4962 0 : rtx src1 = SET_SRC (PATTERN (seq));
4963 0 : rtx src2 = SET_SRC (PATTERN (NEXT_INSN (seq)));
4964 :
4965 : /* Replace the placeholder in SRC2 with SRC1. If we can
4966 : find where in SRC2 it was placed, that can become our
4967 : split point and we can replace this address with SRC2.
4968 : Just try two obvious places. */
4969 :
4970 0 : src2 = replace_rtx (src2, reg, src1);
4971 0 : split = 0;
4972 0 : if (XEXP (src2, 0) == src1)
4973 0 : split = &XEXP (src2, 0);
4974 0 : else if (GET_RTX_FORMAT (GET_CODE (XEXP (src2, 0)))[0] == 'e'
4975 0 : && XEXP (XEXP (src2, 0), 0) == src1)
4976 0 : split = &XEXP (XEXP (src2, 0), 0);
4977 :
4978 0 : if (split)
4979 : {
4980 0 : SUBST (XEXP (x, 0), src2);
4981 104478 : return split;
4982 : }
4983 : }
4984 :
4985 : /* If that didn't work and we have a nested plus, like:
4986 : ((REG1 * CONST1) + REG2) + CONST2 and (REG1 + REG2) + CONST2
4987 : is valid address, try to split (REG1 * CONST1). */
4988 128064 : if (GET_CODE (XEXP (XEXP (x, 0), 0)) == PLUS
4989 94241 : && !OBJECT_P (XEXP (XEXP (XEXP (x, 0), 0), 0))
4990 82121 : && OBJECT_P (XEXP (XEXP (XEXP (x, 0), 0), 1))
4991 82116 : && ! (GET_CODE (XEXP (XEXP (XEXP (x, 0), 0), 0)) == SUBREG
4992 10 : && OBJECT_P (SUBREG_REG (XEXP (XEXP (XEXP (x, 0),
4993 : 0), 0)))))
4994 : {
4995 82116 : rtx tem = XEXP (XEXP (XEXP (x, 0), 0), 0);
4996 82116 : XEXP (XEXP (XEXP (x, 0), 0), 0) = reg;
4997 164232 : if (memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
4998 82116 : MEM_ADDR_SPACE (x)))
4999 : {
5000 71130 : XEXP (XEXP (XEXP (x, 0), 0), 0) = tem;
5001 71130 : return &XEXP (XEXP (XEXP (x, 0), 0), 0);
5002 : }
5003 10986 : XEXP (XEXP (XEXP (x, 0), 0), 0) = tem;
5004 10986 : }
5005 45948 : else if (GET_CODE (XEXP (XEXP (x, 0), 0)) == PLUS
5006 12125 : && OBJECT_P (XEXP (XEXP (XEXP (x, 0), 0), 0))
5007 12120 : && !OBJECT_P (XEXP (XEXP (XEXP (x, 0), 0), 1))
5008 384 : && ! (GET_CODE (XEXP (XEXP (XEXP (x, 0), 0), 1)) == SUBREG
5009 384 : && OBJECT_P (SUBREG_REG (XEXP (XEXP (XEXP (x, 0),
5010 : 0), 1)))))
5011 : {
5012 0 : rtx tem = XEXP (XEXP (XEXP (x, 0), 0), 1);
5013 0 : XEXP (XEXP (XEXP (x, 0), 0), 1) = reg;
5014 0 : if (memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
5015 0 : MEM_ADDR_SPACE (x)))
5016 : {
5017 0 : XEXP (XEXP (XEXP (x, 0), 0), 1) = tem;
5018 0 : return &XEXP (XEXP (XEXP (x, 0), 0), 1);
5019 : }
5020 0 : XEXP (XEXP (XEXP (x, 0), 0), 1) = tem;
5021 : }
5022 :
5023 : /* If that didn't work, perhaps the first operand is complex and
5024 : needs to be computed separately, so make a split point there.
5025 : This will occur on machines that just support REG + CONST
5026 : and have a constant moved through some previous computation. */
5027 56934 : if (!OBJECT_P (XEXP (XEXP (x, 0), 0))
5028 33348 : && ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
5029 0 : && OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
5030 33348 : return &XEXP (XEXP (x, 0), 0);
5031 : }
5032 :
5033 : /* If we have a PLUS whose first operand is complex, try computing it
5034 : separately by making a split there. */
5035 1411756 : if (GET_CODE (XEXP (x, 0)) == PLUS
5036 2476724 : && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
5037 896100 : MEM_ADDR_SPACE (x))
5038 168868 : && ! OBJECT_P (XEXP (XEXP (x, 0), 0))
5039 1531903 : && ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
5040 1058 : && OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
5041 120143 : return &XEXP (XEXP (x, 0), 0);
5042 : break;
5043 :
5044 4612214 : case SET:
5045 : /* See if we can split SET_SRC as it stands. */
5046 4612214 : split = find_split_point (&SET_SRC (x), insn, true);
5047 4612214 : if (split && split != &SET_SRC (x))
5048 : return split;
5049 :
5050 : /* See if we can split SET_DEST as it stands. */
5051 503662 : split = find_split_point (&SET_DEST (x), insn, false);
5052 503662 : if (split && split != &SET_DEST (x))
5053 : return split;
5054 :
5055 : /* See if this is a bitfield assignment with everything constant. If
5056 : so, this is an IOR of an AND, so split it into that. */
5057 469636 : if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
5058 4881 : && is_a <scalar_int_mode> (GET_MODE (XEXP (SET_DEST (x), 0)),
5059 : &inner_mode)
5060 4881 : && HWI_COMPUTABLE_MODE_P (inner_mode)
5061 4881 : && CONST_INT_P (XEXP (SET_DEST (x), 1))
5062 4881 : && CONST_INT_P (XEXP (SET_DEST (x), 2))
5063 4706 : && CONST_INT_P (SET_SRC (x))
5064 421 : && ((INTVAL (XEXP (SET_DEST (x), 1))
5065 421 : + INTVAL (XEXP (SET_DEST (x), 2)))
5066 421 : <= GET_MODE_PRECISION (inner_mode))
5067 470057 : && ! side_effects_p (XEXP (SET_DEST (x), 0)))
5068 : {
5069 404 : HOST_WIDE_INT pos = INTVAL (XEXP (SET_DEST (x), 2));
5070 404 : unsigned HOST_WIDE_INT len = INTVAL (XEXP (SET_DEST (x), 1));
5071 404 : rtx dest = XEXP (SET_DEST (x), 0);
5072 404 : unsigned HOST_WIDE_INT mask = (HOST_WIDE_INT_1U << len) - 1;
5073 404 : unsigned HOST_WIDE_INT src = INTVAL (SET_SRC (x)) & mask;
5074 404 : rtx or_mask;
5075 :
5076 404 : if (BITS_BIG_ENDIAN)
5077 : pos = GET_MODE_PRECISION (inner_mode) - len - pos;
5078 :
5079 404 : or_mask = gen_int_mode (src << pos, inner_mode);
5080 404 : if (src == mask)
5081 0 : SUBST (SET_SRC (x),
5082 : simplify_gen_binary (IOR, inner_mode, dest, or_mask));
5083 : else
5084 : {
5085 404 : rtx negmask = gen_int_mode (~(mask << pos), inner_mode);
5086 404 : SUBST (SET_SRC (x),
5087 : simplify_gen_binary (IOR, inner_mode,
5088 : simplify_gen_binary (AND, inner_mode,
5089 : dest, negmask),
5090 : or_mask));
5091 : }
5092 :
5093 404 : SUBST (SET_DEST (x), dest);
5094 :
5095 404 : split = find_split_point (&SET_SRC (x), insn, true);
5096 404 : if (split && split != &SET_SRC (x))
5097 : return split;
5098 : }
5099 :
5100 : /* Otherwise, see if this is an operation that we can split into two.
5101 : If so, try to split that. */
5102 469232 : code = GET_CODE (SET_SRC (x));
5103 :
5104 469232 : switch (code)
5105 : {
5106 16101 : case AND:
5107 : /* If we are AND'ing with a large constant that is only a single
5108 : bit and the result is only being used in a context where we
5109 : need to know if it is zero or nonzero, replace it with a bit
5110 : extraction. This will avoid the large constant, which might
5111 : have taken more than one insn to make. If the constant were
5112 : not a valid argument to the AND but took only one insn to make,
5113 : this is no worse, but if it took more than one insn, it will
5114 : be better. */
5115 :
5116 16101 : if (CONST_INT_P (XEXP (SET_SRC (x), 1))
5117 10823 : && REG_P (XEXP (SET_SRC (x), 0))
5118 436 : && (pos = exact_log2 (UINTVAL (XEXP (SET_SRC (x), 1)))) >= 7
5119 1 : && REG_P (SET_DEST (x))
5120 0 : && (split = find_single_use (SET_DEST (x), insn, NULL)) != 0
5121 0 : && (GET_CODE (*split) == EQ || GET_CODE (*split) == NE)
5122 0 : && XEXP (*split, 0) == SET_DEST (x)
5123 16101 : && XEXP (*split, 1) == const0_rtx)
5124 : {
5125 0 : rtx extraction = make_extraction (GET_MODE (SET_DEST (x)),
5126 0 : XEXP (SET_SRC (x), 0),
5127 : pos, NULL_RTX, 1,
5128 : true, false, false);
5129 0 : if (extraction != 0)
5130 : {
5131 0 : SUBST (SET_SRC (x), extraction);
5132 0 : return find_split_point (loc, insn, false);
5133 : }
5134 : }
5135 : break;
5136 :
5137 : case NE:
5138 : /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
5139 : is known to be on, this can be converted into a NEG of a shift. */
5140 : if (STORE_FLAG_VALUE == -1 && XEXP (SET_SRC (x), 1) == const0_rtx
5141 : && GET_MODE (SET_SRC (x)) == GET_MODE (XEXP (SET_SRC (x), 0))
5142 : && ((pos = exact_log2 (nonzero_bits (XEXP (SET_SRC (x), 0),
5143 : GET_MODE (XEXP (SET_SRC (x),
5144 : 0))))) >= 1))
5145 : {
5146 : machine_mode mode = GET_MODE (XEXP (SET_SRC (x), 0));
5147 : rtx pos_rtx = gen_int_shift_amount (mode, pos);
5148 : SUBST (SET_SRC (x),
5149 : gen_rtx_NEG (mode,
5150 : gen_rtx_LSHIFTRT (mode,
5151 : XEXP (SET_SRC (x), 0),
5152 : pos_rtx)));
5153 :
5154 : split = find_split_point (&SET_SRC (x), insn, true);
5155 : if (split && split != &SET_SRC (x))
5156 : return split;
5157 : }
5158 : break;
5159 :
5160 734 : case SIGN_EXTEND:
5161 734 : inner = XEXP (SET_SRC (x), 0);
5162 :
5163 : /* We can't optimize if either mode is a partial integer
5164 : mode as we don't know how many bits are significant
5165 : in those modes. */
5166 734 : if (!is_int_mode (GET_MODE (inner), &inner_mode)
5167 728 : || GET_MODE_CLASS (GET_MODE (SET_SRC (x))) == MODE_PARTIAL_INT)
5168 : break;
5169 :
5170 728 : pos = 0;
5171 728 : len = GET_MODE_PRECISION (inner_mode);
5172 728 : unsignedp = false;
5173 728 : break;
5174 :
5175 12887 : case SIGN_EXTRACT:
5176 12887 : case ZERO_EXTRACT:
5177 12887 : if (is_a <scalar_int_mode> (GET_MODE (XEXP (SET_SRC (x), 0)),
5178 : &inner_mode)
5179 12605 : && CONST_INT_P (XEXP (SET_SRC (x), 1))
5180 12605 : && CONST_INT_P (XEXP (SET_SRC (x), 2)))
5181 : {
5182 12072 : inner = XEXP (SET_SRC (x), 0);
5183 12072 : len = INTVAL (XEXP (SET_SRC (x), 1));
5184 12072 : pos = INTVAL (XEXP (SET_SRC (x), 2));
5185 :
5186 12072 : if (BITS_BIG_ENDIAN)
5187 : pos = GET_MODE_PRECISION (inner_mode) - len - pos;
5188 12072 : unsignedp = (code == ZERO_EXTRACT);
5189 : }
5190 : break;
5191 :
5192 : default:
5193 : break;
5194 : }
5195 :
5196 469232 : if (len
5197 12800 : && known_subrange_p (pos, len,
5198 12800 : 0, GET_MODE_PRECISION (GET_MODE (inner)))
5199 482032 : && is_a <scalar_int_mode> (GET_MODE (SET_SRC (x)), &mode))
5200 : {
5201 : /* For unsigned, we have a choice of a shift followed by an
5202 : AND or two shifts. Use two shifts for field sizes where the
5203 : constant might be too large. We assume here that we can
5204 : always at least get 8-bit constants in an AND insn, which is
5205 : true for every current RISC. */
5206 :
5207 12800 : if (unsignedp && len <= 8)
5208 : {
5209 5949 : unsigned HOST_WIDE_INT mask
5210 5949 : = (HOST_WIDE_INT_1U << len) - 1;
5211 5949 : rtx pos_rtx = gen_int_shift_amount (mode, pos);
5212 5949 : SUBST (SET_SRC (x),
5213 : gen_rtx_AND (mode,
5214 : gen_rtx_LSHIFTRT
5215 : (mode, gen_lowpart (mode, inner), pos_rtx),
5216 : gen_int_mode (mask, mode)));
5217 :
5218 5949 : split = find_split_point (&SET_SRC (x), insn, true);
5219 5949 : if (split && split != &SET_SRC (x))
5220 31008341 : return split;
5221 : }
5222 : else
5223 : {
5224 6851 : int left_bits = GET_MODE_PRECISION (mode) - len - pos;
5225 6851 : int right_bits = GET_MODE_PRECISION (mode) - len;
5226 13702 : SUBST (SET_SRC (x),
5227 : gen_rtx_fmt_ee
5228 : (unsignedp ? LSHIFTRT : ASHIFTRT, mode,
5229 : gen_rtx_ASHIFT (mode,
5230 : gen_lowpart (mode, inner),
5231 : gen_int_shift_amount (mode, left_bits)),
5232 : gen_int_shift_amount (mode, right_bits)));
5233 :
5234 6851 : split = find_split_point (&SET_SRC (x), insn, true);
5235 6851 : if (split && split != &SET_SRC (x))
5236 31008341 : return split;
5237 : }
5238 : }
5239 :
5240 : /* See if this is a simple operation with a constant as the second
5241 : operand. It might be that this constant is out of range and hence
5242 : could be used as a split point. */
5243 456432 : if (BINARY_P (SET_SRC (x))
5244 204475 : && CONSTANT_P (XEXP (SET_SRC (x), 1))
5245 110452 : && (OBJECT_P (XEXP (SET_SRC (x), 0))
5246 33987 : || (GET_CODE (XEXP (SET_SRC (x), 0)) == SUBREG
5247 11377 : && OBJECT_P (SUBREG_REG (XEXP (SET_SRC (x), 0))))))
5248 78421 : return &XEXP (SET_SRC (x), 1);
5249 :
5250 : /* Finally, see if this is a simple operation with its first operand
5251 : not in a register. The operation might require this operand in a
5252 : register, so return it as a split point. We can always do this
5253 : because if the first operand were another operation, we would have
5254 : already found it as a split point. */
5255 378011 : if ((BINARY_P (SET_SRC (x)) || UNARY_P (SET_SRC (x)))
5256 378011 : && ! register_operand (XEXP (SET_SRC (x), 0), VOIDmode))
5257 120292 : return &XEXP (SET_SRC (x), 0);
5258 :
5259 : return 0;
5260 :
5261 1198909 : case AND:
5262 1198909 : case IOR:
5263 : /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
5264 : it is better to write this as (not (ior A B)) so we can split it.
5265 : Similarly for IOR. */
5266 1198909 : if (GET_CODE (XEXP (x, 0)) == NOT && GET_CODE (XEXP (x, 1)) == NOT)
5267 : {
5268 1178 : SUBST (*loc,
5269 : gen_rtx_NOT (GET_MODE (x),
5270 : gen_rtx_fmt_ee (code == IOR ? AND : IOR,
5271 : GET_MODE (x),
5272 : XEXP (XEXP (x, 0), 0),
5273 : XEXP (XEXP (x, 1), 0))));
5274 589 : return find_split_point (loc, insn, set_src);
5275 : }
5276 :
5277 : /* Many RISC machines have a large set of logical insns. If the
5278 : second operand is a NOT, put it first so we will try to split the
5279 : other operand first. */
5280 1198320 : if (GET_CODE (XEXP (x, 1)) == NOT)
5281 : {
5282 5113 : rtx tem = XEXP (x, 0);
5283 5113 : SUBST (XEXP (x, 0), XEXP (x, 1));
5284 5113 : SUBST (XEXP (x, 1), tem);
5285 : }
5286 : /* Many targets have a `(and (not X) Y)` and/or `(ior (not X) Y)` instructions.
5287 : Split at that insns. However if this is
5288 : the SET_SRC, we likely do not have such an instruction and it's
5289 : worthless to try this split. */
5290 1198320 : if (!set_src && GET_CODE (XEXP (x, 0)) == NOT)
5291 : return loc;
5292 : break;
5293 :
5294 3125732 : case PLUS:
5295 3125732 : case MINUS:
5296 : /* Canonicalization can produce (minus A (mult B C)), where C is a
5297 : constant. It may be better to try splitting (plus (mult B -C) A)
5298 : instead if this isn't a multiply by a power of two. */
5299 190695 : if (set_src && code == MINUS && GET_CODE (XEXP (x, 1)) == MULT
5300 20956 : && GET_CODE (XEXP (XEXP (x, 1), 1)) == CONST_INT
5301 3131163 : && !pow2p_hwi (INTVAL (XEXP (XEXP (x, 1), 1))))
5302 : {
5303 5431 : machine_mode mode = GET_MODE (x);
5304 5431 : unsigned HOST_WIDE_INT this_int = INTVAL (XEXP (XEXP (x, 1), 1));
5305 5431 : HOST_WIDE_INT other_int = trunc_int_for_mode (-this_int, mode);
5306 5431 : SUBST (*loc, gen_rtx_PLUS (mode,
5307 : gen_rtx_MULT (mode,
5308 : XEXP (XEXP (x, 1), 0),
5309 : gen_int_mode (other_int,
5310 : mode)),
5311 : XEXP (x, 0)));
5312 5431 : return find_split_point (loc, insn, set_src);
5313 : }
5314 :
5315 : /* Split at a multiply-accumulate instruction. However if this is
5316 : the SET_SRC, we likely do not have such an instruction and it's
5317 : worthless to try this split. */
5318 3120301 : if (!set_src
5319 1861639 : && (GET_CODE (XEXP (x, 0)) == MULT
5320 1751715 : || (GET_CODE (XEXP (x, 0)) == ASHIFT
5321 122728 : && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT)))
5322 : return loc;
5323 :
5324 : default:
5325 : break;
5326 : }
5327 :
5328 : /* Otherwise, select our actions depending on our rtx class. */
5329 24951191 : switch (GET_RTX_CLASS (code))
5330 : {
5331 1394563 : case RTX_BITFIELD_OPS: /* This is ZERO_EXTRACT and SIGN_EXTRACT. */
5332 1394563 : case RTX_TERNARY:
5333 1394563 : split = find_split_point (&XEXP (x, 2), insn, false);
5334 1394563 : if (split)
5335 : return split;
5336 : /* fall through */
5337 9867947 : case RTX_BIN_ARITH:
5338 9867947 : case RTX_COMM_ARITH:
5339 9867947 : case RTX_COMPARE:
5340 9867947 : case RTX_COMM_COMPARE:
5341 9867947 : split = find_split_point (&XEXP (x, 1), insn, false);
5342 9867947 : if (split)
5343 : return split;
5344 : /* fall through */
5345 9408365 : case RTX_UNARY:
5346 : /* Some machines have (and (shift ...) ...) insns. If X is not
5347 : an AND, but XEXP (X, 0) is, use it as our split point. */
5348 9408365 : if (GET_CODE (x) != AND && GET_CODE (XEXP (x, 0)) == AND)
5349 370714 : return &XEXP (x, 0);
5350 :
5351 9037651 : split = find_split_point (&XEXP (x, 0), insn, false);
5352 9037651 : if (split)
5353 : return split;
5354 : return loc;
5355 :
5356 : default:
5357 : /* Otherwise, we don't have a split point. */
5358 : return 0;
5359 : }
5360 : }
5361 :
5362 : /* Throughout X, replace FROM with TO, and return the result.
5363 : The result is TO if X is FROM;
5364 : otherwise the result is X, but its contents may have been modified.
5365 : If they were modified, a record was made in undobuf so that
5366 : undo_all will (among other things) return X to its original state.
5367 :
5368 : If the number of changes necessary is too much to record to undo,
5369 : the excess changes are not made, so the result is invalid.
5370 : The changes already made can still be undone.
5371 : undobuf.num_undo is incremented for such changes, so by testing that
5372 : the caller can tell whether the result is valid.
5373 :
5374 : `n_occurrences' is incremented each time FROM is replaced.
5375 :
5376 : IN_DEST is true if we are processing the SET_DEST of a SET.
5377 :
5378 : IN_COND is true if we are at the top level of a condition.
5379 :
5380 : UNIQUE_COPY is true if each substitution must be unique. We do this
5381 : by copying if `n_occurrences' is nonzero. */
5382 :
5383 : static rtx
5384 406525018 : subst (rtx x, rtx from, rtx to, bool in_dest, bool in_cond, bool unique_copy)
5385 : {
5386 406525018 : enum rtx_code code = GET_CODE (x);
5387 406525018 : machine_mode op0_mode = VOIDmode;
5388 406525018 : const char *fmt;
5389 406525018 : int len, i;
5390 406525018 : rtx new_rtx;
5391 :
5392 : /* Two expressions are equal if they are identical copies of a shared
5393 : RTX or if they are both registers with the same register number
5394 : and mode. */
5395 :
5396 : #define COMBINE_RTX_EQUAL_P(X,Y) \
5397 : ((X) == (Y) \
5398 : || (REG_P (X) && REG_P (Y) \
5399 : && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5400 :
5401 : /* Do not substitute into clobbers of regs -- this will never result in
5402 : valid RTL. */
5403 406525018 : if (GET_CODE (x) == CLOBBER && REG_P (XEXP (x, 0)))
5404 : return x;
5405 :
5406 396404231 : if (! in_dest && COMBINE_RTX_EQUAL_P (x, from))
5407 : {
5408 0 : n_occurrences++;
5409 0 : return (unique_copy && n_occurrences > 1 ? copy_rtx (to) : to);
5410 : }
5411 :
5412 : /* If X and FROM are the same register but different modes, they
5413 : will not have been seen as equal above. However, the log links code
5414 : will make a LOG_LINKS entry for that case. If we do nothing, we
5415 : will try to rerecognize our original insn and, when it succeeds,
5416 : we will delete the feeding insn, which is incorrect.
5417 :
5418 : So force this insn not to match in this (rare) case. */
5419 87933721 : if (! in_dest && code == REG && REG_P (from)
5420 428289193 : && reg_overlap_mentioned_p (x, from))
5421 4211 : return gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
5422 :
5423 : /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5424 : of which may contain things that can be combined. */
5425 396400020 : if (code != MEM && code != LO_SUM && OBJECT_P (x))
5426 : return x;
5427 :
5428 : /* It is possible to have a subexpression appear twice in the insn.
5429 : Suppose that FROM is a register that appears within TO.
5430 : Then, after that subexpression has been scanned once by `subst',
5431 : the second time it is scanned, TO may be found. If we were
5432 : to scan TO here, we would find FROM within it and create a
5433 : self-referent rtl structure which is completely wrong. */
5434 212095782 : if (COMBINE_RTX_EQUAL_P (x, to))
5435 : return to;
5436 :
5437 : /* Parallel asm_operands need special attention because all of the
5438 : inputs are shared across the arms. Furthermore, unsharing the
5439 : rtl results in recognition failures. Failure to handle this case
5440 : specially can result in circular rtl.
5441 :
5442 : Solve this by doing a normal pass across the first entry of the
5443 : parallel, and only processing the SET_DESTs of the subsequent
5444 : entries. Ug. */
5445 :
5446 211956694 : if (code == PARALLEL
5447 12500231 : && GET_CODE (XVECEXP (x, 0, 0)) == SET
5448 10700159 : && GET_CODE (SET_SRC (XVECEXP (x, 0, 0))) == ASM_OPERANDS)
5449 : {
5450 20019 : new_rtx = subst (XVECEXP (x, 0, 0), from, to, false, false, unique_copy);
5451 :
5452 : /* If this substitution failed, this whole thing fails. */
5453 20019 : if (GET_CODE (new_rtx) == CLOBBER
5454 0 : && XEXP (new_rtx, 0) == const0_rtx)
5455 : return new_rtx;
5456 :
5457 20019 : SUBST (XVECEXP (x, 0, 0), new_rtx);
5458 :
5459 99796 : for (i = XVECLEN (x, 0) - 1; i >= 1; i--)
5460 : {
5461 79777 : rtx dest = SET_DEST (XVECEXP (x, 0, i));
5462 :
5463 79777 : if (!REG_P (dest) && GET_CODE (dest) != PC)
5464 : {
5465 2248 : new_rtx = subst (dest, from, to, false, false, unique_copy);
5466 :
5467 : /* If this substitution failed, this whole thing fails. */
5468 2248 : if (GET_CODE (new_rtx) == CLOBBER
5469 0 : && XEXP (new_rtx, 0) == const0_rtx)
5470 : return new_rtx;
5471 :
5472 2248 : SUBST (SET_DEST (XVECEXP (x, 0, i)), new_rtx);
5473 : }
5474 : }
5475 : }
5476 : else
5477 : {
5478 211936675 : len = GET_RTX_LENGTH (code);
5479 211936675 : fmt = GET_RTX_FORMAT (code);
5480 :
5481 : /* We don't need to process a SET_DEST that is a register or PC, so
5482 : set up to skip this common case. All other cases where we want
5483 : to suppress replacing something inside a SET_SRC are handled via
5484 : the IN_DEST operand. */
5485 211936675 : if (code == SET
5486 46889619 : && (REG_P (SET_DEST (x))
5487 46889619 : || GET_CODE (SET_DEST (x)) == PC))
5488 211936675 : fmt = "ie";
5489 :
5490 : /* Trying to simplify the operands of a widening MULT is not likely
5491 : to create RTL matching a machine insn. */
5492 211936675 : if (code == MULT
5493 4726979 : && (GET_CODE (XEXP (x, 0)) == ZERO_EXTEND
5494 4726979 : || GET_CODE (XEXP (x, 0)) == SIGN_EXTEND)
5495 278948 : && (GET_CODE (XEXP (x, 1)) == ZERO_EXTEND
5496 278948 : || GET_CODE (XEXP (x, 1)) == SIGN_EXTEND)
5497 209211 : && REG_P (XEXP (XEXP (x, 0), 0))
5498 93072 : && REG_P (XEXP (XEXP (x, 1), 0))
5499 74697 : && from == to)
5500 : return x;
5501 :
5502 :
5503 : /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5504 : constant. */
5505 211894457 : if (fmt[0] == 'e')
5506 156097989 : op0_mode = GET_MODE (XEXP (x, 0));
5507 :
5508 628426838 : for (i = 0; i < len; i++)
5509 : {
5510 417363535 : if (fmt[i] == 'E')
5511 : {
5512 14844079 : int j;
5513 46914157 : for (j = XVECLEN (x, i) - 1; j >= 0; j--)
5514 : {
5515 32170636 : if (COMBINE_RTX_EQUAL_P (XVECEXP (x, i, j), from))
5516 : {
5517 1598 : new_rtx = (unique_copy && n_occurrences
5518 293703 : ? copy_rtx (to) : to);
5519 293679 : n_occurrences++;
5520 : }
5521 : else
5522 : {
5523 31876957 : new_rtx = subst (XVECEXP (x, i, j), from, to,
5524 : false, false, unique_copy);
5525 :
5526 : /* If this substitution failed, this whole thing
5527 : fails. */
5528 31876957 : if (GET_CODE (new_rtx) == CLOBBER
5529 10535417 : && XEXP (new_rtx, 0) == const0_rtx)
5530 : return new_rtx;
5531 : }
5532 :
5533 32070078 : SUBST (XVECEXP (x, i, j), new_rtx);
5534 : }
5535 : }
5536 402519456 : else if (fmt[i] == 'e')
5537 : {
5538 : /* If this is a register being set, ignore it. */
5539 328791403 : new_rtx = XEXP (x, i);
5540 328791403 : if (in_dest
5541 328791403 : && i == 0
5542 5972169 : && (((code == SUBREG || code == ZERO_EXTRACT)
5543 354361 : && REG_P (new_rtx))
5544 5620154 : || code == STRICT_LOW_PART))
5545 : ;
5546 :
5547 328428772 : else if (COMBINE_RTX_EQUAL_P (XEXP (x, i), from))
5548 : {
5549 : /* In general, don't install a subreg involving two
5550 : modes not tieable. It can worsen register
5551 : allocation, and can even make invalid reload
5552 : insns, since the reg inside may need to be copied
5553 : from in the outside mode, and that may be invalid
5554 : if it is an fp reg copied in integer mode.
5555 :
5556 : We allow an exception to this: It is valid if
5557 : it is inside another SUBREG and the mode of that
5558 : SUBREG and the mode of the inside of TO is
5559 : tieable. */
5560 :
5561 47166549 : if (GET_CODE (to) == SUBREG
5562 535979 : && !targetm.modes_tieable_p (GET_MODE (to),
5563 535979 : GET_MODE (SUBREG_REG (to)))
5564 47449732 : && ! (code == SUBREG
5565 22702 : && (targetm.modes_tieable_p
5566 22702 : (GET_MODE (x), GET_MODE (SUBREG_REG (to))))))
5567 258236 : return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5568 :
5569 46908313 : if (code == SUBREG
5570 2425807 : && REG_P (to)
5571 98614 : && REGNO (to) < FIRST_PSEUDO_REGISTER
5572 46908318 : && simplify_subreg_regno (REGNO (to), GET_MODE (to),
5573 5 : SUBREG_BYTE (x),
5574 5 : GET_MODE (x)) < 0)
5575 0 : return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5576 :
5577 46908313 : new_rtx = (unique_copy && n_occurrences ? copy_rtx (to) : to);
5578 46908313 : n_occurrences++;
5579 : }
5580 : else
5581 : /* If we are in a SET_DEST, suppress most cases unless we
5582 : have gone inside a MEM, in which case we want to
5583 : simplify the address. We assume here that things that
5584 : are actually part of the destination have their inner
5585 : parts in the first expression. This is true for SUBREG,
5586 : STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5587 : things aside from REG and MEM that should appear in a
5588 : SET_DEST. */
5589 281262223 : new_rtx = subst (XEXP (x, i), from, to,
5590 : (((in_dest
5591 5331684 : && (code == SUBREG || code == STRICT_LOW_PART
5592 5331684 : || code == ZERO_EXTRACT))
5593 281254670 : || code == SET)
5594 48449355 : && i == 0),
5595 281262223 : code == IF_THEN_ELSE && i == 0,
5596 : unique_copy);
5597 :
5598 : /* If we found that we will have to reject this combination,
5599 : indicate that by returning the CLOBBER ourselves, rather than
5600 : an expression containing it. This will speed things up as
5601 : well as prevent accidents where two CLOBBERs are considered
5602 : to be equal, thus producing an incorrect simplification. */
5603 :
5604 328533167 : if (GET_CODE (new_rtx) == CLOBBER && XEXP (new_rtx, 0) == const0_rtx)
5605 : return new_rtx;
5606 :
5607 328061059 : if (GET_CODE (x) == SUBREG && CONST_SCALAR_INT_P (new_rtx))
5608 : {
5609 31269 : machine_mode mode = GET_MODE (x);
5610 :
5611 62538 : x = simplify_subreg (GET_MODE (x), new_rtx,
5612 31269 : GET_MODE (SUBREG_REG (x)),
5613 31269 : SUBREG_BYTE (x));
5614 31269 : if (! x)
5615 2 : x = gen_rtx_CLOBBER (mode, const0_rtx);
5616 : }
5617 328029790 : else if (CONST_SCALAR_INT_P (new_rtx)
5618 : && (GET_CODE (x) == ZERO_EXTEND
5619 59540984 : || GET_CODE (x) == SIGN_EXTEND
5620 : || GET_CODE (x) == FLOAT
5621 : || GET_CODE (x) == UNSIGNED_FLOAT))
5622 : {
5623 143492 : x = simplify_unary_operation (GET_CODE (x), GET_MODE (x),
5624 : new_rtx,
5625 71746 : GET_MODE (XEXP (x, 0)));
5626 71746 : if (!x)
5627 252 : return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5628 : }
5629 : /* CONST_INTs shouldn't be substituted into PRE_DEC, PRE_MODIFY
5630 : etc. arguments, otherwise we can ICE before trying to recog
5631 : it. See PR104446. */
5632 327958044 : else if (CONST_SCALAR_INT_P (new_rtx)
5633 59469238 : && GET_RTX_CLASS (GET_CODE (x)) == RTX_AUTOINC)
5634 0 : return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5635 : else
5636 327958044 : SUBST (XEXP (x, i), new_rtx);
5637 : }
5638 : }
5639 : }
5640 :
5641 : /* Check if we are loading something from the constant pool via float
5642 : extension; in this case we would undo compress_float_constant
5643 : optimization and degenerate constant load to an immediate value. */
5644 211083322 : if (GET_CODE (x) == FLOAT_EXTEND
5645 315628 : && MEM_P (XEXP (x, 0))
5646 211146359 : && MEM_READONLY_P (XEXP (x, 0)))
5647 : {
5648 35770 : rtx tmp = avoid_constant_pool_reference (x);
5649 35770 : if (x != tmp)
5650 : return x;
5651 : }
5652 :
5653 : /* Try to simplify X. If the simplification changed the code, it is likely
5654 : that further simplification will help, so loop, but limit the number
5655 : of repetitions that will be performed. */
5656 :
5657 218914167 : for (i = 0; i < 4; i++)
5658 : {
5659 : /* If X is sufficiently simple, don't bother trying to do anything
5660 : with it. */
5661 218904564 : if (code != CONST_INT && code != REG && code != CLOBBER)
5662 218244294 : x = combine_simplify_rtx (x, op0_mode, in_dest, in_cond);
5663 :
5664 218904564 : if (GET_CODE (x) == code)
5665 : break;
5666 :
5667 7866485 : code = GET_CODE (x);
5668 :
5669 : /* We no longer know the original mode of operand 0 since we
5670 : have changed the form of X) */
5671 7866485 : op0_mode = VOIDmode;
5672 : }
5673 :
5674 : return x;
5675 : }
5676 :
5677 : /* If X is a commutative operation whose operands are not in the canonical
5678 : order, use substitutions to swap them. */
5679 :
5680 : static void
5681 689563795 : maybe_swap_commutative_operands (rtx x)
5682 : {
5683 689563795 : if (COMMUTATIVE_ARITH_P (x)
5684 689563795 : && swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1)))
5685 : {
5686 3597721 : rtx temp = XEXP (x, 0);
5687 3597721 : SUBST (XEXP (x, 0), XEXP (x, 1));
5688 3597721 : SUBST (XEXP (x, 1), temp);
5689 : }
5690 :
5691 : /* Canonicalize (vec_merge (fma op2 op1 op3) op1 mask) to
5692 : (vec_merge (fma op1 op2 op3) op1 mask). */
5693 689563795 : if (GET_CODE (x) == VEC_MERGE
5694 806748 : && GET_CODE (XEXP (x, 0)) == FMA)
5695 : {
5696 25208 : rtx fma_op1 = XEXP (XEXP (x, 0), 0);
5697 25208 : rtx fma_op2 = XEXP (XEXP (x, 0), 1);
5698 25208 : rtx masked_op = XEXP (x, 1);
5699 25208 : if (rtx_equal_p (masked_op, fma_op2))
5700 : {
5701 218 : if (GET_CODE (fma_op1) == NEG)
5702 : {
5703 : /* Keep the negate canonicalized to the first operand. */
5704 150 : fma_op1 = XEXP (fma_op1, 0);
5705 150 : SUBST (XEXP (XEXP (XEXP (x, 0), 0), 0), fma_op2);
5706 150 : SUBST (XEXP (XEXP (x, 0), 1), fma_op1);
5707 : }
5708 : else
5709 : {
5710 68 : SUBST (XEXP (XEXP (x, 0), 0), fma_op2);
5711 68 : SUBST (XEXP (XEXP (x, 0), 1), fma_op1);
5712 : }
5713 : }
5714 : }
5715 :
5716 689563795 : unsigned n_elts = 0;
5717 689563795 : if (GET_CODE (x) == VEC_MERGE
5718 806748 : && CONST_INT_P (XEXP (x, 2))
5719 861832 : && GET_MODE_NUNITS (GET_MODE (x)).is_constant (&n_elts)
5720 689994711 : && (swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1))
5721 : /* Two operands have same precedence, then
5722 : first bit of mask select first operand. */
5723 396239 : || (!swap_commutative_operands_p (XEXP (x, 1), XEXP (x, 0))
5724 103502 : && !(UINTVAL (XEXP (x, 2)) & 1))))
5725 : {
5726 49900 : rtx temp = XEXP (x, 0);
5727 49900 : unsigned HOST_WIDE_INT sel = UINTVAL (XEXP (x, 2));
5728 49900 : unsigned HOST_WIDE_INT mask = HOST_WIDE_INT_1U;
5729 49900 : if (n_elts == HOST_BITS_PER_WIDE_INT)
5730 : mask = -1;
5731 : else
5732 49761 : mask = (HOST_WIDE_INT_1U << n_elts) - 1;
5733 49900 : SUBST (XEXP (x, 0), XEXP (x, 1));
5734 49900 : SUBST (XEXP (x, 1), temp);
5735 49900 : SUBST (XEXP (x, 2), GEN_INT (~sel & mask));
5736 : }
5737 689563795 : }
5738 :
5739 : /* Simplify X, a piece of RTL. We just operate on the expression at the
5740 : outer level; call `subst' to simplify recursively. Return the new
5741 : expression.
5742 :
5743 : OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is true
5744 : if we are inside a SET_DEST. IN_COND is true if we are at the top level
5745 : of a condition. */
5746 :
5747 : static rtx
5748 218542994 : combine_simplify_rtx (rtx x, machine_mode op0_mode, bool in_dest, bool in_cond)
5749 : {
5750 218542994 : enum rtx_code code = GET_CODE (x);
5751 218542994 : machine_mode mode = GET_MODE (x);
5752 218542994 : scalar_int_mode int_mode;
5753 218542994 : rtx temp;
5754 218542994 : int i;
5755 :
5756 : /* If this is a commutative operation, put a constant last and a complex
5757 : expression first. We don't need to do this for comparisons here. */
5758 218542994 : maybe_swap_commutative_operands (x);
5759 :
5760 : /* Try to fold this expression in case we have constants that weren't
5761 : present before. */
5762 218542994 : temp = 0;
5763 218542994 : switch (GET_RTX_CLASS (code))
5764 : {
5765 6889694 : case RTX_UNARY:
5766 6889694 : if (op0_mode == VOIDmode)
5767 148740 : op0_mode = GET_MODE (XEXP (x, 0));
5768 6889694 : temp = simplify_unary_operation (code, mode, XEXP (x, 0), op0_mode);
5769 6889694 : break;
5770 17889666 : case RTX_COMPARE:
5771 17889666 : case RTX_COMM_COMPARE:
5772 17889666 : {
5773 17889666 : machine_mode cmp_mode = GET_MODE (XEXP (x, 0));
5774 17889666 : if (cmp_mode == VOIDmode)
5775 : {
5776 53446 : cmp_mode = GET_MODE (XEXP (x, 1));
5777 53446 : if (cmp_mode == VOIDmode)
5778 7934 : cmp_mode = op0_mode;
5779 : }
5780 17889666 : temp = simplify_relational_operation (code, mode, cmp_mode,
5781 : XEXP (x, 0), XEXP (x, 1));
5782 : }
5783 17889666 : break;
5784 86334643 : case RTX_COMM_ARITH:
5785 86334643 : case RTX_BIN_ARITH:
5786 86334643 : temp = simplify_binary_operation (code, mode, XEXP (x, 0), XEXP (x, 1));
5787 86334643 : break;
5788 14246086 : case RTX_BITFIELD_OPS:
5789 14246086 : case RTX_TERNARY:
5790 14246086 : temp = simplify_ternary_operation (code, mode, op0_mode, XEXP (x, 0),
5791 : XEXP (x, 1), XEXP (x, 2));
5792 14246086 : break;
5793 : default:
5794 : break;
5795 : }
5796 :
5797 125360089 : if (temp)
5798 : {
5799 16683706 : x = temp;
5800 16683706 : code = GET_CODE (temp);
5801 16683706 : op0_mode = VOIDmode;
5802 16683706 : mode = GET_MODE (temp);
5803 : }
5804 :
5805 : /* If this is a simple operation applied to an IF_THEN_ELSE, try
5806 : applying it to the arms of the IF_THEN_ELSE. This often simplifies
5807 : things. Check for cases where both arms are testing the same
5808 : condition.
5809 :
5810 : Don't do anything if all operands are very simple. */
5811 :
5812 218542994 : if ((BINARY_P (x)
5813 103997537 : && ((!OBJECT_P (XEXP (x, 0))
5814 40392093 : && ! (GET_CODE (XEXP (x, 0)) == SUBREG
5815 4628611 : && OBJECT_P (SUBREG_REG (XEXP (x, 0)))))
5816 66223191 : || (!OBJECT_P (XEXP (x, 1))
5817 4516697 : && ! (GET_CODE (XEXP (x, 1)) == SUBREG
5818 1546225 : && OBJECT_P (SUBREG_REG (XEXP (x, 1)))))))
5819 177536244 : || (UNARY_P (x)
5820 6745284 : && (!OBJECT_P (XEXP (x, 0))
5821 2863254 : && ! (GET_CODE (XEXP (x, 0)) == SUBREG
5822 605462 : && OBJECT_P (SUBREG_REG (XEXP (x, 0)))))))
5823 : {
5824 43334549 : rtx cond, true_rtx, false_rtx;
5825 :
5826 43334549 : cond = if_then_else_cond (x, &true_rtx, &false_rtx);
5827 43334549 : if (cond != 0
5828 : /* If everything is a comparison, what we have is highly unlikely
5829 : to be simpler, so don't use it. */
5830 4141982 : && ! (COMPARISON_P (x)
5831 1263435 : && (COMPARISON_P (true_rtx) || COMPARISON_P (false_rtx)))
5832 : /* Similarly, if we end up with one of the expressions the same
5833 : as the original, it is certainly not simpler. */
5834 3942048 : && ! rtx_equal_p (x, true_rtx)
5835 47276597 : && ! rtx_equal_p (x, false_rtx))
5836 : {
5837 3942048 : rtx cop1 = const0_rtx;
5838 3942048 : enum rtx_code cond_code = simplify_comparison (NE, &cond, &cop1);
5839 :
5840 3942048 : if (cond_code == NE && COMPARISON_P (cond))
5841 644539 : return x;
5842 :
5843 : /* Simplify the alternative arms; this may collapse the true and
5844 : false arms to store-flag values. Be careful to use copy_rtx
5845 : here since true_rtx or false_rtx might share RTL with x as a
5846 : result of the if_then_else_cond call above. */
5847 3297509 : true_rtx = subst (copy_rtx (true_rtx), pc_rtx, pc_rtx,
5848 : false, false, false);
5849 3297509 : false_rtx = subst (copy_rtx (false_rtx), pc_rtx, pc_rtx,
5850 : false, false, false);
5851 :
5852 : /* If true_rtx and false_rtx are not general_operands, an if_then_else
5853 : is unlikely to be simpler. */
5854 3297509 : if (general_operand (true_rtx, VOIDmode)
5855 3297509 : && general_operand (false_rtx, VOIDmode))
5856 : {
5857 1361237 : enum rtx_code reversed;
5858 :
5859 : /* Restarting if we generate a store-flag expression will cause
5860 : us to loop. Just drop through in this case. */
5861 :
5862 : /* If the result values are STORE_FLAG_VALUE and zero, we can
5863 : just make the comparison operation. */
5864 1361237 : if (true_rtx == const_true_rtx && false_rtx == const0_rtx)
5865 500658 : x = simplify_gen_relational (cond_code, mode, VOIDmode,
5866 : cond, cop1);
5867 633712 : else if (true_rtx == const0_rtx && false_rtx == const_true_rtx
5868 860579 : && ((reversed = reversed_comparison_code_parts
5869 571327 : (cond_code, cond, cop1, NULL))
5870 : != UNKNOWN))
5871 571327 : x = simplify_gen_relational (reversed, mode, VOIDmode,
5872 : cond, cop1);
5873 :
5874 : /* Likewise, we can make the negate of a comparison operation
5875 : if the result values are - STORE_FLAG_VALUE and zero. */
5876 289252 : else if (CONST_INT_P (true_rtx)
5877 205514 : && INTVAL (true_rtx) == - STORE_FLAG_VALUE
5878 41134 : && false_rtx == const0_rtx)
5879 39098 : x = simplify_gen_unary (NEG, mode,
5880 : simplify_gen_relational (cond_code,
5881 : mode, VOIDmode,
5882 : cond, cop1),
5883 : mode);
5884 250154 : else if (CONST_INT_P (false_rtx)
5885 200116 : && INTVAL (false_rtx) == - STORE_FLAG_VALUE
5886 22857 : && true_rtx == const0_rtx
5887 250154 : && ((reversed = reversed_comparison_code_parts
5888 20083 : (cond_code, cond, cop1, NULL))
5889 : != UNKNOWN))
5890 20080 : x = simplify_gen_unary (NEG, mode,
5891 : simplify_gen_relational (reversed,
5892 : mode, VOIDmode,
5893 : cond, cop1),
5894 : mode);
5895 :
5896 1361237 : code = GET_CODE (x);
5897 1361237 : op0_mode = VOIDmode;
5898 : }
5899 : }
5900 : }
5901 :
5902 : /* First see if we can apply the inverse distributive law. */
5903 217898455 : if (code == PLUS || code == MINUS
5904 217898455 : || code == AND || code == IOR || code == XOR)
5905 : {
5906 48962586 : x = apply_distributive_law (x);
5907 48962586 : code = GET_CODE (x);
5908 48962586 : op0_mode = VOIDmode;
5909 : }
5910 :
5911 : /* If CODE is an associative operation not otherwise handled, see if we
5912 : can associate some operands. This can win if they are constants or
5913 : if they are logically related (i.e. (a & b) & a). */
5914 217898455 : if ((code == PLUS || code == MINUS || code == MULT || code == DIV
5915 : || code == AND || code == IOR || code == XOR
5916 : || code == SMAX || code == SMIN || code == UMAX || code == UMIN)
5917 52932430 : && ((INTEGRAL_MODE_P (mode) && code != DIV)
5918 4671788 : || (flag_associative_math && FLOAT_MODE_P (mode))))
5919 : {
5920 48897784 : if (GET_CODE (XEXP (x, 0)) == code)
5921 : {
5922 4264224 : rtx other = XEXP (XEXP (x, 0), 0);
5923 4264224 : rtx inner_op0 = XEXP (XEXP (x, 0), 1);
5924 4264224 : rtx inner_op1 = XEXP (x, 1);
5925 4264224 : rtx inner;
5926 :
5927 : /* Make sure we pass the constant operand if any as the second
5928 : one if this is a commutative operation. */
5929 4264224 : if (CONSTANT_P (inner_op0) && COMMUTATIVE_ARITH_P (x))
5930 : std::swap (inner_op0, inner_op1);
5931 4264224 : inner = simplify_binary_operation (code == MINUS ? PLUS
5932 4171861 : : code == DIV ? MULT
5933 : : code,
5934 : mode, inner_op0, inner_op1);
5935 :
5936 : /* For commutative operations, try the other pair if that one
5937 : didn't simplify. */
5938 4264224 : if (inner == 0 && COMMUTATIVE_ARITH_P (x))
5939 : {
5940 4139741 : other = XEXP (XEXP (x, 0), 1);
5941 4139741 : inner = simplify_binary_operation (code, mode,
5942 : XEXP (XEXP (x, 0), 0),
5943 : XEXP (x, 1));
5944 : }
5945 :
5946 4228979 : if (inner)
5947 235881 : return simplify_gen_binary (code, mode, other, inner);
5948 : }
5949 : }
5950 :
5951 : /* A little bit of algebraic simplification here. */
5952 217662574 : switch (code)
5953 : {
5954 22494286 : case MEM:
5955 : /* Ensure that our address has any ASHIFTs converted to MULT in case
5956 : address-recognizing predicates are called later. */
5957 22494286 : temp = make_compound_operation (XEXP (x, 0), MEM);
5958 22494286 : SUBST (XEXP (x, 0), temp);
5959 22494286 : break;
5960 :
5961 8061774 : case SUBREG:
5962 8061774 : if (op0_mode == VOIDmode)
5963 141652 : op0_mode = GET_MODE (SUBREG_REG (x));
5964 :
5965 : /* See if this can be moved to simplify_subreg. */
5966 8061774 : if (CONSTANT_P (SUBREG_REG (x))
5967 19467 : && known_eq (subreg_lowpart_offset (mode, op0_mode), SUBREG_BYTE (x))
5968 : /* Don't call gen_lowpart if the inner mode
5969 : is VOIDmode and we cannot simplify it, as SUBREG without
5970 : inner mode is invalid. */
5971 8081241 : && (GET_MODE (SUBREG_REG (x)) != VOIDmode
5972 0 : || gen_lowpart_common (mode, SUBREG_REG (x))))
5973 19467 : return gen_lowpart (mode, SUBREG_REG (x));
5974 :
5975 8042307 : if (GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_CC)
5976 : break;
5977 8042307 : {
5978 8042307 : rtx temp;
5979 16084614 : temp = simplify_subreg (mode, SUBREG_REG (x), op0_mode,
5980 8042307 : SUBREG_BYTE (x));
5981 8042307 : if (temp)
5982 218542994 : return temp;
5983 :
5984 : /* If op is known to have all lower bits zero, the result is zero. */
5985 7447872 : scalar_int_mode int_mode, int_op0_mode;
5986 7447872 : if (!in_dest
5987 4012501 : && is_a <scalar_int_mode> (mode, &int_mode)
5988 3932907 : && is_a <scalar_int_mode> (op0_mode, &int_op0_mode)
5989 3932907 : && (GET_MODE_PRECISION (int_mode)
5990 3932907 : < GET_MODE_PRECISION (int_op0_mode))
5991 3387667 : && known_eq (subreg_lowpart_offset (int_mode, int_op0_mode),
5992 : SUBREG_BYTE (x))
5993 2914821 : && HWI_COMPUTABLE_MODE_P (int_op0_mode)
5994 2699377 : && ((nonzero_bits (SUBREG_REG (x), int_op0_mode)
5995 2699377 : & GET_MODE_MASK (int_mode)) == 0)
5996 7448678 : && !side_effects_p (SUBREG_REG (x)))
5997 806 : return CONST0_RTX (int_mode);
5998 : }
5999 :
6000 : /* Don't change the mode of the MEM if that would change the meaning
6001 : of the address. */
6002 7447066 : if (MEM_P (SUBREG_REG (x))
6003 7447066 : && (MEM_VOLATILE_P (SUBREG_REG (x))
6004 89575 : || mode_dependent_address_p (XEXP (SUBREG_REG (x), 0),
6005 89642 : MEM_ADDR_SPACE (SUBREG_REG (x)))))
6006 44113 : return gen_rtx_CLOBBER (mode, const0_rtx);
6007 :
6008 : /* Note that we cannot do any narrowing for non-constants since
6009 : we might have been counting on using the fact that some bits were
6010 : zero. We now do this in the SET. */
6011 :
6012 : break;
6013 :
6014 368931 : case NEG:
6015 368931 : temp = expand_compound_operation (XEXP (x, 0));
6016 :
6017 : /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
6018 : replaced by (lshiftrt X C). This will convert
6019 : (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
6020 :
6021 368931 : if (GET_CODE (temp) == ASHIFTRT
6022 14168 : && CONST_INT_P (XEXP (temp, 1))
6023 397205 : && INTVAL (XEXP (temp, 1)) == GET_MODE_UNIT_PRECISION (mode) - 1)
6024 0 : return simplify_shift_const (NULL_RTX, LSHIFTRT, mode, XEXP (temp, 0),
6025 0 : INTVAL (XEXP (temp, 1)));
6026 :
6027 : /* If X has only a single bit that might be nonzero, say, bit I, convert
6028 : (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
6029 : MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
6030 : (sign_extract X 1 Y). But only do this if TEMP isn't a register
6031 : or a SUBREG of one since we'd be making the expression more
6032 : complex if it was just a register. */
6033 :
6034 368931 : if (!REG_P (temp)
6035 179167 : && ! (GET_CODE (temp) == SUBREG
6036 20244 : && REG_P (SUBREG_REG (temp)))
6037 218680018 : && is_a <scalar_int_mode> (mode, &int_mode)
6038 505955 : && (i = exact_log2 (nonzero_bits (temp, int_mode))) >= 0)
6039 : {
6040 60076 : rtx temp1 = simplify_shift_const
6041 60076 : (NULL_RTX, ASHIFTRT, int_mode,
6042 : simplify_shift_const (NULL_RTX, ASHIFT, int_mode, temp,
6043 60076 : GET_MODE_PRECISION (int_mode) - 1 - i),
6044 60076 : GET_MODE_PRECISION (int_mode) - 1 - i);
6045 :
6046 : /* If all we did was surround TEMP with the two shifts, we
6047 : haven't improved anything, so don't use it. Otherwise,
6048 : we are better off with TEMP1. */
6049 60076 : if (GET_CODE (temp1) != ASHIFTRT
6050 59861 : || GET_CODE (XEXP (temp1, 0)) != ASHIFT
6051 59823 : || XEXP (XEXP (temp1, 0), 0) != temp)
6052 : return temp1;
6053 : }
6054 : break;
6055 :
6056 9527 : case TRUNCATE:
6057 : /* We can't handle truncation to a partial integer mode here
6058 : because we don't know the real bitsize of the partial
6059 : integer mode. */
6060 9527 : if (GET_MODE_CLASS (mode) == MODE_PARTIAL_INT)
6061 : break;
6062 :
6063 9527 : if (HWI_COMPUTABLE_MODE_P (mode))
6064 0 : SUBST (XEXP (x, 0),
6065 : force_to_mode (XEXP (x, 0), GET_MODE (XEXP (x, 0)),
6066 : GET_MODE_MASK (mode), false));
6067 :
6068 : /* We can truncate a constant value and return it. */
6069 9527 : {
6070 9527 : poly_int64 c;
6071 9527 : if (poly_int_rtx_p (XEXP (x, 0), &c))
6072 0 : return gen_int_mode (c, mode);
6073 : }
6074 :
6075 : /* Similarly to what we do in simplify-rtx.cc, a truncate of a register
6076 : whose value is a comparison can be replaced with a subreg if
6077 : STORE_FLAG_VALUE permits. */
6078 9527 : if (HWI_COMPUTABLE_MODE_P (mode)
6079 0 : && (STORE_FLAG_VALUE & ~GET_MODE_MASK (mode)) == 0
6080 0 : && (temp = get_last_value (XEXP (x, 0)))
6081 0 : && COMPARISON_P (temp)
6082 9527 : && TRULY_NOOP_TRUNCATION_MODES_P (mode, GET_MODE (XEXP (x, 0))))
6083 0 : return gen_lowpart (mode, XEXP (x, 0));
6084 : break;
6085 :
6086 5424 : case CONST:
6087 : /* (const (const X)) can become (const X). Do it this way rather than
6088 : returning the inner CONST since CONST can be shared with a
6089 : REG_EQUAL note. */
6090 5424 : if (GET_CODE (XEXP (x, 0)) == CONST)
6091 0 : SUBST (XEXP (x, 0), XEXP (XEXP (x, 0), 0));
6092 : break;
6093 :
6094 : case LO_SUM:
6095 : /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
6096 : can add in an offset. find_split_point will split this address up
6097 : again if it doesn't match. */
6098 : if (HAVE_lo_sum && GET_CODE (XEXP (x, 0)) == HIGH
6099 : && rtx_equal_p (XEXP (XEXP (x, 0), 0), XEXP (x, 1)))
6100 : return XEXP (x, 1);
6101 : break;
6102 :
6103 33331020 : case PLUS:
6104 : /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
6105 : when c is (const_int (pow2 + 1) / 2) is a sign extension of a
6106 : bit-field and can be replaced by either a sign_extend or a
6107 : sign_extract. The `and' may be a zero_extend and the two
6108 : <c>, -<c> constants may be reversed. */
6109 33331020 : if (GET_CODE (XEXP (x, 0)) == XOR
6110 33331020 : && is_a <scalar_int_mode> (mode, &int_mode)
6111 13866 : && CONST_INT_P (XEXP (x, 1))
6112 4925 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
6113 4395 : && INTVAL (XEXP (x, 1)) == -INTVAL (XEXP (XEXP (x, 0), 1))
6114 77 : && ((i = exact_log2 (UINTVAL (XEXP (XEXP (x, 0), 1)))) >= 0
6115 2 : || (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0)
6116 39 : && HWI_COMPUTABLE_MODE_P (int_mode)
6117 33331059 : && ((GET_CODE (XEXP (XEXP (x, 0), 0)) == AND
6118 0 : && CONST_INT_P (XEXP (XEXP (XEXP (x, 0), 0), 1))
6119 0 : && (UINTVAL (XEXP (XEXP (XEXP (x, 0), 0), 1))
6120 0 : == (HOST_WIDE_INT_1U << (i + 1)) - 1))
6121 39 : || (GET_CODE (XEXP (XEXP (x, 0), 0)) == ZERO_EXTEND
6122 0 : && known_eq ((GET_MODE_PRECISION
6123 : (GET_MODE (XEXP (XEXP (XEXP (x, 0), 0), 0)))),
6124 : (unsigned int) i + 1))))
6125 0 : return simplify_shift_const
6126 0 : (NULL_RTX, ASHIFTRT, int_mode,
6127 : simplify_shift_const (NULL_RTX, ASHIFT, int_mode,
6128 : XEXP (XEXP (XEXP (x, 0), 0), 0),
6129 0 : GET_MODE_PRECISION (int_mode) - (i + 1)),
6130 0 : GET_MODE_PRECISION (int_mode) - (i + 1));
6131 :
6132 : /* If only the low-order bit of X is possibly nonzero, (plus x -1)
6133 : can become (ashiftrt (ashift (xor x 1) C) C) where C is
6134 : the bitsize of the mode - 1. This allows simplification of
6135 : "a = (b & 8) == 0;" */
6136 33331020 : if (XEXP (x, 1) == constm1_rtx
6137 718678 : && !REG_P (XEXP (x, 0))
6138 304406 : && ! (GET_CODE (XEXP (x, 0)) == SUBREG
6139 37572 : && REG_P (SUBREG_REG (XEXP (x, 0))))
6140 33589272 : && is_a <scalar_int_mode> (mode, &int_mode)
6141 33599289 : && nonzero_bits (XEXP (x, 0), int_mode) == 1)
6142 10017 : return simplify_shift_const
6143 10017 : (NULL_RTX, ASHIFTRT, int_mode,
6144 : simplify_shift_const (NULL_RTX, ASHIFT, int_mode,
6145 : gen_rtx_XOR (int_mode, XEXP (x, 0),
6146 : const1_rtx),
6147 10017 : GET_MODE_PRECISION (int_mode) - 1),
6148 20034 : GET_MODE_PRECISION (int_mode) - 1);
6149 :
6150 : /* If we are adding two things that have no bits in common, convert
6151 : the addition into an IOR. This will often be further simplified,
6152 : for example in cases like ((a & 1) + (a & 2)), which can
6153 : become a & 3. */
6154 :
6155 33321003 : if (HWI_COMPUTABLE_MODE_P (mode)
6156 29431788 : && (nonzero_bits (XEXP (x, 0), mode)
6157 29431788 : & nonzero_bits (XEXP (x, 1), mode)) == 0)
6158 : {
6159 : /* Try to simplify the expression further. */
6160 298700 : rtx tor = simplify_gen_binary (IOR, mode, XEXP (x, 0), XEXP (x, 1));
6161 298700 : temp = combine_simplify_rtx (tor, VOIDmode, in_dest, false);
6162 :
6163 : /* If we could, great. If not, do not go ahead with the IOR
6164 : replacement, since PLUS appears in many special purpose
6165 : address arithmetic instructions. */
6166 298700 : if (GET_CODE (temp) != CLOBBER
6167 298700 : && (GET_CODE (temp) != IOR
6168 294417 : || ((XEXP (temp, 0) != XEXP (x, 0)
6169 293325 : || XEXP (temp, 1) != XEXP (x, 1))
6170 1092 : && (XEXP (temp, 0) != XEXP (x, 1)
6171 0 : || XEXP (temp, 1) != XEXP (x, 0)))))
6172 : return temp;
6173 : }
6174 :
6175 : /* Canonicalize x + x into x << 1. */
6176 33315628 : if (GET_MODE_CLASS (mode) == MODE_INT
6177 29747156 : && rtx_equal_p (XEXP (x, 0), XEXP (x, 1))
6178 33318783 : && !side_effects_p (XEXP (x, 0)))
6179 3150 : return simplify_gen_binary (ASHIFT, mode, XEXP (x, 0), const1_rtx);
6180 :
6181 : break;
6182 :
6183 3663245 : case MINUS:
6184 : /* (minus <foo> (and <foo> (const_int -pow2))) becomes
6185 : (and <foo> (const_int pow2-1)) */
6186 3663245 : if (is_a <scalar_int_mode> (mode, &int_mode)
6187 3083301 : && GET_CODE (XEXP (x, 1)) == AND
6188 107426 : && CONST_INT_P (XEXP (XEXP (x, 1), 1))
6189 104702 : && pow2p_hwi (-UINTVAL (XEXP (XEXP (x, 1), 1)))
6190 48451 : && rtx_equal_p (XEXP (XEXP (x, 1), 0), XEXP (x, 0)))
6191 0 : return simplify_and_const_int (NULL_RTX, int_mode, XEXP (x, 0),
6192 0 : -INTVAL (XEXP (XEXP (x, 1), 1)) - 1);
6193 : break;
6194 :
6195 2981331 : case MULT:
6196 : /* If we have (mult (plus A B) C), apply the distributive law and then
6197 : the inverse distributive law to see if things simplify. This
6198 : occurs mostly in addresses, often when unrolling loops. */
6199 :
6200 2981331 : if (GET_CODE (XEXP (x, 0)) == PLUS)
6201 : {
6202 266518 : rtx result = distribute_and_simplify_rtx (x, 0);
6203 266518 : if (result)
6204 : return result;
6205 : }
6206 :
6207 : /* Try simplify a*(b/c) as (a*b)/c. */
6208 2980766 : if (FLOAT_MODE_P (mode) && flag_associative_math
6209 201353 : && GET_CODE (XEXP (x, 0)) == DIV)
6210 : {
6211 259 : rtx tem = simplify_binary_operation (MULT, mode,
6212 : XEXP (XEXP (x, 0), 0),
6213 : XEXP (x, 1));
6214 259 : if (tem)
6215 33 : return simplify_gen_binary (DIV, mode, tem, XEXP (XEXP (x, 0), 1));
6216 : }
6217 : break;
6218 :
6219 122690 : case UDIV:
6220 : /* If this is a divide by a power of two, treat it as a shift if
6221 : its first operand is a shift. */
6222 122690 : if (is_a <scalar_int_mode> (mode, &int_mode)
6223 122690 : && CONST_INT_P (XEXP (x, 1))
6224 2024 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0
6225 0 : && (GET_CODE (XEXP (x, 0)) == ASHIFT
6226 0 : || GET_CODE (XEXP (x, 0)) == LSHIFTRT
6227 0 : || GET_CODE (XEXP (x, 0)) == ASHIFTRT
6228 0 : || GET_CODE (XEXP (x, 0)) == ROTATE
6229 0 : || GET_CODE (XEXP (x, 0)) == ROTATERT))
6230 0 : return simplify_shift_const (NULL_RTX, LSHIFTRT, int_mode,
6231 0 : XEXP (x, 0), i);
6232 : break;
6233 :
6234 17863177 : case EQ: case NE:
6235 17863177 : case GT: case GTU: case GE: case GEU:
6236 17863177 : case LT: case LTU: case LE: case LEU:
6237 17863177 : case UNEQ: case LTGT:
6238 17863177 : case UNGT: case UNGE:
6239 17863177 : case UNLT: case UNLE:
6240 17863177 : case UNORDERED: case ORDERED:
6241 : /* If the first operand is a condition code, we can't do anything
6242 : with it. */
6243 17863177 : if (GET_CODE (XEXP (x, 0)) == COMPARE
6244 17863177 : || GET_MODE_CLASS (GET_MODE (XEXP (x, 0))) != MODE_CC)
6245 : {
6246 13507201 : rtx op0 = XEXP (x, 0);
6247 13507201 : rtx op1 = XEXP (x, 1);
6248 13507201 : enum rtx_code new_code;
6249 :
6250 13507201 : if (GET_CODE (op0) == COMPARE)
6251 0 : op1 = XEXP (op0, 1), op0 = XEXP (op0, 0);
6252 :
6253 : /* Simplify our comparison, if possible. */
6254 13507201 : new_code = simplify_comparison (code, &op0, &op1);
6255 :
6256 : /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
6257 : if only the low-order bit is possibly nonzero in X (such as when
6258 : X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
6259 : (xor X 1) or (minus 1 X); we use the former. Finally, if X is
6260 : known to be either 0 or -1, NE becomes a NEG and EQ becomes
6261 : (plus X 1).
6262 :
6263 : Remove any ZERO_EXTRACT we made when thinking this was a
6264 : comparison. It may now be simpler to use, e.g., an AND. If a
6265 : ZERO_EXTRACT is indeed appropriate, it will be placed back by
6266 : the call to make_compound_operation in the SET case.
6267 :
6268 : Don't apply these optimizations if the caller would
6269 : prefer a comparison rather than a value.
6270 : E.g., for the condition in an IF_THEN_ELSE most targets need
6271 : an explicit comparison. */
6272 :
6273 13507201 : if (in_cond)
6274 : ;
6275 :
6276 2046480 : else if (STORE_FLAG_VALUE == 1
6277 : && new_code == NE
6278 2455933 : && is_int_mode (mode, &int_mode)
6279 409675 : && op1 == const0_rtx
6280 217865 : && int_mode == GET_MODE (op0)
6281 2133103 : && nonzero_bits (op0, int_mode) == 1)
6282 222 : return gen_lowpart (int_mode,
6283 456124 : expand_compound_operation (op0));
6284 :
6285 2046258 : else if (STORE_FLAG_VALUE == 1
6286 : && new_code == NE
6287 2455046 : && is_int_mode (mode, &int_mode)
6288 409453 : && op1 == const0_rtx
6289 217643 : && int_mode == GET_MODE (op0)
6290 2132659 : && (num_sign_bit_copies (op0, int_mode)
6291 86401 : == GET_MODE_PRECISION (int_mode)))
6292 : {
6293 665 : op0 = expand_compound_operation (op0);
6294 665 : return simplify_gen_unary (NEG, int_mode,
6295 665 : gen_lowpart (int_mode, op0),
6296 665 : int_mode);
6297 : }
6298 :
6299 2045593 : else if (STORE_FLAG_VALUE == 1
6300 : && new_code == EQ
6301 2328319 : && is_int_mode (mode, &int_mode)
6302 284511 : && op1 == const0_rtx
6303 135470 : && int_mode == GET_MODE (op0)
6304 2092869 : && nonzero_bits (op0, int_mode) == 1)
6305 : {
6306 1785 : op0 = expand_compound_operation (op0);
6307 1785 : return simplify_gen_binary (XOR, int_mode,
6308 1785 : gen_lowpart (int_mode, op0),
6309 1785 : const1_rtx);
6310 : }
6311 :
6312 2043808 : else if (STORE_FLAG_VALUE == 1
6313 : && new_code == EQ
6314 13786696 : && is_int_mode (mode, &int_mode)
6315 282726 : && op1 == const0_rtx
6316 133685 : && int_mode == GET_MODE (op0)
6317 2089299 : && (num_sign_bit_copies (op0, int_mode)
6318 45491 : == GET_MODE_PRECISION (int_mode)))
6319 : {
6320 559 : op0 = expand_compound_operation (op0);
6321 559 : return plus_constant (int_mode, gen_lowpart (int_mode, op0), 1);
6322 : }
6323 :
6324 : /* If STORE_FLAG_VALUE is -1, we have cases similar to
6325 : those above. */
6326 13503970 : if (in_cond)
6327 : ;
6328 :
6329 13503970 : else if (STORE_FLAG_VALUE == -1
6330 : && new_code == NE
6331 : && is_int_mode (mode, &int_mode)
6332 : && op1 == const0_rtx
6333 : && int_mode == GET_MODE (op0)
6334 : && (num_sign_bit_copies (op0, int_mode)
6335 : == GET_MODE_PRECISION (int_mode)))
6336 : return gen_lowpart (int_mode, expand_compound_operation (op0));
6337 :
6338 13503970 : else if (STORE_FLAG_VALUE == -1
6339 : && new_code == NE
6340 : && is_int_mode (mode, &int_mode)
6341 : && op1 == const0_rtx
6342 : && int_mode == GET_MODE (op0)
6343 : && nonzero_bits (op0, int_mode) == 1)
6344 : {
6345 : op0 = expand_compound_operation (op0);
6346 : return simplify_gen_unary (NEG, int_mode,
6347 : gen_lowpart (int_mode, op0),
6348 : int_mode);
6349 : }
6350 :
6351 13503970 : else if (STORE_FLAG_VALUE == -1
6352 : && new_code == EQ
6353 : && is_int_mode (mode, &int_mode)
6354 : && op1 == const0_rtx
6355 : && int_mode == GET_MODE (op0)
6356 : && (num_sign_bit_copies (op0, int_mode)
6357 : == GET_MODE_PRECISION (int_mode)))
6358 : {
6359 : op0 = expand_compound_operation (op0);
6360 : return simplify_gen_unary (NOT, int_mode,
6361 : gen_lowpart (int_mode, op0),
6362 : int_mode);
6363 : }
6364 :
6365 : /* If X is 0/1, (eq X 0) is X-1. */
6366 13503970 : else if (STORE_FLAG_VALUE == -1
6367 : && new_code == EQ
6368 : && is_int_mode (mode, &int_mode)
6369 : && op1 == const0_rtx
6370 : && int_mode == GET_MODE (op0)
6371 : && nonzero_bits (op0, int_mode) == 1)
6372 : {
6373 : op0 = expand_compound_operation (op0);
6374 : return plus_constant (int_mode, gen_lowpart (int_mode, op0), -1);
6375 : }
6376 :
6377 : /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
6378 : one bit that might be nonzero, we can convert (ne x 0) to
6379 : (ashift x c) where C puts the bit in the sign bit. Remove any
6380 : AND with STORE_FLAG_VALUE when we are done, since we are only
6381 : going to test the sign bit. */
6382 13503970 : if (new_code == NE
6383 13908855 : && is_int_mode (mode, &int_mode)
6384 408868 : && HWI_COMPUTABLE_MODE_P (int_mode)
6385 404885 : && val_signbit_p (int_mode, STORE_FLAG_VALUE)
6386 0 : && op1 == const0_rtx
6387 0 : && int_mode == GET_MODE (op0)
6388 13503970 : && (i = exact_log2 (nonzero_bits (op0, int_mode))) >= 0)
6389 : {
6390 0 : x = simplify_shift_const (NULL_RTX, ASHIFT, int_mode,
6391 : expand_compound_operation (op0),
6392 0 : GET_MODE_PRECISION (int_mode) - 1 - i);
6393 0 : if (GET_CODE (x) == AND && XEXP (x, 1) == const_true_rtx)
6394 0 : return XEXP (x, 0);
6395 : else
6396 : return x;
6397 : }
6398 :
6399 : /* If the code changed, return a whole new comparison.
6400 : We also need to avoid using SUBST in cases where
6401 : simplify_comparison has widened a comparison with a CONST_INT,
6402 : since in that case the wider CONST_INT may fail the sanity
6403 : checks in do_SUBST. */
6404 13503970 : if (new_code != code
6405 13060865 : || (CONST_INT_P (op1)
6406 7307235 : && GET_MODE (op0) != GET_MODE (XEXP (x, 0))
6407 11122 : && GET_MODE (op0) != GET_MODE (XEXP (x, 1))))
6408 452893 : return gen_rtx_fmt_ee (new_code, mode, op0, op1);
6409 :
6410 : /* Otherwise, keep this operation, but maybe change its operands.
6411 : This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
6412 13051077 : SUBST (XEXP (x, 0), op0);
6413 13051077 : SUBST (XEXP (x, 1), op1);
6414 : }
6415 : break;
6416 :
6417 13096006 : case IF_THEN_ELSE:
6418 13096006 : return simplify_if_then_else (x);
6419 :
6420 4679331 : case ZERO_EXTRACT:
6421 4679331 : case SIGN_EXTRACT:
6422 4679331 : case ZERO_EXTEND:
6423 4679331 : case SIGN_EXTEND:
6424 : /* If we are processing SET_DEST, we are done. */
6425 4679331 : if (in_dest)
6426 : return x;
6427 :
6428 4676654 : return expand_compound_operation (x);
6429 :
6430 46576201 : case SET:
6431 46576201 : return simplify_set (x);
6432 :
6433 11137100 : case AND:
6434 11137100 : case IOR:
6435 11137100 : return simplify_logical (x);
6436 :
6437 13360848 : case ASHIFT:
6438 13360848 : case LSHIFTRT:
6439 13360848 : case ASHIFTRT:
6440 13360848 : case ROTATE:
6441 13360848 : case ROTATERT:
6442 : /* If this is a shift by a constant amount, simplify it. */
6443 13360848 : if (CONST_INT_P (XEXP (x, 1)))
6444 12891278 : return simplify_shift_const (x, code, mode, XEXP (x, 0),
6445 12891278 : INTVAL (XEXP (x, 1)));
6446 :
6447 : else if (SHIFT_COUNT_TRUNCATED && !REG_P (XEXP (x, 1)))
6448 : SUBST (XEXP (x, 1),
6449 : force_to_mode (XEXP (x, 1), GET_MODE (XEXP (x, 1)),
6450 : (HOST_WIDE_INT_1U
6451 : << exact_log2 (GET_MODE_UNIT_BITSIZE
6452 : (GET_MODE (x)))) - 1, false));
6453 : break;
6454 1818375 : case VEC_SELECT:
6455 1818375 : {
6456 1818375 : rtx trueop0 = XEXP (x, 0);
6457 1818375 : mode = GET_MODE (trueop0);
6458 1818375 : rtx trueop1 = XEXP (x, 1);
6459 : /* If we select a low-part subreg, return that. */
6460 1818375 : if (vec_series_lowpart_p (GET_MODE (x), mode, trueop1))
6461 : {
6462 999 : rtx new_rtx = lowpart_subreg (GET_MODE (x), trueop0, mode);
6463 999 : if (new_rtx != NULL_RTX)
6464 : return new_rtx;
6465 : }
6466 : }
6467 :
6468 : default:
6469 : break;
6470 : }
6471 :
6472 : return x;
6473 : }
6474 :
6475 : /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6476 :
6477 : static rtx
6478 13096006 : simplify_if_then_else (rtx x)
6479 : {
6480 13096006 : machine_mode mode = GET_MODE (x);
6481 13096006 : rtx cond = XEXP (x, 0);
6482 13096006 : rtx true_rtx = XEXP (x, 1);
6483 13096006 : rtx false_rtx = XEXP (x, 2);
6484 13096006 : enum rtx_code true_code = GET_CODE (cond);
6485 13096006 : bool comparison_p = COMPARISON_P (cond);
6486 13096006 : rtx temp;
6487 13096006 : int i;
6488 13096006 : enum rtx_code false_code;
6489 13096006 : rtx reversed;
6490 13096006 : scalar_int_mode int_mode, inner_mode;
6491 :
6492 : /* Simplify storing of the truth value. */
6493 13096006 : if (comparison_p && true_rtx == const_true_rtx && false_rtx == const0_rtx)
6494 0 : return simplify_gen_relational (true_code, mode, VOIDmode,
6495 0 : XEXP (cond, 0), XEXP (cond, 1));
6496 :
6497 : /* Also when the truth value has to be reversed. */
6498 13095461 : if (comparison_p
6499 13095461 : && true_rtx == const0_rtx && false_rtx == const_true_rtx
6500 0 : && (reversed = reversed_comparison (cond, mode)))
6501 : return reversed;
6502 :
6503 : /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6504 : in it is being compared against certain values. Get the true and false
6505 : comparisons and see if that says anything about the value of each arm. */
6506 :
6507 13096006 : if (comparison_p
6508 13095461 : && ((false_code = reversed_comparison_code (cond, NULL))
6509 : != UNKNOWN)
6510 26044950 : && REG_P (XEXP (cond, 0)))
6511 : {
6512 8063644 : HOST_WIDE_INT nzb;
6513 8063644 : rtx from = XEXP (cond, 0);
6514 8063644 : rtx true_val = XEXP (cond, 1);
6515 8063644 : rtx false_val = true_val;
6516 8063644 : bool swapped = false;
6517 :
6518 : /* If FALSE_CODE is EQ, swap the codes and arms. */
6519 :
6520 8063644 : if (false_code == EQ)
6521 : {
6522 2928872 : swapped = true, true_code = EQ, false_code = NE;
6523 2928872 : std::swap (true_rtx, false_rtx);
6524 : }
6525 :
6526 8063644 : scalar_int_mode from_mode;
6527 8063644 : if (is_a <scalar_int_mode> (GET_MODE (from), &from_mode))
6528 : {
6529 : /* If we are comparing against zero and the expression being
6530 : tested has only a single bit that might be nonzero, that is
6531 : its value when it is not equal to zero. Similarly if it is
6532 : known to be -1 or 0. */
6533 6787356 : if (true_code == EQ
6534 4941716 : && true_val == const0_rtx
6535 8803984 : && pow2p_hwi (nzb = nonzero_bits (from, from_mode)))
6536 : {
6537 206962 : false_code = EQ;
6538 206962 : false_val = gen_int_mode (nzb, from_mode);
6539 : }
6540 6580394 : else if (true_code == EQ
6541 4734754 : && true_val == const0_rtx
6542 8390060 : && (num_sign_bit_copies (from, from_mode)
6543 1809666 : == GET_MODE_PRECISION (from_mode)))
6544 : {
6545 666 : false_code = EQ;
6546 666 : false_val = constm1_rtx;
6547 : }
6548 : }
6549 :
6550 : /* Now simplify an arm if we know the value of the register in the
6551 : branch and it is used in the arm. Be careful due to the potential
6552 : of locally-shared RTL. */
6553 :
6554 8063644 : if (reg_mentioned_p (from, true_rtx))
6555 300941 : true_rtx = subst (known_cond (copy_rtx (true_rtx), true_code,
6556 : from, true_val),
6557 : pc_rtx, pc_rtx, false, false, false);
6558 8063644 : if (reg_mentioned_p (from, false_rtx))
6559 104512 : false_rtx = subst (known_cond (copy_rtx (false_rtx), false_code,
6560 : from, false_val),
6561 : pc_rtx, pc_rtx, false, false, false);
6562 :
6563 13198416 : SUBST (XEXP (x, 1), swapped ? false_rtx : true_rtx);
6564 13198416 : SUBST (XEXP (x, 2), swapped ? true_rtx : false_rtx);
6565 :
6566 8063644 : true_rtx = XEXP (x, 1);
6567 8063644 : false_rtx = XEXP (x, 2);
6568 8063644 : true_code = GET_CODE (cond);
6569 : }
6570 :
6571 : /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6572 : reversed, do so to avoid needing two sets of patterns for
6573 : subtract-and-branch insns. Similarly if we have a constant in the true
6574 : arm, the false arm is the same as the first operand of the comparison, or
6575 : the false arm is more complicated than the true arm. */
6576 :
6577 13096006 : if (comparison_p
6578 13095461 : && reversed_comparison_code (cond, NULL) != UNKNOWN
6579 26044950 : && (true_rtx == pc_rtx
6580 12948944 : || (CONSTANT_P (true_rtx)
6581 10847642 : && !CONST_INT_P (false_rtx) && false_rtx != pc_rtx)
6582 12916010 : || true_rtx == const0_rtx
6583 12915779 : || (OBJECT_P (true_rtx) && !OBJECT_P (false_rtx))
6584 12877043 : || (GET_CODE (true_rtx) == SUBREG && OBJECT_P (SUBREG_REG (true_rtx))
6585 13927 : && !OBJECT_P (false_rtx))
6586 12874757 : || reg_mentioned_p (true_rtx, false_rtx)
6587 12874627 : || rtx_equal_p (false_rtx, XEXP (cond, 0))))
6588 : {
6589 101969 : SUBST (XEXP (x, 0), reversed_comparison (cond, GET_MODE (cond)));
6590 101969 : SUBST (XEXP (x, 1), false_rtx);
6591 101969 : SUBST (XEXP (x, 2), true_rtx);
6592 :
6593 101969 : std::swap (true_rtx, false_rtx);
6594 101969 : cond = XEXP (x, 0);
6595 :
6596 : /* It is possible that the conditional has been simplified out. */
6597 101969 : true_code = GET_CODE (cond);
6598 101969 : comparison_p = COMPARISON_P (cond);
6599 : }
6600 :
6601 : /* If the two arms are identical, we don't need the comparison. */
6602 :
6603 13096006 : if (rtx_equal_p (true_rtx, false_rtx) && ! side_effects_p (cond))
6604 : return true_rtx;
6605 :
6606 : /* Convert a == b ? b : a to "a". */
6607 4009024 : if (true_code == EQ && ! side_effects_p (cond)
6608 3983023 : && !HONOR_NANS (mode)
6609 3912801 : && rtx_equal_p (XEXP (cond, 0), false_rtx)
6610 13096370 : && rtx_equal_p (XEXP (cond, 1), true_rtx))
6611 : return false_rtx;
6612 4668687 : else if (true_code == NE && ! side_effects_p (cond)
6613 4627264 : && !HONOR_NANS (mode)
6614 4621673 : && rtx_equal_p (XEXP (cond, 0), true_rtx)
6615 13158857 : && rtx_equal_p (XEXP (cond, 1), false_rtx))
6616 : return true_rtx;
6617 :
6618 : /* Look for cases where we have (abs x) or (neg (abs X)). */
6619 :
6620 13096000 : if (GET_MODE_CLASS (mode) == MODE_INT
6621 1975306 : && comparison_p
6622 1975286 : && XEXP (cond, 1) == const0_rtx
6623 1513655 : && GET_CODE (false_rtx) == NEG
6624 136 : && rtx_equal_p (true_rtx, XEXP (false_rtx, 0))
6625 15 : && rtx_equal_p (true_rtx, XEXP (cond, 0))
6626 13096015 : && ! side_effects_p (true_rtx))
6627 15 : switch (true_code)
6628 : {
6629 15 : case GT:
6630 15 : case GE:
6631 15 : return simplify_gen_unary (ABS, mode, true_rtx, mode);
6632 0 : case LT:
6633 0 : case LE:
6634 0 : return
6635 0 : simplify_gen_unary (NEG, mode,
6636 : simplify_gen_unary (ABS, mode, true_rtx, mode),
6637 0 : mode);
6638 : default:
6639 : break;
6640 : }
6641 :
6642 : /* Look for MIN or MAX. */
6643 :
6644 13095985 : if ((! FLOAT_MODE_P (mode)
6645 84774 : || (flag_unsafe_math_optimizations
6646 370 : && !HONOR_NANS (mode)
6647 370 : && !HONOR_SIGNED_ZEROS (mode)))
6648 13011581 : && comparison_p
6649 13011204 : && rtx_equal_p (XEXP (cond, 0), true_rtx)
6650 105579 : && rtx_equal_p (XEXP (cond, 1), false_rtx)
6651 13122 : && ! side_effects_p (cond))
6652 13118 : switch (true_code)
6653 : {
6654 5016 : case GE:
6655 5016 : case GT:
6656 5016 : return simplify_gen_binary (SMAX, mode, true_rtx, false_rtx);
6657 4536 : case LE:
6658 4536 : case LT:
6659 4536 : return simplify_gen_binary (SMIN, mode, true_rtx, false_rtx);
6660 2675 : case GEU:
6661 2675 : case GTU:
6662 2675 : return simplify_gen_binary (UMAX, mode, true_rtx, false_rtx);
6663 891 : case LEU:
6664 891 : case LTU:
6665 891 : return simplify_gen_binary (UMIN, mode, true_rtx, false_rtx);
6666 : default:
6667 : break;
6668 : }
6669 :
6670 : /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6671 : second operand is zero, this can be done as (OP Z (mult COND C2)) where
6672 : C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6673 : SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6674 : We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6675 : neither 1 or -1, but it isn't worth checking for. */
6676 :
6677 13082867 : if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
6678 : && comparison_p
6679 14963878 : && is_int_mode (mode, &int_mode)
6680 15045042 : && ! side_effects_p (x))
6681 : {
6682 1958206 : rtx t = make_compound_operation (true_rtx, SET);
6683 1958206 : rtx f = make_compound_operation (false_rtx, SET);
6684 1958206 : rtx cond_op0 = XEXP (cond, 0);
6685 1958206 : rtx cond_op1 = XEXP (cond, 1);
6686 1958206 : enum rtx_code op = UNKNOWN, extend_op = UNKNOWN;
6687 1958206 : scalar_int_mode m = int_mode;
6688 1958206 : rtx z = 0, c1 = NULL_RTX;
6689 :
6690 1958206 : if ((GET_CODE (t) == PLUS || GET_CODE (t) == MINUS
6691 : || GET_CODE (t) == IOR || GET_CODE (t) == XOR
6692 : || GET_CODE (t) == ASHIFT
6693 : || GET_CODE (t) == LSHIFTRT || GET_CODE (t) == ASHIFTRT)
6694 199889 : && rtx_equal_p (XEXP (t, 0), f))
6695 72798 : c1 = XEXP (t, 1), op = GET_CODE (t), z = f;
6696 :
6697 : /* If an identity-zero op is commutative, check whether there
6698 : would be a match if we swapped the operands. */
6699 1820834 : else if ((GET_CODE (t) == PLUS || GET_CODE (t) == IOR
6700 1807179 : || GET_CODE (t) == XOR)
6701 1899710 : && rtx_equal_p (XEXP (t, 1), f))
6702 8366 : c1 = XEXP (t, 0), op = GET_CODE (t), z = f;
6703 1877042 : else if (GET_CODE (t) == SIGN_EXTEND
6704 1842 : && is_a <scalar_int_mode> (GET_MODE (XEXP (t, 0)), &inner_mode)
6705 1842 : && (GET_CODE (XEXP (t, 0)) == PLUS
6706 1842 : || GET_CODE (XEXP (t, 0)) == MINUS
6707 : || GET_CODE (XEXP (t, 0)) == IOR
6708 : || GET_CODE (XEXP (t, 0)) == XOR
6709 : || GET_CODE (XEXP (t, 0)) == ASHIFT
6710 : || GET_CODE (XEXP (t, 0)) == LSHIFTRT
6711 : || GET_CODE (XEXP (t, 0)) == ASHIFTRT)
6712 80 : && GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
6713 54 : && subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
6714 54 : && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
6715 1877042 : && (num_sign_bit_copies (f, GET_MODE (f))
6716 0 : > (unsigned int)
6717 0 : (GET_MODE_PRECISION (int_mode)
6718 0 : - GET_MODE_PRECISION (inner_mode))))
6719 : {
6720 0 : c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
6721 0 : extend_op = SIGN_EXTEND;
6722 0 : m = inner_mode;
6723 : }
6724 1877042 : else if (GET_CODE (t) == SIGN_EXTEND
6725 1842 : && is_a <scalar_int_mode> (GET_MODE (XEXP (t, 0)), &inner_mode)
6726 1842 : && (GET_CODE (XEXP (t, 0)) == PLUS
6727 1768 : || GET_CODE (XEXP (t, 0)) == IOR
6728 1764 : || GET_CODE (XEXP (t, 0)) == XOR)
6729 78 : && GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
6730 4 : && subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
6731 4 : && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
6732 1877046 : && (num_sign_bit_copies (f, GET_MODE (f))
6733 4 : > (unsigned int)
6734 4 : (GET_MODE_PRECISION (int_mode)
6735 4 : - GET_MODE_PRECISION (inner_mode))))
6736 : {
6737 0 : c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
6738 0 : extend_op = SIGN_EXTEND;
6739 0 : m = inner_mode;
6740 : }
6741 1877042 : else if (GET_CODE (t) == ZERO_EXTEND
6742 4393 : && is_a <scalar_int_mode> (GET_MODE (XEXP (t, 0)), &inner_mode)
6743 4393 : && (GET_CODE (XEXP (t, 0)) == PLUS
6744 4393 : || GET_CODE (XEXP (t, 0)) == MINUS
6745 : || GET_CODE (XEXP (t, 0)) == IOR
6746 : || GET_CODE (XEXP (t, 0)) == XOR
6747 : || GET_CODE (XEXP (t, 0)) == ASHIFT
6748 : || GET_CODE (XEXP (t, 0)) == LSHIFTRT
6749 : || GET_CODE (XEXP (t, 0)) == ASHIFTRT)
6750 973 : && GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
6751 103 : && HWI_COMPUTABLE_MODE_P (int_mode)
6752 103 : && subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
6753 103 : && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
6754 1877042 : && ((nonzero_bits (f, GET_MODE (f))
6755 0 : & ~GET_MODE_MASK (inner_mode))
6756 : == 0))
6757 : {
6758 0 : c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
6759 0 : extend_op = ZERO_EXTEND;
6760 0 : m = inner_mode;
6761 : }
6762 1877042 : else if (GET_CODE (t) == ZERO_EXTEND
6763 4393 : && is_a <scalar_int_mode> (GET_MODE (XEXP (t, 0)), &inner_mode)
6764 4393 : && (GET_CODE (XEXP (t, 0)) == PLUS
6765 3992 : || GET_CODE (XEXP (t, 0)) == IOR
6766 3992 : || GET_CODE (XEXP (t, 0)) == XOR)
6767 401 : && GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
6768 16 : && HWI_COMPUTABLE_MODE_P (int_mode)
6769 16 : && subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
6770 16 : && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
6771 1877042 : && ((nonzero_bits (f, GET_MODE (f))
6772 0 : & ~GET_MODE_MASK (inner_mode))
6773 : == 0))
6774 : {
6775 0 : c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
6776 0 : extend_op = ZERO_EXTEND;
6777 0 : m = inner_mode;
6778 : }
6779 :
6780 81164 : if (z)
6781 : {
6782 81164 : machine_mode cm = m;
6783 81164 : if ((op == ASHIFT || op == LSHIFTRT || op == ASHIFTRT)
6784 2090 : && GET_MODE (c1) != VOIDmode)
6785 1590 : cm = GET_MODE (c1);
6786 81164 : temp = subst (simplify_gen_relational (true_code, cm, VOIDmode,
6787 : cond_op0, cond_op1),
6788 : pc_rtx, pc_rtx, false, false, false);
6789 81164 : temp = simplify_gen_binary (MULT, cm, temp,
6790 : simplify_gen_binary (MULT, cm, c1,
6791 : const_true_rtx));
6792 81164 : temp = subst (temp, pc_rtx, pc_rtx, false, false, false);
6793 81164 : temp = simplify_gen_binary (op, m, gen_lowpart (m, z), temp);
6794 :
6795 81164 : if (extend_op != UNKNOWN)
6796 0 : temp = simplify_gen_unary (extend_op, int_mode, temp, m);
6797 :
6798 81164 : return temp;
6799 : }
6800 : }
6801 :
6802 : /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6803 : 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6804 : negation of a single bit, we can convert this operation to a shift. We
6805 : can actually do this more generally, but it doesn't seem worth it. */
6806 :
6807 13001703 : if (true_code == NE
6808 13001702 : && is_a <scalar_int_mode> (mode, &int_mode)
6809 427644 : && XEXP (cond, 1) == const0_rtx
6810 299061 : && false_rtx == const0_rtx
6811 46443 : && CONST_INT_P (true_rtx)
6812 13002133 : && ((nonzero_bits (XEXP (cond, 0), int_mode) == 1
6813 2 : && (i = exact_log2 (UINTVAL (true_rtx))) >= 0)
6814 429 : || ((num_sign_bit_copies (XEXP (cond, 0), int_mode)
6815 429 : == GET_MODE_PRECISION (int_mode))
6816 0 : && (i = exact_log2 (-UINTVAL (true_rtx))) >= 0)))
6817 1 : return
6818 1 : simplify_shift_const (NULL_RTX, ASHIFT, int_mode,
6819 2 : gen_lowpart (int_mode, XEXP (cond, 0)), i);
6820 :
6821 : /* (IF_THEN_ELSE (NE A 0) C1 0) is A or a zero-extend of A if the only
6822 : non-zero bit in A is C1. */
6823 4650909 : if (true_code == NE && XEXP (cond, 1) == const0_rtx
6824 2057100 : && false_rtx == const0_rtx && CONST_INT_P (true_rtx)
6825 13096435 : && is_a <scalar_int_mode> (mode, &int_mode)
6826 429 : && is_a <scalar_int_mode> (GET_MODE (XEXP (cond, 0)), &inner_mode)
6827 41 : && (UINTVAL (true_rtx) & GET_MODE_MASK (int_mode))
6828 41 : == nonzero_bits (XEXP (cond, 0), inner_mode)
6829 13001702 : && (i = exact_log2 (UINTVAL (true_rtx) & GET_MODE_MASK (int_mode))) >= 0)
6830 : {
6831 0 : rtx val = XEXP (cond, 0);
6832 0 : if (inner_mode == int_mode)
6833 : return val;
6834 0 : else if (GET_MODE_PRECISION (inner_mode) < GET_MODE_PRECISION (int_mode))
6835 0 : return simplify_gen_unary (ZERO_EXTEND, int_mode, val, inner_mode);
6836 : }
6837 :
6838 : return x;
6839 : }
6840 :
6841 : /* Simplify X, a SET expression. Return the new expression. */
6842 :
6843 : static rtx
6844 46576201 : simplify_set (rtx x)
6845 : {
6846 46576201 : rtx src = SET_SRC (x);
6847 46576201 : rtx dest = SET_DEST (x);
6848 104470005 : machine_mode mode
6849 46576201 : = GET_MODE (src) != VOIDmode ? GET_MODE (src) : GET_MODE (dest);
6850 46576201 : rtx_insn *other_insn;
6851 46576201 : rtx *cc_use;
6852 46576201 : scalar_int_mode int_mode;
6853 :
6854 : /* (set (pc) (return)) gets written as (return). */
6855 46576201 : if (GET_CODE (dest) == PC && ANY_RETURN_P (src))
6856 : return src;
6857 :
6858 : /* Now that we know for sure which bits of SRC we are using, see if we can
6859 : simplify the expression for the object knowing that we only need the
6860 : low-order bits. */
6861 :
6862 46576201 : if (GET_MODE_CLASS (mode) == MODE_INT && HWI_COMPUTABLE_MODE_P (mode))
6863 : {
6864 20438219 : src = force_to_mode (src, mode, HOST_WIDE_INT_M1U, false);
6865 20438219 : SUBST (SET_SRC (x), src);
6866 : }
6867 :
6868 : /* If the source is a COMPARE, look for the use of the comparison result
6869 : and try to simplify it unless we already have used undobuf.other_insn. */
6870 39969090 : if ((GET_MODE_CLASS (mode) == MODE_CC || GET_CODE (src) == COMPARE)
6871 6607111 : && (cc_use = find_single_use (dest, subst_insn, &other_insn)) != 0
6872 6025137 : && (undobuf.other_insn == 0 || other_insn == undobuf.other_insn)
6873 6025137 : && COMPARISON_P (*cc_use)
6874 52600854 : && rtx_equal_p (XEXP (*cc_use, 0), dest))
6875 : {
6876 6023117 : enum rtx_code old_code = GET_CODE (*cc_use);
6877 6023117 : enum rtx_code new_code;
6878 6023117 : rtx op0, op1, tmp;
6879 6023117 : bool other_changed = false;
6880 6023117 : rtx inner_compare = NULL_RTX;
6881 6023117 : machine_mode compare_mode = GET_MODE (dest);
6882 :
6883 6023117 : if (GET_CODE (src) == COMPARE)
6884 : {
6885 5592853 : op0 = XEXP (src, 0), op1 = XEXP (src, 1);
6886 5592853 : if (GET_CODE (op0) == COMPARE && op1 == const0_rtx)
6887 : {
6888 0 : inner_compare = op0;
6889 0 : op0 = XEXP (inner_compare, 0), op1 = XEXP (inner_compare, 1);
6890 : }
6891 : }
6892 : else
6893 430264 : op0 = src, op1 = CONST0_RTX (GET_MODE (src));
6894 :
6895 6023117 : tmp = simplify_relational_operation (old_code, compare_mode, VOIDmode,
6896 : op0, op1);
6897 6023117 : if (!tmp)
6898 : new_code = old_code;
6899 459647 : else if (!CONSTANT_P (tmp))
6900 : {
6901 454539 : new_code = GET_CODE (tmp);
6902 454539 : op0 = XEXP (tmp, 0);
6903 454539 : op1 = XEXP (tmp, 1);
6904 : }
6905 : else
6906 : {
6907 5108 : rtx pat = PATTERN (other_insn);
6908 5108 : undobuf.other_insn = other_insn;
6909 5108 : SUBST (*cc_use, tmp);
6910 :
6911 : /* Attempt to simplify CC user. */
6912 5108 : if (GET_CODE (pat) == SET)
6913 : {
6914 4610 : rtx new_rtx = simplify_rtx (SET_SRC (pat));
6915 4610 : if (new_rtx != NULL_RTX)
6916 4020 : SUBST (SET_SRC (pat), new_rtx);
6917 : }
6918 :
6919 : /* Convert X into a no-op move. */
6920 5108 : SUBST (SET_DEST (x), pc_rtx);
6921 5108 : SUBST (SET_SRC (x), pc_rtx);
6922 5108 : return x;
6923 : }
6924 :
6925 : /* Simplify our comparison, if possible. */
6926 6018009 : new_code = simplify_comparison (new_code, &op0, &op1);
6927 :
6928 : #ifdef SELECT_CC_MODE
6929 : /* If this machine has CC modes other than CCmode, check to see if we
6930 : need to use a different CC mode here. */
6931 6018009 : if (GET_MODE_CLASS (GET_MODE (op0)) == MODE_CC)
6932 634670 : compare_mode = GET_MODE (op0);
6933 5383339 : else if (inner_compare
6934 0 : && GET_MODE_CLASS (GET_MODE (inner_compare)) == MODE_CC
6935 0 : && new_code == old_code
6936 0 : && op0 == XEXP (inner_compare, 0)
6937 0 : && op1 == XEXP (inner_compare, 1))
6938 0 : compare_mode = GET_MODE (inner_compare);
6939 : else
6940 5383339 : compare_mode = SELECT_CC_MODE (new_code, op0, op1);
6941 :
6942 : /* If the mode changed, we have to change SET_DEST, the mode in the
6943 : compare, and the mode in the place SET_DEST is used. If SET_DEST is
6944 : a hard register, just build new versions with the proper mode. If it
6945 : is a pseudo, we lose unless it is only time we set the pseudo, in
6946 : which case we can safely change its mode. */
6947 6018009 : if (compare_mode != GET_MODE (dest))
6948 : {
6949 213905 : if (can_change_dest_mode (dest, 0, compare_mode))
6950 : {
6951 213905 : unsigned int regno = REGNO (dest);
6952 213905 : rtx new_dest;
6953 :
6954 213905 : if (regno < FIRST_PSEUDO_REGISTER)
6955 213905 : new_dest = gen_rtx_REG (compare_mode, regno);
6956 : else
6957 : {
6958 0 : subst_mode (regno, compare_mode);
6959 0 : new_dest = regno_reg_rtx[regno];
6960 : }
6961 :
6962 213905 : SUBST (SET_DEST (x), new_dest);
6963 213905 : SUBST (XEXP (*cc_use, 0), new_dest);
6964 213905 : other_changed = true;
6965 :
6966 213905 : dest = new_dest;
6967 : }
6968 : }
6969 : #endif /* SELECT_CC_MODE */
6970 :
6971 : /* If the code changed, we have to build a new comparison in
6972 : undobuf.other_insn. */
6973 6018009 : if (new_code != old_code)
6974 : {
6975 598400 : bool other_changed_previously = other_changed;
6976 598400 : unsigned HOST_WIDE_INT mask;
6977 598400 : rtx old_cc_use = *cc_use;
6978 :
6979 598400 : SUBST (*cc_use, gen_rtx_fmt_ee (new_code, GET_MODE (*cc_use),
6980 : dest, const0_rtx));
6981 598400 : other_changed = true;
6982 :
6983 : /* If the only change we made was to change an EQ into an NE or
6984 : vice versa, OP0 has only one bit that might be nonzero, and OP1
6985 : is zero, check if changing the user of the condition code will
6986 : produce a valid insn. If it won't, we can keep the original code
6987 : in that insn by surrounding our operation with an XOR. */
6988 :
6989 598400 : if (((old_code == NE && new_code == EQ)
6990 569401 : || (old_code == EQ && new_code == NE))
6991 63020 : && ! other_changed_previously && op1 == const0_rtx
6992 60380 : && HWI_COMPUTABLE_MODE_P (GET_MODE (op0))
6993 606439 : && pow2p_hwi (mask = nonzero_bits (op0, GET_MODE (op0))))
6994 : {
6995 8028 : rtx pat = PATTERN (other_insn), note = 0;
6996 :
6997 8028 : if ((recog_for_combine (&pat, other_insn, ¬e) < 0
6998 8028 : && ! check_asm_operands (pat)))
6999 : {
7000 4 : *cc_use = old_cc_use;
7001 4 : other_changed = false;
7002 :
7003 4 : op0 = simplify_gen_binary (XOR, GET_MODE (op0), op0,
7004 4 : gen_int_mode (mask,
7005 4 : GET_MODE (op0)));
7006 : }
7007 : }
7008 : }
7009 :
7010 5427637 : if (other_changed)
7011 620686 : undobuf.other_insn = other_insn;
7012 :
7013 : /* Don't generate a compare of a CC with 0, just use that CC. */
7014 6018009 : if (GET_MODE (op0) == compare_mode && op1 == const0_rtx)
7015 : {
7016 634670 : SUBST (SET_SRC (x), op0);
7017 634670 : src = SET_SRC (x);
7018 : }
7019 : /* Otherwise, if we didn't previously have the same COMPARE we
7020 : want, create it from scratch. */
7021 5383339 : else if (GET_CODE (src) != COMPARE || GET_MODE (src) != compare_mode
7022 5257319 : || XEXP (src, 0) != op0 || XEXP (src, 1) != op1)
7023 : {
7024 1417428 : SUBST (SET_SRC (x), gen_rtx_COMPARE (compare_mode, op0, op1));
7025 1417428 : src = SET_SRC (x);
7026 : }
7027 : }
7028 : else
7029 : {
7030 : /* Get SET_SRC in a form where we have placed back any
7031 : compound expressions. Then do the checks below. */
7032 40553084 : src = make_compound_operation (src, SET);
7033 40553084 : SUBST (SET_SRC (x), src);
7034 : }
7035 :
7036 : /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
7037 : and X being a REG or (subreg (reg)), we may be able to convert this to
7038 : (set (subreg:m2 x) (op)).
7039 :
7040 : We can always do this if M1 is narrower than M2 because that means that
7041 : we only care about the low bits of the result.
7042 :
7043 : However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
7044 : perform a narrower operation than requested since the high-order bits will
7045 : be undefined. On machine where it is defined, this transformation is safe
7046 : as long as M1 and M2 have the same number of words. */
7047 :
7048 403232 : if (GET_CODE (src) == SUBREG && subreg_lowpart_p (src)
7049 387412 : && !OBJECT_P (SUBREG_REG (src))
7050 : && (known_equal_after_align_up
7051 244066 : (GET_MODE_SIZE (GET_MODE (src)),
7052 488132 : GET_MODE_SIZE (GET_MODE (SUBREG_REG (src))),
7053 244066 : UNITS_PER_WORD))
7054 215819 : && (WORD_REGISTER_OPERATIONS || !paradoxical_subreg_p (src))
7055 211901 : && ! (REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER
7056 221 : && !REG_CAN_CHANGE_MODE_P (REGNO (dest),
7057 : GET_MODE (SUBREG_REG (src)),
7058 : GET_MODE (src)))
7059 46782773 : && (REG_P (dest)
7060 98367 : || (GET_CODE (dest) == SUBREG
7061 272 : && REG_P (SUBREG_REG (dest)))))
7062 : {
7063 113585 : SUBST (SET_DEST (x),
7064 : gen_lowpart (GET_MODE (SUBREG_REG (src)),
7065 : dest));
7066 113585 : SUBST (SET_SRC (x), SUBREG_REG (src));
7067 :
7068 113585 : src = SET_SRC (x), dest = SET_DEST (x);
7069 : }
7070 :
7071 : /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
7072 : would require a paradoxical subreg. Replace the subreg with a
7073 : zero_extend to avoid the reload that would otherwise be required.
7074 : Don't do this unless we have a scalar integer mode, otherwise the
7075 : transformation is incorrect. */
7076 :
7077 46571093 : enum rtx_code extend_op;
7078 46571093 : if (paradoxical_subreg_p (src)
7079 : && MEM_P (SUBREG_REG (src))
7080 : && SCALAR_INT_MODE_P (GET_MODE (src))
7081 : && (extend_op = load_extend_op (GET_MODE (SUBREG_REG (src)))) != UNKNOWN)
7082 : {
7083 : SUBST (SET_SRC (x),
7084 : gen_rtx_fmt_e (extend_op, GET_MODE (src), SUBREG_REG (src)));
7085 :
7086 : src = SET_SRC (x);
7087 : }
7088 :
7089 : /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
7090 : are comparing an item known to be 0 or -1 against 0, use a logical
7091 : operation instead. Check for one of the arms being an IOR of the other
7092 : arm with some value. We compute three terms to be IOR'ed together. In
7093 : practice, at most two will be nonzero. Then we do the IOR's. */
7094 :
7095 46571093 : if (GET_CODE (dest) != PC
7096 35649336 : && GET_CODE (src) == IF_THEN_ELSE
7097 1169558 : && is_int_mode (GET_MODE (src), &int_mode)
7098 1074682 : && (GET_CODE (XEXP (src, 0)) == EQ || GET_CODE (XEXP (src, 0)) == NE)
7099 463957 : && XEXP (XEXP (src, 0), 1) == const0_rtx
7100 317989 : && int_mode == GET_MODE (XEXP (XEXP (src, 0), 0))
7101 98624 : && (!HAVE_conditional_move
7102 98624 : || ! can_conditionally_move_p (int_mode))
7103 0 : && (num_sign_bit_copies (XEXP (XEXP (src, 0), 0), int_mode)
7104 0 : == GET_MODE_PRECISION (int_mode))
7105 46571093 : && ! side_effects_p (src))
7106 : {
7107 0 : rtx true_rtx = (GET_CODE (XEXP (src, 0)) == NE
7108 0 : ? XEXP (src, 1) : XEXP (src, 2));
7109 0 : rtx false_rtx = (GET_CODE (XEXP (src, 0)) == NE
7110 0 : ? XEXP (src, 2) : XEXP (src, 1));
7111 0 : rtx term1 = const0_rtx, term2, term3;
7112 :
7113 0 : if (GET_CODE (true_rtx) == IOR
7114 0 : && rtx_equal_p (XEXP (true_rtx, 0), false_rtx))
7115 0 : term1 = false_rtx, true_rtx = XEXP (true_rtx, 1), false_rtx = const0_rtx;
7116 0 : else if (GET_CODE (true_rtx) == IOR
7117 0 : && rtx_equal_p (XEXP (true_rtx, 1), false_rtx))
7118 0 : term1 = false_rtx, true_rtx = XEXP (true_rtx, 0), false_rtx = const0_rtx;
7119 0 : else if (GET_CODE (false_rtx) == IOR
7120 0 : && rtx_equal_p (XEXP (false_rtx, 0), true_rtx))
7121 0 : term1 = true_rtx, false_rtx = XEXP (false_rtx, 1), true_rtx = const0_rtx;
7122 0 : else if (GET_CODE (false_rtx) == IOR
7123 0 : && rtx_equal_p (XEXP (false_rtx, 1), true_rtx))
7124 0 : term1 = true_rtx, false_rtx = XEXP (false_rtx, 0), true_rtx = const0_rtx;
7125 :
7126 0 : term2 = simplify_gen_binary (AND, int_mode,
7127 0 : XEXP (XEXP (src, 0), 0), true_rtx);
7128 0 : term3 = simplify_gen_binary (AND, int_mode,
7129 : simplify_gen_unary (NOT, int_mode,
7130 0 : XEXP (XEXP (src, 0), 0),
7131 : int_mode),
7132 : false_rtx);
7133 :
7134 0 : SUBST (SET_SRC (x),
7135 : simplify_gen_binary (IOR, int_mode,
7136 : simplify_gen_binary (IOR, int_mode,
7137 : term1, term2),
7138 : term3));
7139 :
7140 0 : src = SET_SRC (x);
7141 : }
7142 :
7143 : /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
7144 : whole thing fail. */
7145 46571093 : if (GET_CODE (src) == CLOBBER && XEXP (src, 0) == const0_rtx)
7146 : return src;
7147 46571070 : else if (GET_CODE (dest) == CLOBBER && XEXP (dest, 0) == const0_rtx)
7148 : return dest;
7149 : else
7150 : /* Convert this into a field assignment operation, if possible. */
7151 46571070 : return make_field_assignment (x);
7152 : }
7153 :
7154 : /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
7155 : result. */
7156 :
7157 : static rtx
7158 11137100 : simplify_logical (rtx x)
7159 : {
7160 11137100 : rtx op0 = XEXP (x, 0);
7161 11137100 : rtx op1 = XEXP (x, 1);
7162 11137100 : scalar_int_mode mode;
7163 :
7164 11137100 : switch (GET_CODE (x))
7165 : {
7166 6844488 : case AND:
7167 : /* We can call simplify_and_const_int only if we don't lose
7168 : any (sign) bits when converting INTVAL (op1) to
7169 : "unsigned HOST_WIDE_INT". */
7170 6844488 : if (is_a <scalar_int_mode> (GET_MODE (x), &mode)
7171 6342424 : && CONST_INT_P (op1)
7172 4983039 : && (HWI_COMPUTABLE_MODE_P (mode)
7173 8806 : || INTVAL (op1) > 0))
7174 : {
7175 4979787 : x = simplify_and_const_int (x, mode, op0, INTVAL (op1));
7176 4979787 : if (GET_CODE (x) != AND)
7177 : return x;
7178 :
7179 4947827 : op0 = XEXP (x, 0);
7180 4947827 : op1 = XEXP (x, 1);
7181 : }
7182 :
7183 : /* If we have any of (and (ior A B) C) or (and (xor A B) C),
7184 : apply the distributive law and then the inverse distributive
7185 : law to see if things simplify. */
7186 6812528 : if (GET_CODE (op0) == IOR || GET_CODE (op0) == XOR)
7187 : {
7188 122951 : rtx result = distribute_and_simplify_rtx (x, 0);
7189 122951 : if (result)
7190 : return result;
7191 : }
7192 6799436 : if (GET_CODE (op1) == IOR || GET_CODE (op1) == XOR)
7193 : {
7194 1809 : rtx result = distribute_and_simplify_rtx (x, 1);
7195 1809 : if (result)
7196 : return result;
7197 : }
7198 : break;
7199 :
7200 4292612 : case IOR:
7201 : /* If we have (ior (and A B) C), apply the distributive law and then
7202 : the inverse distributive law to see if things simplify. */
7203 :
7204 4292612 : if (GET_CODE (op0) == AND)
7205 : {
7206 1213997 : rtx result = distribute_and_simplify_rtx (x, 0);
7207 1213997 : if (result)
7208 : return result;
7209 : }
7210 :
7211 4289939 : if (GET_CODE (op1) == AND)
7212 : {
7213 66752 : rtx result = distribute_and_simplify_rtx (x, 1);
7214 66752 : if (result)
7215 : return result;
7216 : }
7217 : break;
7218 :
7219 0 : default:
7220 0 : gcc_unreachable ();
7221 : }
7222 :
7223 : return x;
7224 : }
7225 :
7226 : /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
7227 : operations" because they can be replaced with two more basic operations.
7228 : ZERO_EXTEND is also considered "compound" because it can be replaced with
7229 : an AND operation, which is simpler, though only one operation.
7230 :
7231 : The function expand_compound_operation is called with an rtx expression
7232 : and will convert it to the appropriate shifts and AND operations,
7233 : simplifying at each stage.
7234 :
7235 : The function make_compound_operation is called to convert an expression
7236 : consisting of shifts and ANDs into the equivalent compound expression.
7237 : It is the inverse of this function, loosely speaking. */
7238 :
7239 : static rtx
7240 15645830 : expand_compound_operation (rtx x)
7241 : {
7242 15645830 : unsigned HOST_WIDE_INT pos = 0, len;
7243 15645830 : bool unsignedp = false;
7244 15645830 : unsigned int modewidth;
7245 15645830 : rtx tem;
7246 15645830 : scalar_int_mode inner_mode;
7247 :
7248 15645830 : switch (GET_CODE (x))
7249 : {
7250 4454410 : case ZERO_EXTEND:
7251 4454410 : unsignedp = true;
7252 : /* FALLTHRU */
7253 5802542 : case SIGN_EXTEND:
7254 : /* We can't necessarily use a const_int for a multiword mode;
7255 : it depends on implicitly extending the value.
7256 : Since we don't know the right way to extend it,
7257 : we can't tell whether the implicit way is right.
7258 :
7259 : Even for a mode that is no wider than a const_int,
7260 : we can't win, because we need to sign extend one of its bits through
7261 : the rest of it, and we don't know which bit. */
7262 5802542 : if (CONST_INT_P (XEXP (x, 0)))
7263 : return x;
7264 :
7265 : /* Reject modes that aren't scalar integers because turning vector
7266 : or complex modes into shifts causes problems. */
7267 5802542 : if (!is_a <scalar_int_mode> (GET_MODE (XEXP (x, 0)), &inner_mode))
7268 : return x;
7269 :
7270 : /* Return if (subreg:MODE FROM 0) is not a safe replacement for
7271 : (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
7272 : because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
7273 : reloaded. If not for that, MEM's would very rarely be safe.
7274 :
7275 : Reject modes bigger than a word, because we might not be able
7276 : to reference a two-register group starting with an arbitrary register
7277 : (and currently gen_lowpart might crash for a SUBREG). */
7278 :
7279 11723075 : if (GET_MODE_SIZE (inner_mode) > UNITS_PER_WORD)
7280 : return x;
7281 :
7282 5457307 : len = GET_MODE_PRECISION (inner_mode);
7283 : /* If the inner object has VOIDmode (the only way this can happen
7284 : is if it is an ASM_OPERANDS), we can't do anything since we don't
7285 : know how much masking to do. */
7286 5457307 : if (len == 0)
7287 : return x;
7288 :
7289 : break;
7290 :
7291 944191 : case ZERO_EXTRACT:
7292 944191 : unsignedp = true;
7293 :
7294 : /* fall through */
7295 :
7296 967366 : case SIGN_EXTRACT:
7297 : /* If the operand is a CLOBBER, just return it. */
7298 967366 : if (GET_CODE (XEXP (x, 0)) == CLOBBER)
7299 : return XEXP (x, 0);
7300 :
7301 967366 : if (!CONST_INT_P (XEXP (x, 1))
7302 967231 : || !CONST_INT_P (XEXP (x, 2)))
7303 : return x;
7304 :
7305 : /* Reject modes that aren't scalar integers because turning vector
7306 : or complex modes into shifts causes problems. */
7307 13118072 : if (!is_a <scalar_int_mode> (GET_MODE (XEXP (x, 0)), &inner_mode))
7308 : return x;
7309 :
7310 889982 : len = INTVAL (XEXP (x, 1));
7311 889982 : pos = INTVAL (XEXP (x, 2));
7312 :
7313 : /* This should stay within the object being extracted, fail otherwise. */
7314 889982 : if (len + pos > GET_MODE_PRECISION (inner_mode))
7315 : return x;
7316 :
7317 : if (BITS_BIG_ENDIAN)
7318 : pos = GET_MODE_PRECISION (inner_mode) - len - pos;
7319 :
7320 : break;
7321 :
7322 : default:
7323 : return x;
7324 : }
7325 :
7326 : /* We've rejected non-scalar operations by now. */
7327 6347240 : scalar_int_mode mode = as_a <scalar_int_mode> (GET_MODE (x));
7328 :
7329 : /* Convert sign extension to zero extension, if we know that the high
7330 : bit is not set, as this is easier to optimize. It will be converted
7331 : back to cheaper alternative in make_extraction. */
7332 6347240 : if (GET_CODE (x) == SIGN_EXTEND
7333 1194977 : && HWI_COMPUTABLE_MODE_P (mode)
7334 7427182 : && ((nonzero_bits (XEXP (x, 0), inner_mode)
7335 1079942 : & ~(((unsigned HOST_WIDE_INT) GET_MODE_MASK (inner_mode)) >> 1))
7336 : == 0))
7337 : {
7338 580 : rtx temp = gen_rtx_ZERO_EXTEND (mode, XEXP (x, 0));
7339 580 : rtx temp2 = expand_compound_operation (temp);
7340 :
7341 : /* Make sure this is a profitable operation. */
7342 580 : if (set_src_cost (x, mode, optimize_this_for_speed_p)
7343 580 : > set_src_cost (temp2, mode, optimize_this_for_speed_p))
7344 : return temp2;
7345 566 : else if (set_src_cost (x, mode, optimize_this_for_speed_p)
7346 566 : > set_src_cost (temp, mode, optimize_this_for_speed_p))
7347 : return temp;
7348 : else
7349 : return x;
7350 : }
7351 :
7352 : /* We can optimize some special cases of ZERO_EXTEND. */
7353 6346660 : if (GET_CODE (x) == ZERO_EXTEND)
7354 : {
7355 : /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
7356 : know that the last value didn't have any inappropriate bits
7357 : set. */
7358 4262330 : if (GET_CODE (XEXP (x, 0)) == TRUNCATE
7359 188 : && GET_MODE (XEXP (XEXP (x, 0), 0)) == mode
7360 188 : && HWI_COMPUTABLE_MODE_P (mode)
7361 4262518 : && (nonzero_bits (XEXP (XEXP (x, 0), 0), mode)
7362 188 : & ~GET_MODE_MASK (inner_mode)) == 0)
7363 36 : return XEXP (XEXP (x, 0), 0);
7364 :
7365 : /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7366 4262294 : if (GET_CODE (XEXP (x, 0)) == SUBREG
7367 638024 : && GET_MODE (SUBREG_REG (XEXP (x, 0))) == mode
7368 602779 : && subreg_lowpart_p (XEXP (x, 0))
7369 257302 : && HWI_COMPUTABLE_MODE_P (mode)
7370 4496966 : && (nonzero_bits (SUBREG_REG (XEXP (x, 0)), mode)
7371 234672 : & ~GET_MODE_MASK (inner_mode)) == 0)
7372 88 : return SUBREG_REG (XEXP (x, 0));
7373 :
7374 : /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
7375 : is a comparison and STORE_FLAG_VALUE permits. This is like
7376 : the first case, but it works even when MODE is larger
7377 : than HOST_WIDE_INT. */
7378 4262206 : if (GET_CODE (XEXP (x, 0)) == TRUNCATE
7379 152 : && GET_MODE (XEXP (XEXP (x, 0), 0)) == mode
7380 152 : && COMPARISON_P (XEXP (XEXP (x, 0), 0))
7381 0 : && GET_MODE_PRECISION (inner_mode) <= HOST_BITS_PER_WIDE_INT
7382 4262206 : && (STORE_FLAG_VALUE & ~GET_MODE_MASK (inner_mode)) == 0)
7383 : return XEXP (XEXP (x, 0), 0);
7384 :
7385 : /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7386 4262206 : if (GET_CODE (XEXP (x, 0)) == SUBREG
7387 637936 : && GET_MODE (SUBREG_REG (XEXP (x, 0))) == mode
7388 602691 : && subreg_lowpart_p (XEXP (x, 0))
7389 257214 : && COMPARISON_P (SUBREG_REG (XEXP (x, 0)))
7390 0 : && GET_MODE_PRECISION (inner_mode) <= HOST_BITS_PER_WIDE_INT
7391 4262206 : && (STORE_FLAG_VALUE & ~GET_MODE_MASK (inner_mode)) == 0)
7392 : return SUBREG_REG (XEXP (x, 0));
7393 :
7394 : }
7395 :
7396 : /* If we reach here, we want to return a pair of shifts. The inner
7397 : shift is a left shift of BITSIZE - POS - LEN bits. The outer
7398 : shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
7399 : logical depending on the value of UNSIGNEDP.
7400 :
7401 : If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
7402 : converted into an AND of a shift.
7403 :
7404 : We must check for the case where the left shift would have a negative
7405 : count. This can happen in a case like (x >> 31) & 255 on machines
7406 : that can't shift by a constant. On those machines, we would first
7407 : combine the shift with the AND to produce a variable-position
7408 : extraction. Then the constant of 31 would be substituted in
7409 : to produce such a position. */
7410 :
7411 6346536 : modewidth = GET_MODE_PRECISION (mode);
7412 6346536 : if (modewidth >= pos + len)
7413 : {
7414 6346535 : tem = gen_lowpart (mode, XEXP (x, 0));
7415 6346535 : if (!tem || GET_CODE (tem) == CLOBBER)
7416 : return x;
7417 6834194 : tem = simplify_shift_const (NULL_RTX, ASHIFT, mode,
7418 3417097 : tem, modewidth - pos - len);
7419 3417097 : tem = simplify_shift_const (NULL_RTX, unsignedp ? LSHIFTRT : ASHIFTRT,
7420 3417097 : mode, tem, modewidth - len);
7421 : }
7422 1 : else if (unsignedp && len < HOST_BITS_PER_WIDE_INT)
7423 : {
7424 0 : tem = simplify_shift_const (NULL_RTX, LSHIFTRT, inner_mode,
7425 : XEXP (x, 0), pos);
7426 0 : tem = gen_lowpart (mode, tem);
7427 0 : if (!tem || GET_CODE (tem) == CLOBBER)
7428 : return x;
7429 0 : tem = simplify_and_const_int (NULL_RTX, mode, tem,
7430 0 : (HOST_WIDE_INT_1U << len) - 1);
7431 : }
7432 : else
7433 : /* Any other cases we can't handle. */
7434 : return x;
7435 :
7436 : /* If we couldn't do this for some reason, return the original
7437 : expression. */
7438 3417097 : if (GET_CODE (tem) == CLOBBER)
7439 : return x;
7440 :
7441 : return tem;
7442 : }
7443 :
7444 : /* X is a SET which contains an assignment of one object into
7445 : a part of another (such as a bit-field assignment, STRICT_LOW_PART,
7446 : or certain SUBREGS). If possible, convert it into a series of
7447 : logical operations.
7448 :
7449 : We half-heartedly support variable positions, but do not at all
7450 : support variable lengths. */
7451 :
7452 : static const_rtx
7453 83376796 : expand_field_assignment (const_rtx x)
7454 : {
7455 83376796 : rtx inner;
7456 83376796 : rtx pos; /* Always counts from low bit. */
7457 83376796 : int len, inner_len;
7458 83376796 : rtx mask, cleared, masked;
7459 83376796 : scalar_int_mode compute_mode;
7460 :
7461 : /* Loop until we find something we can't simplify. */
7462 83640967 : while (1)
7463 : {
7464 83640967 : if (GET_CODE (SET_DEST (x)) == STRICT_LOW_PART
7465 15571 : && GET_CODE (XEXP (SET_DEST (x), 0)) == SUBREG)
7466 : {
7467 15571 : rtx x0 = XEXP (SET_DEST (x), 0);
7468 15571 : if (!GET_MODE_PRECISION (GET_MODE (x0)).is_constant (&len))
7469 : break;
7470 15571 : inner = SUBREG_REG (XEXP (SET_DEST (x), 0));
7471 15571 : pos = gen_int_mode (subreg_lsb (XEXP (SET_DEST (x), 0)),
7472 : MAX_MODE_INT);
7473 15571 : }
7474 83625396 : else if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
7475 4219 : && CONST_INT_P (XEXP (SET_DEST (x), 1)))
7476 : {
7477 4219 : inner = XEXP (SET_DEST (x), 0);
7478 4219 : if (!GET_MODE_PRECISION (GET_MODE (inner)).is_constant (&inner_len))
7479 : break;
7480 :
7481 4219 : len = INTVAL (XEXP (SET_DEST (x), 1));
7482 4219 : pos = XEXP (SET_DEST (x), 2);
7483 :
7484 : /* A constant position should stay within the width of INNER. */
7485 4219 : if (CONST_INT_P (pos) && INTVAL (pos) + len > inner_len)
7486 : break;
7487 :
7488 : if (BITS_BIG_ENDIAN)
7489 : {
7490 : if (CONST_INT_P (pos))
7491 : pos = GEN_INT (inner_len - len - INTVAL (pos));
7492 : else if (GET_CODE (pos) == MINUS
7493 : && CONST_INT_P (XEXP (pos, 1))
7494 : && INTVAL (XEXP (pos, 1)) == inner_len - len)
7495 : /* If position is ADJUST - X, new position is X. */
7496 : pos = XEXP (pos, 0);
7497 : else
7498 : pos = simplify_gen_binary (MINUS, GET_MODE (pos),
7499 : gen_int_mode (inner_len - len,
7500 : GET_MODE (pos)),
7501 : pos);
7502 : }
7503 : }
7504 :
7505 : /* If the destination is a subreg that overwrites the whole of the inner
7506 : register, we can move the subreg to the source. */
7507 83871622 : else if (GET_CODE (SET_DEST (x)) == SUBREG
7508 : /* We need SUBREGs to compute nonzero_bits properly. */
7509 847198 : && nonzero_sign_valid
7510 84383722 : && !read_modify_subreg_p (SET_DEST (x)))
7511 : {
7512 250445 : x = gen_rtx_SET (SUBREG_REG (SET_DEST (x)),
7513 : gen_lowpart
7514 : (GET_MODE (SUBREG_REG (SET_DEST (x))),
7515 : SET_SRC (x)));
7516 250445 : continue;
7517 : }
7518 : else
7519 : break;
7520 :
7521 21681 : while (GET_CODE (inner) == SUBREG && subreg_lowpart_p (inner))
7522 1891 : inner = SUBREG_REG (inner);
7523 :
7524 : /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7525 19790 : if (!is_a <scalar_int_mode> (GET_MODE (inner), &compute_mode))
7526 : {
7527 : /* Don't do anything for vector or complex integral types. */
7528 4115 : if (! FLOAT_MODE_P (GET_MODE (inner)))
7529 : break;
7530 :
7531 : /* Try to find an integral mode to pun with. */
7532 38 : if (!int_mode_for_size (GET_MODE_BITSIZE (GET_MODE (inner)), 0)
7533 0 : .exists (&compute_mode))
7534 : break;
7535 :
7536 19 : inner = gen_lowpart (compute_mode, inner);
7537 : }
7538 :
7539 : /* Compute a mask of LEN bits, if we can do this on the host machine. */
7540 15694 : if (len >= HOST_BITS_PER_WIDE_INT)
7541 : break;
7542 :
7543 : /* Don't try to compute in too wide unsupported modes. */
7544 15694 : if (!targetm.scalar_mode_supported_p (compute_mode))
7545 : break;
7546 :
7547 : /* gen_lowpart_for_combine returns CLOBBER on failure. */
7548 15694 : rtx lowpart = gen_lowpart (compute_mode, SET_SRC (x));
7549 15694 : if (GET_CODE (lowpart) == CLOBBER)
7550 : break;
7551 :
7552 : /* Now compute the equivalent expression. Make a copy of INNER
7553 : for the SET_DEST in case it is a MEM into which we will substitute;
7554 : we don't want shared RTL in that case. */
7555 13726 : mask = gen_int_mode ((HOST_WIDE_INT_1U << len) - 1,
7556 : compute_mode);
7557 13726 : cleared = simplify_gen_binary (AND, compute_mode,
7558 : simplify_gen_unary (NOT, compute_mode,
7559 : simplify_gen_binary (ASHIFT,
7560 : compute_mode,
7561 : mask, pos),
7562 : compute_mode),
7563 : inner);
7564 13726 : masked = simplify_gen_binary (ASHIFT, compute_mode,
7565 : simplify_gen_binary (
7566 : AND, compute_mode, lowpart, mask),
7567 : pos);
7568 :
7569 13726 : x = gen_rtx_SET (copy_rtx (inner),
7570 : simplify_gen_binary (IOR, compute_mode,
7571 : cleared, masked));
7572 : }
7573 :
7574 83376796 : return x;
7575 : }
7576 :
7577 : /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7578 : it is an RTX that represents the (variable) starting position; otherwise,
7579 : POS is the (constant) starting bit position. Both are counted from the LSB.
7580 :
7581 : UNSIGNEDP is true for an unsigned reference and zero for a signed one.
7582 :
7583 : IN_DEST is true if this is a reference in the destination of a SET.
7584 : This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7585 : a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7586 : be used.
7587 :
7588 : IN_COMPARE is true if we are in a COMPARE. This means that a
7589 : ZERO_EXTRACT should be built even for bits starting at bit 0.
7590 :
7591 : MODE is the desired mode of the result (if IN_DEST == 0).
7592 :
7593 : The result is an RTX for the extraction or NULL_RTX if the target
7594 : can't handle it. */
7595 :
7596 : static rtx
7597 5071333 : make_extraction (machine_mode mode, rtx inner, HOST_WIDE_INT pos,
7598 : rtx pos_rtx, unsigned HOST_WIDE_INT len, bool unsignedp,
7599 : bool in_dest, bool in_compare)
7600 : {
7601 : /* This mode describes the size of the storage area
7602 : to fetch the overall value from. Within that, we
7603 : ignore the POS lowest bits, etc. */
7604 5071333 : machine_mode is_mode = GET_MODE (inner);
7605 5071333 : machine_mode inner_mode;
7606 5071333 : scalar_int_mode wanted_inner_mode;
7607 5071333 : scalar_int_mode wanted_inner_reg_mode = word_mode;
7608 5071333 : scalar_int_mode pos_mode = word_mode;
7609 5071333 : machine_mode extraction_mode = word_mode;
7610 5071333 : rtx new_rtx = 0;
7611 5071333 : rtx orig_pos_rtx = pos_rtx;
7612 5071333 : HOST_WIDE_INT orig_pos;
7613 :
7614 5071333 : if (pos_rtx && CONST_INT_P (pos_rtx))
7615 926252 : pos = INTVAL (pos_rtx), pos_rtx = 0;
7616 :
7617 5071333 : if (GET_CODE (inner) == SUBREG
7618 2609334 : && subreg_lowpart_p (inner)
7619 7676799 : && (paradoxical_subreg_p (inner)
7620 : /* If trying or potentially trying to extract
7621 : bits outside of is_mode, don't look through
7622 : non-paradoxical SUBREGs. See PR82192. */
7623 151517 : || (pos_rtx == NULL_RTX
7624 151462 : && known_le (pos + len, GET_MODE_PRECISION (is_mode)))))
7625 : {
7626 : /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7627 : consider just the QI as the memory to extract from.
7628 : The subreg adds or removes high bits; its mode is
7629 : irrelevant to the meaning of this extraction,
7630 : since POS and LEN count from the lsb. */
7631 2605411 : if (MEM_P (SUBREG_REG (inner)))
7632 533409 : is_mode = GET_MODE (SUBREG_REG (inner));
7633 : inner = SUBREG_REG (inner);
7634 : }
7635 2465922 : else if (GET_CODE (inner) == ASHIFT
7636 134389 : && CONST_INT_P (XEXP (inner, 1))
7637 133180 : && pos_rtx == 0 && pos == 0
7638 133148 : && len > UINTVAL (XEXP (inner, 1)))
7639 : {
7640 : /* We're extracting the least significant bits of an rtx
7641 : (ashift X (const_int C)), where LEN > C. Extract the
7642 : least significant (LEN - C) bits of X, giving an rtx
7643 : whose mode is MODE, then shift it left C times. */
7644 133148 : new_rtx = make_extraction (mode, XEXP (inner, 0),
7645 : 0, 0, len - INTVAL (XEXP (inner, 1)),
7646 : unsignedp, in_dest, in_compare);
7647 133148 : if (new_rtx != 0)
7648 131502 : return gen_rtx_ASHIFT (mode, new_rtx, XEXP (inner, 1));
7649 : }
7650 2332774 : else if (GET_CODE (inner) == MULT
7651 171438 : && CONST_INT_P (XEXP (inner, 1))
7652 130832 : && pos_rtx == 0 && pos == 0)
7653 : {
7654 : /* We're extracting the least significant bits of an rtx
7655 : (mult X (const_int 2^C)), where LEN > C. Extract the
7656 : least significant (LEN - C) bits of X, giving an rtx
7657 : whose mode is MODE, then multiply it by 2^C. */
7658 109108 : const HOST_WIDE_INT shift_amt = exact_log2 (INTVAL (XEXP (inner, 1)));
7659 109108 : if (len > 1 && IN_RANGE (shift_amt, 1, len - 1))
7660 : {
7661 104274 : new_rtx = make_extraction (mode, XEXP (inner, 0),
7662 : 0, 0, len - shift_amt,
7663 : unsignedp, in_dest, in_compare);
7664 104274 : if (new_rtx)
7665 104274 : return gen_rtx_MULT (mode, new_rtx, XEXP (inner, 1));
7666 : }
7667 : }
7668 2223666 : else if (GET_CODE (inner) == TRUNCATE
7669 : /* If trying or potentially trying to extract
7670 : bits outside of is_mode, don't look through
7671 : TRUNCATE. See PR82192. */
7672 0 : && pos_rtx == NULL_RTX
7673 2223666 : && known_le (pos + len, GET_MODE_PRECISION (is_mode)))
7674 0 : inner = XEXP (inner, 0);
7675 :
7676 4835557 : inner_mode = GET_MODE (inner);
7677 :
7678 : /* See if this can be done without an extraction. We never can if the
7679 : width of the field is not the same as that of some integer mode. For
7680 : registers, we can only avoid the extraction if the position is at the
7681 : low-order bit and this is either not in the destination or we have the
7682 : appropriate STRICT_LOW_PART operation available.
7683 :
7684 : For MEM, we can avoid an extract if the field starts on an appropriate
7685 : boundary and we can change the mode of the memory reference. */
7686 :
7687 4835557 : scalar_int_mode tmode;
7688 4835557 : if (int_mode_for_size (len, 1).exists (&tmode)
7689 2351986 : && ((pos_rtx == 0 && (pos % BITS_PER_WORD) == 0
7690 2076112 : && !MEM_P (inner)
7691 1693850 : && (pos == 0 || REG_P (inner))
7692 1693850 : && (inner_mode == tmode
7693 245884 : || !REG_P (inner)
7694 221538 : || TRULY_NOOP_TRUNCATION_MODES_P (tmode, inner_mode)
7695 0 : || reg_truncated_to_mode (tmode, inner))
7696 1693850 : && (! in_dest
7697 23 : || (REG_P (inner)
7698 23 : && have_insn_for (STRICT_LOW_PART, tmode))))
7699 530539 : || (MEM_P (inner) && pos_rtx == 0
7700 383470 : && (pos
7701 : % (STRICT_ALIGNMENT ? GET_MODE_ALIGNMENT (tmode)
7702 : : BITS_PER_UNIT)) == 0
7703 : /* We can't do this if we are widening INNER_MODE (it
7704 : may not be aligned, for one thing). */
7705 382603 : && !paradoxical_subreg_p (tmode, inner_mode)
7706 382603 : && known_le (pos + len, GET_MODE_PRECISION (is_mode))
7707 382603 : && (inner_mode == tmode
7708 1127 : || (! mode_dependent_address_p (XEXP (inner, 0),
7709 1127 : MEM_ADDR_SPACE (inner))
7710 1127 : && ! MEM_VOLATILE_P (inner))))))
7711 : {
7712 : /* If INNER is a MEM, make a new MEM that encompasses just the desired
7713 : field. If the original and current mode are the same, we need not
7714 : adjust the offset. Otherwise, we do if bytes big endian.
7715 :
7716 : If INNER is not a MEM, get a piece consisting of just the field
7717 : of interest (in this case POS % BITS_PER_WORD must be 0). */
7718 :
7719 2076430 : if (MEM_P (inner))
7720 : {
7721 382590 : poly_int64 offset;
7722 :
7723 : /* POS counts from lsb, but make OFFSET count in memory order. */
7724 382590 : if (BYTES_BIG_ENDIAN)
7725 : offset = bits_to_bytes_round_down (GET_MODE_PRECISION (is_mode)
7726 : - len - pos);
7727 : else
7728 382590 : offset = pos / BITS_PER_UNIT;
7729 :
7730 382590 : new_rtx = adjust_address_nv (inner, tmode, offset);
7731 : }
7732 1693840 : else if (REG_P (inner))
7733 : {
7734 1121157 : if (tmode != inner_mode)
7735 : {
7736 : /* We can't call gen_lowpart in a DEST since we
7737 : always want a SUBREG (see below) and it would sometimes
7738 : return a new hard register. */
7739 221528 : if (pos || in_dest)
7740 : {
7741 16 : poly_uint64 offset
7742 16 : = subreg_offset_from_lsb (tmode, inner_mode, pos);
7743 :
7744 : /* Avoid creating invalid subregs, for example when
7745 : simplifying (x>>32)&255. */
7746 16 : if (!validate_subreg (tmode, inner_mode, inner, offset))
7747 0 : return NULL_RTX;
7748 :
7749 16 : new_rtx = gen_rtx_SUBREG (tmode, inner, offset);
7750 16 : }
7751 : else
7752 221512 : new_rtx = gen_lowpart (tmode, inner);
7753 : }
7754 : else
7755 : new_rtx = inner;
7756 : }
7757 : else
7758 1145366 : new_rtx = force_to_mode (inner, tmode,
7759 : len >= HOST_BITS_PER_WIDE_INT
7760 : ? HOST_WIDE_INT_M1U
7761 572683 : : (HOST_WIDE_INT_1U << len) - 1, false);
7762 :
7763 : /* If this extraction is going into the destination of a SET,
7764 : make a STRICT_LOW_PART unless we made a MEM. */
7765 :
7766 2076430 : if (in_dest)
7767 49 : return (MEM_P (new_rtx) ? new_rtx
7768 : : (GET_CODE (new_rtx) != SUBREG
7769 13 : ? gen_rtx_CLOBBER (tmode, const0_rtx)
7770 13 : : gen_rtx_STRICT_LOW_PART (VOIDmode, new_rtx)));
7771 :
7772 2076381 : if (mode == tmode)
7773 : return new_rtx;
7774 :
7775 2076352 : if (CONST_SCALAR_INT_P (new_rtx))
7776 5 : return simplify_unary_operation (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
7777 5 : mode, new_rtx, tmode);
7778 :
7779 : /* If we know that no extraneous bits are set, and that the high
7780 : bit is not set, convert the extraction to the cheaper of
7781 : sign and zero extension, that are equivalent in these cases. */
7782 2076347 : if (flag_expensive_optimizations
7783 2076347 : && (HWI_COMPUTABLE_MODE_P (tmode)
7784 1929274 : && ((nonzero_bits (new_rtx, tmode)
7785 1929274 : & ~(((unsigned HOST_WIDE_INT)GET_MODE_MASK (tmode)) >> 1))
7786 : == 0)))
7787 : {
7788 8516 : rtx temp = gen_rtx_ZERO_EXTEND (mode, new_rtx);
7789 8516 : rtx temp1 = gen_rtx_SIGN_EXTEND (mode, new_rtx);
7790 :
7791 : /* Prefer ZERO_EXTENSION, since it gives more information to
7792 : backends. */
7793 8516 : if (set_src_cost (temp, mode, optimize_this_for_speed_p)
7794 8516 : <= set_src_cost (temp1, mode, optimize_this_for_speed_p))
7795 : return temp;
7796 0 : return temp1;
7797 : }
7798 :
7799 : /* Otherwise, sign- or zero-extend unless we already are in the
7800 : proper mode. */
7801 :
7802 2067831 : return (gen_rtx_fmt_e (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
7803 2067831 : mode, new_rtx));
7804 : }
7805 :
7806 : /* Unless this is a COMPARE or we have a funny memory reference,
7807 : don't do anything with zero-extending field extracts starting at
7808 : the low-order bit since they are simple AND operations. */
7809 2759127 : if (pos_rtx == 0 && pos == 0 && ! in_dest
7810 1700805 : && ! in_compare && unsignedp)
7811 : return 0;
7812 :
7813 : /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7814 : if the position is not a constant and the length is not 1. In all
7815 : other cases, we would only be going outside our object in cases when
7816 : an original shift would have been undefined. */
7817 1447966 : if (MEM_P (inner)
7818 1447966 : && ((pos_rtx == 0 && maybe_gt (pos + len, GET_MODE_PRECISION (is_mode)))
7819 3031 : || (pos_rtx != 0 && len != 1)))
7820 : return 0;
7821 :
7822 1555108 : enum extraction_pattern pattern = (in_dest ? EP_insv
7823 1440757 : : unsignedp ? EP_extzv : EP_extv);
7824 :
7825 : /* If INNER is not from memory, we want it to have the mode of a register
7826 : extraction pattern's structure operand, or word_mode if there is no
7827 : such pattern. The same applies to extraction_mode and pos_mode
7828 : and their respective operands.
7829 :
7830 : For memory, assume that the desired extraction_mode and pos_mode
7831 : are the same as for a register operation, since at present we don't
7832 : have named patterns for aligned memory structures. */
7833 1447928 : class extraction_insn insn;
7834 1447928 : unsigned int inner_size;
7835 2895856 : if (GET_MODE_BITSIZE (inner_mode).is_constant (&inner_size)
7836 1447928 : && get_best_reg_extraction_insn (&insn, pattern, inner_size, mode))
7837 : {
7838 1339665 : wanted_inner_reg_mode = insn.struct_mode.require ();
7839 1339665 : pos_mode = insn.pos_mode;
7840 1339665 : extraction_mode = insn.field_mode;
7841 : }
7842 :
7843 : /* Never narrow an object, since that might not be safe. */
7844 :
7845 1447928 : if (mode != VOIDmode
7846 1447928 : && partial_subreg_p (extraction_mode, mode))
7847 : extraction_mode = mode;
7848 :
7849 : /* Punt if len is too large for extraction_mode. */
7850 1447928 : if (maybe_gt (len, GET_MODE_PRECISION (extraction_mode)))
7851 : return NULL_RTX;
7852 :
7853 1447916 : if (!MEM_P (inner))
7854 1265695 : wanted_inner_mode = wanted_inner_reg_mode;
7855 : else
7856 : {
7857 : /* Be careful not to go beyond the extracted object and maintain the
7858 : natural alignment of the memory. */
7859 182221 : wanted_inner_mode = smallest_int_mode_for_size (len).require ();
7860 367507 : while (pos % GET_MODE_BITSIZE (wanted_inner_mode) + len
7861 370572 : > GET_MODE_BITSIZE (wanted_inner_mode))
7862 3065 : wanted_inner_mode = GET_MODE_WIDER_MODE (wanted_inner_mode).require ();
7863 : }
7864 :
7865 1447916 : orig_pos = pos;
7866 :
7867 1447916 : if (BITS_BIG_ENDIAN)
7868 : {
7869 : /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7870 : BITS_BIG_ENDIAN style. If position is constant, compute new
7871 : position. Otherwise, build subtraction.
7872 : Note that POS is relative to the mode of the original argument.
7873 : If it's a MEM we need to recompute POS relative to that.
7874 : However, if we're extracting from (or inserting into) a register,
7875 : we want to recompute POS relative to wanted_inner_mode. */
7876 : int width;
7877 : if (!MEM_P (inner))
7878 : width = GET_MODE_BITSIZE (wanted_inner_mode);
7879 : else if (!GET_MODE_BITSIZE (is_mode).is_constant (&width))
7880 : return NULL_RTX;
7881 :
7882 : if (pos_rtx == 0)
7883 : pos = width - len - pos;
7884 : else
7885 : pos_rtx
7886 : = gen_rtx_MINUS (GET_MODE (pos_rtx),
7887 : gen_int_mode (width - len, GET_MODE (pos_rtx)),
7888 : pos_rtx);
7889 : /* POS may be less than 0 now, but we check for that below.
7890 : Note that it can only be less than 0 if !MEM_P (inner). */
7891 : }
7892 :
7893 : /* If INNER has a wider mode, and this is a constant extraction, try to
7894 : make it smaller and adjust the byte to point to the byte containing
7895 : the value. */
7896 1447916 : if (wanted_inner_mode != VOIDmode
7897 1447916 : && inner_mode != wanted_inner_mode
7898 203048 : && ! pos_rtx
7899 195006 : && partial_subreg_p (wanted_inner_mode, is_mode)
7900 115286 : && MEM_P (inner)
7901 29305 : && ! mode_dependent_address_p (XEXP (inner, 0), MEM_ADDR_SPACE (inner))
7902 1477221 : && ! MEM_VOLATILE_P (inner))
7903 : {
7904 27695 : poly_int64 offset = 0;
7905 :
7906 : /* The computations below will be correct if the machine is big
7907 : endian in both bits and bytes or little endian in bits and bytes.
7908 : If it is mixed, we must adjust. */
7909 :
7910 : /* If bytes are big endian and we had a paradoxical SUBREG, we must
7911 : adjust OFFSET to compensate. */
7912 27695 : if (BYTES_BIG_ENDIAN
7913 : && paradoxical_subreg_p (is_mode, inner_mode))
7914 : offset -= GET_MODE_SIZE (is_mode) - GET_MODE_SIZE (inner_mode);
7915 :
7916 : /* We can now move to the desired byte. */
7917 55390 : offset += (pos / GET_MODE_BITSIZE (wanted_inner_mode))
7918 27695 : * GET_MODE_SIZE (wanted_inner_mode);
7919 27695 : pos %= GET_MODE_BITSIZE (wanted_inner_mode);
7920 :
7921 27695 : if (BYTES_BIG_ENDIAN != BITS_BIG_ENDIAN
7922 : && is_mode != wanted_inner_mode)
7923 : offset = (GET_MODE_SIZE (is_mode)
7924 : - GET_MODE_SIZE (wanted_inner_mode) - offset);
7925 :
7926 27695 : inner = adjust_address_nv (inner, wanted_inner_mode, offset);
7927 : }
7928 :
7929 : /* If INNER is not memory, get it into the proper mode. If we are changing
7930 : its mode, POS must be a constant and smaller than the size of the new
7931 : mode. */
7932 1420221 : else if (!MEM_P (inner))
7933 : {
7934 : /* On the LHS, don't create paradoxical subregs implicitly truncating
7935 : the register unless TARGET_TRULY_NOOP_TRUNCATION. */
7936 1265695 : if (in_dest
7937 1265695 : && !TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (inner),
7938 : wanted_inner_mode))
7939 0 : return NULL_RTX;
7940 :
7941 1265695 : if (GET_MODE (inner) != wanted_inner_mode
7942 1265695 : && (pos_rtx != 0
7943 331402 : || orig_pos + len > GET_MODE_BITSIZE (wanted_inner_mode)))
7944 : return NULL_RTX;
7945 :
7946 1197800 : if (orig_pos < 0)
7947 : return NULL_RTX;
7948 :
7949 2375482 : inner = force_to_mode (inner, wanted_inner_mode,
7950 : pos_rtx
7951 1177682 : || len + orig_pos >= HOST_BITS_PER_WIDE_INT
7952 : ? HOST_WIDE_INT_M1U
7953 1045168 : : (((HOST_WIDE_INT_1U << len) - 1)
7954 1045168 : << orig_pos), false);
7955 : }
7956 :
7957 : /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7958 : have to zero extend. Otherwise, we can just use a SUBREG.
7959 :
7960 : We dealt with constant rtxes earlier, so pos_rtx cannot
7961 : have VOIDmode at this point. */
7962 1380021 : if (pos_rtx != 0
7963 1380021 : && (GET_MODE_SIZE (pos_mode)
7964 1403132 : > GET_MODE_SIZE (as_a <scalar_int_mode> (GET_MODE (pos_rtx)))))
7965 : {
7966 74 : rtx temp = simplify_gen_unary (ZERO_EXTEND, pos_mode, pos_rtx,
7967 : GET_MODE (pos_rtx));
7968 :
7969 : /* If we know that no extraneous bits are set, and that the high
7970 : bit is not set, convert extraction to cheaper one - either
7971 : SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7972 : cases. */
7973 74 : if (flag_expensive_optimizations
7974 74 : && (HWI_COMPUTABLE_MODE_P (GET_MODE (pos_rtx))
7975 74 : && ((nonzero_bits (pos_rtx, GET_MODE (pos_rtx))
7976 74 : & ~(((unsigned HOST_WIDE_INT)
7977 74 : GET_MODE_MASK (GET_MODE (pos_rtx)))
7978 74 : >> 1))
7979 : == 0)))
7980 : {
7981 46 : rtx temp1 = simplify_gen_unary (SIGN_EXTEND, pos_mode, pos_rtx,
7982 : GET_MODE (pos_rtx));
7983 :
7984 : /* Prefer ZERO_EXTENSION, since it gives more information to
7985 : backends. */
7986 46 : if (set_src_cost (temp1, pos_mode, optimize_this_for_speed_p)
7987 46 : < set_src_cost (temp, pos_mode, optimize_this_for_speed_p))
7988 1380021 : temp = temp1;
7989 : }
7990 : pos_rtx = temp;
7991 : }
7992 :
7993 : /* Make POS_RTX unless we already have it and it is correct. If we don't
7994 : have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7995 : be a CONST_INT. */
7996 1380021 : if (pos_rtx == 0 && orig_pos_rtx != 0 && INTVAL (orig_pos_rtx) == pos)
7997 : pos_rtx = orig_pos_rtx;
7998 :
7999 480240 : else if (pos_rtx == 0)
8000 457129 : pos_rtx = GEN_INT (pos);
8001 :
8002 : /* Make the required operation. See if we can use existing rtx. */
8003 1380021 : new_rtx = gen_rtx_fmt_eee (unsignedp ? ZERO_EXTRACT : SIGN_EXTRACT,
8004 : extraction_mode, inner, GEN_INT (len), pos_rtx);
8005 1380021 : if (! in_dest)
8006 1372894 : new_rtx = gen_lowpart (mode, new_rtx);
8007 :
8008 : return new_rtx;
8009 : }
8010 :
8011 : /* See if X (of mode MODE) contains an ASHIFT of COUNT or more bits that
8012 : can be commuted with any other operations in X. Return X without
8013 : that shift if so. */
8014 :
8015 : static rtx
8016 1583002 : extract_left_shift (scalar_int_mode mode, rtx x, int count)
8017 : {
8018 1583002 : enum rtx_code code = GET_CODE (x);
8019 1583002 : rtx tem;
8020 :
8021 1583002 : switch (code)
8022 : {
8023 252922 : case ASHIFT:
8024 : /* This is the shift itself. If it is wide enough, we will return
8025 : either the value being shifted if the shift count is equal to
8026 : COUNT or a shift for the difference. */
8027 252922 : if (CONST_INT_P (XEXP (x, 1))
8028 247672 : && INTVAL (XEXP (x, 1)) >= count)
8029 246487 : return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (x, 0),
8030 246487 : INTVAL (XEXP (x, 1)) - count);
8031 : break;
8032 :
8033 5388 : case NEG: case NOT:
8034 5388 : if ((tem = extract_left_shift (mode, XEXP (x, 0), count)) != 0)
8035 2585 : return simplify_gen_unary (code, mode, tem, mode);
8036 :
8037 : break;
8038 :
8039 555142 : case PLUS: case IOR: case XOR: case AND:
8040 : /* If we can safely shift this constant and we find the inner shift,
8041 : make a new operation. */
8042 555142 : if (CONST_INT_P (XEXP (x, 1))
8043 284650 : && (UINTVAL (XEXP (x, 1))
8044 284650 : & (((HOST_WIDE_INT_1U << count)) - 1)) == 0
8045 682166 : && (tem = extract_left_shift (mode, XEXP (x, 0), count)) != 0)
8046 : {
8047 6715 : HOST_WIDE_INT val = INTVAL (XEXP (x, 1)) >> count;
8048 6715 : return simplify_gen_binary (code, mode, tem,
8049 6715 : gen_int_mode (val, mode));
8050 : }
8051 : break;
8052 :
8053 : default:
8054 : break;
8055 : }
8056 :
8057 : return 0;
8058 : }
8059 :
8060 : /* Subroutine of make_compound_operation. *X_PTR is the rtx at the current
8061 : level of the expression and MODE is its mode. IN_CODE is as for
8062 : make_compound_operation. *NEXT_CODE_PTR is the value of IN_CODE
8063 : that should be used when recursing on operands of *X_PTR.
8064 :
8065 : There are two possible actions:
8066 :
8067 : - Return null. This tells the caller to recurse on *X_PTR with IN_CODE
8068 : equal to *NEXT_CODE_PTR, after which *X_PTR holds the final value.
8069 :
8070 : - Return a new rtx, which the caller returns directly. */
8071 :
8072 : static rtx
8073 275419182 : make_compound_operation_int (scalar_int_mode mode, rtx *x_ptr,
8074 : enum rtx_code in_code,
8075 : enum rtx_code *next_code_ptr)
8076 : {
8077 275419182 : rtx x = *x_ptr;
8078 275419182 : enum rtx_code next_code = *next_code_ptr;
8079 275419182 : enum rtx_code code = GET_CODE (x);
8080 275419182 : int mode_width = GET_MODE_PRECISION (mode);
8081 275419182 : rtx rhs, lhs;
8082 275419182 : rtx new_rtx = 0;
8083 275419182 : int i;
8084 275419182 : rtx tem;
8085 275419182 : scalar_int_mode inner_mode;
8086 275419182 : bool equality_comparison = false;
8087 :
8088 275419182 : if (in_code == EQ)
8089 : {
8090 8677913 : equality_comparison = true;
8091 8677913 : in_code = COMPARE;
8092 : }
8093 :
8094 : /* Process depending on the code of this operation. If NEW is set
8095 : nonzero, it will be returned. */
8096 :
8097 275419182 : switch (code)
8098 : {
8099 6487029 : case ASHIFT:
8100 : /* Convert shifts by constants into multiplications if inside
8101 : an address. */
8102 6487029 : if (in_code == MEM && CONST_INT_P (XEXP (x, 1))
8103 2071283 : && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
8104 2071283 : && INTVAL (XEXP (x, 1)) >= 0)
8105 : {
8106 2071283 : HOST_WIDE_INT count = INTVAL (XEXP (x, 1));
8107 2071283 : HOST_WIDE_INT multval = HOST_WIDE_INT_1 << count;
8108 :
8109 2071283 : new_rtx = make_compound_operation (XEXP (x, 0), next_code);
8110 2071283 : if (GET_CODE (new_rtx) == NEG)
8111 : {
8112 9 : new_rtx = XEXP (new_rtx, 0);
8113 9 : multval = -multval;
8114 : }
8115 2071283 : multval = trunc_int_for_mode (multval, mode);
8116 2071283 : new_rtx = gen_rtx_MULT (mode, new_rtx, gen_int_mode (multval, mode));
8117 : }
8118 : break;
8119 :
8120 50770565 : case PLUS:
8121 50770565 : lhs = XEXP (x, 0);
8122 50770565 : rhs = XEXP (x, 1);
8123 50770565 : lhs = make_compound_operation (lhs, next_code);
8124 50770565 : rhs = make_compound_operation (rhs, next_code);
8125 50770565 : if (GET_CODE (lhs) == MULT && GET_CODE (XEXP (lhs, 0)) == NEG)
8126 : {
8127 0 : tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (lhs, 0), 0),
8128 : XEXP (lhs, 1));
8129 0 : new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
8130 : }
8131 50770565 : else if (GET_CODE (lhs) == MULT
8132 4918795 : && (CONST_INT_P (XEXP (lhs, 1)) && INTVAL (XEXP (lhs, 1)) < 0))
8133 : {
8134 32673 : tem = simplify_gen_binary (MULT, mode, XEXP (lhs, 0),
8135 : simplify_gen_unary (NEG, mode,
8136 : XEXP (lhs, 1),
8137 : mode));
8138 32673 : new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
8139 : }
8140 : else
8141 : {
8142 50737892 : SUBST (XEXP (x, 0), lhs);
8143 50737892 : SUBST (XEXP (x, 1), rhs);
8144 : }
8145 50770565 : maybe_swap_commutative_operands (x);
8146 50770565 : return x;
8147 :
8148 3748459 : case MINUS:
8149 3748459 : lhs = XEXP (x, 0);
8150 3748459 : rhs = XEXP (x, 1);
8151 3748459 : lhs = make_compound_operation (lhs, next_code);
8152 3748459 : rhs = make_compound_operation (rhs, next_code);
8153 3748459 : if (GET_CODE (rhs) == MULT && GET_CODE (XEXP (rhs, 0)) == NEG)
8154 : {
8155 0 : tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (rhs, 0), 0),
8156 : XEXP (rhs, 1));
8157 0 : return simplify_gen_binary (PLUS, mode, tem, lhs);
8158 : }
8159 3748459 : else if (GET_CODE (rhs) == MULT
8160 57920 : && (CONST_INT_P (XEXP (rhs, 1)) && INTVAL (XEXP (rhs, 1)) < 0))
8161 : {
8162 192 : tem = simplify_gen_binary (MULT, mode, XEXP (rhs, 0),
8163 : simplify_gen_unary (NEG, mode,
8164 : XEXP (rhs, 1),
8165 : mode));
8166 192 : return simplify_gen_binary (PLUS, mode, tem, lhs);
8167 : }
8168 : else
8169 : {
8170 3748267 : SUBST (XEXP (x, 0), lhs);
8171 3748267 : SUBST (XEXP (x, 1), rhs);
8172 3748267 : return x;
8173 : }
8174 :
8175 7404085 : case AND:
8176 : /* If the second operand is not a constant, we can't do anything
8177 : with it. */
8178 7404085 : if (!CONST_INT_P (XEXP (x, 1)))
8179 : break;
8180 :
8181 : /* If the constant is a power of two minus one and the first operand
8182 : is a logical right shift, make an extraction. */
8183 5898690 : if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
8184 5898690 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
8185 : {
8186 671644 : new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
8187 671644 : new_rtx = make_extraction (mode, new_rtx, 0, XEXP (XEXP (x, 0), 1),
8188 : i, true, false, in_code == COMPARE);
8189 : }
8190 :
8191 : /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
8192 5227046 : else if (GET_CODE (XEXP (x, 0)) == SUBREG
8193 1337511 : && subreg_lowpart_p (XEXP (x, 0))
8194 6529414 : && is_a <scalar_int_mode> (GET_MODE (SUBREG_REG (XEXP (x, 0))),
8195 : &inner_mode)
8196 1332953 : && GET_CODE (SUBREG_REG (XEXP (x, 0))) == LSHIFTRT
8197 5258737 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
8198 : {
8199 30585 : rtx inner_x0 = SUBREG_REG (XEXP (x, 0));
8200 30585 : new_rtx = make_compound_operation (XEXP (inner_x0, 0), next_code);
8201 30585 : new_rtx = make_extraction (inner_mode, new_rtx, 0,
8202 : XEXP (inner_x0, 1),
8203 : i, true, false, in_code == COMPARE);
8204 :
8205 : /* If we narrowed the mode when dropping the subreg, then we lose. */
8206 91755 : if (GET_MODE_SIZE (inner_mode) < GET_MODE_SIZE (mode))
8207 30585 : new_rtx = NULL;
8208 :
8209 : /* If that didn't give anything, see if the AND simplifies on
8210 : its own. */
8211 30585 : if (!new_rtx && i >= 0)
8212 : {
8213 3642 : new_rtx = make_compound_operation (XEXP (x, 0), next_code);
8214 3642 : new_rtx = make_extraction (mode, new_rtx, 0, NULL_RTX, i,
8215 : true, false, in_code == COMPARE);
8216 : }
8217 : }
8218 : /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
8219 5196461 : else if ((GET_CODE (XEXP (x, 0)) == XOR
8220 5196461 : || GET_CODE (XEXP (x, 0)) == IOR)
8221 28052 : && GET_CODE (XEXP (XEXP (x, 0), 0)) == LSHIFTRT
8222 2509 : && GET_CODE (XEXP (XEXP (x, 0), 1)) == LSHIFTRT
8223 5196471 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
8224 : {
8225 : /* Apply the distributive law, and then try to make extractions. */
8226 10 : new_rtx = gen_rtx_fmt_ee (GET_CODE (XEXP (x, 0)), mode,
8227 : gen_rtx_AND (mode, XEXP (XEXP (x, 0), 0),
8228 : XEXP (x, 1)),
8229 : gen_rtx_AND (mode, XEXP (XEXP (x, 0), 1),
8230 : XEXP (x, 1)));
8231 10 : new_rtx = make_compound_operation (new_rtx, in_code);
8232 : }
8233 :
8234 : /* If we are have (and (rotate X C) M) and C is larger than the number
8235 : of bits in M, this is an extraction. */
8236 :
8237 5196451 : else if (GET_CODE (XEXP (x, 0)) == ROTATE
8238 1689 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
8239 1689 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0
8240 5196483 : && i <= INTVAL (XEXP (XEXP (x, 0), 1)))
8241 : {
8242 0 : new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
8243 0 : new_rtx = make_extraction (mode, new_rtx,
8244 0 : (GET_MODE_PRECISION (mode)
8245 0 : - INTVAL (XEXP (XEXP (x, 0), 1))),
8246 : NULL_RTX, i, true, false,
8247 : in_code == COMPARE);
8248 : }
8249 :
8250 : /* On machines without logical shifts, if the operand of the AND is
8251 : a logical shift and our mask turns off all the propagated sign
8252 : bits, we can replace the logical shift with an arithmetic shift. */
8253 5196451 : else if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
8254 84527 : && !have_insn_for (LSHIFTRT, mode)
8255 0 : && have_insn_for (ASHIFTRT, mode)
8256 0 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
8257 0 : && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
8258 0 : && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
8259 5196451 : && mode_width <= HOST_BITS_PER_WIDE_INT)
8260 : {
8261 0 : unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
8262 :
8263 0 : mask >>= INTVAL (XEXP (XEXP (x, 0), 1));
8264 0 : if ((INTVAL (XEXP (x, 1)) & ~mask) == 0)
8265 0 : SUBST (XEXP (x, 0),
8266 : gen_rtx_ASHIFTRT (mode,
8267 : make_compound_operation (XEXP (XEXP (x,
8268 : 0),
8269 : 0),
8270 : next_code),
8271 : XEXP (XEXP (x, 0), 1)));
8272 : }
8273 :
8274 : /* If the constant is one less than a power of two, this might be
8275 : representable by an extraction even if no shift is present.
8276 : If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
8277 : we are in a COMPARE. */
8278 5196451 : else if ((i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
8279 2624916 : new_rtx = make_extraction (mode,
8280 : make_compound_operation (XEXP (x, 0),
8281 : next_code),
8282 : 0, NULL_RTX, i,
8283 : true, false, in_code == COMPARE);
8284 :
8285 : /* If we are in a comparison and this is an AND with a power of two,
8286 : convert this into the appropriate bit extract. */
8287 2571535 : else if (in_code == COMPARE
8288 493174 : && (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0
8289 2660618 : && (equality_comparison || i < GET_MODE_PRECISION (mode) - 1))
8290 89083 : new_rtx = make_extraction (mode,
8291 : make_compound_operation (XEXP (x, 0),
8292 : next_code),
8293 : i, NULL_RTX, 1, true, false, true);
8294 :
8295 : /* If the one operand is a paradoxical subreg of a register or memory and
8296 : the constant (limited to the smaller mode) has only zero bits where
8297 : the sub expression has known zero bits, this can be expressed as
8298 : a zero_extend. */
8299 2482452 : else if (GET_CODE (XEXP (x, 0)) == SUBREG)
8300 : {
8301 63676 : rtx sub;
8302 :
8303 63676 : sub = XEXP (XEXP (x, 0), 0);
8304 63676 : machine_mode sub_mode = GET_MODE (sub);
8305 63676 : int sub_width;
8306 29784 : if ((REG_P (sub) || MEM_P (sub))
8307 34391 : && GET_MODE_PRECISION (sub_mode).is_constant (&sub_width)
8308 34391 : && sub_width < mode_width
8309 63676 : && (!WORD_REGISTER_OPERATIONS
8310 : || sub_width >= BITS_PER_WORD
8311 : /* On WORD_REGISTER_OPERATIONS targets the bits
8312 : beyond sub_mode aren't considered undefined,
8313 : so optimize only if it is a MEM load when MEM loads
8314 : zero extend, because then the upper bits are all zero. */
8315 : || (MEM_P (sub)
8316 : && load_extend_op (sub_mode) == ZERO_EXTEND)))
8317 : {
8318 26957 : unsigned HOST_WIDE_INT mode_mask = GET_MODE_MASK (sub_mode);
8319 26957 : unsigned HOST_WIDE_INT mask;
8320 :
8321 : /* Original AND constant with all the known zero bits set. */
8322 26957 : mask = UINTVAL (XEXP (x, 1)) | (~nonzero_bits (sub, sub_mode));
8323 26957 : if ((mask & mode_mask) == mode_mask)
8324 : {
8325 23722 : new_rtx = make_compound_operation (sub, next_code);
8326 23722 : new_rtx = make_extraction (mode, new_rtx, 0, 0, sub_width,
8327 : true, false, in_code == COMPARE);
8328 : }
8329 : }
8330 : }
8331 :
8332 : break;
8333 :
8334 1927502 : case LSHIFTRT:
8335 : /* If the sign bit is known to be zero, replace this with an
8336 : arithmetic shift. */
8337 1927502 : if (have_insn_for (ASHIFTRT, mode)
8338 1927502 : && ! have_insn_for (LSHIFTRT, mode)
8339 0 : && mode_width <= HOST_BITS_PER_WIDE_INT
8340 1927502 : && (nonzero_bits (XEXP (x, 0), mode) & (1 << (mode_width - 1))) == 0)
8341 : {
8342 0 : new_rtx = gen_rtx_ASHIFTRT (mode,
8343 : make_compound_operation (XEXP (x, 0),
8344 : next_code),
8345 : XEXP (x, 1));
8346 0 : break;
8347 : }
8348 :
8349 : /* fall through */
8350 :
8351 4406344 : case ASHIFTRT:
8352 4406344 : lhs = XEXP (x, 0);
8353 4406344 : rhs = XEXP (x, 1);
8354 :
8355 : /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
8356 : this is a SIGN_EXTRACT. */
8357 4406344 : if (CONST_INT_P (rhs)
8358 4249110 : && GET_CODE (lhs) == ASHIFT
8359 1127063 : && CONST_INT_P (XEXP (lhs, 1))
8360 1121817 : && INTVAL (rhs) >= INTVAL (XEXP (lhs, 1))
8361 884370 : && INTVAL (XEXP (lhs, 1)) >= 0
8362 884366 : && INTVAL (rhs) < mode_width)
8363 : {
8364 884366 : new_rtx = make_compound_operation (XEXP (lhs, 0), next_code);
8365 884366 : new_rtx = make_extraction (mode, new_rtx,
8366 884366 : INTVAL (rhs) - INTVAL (XEXP (lhs, 1)),
8367 884366 : NULL_RTX, mode_width - INTVAL (rhs),
8368 : code == LSHIFTRT, false,
8369 : in_code == COMPARE);
8370 884366 : break;
8371 : }
8372 :
8373 : /* See if we have operations between an ASHIFTRT and an ASHIFT.
8374 : If so, try to merge the shifts into a SIGN_EXTEND. We could
8375 : also do this for some cases of SIGN_EXTRACT, but it doesn't
8376 : seem worth the effort; the case checked for occurs on Alpha. */
8377 :
8378 3521978 : if (!OBJECT_P (lhs)
8379 1546993 : && ! (GET_CODE (lhs) == SUBREG
8380 84979 : && (OBJECT_P (SUBREG_REG (lhs))))
8381 1477028 : && CONST_INT_P (rhs)
8382 1455253 : && INTVAL (rhs) >= 0
8383 1455253 : && INTVAL (rhs) < HOST_BITS_PER_WIDE_INT
8384 1450590 : && INTVAL (rhs) < mode_width
8385 4972568 : && (new_rtx = extract_left_shift (mode, lhs, INTVAL (rhs))) != 0)
8386 246487 : new_rtx = make_extraction (mode, make_compound_operation (new_rtx,
8387 : next_code),
8388 246487 : 0, NULL_RTX, mode_width - INTVAL (rhs),
8389 : code == LSHIFTRT, false, in_code == COMPARE);
8390 :
8391 : break;
8392 :
8393 8726953 : case SUBREG:
8394 : /* Call ourselves recursively on the inner expression. If we are
8395 : narrowing the object and it has a different RTL code from
8396 : what it originally did, do this SUBREG as a force_to_mode. */
8397 8726953 : {
8398 8726953 : rtx inner = SUBREG_REG (x), simplified;
8399 8726953 : enum rtx_code subreg_code = in_code;
8400 :
8401 : /* If the SUBREG is masking of a logical right shift,
8402 : make an extraction. */
8403 8726953 : if (GET_CODE (inner) == LSHIFTRT
8404 8741661 : && is_a <scalar_int_mode> (GET_MODE (inner), &inner_mode)
8405 533908 : && GET_MODE_SIZE (mode) < GET_MODE_SIZE (inner_mode)
8406 259249 : && CONST_INT_P (XEXP (inner, 1))
8407 254340 : && UINTVAL (XEXP (inner, 1)) < GET_MODE_PRECISION (inner_mode)
8408 8981293 : && subreg_lowpart_p (x))
8409 : {
8410 252246 : new_rtx = make_compound_operation (XEXP (inner, 0), next_code);
8411 252246 : int width = GET_MODE_PRECISION (inner_mode)
8412 252246 : - INTVAL (XEXP (inner, 1));
8413 252246 : if (width > mode_width)
8414 : width = mode_width;
8415 252246 : new_rtx = make_extraction (mode, new_rtx, 0, XEXP (inner, 1),
8416 : width, true, false, in_code == COMPARE);
8417 252246 : break;
8418 : }
8419 :
8420 : /* If in_code is COMPARE, it isn't always safe to pass it through
8421 : to the recursive make_compound_operation call. */
8422 8474707 : if (subreg_code == COMPARE
8423 8474707 : && (!subreg_lowpart_p (x)
8424 147848 : || GET_CODE (inner) == SUBREG
8425 : /* (subreg:SI (and:DI (reg:DI) (const_int 0x800000000)) 0)
8426 : is (const_int 0), rather than
8427 : (subreg:SI (lshiftrt:DI (reg:DI) (const_int 35)) 0).
8428 : Similarly (subreg:QI (and:SI (reg:SI) (const_int 0x80)) 0)
8429 : for non-equality comparisons against 0 is not equivalent
8430 : to (subreg:QI (lshiftrt:SI (reg:SI) (const_int 7)) 0). */
8431 147848 : || (GET_CODE (inner) == AND
8432 1199 : && CONST_INT_P (XEXP (inner, 1))
8433 139 : && partial_subreg_p (x)
8434 278 : && exact_log2 (UINTVAL (XEXP (inner, 1)))
8435 139 : >= GET_MODE_BITSIZE (mode) - 1)))
8436 : subreg_code = SET;
8437 :
8438 8474707 : tem = make_compound_operation (inner, subreg_code);
8439 :
8440 8474707 : simplified
8441 8474707 : = simplify_subreg (mode, tem, GET_MODE (inner), SUBREG_BYTE (x));
8442 8474707 : if (simplified)
8443 8991 : tem = simplified;
8444 :
8445 8474707 : if (GET_CODE (tem) != GET_CODE (inner)
8446 13744 : && partial_subreg_p (x)
8447 8486473 : && subreg_lowpart_p (x))
8448 : {
8449 11750 : rtx newer
8450 11750 : = force_to_mode (tem, mode, HOST_WIDE_INT_M1U, false);
8451 :
8452 : /* If we have something other than a SUBREG, we might have
8453 : done an expansion, so rerun ourselves. */
8454 11750 : if (GET_CODE (newer) != SUBREG)
8455 9524 : newer = make_compound_operation (newer, in_code);
8456 :
8457 : /* force_to_mode can expand compounds. If it just re-expanded
8458 : the compound, use gen_lowpart to convert to the desired
8459 : mode. */
8460 11750 : if (rtx_equal_p (newer, x)
8461 : /* Likewise if it re-expanded the compound only partially.
8462 : This happens for SUBREG of ZERO_EXTRACT if they extract
8463 : the same number of bits. */
8464 11750 : || (GET_CODE (newer) == SUBREG
8465 1859 : && (GET_CODE (SUBREG_REG (newer)) == LSHIFTRT
8466 1859 : || GET_CODE (SUBREG_REG (newer)) == ASHIFTRT)
8467 113 : && GET_CODE (inner) == AND
8468 0 : && rtx_equal_p (SUBREG_REG (newer), XEXP (inner, 0))))
8469 1783 : return gen_lowpart (GET_MODE (x), tem);
8470 :
8471 9967 : return newer;
8472 : }
8473 :
8474 8462957 : if (simplified)
8475 : return tem;
8476 : }
8477 : break;
8478 :
8479 : default:
8480 : break;
8481 : }
8482 :
8483 10209787 : if (new_rtx)
8484 5520247 : *x_ptr = gen_lowpart (mode, new_rtx);
8485 220887831 : *next_code_ptr = next_code;
8486 220887831 : return NULL_RTX;
8487 : }
8488 :
8489 : /* Look at the expression rooted at X. Look for expressions
8490 : equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
8491 : Form these expressions.
8492 :
8493 : Return the new rtx, usually just X.
8494 :
8495 : Also, for machines like the VAX that don't have logical shift insns,
8496 : try to convert logical to arithmetic shift operations in cases where
8497 : they are equivalent. This undoes the canonicalizations to logical
8498 : shifts done elsewhere.
8499 :
8500 : We try, as much as possible, to re-use rtl expressions to save memory.
8501 :
8502 : IN_CODE says what kind of expression we are processing. Normally, it is
8503 : SET. In a memory address it is MEM. When processing the arguments of
8504 : a comparison or a COMPARE against zero, it is COMPARE, or EQ if more
8505 : precisely it is an equality comparison against zero. */
8506 :
8507 : rtx
8508 469909628 : make_compound_operation (rtx x, enum rtx_code in_code)
8509 : {
8510 469909628 : enum rtx_code code = GET_CODE (x);
8511 469909628 : const char *fmt;
8512 469909628 : int i, j;
8513 469909628 : enum rtx_code next_code;
8514 469909628 : rtx new_rtx, tem;
8515 :
8516 : /* Select the code to be used in recursive calls. Once we are inside an
8517 : address, we stay there. If we have a comparison, set to COMPARE,
8518 : but once inside, go back to our default of SET. */
8519 :
8520 469909628 : next_code = (code == MEM ? MEM
8521 441840950 : : ((code == COMPARE || COMPARISON_P (x))
8522 462093029 : && XEXP (x, 1) == const0_rtx) ? COMPARE
8523 433875446 : : in_code == COMPARE || in_code == EQ ? SET : in_code);
8524 :
8525 469909628 : scalar_int_mode mode;
8526 469909628 : if (is_a <scalar_int_mode> (GET_MODE (x), &mode))
8527 : {
8528 275419182 : rtx new_rtx = make_compound_operation_int (mode, &x, in_code,
8529 : &next_code);
8530 275419182 : if (new_rtx)
8531 : return new_rtx;
8532 220887831 : code = GET_CODE (x);
8533 : }
8534 :
8535 : /* Now recursively process each operand of this operation. We need to
8536 : handle ZERO_EXTEND specially so that we don't lose track of the
8537 : inner mode. */
8538 415378277 : if (code == ZERO_EXTEND)
8539 : {
8540 3363606 : new_rtx = make_compound_operation (XEXP (x, 0), next_code);
8541 6727212 : tem = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
8542 3363606 : new_rtx, GET_MODE (XEXP (x, 0)));
8543 3363606 : if (tem)
8544 : return tem;
8545 3353439 : SUBST (XEXP (x, 0), new_rtx);
8546 3353439 : return x;
8547 : }
8548 :
8549 412014671 : fmt = GET_RTX_FORMAT (code);
8550 958412239 : for (i = 0; i < GET_RTX_LENGTH (code); i++)
8551 546397568 : if (fmt[i] == 'e')
8552 : {
8553 209740604 : new_rtx = make_compound_operation (XEXP (x, i), next_code);
8554 209740604 : SUBST (XEXP (x, i), new_rtx);
8555 : }
8556 336656964 : else if (fmt[i] == 'E')
8557 25291743 : for (j = 0; j < XVECLEN (x, i); j++)
8558 : {
8559 18304283 : new_rtx = make_compound_operation (XVECEXP (x, i, j), next_code);
8560 18304283 : SUBST (XVECEXP (x, i, j), new_rtx);
8561 : }
8562 :
8563 412014671 : maybe_swap_commutative_operands (x);
8564 412014671 : return x;
8565 : }
8566 :
8567 : /* Given M see if it is a value that would select a field of bits
8568 : within an item, but not the entire word. Return -1 if not.
8569 : Otherwise, return the starting position of the field, where 0 is the
8570 : low-order bit.
8571 :
8572 : *PLEN is set to the length of the field. */
8573 :
8574 : static int
8575 9672 : get_pos_from_mask (unsigned HOST_WIDE_INT m, unsigned HOST_WIDE_INT *plen)
8576 : {
8577 : /* Get the bit number of the first 1 bit from the right, -1 if none. */
8578 9672 : int pos = m ? ctz_hwi (m) : -1;
8579 9672 : int len = 0;
8580 :
8581 9672 : if (pos >= 0)
8582 : /* Now shift off the low-order zero bits and see if we have a
8583 : power of two minus 1. */
8584 9672 : len = exact_log2 ((m >> pos) + 1);
8585 :
8586 7101 : if (len <= 0)
8587 : pos = -1;
8588 :
8589 9672 : *plen = len;
8590 9672 : return pos;
8591 : }
8592 :
8593 : /* If X refers to a register that equals REG in value, replace these
8594 : references with REG. */
8595 : static rtx
8596 10083 : canon_reg_for_combine (rtx x, rtx reg)
8597 : {
8598 10083 : rtx op0, op1, op2;
8599 10083 : const char *fmt;
8600 10083 : int i;
8601 10083 : bool copied;
8602 :
8603 10083 : enum rtx_code code = GET_CODE (x);
8604 10083 : switch (GET_RTX_CLASS (code))
8605 : {
8606 0 : case RTX_UNARY:
8607 0 : op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8608 0 : if (op0 != XEXP (x, 0))
8609 0 : return simplify_gen_unary (GET_CODE (x), GET_MODE (x), op0,
8610 0 : GET_MODE (reg));
8611 : break;
8612 :
8613 1764 : case RTX_BIN_ARITH:
8614 1764 : case RTX_COMM_ARITH:
8615 1764 : op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8616 1764 : op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8617 1764 : if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
8618 0 : return simplify_gen_binary (GET_CODE (x), GET_MODE (x), op0, op1);
8619 : break;
8620 :
8621 14 : case RTX_COMPARE:
8622 14 : case RTX_COMM_COMPARE:
8623 14 : op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8624 14 : op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8625 14 : if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
8626 0 : return simplify_gen_relational (GET_CODE (x), GET_MODE (x),
8627 0 : GET_MODE (op0), op0, op1);
8628 : break;
8629 :
8630 0 : case RTX_TERNARY:
8631 0 : case RTX_BITFIELD_OPS:
8632 0 : op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8633 0 : op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8634 0 : op2 = canon_reg_for_combine (XEXP (x, 2), reg);
8635 0 : if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1) || op2 != XEXP (x, 2))
8636 0 : return simplify_gen_ternary (GET_CODE (x), GET_MODE (x),
8637 0 : GET_MODE (op0), op0, op1, op2);
8638 : /* FALLTHRU */
8639 :
8640 6026 : case RTX_OBJ:
8641 6026 : if (REG_P (x))
8642 : {
8643 6020 : if (rtx_equal_p (get_last_value (reg), x)
8644 6020 : || rtx_equal_p (reg, get_last_value (x)))
8645 0 : return reg;
8646 : else
8647 : break;
8648 : }
8649 :
8650 : /* fall through */
8651 :
8652 2285 : default:
8653 2285 : fmt = GET_RTX_FORMAT (code);
8654 2285 : copied = false;
8655 4631 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
8656 2346 : if (fmt[i] == 'e')
8657 : {
8658 61 : rtx op = canon_reg_for_combine (XEXP (x, i), reg);
8659 61 : if (op != XEXP (x, i))
8660 : {
8661 0 : if (!copied)
8662 : {
8663 0 : copied = true;
8664 0 : x = copy_rtx (x);
8665 : }
8666 0 : XEXP (x, i) = op;
8667 : }
8668 : }
8669 2285 : else if (fmt[i] == 'E')
8670 : {
8671 : int j;
8672 0 : for (j = 0; j < XVECLEN (x, i); j++)
8673 : {
8674 0 : rtx op = canon_reg_for_combine (XVECEXP (x, i, j), reg);
8675 0 : if (op != XVECEXP (x, i, j))
8676 : {
8677 0 : if (!copied)
8678 : {
8679 0 : copied = true;
8680 0 : x = copy_rtx (x);
8681 : }
8682 0 : XVECEXP (x, i, j) = op;
8683 : }
8684 : }
8685 : }
8686 :
8687 : break;
8688 : }
8689 :
8690 : return x;
8691 : }
8692 :
8693 : /* Return X converted to MODE. If the value is already truncated to
8694 : MODE we can just return a subreg even though in the general case we
8695 : would need an explicit truncation. */
8696 :
8697 : static rtx
8698 114926487 : gen_lowpart_or_truncate (machine_mode mode, rtx x)
8699 : {
8700 114926487 : if (!CONST_INT_P (x)
8701 109297402 : && partial_subreg_p (mode, GET_MODE (x))
8702 114926487 : && !TRULY_NOOP_TRUNCATION_MODES_P (mode, GET_MODE (x))
8703 114926487 : && !(REG_P (x) && reg_truncated_to_mode (mode, x)))
8704 : {
8705 : /* Bit-cast X into an integer mode. */
8706 0 : if (!SCALAR_INT_MODE_P (GET_MODE (x)))
8707 0 : x = gen_lowpart (int_mode_for_mode (GET_MODE (x)).require (), x);
8708 0 : x = simplify_gen_unary (TRUNCATE, int_mode_for_mode (mode).require (),
8709 0 : x, GET_MODE (x));
8710 : }
8711 :
8712 114926487 : return gen_lowpart (mode, x);
8713 : }
8714 :
8715 : /* See if X can be simplified knowing that we will only refer to it in
8716 : MODE and will only refer to those bits that are nonzero in MASK.
8717 : If other bits are being computed or if masking operations are done
8718 : that select a superset of the bits in MASK, they can sometimes be
8719 : ignored.
8720 :
8721 : Return a possibly simplified expression, but always convert X to
8722 : MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8723 :
8724 : If JUST_SELECT is true, don't optimize by noticing that bits in MASK
8725 : are all off in X. This is used when X will be complemented, by either
8726 : NOT, NEG, or XOR. */
8727 :
8728 : static rtx
8729 86093555 : force_to_mode (rtx x, machine_mode mode, unsigned HOST_WIDE_INT mask,
8730 : bool just_select)
8731 : {
8732 86093555 : enum rtx_code code = GET_CODE (x);
8733 86093555 : bool next_select = just_select || code == XOR || code == NOT || code == NEG;
8734 86093555 : machine_mode op_mode;
8735 86093555 : unsigned HOST_WIDE_INT nonzero;
8736 :
8737 : /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8738 : code below will do the wrong thing since the mode of such an
8739 : expression is VOIDmode.
8740 :
8741 : Also do nothing if X is a CLOBBER; this can happen if X was
8742 : the return value from a call to gen_lowpart. */
8743 86093555 : if (code == CALL || code == ASM_OPERANDS || code == CLOBBER)
8744 : return x;
8745 :
8746 : /* We want to perform the operation in its present mode unless we know
8747 : that the operation is valid in MODE, in which case we do the operation
8748 : in MODE. */
8749 140895962 : op_mode = ((GET_MODE_CLASS (mode) == GET_MODE_CLASS (GET_MODE (x))
8750 79929462 : && have_insn_for (code, mode))
8751 134810353 : ? mode : GET_MODE (x));
8752 :
8753 : /* It is not valid to do a right-shift in a narrower mode
8754 : than the one it came in with. */
8755 86015071 : if ((code == LSHIFTRT || code == ASHIFTRT)
8756 86015071 : && partial_subreg_p (mode, GET_MODE (x)))
8757 362547 : op_mode = GET_MODE (x);
8758 :
8759 : /* Truncate MASK to fit OP_MODE. */
8760 86015071 : if (op_mode)
8761 79960660 : mask &= GET_MODE_MASK (op_mode);
8762 :
8763 : /* Determine what bits of X are guaranteed to be (non)zero. */
8764 86015071 : nonzero = nonzero_bits (x, mode);
8765 :
8766 : /* If none of the bits in X are needed, return a zero. */
8767 86015071 : if (!just_select && (nonzero & mask) == 0 && !side_effects_p (x))
8768 614667 : x = const0_rtx;
8769 :
8770 : /* If X is a CONST_INT, return a new one. Do this here since the
8771 : test below will fail. */
8772 86015071 : if (CONST_INT_P (x))
8773 : {
8774 6249615 : if (SCALAR_INT_MODE_P (mode))
8775 6249615 : return gen_int_mode (INTVAL (x) & mask, mode);
8776 : else
8777 : {
8778 0 : x = GEN_INT (INTVAL (x) & mask);
8779 0 : return gen_lowpart_common (mode, x);
8780 : }
8781 : }
8782 :
8783 : /* If X is narrower than MODE and we want all the bits in X's mode, just
8784 : get X in the proper mode. */
8785 79765456 : if (paradoxical_subreg_p (mode, GET_MODE (x))
8786 79765456 : && (GET_MODE_MASK (GET_MODE (x)) & ~mask) == 0)
8787 3142411 : return gen_lowpart (mode, x);
8788 :
8789 : /* We can ignore the effect of a SUBREG if it narrows the mode or
8790 : if the constant masks to zero all the bits the mode doesn't have. */
8791 76623045 : if (GET_CODE (x) == SUBREG
8792 6843134 : && subreg_lowpart_p (x)
8793 83312412 : && (partial_subreg_p (x)
8794 5136328 : || (mask
8795 5136328 : & GET_MODE_MASK (GET_MODE (x))
8796 5136328 : & ~GET_MODE_MASK (GET_MODE (SUBREG_REG (x)))) == 0))
8797 6662230 : return force_to_mode (SUBREG_REG (x), mode, mask, next_select);
8798 :
8799 69960815 : scalar_int_mode int_mode, xmode;
8800 69960815 : if (is_a <scalar_int_mode> (mode, &int_mode)
8801 69960815 : && is_a <scalar_int_mode> (GET_MODE (x), &xmode))
8802 : /* OP_MODE is either MODE or XMODE, so it must be a scalar
8803 : integer too. */
8804 69930014 : return force_int_to_mode (x, int_mode, xmode,
8805 : as_a <scalar_int_mode> (op_mode),
8806 69930014 : mask, just_select);
8807 :
8808 30801 : return gen_lowpart_or_truncate (mode, x);
8809 : }
8810 :
8811 : /* Subroutine of force_to_mode that handles cases in which both X and
8812 : the result are scalar integers. MODE is the mode of the result,
8813 : XMODE is the mode of X, and OP_MODE says which of MODE or XMODE
8814 : is preferred for simplified versions of X. The other arguments
8815 : are as for force_to_mode. */
8816 :
8817 : static rtx
8818 69930014 : force_int_to_mode (rtx x, scalar_int_mode mode, scalar_int_mode xmode,
8819 : scalar_int_mode op_mode, unsigned HOST_WIDE_INT mask,
8820 : bool just_select)
8821 : {
8822 69930014 : enum rtx_code code = GET_CODE (x);
8823 69930014 : bool next_select = just_select || code == XOR || code == NOT || code == NEG;
8824 69930014 : unsigned HOST_WIDE_INT fuller_mask;
8825 69930014 : rtx op0, op1, temp;
8826 69930014 : poly_int64 const_op0;
8827 :
8828 : /* When we have an arithmetic operation, or a shift whose count we
8829 : do not know, we need to assume that all bits up to the highest-order
8830 : bit in MASK will be needed. This is how we form such a mask. */
8831 69930014 : if (mask & (HOST_WIDE_INT_1U << (HOST_BITS_PER_WIDE_INT - 1)))
8832 : fuller_mask = HOST_WIDE_INT_M1U;
8833 : else
8834 77378457 : fuller_mask = ((HOST_WIDE_INT_1U << (floor_log2 (mask) + 1)) - 1);
8835 :
8836 69930014 : switch (code)
8837 : {
8838 : case CLOBBER:
8839 : /* If X is a (clobber (const_int)), return it since we know we are
8840 : generating something that won't match. */
8841 : return x;
8842 :
8843 322856 : case SIGN_EXTEND:
8844 322856 : case ZERO_EXTEND:
8845 322856 : case ZERO_EXTRACT:
8846 322856 : case SIGN_EXTRACT:
8847 322856 : x = expand_compound_operation (x);
8848 322856 : if (GET_CODE (x) != code)
8849 192398 : return force_to_mode (x, mode, mask, next_select);
8850 : break;
8851 :
8852 141 : case TRUNCATE:
8853 : /* Similarly for a truncate. */
8854 141 : return force_to_mode (XEXP (x, 0), mode, mask, next_select);
8855 :
8856 3507406 : case AND:
8857 : /* If this is an AND with a constant, convert it into an AND
8858 : whose constant is the AND of that constant with MASK. If it
8859 : remains an AND of MASK, delete it since it is redundant. */
8860 :
8861 3507406 : if (CONST_INT_P (XEXP (x, 1)))
8862 : {
8863 5612544 : x = simplify_and_const_int (x, op_mode, XEXP (x, 0),
8864 2806272 : mask & INTVAL (XEXP (x, 1)));
8865 2806272 : xmode = op_mode;
8866 :
8867 : /* If X is still an AND, see if it is an AND with a mask that
8868 : is just some low-order bits. If so, and it is MASK, we don't
8869 : need it. */
8870 :
8871 2783103 : if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
8872 5589375 : && (INTVAL (XEXP (x, 1)) & GET_MODE_MASK (xmode)) == mask)
8873 19110 : x = XEXP (x, 0);
8874 :
8875 : /* If it remains an AND, try making another AND with the bits
8876 : in the mode mask that aren't in MASK turned on. If the
8877 : constant in the AND is wide enough, this might make a
8878 : cheaper constant. */
8879 :
8880 2763997 : if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
8881 2763993 : && GET_MODE_MASK (xmode) != mask
8882 2890042 : && HWI_COMPUTABLE_MODE_P (xmode))
8883 : {
8884 83770 : unsigned HOST_WIDE_INT cval
8885 83770 : = UINTVAL (XEXP (x, 1)) | (GET_MODE_MASK (xmode) & ~mask);
8886 83770 : rtx y;
8887 :
8888 83770 : y = simplify_gen_binary (AND, xmode, XEXP (x, 0),
8889 83770 : gen_int_mode (cval, xmode));
8890 83770 : if (set_src_cost (y, xmode, optimize_this_for_speed_p)
8891 83770 : < set_src_cost (x, xmode, optimize_this_for_speed_p))
8892 69644226 : x = y;
8893 : }
8894 :
8895 : break;
8896 : }
8897 :
8898 701134 : goto binop;
8899 :
8900 9703115 : case PLUS:
8901 : /* In (and (plus FOO C1) M), if M is a mask that just turns off
8902 : low-order bits (as in an alignment operation) and FOO is already
8903 : aligned to that boundary, mask C1 to that boundary as well.
8904 : This may eliminate that PLUS and, later, the AND. */
8905 :
8906 9703115 : {
8907 9703115 : unsigned int width = GET_MODE_PRECISION (mode);
8908 9703115 : unsigned HOST_WIDE_INT smask = mask;
8909 :
8910 : /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8911 : number, sign extend it. */
8912 :
8913 9703115 : if (width < HOST_BITS_PER_WIDE_INT
8914 3015104 : && (smask & (HOST_WIDE_INT_1U << (width - 1))) != 0)
8915 2683432 : smask |= HOST_WIDE_INT_M1U << width;
8916 :
8917 9703115 : if (CONST_INT_P (XEXP (x, 1))
8918 3619122 : && pow2p_hwi (- smask)
8919 3069524 : && (nonzero_bits (XEXP (x, 0), mode) & ~smask) == 0
8920 12389471 : && (INTVAL (XEXP (x, 1)) & ~smask) != 0)
8921 11468 : return force_to_mode (plus_constant (xmode, XEXP (x, 0),
8922 11468 : (INTVAL (XEXP (x, 1)) & smask)),
8923 : mode, smask, next_select);
8924 : }
8925 :
8926 : /* fall through */
8927 :
8928 11546489 : case MULT:
8929 : /* Substituting into the operands of a widening MULT is not likely to
8930 : create RTL matching a machine insn. */
8931 11546489 : if (code == MULT
8932 1854842 : && (GET_CODE (XEXP (x, 0)) == ZERO_EXTEND
8933 1854842 : || GET_CODE (XEXP (x, 0)) == SIGN_EXTEND)
8934 82230 : && (GET_CODE (XEXP (x, 1)) == ZERO_EXTEND
8935 82230 : || GET_CODE (XEXP (x, 1)) == SIGN_EXTEND)
8936 39776 : && REG_P (XEXP (XEXP (x, 0), 0))
8937 30098 : && REG_P (XEXP (XEXP (x, 1), 0)))
8938 22354 : return gen_lowpart_or_truncate (mode, x);
8939 :
8940 : /* For PLUS, MINUS and MULT, we need any bits less significant than the
8941 : most significant bit in MASK since carries from those bits will
8942 : affect the bits we are interested in. */
8943 11524135 : mask = fuller_mask;
8944 11524135 : goto binop;
8945 :
8946 2229018 : case MINUS:
8947 : /* If X is (minus C Y) where C's least set bit is larger than any bit
8948 : in the mask, then we may replace with (neg Y). */
8949 2229018 : if (poly_int_rtx_p (XEXP (x, 0), &const_op0)
8950 170475 : && known_alignment (poly_uint64 (const_op0)) > mask)
8951 : {
8952 19 : x = simplify_gen_unary (NEG, xmode, XEXP (x, 1), xmode);
8953 19 : return force_to_mode (x, mode, mask, next_select);
8954 : }
8955 :
8956 : /* Similarly, if C contains every bit in the fuller_mask, then we may
8957 : replace with (not Y). */
8958 2228999 : if (CONST_INT_P (XEXP (x, 0))
8959 170456 : && ((UINTVAL (XEXP (x, 0)) | fuller_mask) == UINTVAL (XEXP (x, 0))))
8960 : {
8961 455 : x = simplify_gen_unary (NOT, xmode, XEXP (x, 1), xmode);
8962 455 : return force_to_mode (x, mode, mask, next_select);
8963 : }
8964 :
8965 2228544 : mask = fuller_mask;
8966 2228544 : goto binop;
8967 :
8968 2519669 : case IOR:
8969 2519669 : case XOR:
8970 : /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8971 : LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8972 : operation which may be a bitfield extraction. Ensure that the
8973 : constant we form is not wider than the mode of X. */
8974 :
8975 2519669 : if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
8976 74366 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
8977 64285 : && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
8978 64285 : && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
8979 64285 : && CONST_INT_P (XEXP (x, 1))
8980 7717 : && ((INTVAL (XEXP (XEXP (x, 0), 1))
8981 15434 : + floor_log2 (INTVAL (XEXP (x, 1))))
8982 7717 : < GET_MODE_PRECISION (xmode))
8983 2519669 : && (UINTVAL (XEXP (x, 1))
8984 4453 : & ~nonzero_bits (XEXP (x, 0), xmode)) == 0)
8985 : {
8986 8170 : temp = gen_int_mode ((INTVAL (XEXP (x, 1)) & mask)
8987 4085 : << INTVAL (XEXP (XEXP (x, 0), 1)),
8988 : xmode);
8989 8170 : temp = simplify_gen_binary (GET_CODE (x), xmode,
8990 4085 : XEXP (XEXP (x, 0), 0), temp);
8991 8170 : x = simplify_gen_binary (LSHIFTRT, xmode, temp,
8992 4085 : XEXP (XEXP (x, 0), 1));
8993 4085 : return force_to_mode (x, mode, mask, next_select);
8994 : }
8995 :
8996 16969397 : binop:
8997 : /* For most binary operations, just propagate into the operation and
8998 : change the mode if we have an operation of that mode. */
8999 :
9000 16969397 : op0 = force_to_mode (XEXP (x, 0), mode, mask, next_select);
9001 16969397 : op1 = force_to_mode (XEXP (x, 1), mode, mask, next_select);
9002 :
9003 : /* If we ended up truncating both operands, truncate the result of the
9004 : operation instead. */
9005 16969397 : if (GET_CODE (op0) == TRUNCATE
9006 0 : && GET_CODE (op1) == TRUNCATE)
9007 : {
9008 0 : op0 = XEXP (op0, 0);
9009 0 : op1 = XEXP (op1, 0);
9010 : }
9011 :
9012 16969397 : op0 = gen_lowpart_or_truncate (op_mode, op0);
9013 16969397 : op1 = gen_lowpart_or_truncate (op_mode, op1);
9014 :
9015 16969397 : if (op_mode != xmode || op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
9016 : {
9017 2026162 : x = simplify_gen_binary (code, op_mode, op0, op1);
9018 2026162 : xmode = op_mode;
9019 : }
9020 : break;
9021 :
9022 4240262 : case ASHIFT:
9023 : /* For left shifts, do the same, but just for the first operand.
9024 : However, we cannot do anything with shifts where we cannot
9025 : guarantee that the counts are smaller than the size of the mode
9026 : because such a count will have a different meaning in a
9027 : wider mode. */
9028 :
9029 4043257 : if (! (CONST_INT_P (XEXP (x, 1))
9030 4043282 : && INTVAL (XEXP (x, 1)) >= 0
9031 4043257 : && INTVAL (XEXP (x, 1)) < GET_MODE_PRECISION (mode))
9032 4242854 : && ! (GET_MODE (XEXP (x, 1)) != VOIDmode
9033 196980 : && (nonzero_bits (XEXP (x, 1), GET_MODE (XEXP (x, 1)))
9034 196980 : < (unsigned HOST_WIDE_INT) GET_MODE_PRECISION (mode))))
9035 : break;
9036 :
9037 : /* If the shift count is a constant and we can do arithmetic in
9038 : the mode of the shift, refine which bits we need. Otherwise, use the
9039 : conservative form of the mask. */
9040 4105781 : if (CONST_INT_P (XEXP (x, 1))
9041 4040690 : && INTVAL (XEXP (x, 1)) >= 0
9042 4040690 : && INTVAL (XEXP (x, 1)) < GET_MODE_PRECISION (op_mode)
9043 8146471 : && HWI_COMPUTABLE_MODE_P (op_mode))
9044 4039748 : mask >>= INTVAL (XEXP (x, 1));
9045 : else
9046 : mask = fuller_mask;
9047 :
9048 4105781 : op0 = gen_lowpart_or_truncate (op_mode,
9049 : force_to_mode (XEXP (x, 0), mode,
9050 : mask, next_select));
9051 :
9052 4105781 : if (op_mode != xmode || op0 != XEXP (x, 0))
9053 : {
9054 1012754 : x = simplify_gen_binary (code, op_mode, op0, XEXP (x, 1));
9055 1012754 : xmode = op_mode;
9056 : }
9057 : break;
9058 :
9059 3145041 : case LSHIFTRT:
9060 : /* Here we can only do something if the shift count is a constant,
9061 : this shift constant is valid for the host, and we can do arithmetic
9062 : in OP_MODE. */
9063 :
9064 3145041 : if (CONST_INT_P (XEXP (x, 1))
9065 3035158 : && INTVAL (XEXP (x, 1)) >= 0
9066 3035157 : && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
9067 6180180 : && HWI_COMPUTABLE_MODE_P (op_mode))
9068 : {
9069 3031141 : rtx inner = XEXP (x, 0);
9070 3031141 : unsigned HOST_WIDE_INT inner_mask;
9071 :
9072 : /* Select the mask of the bits we need for the shift operand. */
9073 3031141 : inner_mask = mask << INTVAL (XEXP (x, 1));
9074 :
9075 : /* We can only change the mode of the shift if we can do arithmetic
9076 : in the mode of the shift and INNER_MASK is no wider than the
9077 : width of X's mode. */
9078 3031141 : if ((inner_mask & ~GET_MODE_MASK (xmode)) != 0)
9079 302592 : op_mode = xmode;
9080 :
9081 3031141 : inner = force_to_mode (inner, op_mode, inner_mask, next_select);
9082 :
9083 3031141 : if (xmode != op_mode || inner != XEXP (x, 0))
9084 : {
9085 816829 : x = simplify_gen_binary (LSHIFTRT, op_mode, inner, XEXP (x, 1));
9086 816829 : xmode = op_mode;
9087 : }
9088 : }
9089 :
9090 : /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
9091 : shift and AND produces only copies of the sign bit (C2 is one less
9092 : than a power of two), we can do this with just a shift. */
9093 :
9094 3145041 : if (GET_CODE (x) == LSHIFTRT
9095 3144987 : && CONST_INT_P (XEXP (x, 1))
9096 : /* The shift puts one of the sign bit copies in the least significant
9097 : bit. */
9098 6070208 : && ((INTVAL (XEXP (x, 1))
9099 3035104 : + num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0))))
9100 3035104 : >= GET_MODE_PRECISION (xmode))
9101 246963 : && pow2p_hwi (mask + 1)
9102 : /* Number of bits left after the shift must be more than the mask
9103 : needs. */
9104 74284 : && ((INTVAL (XEXP (x, 1)) + exact_log2 (mask + 1))
9105 74284 : <= GET_MODE_PRECISION (xmode))
9106 : /* Must be more sign bit copies than the mask needs. */
9107 3174475 : && ((int) num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0)))
9108 29434 : >= exact_log2 (mask + 1)))
9109 : {
9110 29434 : int nbits = GET_MODE_PRECISION (xmode) - exact_log2 (mask + 1);
9111 29434 : x = simplify_gen_binary (LSHIFTRT, xmode, XEXP (x, 0),
9112 29434 : gen_int_shift_amount (xmode, nbits));
9113 : }
9114 3145041 : goto shiftrt;
9115 :
9116 1887705 : case ASHIFTRT:
9117 : /* If we are just looking for the sign bit, we don't need this shift at
9118 : all, even if it has a variable count. */
9119 1887705 : if (val_signbit_p (xmode, mask))
9120 1282 : return force_to_mode (XEXP (x, 0), mode, mask, next_select);
9121 :
9122 : /* If this is a shift by a constant, get a mask that contains those bits
9123 : that are not copies of the sign bit. We then have two cases: If
9124 : MASK only includes those bits, this can be a logical shift, which may
9125 : allow simplifications. If MASK is a single-bit field not within
9126 : those bits, we are requesting a copy of the sign bit and hence can
9127 : shift the sign bit to the appropriate location. */
9128 :
9129 1886423 : if (CONST_INT_P (XEXP (x, 1)) && INTVAL (XEXP (x, 1)) >= 0
9130 1848281 : && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT)
9131 : {
9132 1848170 : unsigned HOST_WIDE_INT nonzero;
9133 1848170 : int i;
9134 :
9135 : /* If the considered data is wider than HOST_WIDE_INT, we can't
9136 : represent a mask for all its bits in a single scalar.
9137 : But we only care about the lower bits, so calculate these. */
9138 :
9139 1848170 : if (GET_MODE_PRECISION (xmode) > HOST_BITS_PER_WIDE_INT)
9140 : {
9141 408 : nonzero = HOST_WIDE_INT_M1U;
9142 :
9143 : /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
9144 : is the number of bits a full-width mask would have set.
9145 : We need only shift if these are fewer than nonzero can
9146 : hold. If not, we must keep all bits set in nonzero. */
9147 :
9148 408 : if (GET_MODE_PRECISION (xmode) - INTVAL (XEXP (x, 1))
9149 : < HOST_BITS_PER_WIDE_INT)
9150 0 : nonzero >>= INTVAL (XEXP (x, 1))
9151 0 : + HOST_BITS_PER_WIDE_INT
9152 0 : - GET_MODE_PRECISION (xmode);
9153 : }
9154 : else
9155 : {
9156 1847762 : nonzero = GET_MODE_MASK (xmode);
9157 1847762 : nonzero >>= INTVAL (XEXP (x, 1));
9158 : }
9159 :
9160 1848170 : if ((mask & ~nonzero) == 0)
9161 : {
9162 44423 : x = simplify_shift_const (NULL_RTX, LSHIFTRT, xmode,
9163 : XEXP (x, 0), INTVAL (XEXP (x, 1)));
9164 44423 : if (GET_CODE (x) != ASHIFTRT)
9165 44423 : return force_to_mode (x, mode, mask, next_select);
9166 : }
9167 :
9168 1803747 : else if ((i = exact_log2 (mask)) >= 0)
9169 : {
9170 73 : x = simplify_shift_const
9171 146 : (NULL_RTX, LSHIFTRT, xmode, XEXP (x, 0),
9172 73 : GET_MODE_PRECISION (xmode) - 1 - i);
9173 :
9174 73 : if (GET_CODE (x) != ASHIFTRT)
9175 73 : return force_to_mode (x, mode, mask, next_select);
9176 : }
9177 : }
9178 :
9179 : /* If MASK is 1, convert this to an LSHIFTRT. This can be done
9180 : even if the shift count isn't a constant. */
9181 1841927 : if (mask == 1)
9182 3147 : x = simplify_gen_binary (LSHIFTRT, xmode, XEXP (x, 0), XEXP (x, 1));
9183 :
9184 1838780 : shiftrt:
9185 :
9186 : /* If this is a zero- or sign-extension operation that just affects bits
9187 : we don't care about, remove it. Be sure the call above returned
9188 : something that is still a shift. */
9189 :
9190 4986968 : if ((GET_CODE (x) == LSHIFTRT || GET_CODE (x) == ASHIFTRT)
9191 4986914 : && CONST_INT_P (XEXP (x, 1))
9192 4838889 : && INTVAL (XEXP (x, 1)) >= 0
9193 4838888 : && (INTVAL (XEXP (x, 1))
9194 9677776 : <= GET_MODE_PRECISION (xmode) - (floor_log2 (mask) + 1))
9195 1777487 : && GET_CODE (XEXP (x, 0)) == ASHIFT
9196 4987906 : && XEXP (XEXP (x, 0), 1) == XEXP (x, 1))
9197 770 : return force_to_mode (XEXP (XEXP (x, 0), 0), mode, mask, next_select);
9198 :
9199 : break;
9200 :
9201 38526 : case ROTATE:
9202 38526 : case ROTATERT:
9203 : /* If the shift count is constant and we can do computations
9204 : in the mode of X, compute where the bits we care about are.
9205 : Otherwise, we can't do anything. Don't change the mode of
9206 : the shift or propagate MODE into the shift, though. */
9207 38526 : if (CONST_INT_P (XEXP (x, 1))
9208 29423 : && INTVAL (XEXP (x, 1)) >= 0)
9209 : {
9210 29421 : temp = simplify_binary_operation (code == ROTATE ? ROTATERT : ROTATE,
9211 29421 : xmode, gen_int_mode (mask, xmode),
9212 : XEXP (x, 1));
9213 29421 : if (temp && CONST_INT_P (temp))
9214 29421 : x = simplify_gen_binary (code, xmode,
9215 : force_to_mode (XEXP (x, 0), xmode,
9216 29421 : INTVAL (temp), next_select),
9217 : XEXP (x, 1));
9218 : }
9219 : break;
9220 :
9221 149914 : case NEG:
9222 : /* If we just want the low-order bit, the NEG isn't needed since it
9223 : won't change the low-order bit. */
9224 149914 : if (mask == 1)
9225 286 : return force_to_mode (XEXP (x, 0), mode, mask, just_select);
9226 :
9227 : /* We need any bits less significant than the most significant bit in
9228 : MASK since carries from those bits will affect the bits we are
9229 : interested in. */
9230 149628 : mask = fuller_mask;
9231 149628 : goto unop;
9232 :
9233 424537 : case NOT:
9234 : /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
9235 : same as the XOR case above. Ensure that the constant we form is not
9236 : wider than the mode of X. */
9237 :
9238 424537 : if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
9239 15866 : && CONST_INT_P (XEXP (XEXP (x, 0), 1))
9240 15234 : && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
9241 30468 : && (INTVAL (XEXP (XEXP (x, 0), 1)) + floor_log2 (mask)
9242 15234 : < GET_MODE_PRECISION (xmode))
9243 432571 : && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT)
9244 : {
9245 8034 : temp = gen_int_mode (mask << INTVAL (XEXP (XEXP (x, 0), 1)), xmode);
9246 8034 : temp = simplify_gen_binary (XOR, xmode, XEXP (XEXP (x, 0), 0), temp);
9247 16068 : x = simplify_gen_binary (LSHIFTRT, xmode,
9248 8034 : temp, XEXP (XEXP (x, 0), 1));
9249 :
9250 8034 : return force_to_mode (x, mode, mask, next_select);
9251 : }
9252 :
9253 : /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
9254 : use the full mask inside the NOT. */
9255 : mask = fuller_mask;
9256 :
9257 566131 : unop:
9258 566131 : op0 = gen_lowpart_or_truncate (op_mode,
9259 : force_to_mode (XEXP (x, 0), mode, mask,
9260 : next_select));
9261 566131 : if (op_mode != xmode || op0 != XEXP (x, 0))
9262 : {
9263 65567 : x = simplify_gen_unary (code, op_mode, op0, op_mode);
9264 65567 : xmode = op_mode;
9265 : }
9266 : break;
9267 :
9268 544069 : case NE:
9269 : /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
9270 : in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
9271 : which is equal to STORE_FLAG_VALUE. */
9272 544069 : if ((mask & ~STORE_FLAG_VALUE) == 0
9273 3002 : && XEXP (x, 1) == const0_rtx
9274 2981 : && GET_MODE (XEXP (x, 0)) == mode
9275 2 : && pow2p_hwi (nonzero_bits (XEXP (x, 0), mode))
9276 544069 : && (nonzero_bits (XEXP (x, 0), mode)
9277 : == (unsigned HOST_WIDE_INT) STORE_FLAG_VALUE))
9278 0 : return force_to_mode (XEXP (x, 0), mode, mask, next_select);
9279 :
9280 : break;
9281 :
9282 1410605 : case IF_THEN_ELSE:
9283 : /* We have no way of knowing if the IF_THEN_ELSE can itself be
9284 : written in a narrower mode. We play it safe and do not do so. */
9285 :
9286 1410605 : op0 = gen_lowpart_or_truncate (xmode,
9287 : force_to_mode (XEXP (x, 1), mode,
9288 : mask, next_select));
9289 1410605 : op1 = gen_lowpart_or_truncate (xmode,
9290 : force_to_mode (XEXP (x, 2), mode,
9291 : mask, next_select));
9292 1410605 : if (op0 != XEXP (x, 1) || op1 != XEXP (x, 2))
9293 236674 : x = simplify_gen_ternary (IF_THEN_ELSE, xmode,
9294 236674 : GET_MODE (XEXP (x, 0)), XEXP (x, 0),
9295 : op0, op1);
9296 : break;
9297 :
9298 : default:
9299 : break;
9300 : }
9301 :
9302 : /* Ensure we return a value of the proper mode. */
9303 69644226 : return gen_lowpart_or_truncate (mode, x);
9304 : }
9305 :
9306 : /* Return nonzero if X is an expression that has one of two values depending on
9307 : whether some other value is zero or nonzero. In that case, we return the
9308 : value that is being tested, *PTRUE is set to the value if the rtx being
9309 : returned has a nonzero value, and *PFALSE is set to the other alternative.
9310 :
9311 : If we return zero, we set *PTRUE and *PFALSE to X. */
9312 :
9313 : static rtx
9314 231272330 : if_then_else_cond (rtx x, rtx *ptrue, rtx *pfalse)
9315 : {
9316 231272330 : machine_mode mode = GET_MODE (x);
9317 231272330 : enum rtx_code code = GET_CODE (x);
9318 231272330 : rtx cond0, cond1, true0, true1, false0, false1;
9319 231272330 : unsigned HOST_WIDE_INT nz;
9320 231272330 : scalar_int_mode int_mode;
9321 :
9322 : /* If we are comparing a value against zero, we are done. */
9323 231272330 : if ((code == NE || code == EQ)
9324 2614050 : && XEXP (x, 1) == const0_rtx)
9325 : {
9326 1556164 : *ptrue = (code == NE) ? const_true_rtx : const0_rtx;
9327 1556164 : *pfalse = (code == NE) ? const0_rtx : const_true_rtx;
9328 1556164 : return XEXP (x, 0);
9329 : }
9330 :
9331 : /* If this is a unary operation whose operand has one of two values, apply
9332 : our opcode to compute those values. */
9333 229716166 : else if (UNARY_P (x)
9334 229716166 : && (cond0 = if_then_else_cond (XEXP (x, 0), &true0, &false0)) != 0)
9335 : {
9336 428998 : *ptrue = simplify_gen_unary (code, mode, true0, GET_MODE (XEXP (x, 0)));
9337 857996 : *pfalse = simplify_gen_unary (code, mode, false0,
9338 428998 : GET_MODE (XEXP (x, 0)));
9339 428998 : return cond0;
9340 : }
9341 :
9342 : /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
9343 : make can't possibly match and would suppress other optimizations. */
9344 229287168 : else if (code == COMPARE)
9345 : ;
9346 :
9347 : /* If this is a binary operation, see if either side has only one of two
9348 : values. If either one does or if both do and they are conditional on
9349 : the same value, compute the new true and false values. */
9350 224978364 : else if (BINARY_P (x))
9351 : {
9352 83928575 : rtx op0 = XEXP (x, 0);
9353 83928575 : rtx op1 = XEXP (x, 1);
9354 83928575 : cond0 = if_then_else_cond (op0, &true0, &false0);
9355 83928575 : cond1 = if_then_else_cond (op1, &true1, &false1);
9356 :
9357 521859 : if ((cond0 != 0 && cond1 != 0 && !rtx_equal_p (cond0, cond1))
9358 84396347 : && (REG_P (op0) || REG_P (op1)))
9359 : {
9360 : /* Try to enable a simplification by undoing work done by
9361 : if_then_else_cond if it converted a REG into something more
9362 : complex. */
9363 400795 : if (REG_P (op0))
9364 : {
9365 104975 : cond0 = 0;
9366 104975 : true0 = false0 = op0;
9367 : }
9368 : else
9369 : {
9370 295820 : cond1 = 0;
9371 295820 : true1 = false1 = op1;
9372 : }
9373 : }
9374 :
9375 83928575 : if ((cond0 != 0 || cond1 != 0)
9376 83928575 : && ! (cond0 != 0 && cond1 != 0 && !rtx_equal_p (cond0, cond1)))
9377 : {
9378 : /* If if_then_else_cond returned zero, then true/false are the
9379 : same rtl. We must copy one of them to prevent invalid rtl
9380 : sharing. */
9381 4221109 : if (cond0 == 0)
9382 1203581 : true0 = copy_rtx (true0);
9383 3017528 : else if (cond1 == 0)
9384 2963441 : true1 = copy_rtx (true1);
9385 :
9386 4221109 : if (COMPARISON_P (x))
9387 : {
9388 358221 : *ptrue = simplify_gen_relational (code, mode, VOIDmode,
9389 : true0, true1);
9390 358221 : *pfalse = simplify_gen_relational (code, mode, VOIDmode,
9391 : false0, false1);
9392 : }
9393 : else
9394 : {
9395 3862888 : *ptrue = simplify_gen_binary (code, mode, true0, true1);
9396 3862888 : *pfalse = simplify_gen_binary (code, mode, false0, false1);
9397 : }
9398 :
9399 5424690 : return cond0 ? cond0 : cond1;
9400 : }
9401 :
9402 : /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
9403 : operands is zero when the other is nonzero, and vice-versa,
9404 : and STORE_FLAG_VALUE is 1 or -1. */
9405 :
9406 79707466 : if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
9407 79707466 : && (code == PLUS || code == IOR || code == XOR || code == MINUS
9408 : || code == UMAX)
9409 33699755 : && GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
9410 : {
9411 35889 : rtx op0 = XEXP (XEXP (x, 0), 1);
9412 35889 : rtx op1 = XEXP (XEXP (x, 1), 1);
9413 :
9414 35889 : cond0 = XEXP (XEXP (x, 0), 0);
9415 35889 : cond1 = XEXP (XEXP (x, 1), 0);
9416 :
9417 35889 : if (COMPARISON_P (cond0)
9418 1 : && COMPARISON_P (cond1)
9419 0 : && SCALAR_INT_MODE_P (mode)
9420 0 : && ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
9421 0 : && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
9422 0 : && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
9423 0 : || ((swap_condition (GET_CODE (cond0))
9424 0 : == reversed_comparison_code (cond1, NULL))
9425 0 : && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
9426 0 : && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
9427 35889 : && ! side_effects_p (x))
9428 : {
9429 0 : *ptrue = simplify_gen_binary (MULT, mode, op0, const_true_rtx);
9430 0 : *pfalse = simplify_gen_binary (MULT, mode,
9431 : (code == MINUS
9432 0 : ? simplify_gen_unary (NEG, mode,
9433 : op1, mode)
9434 : : op1),
9435 : const_true_rtx);
9436 0 : return cond0;
9437 : }
9438 : }
9439 :
9440 : /* Similarly for MULT, AND and UMIN, except that for these the result
9441 : is always zero. */
9442 79707466 : if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
9443 79707466 : && (code == MULT || code == AND || code == UMIN)
9444 11221061 : && GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
9445 : {
9446 886 : cond0 = XEXP (XEXP (x, 0), 0);
9447 886 : cond1 = XEXP (XEXP (x, 1), 0);
9448 :
9449 886 : if (COMPARISON_P (cond0)
9450 0 : && COMPARISON_P (cond1)
9451 0 : && ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
9452 0 : && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
9453 0 : && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
9454 0 : || ((swap_condition (GET_CODE (cond0))
9455 0 : == reversed_comparison_code (cond1, NULL))
9456 0 : && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
9457 0 : && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
9458 886 : && ! side_effects_p (x))
9459 : {
9460 0 : *ptrue = *pfalse = const0_rtx;
9461 0 : return cond0;
9462 : }
9463 : }
9464 : }
9465 :
9466 141049789 : else if (code == IF_THEN_ELSE)
9467 : {
9468 : /* If we have IF_THEN_ELSE already, extract the condition and
9469 : canonicalize it if it is NE or EQ. */
9470 624588 : cond0 = XEXP (x, 0);
9471 624588 : *ptrue = XEXP (x, 1), *pfalse = XEXP (x, 2);
9472 624588 : if (GET_CODE (cond0) == NE && XEXP (cond0, 1) == const0_rtx)
9473 256510 : return XEXP (cond0, 0);
9474 368078 : else if (GET_CODE (cond0) == EQ && XEXP (cond0, 1) == const0_rtx)
9475 : {
9476 24368 : *ptrue = XEXP (x, 2), *pfalse = XEXP (x, 1);
9477 24368 : return XEXP (cond0, 0);
9478 : }
9479 : else
9480 : return cond0;
9481 : }
9482 :
9483 : /* If X is a SUBREG, we can narrow both the true and false values
9484 : if the inner expression, if there is a condition. */
9485 140425201 : else if (code == SUBREG
9486 140425201 : && (cond0 = if_then_else_cond (SUBREG_REG (x), &true0,
9487 : &false0)) != 0)
9488 : {
9489 691692 : true0 = simplify_gen_subreg (mode, true0,
9490 345846 : GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
9491 691692 : false0 = simplify_gen_subreg (mode, false0,
9492 345846 : GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
9493 345846 : if (true0 && false0)
9494 : {
9495 345846 : *ptrue = true0;
9496 345846 : *pfalse = false0;
9497 345846 : return cond0;
9498 : }
9499 : }
9500 :
9501 : /* If X is a constant, this isn't special and will cause confusions
9502 : if we treat it as such. Likewise if it is equivalent to a constant. */
9503 140079355 : else if (CONSTANT_P (x)
9504 140079355 : || ((cond0 = get_last_value (x)) != 0 && CONSTANT_P (cond0)))
9505 : ;
9506 :
9507 : /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
9508 : will be least confusing to the rest of the compiler. */
9509 93745047 : else if (mode == BImode)
9510 : {
9511 0 : *ptrue = GEN_INT (STORE_FLAG_VALUE), *pfalse = const0_rtx;
9512 0 : return x;
9513 : }
9514 :
9515 : /* If X is known to be either 0 or -1, those are the true and
9516 : false values when testing X. */
9517 93745047 : else if (x == constm1_rtx || x == const0_rtx
9518 93745047 : || (is_a <scalar_int_mode> (mode, &int_mode)
9519 66462743 : && (num_sign_bit_copies (x, int_mode)
9520 66462743 : == GET_MODE_PRECISION (int_mode))))
9521 : {
9522 776619 : *ptrue = constm1_rtx, *pfalse = const0_rtx;
9523 776619 : return x;
9524 : }
9525 :
9526 : /* Likewise for 0 or a single bit. */
9527 92968428 : else if (HWI_COMPUTABLE_MODE_P (mode)
9528 62362272 : && pow2p_hwi (nz = nonzero_bits (x, mode)))
9529 : {
9530 1773447 : *ptrue = gen_int_mode (nz, mode), *pfalse = const0_rtx;
9531 1773447 : return x;
9532 : }
9533 :
9534 : /* Otherwise fail; show no condition with true and false values the same. */
9535 221545559 : *ptrue = *pfalse = x;
9536 221545559 : return 0;
9537 : }
9538 :
9539 : /* Return the value of expression X given the fact that condition COND
9540 : is known to be true when applied to REG as its first operand and VAL
9541 : as its second. X is known to not be shared and so can be modified in
9542 : place.
9543 :
9544 : We only handle the simplest cases, and specifically those cases that
9545 : arise with IF_THEN_ELSE expressions. */
9546 :
9547 : static rtx
9548 612831 : known_cond (rtx x, enum rtx_code cond, rtx reg, rtx val)
9549 : {
9550 612831 : enum rtx_code code = GET_CODE (x);
9551 612831 : const char *fmt;
9552 612831 : int i, j;
9553 :
9554 612831 : if (side_effects_p (x))
9555 : return x;
9556 :
9557 : /* If either operand of the condition is a floating point value,
9558 : then we have to avoid collapsing an EQ comparison. */
9559 612831 : if (cond == EQ
9560 123267 : && rtx_equal_p (x, reg)
9561 82152 : && ! FLOAT_MODE_P (GET_MODE (x))
9562 694983 : && ! FLOAT_MODE_P (GET_MODE (val)))
9563 : return val;
9564 :
9565 530679 : if (cond == UNEQ && rtx_equal_p (x, reg))
9566 : return val;
9567 :
9568 : /* If X is (abs REG) and we know something about REG's relationship
9569 : with zero, we may be able to simplify this. */
9570 :
9571 530679 : if (code == ABS && rtx_equal_p (XEXP (x, 0), reg) && val == const0_rtx)
9572 3 : switch (cond)
9573 : {
9574 1 : case GE: case GT: case EQ:
9575 1 : return XEXP (x, 0);
9576 2 : case LT: case LE:
9577 4 : return simplify_gen_unary (NEG, GET_MODE (XEXP (x, 0)),
9578 : XEXP (x, 0),
9579 2 : GET_MODE (XEXP (x, 0)));
9580 : default:
9581 : break;
9582 : }
9583 :
9584 : /* The only other cases we handle are MIN, MAX, and comparisons if the
9585 : operands are the same as REG and VAL. */
9586 :
9587 530676 : else if (COMPARISON_P (x) || COMMUTATIVE_ARITH_P (x))
9588 : {
9589 242288 : if (rtx_equal_p (XEXP (x, 0), val))
9590 : {
9591 2 : std::swap (val, reg);
9592 2 : cond = swap_condition (cond);
9593 : }
9594 :
9595 242288 : if (rtx_equal_p (XEXP (x, 0), reg) && rtx_equal_p (XEXP (x, 1), val))
9596 : {
9597 220396 : if (COMPARISON_P (x))
9598 : {
9599 220179 : if (comparison_dominates_p (cond, code))
9600 348 : return VECTOR_MODE_P (GET_MODE (x)) ? x : const_true_rtx;
9601 :
9602 219831 : code = reversed_comparison_code (x, NULL);
9603 219831 : if (code != UNKNOWN
9604 219831 : && comparison_dominates_p (cond, code))
9605 42 : return CONST0_RTX (GET_MODE (x));
9606 : else
9607 219789 : return x;
9608 : }
9609 217 : else if (code == SMAX || code == SMIN
9610 217 : || code == UMIN || code == UMAX)
9611 : {
9612 39 : int unsignedp = (code == UMIN || code == UMAX);
9613 :
9614 : /* Do not reverse the condition when it is NE or EQ.
9615 : This is because we cannot conclude anything about
9616 : the value of 'SMAX (x, y)' when x is not equal to y,
9617 : but we can when x equals y. */
9618 39 : if ((code == SMAX || code == UMAX)
9619 36 : && ! (cond == EQ || cond == NE))
9620 3 : cond = reverse_condition (cond);
9621 :
9622 6 : switch (cond)
9623 : {
9624 2 : case GE: case GT:
9625 2 : return unsignedp ? x : XEXP (x, 1);
9626 4 : case LE: case LT:
9627 4 : return unsignedp ? x : XEXP (x, 0);
9628 0 : case GEU: case GTU:
9629 0 : return unsignedp ? XEXP (x, 1) : x;
9630 0 : case LEU: case LTU:
9631 0 : return unsignedp ? XEXP (x, 0) : x;
9632 : default:
9633 : break;
9634 : }
9635 : }
9636 : }
9637 : }
9638 288388 : else if (code == SUBREG)
9639 : {
9640 9754 : machine_mode inner_mode = GET_MODE (SUBREG_REG (x));
9641 9754 : rtx new_rtx, r = known_cond (SUBREG_REG (x), cond, reg, val);
9642 :
9643 9754 : if (SUBREG_REG (x) != r)
9644 : {
9645 : /* We must simplify subreg here, before we lose track of the
9646 : original inner_mode. */
9647 34 : new_rtx = simplify_subreg (GET_MODE (x), r,
9648 17 : inner_mode, SUBREG_BYTE (x));
9649 17 : if (new_rtx)
9650 : return new_rtx;
9651 : else
9652 17 : SUBST (SUBREG_REG (x), r);
9653 : }
9654 :
9655 9754 : return x;
9656 : }
9657 : /* We don't have to handle SIGN_EXTEND here, because even in the
9658 : case of replacing something with a modeless CONST_INT, a
9659 : CONST_INT is already (supposed to be) a valid sign extension for
9660 : its narrower mode, which implies it's already properly
9661 : sign-extended for the wider mode. Now, for ZERO_EXTEND, the
9662 : story is different. */
9663 278634 : else if (code == ZERO_EXTEND)
9664 : {
9665 1276 : machine_mode inner_mode = GET_MODE (XEXP (x, 0));
9666 1276 : rtx new_rtx, r = known_cond (XEXP (x, 0), cond, reg, val);
9667 :
9668 1276 : if (XEXP (x, 0) != r)
9669 : {
9670 : /* We must simplify the zero_extend here, before we lose
9671 : track of the original inner_mode. */
9672 0 : new_rtx = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
9673 : r, inner_mode);
9674 0 : if (new_rtx)
9675 : return new_rtx;
9676 : else
9677 0 : SUBST (XEXP (x, 0), r);
9678 : }
9679 :
9680 1276 : return x;
9681 : }
9682 :
9683 299461 : fmt = GET_RTX_FORMAT (code);
9684 688636 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
9685 : {
9686 389175 : if (fmt[i] == 'e')
9687 184136 : SUBST (XEXP (x, i), known_cond (XEXP (x, i), cond, reg, val));
9688 205039 : else if (fmt[i] == 'E')
9689 15148 : for (j = XVECLEN (x, i) - 1; j >= 0; j--)
9690 12212 : SUBST (XVECEXP (x, i, j), known_cond (XVECEXP (x, i, j),
9691 : cond, reg, val));
9692 : }
9693 :
9694 : return x;
9695 : }
9696 :
9697 : /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9698 : assignment as a field assignment. */
9699 :
9700 : static bool
9701 576400 : rtx_equal_for_field_assignment_p (rtx x, rtx y, bool widen_x)
9702 : {
9703 576400 : if (widen_x && GET_MODE (x) != GET_MODE (y))
9704 : {
9705 56057 : if (paradoxical_subreg_p (GET_MODE (x), GET_MODE (y)))
9706 : return false;
9707 56057 : if (BYTES_BIG_ENDIAN != WORDS_BIG_ENDIAN)
9708 : return false;
9709 56057 : x = adjust_address_nv (x, GET_MODE (y),
9710 : byte_lowpart_offset (GET_MODE (y),
9711 : GET_MODE (x)));
9712 : }
9713 :
9714 576400 : if (x == y || rtx_equal_p (x, y))
9715 10363 : return true;
9716 :
9717 566037 : if (x == 0 || y == 0 || GET_MODE (x) != GET_MODE (y))
9718 : return false;
9719 :
9720 : /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9721 : Note that all SUBREGs of MEM are paradoxical; otherwise they
9722 : would have been rewritten. */
9723 95284 : if (MEM_P (x) && GET_CODE (y) == SUBREG
9724 6134 : && MEM_P (SUBREG_REG (y))
9725 566037 : && rtx_equal_p (SUBREG_REG (y),
9726 0 : gen_lowpart (GET_MODE (SUBREG_REG (y)), x)))
9727 : return true;
9728 :
9729 66322 : if (MEM_P (y) && GET_CODE (x) == SUBREG
9730 4915 : && MEM_P (SUBREG_REG (x))
9731 566225 : && rtx_equal_p (SUBREG_REG (x),
9732 188 : gen_lowpart (GET_MODE (SUBREG_REG (x)), y)))
9733 : return true;
9734 :
9735 : /* We used to see if get_last_value of X and Y were the same but that's
9736 : not correct. In one direction, we'll cause the assignment to have
9737 : the wrong destination and in the case, we'll import a register into this
9738 : insn that might have already have been dead. So fail if none of the
9739 : above cases are true. */
9740 : return false;
9741 : }
9742 :
9743 : /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9744 : Return that assignment if so.
9745 :
9746 : We only handle the most common cases. */
9747 :
9748 : static rtx
9749 46571070 : make_field_assignment (rtx x)
9750 : {
9751 46571070 : rtx dest = SET_DEST (x);
9752 46571070 : rtx src = SET_SRC (x);
9753 46571070 : rtx assign;
9754 46571070 : rtx rhs, lhs;
9755 46571070 : HOST_WIDE_INT c1;
9756 46571070 : HOST_WIDE_INT pos;
9757 46571070 : unsigned HOST_WIDE_INT len;
9758 46571070 : rtx other;
9759 :
9760 : /* All the rules in this function are specific to scalar integers. */
9761 46571070 : scalar_int_mode mode;
9762 67615898 : if (!is_a <scalar_int_mode> (GET_MODE (dest), &mode))
9763 : return x;
9764 :
9765 : /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9766 : a clear of a one-bit field. We will have changed it to
9767 : (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9768 : for a SUBREG. */
9769 :
9770 1252855 : if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == ROTATE
9771 2526 : && CONST_INT_P (XEXP (XEXP (src, 0), 0))
9772 553 : && INTVAL (XEXP (XEXP (src, 0), 0)) == -2
9773 21053355 : && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9774 : {
9775 176 : assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
9776 : 1, true, true, false);
9777 176 : if (assign != 0)
9778 173 : return gen_rtx_SET (assign, const0_rtx);
9779 : return x;
9780 : }
9781 :
9782 1252679 : if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == SUBREG
9783 84968 : && subreg_lowpart_p (XEXP (src, 0))
9784 84935 : && partial_subreg_p (XEXP (src, 0))
9785 19485 : && GET_CODE (SUBREG_REG (XEXP (src, 0))) == ROTATE
9786 125 : && CONST_INT_P (XEXP (SUBREG_REG (XEXP (src, 0)), 0))
9787 57 : && INTVAL (XEXP (SUBREG_REG (XEXP (src, 0)), 0)) == -2
9788 21052683 : && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9789 : {
9790 14 : assign = make_extraction (VOIDmode, dest, 0,
9791 7 : XEXP (SUBREG_REG (XEXP (src, 0)), 1),
9792 : 1, true, true, false);
9793 7 : if (assign != 0)
9794 7 : return gen_rtx_SET (assign, const0_rtx);
9795 : return x;
9796 : }
9797 :
9798 : /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9799 : one-bit field. */
9800 1807447 : if (GET_CODE (src) == IOR && GET_CODE (XEXP (src, 0)) == ASHIFT
9801 408523 : && XEXP (XEXP (src, 0), 0) == const1_rtx
9802 21054885 : && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9803 : {
9804 559 : assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
9805 : 1, true, true, false);
9806 559 : if (assign != 0)
9807 530 : return gen_rtx_SET (assign, const1_rtx);
9808 : return x;
9809 : }
9810 :
9811 : /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9812 : SRC is an AND with all bits of that field set, then we can discard
9813 : the AND. */
9814 21052060 : if (GET_CODE (dest) == ZERO_EXTRACT
9815 2676 : && CONST_INT_P (XEXP (dest, 1))
9816 2676 : && GET_CODE (src) == AND
9817 816 : && CONST_INT_P (XEXP (src, 1)))
9818 : {
9819 816 : HOST_WIDE_INT width = INTVAL (XEXP (dest, 1));
9820 816 : unsigned HOST_WIDE_INT and_mask = INTVAL (XEXP (src, 1));
9821 816 : unsigned HOST_WIDE_INT ze_mask;
9822 :
9823 816 : if (width >= HOST_BITS_PER_WIDE_INT)
9824 : ze_mask = -1;
9825 : else
9826 816 : ze_mask = (HOST_WIDE_INT_1U << width) - 1;
9827 :
9828 : /* Complete overlap. We can remove the source AND. */
9829 816 : if ((and_mask & ze_mask) == ze_mask)
9830 792 : return gen_rtx_SET (dest, XEXP (src, 0));
9831 :
9832 : /* Partial overlap. We can reduce the source AND. */
9833 24 : if ((and_mask & ze_mask) != and_mask)
9834 : {
9835 6 : src = gen_rtx_AND (mode, XEXP (src, 0),
9836 : gen_int_mode (and_mask & ze_mask, mode));
9837 6 : return gen_rtx_SET (dest, src);
9838 : }
9839 : }
9840 :
9841 : /* The other case we handle is assignments into a constant-position
9842 : field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9843 : a mask that has all one bits except for a group of zero bits and
9844 : OTHER is known to have zeros where C1 has ones, this is such an
9845 : assignment. Compute the position and length from C1. Shift OTHER
9846 : to the appropriate position, force it to the required mode, and
9847 : make the extraction. Check for the AND in both operands. */
9848 :
9849 : /* One or more SUBREGs might obscure the constant-position field
9850 : assignment. The first one we are likely to encounter is an outer
9851 : narrowing SUBREG, which we can just strip for the purposes of
9852 : identifying the constant-field assignment. */
9853 21051262 : scalar_int_mode src_mode = mode;
9854 21051262 : if (GET_CODE (src) == SUBREG
9855 200958 : && subreg_lowpart_p (src)
9856 21236392 : && is_a <scalar_int_mode> (GET_MODE (SUBREG_REG (src)), &src_mode))
9857 : src = SUBREG_REG (src);
9858 :
9859 21051262 : if (GET_CODE (src) != IOR && GET_CODE (src) != XOR)
9860 : return x;
9861 :
9862 1987846 : rhs = expand_compound_operation (XEXP (src, 0));
9863 1987846 : lhs = expand_compound_operation (XEXP (src, 1));
9864 :
9865 1987846 : if (GET_CODE (rhs) == AND
9866 785413 : && CONST_INT_P (XEXP (rhs, 1))
9867 2427230 : && rtx_equal_for_field_assignment_p (XEXP (rhs, 0), dest))
9868 9643 : c1 = INTVAL (XEXP (rhs, 1)), other = lhs;
9869 : /* The second SUBREG that might get in the way is a paradoxical
9870 : SUBREG around the first operand of the AND. We want to
9871 : pretend the operand is as wide as the destination here. We
9872 : do this by adjusting the MEM to wider mode for the sole
9873 : purpose of the call to rtx_equal_for_field_assignment_p. Also
9874 : note this trick only works for MEMs. */
9875 1978203 : else if (GET_CODE (rhs) == AND
9876 775770 : && paradoxical_subreg_p (XEXP (rhs, 0))
9877 68455 : && MEM_P (SUBREG_REG (XEXP (rhs, 0)))
9878 30817 : && CONST_INT_P (XEXP (rhs, 1))
9879 2009020 : && rtx_equal_for_field_assignment_p (SUBREG_REG (XEXP (rhs, 0)),
9880 : dest, true))
9881 0 : c1 = INTVAL (XEXP (rhs, 1)), other = lhs;
9882 1978203 : else if (GET_CODE (lhs) == AND
9883 87822 : && CONST_INT_P (XEXP (lhs, 1))
9884 2056286 : && rtx_equal_for_field_assignment_p (XEXP (lhs, 0), dest))
9885 29 : c1 = INTVAL (XEXP (lhs, 1)), other = rhs;
9886 : /* The second SUBREG that might get in the way is a paradoxical
9887 : SUBREG around the first operand of the AND. We want to
9888 : pretend the operand is as wide as the destination here. We
9889 : do this by adjusting the MEM to wider mode for the sole
9890 : purpose of the call to rtx_equal_for_field_assignment_p. Also
9891 : note this trick only works for MEMs. */
9892 1978174 : else if (GET_CODE (lhs) == AND
9893 87793 : && paradoxical_subreg_p (XEXP (lhs, 0))
9894 37431 : && MEM_P (SUBREG_REG (XEXP (lhs, 0)))
9895 25240 : && CONST_INT_P (XEXP (lhs, 1))
9896 2003414 : && rtx_equal_for_field_assignment_p (SUBREG_REG (XEXP (lhs, 0)),
9897 : dest, true))
9898 0 : c1 = INTVAL (XEXP (lhs, 1)), other = rhs;
9899 : else
9900 1978174 : return x;
9901 :
9902 9672 : pos = get_pos_from_mask ((~c1) & GET_MODE_MASK (mode), &len);
9903 9672 : if (pos < 0
9904 7101 : || pos + len > GET_MODE_PRECISION (mode)
9905 7101 : || GET_MODE_PRECISION (mode) > HOST_BITS_PER_WIDE_INT
9906 16769 : || (c1 & nonzero_bits (other, mode)) != 0)
9907 3194 : return x;
9908 :
9909 6478 : assign = make_extraction (VOIDmode, dest, pos, NULL_RTX, len,
9910 : true, true, false);
9911 6478 : if (assign == 0)
9912 : return x;
9913 :
9914 : /* The mode to use for the source is the mode of the assignment, or of
9915 : what is inside a possible STRICT_LOW_PART. */
9916 12932 : machine_mode new_mode = (GET_CODE (assign) == STRICT_LOW_PART
9917 6466 : ? GET_MODE (XEXP (assign, 0)) : GET_MODE (assign));
9918 :
9919 : /* Shift OTHER right POS places and make it the source, restricting it
9920 : to the proper length and mode. */
9921 :
9922 6466 : src = canon_reg_for_combine (simplify_shift_const (NULL_RTX, LSHIFTRT,
9923 : src_mode, other, pos),
9924 : dest);
9925 12932 : src = force_to_mode (src, new_mode,
9926 : len >= HOST_BITS_PER_WIDE_INT
9927 : ? HOST_WIDE_INT_M1U
9928 6466 : : (HOST_WIDE_INT_1U << len) - 1, false);
9929 :
9930 : /* If SRC is masked by an AND that does not make a difference in
9931 : the value being stored, strip it. */
9932 6466 : if (GET_CODE (assign) == ZERO_EXTRACT
9933 6417 : && CONST_INT_P (XEXP (assign, 1))
9934 6417 : && INTVAL (XEXP (assign, 1)) < HOST_BITS_PER_WIDE_INT
9935 6417 : && GET_CODE (src) == AND
9936 0 : && CONST_INT_P (XEXP (src, 1))
9937 0 : && UINTVAL (XEXP (src, 1))
9938 0 : == (HOST_WIDE_INT_1U << INTVAL (XEXP (assign, 1))) - 1)
9939 0 : src = XEXP (src, 0);
9940 :
9941 6466 : return gen_rtx_SET (assign, src);
9942 : }
9943 :
9944 : /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9945 : if so. */
9946 :
9947 : static rtx
9948 50638360 : apply_distributive_law (rtx x)
9949 : {
9950 50638360 : enum rtx_code code = GET_CODE (x);
9951 50638360 : enum rtx_code inner_code;
9952 50638360 : rtx lhs, rhs, other;
9953 50638360 : rtx tem;
9954 :
9955 : /* Distributivity is not true for floating point as it can change the
9956 : value. So we don't do it unless -funsafe-math-optimizations. */
9957 50638360 : if (FLOAT_MODE_P (GET_MODE (x))
9958 3629954 : && ! flag_unsafe_math_optimizations)
9959 : return x;
9960 :
9961 : /* The outer operation can only be one of the following: */
9962 47441291 : if (code != IOR && code != AND && code != XOR
9963 47441291 : && code != PLUS && code != MINUS)
9964 : return x;
9965 :
9966 47427918 : lhs = XEXP (x, 0);
9967 47427918 : rhs = XEXP (x, 1);
9968 :
9969 : /* If either operand is a primitive we can't do anything, so get out
9970 : fast. */
9971 47427918 : if (OBJECT_P (lhs) || OBJECT_P (rhs))
9972 : return x;
9973 :
9974 2721189 : lhs = expand_compound_operation (lhs);
9975 2721189 : rhs = expand_compound_operation (rhs);
9976 2721189 : inner_code = GET_CODE (lhs);
9977 2721189 : if (inner_code != GET_CODE (rhs))
9978 : return x;
9979 :
9980 : /* See if the inner and outer operations distribute. */
9981 710778 : switch (inner_code)
9982 : {
9983 284048 : case LSHIFTRT:
9984 284048 : case ASHIFTRT:
9985 284048 : case AND:
9986 284048 : case IOR:
9987 : /* These all distribute except over PLUS. */
9988 284048 : if (code == PLUS || code == MINUS)
9989 : return x;
9990 : break;
9991 :
9992 92396 : case MULT:
9993 92396 : if (code != PLUS && code != MINUS)
9994 : return x;
9995 : break;
9996 :
9997 : case ASHIFT:
9998 : /* This is also a multiply, so it distributes over everything. */
9999 : break;
10000 :
10001 : /* This used to handle SUBREG, but this turned out to be counter-
10002 : productive, since (subreg (op ...)) usually is not handled by
10003 : insn patterns, and this "optimization" therefore transformed
10004 : recognizable patterns into unrecognizable ones. Therefore the
10005 : SUBREG case was removed from here.
10006 :
10007 : It is possible that distributing SUBREG over arithmetic operations
10008 : leads to an intermediate result than can then be optimized further,
10009 : e.g. by moving the outer SUBREG to the other side of a SET as done
10010 : in simplify_set. This seems to have been the original intent of
10011 : handling SUBREGs here.
10012 :
10013 : However, with current GCC this does not appear to actually happen,
10014 : at least on major platforms. If some case is found where removing
10015 : the SUBREG case here prevents follow-on optimizations, distributing
10016 : SUBREGs ought to be re-added at that place, e.g. in simplify_set. */
10017 :
10018 : default:
10019 : return x;
10020 : }
10021 :
10022 : /* Set LHS and RHS to the inner operands (A and B in the example
10023 : above) and set OTHER to the common operand (C in the example).
10024 : There is only one way to do this unless the inner operation is
10025 : commutative. */
10026 307913 : if (COMMUTATIVE_ARITH_P (lhs)
10027 307913 : && rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 0)))
10028 2167 : other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 1);
10029 305746 : else if (COMMUTATIVE_ARITH_P (lhs)
10030 305746 : && rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 1)))
10031 15 : other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 0);
10032 305731 : else if (COMMUTATIVE_ARITH_P (lhs)
10033 305731 : && rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 0)))
10034 10570 : other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 1);
10035 295161 : else if (rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 1)))
10036 63366 : other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 0);
10037 : else
10038 : return x;
10039 :
10040 : /* Form the new inner operation, seeing if it simplifies first. */
10041 76118 : tem = simplify_gen_binary (code, GET_MODE (x), lhs, rhs);
10042 :
10043 : /* There is one exception to the general way of distributing:
10044 : (a | c) ^ (b | c) -> (a ^ b) & ~c */
10045 76118 : if (code == XOR && inner_code == IOR)
10046 : {
10047 76 : inner_code = AND;
10048 76 : other = simplify_gen_unary (NOT, GET_MODE (x), other, GET_MODE (x));
10049 : }
10050 :
10051 : /* We may be able to continuing distributing the result, so call
10052 : ourselves recursively on the inner operation before forming the
10053 : outer operation, which we return. */
10054 76118 : return simplify_gen_binary (inner_code, GET_MODE (x),
10055 76118 : apply_distributive_law (tem), other);
10056 : }
10057 :
10058 : /* See if X is of the form (* (+ A B) C), and if so convert to
10059 : (+ (* A C) (* B C)) and try to simplify.
10060 :
10061 : Most of the time, this results in no change. However, if some of
10062 : the operands are the same or inverses of each other, simplifications
10063 : will result.
10064 :
10065 : For example, (and (ior A B) (not B)) can occur as the result of
10066 : expanding a bit field assignment. When we apply the distributive
10067 : law to this, we get (ior (and (A (not B))) (and (B (not B)))),
10068 : which then simplifies to (and (A (not B))).
10069 :
10070 : Note that no checks happen on the validity of applying the inverse
10071 : distributive law. This is pointless since we can do it in the
10072 : few places where this routine is called.
10073 :
10074 : N is the index of the term that is decomposed (the arithmetic operation,
10075 : i.e. (+ A B) in the first example above). !N is the index of the term that
10076 : is distributed, i.e. of C in the first example above. */
10077 : static rtx
10078 1672027 : distribute_and_simplify_rtx (rtx x, int n)
10079 : {
10080 1672027 : machine_mode mode;
10081 1672027 : enum rtx_code outer_code, inner_code;
10082 1672027 : rtx decomposed, distributed, inner_op0, inner_op1, new_op0, new_op1, tmp;
10083 :
10084 : /* Distributivity is not true for floating point as it can change the
10085 : value. So we don't do it unless -funsafe-math-optimizations. */
10086 1672027 : if (FLOAT_MODE_P (GET_MODE (x))
10087 163942 : && ! flag_unsafe_math_optimizations)
10088 : return NULL_RTX;
10089 :
10090 1511679 : decomposed = XEXP (x, n);
10091 1511679 : if (!ARITHMETIC_P (decomposed))
10092 : return NULL_RTX;
10093 :
10094 1511679 : mode = GET_MODE (x);
10095 1511679 : outer_code = GET_CODE (x);
10096 1511679 : distributed = XEXP (x, !n);
10097 :
10098 1511679 : inner_code = GET_CODE (decomposed);
10099 1511679 : inner_op0 = XEXP (decomposed, 0);
10100 1511679 : inner_op1 = XEXP (decomposed, 1);
10101 :
10102 : /* Special case (and (xor B C) (not A)), which is equivalent to
10103 : (xor (ior A B) (ior A C)) */
10104 1511679 : if (outer_code == AND && inner_code == XOR && GET_CODE (distributed) == NOT)
10105 : {
10106 71 : distributed = XEXP (distributed, 0);
10107 71 : outer_code = IOR;
10108 : }
10109 :
10110 1511679 : if (n == 0)
10111 : {
10112 : /* Distribute the second term. */
10113 1451366 : new_op0 = simplify_gen_binary (outer_code, mode, inner_op0, distributed);
10114 1451366 : new_op1 = simplify_gen_binary (outer_code, mode, inner_op1, distributed);
10115 : }
10116 : else
10117 : {
10118 : /* Distribute the first term. */
10119 60313 : new_op0 = simplify_gen_binary (outer_code, mode, distributed, inner_op0);
10120 60313 : new_op1 = simplify_gen_binary (outer_code, mode, distributed, inner_op1);
10121 : }
10122 :
10123 1511679 : tmp = apply_distributive_law (simplify_gen_binary (inner_code, mode,
10124 : new_op0, new_op1));
10125 1511679 : if (GET_CODE (tmp) != outer_code
10126 1511679 : && (set_src_cost (tmp, mode, optimize_this_for_speed_p)
10127 251772 : < set_src_cost (x, mode, optimize_this_for_speed_p)))
10128 : return tmp;
10129 :
10130 : return NULL_RTX;
10131 : }
10132 :
10133 : /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
10134 : in MODE. Return an equivalent form, if different from (and VAROP
10135 : (const_int CONSTOP)). Otherwise, return NULL_RTX. */
10136 :
10137 : static rtx
10138 11995079 : simplify_and_const_int_1 (scalar_int_mode mode, rtx varop,
10139 : unsigned HOST_WIDE_INT constop)
10140 : {
10141 11995079 : unsigned HOST_WIDE_INT nonzero;
10142 11995079 : unsigned HOST_WIDE_INT orig_constop;
10143 11995079 : rtx orig_varop;
10144 11995079 : int i;
10145 :
10146 11995079 : orig_varop = varop;
10147 11995079 : orig_constop = constop;
10148 11995079 : if (GET_CODE (varop) == CLOBBER)
10149 : return NULL_RTX;
10150 :
10151 : /* Simplify VAROP knowing that we will be only looking at some of the
10152 : bits in it.
10153 :
10154 : Note by passing in CONSTOP, we guarantee that the bits not set in
10155 : CONSTOP are not significant and will never be examined. We must
10156 : ensure that is the case by explicitly masking out those bits
10157 : before returning. */
10158 11995073 : varop = force_to_mode (varop, mode, constop, false);
10159 :
10160 : /* If VAROP is a CLOBBER, we will fail so return it. */
10161 11995073 : if (GET_CODE (varop) == CLOBBER)
10162 : return varop;
10163 :
10164 : /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
10165 : to VAROP and return the new constant. */
10166 11995063 : if (CONST_INT_P (varop))
10167 312886 : return gen_int_mode (INTVAL (varop) & constop, mode);
10168 :
10169 : /* See what bits may be nonzero in VAROP. Unlike the general case of
10170 : a call to nonzero_bits, here we don't care about bits outside
10171 : MODE unless WORD_REGISTER_OPERATIONS is true. */
10172 :
10173 11682177 : scalar_int_mode tmode = mode;
10174 11682177 : if (WORD_REGISTER_OPERATIONS && GET_MODE_BITSIZE (mode) < BITS_PER_WORD)
10175 : tmode = word_mode;
10176 11682177 : nonzero = nonzero_bits (varop, tmode) & GET_MODE_MASK (tmode);
10177 :
10178 : /* Turn off all bits in the constant that are known to already be zero.
10179 : Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
10180 : which is tested below. */
10181 :
10182 11682177 : constop &= nonzero;
10183 :
10184 : /* If we don't have any bits left, return zero. */
10185 11682177 : if (constop == 0 && !side_effects_p (varop))
10186 0 : return const0_rtx;
10187 :
10188 : /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
10189 : a power of two, we can replace this with an ASHIFT. */
10190 34430 : if (GET_CODE (varop) == NEG && nonzero_bits (XEXP (varop, 0), tmode) == 1
10191 11688406 : && (i = exact_log2 (constop)) >= 0)
10192 121 : return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (varop, 0), i);
10193 :
10194 : /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
10195 : or XOR, then try to apply the distributive law. This may eliminate
10196 : operations if either branch can be simplified because of the AND.
10197 : It may also make some cases more complex, but those cases probably
10198 : won't match a pattern either with or without this. */
10199 :
10200 11682056 : if (GET_CODE (varop) == IOR || GET_CODE (varop) == XOR)
10201 : {
10202 86399 : scalar_int_mode varop_mode = as_a <scalar_int_mode> (GET_MODE (varop));
10203 86399 : return
10204 86399 : gen_lowpart
10205 86399 : (mode,
10206 : apply_distributive_law
10207 86399 : (simplify_gen_binary (GET_CODE (varop), varop_mode,
10208 : simplify_and_const_int (NULL_RTX, varop_mode,
10209 : XEXP (varop, 0),
10210 : constop),
10211 : simplify_and_const_int (NULL_RTX, varop_mode,
10212 : XEXP (varop, 1),
10213 : constop))));
10214 : }
10215 :
10216 : /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
10217 : the AND and see if one of the operands simplifies to zero. If so, we
10218 : may eliminate it. */
10219 :
10220 11595657 : if (GET_CODE (varop) == PLUS
10221 11595657 : && pow2p_hwi (constop + 1))
10222 : {
10223 439698 : rtx o0, o1;
10224 :
10225 439698 : o0 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 0), constop);
10226 439698 : o1 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 1), constop);
10227 439698 : if (o0 == const0_rtx)
10228 : return o1;
10229 439698 : if (o1 == const0_rtx)
10230 : return o0;
10231 : }
10232 :
10233 : /* Make a SUBREG if necessary. If we can't make it, fail. */
10234 11595594 : varop = gen_lowpart (mode, varop);
10235 11595594 : if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
10236 : return NULL_RTX;
10237 :
10238 : /* If we are only masking insignificant bits, return VAROP. */
10239 11595594 : if (constop == nonzero)
10240 : return varop;
10241 :
10242 11162725 : if (varop == orig_varop && constop == orig_constop)
10243 : return NULL_RTX;
10244 :
10245 : /* Otherwise, return an AND. */
10246 6109715 : return simplify_gen_binary (AND, mode, varop, gen_int_mode (constop, mode));
10247 : }
10248 :
10249 :
10250 : /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
10251 : in MODE.
10252 :
10253 : Return an equivalent form, if different from X. Otherwise, return X. If
10254 : X is zero, we are to always construct the equivalent form. */
10255 :
10256 : static rtx
10257 11995079 : simplify_and_const_int (rtx x, scalar_int_mode mode, rtx varop,
10258 : unsigned HOST_WIDE_INT constop)
10259 : {
10260 11995079 : rtx tem = simplify_and_const_int_1 (mode, varop, constop);
10261 11995079 : if (tem)
10262 : return tem;
10263 :
10264 5053016 : if (!x)
10265 1291909 : x = simplify_gen_binary (AND, GET_MODE (varop), varop,
10266 1291909 : gen_int_mode (constop, mode));
10267 5053016 : if (GET_MODE (x) != mode)
10268 0 : x = gen_lowpart (mode, x);
10269 : return x;
10270 : }
10271 :
10272 : /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
10273 : We don't care about bits outside of those defined in MODE.
10274 : We DO care about all the bits in MODE, even if XMODE is smaller than MODE.
10275 :
10276 : For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
10277 : a shift, AND, or zero_extract, we can do better. */
10278 :
10279 : static rtx
10280 436540883 : reg_nonzero_bits_for_combine (const_rtx x, scalar_int_mode xmode,
10281 : scalar_int_mode mode,
10282 : unsigned HOST_WIDE_INT *nonzero)
10283 : {
10284 436540883 : rtx tem;
10285 436540883 : reg_stat_type *rsp;
10286 :
10287 : /* If X is a register whose nonzero bits value is current, use it.
10288 : Otherwise, if X is a register whose value we can find, use that
10289 : value. Otherwise, use the previously-computed global nonzero bits
10290 : for this register. */
10291 :
10292 436540883 : rsp = ®_stat[REGNO (x)];
10293 436540883 : if (rsp->last_set_value != 0
10294 406382492 : && (rsp->last_set_mode == mode
10295 1278 : || (REGNO (x) >= FIRST_PSEUDO_REGISTER
10296 0 : && GET_MODE_CLASS (rsp->last_set_mode) == MODE_INT
10297 0 : && GET_MODE_CLASS (mode) == MODE_INT))
10298 842922097 : && ((rsp->last_set_label >= label_tick_ebb_start
10299 303329590 : && rsp->last_set_label < label_tick)
10300 384985054 : || (rsp->last_set_label == label_tick
10301 281933430 : && DF_INSN_LUID (rsp->last_set) < subst_low_luid)
10302 130769994 : || (REGNO (x) >= FIRST_PSEUDO_REGISTER
10303 130704002 : && REGNO (x) < reg_n_sets_max
10304 130703876 : && REG_N_SETS (REGNO (x)) == 1
10305 147734218 : && !REGNO_REG_SET_P
10306 : (DF_LR_IN (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb),
10307 : REGNO (x)))))
10308 : {
10309 : /* Note that, even if the precision of last_set_mode is lower than that
10310 : of mode, record_value_for_reg invoked nonzero_bits on the register
10311 : with nonzero_bits_mode (because last_set_mode is necessarily integral
10312 : and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
10313 : are all valid, hence in mode too since nonzero_bits_mode is defined
10314 : to the largest HWI_COMPUTABLE_MODE_P mode. */
10315 349388879 : *nonzero &= rsp->last_set_nonzero_bits;
10316 349388879 : return NULL;
10317 : }
10318 :
10319 87152004 : tem = get_last_value (x);
10320 87152004 : if (tem)
10321 : {
10322 : if (SHORT_IMMEDIATES_SIGN_EXTEND)
10323 : tem = sign_extend_short_imm (tem, xmode, GET_MODE_PRECISION (mode));
10324 :
10325 : return tem;
10326 : }
10327 :
10328 87151998 : if (nonzero_sign_valid && rsp->nonzero_bits)
10329 : {
10330 53999741 : unsigned HOST_WIDE_INT mask = rsp->nonzero_bits;
10331 :
10332 53999741 : if (GET_MODE_PRECISION (xmode) < GET_MODE_PRECISION (mode))
10333 : /* We don't know anything about the upper bits. */
10334 0 : mask |= GET_MODE_MASK (mode) ^ GET_MODE_MASK (xmode);
10335 :
10336 53999741 : *nonzero &= mask;
10337 : }
10338 :
10339 : return NULL;
10340 : }
10341 :
10342 : /* Given a reg X of mode XMODE, return the number of bits at the high-order
10343 : end of X that are known to be equal to the sign bit. X will be used
10344 : in mode MODE; the returned value will always be between 1 and the
10345 : number of bits in MODE. */
10346 :
10347 : static rtx
10348 127032738 : reg_num_sign_bit_copies_for_combine (const_rtx x, scalar_int_mode xmode,
10349 : scalar_int_mode mode,
10350 : unsigned int *result)
10351 : {
10352 127032738 : rtx tem;
10353 127032738 : reg_stat_type *rsp;
10354 :
10355 127032738 : rsp = ®_stat[REGNO (x)];
10356 127032738 : if (rsp->last_set_value != 0
10357 116109290 : && rsp->last_set_mode == mode
10358 243141859 : && ((rsp->last_set_label >= label_tick_ebb_start
10359 87058294 : && rsp->last_set_label < label_tick)
10360 110547987 : || (rsp->last_set_label == label_tick
10361 81497160 : && DF_INSN_LUID (rsp->last_set) < subst_low_luid)
10362 36735309 : || (REGNO (x) >= FIRST_PSEUDO_REGISTER
10363 36722720 : && REGNO (x) < reg_n_sets_max
10364 36722638 : && REG_N_SETS (REGNO (x)) == 1
10365 41582170 : && !REGNO_REG_SET_P
10366 : (DF_LR_IN (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb),
10367 : REGNO (x)))))
10368 : {
10369 100141968 : *result = rsp->last_set_sign_bit_copies;
10370 100141968 : return NULL;
10371 : }
10372 :
10373 26890770 : tem = get_last_value (x);
10374 26890770 : if (tem != 0)
10375 : return tem;
10376 :
10377 17572428 : if (nonzero_sign_valid && rsp->sign_bit_copies != 0
10378 40359976 : && GET_MODE_PRECISION (xmode) == GET_MODE_PRECISION (mode))
10379 13469211 : *result = rsp->sign_bit_copies;
10380 :
10381 : return NULL;
10382 : }
10383 :
10384 : /* Return the number of "extended" bits there are in X, when interpreted
10385 : as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
10386 : unsigned quantities, this is the number of high-order zero bits.
10387 : For signed quantities, this is the number of copies of the sign bit
10388 : minus 1. In both case, this function returns the number of "spare"
10389 : bits. For example, if two quantities for which this function returns
10390 : at least 1 are added, the addition is known not to overflow.
10391 :
10392 : This function will always return 0 unless called during combine, which
10393 : implies that it must be called from a define_split. */
10394 :
10395 : unsigned int
10396 0 : extended_count (const_rtx x, machine_mode mode, bool unsignedp)
10397 : {
10398 0 : if (nonzero_sign_valid == 0)
10399 : return 0;
10400 :
10401 0 : scalar_int_mode int_mode;
10402 0 : return (unsignedp
10403 0 : ? (is_a <scalar_int_mode> (mode, &int_mode)
10404 0 : && HWI_COMPUTABLE_MODE_P (int_mode)
10405 0 : ? (unsigned int) (GET_MODE_PRECISION (int_mode) - 1
10406 0 : - floor_log2 (nonzero_bits (x, int_mode)))
10407 : : 0)
10408 0 : : num_sign_bit_copies (x, mode) - 1);
10409 : }
10410 :
10411 : /* This function is called from `simplify_shift_const' to merge two
10412 : outer operations. Specifically, we have already found that we need
10413 : to perform operation *POP0 with constant *PCONST0 at the outermost
10414 : position. We would now like to also perform OP1 with constant CONST1
10415 : (with *POP0 being done last).
10416 :
10417 : Return true if we can do the operation and update *POP0 and *PCONST0 with
10418 : the resulting operation. *PCOMP_P is set to true if we would need to
10419 : complement the innermost operand, otherwise it is unchanged.
10420 :
10421 : MODE is the mode in which the operation will be done. No bits outside
10422 : the width of this mode matter. It is assumed that the width of this mode
10423 : is smaller than or equal to HOST_BITS_PER_WIDE_INT.
10424 :
10425 : If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
10426 : IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
10427 : result is simply *PCONST0.
10428 :
10429 : If the resulting operation cannot be expressed as one operation, we
10430 : return false and do not change *POP0, *PCONST0, and *PCOMP_P. */
10431 :
10432 : static bool
10433 3682544 : merge_outer_ops (enum rtx_code *pop0, HOST_WIDE_INT *pconst0,
10434 : enum rtx_code op1, HOST_WIDE_INT const1,
10435 : machine_mode mode, bool *pcomp_p)
10436 : {
10437 3682544 : enum rtx_code op0 = *pop0;
10438 3682544 : HOST_WIDE_INT const0 = *pconst0;
10439 :
10440 3682544 : const0 &= GET_MODE_MASK (mode);
10441 3682544 : const1 &= GET_MODE_MASK (mode);
10442 :
10443 : /* If OP0 is an AND, clear unimportant bits in CONST1. */
10444 3682544 : if (op0 == AND)
10445 13219 : const1 &= const0;
10446 :
10447 : /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
10448 : if OP0 is SET. */
10449 :
10450 3682544 : if (op1 == UNKNOWN || op0 == SET)
10451 : return true;
10452 :
10453 3682544 : else if (op0 == UNKNOWN)
10454 : op0 = op1, const0 = const1;
10455 :
10456 64570 : else if (op0 == op1)
10457 : {
10458 13064 : switch (op0)
10459 : {
10460 13059 : case AND:
10461 13059 : const0 &= const1;
10462 13059 : break;
10463 5 : case IOR:
10464 5 : const0 |= const1;
10465 5 : break;
10466 0 : case XOR:
10467 0 : const0 ^= const1;
10468 0 : break;
10469 0 : case PLUS:
10470 0 : const0 += const1;
10471 0 : break;
10472 : case NEG:
10473 3654127 : op0 = UNKNOWN;
10474 : break;
10475 : default:
10476 : break;
10477 : }
10478 : }
10479 :
10480 : /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
10481 51506 : else if (op0 == PLUS || op1 == PLUS || op0 == NEG || op1 == NEG)
10482 : return false;
10483 :
10484 : /* If the two constants aren't the same, we can't do anything. The
10485 : remaining six cases can all be done. */
10486 24344 : else if (const0 != const1)
10487 : return false;
10488 :
10489 : else
10490 23089 : switch (op0)
10491 : {
10492 8 : case IOR:
10493 8 : if (op1 == AND)
10494 : /* (a & b) | b == b */
10495 0 : op0 = SET;
10496 : else /* op1 == XOR */
10497 : /* (a ^ b) | b == a | b */
10498 : {;}
10499 : break;
10500 :
10501 22925 : case XOR:
10502 22925 : if (op1 == AND)
10503 : /* (a & b) ^ b == (~a) & b */
10504 22925 : op0 = AND, *pcomp_p = true;
10505 : else /* op1 == IOR */
10506 : /* (a | b) ^ b == a & ~b */
10507 0 : op0 = AND, const0 = ~const0;
10508 : break;
10509 :
10510 156 : case AND:
10511 156 : if (op1 == IOR)
10512 : /* (a | b) & b == b */
10513 : op0 = SET;
10514 : else /* op1 == XOR */
10515 : /* (a ^ b) & b) == (~a) & b */
10516 156 : *pcomp_p = true;
10517 : break;
10518 : default:
10519 : break;
10520 : }
10521 :
10522 : /* Check for NO-OP cases. */
10523 3654127 : const0 &= GET_MODE_MASK (mode);
10524 3654127 : if (const0 == 0
10525 20869 : && (op0 == IOR || op0 == XOR || op0 == PLUS))
10526 : op0 = UNKNOWN;
10527 3651856 : else if (const0 == 0 && op0 == AND)
10528 : op0 = SET;
10529 3651856 : else if ((unsigned HOST_WIDE_INT) const0 == GET_MODE_MASK (mode)
10530 16115 : && op0 == AND)
10531 3654127 : op0 = UNKNOWN;
10532 :
10533 3654127 : *pop0 = op0;
10534 :
10535 : /* ??? Slightly redundant with the above mask, but not entirely.
10536 : Moving this above means we'd have to sign-extend the mode mask
10537 : for the final test. */
10538 3654127 : if (op0 != UNKNOWN && op0 != NEG)
10539 3624243 : *pconst0 = trunc_int_for_mode (const0, mode);
10540 :
10541 : return true;
10542 : }
10543 :
10544 : /* A helper to simplify_shift_const_1 to determine the mode we can perform
10545 : the shift in. The original shift operation CODE is performed on OP in
10546 : ORIG_MODE. Return the wider mode MODE if we can perform the operation
10547 : in that mode. Return ORIG_MODE otherwise. We can also assume that the
10548 : result of the shift is subject to operation OUTER_CODE with operand
10549 : OUTER_CONST. */
10550 :
10551 : static scalar_int_mode
10552 296430 : try_widen_shift_mode (enum rtx_code code, rtx op, int count,
10553 : scalar_int_mode orig_mode, scalar_int_mode mode,
10554 : enum rtx_code outer_code, HOST_WIDE_INT outer_const)
10555 : {
10556 296430 : gcc_assert (GET_MODE_PRECISION (mode) > GET_MODE_PRECISION (orig_mode));
10557 :
10558 : /* In general we can't perform in wider mode for right shift and rotate. */
10559 296430 : switch (code)
10560 : {
10561 33134 : case ASHIFTRT:
10562 : /* We can still widen if the bits brought in from the left are identical
10563 : to the sign bit of ORIG_MODE. */
10564 33134 : if (num_sign_bit_copies (op, mode)
10565 33134 : > (unsigned) (GET_MODE_PRECISION (mode)
10566 33134 : - GET_MODE_PRECISION (orig_mode)))
10567 347 : return mode;
10568 32787 : return orig_mode;
10569 :
10570 34686 : case LSHIFTRT:
10571 : /* Similarly here but with zero bits. */
10572 34686 : if (HWI_COMPUTABLE_MODE_P (mode)
10573 34686 : && (nonzero_bits (op, mode) & ~GET_MODE_MASK (orig_mode)) == 0)
10574 3990 : return mode;
10575 :
10576 : /* We can also widen if the bits brought in will be masked off. This
10577 : operation is performed in ORIG_MODE. */
10578 30696 : if (outer_code == AND)
10579 : {
10580 8080 : int care_bits = low_bitmask_len (orig_mode, outer_const);
10581 :
10582 8080 : if (care_bits >= 0
10583 8080 : && GET_MODE_PRECISION (orig_mode) - care_bits >= count)
10584 8062 : return mode;
10585 : }
10586 : /* fall through */
10587 :
10588 23298 : case ROTATE:
10589 23298 : return orig_mode;
10590 :
10591 0 : case ROTATERT:
10592 0 : gcc_unreachable ();
10593 :
10594 227946 : default:
10595 227946 : return mode;
10596 : }
10597 : }
10598 :
10599 : /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
10600 : of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
10601 : if we cannot simplify it. Otherwise, return a simplified value.
10602 :
10603 : The shift is normally computed in the widest mode we find in VAROP, as
10604 : long as it isn't a different number of words than RESULT_MODE. Exceptions
10605 : are ASHIFTRT and ROTATE, which are always done in their original mode. */
10606 :
10607 : static rtx
10608 23796976 : simplify_shift_const_1 (enum rtx_code code, machine_mode result_mode,
10609 : rtx varop, int orig_count)
10610 : {
10611 23796976 : enum rtx_code orig_code = code;
10612 23796976 : rtx orig_varop = varop;
10613 23796976 : int count, log2;
10614 23796976 : machine_mode mode = result_mode;
10615 23796976 : machine_mode shift_mode;
10616 23796976 : scalar_int_mode tmode, inner_mode, int_mode, int_varop_mode, int_result_mode;
10617 : /* We form (outer_op (code varop count) (outer_const)). */
10618 23796976 : enum rtx_code outer_op = UNKNOWN;
10619 23796976 : HOST_WIDE_INT outer_const = 0;
10620 23796976 : bool complement_p = false;
10621 23796976 : rtx new_rtx, x;
10622 :
10623 : /* Make sure and truncate the "natural" shift on the way in. We don't
10624 : want to do this inside the loop as it makes it more difficult to
10625 : combine shifts. */
10626 23796976 : if (SHIFT_COUNT_TRUNCATED)
10627 : orig_count &= GET_MODE_UNIT_BITSIZE (mode) - 1;
10628 :
10629 : /* If we were given an invalid count, don't do anything except exactly
10630 : what was requested. */
10631 :
10632 47593852 : if (orig_count < 0 || orig_count >= (int) GET_MODE_UNIT_PRECISION (mode))
10633 : return NULL_RTX;
10634 :
10635 : count = orig_count;
10636 :
10637 : /* Unless one of the branches of the `if' in this loop does a `continue',
10638 : we will `break' the loop after the `if'. */
10639 :
10640 27754830 : while (count != 0)
10641 : {
10642 : /* If we have an operand of (clobber (const_int 0)), fail. */
10643 24014142 : if (GET_CODE (varop) == CLOBBER)
10644 23796976 : return NULL_RTX;
10645 :
10646 : /* Convert ROTATERT to ROTATE. */
10647 24014142 : if (code == ROTATERT)
10648 : {
10649 11029 : unsigned int bitsize = GET_MODE_UNIT_PRECISION (result_mode);
10650 11029 : code = ROTATE;
10651 11029 : count = bitsize - count;
10652 : }
10653 :
10654 24014142 : shift_mode = result_mode;
10655 24014142 : if (shift_mode != mode)
10656 : {
10657 : /* We only change the modes of scalar shifts. */
10658 150378 : int_mode = as_a <scalar_int_mode> (mode);
10659 150378 : int_result_mode = as_a <scalar_int_mode> (result_mode);
10660 150378 : shift_mode = try_widen_shift_mode (code, varop, count,
10661 : int_result_mode, int_mode,
10662 : outer_op, outer_const);
10663 : }
10664 :
10665 24014142 : scalar_int_mode shift_unit_mode;
10666 68044170 : if (!is_a <scalar_int_mode> (GET_MODE_INNER (shift_mode),
10667 : &shift_unit_mode))
10668 : return NULL_RTX;
10669 :
10670 : /* Handle cases where the count is greater than the size of the mode
10671 : minus 1. For ASHIFT, use the size minus one as the count (this can
10672 : occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
10673 : take the count modulo the size. For other shifts, the result is
10674 : zero.
10675 :
10676 : Since these shifts are being produced by the compiler by combining
10677 : multiple operations, each of which are defined, we know what the
10678 : result is supposed to be. */
10679 :
10680 24014142 : if (count > (GET_MODE_PRECISION (shift_unit_mode) - 1))
10681 : {
10682 8151 : if (code == ASHIFTRT)
10683 8145 : count = GET_MODE_PRECISION (shift_unit_mode) - 1;
10684 6 : else if (code == ROTATE || code == ROTATERT)
10685 6 : count %= GET_MODE_PRECISION (shift_unit_mode);
10686 : else
10687 : {
10688 : /* We can't simply return zero because there may be an
10689 : outer op. */
10690 0 : varop = const0_rtx;
10691 0 : count = 0;
10692 0 : break;
10693 : }
10694 : }
10695 :
10696 : /* If we discovered we had to complement VAROP, leave. Making a NOT
10697 : here would cause an infinite loop. */
10698 24014142 : if (complement_p)
10699 : break;
10700 :
10701 24000423 : if (shift_mode == shift_unit_mode)
10702 : {
10703 : /* An arithmetic right shift of a quantity known to be -1 or 0
10704 : is a no-op. */
10705 23402707 : if (code == ASHIFTRT
10706 23402707 : && (num_sign_bit_copies (varop, shift_unit_mode)
10707 4254694 : == GET_MODE_PRECISION (shift_unit_mode)))
10708 : {
10709 : count = 0;
10710 : break;
10711 : }
10712 :
10713 : /* If we are doing an arithmetic right shift and discarding all but
10714 : the sign bit copies, this is equivalent to doing a shift by the
10715 : bitsize minus one. Convert it into that shift because it will
10716 : often allow other simplifications. */
10717 :
10718 23402636 : if (code == ASHIFTRT
10719 23402636 : && (count + num_sign_bit_copies (varop, shift_unit_mode)
10720 4254623 : >= GET_MODE_PRECISION (shift_unit_mode)))
10721 256736 : count = GET_MODE_PRECISION (shift_unit_mode) - 1;
10722 :
10723 : /* We simplify the tests below and elsewhere by converting
10724 : ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10725 : `make_compound_operation' will convert it to an ASHIFTRT for
10726 : those machines (such as VAX) that don't have an LSHIFTRT. */
10727 23402636 : if (code == ASHIFTRT
10728 4254623 : && HWI_COMPUTABLE_MODE_P (shift_unit_mode)
10729 27632044 : && val_signbit_known_clear_p (shift_unit_mode,
10730 : nonzero_bits (varop,
10731 : shift_unit_mode)))
10732 : code = LSHIFTRT;
10733 :
10734 23373470 : if (((code == LSHIFTRT
10735 5832853 : && HWI_COMPUTABLE_MODE_P (shift_unit_mode)
10736 5812342 : && !(nonzero_bits (varop, shift_unit_mode) >> count))
10737 23400873 : || (code == ASHIFT
10738 13313709 : && HWI_COMPUTABLE_MODE_P (shift_unit_mode)
10739 12867864 : && !((nonzero_bits (varop, shift_unit_mode) << count)
10740 12867864 : & GET_MODE_MASK (shift_unit_mode))))
10741 23377447 : && !side_effects_p (varop))
10742 3977 : varop = const0_rtx;
10743 : }
10744 :
10745 24000352 : switch (GET_CODE (varop))
10746 : {
10747 481627 : case SIGN_EXTEND:
10748 481627 : case ZERO_EXTEND:
10749 481627 : case SIGN_EXTRACT:
10750 481627 : case ZERO_EXTRACT:
10751 481627 : new_rtx = expand_compound_operation (varop);
10752 481627 : if (new_rtx != varop)
10753 : {
10754 68169 : varop = new_rtx;
10755 27822999 : continue;
10756 : }
10757 : break;
10758 :
10759 256666 : case MEM:
10760 : /* The following rules apply only to scalars. */
10761 256666 : if (shift_mode != shift_unit_mode)
10762 : break;
10763 242366 : int_mode = as_a <scalar_int_mode> (mode);
10764 :
10765 : /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10766 : minus the width of a smaller mode, we can do this with a
10767 : SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10768 247091 : if ((code == ASHIFTRT || code == LSHIFTRT)
10769 78042 : && ! mode_dependent_address_p (XEXP (varop, 0),
10770 78042 : MEM_ADDR_SPACE (varop))
10771 78042 : && ! MEM_VOLATILE_P (varop)
10772 318846 : && (int_mode_for_size (GET_MODE_BITSIZE (int_mode) - count, 1)
10773 237641 : .exists (&tmode)))
10774 : {
10775 4725 : new_rtx = adjust_address_nv (varop, tmode,
10776 : BYTES_BIG_ENDIAN ? 0
10777 : : count / BITS_PER_UNIT);
10778 :
10779 4725 : varop = gen_rtx_fmt_e (code == ASHIFTRT ? SIGN_EXTEND
10780 : : ZERO_EXTEND, int_mode, new_rtx);
10781 4725 : count = 0;
10782 4725 : continue;
10783 : }
10784 : break;
10785 :
10786 4842079 : case SUBREG:
10787 : /* The following rules apply only to scalars. */
10788 4842079 : if (shift_mode != shift_unit_mode)
10789 : break;
10790 4434048 : int_mode = as_a <scalar_int_mode> (mode);
10791 4434048 : int_varop_mode = as_a <scalar_int_mode> (GET_MODE (varop));
10792 :
10793 : /* If VAROP is a SUBREG, strip it as long as the inner operand has
10794 : the same number of words as what we've seen so far. Then store
10795 : the widest mode in MODE. */
10796 4434048 : if (subreg_lowpart_p (varop)
10797 28216582 : && is_int_mode (GET_MODE (SUBREG_REG (varop)), &inner_mode)
10798 8820454 : && GET_MODE_SIZE (inner_mode) > GET_MODE_SIZE (int_varop_mode)
10799 179832 : && (CEIL (GET_MODE_SIZE (inner_mode), UNITS_PER_WORD)
10800 163513 : == CEIL (GET_MODE_SIZE (int_mode), UNITS_PER_WORD))
10801 4580113 : && GET_MODE_CLASS (int_varop_mode) == MODE_INT)
10802 : {
10803 146065 : varop = SUBREG_REG (varop);
10804 438195 : if (GET_MODE_SIZE (inner_mode) > GET_MODE_SIZE (int_mode))
10805 146065 : mode = inner_mode;
10806 146065 : continue;
10807 : }
10808 : break;
10809 :
10810 423462 : case MULT:
10811 : /* Some machines use MULT instead of ASHIFT because MULT
10812 : is cheaper. But it is still better on those machines to
10813 : merge two shifts into one. */
10814 423462 : if (CONST_INT_P (XEXP (varop, 1))
10815 423462 : && (log2 = exact_log2 (UINTVAL (XEXP (varop, 1)))) >= 0)
10816 : {
10817 0 : rtx log2_rtx = gen_int_shift_amount (GET_MODE (varop), log2);
10818 0 : varop = simplify_gen_binary (ASHIFT, GET_MODE (varop),
10819 : XEXP (varop, 0), log2_rtx);
10820 0 : continue;
10821 0 : }
10822 : break;
10823 :
10824 8822 : case UDIV:
10825 : /* Similar, for when divides are cheaper. */
10826 8822 : if (CONST_INT_P (XEXP (varop, 1))
10827 8822 : && (log2 = exact_log2 (UINTVAL (XEXP (varop, 1)))) >= 0)
10828 : {
10829 9 : rtx log2_rtx = gen_int_shift_amount (GET_MODE (varop), log2);
10830 9 : varop = simplify_gen_binary (LSHIFTRT, GET_MODE (varop),
10831 : XEXP (varop, 0), log2_rtx);
10832 9 : continue;
10833 9 : }
10834 : break;
10835 :
10836 348896 : case ASHIFTRT:
10837 : /* If we are extracting just the sign bit of an arithmetic
10838 : right shift, that shift is not needed. However, the sign
10839 : bit of a wider mode may be different from what would be
10840 : interpreted as the sign bit in a narrower mode, so, if
10841 : the result is narrower, don't discard the shift. */
10842 350801 : if (code == LSHIFTRT
10843 14751 : && count == (GET_MODE_UNIT_BITSIZE (result_mode) - 1)
10844 348896 : && (GET_MODE_UNIT_BITSIZE (result_mode)
10845 3836 : >= GET_MODE_UNIT_BITSIZE (GET_MODE (varop))))
10846 : {
10847 1905 : varop = XEXP (varop, 0);
10848 1905 : continue;
10849 : }
10850 :
10851 : /* fall through */
10852 :
10853 5816014 : case LSHIFTRT:
10854 5816014 : case ASHIFT:
10855 5816014 : case ROTATE:
10856 : /* The following rules apply only to scalars. */
10857 5816014 : if (shift_mode != shift_unit_mode)
10858 : break;
10859 5808249 : int_mode = as_a <scalar_int_mode> (mode);
10860 5808249 : int_varop_mode = as_a <scalar_int_mode> (GET_MODE (varop));
10861 5808249 : int_result_mode = as_a <scalar_int_mode> (result_mode);
10862 :
10863 : /* Here we have two nested shifts. The result is usually the
10864 : AND of a new shift with a mask. We compute the result below. */
10865 5808249 : if (CONST_INT_P (XEXP (varop, 1))
10866 5788066 : && INTVAL (XEXP (varop, 1)) >= 0
10867 5788063 : && INTVAL (XEXP (varop, 1)) < GET_MODE_PRECISION (int_varop_mode)
10868 5788063 : && HWI_COMPUTABLE_MODE_P (int_result_mode)
10869 11563131 : && HWI_COMPUTABLE_MODE_P (int_mode))
10870 : {
10871 5754882 : enum rtx_code first_code = GET_CODE (varop);
10872 5754882 : unsigned int first_count = INTVAL (XEXP (varop, 1));
10873 5754882 : unsigned HOST_WIDE_INT mask;
10874 5754882 : rtx mask_rtx;
10875 :
10876 : /* We have one common special case. We can't do any merging if
10877 : the inner code is an ASHIFTRT of a smaller mode. However, if
10878 : we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10879 : with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10880 : we can convert it to
10881 : (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10882 : This simplifies certain SIGN_EXTEND operations. */
10883 5754882 : if (code == ASHIFT && first_code == ASHIFTRT
10884 5754882 : && count == (GET_MODE_PRECISION (int_result_mode)
10885 315921 : - GET_MODE_PRECISION (int_varop_mode)))
10886 : {
10887 : /* C3 has the low-order C1 bits zero. */
10888 :
10889 0 : mask = GET_MODE_MASK (int_mode)
10890 0 : & ~((HOST_WIDE_INT_1U << first_count) - 1);
10891 :
10892 0 : varop = simplify_and_const_int (NULL_RTX, int_result_mode,
10893 : XEXP (varop, 0), mask);
10894 0 : varop = simplify_shift_const (NULL_RTX, ASHIFT,
10895 : int_result_mode, varop, count);
10896 0 : count = first_count;
10897 0 : code = ASHIFTRT;
10898 0 : continue;
10899 : }
10900 :
10901 : /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10902 : than C1 high-order bits equal to the sign bit, we can convert
10903 : this to either an ASHIFT or an ASHIFTRT depending on the
10904 : two counts.
10905 :
10906 : We cannot do this if VAROP's mode is not SHIFT_UNIT_MODE. */
10907 :
10908 5756354 : if (code == ASHIFTRT && first_code == ASHIFT
10909 2861194 : && int_varop_mode == shift_unit_mode
10910 8608581 : && (num_sign_bit_copies (XEXP (varop, 0), shift_unit_mode)
10911 : > first_count))
10912 : {
10913 1472 : varop = XEXP (varop, 0);
10914 1472 : count -= first_count;
10915 1472 : if (count < 0)
10916 : {
10917 0 : count = -count;
10918 0 : code = ASHIFT;
10919 : }
10920 :
10921 1472 : continue;
10922 : }
10923 :
10924 : /* There are some cases we can't do. If CODE is ASHIFTRT,
10925 : we can only do this if FIRST_CODE is also ASHIFTRT.
10926 :
10927 : We can't do the case when CODE is ROTATE and FIRST_CODE is
10928 : ASHIFTRT.
10929 :
10930 : If the mode of this shift is not the mode of the outer shift,
10931 : we can't do this if either shift is a right shift or ROTATE.
10932 :
10933 : Finally, we can't do any of these if the mode is too wide
10934 : unless the codes are the same.
10935 :
10936 : Handle the case where the shift codes are the same
10937 : first. */
10938 :
10939 5753410 : if (code == first_code)
10940 : {
10941 21373 : if (int_varop_mode != int_result_mode
10942 21373 : && (code == ASHIFTRT || code == LSHIFTRT
10943 69 : || code == ROTATE))
10944 : break;
10945 :
10946 21337 : count += first_count;
10947 21337 : varop = XEXP (varop, 0);
10948 21337 : continue;
10949 : }
10950 :
10951 5732037 : if (code == ASHIFTRT
10952 2872267 : || (code == ROTATE && first_code == ASHIFTRT)
10953 2872237 : || GET_MODE_PRECISION (int_mode) > HOST_BITS_PER_WIDE_INT
10954 8604274 : || (int_varop_mode != int_result_mode
10955 52504 : && (first_code == ASHIFTRT || first_code == LSHIFTRT
10956 52504 : || first_code == ROTATE
10957 8407 : || code == ROTATE)))
10958 : break;
10959 :
10960 : /* To compute the mask to apply after the shift, shift the
10961 : nonzero bits of the inner shift the same way the
10962 : outer shift will. */
10963 :
10964 2828140 : mask_rtx = gen_int_mode (nonzero_bits (varop, int_varop_mode),
10965 : int_result_mode);
10966 2828140 : rtx count_rtx = gen_int_shift_amount (int_result_mode, count);
10967 2828140 : mask_rtx
10968 2828140 : = simplify_const_binary_operation (code, int_result_mode,
10969 : mask_rtx, count_rtx);
10970 :
10971 : /* Give up if we can't compute an outer operation to use. */
10972 2828140 : if (mask_rtx == 0
10973 2828140 : || !CONST_INT_P (mask_rtx)
10974 5656280 : || ! merge_outer_ops (&outer_op, &outer_const, AND,
10975 : INTVAL (mask_rtx),
10976 : int_result_mode, &complement_p))
10977 : break;
10978 :
10979 : /* If the shifts are in the same direction, we add the
10980 : counts. Otherwise, we subtract them. */
10981 2801477 : if ((code == ASHIFTRT || code == LSHIFTRT)
10982 2801477 : == (first_code == ASHIFTRT || first_code == LSHIFTRT))
10983 11473 : count += first_count;
10984 : else
10985 2790004 : count -= first_count;
10986 :
10987 : /* If COUNT is positive, the new shift is usually CODE,
10988 : except for the two exceptions below, in which case it is
10989 : FIRST_CODE. If the count is negative, FIRST_CODE should
10990 : always be used */
10991 2801477 : if (count > 0
10992 683548 : && ((first_code == ROTATE && code == ASHIFT)
10993 682773 : || (first_code == ASHIFTRT && code == LSHIFTRT)))
10994 : code = first_code;
10995 2790010 : else if (count < 0)
10996 291024 : code = first_code, count = -count;
10997 :
10998 2801477 : varop = XEXP (varop, 0);
10999 2801477 : continue;
11000 2801477 : }
11001 :
11002 : /* If we have (A << B << C) for any shift, we can convert this to
11003 : (A << C << B). This wins if A is a constant. Only try this if
11004 : B is not a constant. */
11005 :
11006 53367 : else if (GET_CODE (varop) == code
11007 5174 : && CONST_INT_P (XEXP (varop, 0))
11008 1013 : && !CONST_INT_P (XEXP (varop, 1)))
11009 : {
11010 : /* For ((unsigned) (cstULL >> count)) >> cst2 we have to make
11011 : sure the result will be masked. See PR70222. */
11012 1013 : if (code == LSHIFTRT
11013 7 : && int_mode != int_result_mode
11014 1020 : && !merge_outer_ops (&outer_op, &outer_const, AND,
11015 7 : GET_MODE_MASK (int_result_mode)
11016 7 : >> orig_count, int_result_mode,
11017 : &complement_p))
11018 : break;
11019 : /* For ((int) (cstLL >> count)) >> cst2 just give up. Queuing
11020 : up outer sign extension (often left and right shift) is
11021 : hardly more efficient than the original. See PR70429.
11022 : Similarly punt for rotates with different modes.
11023 : See PR97386. */
11024 1013 : if ((code == ASHIFTRT || code == ROTATE)
11025 1013 : && int_mode != int_result_mode)
11026 : break;
11027 :
11028 999 : rtx count_rtx = gen_int_shift_amount (int_result_mode, count);
11029 999 : rtx new_rtx = simplify_const_binary_operation (code, int_mode,
11030 : XEXP (varop, 0),
11031 : count_rtx);
11032 999 : varop = gen_rtx_fmt_ee (code, int_mode, new_rtx, XEXP (varop, 1));
11033 999 : count = 0;
11034 999 : continue;
11035 999 : }
11036 : break;
11037 :
11038 53863 : case NOT:
11039 : /* The following rules apply only to scalars. */
11040 53863 : if (shift_mode != shift_unit_mode)
11041 : break;
11042 :
11043 : /* Make this fit the case below. */
11044 53861 : varop = gen_rtx_XOR (mode, XEXP (varop, 0), constm1_rtx);
11045 53861 : continue;
11046 :
11047 779471 : case IOR:
11048 779471 : case AND:
11049 779471 : case XOR:
11050 : /* The following rules apply only to scalars. */
11051 779471 : if (shift_mode != shift_unit_mode)
11052 : break;
11053 777771 : int_varop_mode = as_a <scalar_int_mode> (GET_MODE (varop));
11054 777771 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11055 :
11056 : /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
11057 : with C the size of VAROP - 1 and the shift is logical if
11058 : STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
11059 : we have an (le X 0) operation. If we have an arithmetic shift
11060 : and STORE_FLAG_VALUE is 1 or we have a logical shift with
11061 : STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
11062 :
11063 236207 : if (GET_CODE (varop) == IOR && GET_CODE (XEXP (varop, 0)) == PLUS
11064 1962 : && XEXP (XEXP (varop, 0), 1) == constm1_rtx
11065 : && (STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
11066 221 : && (code == LSHIFTRT || code == ASHIFTRT)
11067 221 : && count == (GET_MODE_PRECISION (int_varop_mode) - 1)
11068 777992 : && rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
11069 : {
11070 53 : count = 0;
11071 53 : varop = gen_rtx_LE (int_varop_mode, XEXP (varop, 1),
11072 : const0_rtx);
11073 :
11074 53 : if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
11075 53 : varop = gen_rtx_NEG (int_varop_mode, varop);
11076 :
11077 53 : continue;
11078 : }
11079 :
11080 : /* If we have (shift (logical)), move the logical to the outside
11081 : to allow it to possibly combine with another logical and the
11082 : shift to combine with another shift. This also canonicalizes to
11083 : what a ZERO_EXTRACT looks like. Also, some machines have
11084 : (and (shift)) insns. */
11085 :
11086 1212584 : if (CONST_INT_P (XEXP (varop, 1))
11087 : /* We can't do this if we have (ashiftrt (xor)) and the
11088 : constant has its sign bit set in shift_unit_mode with
11089 : shift_unit_mode wider than result_mode. */
11090 436444 : && !(code == ASHIFTRT && GET_CODE (varop) == XOR
11091 7305 : && int_result_mode != shift_unit_mode
11092 0 : && trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
11093 : shift_unit_mode) < 0)
11094 436444 : && (new_rtx = simplify_const_binary_operation
11095 436444 : (code, int_result_mode,
11096 436444 : gen_int_mode (INTVAL (XEXP (varop, 1)), int_result_mode),
11097 436444 : gen_int_shift_amount (int_result_mode, count))) != 0
11098 436444 : && CONST_INT_P (new_rtx)
11099 1214162 : && merge_outer_ops (&outer_op, &outer_const, GET_CODE (varop),
11100 : INTVAL (new_rtx), int_result_mode,
11101 : &complement_p))
11102 : {
11103 434866 : varop = XEXP (varop, 0);
11104 434866 : continue;
11105 : }
11106 :
11107 : /* If we can't do that, try to simplify the shift in each arm of the
11108 : logical expression, make a new logical expression, and apply
11109 : the inverse distributive law. This also can't be done for
11110 : (ashiftrt (xor)) where we've widened the shift and the constant
11111 : changes the sign bit. */
11112 342852 : if (CONST_INT_P (XEXP (varop, 1))
11113 342852 : && !(code == ASHIFTRT && GET_CODE (varop) == XOR
11114 48 : && int_result_mode != shift_unit_mode
11115 0 : && trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
11116 : shift_unit_mode) < 0))
11117 : {
11118 1578 : rtx lhs = simplify_shift_const (NULL_RTX, code, shift_unit_mode,
11119 : XEXP (varop, 0), count);
11120 1578 : rtx rhs = simplify_shift_const (NULL_RTX, code, shift_unit_mode,
11121 : XEXP (varop, 1), count);
11122 :
11123 1578 : varop = simplify_gen_binary (GET_CODE (varop), shift_unit_mode,
11124 : lhs, rhs);
11125 1578 : varop = apply_distributive_law (varop);
11126 :
11127 1578 : count = 0;
11128 1578 : continue;
11129 1578 : }
11130 : break;
11131 :
11132 30598 : case EQ:
11133 : /* The following rules apply only to scalars. */
11134 30598 : if (shift_mode != shift_unit_mode)
11135 : break;
11136 30598 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11137 :
11138 : /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
11139 : says that the sign bit can be tested, FOO has mode MODE, C is
11140 : GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
11141 : that may be nonzero. */
11142 30598 : if (code == LSHIFTRT
11143 : && XEXP (varop, 1) == const0_rtx
11144 : && GET_MODE (XEXP (varop, 0)) == int_result_mode
11145 : && count == (GET_MODE_PRECISION (int_result_mode) - 1)
11146 : && HWI_COMPUTABLE_MODE_P (int_result_mode)
11147 : && STORE_FLAG_VALUE == -1
11148 : && nonzero_bits (XEXP (varop, 0), int_result_mode) == 1
11149 : && merge_outer_ops (&outer_op, &outer_const, XOR, 1,
11150 : int_result_mode, &complement_p))
11151 : {
11152 : varop = XEXP (varop, 0);
11153 : count = 0;
11154 : continue;
11155 : }
11156 : break;
11157 :
11158 27982 : case NEG:
11159 : /* The following rules apply only to scalars. */
11160 27982 : if (shift_mode != shift_unit_mode)
11161 : break;
11162 27839 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11163 :
11164 : /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
11165 : than the number of bits in the mode is equivalent to A. */
11166 27844 : if (code == LSHIFTRT
11167 5930 : && count == (GET_MODE_PRECISION (int_result_mode) - 1)
11168 30459 : && nonzero_bits (XEXP (varop, 0), int_result_mode) == 1)
11169 : {
11170 5 : varop = XEXP (varop, 0);
11171 5 : count = 0;
11172 5 : continue;
11173 : }
11174 :
11175 : /* NEG commutes with ASHIFT since it is multiplication. Move the
11176 : NEG outside to allow shifts to combine. */
11177 46432 : if (code == ASHIFT
11178 27834 : && merge_outer_ops (&outer_op, &outer_const, NEG, 0,
11179 : int_result_mode, &complement_p))
11180 : {
11181 18598 : varop = XEXP (varop, 0);
11182 18598 : continue;
11183 : }
11184 : break;
11185 :
11186 1937936 : case PLUS:
11187 : /* The following rules apply only to scalars. */
11188 1937936 : if (shift_mode != shift_unit_mode)
11189 : break;
11190 1893204 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11191 :
11192 : /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
11193 : is one less than the number of bits in the mode is
11194 : equivalent to (xor A 1). */
11195 1893204 : if (code == LSHIFTRT
11196 384364 : && count == (GET_MODE_PRECISION (int_result_mode) - 1)
11197 31201 : && XEXP (varop, 1) == constm1_rtx
11198 13770 : && nonzero_bits (XEXP (varop, 0), int_result_mode) == 1
11199 1893204 : && merge_outer_ops (&outer_op, &outer_const, XOR, 1,
11200 : int_result_mode, &complement_p))
11201 : {
11202 0 : count = 0;
11203 0 : varop = XEXP (varop, 0);
11204 0 : continue;
11205 : }
11206 :
11207 : /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
11208 : that might be nonzero in BAR are those being shifted out and those
11209 : bits are known zero in FOO, we can replace the PLUS with FOO.
11210 : Similarly in the other operand order. This code occurs when
11211 : we are computing the size of a variable-size array. */
11212 :
11213 1896362 : if ((code == ASHIFTRT || code == LSHIFTRT)
11214 544824 : && count < HOST_BITS_PER_WIDE_INT
11215 544740 : && nonzero_bits (XEXP (varop, 1), int_result_mode) >> count == 0
11216 2041525 : && (nonzero_bits (XEXP (varop, 1), int_result_mode)
11217 148321 : & nonzero_bits (XEXP (varop, 0), int_result_mode)) == 0)
11218 : {
11219 3158 : varop = XEXP (varop, 0);
11220 3158 : continue;
11221 : }
11222 1890077 : else if ((code == ASHIFTRT || code == LSHIFTRT)
11223 541666 : && count < HOST_BITS_PER_WIDE_INT
11224 541582 : && HWI_COMPUTABLE_MODE_P (int_result_mode)
11225 540353 : && (nonzero_bits (XEXP (varop, 0), int_result_mode)
11226 540353 : >> count) == 0
11227 1954329 : && (nonzero_bits (XEXP (varop, 0), int_result_mode)
11228 64283 : & nonzero_bits (XEXP (varop, 1), int_result_mode)) == 0)
11229 : {
11230 31 : varop = XEXP (varop, 1);
11231 31 : continue;
11232 : }
11233 :
11234 : /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
11235 2280548 : if (code == ASHIFT
11236 1339695 : && CONST_INT_P (XEXP (varop, 1))
11237 390705 : && (new_rtx = simplify_const_binary_operation
11238 390705 : (ASHIFT, int_result_mode,
11239 390705 : gen_int_mode (INTVAL (XEXP (varop, 1)), int_result_mode),
11240 390705 : gen_int_shift_amount (int_result_mode, count))) != 0
11241 390705 : && CONST_INT_P (new_rtx)
11242 2280720 : && merge_outer_ops (&outer_op, &outer_const, PLUS,
11243 : INTVAL (new_rtx), int_result_mode,
11244 : &complement_p))
11245 : {
11246 390533 : varop = XEXP (varop, 0);
11247 390533 : continue;
11248 : }
11249 :
11250 : /* Check for 'PLUS signbit', which is the canonical form of 'XOR
11251 : signbit', and attempt to change the PLUS to an XOR and move it to
11252 : the outer operation as is done above in the AND/IOR/XOR case
11253 : leg for shift(logical). See details in logical handling above
11254 : for reasoning in doing so. */
11255 1508128 : if (code == LSHIFTRT
11256 381255 : && CONST_INT_P (XEXP (varop, 1))
11257 281235 : && mode_signbit_p (int_result_mode, XEXP (varop, 1))
11258 8646 : && (new_rtx = simplify_const_binary_operation
11259 1499482 : (code, int_result_mode,
11260 8646 : gen_int_mode (INTVAL (XEXP (varop, 1)), int_result_mode),
11261 8646 : gen_int_shift_amount (int_result_mode, count))) != 0
11262 8646 : && CONST_INT_P (new_rtx)
11263 1508128 : && merge_outer_ops (&outer_op, &outer_const, XOR,
11264 : INTVAL (new_rtx), int_result_mode,
11265 : &complement_p))
11266 : {
11267 8646 : varop = XEXP (varop, 0);
11268 8646 : continue;
11269 : }
11270 :
11271 : break;
11272 :
11273 610585 : case MINUS:
11274 : /* The following rules apply only to scalars. */
11275 610585 : if (shift_mode != shift_unit_mode)
11276 : break;
11277 599265 : int_varop_mode = as_a <scalar_int_mode> (GET_MODE (varop));
11278 :
11279 : /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
11280 : with C the size of VAROP - 1 and the shift is logical if
11281 : STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
11282 : we have a (gt X 0) operation. If the shift is arithmetic with
11283 : STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
11284 : we have a (neg (gt X 0)) operation. */
11285 :
11286 599265 : if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
11287 599265 : && GET_CODE (XEXP (varop, 0)) == ASHIFTRT
11288 13364 : && count == (GET_MODE_PRECISION (int_varop_mode) - 1)
11289 45 : && (code == LSHIFTRT || code == ASHIFTRT)
11290 14 : && CONST_INT_P (XEXP (XEXP (varop, 0), 1))
11291 14 : && INTVAL (XEXP (XEXP (varop, 0), 1)) == count
11292 599265 : && rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
11293 : {
11294 0 : count = 0;
11295 0 : varop = gen_rtx_GT (int_varop_mode, XEXP (varop, 1),
11296 : const0_rtx);
11297 :
11298 0 : if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
11299 0 : varop = gen_rtx_NEG (int_varop_mode, varop);
11300 :
11301 0 : continue;
11302 : }
11303 : break;
11304 :
11305 667 : case TRUNCATE:
11306 : /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
11307 : if the truncate does not affect the value. */
11308 667 : if (code == LSHIFTRT
11309 509 : && GET_CODE (XEXP (varop, 0)) == LSHIFTRT
11310 509 : && CONST_INT_P (XEXP (XEXP (varop, 0), 1))
11311 667 : && (INTVAL (XEXP (XEXP (varop, 0), 1))
11312 509 : >= (GET_MODE_UNIT_PRECISION (GET_MODE (XEXP (varop, 0)))
11313 1018 : - GET_MODE_UNIT_PRECISION (GET_MODE (varop)))))
11314 : {
11315 509 : rtx varop_inner = XEXP (varop, 0);
11316 509 : int new_count = count + INTVAL (XEXP (varop_inner, 1));
11317 509 : rtx new_count_rtx = gen_int_shift_amount (GET_MODE (varop_inner),
11318 509 : new_count);
11319 509 : varop_inner = gen_rtx_LSHIFTRT (GET_MODE (varop_inner),
11320 : XEXP (varop_inner, 0),
11321 : new_count_rtx);
11322 509 : varop = gen_rtx_TRUNCATE (GET_MODE (varop), varop_inner);
11323 509 : count = 0;
11324 509 : continue;
11325 509 : }
11326 : break;
11327 :
11328 : default:
11329 : break;
11330 53861 : }
11331 :
11332 : break;
11333 : }
11334 :
11335 23796834 : shift_mode = result_mode;
11336 23796834 : if (shift_mode != mode)
11337 : {
11338 : /* We only change the modes of scalar shifts. */
11339 146052 : int_mode = as_a <scalar_int_mode> (mode);
11340 146052 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11341 146052 : shift_mode = try_widen_shift_mode (code, varop, count, int_result_mode,
11342 : int_mode, outer_op, outer_const);
11343 : }
11344 :
11345 : /* We have now finished analyzing the shift. The result should be
11346 : a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
11347 : OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
11348 : to the result of the shift. OUTER_CONST is the relevant constant,
11349 : but we must turn off all bits turned off in the shift. */
11350 :
11351 23796834 : if (outer_op == UNKNOWN
11352 20190146 : && orig_code == code && orig_count == count
11353 20148768 : && varop == orig_varop
11354 20018551 : && shift_mode == GET_MODE (varop))
11355 : return NULL_RTX;
11356 :
11357 : /* Make a SUBREG if necessary. If we can't make it, fail. */
11358 3781090 : varop = gen_lowpart (shift_mode, varop);
11359 3781090 : if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
11360 : return NULL_RTX;
11361 :
11362 : /* If we have an outer operation and we just made a shift, it is
11363 : possible that we could have simplified the shift were it not
11364 : for the outer operation. So try to do the simplification
11365 : recursively. */
11366 :
11367 3781090 : if (outer_op != UNKNOWN)
11368 3606688 : x = simplify_shift_const_1 (code, shift_mode, varop, count);
11369 : else
11370 : x = NULL_RTX;
11371 :
11372 3606688 : if (x == NULL_RTX)
11373 3745439 : x = simplify_gen_binary (code, shift_mode, varop,
11374 3745439 : gen_int_shift_amount (shift_mode, count));
11375 :
11376 : /* If we were doing an LSHIFTRT in a wider mode than it was originally,
11377 : turn off all the bits that the shift would have turned off. */
11378 3781090 : if (orig_code == LSHIFTRT && result_mode != shift_mode)
11379 : /* We only change the modes of scalar shifts. */
11380 9116 : x = simplify_and_const_int (NULL_RTX, as_a <scalar_int_mode> (shift_mode),
11381 9116 : x, GET_MODE_MASK (result_mode) >> orig_count);
11382 :
11383 : /* Do the remainder of the processing in RESULT_MODE. */
11384 3781090 : x = gen_lowpart_or_truncate (result_mode, x);
11385 :
11386 : /* If COMPLEMENT_P is set, we have to complement X before doing the outer
11387 : operation. */
11388 3781090 : if (complement_p)
11389 23081 : x = simplify_gen_unary (NOT, result_mode, x, result_mode);
11390 :
11391 3781090 : if (outer_op != UNKNOWN)
11392 : {
11393 3606688 : int_result_mode = as_a <scalar_int_mode> (result_mode);
11394 :
11395 3606688 : if (GET_RTX_CLASS (outer_op) != RTX_UNARY
11396 3606688 : && GET_MODE_PRECISION (int_result_mode) < HOST_BITS_PER_WIDE_INT)
11397 1302836 : outer_const = trunc_int_for_mode (outer_const, int_result_mode);
11398 :
11399 3606688 : if (outer_op == AND)
11400 3131870 : x = simplify_and_const_int (NULL_RTX, int_result_mode, x, outer_const);
11401 474818 : else if (outer_op == SET)
11402 : {
11403 : /* This means that we have determined that the result is
11404 : equivalent to a constant. This should be rare. */
11405 0 : if (!side_effects_p (x))
11406 0 : x = GEN_INT (outer_const);
11407 : }
11408 474818 : else if (GET_RTX_CLASS (outer_op) == RTX_UNARY)
11409 18598 : x = simplify_gen_unary (outer_op, int_result_mode, x, int_result_mode);
11410 : else
11411 456220 : x = simplify_gen_binary (outer_op, int_result_mode, x,
11412 : GEN_INT (outer_const));
11413 : }
11414 :
11415 : return x;
11416 : }
11417 :
11418 : /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
11419 : The result of the shift is RESULT_MODE. If we cannot simplify it,
11420 : return X or, if it is NULL, synthesize the expression with
11421 : simplify_gen_binary. Otherwise, return a simplified value.
11422 :
11423 : The shift is normally computed in the widest mode we find in VAROP, as
11424 : long as it isn't a different number of words than RESULT_MODE. Exceptions
11425 : are ASHIFTRT and ROTATE, which are always done in their original mode. */
11426 :
11427 : static rtx
11428 20190288 : simplify_shift_const (rtx x, enum rtx_code code, machine_mode result_mode,
11429 : rtx varop, int count)
11430 : {
11431 20190288 : rtx tem = simplify_shift_const_1 (code, result_mode, varop, count);
11432 20190288 : if (tem)
11433 : return tem;
11434 :
11435 16444849 : if (!x)
11436 4882480 : x = simplify_gen_binary (code, GET_MODE (varop), varop,
11437 4882480 : gen_int_shift_amount (GET_MODE (varop), count));
11438 16444849 : if (GET_MODE (x) != result_mode)
11439 0 : x = gen_lowpart (result_mode, x);
11440 : return x;
11441 : }
11442 :
11443 :
11444 : /* A subroutine of recog_for_combine. See there for arguments and
11445 : return value. */
11446 :
11447 : static int
11448 48581769 : recog_for_combine_1 (rtx *pnewpat, rtx_insn *insn, rtx *pnotes,
11449 : unsigned old_nregs, unsigned new_nregs)
11450 : {
11451 48581769 : rtx pat = *pnewpat;
11452 48581769 : rtx pat_without_clobbers;
11453 48581769 : int insn_code_number;
11454 48581769 : int num_clobbers_to_add = 0;
11455 48581769 : int i;
11456 48581769 : rtx notes = NULL_RTX;
11457 48581769 : rtx old_notes, old_pat;
11458 48581769 : int old_icode;
11459 :
11460 : /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
11461 : we use to indicate that something didn't match. If we find such a
11462 : thing, force rejection. */
11463 48581769 : if (GET_CODE (pat) == PARALLEL)
11464 52414515 : for (i = XVECLEN (pat, 0) - 1; i >= 0; i--)
11465 36093560 : if (GET_CODE (XVECEXP (pat, 0, i)) == CLOBBER
11466 7215457 : && XEXP (XVECEXP (pat, 0, i), 0) == const0_rtx)
11467 : return -1;
11468 :
11469 48579752 : old_pat = PATTERN (insn);
11470 48579752 : old_notes = REG_NOTES (insn);
11471 48579752 : PATTERN (insn) = pat;
11472 48579752 : REG_NOTES (insn) = NULL_RTX;
11473 :
11474 48579752 : insn_code_number = recog (pat, insn, &num_clobbers_to_add);
11475 48579752 : if (dump_file && (dump_flags & TDF_DETAILS))
11476 : {
11477 277 : if (insn_code_number < 0)
11478 177 : fputs ("Failed to match this instruction:\n", dump_file);
11479 : else
11480 100 : fputs ("Successfully matched this instruction:\n", dump_file);
11481 277 : print_rtl_single (dump_file, pat);
11482 : }
11483 :
11484 : /* If it isn't, there is the possibility that we previously had an insn
11485 : that clobbered some register as a side effect, but the combined
11486 : insn doesn't need to do that. So try once more without the clobbers
11487 : unless this represents an ASM insn. */
11488 :
11489 38717157 : if (insn_code_number < 0 && ! check_asm_operands (pat)
11490 87295406 : && GET_CODE (pat) == PARALLEL)
11491 : {
11492 : int pos;
11493 :
11494 50958349 : for (pos = 0, i = 0; i < XVECLEN (pat, 0); i++)
11495 35103823 : if (GET_CODE (XVECEXP (pat, 0, i)) != CLOBBER)
11496 : {
11497 28288550 : if (i != pos)
11498 2248820 : SUBST (XVECEXP (pat, 0, pos), XVECEXP (pat, 0, i));
11499 28288550 : pos++;
11500 : }
11501 :
11502 15854526 : SUBST_INT (XVECLEN (pat, 0), pos);
11503 :
11504 15854526 : if (pos == 1)
11505 4628076 : pat = XVECEXP (pat, 0, 0);
11506 :
11507 15854526 : PATTERN (insn) = pat;
11508 15854526 : insn_code_number = recog (pat, insn, &num_clobbers_to_add);
11509 15854526 : if (dump_file && (dump_flags & TDF_DETAILS))
11510 : {
11511 82 : if (insn_code_number < 0)
11512 81 : fputs ("Failed to match this instruction:\n", dump_file);
11513 : else
11514 1 : fputs ("Successfully matched this instruction:\n", dump_file);
11515 82 : print_rtl_single (dump_file, pat);
11516 : }
11517 : }
11518 :
11519 48579752 : pat_without_clobbers = pat;
11520 :
11521 48579752 : PATTERN (insn) = old_pat;
11522 48579752 : REG_NOTES (insn) = old_notes;
11523 :
11524 : /* Recognize all noop sets, these will be killed by followup pass. */
11525 48579752 : if (insn_code_number < 0 && GET_CODE (pat) == SET && set_noop_p (pat))
11526 210333 : insn_code_number = NOOP_MOVE_INSN_CODE, num_clobbers_to_add = 0;
11527 :
11528 : /* If we had any clobbers to add, make a new pattern than contains
11529 : them. Then check to make sure that all of them are dead. */
11530 48579752 : if (num_clobbers_to_add)
11531 : {
11532 1578668 : rtx newpat = gen_rtx_PARALLEL (VOIDmode,
11533 : rtvec_alloc (GET_CODE (pat) == PARALLEL
11534 : ? (XVECLEN (pat, 0)
11535 : + num_clobbers_to_add)
11536 : : num_clobbers_to_add + 1));
11537 :
11538 1578668 : if (GET_CODE (pat) == PARALLEL)
11539 1461 : for (i = 0; i < XVECLEN (pat, 0); i++)
11540 974 : XVECEXP (newpat, 0, i) = XVECEXP (pat, 0, i);
11541 : else
11542 1578181 : XVECEXP (newpat, 0, 0) = pat;
11543 :
11544 1578668 : add_clobbers (newpat, insn_code_number);
11545 :
11546 3034468 : for (i = XVECLEN (newpat, 0) - num_clobbers_to_add;
11547 3034468 : i < XVECLEN (newpat, 0); i++)
11548 : {
11549 1598619 : if (REG_P (XEXP (XVECEXP (newpat, 0, i), 0))
11550 1598619 : && ! reg_dead_at_p (XEXP (XVECEXP (newpat, 0, i), 0), insn))
11551 : return -1;
11552 1455800 : if (GET_CODE (XEXP (XVECEXP (newpat, 0, i), 0)) != SCRATCH)
11553 : {
11554 1410509 : gcc_assert (REG_P (XEXP (XVECEXP (newpat, 0, i), 0)));
11555 1410509 : notes = alloc_reg_note (REG_UNUSED,
11556 : XEXP (XVECEXP (newpat, 0, i), 0), notes);
11557 : }
11558 : }
11559 : pat = newpat;
11560 : }
11561 :
11562 48436933 : if (insn_code_number >= 0
11563 48436933 : && insn_code_number != NOOP_MOVE_INSN_CODE)
11564 : {
11565 : /* Create the reg dead notes if needed for the regs that were created via split. */
11566 10079544 : for (; old_nregs < new_nregs; old_nregs++)
11567 3236 : notes = alloc_reg_note (REG_DEAD, regno_reg_rtx[old_nregs], notes);
11568 10076308 : old_pat = PATTERN (insn);
11569 10076308 : old_notes = REG_NOTES (insn);
11570 10076308 : old_icode = INSN_CODE (insn);
11571 10076308 : PATTERN (insn) = pat;
11572 10076308 : REG_NOTES (insn) = notes;
11573 10076308 : INSN_CODE (insn) = insn_code_number;
11574 :
11575 : /* Allow targets to reject combined insn. */
11576 10076308 : if (!targetm.legitimate_combined_insn (insn))
11577 : {
11578 3519 : if (dump_file && (dump_flags & TDF_DETAILS))
11579 0 : fputs ("Instruction not appropriate for target.",
11580 : dump_file);
11581 :
11582 : /* Callers expect recog_for_combine to strip
11583 : clobbers from the pattern on failure. */
11584 : pat = pat_without_clobbers;
11585 : notes = NULL_RTX;
11586 :
11587 : insn_code_number = -1;
11588 : }
11589 :
11590 10076308 : PATTERN (insn) = old_pat;
11591 10076308 : REG_NOTES (insn) = old_notes;
11592 10076308 : INSN_CODE (insn) = old_icode;
11593 : }
11594 :
11595 48436933 : *pnewpat = pat;
11596 48436933 : *pnotes = notes;
11597 :
11598 48436933 : return insn_code_number;
11599 : }
11600 :
11601 : /* Change every ZERO_EXTRACT and ZERO_EXTEND of a SUBREG that can be
11602 : expressed as an AND and maybe an LSHIFTRT, to that formulation.
11603 : Return whether anything was so changed. */
11604 :
11605 : static bool
11606 48762140 : change_zero_ext (rtx pat)
11607 : {
11608 48762140 : bool changed = false;
11609 48762140 : rtx *src = &SET_SRC (pat);
11610 :
11611 48762140 : subrtx_ptr_iterator::array_type array;
11612 337906869 : FOR_EACH_SUBRTX_PTR (iter, array, src, NONCONST)
11613 : {
11614 289144729 : rtx x = **iter;
11615 289144729 : scalar_int_mode mode, inner_mode;
11616 289144729 : if (!is_a <scalar_int_mode> (GET_MODE (x), &mode))
11617 425409806 : continue;
11618 152879652 : int size;
11619 :
11620 152879652 : if (GET_CODE (x) == ZERO_EXTRACT
11621 737220 : && CONST_INT_P (XEXP (x, 1))
11622 737198 : && CONST_INT_P (XEXP (x, 2))
11623 693393 : && is_a <scalar_int_mode> (GET_MODE (XEXP (x, 0)), &inner_mode)
11624 153573045 : && GET_MODE_PRECISION (inner_mode) <= GET_MODE_PRECISION (mode))
11625 : {
11626 693385 : size = INTVAL (XEXP (x, 1));
11627 :
11628 693385 : int start = INTVAL (XEXP (x, 2));
11629 693385 : if (BITS_BIG_ENDIAN)
11630 : start = GET_MODE_PRECISION (inner_mode) - size - start;
11631 :
11632 693385 : if (start != 0)
11633 593712 : x = gen_rtx_LSHIFTRT (inner_mode, XEXP (x, 0),
11634 : gen_int_shift_amount (inner_mode, start));
11635 : else
11636 : x = XEXP (x, 0);
11637 :
11638 693385 : if (mode != inner_mode)
11639 : {
11640 157 : if (REG_P (x) && HARD_REGISTER_P (x)
11641 209189 : && !can_change_dest_mode (x, 0, mode))
11642 0 : continue;
11643 :
11644 209189 : x = gen_lowpart_SUBREG (mode, x);
11645 : }
11646 : }
11647 152186267 : else if (GET_CODE (x) == ZERO_EXTEND
11648 2309336 : && GET_CODE (XEXP (x, 0)) == SUBREG
11649 443919 : && SCALAR_INT_MODE_P (GET_MODE (SUBREG_REG (XEXP (x, 0))))
11650 442710 : && !paradoxical_subreg_p (XEXP (x, 0))
11651 152628977 : && subreg_lowpart_p (XEXP (x, 0)))
11652 : {
11653 327790 : inner_mode = as_a <scalar_int_mode> (GET_MODE (XEXP (x, 0)));
11654 327790 : size = GET_MODE_PRECISION (inner_mode);
11655 327790 : x = SUBREG_REG (XEXP (x, 0));
11656 327790 : if (GET_MODE (x) != mode)
11657 : {
11658 11459 : if (REG_P (x) && HARD_REGISTER_P (x)
11659 11851 : && !can_change_dest_mode (x, 0, mode))
11660 0 : continue;
11661 :
11662 11851 : x = gen_lowpart_SUBREG (mode, x);
11663 : }
11664 : }
11665 303716881 : else if (GET_CODE (x) == ZERO_EXTEND
11666 1981546 : && REG_P (XEXP (x, 0))
11667 1077997 : && HARD_REGISTER_P (XEXP (x, 0))
11668 151858550 : && can_change_dest_mode (XEXP (x, 0), 0, mode))
11669 : {
11670 73 : inner_mode = as_a <scalar_int_mode> (GET_MODE (XEXP (x, 0)));
11671 73 : size = GET_MODE_PRECISION (inner_mode);
11672 73 : x = gen_rtx_REG (mode, REGNO (XEXP (x, 0)));
11673 : }
11674 : else
11675 151858404 : continue;
11676 :
11677 1433104 : if (!(GET_CODE (x) == LSHIFTRT
11678 411856 : && CONST_INT_P (XEXP (x, 1))
11679 411856 : && size + INTVAL (XEXP (x, 1)) == GET_MODE_PRECISION (mode)))
11680 : {
11681 868287 : wide_int mask = wi::mask (size, false, GET_MODE_PRECISION (mode));
11682 868287 : x = gen_rtx_AND (mode, x, immed_wide_int_const (mask, mode));
11683 868287 : }
11684 :
11685 1021248 : SUBST (**iter, x);
11686 1021248 : changed = true;
11687 : }
11688 :
11689 48762140 : if (changed)
11690 9242445 : FOR_EACH_SUBRTX_PTR (iter, array, src, NONCONST)
11691 8235565 : maybe_swap_commutative_operands (**iter);
11692 :
11693 48762140 : rtx *dst = &SET_DEST (pat);
11694 48762140 : scalar_int_mode mode;
11695 48762140 : if (GET_CODE (*dst) == ZERO_EXTRACT
11696 9879 : && REG_P (XEXP (*dst, 0))
11697 371 : && is_a <scalar_int_mode> (GET_MODE (XEXP (*dst, 0)), &mode)
11698 371 : && CONST_INT_P (XEXP (*dst, 1))
11699 48762511 : && CONST_INT_P (XEXP (*dst, 2)))
11700 : {
11701 244 : rtx reg = XEXP (*dst, 0);
11702 244 : int width = INTVAL (XEXP (*dst, 1));
11703 244 : int offset = INTVAL (XEXP (*dst, 2));
11704 244 : int reg_width = GET_MODE_PRECISION (mode);
11705 244 : if (BITS_BIG_ENDIAN)
11706 : offset = reg_width - width - offset;
11707 :
11708 244 : rtx x, y, z, w;
11709 244 : wide_int mask = wi::shifted_mask (offset, width, true, reg_width);
11710 244 : wide_int mask2 = wi::shifted_mask (offset, width, false, reg_width);
11711 244 : x = gen_rtx_AND (mode, reg, immed_wide_int_const (mask, mode));
11712 244 : if (offset)
11713 200 : y = gen_rtx_ASHIFT (mode, SET_SRC (pat), GEN_INT (offset));
11714 : else
11715 44 : y = SET_SRC (pat);
11716 244 : z = gen_rtx_AND (mode, y, immed_wide_int_const (mask2, mode));
11717 244 : w = gen_rtx_IOR (mode, x, z);
11718 244 : SUBST (SET_DEST (pat), reg);
11719 244 : SUBST (SET_SRC (pat), w);
11720 :
11721 244 : changed = true;
11722 244 : }
11723 :
11724 48762140 : return changed;
11725 48762140 : }
11726 :
11727 : /* Like recog, but we receive the address of a pointer to a new pattern.
11728 : We try to match the rtx that the pointer points to.
11729 : If that fails, we may try to modify or replace the pattern,
11730 : storing the replacement into the same pointer object.
11731 :
11732 : Modifications include deletion or addition of CLOBBERs. If the
11733 : instruction will still not match, we change ZERO_EXTEND and ZERO_EXTRACT
11734 : to the equivalent AND and perhaps LSHIFTRT patterns, and try with that
11735 : (and undo if that fails).
11736 :
11737 : PNOTES is a pointer to a location where any REG_UNUSED notes added for
11738 : the CLOBBERs are placed.
11739 : If OLD_NREGS != NEW_NREGS, then PNOTES also includes REG_DEAD notes added.
11740 :
11741 : The value is the final insn code from the pattern ultimately matched,
11742 : or -1. */
11743 :
11744 : static int
11745 47341367 : recog_for_combine (rtx *pnewpat, rtx_insn *insn, rtx *pnotes,
11746 : unsigned int old_nregs, unsigned int new_nregs)
11747 : {
11748 47341367 : rtx pat = *pnewpat;
11749 47341367 : int insn_code_number = recog_for_combine_1 (pnewpat, insn, pnotes,
11750 : old_nregs, new_nregs);
11751 47341367 : if (insn_code_number >= 0 || check_asm_operands (pat))
11752 10136178 : return insn_code_number;
11753 :
11754 37205189 : void *marker = get_undo_marker ();
11755 37205189 : bool changed = false;
11756 :
11757 37205189 : if (GET_CODE (pat) == SET)
11758 : {
11759 : /* For an unrecognized single set of a constant, try placing it in
11760 : the constant pool, if this function already uses one. */
11761 21938624 : rtx src = SET_SRC (pat);
11762 21938624 : if (CONSTANT_P (src)
11763 452924 : && !CONST_INT_P (src)
11764 403752 : && crtl->uses_const_pool
11765 354792 : && SET_DEST (pat) != pc_rtx)
11766 : {
11767 354790 : machine_mode mode = GET_MODE (src);
11768 354790 : if (mode == VOIDmode)
11769 1215 : mode = GET_MODE (SET_DEST (pat));
11770 354790 : src = force_const_mem (mode, src);
11771 354790 : if (src)
11772 : {
11773 354780 : SUBST (SET_SRC (pat), src);
11774 354780 : changed = true;
11775 : }
11776 : }
11777 : else
11778 21583834 : changed = change_zero_ext (pat);
11779 : }
11780 15266565 : else if (GET_CODE (pat) == PARALLEL)
11781 : {
11782 : int i;
11783 42652903 : for (i = 0; i < XVECLEN (pat, 0); i++)
11784 : {
11785 27405485 : rtx set = XVECEXP (pat, 0, i);
11786 27405485 : if (GET_CODE (set) == SET)
11787 27178306 : changed |= change_zero_ext (set);
11788 : }
11789 : }
11790 :
11791 37186032 : if (changed)
11792 : {
11793 1240402 : insn_code_number = recog_for_combine_1 (pnewpat, insn, pnotes,
11794 : old_nregs, new_nregs);
11795 :
11796 1240402 : if (insn_code_number < 0)
11797 1091955 : undo_to_marker (marker);
11798 : }
11799 :
11800 : return insn_code_number;
11801 : }
11802 :
11803 : /* Like gen_lowpart_general but for use by combine. In combine it
11804 : is not possible to create any new pseudoregs. However, it is
11805 : safe to create invalid memory addresses, because combine will
11806 : try to recognize them and all they will do is make the combine
11807 : attempt fail.
11808 :
11809 : If for some reason this cannot do its job, an rtx
11810 : (clobber (const_int 0)) is returned.
11811 : An insn containing that will not be recognized. */
11812 :
11813 : static rtx
11814 151973896 : gen_lowpart_for_combine (machine_mode omode, rtx x)
11815 : {
11816 151973896 : machine_mode imode = GET_MODE (x);
11817 151973896 : rtx result;
11818 :
11819 151973896 : if (omode == imode)
11820 : return x;
11821 :
11822 : /* We can only support MODE being wider than a word if X is a
11823 : constant integer or has a mode the same size. */
11824 50358929 : if (maybe_gt (GET_MODE_SIZE (omode), UNITS_PER_WORD)
11825 23843231 : && ! (CONST_SCALAR_INT_P (x)
11826 9735480 : || known_eq (GET_MODE_SIZE (imode), GET_MODE_SIZE (omode))))
11827 2913645 : goto fail;
11828 :
11829 : /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
11830 : won't know what to do. So we will strip off the SUBREG here and
11831 : process normally. */
11832 20929586 : if (GET_CODE (x) == SUBREG && MEM_P (SUBREG_REG (x)))
11833 : {
11834 15021 : x = SUBREG_REG (x);
11835 :
11836 : /* For use in case we fall down into the address adjustments
11837 : further below, we need to adjust the known mode and size of
11838 : x; imode and isize, since we just adjusted x. */
11839 15021 : imode = GET_MODE (x);
11840 :
11841 15021 : if (imode == omode)
11842 : return x;
11843 : }
11844 :
11845 20920421 : result = gen_lowpart_common (omode, x);
11846 :
11847 20920421 : if (result)
11848 : return result;
11849 :
11850 8261873 : if (MEM_P (x))
11851 : {
11852 : /* Refuse to work on a volatile memory ref or one with a mode-dependent
11853 : address. */
11854 2035606 : if (MEM_VOLATILE_P (x)
11855 4026775 : || mode_dependent_address_p (XEXP (x, 0), MEM_ADDR_SPACE (x)))
11856 44467 : goto fail;
11857 :
11858 : /* If we want to refer to something bigger than the original memref,
11859 : generate a paradoxical subreg instead. That will force a reload
11860 : of the original memref X. */
11861 1991139 : if (paradoxical_subreg_p (omode, imode)
11862 1991139 : && validate_subreg (omode, GET_MODE (x), x, 0))
11863 1774502 : return gen_rtx_SUBREG (omode, x, 0);
11864 :
11865 216637 : poly_int64 offset = byte_lowpart_offset (omode, imode);
11866 216637 : return adjust_address_nv (x, omode, offset);
11867 : }
11868 :
11869 : /* If X is a comparison operator, rewrite it in a new mode. This
11870 : probably won't match, but may allow further simplifications. */
11871 6226267 : else if (COMPARISON_P (x)
11872 138957 : && SCALAR_INT_MODE_P (imode)
11873 52280 : && SCALAR_INT_MODE_P (omode))
11874 52269 : return gen_rtx_fmt_ee (GET_CODE (x), omode, XEXP (x, 0), XEXP (x, 1));
11875 :
11876 : /* If we couldn't simplify X any other way, just enclose it in a
11877 : SUBREG. Normally, this SUBREG won't match, but some patterns may
11878 : include an explicit SUBREG or we may simplify it further in combine. */
11879 : else
11880 : {
11881 6173998 : rtx res;
11882 :
11883 6173998 : if (imode == VOIDmode)
11884 : {
11885 8 : imode = int_mode_for_mode (omode).require ();
11886 8 : x = gen_lowpart_common (imode, x);
11887 8 : if (x == NULL)
11888 0 : goto fail;
11889 : }
11890 6173998 : res = lowpart_subreg (omode, x, imode);
11891 6173998 : if (res)
11892 : return res;
11893 : }
11894 :
11895 19601 : fail:
11896 2977713 : return gen_rtx_CLOBBER (omode, const0_rtx);
11897 : }
11898 :
11899 : /* Like gen_lowpart_for_combine but returns NULL_RTX
11900 : for an error instead of CLOBBER.
11901 : Note no_emit is not called directly from combine but rather from
11902 : simplify_rtx and is expecting a NULL on failure rather than
11903 : a CLOBBER. */
11904 :
11905 : static rtx
11906 757029 : gen_lowpart_for_combine_no_emit (machine_mode omode, rtx x)
11907 : {
11908 757029 : rtx tem = gen_lowpart_for_combine (omode, x);
11909 757029 : if (!tem || GET_CODE (tem) == CLOBBER)
11910 15307 : return NULL_RTX;
11911 : return tem;
11912 : }
11913 :
11914 :
11915 : /* Try to simplify a comparison between OP0 and a constant OP1,
11916 : where CODE is the comparison code that will be tested, into a
11917 : (CODE OP0 const0_rtx) form.
11918 :
11919 : The result is a possibly different comparison code to use.
11920 : *POP0 and *POP1 may be updated. */
11921 :
11922 : static enum rtx_code
11923 15466087 : simplify_compare_const (enum rtx_code code, machine_mode mode,
11924 : rtx *pop0, rtx *pop1)
11925 : {
11926 15466087 : scalar_int_mode int_mode;
11927 15466087 : rtx op0 = *pop0;
11928 15466087 : HOST_WIDE_INT const_op = INTVAL (*pop1);
11929 :
11930 : /* Get the constant we are comparing against and turn off all bits
11931 : not on in our mode. */
11932 15466087 : if (mode != VOIDmode)
11933 15170377 : const_op = trunc_int_for_mode (const_op, mode);
11934 :
11935 : /* If we are comparing against a constant power of two and the value
11936 : being compared can only have that single bit nonzero (e.g., it was
11937 : `and'ed with that bit), we can replace this with a comparison
11938 : with zero. */
11939 15466087 : if (const_op
11940 4139508 : && (code == EQ || code == NE || code == GEU || code == LTU
11941 : /* This optimization is incorrect for signed >= INT_MIN or
11942 : < INT_MIN, those are always true or always false. */
11943 24742 : || ((code == GE || code == LT) && const_op > 0))
11944 2790289 : && is_a <scalar_int_mode> (mode, &int_mode)
11945 2790289 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
11946 2771130 : && pow2p_hwi (const_op & GET_MODE_MASK (int_mode))
11947 16392654 : && (nonzero_bits (op0, int_mode)
11948 926567 : == (unsigned HOST_WIDE_INT) (const_op & GET_MODE_MASK (int_mode))))
11949 : {
11950 4534 : code = (code == EQ || code == GE || code == GEU ? NE : EQ);
11951 : const_op = 0;
11952 : }
11953 :
11954 : /* Similarly, if we are comparing a value known to be either -1 or
11955 : 0 with -1, change it to the opposite comparison against zero. */
11956 15461553 : if (const_op == -1
11957 229523 : && (code == EQ || code == NE || code == GT || code == LE
11958 : || code == GEU || code == LTU)
11959 15682106 : && is_a <scalar_int_mode> (mode, &int_mode)
11960 15688522 : && num_sign_bit_copies (op0, int_mode) == GET_MODE_PRECISION (int_mode))
11961 : {
11962 12211 : code = (code == EQ || code == LE || code == GEU ? NE : EQ);
11963 : const_op = 0;
11964 : }
11965 :
11966 : /* Do some canonicalizations based on the comparison code. We prefer
11967 : comparisons against zero and then prefer equality comparisons.
11968 : If we can reduce the size of a constant, we will do that too. */
11969 15455137 : switch (code)
11970 : {
11971 268882 : case LT:
11972 : /* < C is equivalent to <= (C - 1) */
11973 268882 : if (const_op > 0)
11974 : {
11975 5004 : const_op -= 1;
11976 5004 : code = LE;
11977 : /* ... fall through to LE case below. */
11978 402284 : gcc_fallthrough ();
11979 : }
11980 : else
11981 : break;
11982 :
11983 402284 : case LE:
11984 : /* <= C is equivalent to < (C + 1); we do this for C < 0 */
11985 402284 : if (const_op < 0)
11986 : {
11987 52 : const_op += 1;
11988 52 : code = LT;
11989 : }
11990 :
11991 : /* If we are doing a <= 0 comparison on a value known to have
11992 : a zero sign bit, we can replace this with == 0. */
11993 402232 : else if (const_op == 0
11994 276200 : && is_a <scalar_int_mode> (mode, &int_mode)
11995 276200 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
11996 678432 : && (nonzero_bits (op0, int_mode)
11997 276200 : & (HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1)))
11998 276200 : == 0)
11999 : code = EQ;
12000 : break;
12001 :
12002 235982 : case GE:
12003 : /* >= C is equivalent to > (C - 1). */
12004 235982 : if (const_op > 0)
12005 : {
12006 2411 : const_op -= 1;
12007 2411 : code = GT;
12008 : /* ... fall through to GT below. */
12009 246868 : gcc_fallthrough ();
12010 : }
12011 : else
12012 : break;
12013 :
12014 246868 : case GT:
12015 : /* > C is equivalent to >= (C + 1); we do this for C < 0. */
12016 246868 : if (const_op < 0)
12017 : {
12018 155 : const_op += 1;
12019 155 : code = GE;
12020 : }
12021 :
12022 : /* If we are doing a > 0 comparison on a value known to have
12023 : a zero sign bit, we can replace this with != 0. */
12024 246713 : else if (const_op == 0
12025 123726 : && is_a <scalar_int_mode> (mode, &int_mode)
12026 123726 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
12027 370439 : && (nonzero_bits (op0, int_mode)
12028 123726 : & (HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1)))
12029 123726 : == 0)
12030 : code = NE;
12031 : break;
12032 :
12033 99552 : case LTU:
12034 : /* < C is equivalent to <= (C - 1). */
12035 99552 : if (const_op > 0)
12036 : {
12037 89457 : const_op -= 1;
12038 89457 : code = LEU;
12039 : /* ... fall through ... */
12040 89457 : gcc_fallthrough ();
12041 : }
12042 : /* (unsigned) < 0x80000000 is equivalent to >= 0. */
12043 10095 : else if (is_a <scalar_int_mode> (mode, &int_mode)
12044 10095 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
12045 9325 : && (((unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode))
12046 9325 : == HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1)))
12047 : {
12048 : const_op = 0;
12049 : code = GE;
12050 : break;
12051 : }
12052 : else
12053 : break;
12054 :
12055 684981 : case LEU:
12056 : /* unsigned <= 0 is equivalent to == 0 */
12057 684981 : if (const_op == 0)
12058 : code = EQ;
12059 : /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
12060 684643 : else if (is_a <scalar_int_mode> (mode, &int_mode)
12061 684643 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
12062 682702 : && ((unsigned HOST_WIDE_INT) const_op
12063 : == ((HOST_WIDE_INT_1U
12064 682702 : << (GET_MODE_PRECISION (int_mode) - 1)) - 1)))
12065 : {
12066 : const_op = 0;
12067 : code = GE;
12068 : }
12069 : break;
12070 :
12071 31484 : case GEU:
12072 : /* >= C is equivalent to > (C - 1). */
12073 31484 : if (const_op > 1)
12074 : {
12075 22540 : const_op -= 1;
12076 22540 : code = GTU;
12077 : /* ... fall through ... */
12078 22540 : gcc_fallthrough ();
12079 : }
12080 :
12081 : /* (unsigned) >= 0x80000000 is equivalent to < 0. */
12082 8944 : else if (is_a <scalar_int_mode> (mode, &int_mode)
12083 8944 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
12084 7673 : && (((unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode))
12085 7673 : == HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1)))
12086 : {
12087 : const_op = 0;
12088 : code = LT;
12089 : break;
12090 : }
12091 : else
12092 : break;
12093 :
12094 517097 : case GTU:
12095 : /* unsigned > 0 is equivalent to != 0 */
12096 517097 : if (const_op == 0)
12097 : code = NE;
12098 : /* (unsigned) > 0x7fffffff is equivalent to < 0. */
12099 517097 : else if (is_a <scalar_int_mode> (mode, &int_mode)
12100 517097 : && GET_MODE_PRECISION (int_mode) - 1 < HOST_BITS_PER_WIDE_INT
12101 516001 : && ((unsigned HOST_WIDE_INT) const_op
12102 : == (HOST_WIDE_INT_1U
12103 516001 : << (GET_MODE_PRECISION (int_mode) - 1)) - 1))
12104 : {
12105 : const_op = 0;
12106 : code = LT;
12107 : }
12108 : break;
12109 :
12110 : default:
12111 : break;
12112 : }
12113 :
12114 : /* Narrow non-symmetric comparison of memory and constant as e.g.
12115 : x0...x7 <= 0x3fffffffffffffff into x0 <= 0x3f where x0 is the most
12116 : significant byte. Likewise, transform x0...x7 >= 0x4000000000000000 into
12117 : x0 >= 0x40. */
12118 14763395 : if ((code == LEU || code == LTU || code == GEU || code == GTU)
12119 1217387 : && is_a <scalar_int_mode> (GET_MODE (op0), &int_mode)
12120 1217366 : && HWI_COMPUTABLE_MODE_P (int_mode)
12121 1212288 : && MEM_P (op0)
12122 85718 : && !MEM_VOLATILE_P (op0)
12123 : /* The optimization makes only sense for constants which are big enough
12124 : so that we have a chance to chop off something at all. */
12125 84856 : && ((unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode)) > 0xff
12126 : /* Ensure that we do not overflow during normalization. */
12127 25447 : && (code != GTU
12128 3649 : || ((unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode))
12129 : < HOST_WIDE_INT_M1U)
12130 15491534 : && trunc_int_for_mode (const_op, int_mode) == const_op)
12131 : {
12132 25447 : unsigned HOST_WIDE_INT n
12133 25447 : = (unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode);
12134 25447 : enum rtx_code adjusted_code;
12135 :
12136 : /* Normalize code to either LEU or GEU. */
12137 25447 : if (code == LTU)
12138 : {
12139 323 : --n;
12140 323 : adjusted_code = LEU;
12141 : }
12142 25124 : else if (code == GTU)
12143 : {
12144 3649 : ++n;
12145 3649 : adjusted_code = GEU;
12146 : }
12147 : else
12148 : adjusted_code = code;
12149 :
12150 25447 : scalar_int_mode narrow_mode_iter;
12151 78963 : FOR_EACH_MODE_UNTIL (narrow_mode_iter, int_mode)
12152 : {
12153 54157 : unsigned nbits = GET_MODE_PRECISION (int_mode)
12154 54157 : - GET_MODE_PRECISION (narrow_mode_iter);
12155 54157 : unsigned HOST_WIDE_INT mask = (HOST_WIDE_INT_1U << nbits) - 1;
12156 54157 : unsigned HOST_WIDE_INT lower_bits = n & mask;
12157 54157 : if ((adjusted_code == LEU && lower_bits == mask)
12158 53912 : || (adjusted_code == GEU && lower_bits == 0))
12159 : {
12160 641 : n >>= nbits;
12161 641 : break;
12162 : }
12163 : }
12164 :
12165 25447 : if (narrow_mode_iter < int_mode)
12166 : {
12167 641 : if (dump_file && (dump_flags & TDF_DETAILS))
12168 : {
12169 12 : fprintf (
12170 : dump_file, "narrow comparison from mode %s to %s: (MEM %s "
12171 : HOST_WIDE_INT_PRINT_HEX ") to (MEM %s "
12172 12 : HOST_WIDE_INT_PRINT_HEX ").\n", GET_MODE_NAME (int_mode),
12173 12 : GET_MODE_NAME (narrow_mode_iter), GET_RTX_NAME (code),
12174 12 : (unsigned HOST_WIDE_INT) const_op & GET_MODE_MASK (int_mode),
12175 12 : GET_RTX_NAME (adjusted_code), n);
12176 : }
12177 641 : poly_int64 offset = (BYTES_BIG_ENDIAN
12178 641 : ? 0
12179 641 : : (GET_MODE_SIZE (int_mode)
12180 641 : - GET_MODE_SIZE (narrow_mode_iter)));
12181 641 : *pop0 = adjust_address_nv (op0, narrow_mode_iter, offset);
12182 641 : *pop1 = gen_int_mode (n, narrow_mode_iter);
12183 641 : return adjusted_code;
12184 : }
12185 : }
12186 :
12187 15465446 : *pop1 = GEN_INT (const_op);
12188 15465446 : return code;
12189 : }
12190 :
12191 : /* Simplify a comparison between *POP0 and *POP1 where CODE is the
12192 : comparison code that will be tested.
12193 :
12194 : The result is a possibly different comparison code to use. *POP0 and
12195 : *POP1 may be updated.
12196 :
12197 : It is possible that we might detect that a comparison is either always
12198 : true or always false. However, we do not perform general constant
12199 : folding in combine, so this knowledge isn't useful. Such tautologies
12200 : should have been detected earlier. Hence we ignore all such cases. */
12201 :
12202 : static enum rtx_code
12203 23467258 : simplify_comparison (enum rtx_code code, rtx *pop0, rtx *pop1)
12204 : {
12205 23467258 : rtx op0 = *pop0;
12206 23467258 : rtx op1 = *pop1;
12207 23467258 : rtx tem, tem1;
12208 23467258 : int i;
12209 23467258 : scalar_int_mode mode, inner_mode, tmode;
12210 23467258 : opt_scalar_int_mode tmode_iter;
12211 :
12212 : /* Try a few ways of applying the same transformation to both operands. */
12213 23467542 : while (1)
12214 : {
12215 : /* The test below this one won't handle SIGN_EXTENDs on these machines,
12216 : so check specially. */
12217 23467542 : if (!WORD_REGISTER_OPERATIONS
12218 23467542 : && code != GTU && code != GEU && code != LTU && code != LEU
12219 20242316 : && GET_CODE (op0) == ASHIFTRT && GET_CODE (op1) == ASHIFTRT
12220 1663 : && GET_CODE (XEXP (op0, 0)) == ASHIFT
12221 1240 : && GET_CODE (XEXP (op1, 0)) == ASHIFT
12222 724 : && GET_CODE (XEXP (XEXP (op0, 0), 0)) == SUBREG
12223 724 : && GET_CODE (XEXP (XEXP (op1, 0), 0)) == SUBREG
12224 724 : && is_a <scalar_int_mode> (GET_MODE (op0), &mode)
12225 : && (is_a <scalar_int_mode>
12226 724 : (GET_MODE (SUBREG_REG (XEXP (XEXP (op0, 0), 0))), &inner_mode))
12227 724 : && inner_mode == GET_MODE (SUBREG_REG (XEXP (XEXP (op1, 0), 0)))
12228 724 : && CONST_INT_P (XEXP (op0, 1))
12229 724 : && XEXP (op0, 1) == XEXP (op1, 1)
12230 91 : && XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
12231 91 : && XEXP (op0, 1) == XEXP (XEXP (op1, 0), 1)
12232 91 : && (INTVAL (XEXP (op0, 1))
12233 91 : == (GET_MODE_PRECISION (mode)
12234 91 : - GET_MODE_PRECISION (inner_mode))))
12235 : {
12236 91 : op0 = SUBREG_REG (XEXP (XEXP (op0, 0), 0));
12237 91 : op1 = SUBREG_REG (XEXP (XEXP (op1, 0), 0));
12238 : }
12239 :
12240 : /* If both operands are the same constant shift, see if we can ignore the
12241 : shift. We can if the shift is a rotate or if the bits shifted out of
12242 : this shift are known to be zero for both inputs and if the type of
12243 : comparison is compatible with the shift. */
12244 23467542 : if (GET_CODE (op0) == GET_CODE (op1)
12245 3543762 : && HWI_COMPUTABLE_MODE_P (GET_MODE (op0))
12246 3224592 : && ((GET_CODE (op0) == ROTATE && (code == NE || code == EQ))
12247 3224592 : || ((GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFT)
12248 1064 : && (code != GT && code != LT && code != GE && code != LE))
12249 3223578 : || (GET_CODE (op0) == ASHIFTRT
12250 1584 : && (code != GTU && code != LTU
12251 1576 : && code != GEU && code != LEU)))
12252 2586 : && CONST_INT_P (XEXP (op0, 1))
12253 2551 : && INTVAL (XEXP (op0, 1)) >= 0
12254 2551 : && INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
12255 23470093 : && XEXP (op0, 1) == XEXP (op1, 1))
12256 : {
12257 1321 : machine_mode mode = GET_MODE (op0);
12258 1321 : unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
12259 1321 : int shift_count = INTVAL (XEXP (op0, 1));
12260 :
12261 1321 : if (GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFTRT)
12262 776 : mask &= (mask >> shift_count) << shift_count;
12263 545 : else if (GET_CODE (op0) == ASHIFT)
12264 545 : mask = (mask & (mask << shift_count)) >> shift_count;
12265 :
12266 1321 : if ((nonzero_bits (XEXP (op0, 0), mode) & ~mask) == 0
12267 1321 : && (nonzero_bits (XEXP (op1, 0), mode) & ~mask) == 0)
12268 127 : op0 = XEXP (op0, 0), op1 = XEXP (op1, 0);
12269 : else
12270 : break;
12271 : }
12272 :
12273 : /* If both operands are AND's of a paradoxical SUBREG by constant, the
12274 : SUBREGs are of the same mode, and, in both cases, the AND would
12275 : be redundant if the comparison was done in the narrower mode,
12276 : do the comparison in the narrower mode (e.g., we are AND'ing with 1
12277 : and the operand's possibly nonzero bits are 0xffffff01; in that case
12278 : if we only care about QImode, we don't need the AND). This case
12279 : occurs if the output mode of an scc insn is not SImode and
12280 : STORE_FLAG_VALUE == 1 (e.g., the 386).
12281 :
12282 : Similarly, check for a case where the AND's are ZERO_EXTEND
12283 : operations from some narrower mode even though a SUBREG is not
12284 : present. */
12285 :
12286 23466221 : else if (GET_CODE (op0) == AND && GET_CODE (op1) == AND
12287 2612 : && CONST_INT_P (XEXP (op0, 1))
12288 2517 : && CONST_INT_P (XEXP (op1, 1)))
12289 : {
12290 2501 : rtx inner_op0 = XEXP (op0, 0);
12291 2501 : rtx inner_op1 = XEXP (op1, 0);
12292 2501 : HOST_WIDE_INT c0 = INTVAL (XEXP (op0, 1));
12293 2501 : HOST_WIDE_INT c1 = INTVAL (XEXP (op1, 1));
12294 2501 : bool changed = false;
12295 :
12296 2501 : if (paradoxical_subreg_p (inner_op0)
12297 1003 : && GET_CODE (inner_op1) == SUBREG
12298 465 : && HWI_COMPUTABLE_MODE_P (GET_MODE (SUBREG_REG (inner_op0)))
12299 465 : && (GET_MODE (SUBREG_REG (inner_op0))
12300 465 : == GET_MODE (SUBREG_REG (inner_op1)))
12301 196 : && ((~c0) & nonzero_bits (SUBREG_REG (inner_op0),
12302 : GET_MODE (SUBREG_REG (inner_op0)))) == 0
12303 1710 : && ((~c1) & nonzero_bits (SUBREG_REG (inner_op1),
12304 126 : GET_MODE (SUBREG_REG (inner_op1)))) == 0)
12305 : {
12306 110 : op0 = SUBREG_REG (inner_op0);
12307 110 : op1 = SUBREG_REG (inner_op1);
12308 :
12309 : /* The resulting comparison is always unsigned since we masked
12310 : off the original sign bit. */
12311 110 : code = unsigned_condition (code);
12312 :
12313 110 : changed = true;
12314 : }
12315 :
12316 2391 : else if (c0 == c1)
12317 5140 : FOR_EACH_MODE_UNTIL (tmode,
12318 : as_a <scalar_int_mode> (GET_MODE (op0)))
12319 3126 : if ((unsigned HOST_WIDE_INT) c0 == GET_MODE_MASK (tmode))
12320 : {
12321 35 : op0 = gen_lowpart_or_truncate (tmode, inner_op0);
12322 35 : op1 = gen_lowpart_or_truncate (tmode, inner_op1);
12323 35 : code = unsigned_condition (code);
12324 35 : changed = true;
12325 35 : break;
12326 : }
12327 :
12328 2159 : if (! changed)
12329 : break;
12330 : }
12331 :
12332 : /* If both operands are NOT, we can strip off the outer operation
12333 : and adjust the comparison code for swapped operands; similarly for
12334 : NEG, except that this must be an equality comparison. */
12335 23463720 : else if ((GET_CODE (op0) == NOT && GET_CODE (op1) == NOT)
12336 23463719 : || (GET_CODE (op0) == NEG && GET_CODE (op1) == NEG
12337 11 : && (code == EQ || code == NE)))
12338 12 : op0 = XEXP (op0, 0), op1 = XEXP (op1, 0), code = swap_condition (code);
12339 :
12340 : else
12341 : break;
12342 : }
12343 :
12344 : /* If the first operand is a constant, swap the operands and adjust the
12345 : comparison code appropriately, but don't do this if the second operand
12346 : is already a constant integer. */
12347 23467258 : if (swap_commutative_operands_p (op0, op1))
12348 : {
12349 1527696 : std::swap (op0, op1);
12350 1527696 : code = swap_condition (code);
12351 : }
12352 :
12353 : /* We now enter a loop during which we will try to simplify the comparison.
12354 : For the most part, we only are concerned with comparisons with zero,
12355 : but some things may really be comparisons with zero but not start
12356 : out looking that way. */
12357 :
12358 24689295 : while (CONST_INT_P (op1))
12359 : {
12360 15855064 : machine_mode raw_mode = GET_MODE (op0);
12361 15855064 : scalar_int_mode int_mode;
12362 15855064 : int equality_comparison_p;
12363 15855064 : int sign_bit_comparison_p;
12364 15855064 : int unsigned_comparison_p;
12365 15855064 : HOST_WIDE_INT const_op;
12366 :
12367 : /* We only want to handle integral modes. This catches VOIDmode,
12368 : CCmode, and the floating-point modes. An exception is that we
12369 : can handle VOIDmode if OP0 is a COMPARE or a comparison
12370 : operation. */
12371 :
12372 15855064 : if (GET_MODE_CLASS (raw_mode) != MODE_INT
12373 1501214 : && ! (raw_mode == VOIDmode
12374 295750 : && (GET_CODE (op0) == COMPARE || COMPARISON_P (op0))))
12375 : break;
12376 :
12377 : /* Try to simplify the compare to constant, possibly changing the
12378 : comparison op, and/or changing op1 to zero. */
12379 14649560 : code = simplify_compare_const (code, raw_mode, &op0, &op1);
12380 14649560 : const_op = INTVAL (op1);
12381 :
12382 : /* Compute some predicates to simplify code below. */
12383 :
12384 14649560 : equality_comparison_p = (code == EQ || code == NE);
12385 14649560 : sign_bit_comparison_p = ((code == LT || code == GE) && const_op == 0);
12386 14649560 : unsigned_comparison_p = (code == LTU || code == LEU || code == GTU
12387 14649560 : || code == GEU);
12388 :
12389 : /* If this is a sign bit comparison and we can do arithmetic in
12390 : MODE, say that we will only be needing the sign bit of OP0. */
12391 14649560 : if (sign_bit_comparison_p
12392 453814 : && is_a <scalar_int_mode> (raw_mode, &int_mode)
12393 15103374 : && HWI_COMPUTABLE_MODE_P (int_mode))
12394 453422 : op0 = force_to_mode (op0, int_mode,
12395 : HOST_WIDE_INT_1U
12396 453422 : << (GET_MODE_PRECISION (int_mode) - 1), false);
12397 :
12398 14649560 : if (COMPARISON_P (op0))
12399 : {
12400 : /* We can't do anything if OP0 is a condition code value, rather
12401 : than an actual data value. */
12402 688268 : if (const_op != 0
12403 688268 : || GET_MODE_CLASS (GET_MODE (XEXP (op0, 0))) == MODE_CC)
12404 : break;
12405 :
12406 : /* Get the two operands being compared. */
12407 105958 : if (GET_CODE (XEXP (op0, 0)) == COMPARE)
12408 0 : tem = XEXP (XEXP (op0, 0), 0), tem1 = XEXP (XEXP (op0, 0), 1);
12409 : else
12410 105958 : tem = XEXP (op0, 0), tem1 = XEXP (op0, 1);
12411 :
12412 : /* Check for the cases where we simply want the result of the
12413 : earlier test or the opposite of that result. */
12414 105958 : if (code == NE || code == EQ
12415 105958 : || (val_signbit_known_set_p (raw_mode, STORE_FLAG_VALUE)
12416 0 : && (code == LT || code == GE)))
12417 : {
12418 105958 : enum rtx_code new_code;
12419 105958 : if (code == LT || code == NE)
12420 105958 : new_code = GET_CODE (op0);
12421 : else
12422 0 : new_code = reversed_comparison_code (op0, NULL);
12423 :
12424 105958 : if (new_code != UNKNOWN)
12425 : {
12426 105958 : code = new_code;
12427 105958 : op0 = tem;
12428 105958 : op1 = tem1;
12429 24795253 : continue;
12430 : }
12431 : }
12432 : break;
12433 : }
12434 :
12435 13961292 : if (raw_mode == VOIDmode)
12436 : break;
12437 13961292 : scalar_int_mode mode = as_a <scalar_int_mode> (raw_mode);
12438 :
12439 : /* Now try cases based on the opcode of OP0. If none of the cases
12440 : does a "continue", we exit this loop immediately after the
12441 : switch. */
12442 :
12443 13961292 : unsigned int mode_width = GET_MODE_PRECISION (mode);
12444 13961292 : unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
12445 13961292 : switch (GET_CODE (op0))
12446 : {
12447 373877 : case ZERO_EXTRACT:
12448 : /* If we are extracting a single bit from a variable position in
12449 : a constant that has only a single bit set and are comparing it
12450 : with zero, we can convert this into an equality comparison
12451 : between the position and the location of the single bit. */
12452 : /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
12453 : have already reduced the shift count modulo the word size. */
12454 373877 : if (!SHIFT_COUNT_TRUNCATED
12455 373877 : && CONST_INT_P (XEXP (op0, 0))
12456 9492 : && XEXP (op0, 1) == const1_rtx
12457 9474 : && equality_comparison_p && const_op == 0
12458 383351 : && (i = exact_log2 (UINTVAL (XEXP (op0, 0)))) >= 0)
12459 : {
12460 0 : if (BITS_BIG_ENDIAN)
12461 : i = BITS_PER_WORD - 1 - i;
12462 :
12463 0 : op0 = XEXP (op0, 2);
12464 0 : op1 = GEN_INT (i);
12465 0 : const_op = i;
12466 :
12467 : /* Result is nonzero iff shift count is equal to I. */
12468 0 : code = reverse_condition (code);
12469 0 : continue;
12470 : }
12471 :
12472 : /* fall through */
12473 :
12474 373881 : case SIGN_EXTRACT:
12475 373881 : tem = expand_compound_operation (op0);
12476 373881 : if (tem != op0)
12477 : {
12478 338949 : op0 = tem;
12479 338949 : continue;
12480 : }
12481 : break;
12482 :
12483 27452 : case NOT:
12484 : /* If testing for equality, we can take the NOT of the constant. */
12485 38447 : if (equality_comparison_p
12486 27452 : && (tem = simplify_unary_operation (NOT, mode, op1, mode)) != 0)
12487 : {
12488 10995 : op0 = XEXP (op0, 0);
12489 10995 : op1 = tem;
12490 10995 : continue;
12491 : }
12492 :
12493 : /* If just looking at the sign bit, reverse the sense of the
12494 : comparison. */
12495 16457 : if (sign_bit_comparison_p)
12496 : {
12497 16097 : op0 = XEXP (op0, 0);
12498 16097 : code = (code == GE ? LT : GE);
12499 16097 : continue;
12500 : }
12501 : break;
12502 :
12503 320788 : case NEG:
12504 : /* If testing for equality, we can take the NEG of the constant. */
12505 638118 : if (equality_comparison_p
12506 320788 : && (tem = simplify_unary_operation (NEG, mode, op1, mode)) != 0)
12507 : {
12508 317330 : op0 = XEXP (op0, 0);
12509 317330 : op1 = tem;
12510 317330 : continue;
12511 : }
12512 :
12513 : /* The remaining cases only apply to comparisons with zero. */
12514 3458 : if (const_op != 0)
12515 : break;
12516 :
12517 : /* When X is ABS or is known positive,
12518 : (neg X) is < 0 if and only if X != 0. */
12519 :
12520 2946 : if (sign_bit_comparison_p
12521 2908 : && (GET_CODE (XEXP (op0, 0)) == ABS
12522 2907 : || (mode_width <= HOST_BITS_PER_WIDE_INT
12523 2907 : && (nonzero_bits (XEXP (op0, 0), mode)
12524 2907 : & (HOST_WIDE_INT_1U << (mode_width - 1)))
12525 2907 : == 0)))
12526 : {
12527 38 : op0 = XEXP (op0, 0);
12528 38 : code = (code == LT ? NE : EQ);
12529 38 : continue;
12530 : }
12531 :
12532 : /* If we have NEG of something whose two high-order bits are the
12533 : same, we know that "(-a) < 0" is equivalent to "a > 0". */
12534 2870 : if (num_sign_bit_copies (op0, mode) >= 2)
12535 : {
12536 22 : op0 = XEXP (op0, 0);
12537 22 : code = swap_condition (code);
12538 22 : continue;
12539 : }
12540 : break;
12541 :
12542 146 : case ROTATE:
12543 : /* If we are testing equality and our count is a constant, we
12544 : can perform the inverse operation on our RHS. */
12545 146 : if (equality_comparison_p && CONST_INT_P (XEXP (op0, 1))
12546 146 : && (tem = simplify_binary_operation (ROTATERT, mode,
12547 : op1, XEXP (op0, 1))) != 0)
12548 : {
12549 0 : op0 = XEXP (op0, 0);
12550 0 : op1 = tem;
12551 0 : continue;
12552 : }
12553 :
12554 : /* If we are doing a < 0 or >= 0 comparison, it means we are testing
12555 : a particular bit. Convert it to an AND of a constant of that
12556 : bit. This will be converted into a ZERO_EXTRACT. */
12557 146 : if (const_op == 0 && sign_bit_comparison_p
12558 0 : && CONST_INT_P (XEXP (op0, 1))
12559 0 : && mode_width <= HOST_BITS_PER_WIDE_INT
12560 0 : && UINTVAL (XEXP (op0, 1)) < mode_width)
12561 : {
12562 0 : op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
12563 : (HOST_WIDE_INT_1U
12564 : << (mode_width - 1
12565 0 : - INTVAL (XEXP (op0, 1)))));
12566 0 : code = (code == LT ? NE : EQ);
12567 0 : continue;
12568 : }
12569 :
12570 : /* Fall through. */
12571 :
12572 2376 : case ABS:
12573 : /* ABS is ignorable inside an equality comparison with zero. */
12574 2376 : if (const_op == 0 && equality_comparison_p)
12575 : {
12576 1 : op0 = XEXP (op0, 0);
12577 1 : continue;
12578 : }
12579 : break;
12580 :
12581 1666 : case SIGN_EXTEND:
12582 : /* Can simplify (compare (zero/sign_extend FOO) CONST) to
12583 : (compare FOO CONST) if CONST fits in FOO's mode and we
12584 : are either testing inequality or have an unsigned
12585 : comparison with ZERO_EXTEND or a signed comparison with
12586 : SIGN_EXTEND. But don't do it if we don't have a compare
12587 : insn of the given mode, since we'd have to revert it
12588 : later on, and then we wouldn't know whether to sign- or
12589 : zero-extend. */
12590 1666 : if (is_int_mode (GET_MODE (XEXP (op0, 0)), &mode)
12591 1666 : && ! unsigned_comparison_p
12592 928 : && HWI_COMPUTABLE_MODE_P (mode)
12593 928 : && trunc_int_for_mode (const_op, mode) == const_op
12594 928 : && have_insn_for (COMPARE, mode))
12595 : {
12596 928 : op0 = XEXP (op0, 0);
12597 928 : continue;
12598 : }
12599 : break;
12600 :
12601 387624 : case SUBREG:
12602 : /* Check for the case where we are comparing A - C1 with C2, that is
12603 :
12604 : (subreg:MODE (plus (A) (-C1))) op (C2)
12605 :
12606 : with C1 a constant, and try to lift the SUBREG, i.e. to do the
12607 : comparison in the wider mode. One of the following two conditions
12608 : must be true in order for this to be valid:
12609 :
12610 : 1. The mode extension results in the same bit pattern being added
12611 : on both sides and the comparison is equality or unsigned. As
12612 : C2 has been truncated to fit in MODE, the pattern can only be
12613 : all 0s or all 1s.
12614 :
12615 : 2. The mode extension results in the sign bit being copied on
12616 : each side.
12617 :
12618 : The difficulty here is that we have predicates for A but not for
12619 : (A - C1) so we need to check that C1 is within proper bounds so
12620 : as to perturbate A as little as possible. */
12621 :
12622 387624 : if (mode_width <= HOST_BITS_PER_WIDE_INT
12623 387520 : && subreg_lowpart_p (op0)
12624 356588 : && is_a <scalar_int_mode> (GET_MODE (SUBREG_REG (op0)),
12625 : &inner_mode)
12626 355046 : && GET_MODE_PRECISION (inner_mode) > mode_width
12627 355046 : && GET_CODE (SUBREG_REG (op0)) == PLUS
12628 387624 : && CONST_INT_P (XEXP (SUBREG_REG (op0), 1)))
12629 : {
12630 0 : rtx a = XEXP (SUBREG_REG (op0), 0);
12631 0 : HOST_WIDE_INT c1 = -INTVAL (XEXP (SUBREG_REG (op0), 1));
12632 :
12633 0 : if ((c1 > 0
12634 0 : && (unsigned HOST_WIDE_INT) c1
12635 0 : < HOST_WIDE_INT_1U << (mode_width - 1)
12636 0 : && (equality_comparison_p || unsigned_comparison_p)
12637 : /* (A - C1) zero-extends if it is positive and sign-extends
12638 : if it is negative, C2 both zero- and sign-extends. */
12639 0 : && (((nonzero_bits (a, inner_mode)
12640 0 : & ~GET_MODE_MASK (mode)) == 0
12641 0 : && const_op >= 0)
12642 : /* (A - C1) sign-extends if it is positive and 1-extends
12643 : if it is negative, C2 both sign- and 1-extends. */
12644 0 : || (num_sign_bit_copies (a, inner_mode)
12645 0 : > (unsigned int) (GET_MODE_PRECISION (inner_mode)
12646 0 : - mode_width)
12647 0 : && const_op < 0)))
12648 0 : || ((unsigned HOST_WIDE_INT) c1
12649 0 : < HOST_WIDE_INT_1U << (mode_width - 2)
12650 : /* (A - C1) always sign-extends, like C2. */
12651 0 : && num_sign_bit_copies (a, inner_mode)
12652 0 : > (unsigned int) (GET_MODE_PRECISION (inner_mode)
12653 0 : - (mode_width - 1))))
12654 : {
12655 0 : op0 = SUBREG_REG (op0);
12656 0 : continue;
12657 : }
12658 : }
12659 :
12660 : /* If the inner mode is narrower and we are extracting the low part,
12661 : we can treat the SUBREG as if it were a ZERO_EXTEND ... */
12662 387624 : if (paradoxical_subreg_p (op0))
12663 : {
12664 : if (WORD_REGISTER_OPERATIONS
12665 : && is_a <scalar_int_mode> (GET_MODE (SUBREG_REG (op0)),
12666 : &inner_mode)
12667 : && GET_MODE_PRECISION (inner_mode) < BITS_PER_WORD
12668 : /* On WORD_REGISTER_OPERATIONS targets the bits
12669 : beyond sub_mode aren't considered undefined,
12670 : so optimize only if it is a MEM load when MEM loads
12671 : zero extend, because then the upper bits are all zero. */
12672 : && !(MEM_P (SUBREG_REG (op0))
12673 : && load_extend_op (inner_mode) == ZERO_EXTEND))
12674 : break;
12675 : /* FALLTHROUGH to case ZERO_EXTEND */
12676 : }
12677 387624 : else if (subreg_lowpart_p (op0)
12678 356692 : && GET_MODE_CLASS (mode) == MODE_INT
12679 356692 : && is_int_mode (GET_MODE (SUBREG_REG (op0)), &inner_mode)
12680 355046 : && (code == NE || code == EQ)
12681 268522 : && GET_MODE_PRECISION (inner_mode) <= HOST_BITS_PER_WIDE_INT
12682 262692 : && !paradoxical_subreg_p (op0)
12683 650316 : && (nonzero_bits (SUBREG_REG (op0), inner_mode)
12684 262692 : & ~GET_MODE_MASK (mode)) == 0)
12685 : {
12686 : /* Remove outer subregs that don't do anything. */
12687 104914 : tem = gen_lowpart (inner_mode, op1);
12688 :
12689 104914 : if ((nonzero_bits (tem, inner_mode)
12690 104914 : & ~GET_MODE_MASK (mode)) == 0)
12691 : {
12692 104295 : op0 = SUBREG_REG (op0);
12693 104295 : op1 = tem;
12694 104295 : continue;
12695 : }
12696 : break;
12697 : }
12698 : else
12699 : break;
12700 :
12701 : /* FALLTHROUGH */
12702 :
12703 40560 : case ZERO_EXTEND:
12704 40560 : if (is_int_mode (GET_MODE (XEXP (op0, 0)), &mode)
12705 40560 : && (unsigned_comparison_p || equality_comparison_p)
12706 40518 : && HWI_COMPUTABLE_MODE_P (mode)
12707 40518 : && (unsigned HOST_WIDE_INT) const_op <= GET_MODE_MASK (mode)
12708 40518 : && const_op >= 0
12709 40509 : && have_insn_for (COMPARE, mode))
12710 : {
12711 40509 : op0 = XEXP (op0, 0);
12712 40509 : continue;
12713 : }
12714 : break;
12715 :
12716 450276 : case PLUS:
12717 : /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
12718 : this for equality comparisons due to pathological cases involving
12719 : overflows. */
12720 497989 : if (equality_comparison_p
12721 450276 : && (tem = simplify_binary_operation (MINUS, mode,
12722 : op1, XEXP (op0, 1))) != 0)
12723 : {
12724 47713 : op0 = XEXP (op0, 0);
12725 47713 : op1 = tem;
12726 47713 : continue;
12727 : }
12728 :
12729 : /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
12730 402563 : if (const_op == 0 && XEXP (op0, 1) == constm1_rtx
12731 12547 : && GET_CODE (XEXP (op0, 0)) == ABS && sign_bit_comparison_p)
12732 : {
12733 0 : op0 = XEXP (XEXP (op0, 0), 0);
12734 0 : code = (code == LT ? EQ : NE);
12735 0 : continue;
12736 : }
12737 : break;
12738 :
12739 178844 : case MINUS:
12740 : /* We used to optimize signed comparisons against zero, but that
12741 : was incorrect. Unsigned comparisons against zero (GTU, LEU)
12742 : arrive here as equality comparisons, or (GEU, LTU) are
12743 : optimized away. No need to special-case them. */
12744 :
12745 : /* (eq (minus A B) C) -> (eq A (plus B C)) or
12746 : (eq B (minus A C)), whichever simplifies. We can only do
12747 : this for equality comparisons due to pathological cases involving
12748 : overflows. */
12749 212358 : if (equality_comparison_p
12750 178844 : && (tem = simplify_binary_operation (PLUS, mode,
12751 : XEXP (op0, 1), op1)) != 0)
12752 : {
12753 33514 : op0 = XEXP (op0, 0);
12754 33514 : op1 = tem;
12755 33514 : continue;
12756 : }
12757 :
12758 177536 : if (equality_comparison_p
12759 145330 : && (tem = simplify_binary_operation (MINUS, mode,
12760 : XEXP (op0, 0), op1)) != 0)
12761 : {
12762 32206 : op0 = XEXP (op0, 1);
12763 32206 : op1 = tem;
12764 32206 : continue;
12765 : }
12766 :
12767 : /* The sign bit of (minus (ashiftrt X C) X), where C is the number
12768 : of bits in X minus 1, is one iff X > 0. */
12769 17308 : if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == ASHIFTRT
12770 462 : && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
12771 462 : && UINTVAL (XEXP (XEXP (op0, 0), 1)) == mode_width - 1
12772 113148 : && rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
12773 : {
12774 0 : op0 = XEXP (op0, 1);
12775 0 : code = (code == GE ? LE : GT);
12776 0 : continue;
12777 : }
12778 : break;
12779 :
12780 8505 : case XOR:
12781 : /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
12782 : if C is zero or B is a constant. */
12783 8652 : if (equality_comparison_p
12784 8505 : && (tem = simplify_binary_operation (XOR, mode,
12785 : XEXP (op0, 1), op1)) != 0)
12786 : {
12787 147 : op0 = XEXP (op0, 0);
12788 147 : op1 = tem;
12789 147 : continue;
12790 : }
12791 : break;
12792 :
12793 :
12794 378651 : case IOR:
12795 : /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
12796 : iff X <= 0. */
12797 7217 : if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == PLUS
12798 1270 : && XEXP (XEXP (op0, 0), 1) == constm1_rtx
12799 378699 : && rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
12800 : {
12801 48 : op0 = XEXP (op0, 1);
12802 48 : code = (code == GE ? GT : LE);
12803 48 : continue;
12804 : }
12805 : break;
12806 :
12807 1721789 : case AND:
12808 : /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
12809 : will be converted to a ZERO_EXTRACT later. */
12810 1721789 : if (const_op == 0 && equality_comparison_p
12811 1606670 : && GET_CODE (XEXP (op0, 0)) == ASHIFT
12812 60349 : && XEXP (XEXP (op0, 0), 0) == const1_rtx)
12813 : {
12814 6808 : op0 = gen_rtx_LSHIFTRT (mode, XEXP (op0, 1),
12815 : XEXP (XEXP (op0, 0), 1));
12816 6808 : op0 = simplify_and_const_int (NULL_RTX, mode, op0, 1);
12817 6808 : continue;
12818 : }
12819 :
12820 : /* If we are comparing (and (lshiftrt X C1) C2) for equality with
12821 : zero and X is a comparison and C1 and C2 describe only bits set
12822 : in STORE_FLAG_VALUE, we can compare with X. */
12823 1714981 : if (const_op == 0 && equality_comparison_p
12824 1599862 : && mode_width <= HOST_BITS_PER_WIDE_INT
12825 1595969 : && CONST_INT_P (XEXP (op0, 1))
12826 1254872 : && GET_CODE (XEXP (op0, 0)) == LSHIFTRT
12827 526637 : && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
12828 513100 : && INTVAL (XEXP (XEXP (op0, 0), 1)) >= 0
12829 513100 : && INTVAL (XEXP (XEXP (op0, 0), 1)) < HOST_BITS_PER_WIDE_INT)
12830 : {
12831 513100 : mask = ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
12832 513100 : << INTVAL (XEXP (XEXP (op0, 0), 1)));
12833 513100 : if ((~STORE_FLAG_VALUE & mask) == 0
12834 513100 : && (COMPARISON_P (XEXP (XEXP (op0, 0), 0))
12835 0 : || ((tem = get_last_value (XEXP (XEXP (op0, 0), 0))) != 0
12836 0 : && COMPARISON_P (tem))))
12837 : {
12838 0 : op0 = XEXP (XEXP (op0, 0), 0);
12839 0 : continue;
12840 : }
12841 : }
12842 :
12843 : /* If we are doing an equality comparison of an AND of a bit equal
12844 : to the sign bit, replace this with a LT or GE comparison of
12845 : the underlying value. */
12846 1715506 : if (equality_comparison_p
12847 : && const_op == 0
12848 1599862 : && CONST_INT_P (XEXP (op0, 1))
12849 1255187 : && mode_width <= HOST_BITS_PER_WIDE_INT
12850 1714981 : && ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
12851 1254872 : == HOST_WIDE_INT_1U << (mode_width - 1)))
12852 : {
12853 525 : op0 = XEXP (op0, 0);
12854 525 : code = (code == EQ ? GE : LT);
12855 525 : continue;
12856 : }
12857 :
12858 : /* If this AND operation is really a ZERO_EXTEND from a narrower
12859 : mode, the constant fits within that mode, and this is either an
12860 : equality or unsigned comparison, try to do this comparison in
12861 : the narrower mode.
12862 :
12863 : Note that in:
12864 :
12865 : (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
12866 : -> (ne:DI (reg:SI 4) (const_int 0))
12867 :
12868 : unless TARGET_TRULY_NOOP_TRUNCATION allows it or the register is
12869 : known to hold a value of the required mode the
12870 : transformation is invalid. */
12871 1730486 : if ((equality_comparison_p || unsigned_comparison_p)
12872 1698913 : && CONST_INT_P (XEXP (op0, 1))
12873 3960592 : && (i = exact_log2 ((UINTVAL (XEXP (op0, 1))
12874 1349587 : & GET_MODE_MASK (mode))
12875 : + 1)) >= 0
12876 912579 : && const_op >> i == 0
12877 4309918 : && int_mode_for_size (i, 1).exists (&tmode))
12878 : {
12879 16030 : op0 = gen_lowpart_or_truncate (tmode, XEXP (op0, 0));
12880 16030 : continue;
12881 : }
12882 :
12883 : /* If this is (and:M1 (subreg:M1 X:M2 0) (const_int C1)) where C1
12884 : fits in both M1 and M2 and the SUBREG is either paradoxical
12885 : or represents the low part, permute the SUBREG and the AND
12886 : and try again. */
12887 1698426 : if (GET_CODE (XEXP (op0, 0)) == SUBREG
12888 110824 : && CONST_INT_P (XEXP (op0, 1)))
12889 : {
12890 105445 : unsigned HOST_WIDE_INT c1 = INTVAL (XEXP (op0, 1));
12891 : /* Require an integral mode, to avoid creating something like
12892 : (AND:SF ...). */
12893 148484 : if ((is_a <scalar_int_mode>
12894 105445 : (GET_MODE (SUBREG_REG (XEXP (op0, 0))), &tmode))
12895 : /* It is unsafe to commute the AND into the SUBREG if the
12896 : SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
12897 : not defined. As originally written the upper bits
12898 : have a defined value due to the AND operation.
12899 : However, if we commute the AND inside the SUBREG then
12900 : they no longer have defined values and the meaning of
12901 : the code has been changed.
12902 : Also C1 should not change value in the smaller mode,
12903 : see PR67028 (a positive C1 can become negative in the
12904 : smaller mode, so that the AND does no longer mask the
12905 : upper bits). */
12906 105412 : && ((WORD_REGISTER_OPERATIONS
12907 : && mode_width > GET_MODE_PRECISION (tmode)
12908 : && mode_width <= BITS_PER_WORD
12909 : && trunc_int_for_mode (c1, tmode) == (HOST_WIDE_INT) c1)
12910 105412 : || (mode_width <= GET_MODE_PRECISION (tmode)
12911 44733 : && subreg_lowpart_p (XEXP (op0, 0))))
12912 44707 : && mode_width <= HOST_BITS_PER_WIDE_INT
12913 44707 : && HWI_COMPUTABLE_MODE_P (tmode)
12914 44590 : && (c1 & ~mask) == 0
12915 43039 : && (c1 & ~GET_MODE_MASK (tmode)) == 0
12916 43039 : && c1 != mask
12917 43039 : && c1 != GET_MODE_MASK (tmode))
12918 : {
12919 43039 : op0 = simplify_gen_binary (AND, tmode,
12920 43039 : SUBREG_REG (XEXP (op0, 0)),
12921 43039 : gen_int_mode (c1, tmode));
12922 43039 : op0 = gen_lowpart (mode, op0);
12923 43039 : continue;
12924 : }
12925 : }
12926 :
12927 : /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
12928 1655387 : if (const_op == 0 && equality_comparison_p
12929 1548032 : && XEXP (op0, 1) == const1_rtx
12930 664751 : && GET_CODE (XEXP (op0, 0)) == NOT)
12931 : {
12932 4618 : op0 = simplify_and_const_int (NULL_RTX, mode,
12933 : XEXP (XEXP (op0, 0), 0), 1);
12934 4618 : code = (code == NE ? EQ : NE);
12935 4618 : continue;
12936 : }
12937 :
12938 : /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
12939 : (eq (and (lshiftrt X) 1) 0).
12940 : Also handle the case where (not X) is expressed using xor. */
12941 1650769 : if (const_op == 0 && equality_comparison_p
12942 1543414 : && XEXP (op0, 1) == const1_rtx
12943 660133 : && GET_CODE (XEXP (op0, 0)) == LSHIFTRT)
12944 : {
12945 511693 : rtx shift_op = XEXP (XEXP (op0, 0), 0);
12946 511693 : rtx shift_count = XEXP (XEXP (op0, 0), 1);
12947 :
12948 514391 : if (GET_CODE (shift_op) == NOT
12949 511693 : || (GET_CODE (shift_op) == XOR
12950 4555 : && CONST_INT_P (XEXP (shift_op, 1))
12951 2698 : && CONST_INT_P (shift_count)
12952 2698 : && HWI_COMPUTABLE_MODE_P (mode)
12953 2698 : && (UINTVAL (XEXP (shift_op, 1))
12954 : == HOST_WIDE_INT_1U
12955 2698 : << INTVAL (shift_count))))
12956 : {
12957 2698 : op0
12958 2698 : = gen_rtx_LSHIFTRT (mode, XEXP (shift_op, 0), shift_count);
12959 2698 : op0 = simplify_and_const_int (NULL_RTX, mode, op0, 1);
12960 2698 : code = (code == NE ? EQ : NE);
12961 2698 : continue;
12962 : }
12963 : }
12964 : break;
12965 :
12966 48947 : case ASHIFT:
12967 : /* If we have (compare (ashift FOO N) (const_int C)) and
12968 : the high order N bits of FOO (N+1 if an inequality comparison)
12969 : are known to be zero, we can do this by comparing FOO with C
12970 : shifted right N bits so long as the low-order N bits of C are
12971 : zero. */
12972 48947 : if (CONST_INT_P (XEXP (op0, 1))
12973 45397 : && INTVAL (XEXP (op0, 1)) >= 0
12974 45397 : && ((INTVAL (XEXP (op0, 1)) + ! equality_comparison_p)
12975 : < HOST_BITS_PER_WIDE_INT)
12976 45397 : && (((unsigned HOST_WIDE_INT) const_op
12977 45397 : & ((HOST_WIDE_INT_1U << INTVAL (XEXP (op0, 1)))
12978 : - 1)) == 0)
12979 34176 : && mode_width <= HOST_BITS_PER_WIDE_INT
12980 83093 : && (nonzero_bits (XEXP (op0, 0), mode)
12981 34146 : & ~(mask >> (INTVAL (XEXP (op0, 1))
12982 34146 : + ! equality_comparison_p))) == 0)
12983 : {
12984 : /* We must perform a logical shift, not an arithmetic one,
12985 : as we want the top N bits of C to be zero. */
12986 406 : unsigned HOST_WIDE_INT temp = const_op & GET_MODE_MASK (mode);
12987 :
12988 406 : temp >>= INTVAL (XEXP (op0, 1));
12989 406 : op1 = gen_int_mode (temp, mode);
12990 406 : op0 = XEXP (op0, 0);
12991 406 : continue;
12992 406 : }
12993 :
12994 : /* If we are doing a sign bit comparison, it means we are testing
12995 : a particular bit. Convert it to the appropriate AND. */
12996 48541 : if (sign_bit_comparison_p && CONST_INT_P (XEXP (op0, 1))
12997 1611 : && mode_width <= HOST_BITS_PER_WIDE_INT)
12998 : {
12999 3222 : op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
13000 : (HOST_WIDE_INT_1U
13001 : << (mode_width - 1
13002 1611 : - INTVAL (XEXP (op0, 1)))));
13003 1611 : code = (code == LT ? NE : EQ);
13004 1611 : continue;
13005 : }
13006 :
13007 : /* If this an equality comparison with zero and we are shifting
13008 : the low bit to the sign bit, we can convert this to an AND of the
13009 : low-order bit. */
13010 46930 : if (const_op == 0 && equality_comparison_p
13011 10121 : && CONST_INT_P (XEXP (op0, 1))
13012 7738 : && UINTVAL (XEXP (op0, 1)) == mode_width - 1)
13013 : {
13014 105 : op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0), 1);
13015 105 : continue;
13016 : }
13017 : break;
13018 :
13019 45462 : case ASHIFTRT:
13020 : /* If this is an equality comparison with zero, we can do this
13021 : as a logical shift, which might be much simpler. */
13022 45462 : if (equality_comparison_p && const_op == 0
13023 24611 : && CONST_INT_P (XEXP (op0, 1)))
13024 : {
13025 47806 : op0 = simplify_shift_const (NULL_RTX, LSHIFTRT, mode,
13026 : XEXP (op0, 0),
13027 23903 : INTVAL (XEXP (op0, 1)));
13028 23903 : continue;
13029 : }
13030 :
13031 : /* If OP0 is a sign extension and CODE is not an unsigned comparison,
13032 : do the comparison in a narrower mode. */
13033 26663 : if (! unsigned_comparison_p
13034 19022 : && CONST_INT_P (XEXP (op0, 1))
13035 18278 : && GET_CODE (XEXP (op0, 0)) == ASHIFT
13036 5810 : && XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
13037 5592 : && (int_mode_for_size (mode_width - INTVAL (XEXP (op0, 1)), 1)
13038 21559 : .exists (&tmode))
13039 21559 : && (((unsigned HOST_WIDE_INT) const_op
13040 5104 : + (GET_MODE_MASK (tmode) >> 1) + 1)
13041 5104 : <= GET_MODE_MASK (tmode)))
13042 : {
13043 5104 : op0 = gen_lowpart (tmode, XEXP (XEXP (op0, 0), 0));
13044 5104 : continue;
13045 : }
13046 :
13047 : /* Likewise if OP0 is a PLUS of a sign extension with a
13048 : constant, which is usually represented with the PLUS
13049 : between the shifts. */
13050 16455 : if (! unsigned_comparison_p
13051 13918 : && CONST_INT_P (XEXP (op0, 1))
13052 13174 : && GET_CODE (XEXP (op0, 0)) == PLUS
13053 54 : && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
13054 22 : && GET_CODE (XEXP (XEXP (op0, 0), 0)) == ASHIFT
13055 2 : && XEXP (op0, 1) == XEXP (XEXP (XEXP (op0, 0), 0), 1)
13056 0 : && (int_mode_for_size (mode_width - INTVAL (XEXP (op0, 1)), 1)
13057 16455 : .exists (&tmode))
13058 16455 : && (((unsigned HOST_WIDE_INT) const_op
13059 0 : + (GET_MODE_MASK (tmode) >> 1) + 1)
13060 0 : <= GET_MODE_MASK (tmode)))
13061 : {
13062 0 : rtx inner = XEXP (XEXP (XEXP (op0, 0), 0), 0);
13063 0 : rtx add_const = XEXP (XEXP (op0, 0), 1);
13064 0 : rtx new_const = simplify_gen_binary (ASHIFTRT, mode,
13065 : add_const, XEXP (op0, 1));
13066 :
13067 0 : op0 = simplify_gen_binary (PLUS, tmode,
13068 0 : gen_lowpart (tmode, inner),
13069 : new_const);
13070 0 : continue;
13071 0 : }
13072 :
13073 : /* FALLTHROUGH */
13074 132424 : case LSHIFTRT:
13075 : /* If we have (compare (xshiftrt FOO N) (const_int C)) and
13076 : the low order N bits of FOO are known to be zero, we can do this
13077 : by comparing FOO with C shifted left N bits so long as no
13078 : overflow occurs. Even if the low order N bits of FOO aren't known
13079 : to be zero, if the comparison is >= or < we can use the same
13080 : optimization and for > or <= by setting all the low
13081 : order N bits in the comparison constant. */
13082 132424 : if (CONST_INT_P (XEXP (op0, 1))
13083 127816 : && INTVAL (XEXP (op0, 1)) > 0
13084 127816 : && INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
13085 127456 : && mode_width <= HOST_BITS_PER_WIDE_INT
13086 132424 : && (((unsigned HOST_WIDE_INT) const_op
13087 253390 : + (GET_CODE (op0) != LSHIFTRT
13088 126695 : ? ((GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1)) >> 1)
13089 : + 1)
13090 : : 0))
13091 126695 : <= GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1))))
13092 : {
13093 126545 : unsigned HOST_WIDE_INT low_bits
13094 126545 : = (nonzero_bits (XEXP (op0, 0), mode)
13095 126545 : & ((HOST_WIDE_INT_1U
13096 126545 : << INTVAL (XEXP (op0, 1))) - 1));
13097 126545 : if (low_bits == 0 || !equality_comparison_p)
13098 : {
13099 : /* If the shift was logical, then we must make the condition
13100 : unsigned. */
13101 21887 : if (GET_CODE (op0) == LSHIFTRT)
13102 17815 : code = unsigned_condition (code);
13103 :
13104 21887 : const_op = (unsigned HOST_WIDE_INT) const_op
13105 21887 : << INTVAL (XEXP (op0, 1));
13106 21887 : if (low_bits != 0
13107 3849 : && (code == GT || code == GTU
13108 928 : || code == LE || code == LEU))
13109 3781 : const_op
13110 3781 : |= ((HOST_WIDE_INT_1 << INTVAL (XEXP (op0, 1))) - 1);
13111 21887 : op1 = GEN_INT (const_op);
13112 21887 : op0 = XEXP (op0, 0);
13113 21887 : continue;
13114 : }
13115 : }
13116 :
13117 : /* If we are using this shift to extract just the sign bit, we
13118 : can replace this with an LT or GE comparison. */
13119 110537 : if (const_op == 0
13120 94026 : && (equality_comparison_p || sign_bit_comparison_p)
13121 93990 : && CONST_INT_P (XEXP (op0, 1))
13122 89597 : && UINTVAL (XEXP (op0, 1)) == mode_width - 1)
13123 : {
13124 46553 : op0 = XEXP (op0, 0);
13125 46553 : code = (code == NE || code == GT ? LT : GE);
13126 46553 : continue;
13127 : }
13128 : break;
13129 :
13130 : default:
13131 : break;
13132 : }
13133 :
13134 : break;
13135 : }
13136 :
13137 : /* Now make any compound operations involved in this comparison. Then,
13138 : check for an outermost SUBREG on OP0 that is not doing anything or is
13139 : paradoxical. The latter transformation must only be performed when
13140 : it is known that the "extra" bits will be the same in op0 and op1 or
13141 : that they don't matter. There are three cases to consider:
13142 :
13143 : 1. SUBREG_REG (op0) is a register. In this case the bits are don't
13144 : care bits and we can assume they have any convenient value. So
13145 : making the transformation is safe.
13146 :
13147 : 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is UNKNOWN.
13148 : In this case the upper bits of op0 are undefined. We should not make
13149 : the simplification in that case as we do not know the contents of
13150 : those bits.
13151 :
13152 : 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not UNKNOWN.
13153 : In that case we know those bits are zeros or ones. We must also be
13154 : sure that they are the same as the upper bits of op1.
13155 :
13156 : We can never remove a SUBREG for a non-equality comparison because
13157 : the sign bit is in a different place in the underlying object. */
13158 :
13159 23467258 : rtx_code op0_mco_code = SET;
13160 23467258 : if (op1 == const0_rtx)
13161 10877302 : op0_mco_code = code == NE || code == EQ ? EQ : COMPARE;
13162 :
13163 23467258 : op0 = make_compound_operation (op0, op0_mco_code);
13164 23467258 : op1 = make_compound_operation (op1, SET);
13165 :
13166 435749 : if (GET_CODE (op0) == SUBREG && subreg_lowpart_p (op0)
13167 403400 : && is_int_mode (GET_MODE (op0), &mode)
13168 378421 : && is_int_mode (GET_MODE (SUBREG_REG (op0)), &inner_mode)
13169 23842469 : && (code == NE || code == EQ))
13170 : {
13171 211406 : if (paradoxical_subreg_p (op0))
13172 : {
13173 : /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
13174 : implemented. */
13175 0 : if (REG_P (SUBREG_REG (op0)))
13176 : {
13177 0 : op0 = SUBREG_REG (op0);
13178 0 : op1 = gen_lowpart (inner_mode, op1);
13179 : }
13180 : }
13181 211406 : else if (GET_MODE_PRECISION (inner_mode) <= HOST_BITS_PER_WIDE_INT
13182 211406 : && (nonzero_bits (SUBREG_REG (op0), inner_mode)
13183 204517 : & ~GET_MODE_MASK (mode)) == 0)
13184 : {
13185 32337 : tem = gen_lowpart (inner_mode, op1);
13186 :
13187 32337 : if ((nonzero_bits (tem, inner_mode) & ~GET_MODE_MASK (mode)) == 0)
13188 23737 : op0 = SUBREG_REG (op0), op1 = tem;
13189 : }
13190 : }
13191 :
13192 : /* We now do the opposite procedure: Some machines don't have compare
13193 : insns in all modes. If OP0's mode is an integer mode smaller than a
13194 : word and we can't do a compare in that mode, see if there is a larger
13195 : mode for which we can do the compare. There are a number of cases in
13196 : which we can use the wider mode. */
13197 :
13198 23467258 : if (is_int_mode (GET_MODE (op0), &mode)
13199 24282694 : && GET_MODE_SIZE (mode) < UNITS_PER_WORD
13200 8544695 : && ! have_insn_for (COMPARE, mode))
13201 0 : FOR_EACH_WIDER_MODE (tmode_iter, mode)
13202 : {
13203 0 : tmode = tmode_iter.require ();
13204 0 : if (!HWI_COMPUTABLE_MODE_P (tmode))
13205 : break;
13206 0 : if (have_insn_for (COMPARE, tmode))
13207 : {
13208 0 : int zero_extended;
13209 :
13210 : /* If this is a test for negative, we can make an explicit
13211 : test of the sign bit. Test this first so we can use
13212 : a paradoxical subreg to extend OP0. */
13213 :
13214 0 : if (op1 == const0_rtx && (code == LT || code == GE)
13215 0 : && HWI_COMPUTABLE_MODE_P (mode))
13216 : {
13217 0 : unsigned HOST_WIDE_INT sign
13218 0 : = HOST_WIDE_INT_1U << (GET_MODE_BITSIZE (mode) - 1);
13219 0 : op0 = simplify_gen_binary (AND, tmode,
13220 0 : gen_lowpart (tmode, op0),
13221 0 : gen_int_mode (sign, tmode));
13222 0 : code = (code == LT) ? NE : EQ;
13223 : break;
13224 : }
13225 :
13226 : /* If the only nonzero bits in OP0 and OP1 are those in the
13227 : narrower mode and this is an equality or unsigned comparison,
13228 : we can use the wider mode. Similarly for sign-extended
13229 : values, in which case it is true for all comparisons. */
13230 0 : zero_extended = ((code == EQ || code == NE
13231 0 : || code == GEU || code == GTU
13232 0 : || code == LEU || code == LTU)
13233 0 : && (nonzero_bits (op0, tmode)
13234 0 : & ~GET_MODE_MASK (mode)) == 0
13235 0 : && ((CONST_INT_P (op1)
13236 0 : || (nonzero_bits (op1, tmode)
13237 0 : & ~GET_MODE_MASK (mode)) == 0)));
13238 :
13239 0 : if (zero_extended
13240 0 : || ((num_sign_bit_copies (op0, tmode)
13241 0 : > (unsigned int) (GET_MODE_PRECISION (tmode)
13242 0 : - GET_MODE_PRECISION (mode)))
13243 0 : && (num_sign_bit_copies (op1, tmode)
13244 0 : > (unsigned int) (GET_MODE_PRECISION (tmode)
13245 0 : - GET_MODE_PRECISION (mode)))))
13246 : {
13247 : /* If OP0 is an AND and we don't have an AND in MODE either,
13248 : make a new AND in the proper mode. */
13249 0 : if (GET_CODE (op0) == AND
13250 0 : && !have_insn_for (AND, mode))
13251 0 : op0 = simplify_gen_binary (AND, tmode,
13252 0 : gen_lowpart (tmode,
13253 : XEXP (op0, 0)),
13254 0 : gen_lowpart (tmode,
13255 : XEXP (op0, 1)));
13256 : else
13257 : {
13258 0 : if (zero_extended)
13259 : {
13260 0 : op0 = simplify_gen_unary (ZERO_EXTEND, tmode,
13261 : op0, mode);
13262 0 : op1 = simplify_gen_unary (ZERO_EXTEND, tmode,
13263 : op1, mode);
13264 : }
13265 : else
13266 : {
13267 0 : op0 = simplify_gen_unary (SIGN_EXTEND, tmode,
13268 : op0, mode);
13269 0 : op1 = simplify_gen_unary (SIGN_EXTEND, tmode,
13270 : op1, mode);
13271 : }
13272 : break;
13273 : }
13274 : }
13275 : }
13276 : }
13277 :
13278 : /* We may have changed the comparison operands. Re-canonicalize. */
13279 23467258 : if (swap_commutative_operands_p (op0, op1))
13280 : {
13281 85342 : std::swap (op0, op1);
13282 85342 : code = swap_condition (code);
13283 : }
13284 :
13285 : /* If this machine only supports a subset of valid comparisons, see if we
13286 : can convert an unsupported one into a supported one. */
13287 23467258 : target_canonicalize_comparison (&code, &op0, &op1, 0);
13288 :
13289 23467258 : *pop0 = op0;
13290 23467258 : *pop1 = op1;
13291 :
13292 23467258 : return code;
13293 : }
13294 :
13295 : /* Utility function for record_value_for_reg. Count number of
13296 : rtxs in X. */
13297 : static int
13298 1938 : count_rtxs (rtx x)
13299 : {
13300 1938 : enum rtx_code code = GET_CODE (x);
13301 1938 : const char *fmt;
13302 1938 : int i, j, ret = 1;
13303 :
13304 1938 : if (GET_RTX_CLASS (code) == RTX_BIN_ARITH
13305 1938 : || GET_RTX_CLASS (code) == RTX_COMM_ARITH)
13306 : {
13307 78 : rtx x0 = XEXP (x, 0);
13308 78 : rtx x1 = XEXP (x, 1);
13309 :
13310 78 : if (x0 == x1)
13311 0 : return 1 + 2 * count_rtxs (x0);
13312 :
13313 78 : if ((GET_RTX_CLASS (GET_CODE (x1)) == RTX_BIN_ARITH
13314 78 : || GET_RTX_CLASS (GET_CODE (x1)) == RTX_COMM_ARITH)
13315 4 : && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
13316 0 : return 2 + 2 * count_rtxs (x0)
13317 0 : + count_rtxs (x == XEXP (x1, 0)
13318 0 : ? XEXP (x1, 1) : XEXP (x1, 0));
13319 :
13320 78 : if ((GET_RTX_CLASS (GET_CODE (x0)) == RTX_BIN_ARITH
13321 78 : || GET_RTX_CLASS (GET_CODE (x0)) == RTX_COMM_ARITH)
13322 8 : && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
13323 0 : return 2 + 2 * count_rtxs (x1)
13324 0 : + count_rtxs (x == XEXP (x0, 0)
13325 0 : ? XEXP (x0, 1) : XEXP (x0, 0));
13326 : }
13327 :
13328 1938 : fmt = GET_RTX_FORMAT (code);
13329 4666 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
13330 2728 : if (fmt[i] == 'e')
13331 1109 : ret += count_rtxs (XEXP (x, i));
13332 1619 : else if (fmt[i] == 'E')
13333 280 : for (j = 0; j < XVECLEN (x, i); j++)
13334 220 : ret += count_rtxs (XVECEXP (x, i, j));
13335 :
13336 : return ret;
13337 : }
13338 :
13339 : /* Utility function for following routine. Called when X is part of a value
13340 : being stored into last_set_value. Sets last_set_table_tick
13341 : for each register mentioned. Similar to mention_regs in cse.cc */
13342 :
13343 : static void
13344 242398346 : update_table_tick (rtx x)
13345 : {
13346 243033979 : enum rtx_code code = GET_CODE (x);
13347 243033979 : const char *fmt = GET_RTX_FORMAT (code);
13348 243033979 : int i, j;
13349 :
13350 243033979 : if (code == REG)
13351 : {
13352 81715947 : unsigned int regno = REGNO (x);
13353 81715947 : unsigned int endregno = END_REGNO (x);
13354 81715947 : unsigned int r;
13355 :
13356 163544732 : for (r = regno; r < endregno; r++)
13357 : {
13358 81828785 : reg_stat_type *rsp = ®_stat[r];
13359 81828785 : rsp->last_set_table_tick = label_tick;
13360 : }
13361 :
13362 : return;
13363 : }
13364 :
13365 415970657 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
13366 255410062 : if (fmt[i] == 'e')
13367 : {
13368 : /* Check for identical subexpressions. If x contains
13369 : identical subexpression we only have to traverse one of
13370 : them. */
13371 150438012 : if (i == 0 && ARITHMETIC_P (x))
13372 : {
13373 : /* Note that at this point x1 has already been
13374 : processed. */
13375 58628578 : rtx x0 = XEXP (x, 0);
13376 58628578 : rtx x1 = XEXP (x, 1);
13377 :
13378 : /* If x0 and x1 are identical then there is no need to
13379 : process x0. */
13380 58628578 : if (x0 == x1)
13381 : break;
13382 :
13383 : /* If x0 is identical to a subexpression of x1 then while
13384 : processing x1, x0 has already been processed. Thus we
13385 : are done with x. */
13386 58506919 : if (ARITHMETIC_P (x1)
13387 407277 : && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
13388 : break;
13389 :
13390 : /* If x1 is identical to a subexpression of x0 then we
13391 : still have to process the rest of x0. */
13392 58506774 : if (ARITHMETIC_P (x0)
13393 16090992 : && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
13394 : {
13395 635633 : update_table_tick (XEXP (x0, x1 == XEXP (x0, 0) ? 1 : 0));
13396 635633 : break;
13397 : }
13398 : }
13399 :
13400 149680575 : update_table_tick (XEXP (x, i));
13401 : }
13402 104972050 : else if (fmt[i] == 'E')
13403 9657984 : for (j = 0; j < XVECLEN (x, i); j++)
13404 7083837 : update_table_tick (XVECEXP (x, i, j));
13405 : }
13406 :
13407 : /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
13408 : are saying that the register is clobbered and we no longer know its
13409 : value. If INSN is zero, don't update reg_stat[].last_set; this is
13410 : only permitted with VALUE also zero and is used to invalidate the
13411 : register. */
13412 :
13413 : static void
13414 113064053 : record_value_for_reg (rtx reg, rtx_insn *insn, rtx value)
13415 : {
13416 113064053 : unsigned int regno = REGNO (reg);
13417 113064053 : unsigned int endregno = END_REGNO (reg);
13418 113064053 : unsigned int i;
13419 113064053 : reg_stat_type *rsp;
13420 :
13421 : /* If VALUE contains REG and we have a previous value for REG, substitute
13422 : the previous value. */
13423 113064053 : if (value && insn && reg_overlap_mentioned_p (reg, value))
13424 : {
13425 6144506 : rtx tem;
13426 :
13427 : /* Set things up so get_last_value is allowed to see anything set up to
13428 : our insn. */
13429 6144506 : subst_low_luid = DF_INSN_LUID (insn);
13430 6144506 : tem = get_last_value (reg);
13431 :
13432 : /* If TEM is simply a binary operation with two CLOBBERs as operands,
13433 : it isn't going to be useful and will take a lot of time to process,
13434 : so just use the CLOBBER. */
13435 :
13436 6144506 : if (tem)
13437 : {
13438 2459220 : if (ARITHMETIC_P (tem)
13439 2236642 : && GET_CODE (XEXP (tem, 0)) == CLOBBER
13440 1062925 : && GET_CODE (XEXP (tem, 1)) == CLOBBER)
13441 : tem = XEXP (tem, 0);
13442 2458034 : else if (count_occurrences (value, reg, 1) >= 2)
13443 : {
13444 : /* If there are two or more occurrences of REG in VALUE,
13445 : prevent the value from growing too much. */
13446 609 : if (count_rtxs (tem) > param_max_last_value_rtl)
13447 0 : tem = gen_rtx_CLOBBER (GET_MODE (tem), const0_rtx);
13448 : }
13449 :
13450 2459220 : value = replace_rtx (copy_rtx (value), reg, tem);
13451 : }
13452 : }
13453 :
13454 : /* For each register modified, show we don't know its value, that
13455 : we don't know about its bitwise content, that its value has been
13456 : updated, and that we don't know the location of the death of the
13457 : register. */
13458 226477957 : for (i = regno; i < endregno; i++)
13459 : {
13460 113413904 : rsp = ®_stat[i];
13461 :
13462 113413904 : if (insn)
13463 103300604 : rsp->last_set = insn;
13464 :
13465 113413904 : rsp->last_set_value = 0;
13466 113413904 : rsp->last_set_mode = VOIDmode;
13467 113413904 : rsp->last_set_nonzero_bits = 0;
13468 113413904 : rsp->last_set_sign_bit_copies = 0;
13469 113413904 : rsp->last_death = 0;
13470 113413904 : rsp->truncated_to_mode = VOIDmode;
13471 : }
13472 :
13473 : /* Mark registers that are being referenced in this value. */
13474 113064053 : if (value)
13475 85633934 : update_table_tick (value);
13476 :
13477 : /* Now update the status of each register being set.
13478 : If someone is using this register in this block, set this register
13479 : to invalid since we will get confused between the two lives in this
13480 : basic block. This makes using this register always invalid. In cse, we
13481 : scan the table to invalidate all entries using this register, but this
13482 : is too much work for us. */
13483 :
13484 226477957 : for (i = regno; i < endregno; i++)
13485 : {
13486 113413904 : rsp = ®_stat[i];
13487 113413904 : rsp->last_set_label = label_tick;
13488 113413904 : if (!insn
13489 103300604 : || (value && rsp->last_set_table_tick >= label_tick_ebb_start))
13490 20408701 : rsp->last_set_invalid = true;
13491 : else
13492 93005203 : rsp->last_set_invalid = false;
13493 : }
13494 :
13495 : /* The value being assigned might refer to X (like in "x++;"). In that
13496 : case, we must replace it with (clobber (const_int 0)) to prevent
13497 : infinite loops. */
13498 113064053 : rsp = ®_stat[regno];
13499 113064053 : if (value && !get_last_value_validate (&value, insn, label_tick, false))
13500 : {
13501 11083790 : value = copy_rtx (value);
13502 11083790 : if (!get_last_value_validate (&value, insn, label_tick, true))
13503 0 : value = 0;
13504 : }
13505 :
13506 : /* For the main register being modified, update the value, the mode, the
13507 : nonzero bits, and the number of sign bit copies. */
13508 :
13509 113064053 : rsp->last_set_value = value;
13510 :
13511 113064053 : if (value)
13512 : {
13513 85633934 : machine_mode mode = GET_MODE (reg);
13514 85633934 : subst_low_luid = DF_INSN_LUID (insn);
13515 85633934 : rsp->last_set_mode = mode;
13516 85633934 : if (GET_MODE_CLASS (mode) == MODE_INT
13517 85633934 : && HWI_COMPUTABLE_MODE_P (mode))
13518 64330364 : mode = nonzero_bits_mode;
13519 85633934 : rsp->last_set_nonzero_bits = nonzero_bits (value, mode);
13520 85633934 : rsp->last_set_sign_bit_copies
13521 85633934 : = num_sign_bit_copies (value, GET_MODE (reg));
13522 : }
13523 113064053 : }
13524 :
13525 : /* Called via note_stores from record_dead_and_set_regs to handle one
13526 : SET or CLOBBER in an insn. DATA is the instruction in which the
13527 : set is occurring. */
13528 :
13529 : static void
13530 135203106 : record_dead_and_set_regs_1 (rtx dest, const_rtx setter, void *data)
13531 : {
13532 135203106 : rtx_insn *record_dead_insn = (rtx_insn *) data;
13533 :
13534 135203106 : if (GET_CODE (dest) == SUBREG)
13535 5 : dest = SUBREG_REG (dest);
13536 :
13537 135203106 : if (!record_dead_insn)
13538 : {
13539 4998899 : if (REG_P (dest))
13540 4998899 : record_value_for_reg (dest, NULL, NULL_RTX);
13541 4998899 : return;
13542 : }
13543 :
13544 130204207 : if (REG_P (dest))
13545 : {
13546 : /* If we are setting the whole register, we know its value. */
13547 103130514 : if (GET_CODE (setter) == SET && dest == SET_DEST (setter))
13548 85482909 : record_value_for_reg (dest, record_dead_insn, SET_SRC (setter));
13549 : /* We can handle a SUBREG if it's the low part, but we must be
13550 : careful with paradoxical SUBREGs on RISC architectures because
13551 : we cannot strip e.g. an extension around a load and record the
13552 : naked load since the RTL middle-end considers that the upper bits
13553 : are defined according to LOAD_EXTEND_OP. */
13554 17647605 : else if (GET_CODE (setter) == SET
13555 582201 : && GET_CODE (SET_DEST (setter)) == SUBREG
13556 568796 : && SUBREG_REG (SET_DEST (setter)) == dest
13557 920371 : && known_le (GET_MODE_PRECISION (GET_MODE (dest)),
13558 : BITS_PER_WORD)
13559 17752645 : && subreg_lowpart_p (SET_DEST (setter)))
13560 : {
13561 105040 : if (WORD_REGISTER_OPERATIONS
13562 : && word_register_operation_p (SET_SRC (setter))
13563 : && paradoxical_subreg_p (SET_DEST (setter)))
13564 : record_value_for_reg (dest, record_dead_insn, SET_SRC (setter));
13565 105040 : else if (!partial_subreg_p (SET_DEST (setter)))
13566 93042 : record_value_for_reg (dest, record_dead_insn,
13567 93042 : gen_lowpart (GET_MODE (dest),
13568 93042 : SET_SRC (setter)));
13569 : else
13570 : {
13571 11998 : record_value_for_reg (dest, record_dead_insn,
13572 11998 : gen_lowpart (GET_MODE (dest),
13573 11998 : SET_SRC (setter)));
13574 :
13575 11998 : unsigned HOST_WIDE_INT mask;
13576 11998 : reg_stat_type *rsp = ®_stat[REGNO (dest)];
13577 11998 : mask = GET_MODE_MASK (GET_MODE (SET_DEST (setter)));
13578 11998 : rsp->last_set_nonzero_bits |= ~mask;
13579 11998 : rsp->last_set_sign_bit_copies = 1;
13580 : }
13581 : }
13582 : /* Otherwise show that we don't know the value. */
13583 : else
13584 17542565 : record_value_for_reg (dest, record_dead_insn, NULL_RTX);
13585 : }
13586 27073693 : else if (MEM_P (dest)
13587 : /* Ignore pushes, they clobber nothing. */
13588 27073693 : && ! push_operand (dest, GET_MODE (dest)))
13589 13874209 : mem_last_set = DF_INSN_LUID (record_dead_insn);
13590 : }
13591 :
13592 : /* Update the records of when each REG was most recently set or killed
13593 : for the things done by INSN. This is the last thing done in processing
13594 : INSN in the combiner loop.
13595 :
13596 : We update reg_stat[], in particular fields last_set, last_set_value,
13597 : last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
13598 : last_death, and also the similar information mem_last_set (which insn
13599 : most recently modified memory) and last_call_luid (which insn was the
13600 : most recent subroutine call). */
13601 :
13602 : static void
13603 172060997 : record_dead_and_set_regs (rtx_insn *insn)
13604 : {
13605 172060997 : rtx link;
13606 172060997 : unsigned int i;
13607 :
13608 307380547 : for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
13609 : {
13610 135319550 : if (REG_NOTE_KIND (link) == REG_DEAD
13611 77363335 : && REG_P (XEXP (link, 0)))
13612 : {
13613 77363335 : unsigned int regno = REGNO (XEXP (link, 0));
13614 77363335 : unsigned int endregno = END_REGNO (XEXP (link, 0));
13615 :
13616 154922943 : for (i = regno; i < endregno; i++)
13617 : {
13618 77559608 : reg_stat_type *rsp;
13619 :
13620 77559608 : rsp = ®_stat[i];
13621 77559608 : rsp->last_death = insn;
13622 : }
13623 : }
13624 57956215 : else if (REG_NOTE_KIND (link) == REG_INC)
13625 0 : record_value_for_reg (XEXP (link, 0), insn, NULL_RTX);
13626 : }
13627 :
13628 172060997 : if (CALL_P (insn))
13629 : {
13630 9256681 : HARD_REG_SET callee_clobbers
13631 9256681 : = insn_callee_abi (insn).full_and_partial_reg_clobbers ();
13632 9256681 : hard_reg_set_iterator hrsi;
13633 764204654 : EXECUTE_IF_SET_IN_HARD_REG_SET (callee_clobbers, 0, i, hrsi)
13634 : {
13635 754947973 : reg_stat_type *rsp;
13636 :
13637 : /* ??? We could try to preserve some information from the last
13638 : set of register I if the call doesn't actually clobber
13639 : (reg:last_set_mode I), which might be true for ABIs with
13640 : partial clobbers. However, it would be difficult to
13641 : update last_set_nonzero_bits and last_sign_bit_copies
13642 : to account for the part of I that actually was clobbered.
13643 : It wouldn't help much anyway, since we rarely see this
13644 : situation before RA. */
13645 754947973 : rsp = ®_stat[i];
13646 754947973 : rsp->last_set_invalid = true;
13647 754947973 : rsp->last_set = insn;
13648 754947973 : rsp->last_set_value = 0;
13649 754947973 : rsp->last_set_mode = VOIDmode;
13650 754947973 : rsp->last_set_nonzero_bits = 0;
13651 754947973 : rsp->last_set_sign_bit_copies = 0;
13652 754947973 : rsp->last_death = 0;
13653 754947973 : rsp->truncated_to_mode = VOIDmode;
13654 : }
13655 :
13656 9256681 : last_call_luid = mem_last_set = DF_INSN_LUID (insn);
13657 :
13658 : /* We can't combine into a call pattern. Remember, though, that
13659 : the return value register is set at this LUID. We could
13660 : still replace a register with the return value from the
13661 : wrong subroutine call! */
13662 9256681 : note_stores (insn, record_dead_and_set_regs_1, NULL_RTX);
13663 : }
13664 : else
13665 162804316 : note_stores (insn, record_dead_and_set_regs_1, insn);
13666 172060997 : }
13667 :
13668 : /* If a SUBREG has the promoted bit set, it is in fact a property of the
13669 : register present in the SUBREG, so for each such SUBREG go back and
13670 : adjust nonzero and sign bit information of the registers that are
13671 : known to have some zero/sign bits set.
13672 :
13673 : This is needed because when combine blows the SUBREGs away, the
13674 : information on zero/sign bits is lost and further combines can be
13675 : missed because of that. */
13676 :
13677 : static void
13678 5833 : record_promoted_value (rtx_insn *insn, rtx subreg)
13679 : {
13680 5833 : struct insn_link *links;
13681 5833 : rtx set;
13682 5833 : unsigned int regno = REGNO (SUBREG_REG (subreg));
13683 5833 : machine_mode mode = GET_MODE (subreg);
13684 :
13685 5833 : if (!HWI_COMPUTABLE_MODE_P (mode))
13686 : return;
13687 :
13688 6443 : for (links = LOG_LINKS (insn); links;)
13689 : {
13690 5783 : reg_stat_type *rsp;
13691 :
13692 5783 : insn = links->insn;
13693 5783 : set = single_set (insn);
13694 :
13695 5783 : if (! set || !REG_P (SET_DEST (set))
13696 5783 : || REGNO (SET_DEST (set)) != regno
13697 11092 : || GET_MODE (SET_DEST (set)) != GET_MODE (SUBREG_REG (subreg)))
13698 : {
13699 474 : links = links->next;
13700 474 : continue;
13701 : }
13702 :
13703 5309 : rsp = ®_stat[regno];
13704 5309 : if (rsp->last_set == insn)
13705 : {
13706 5309 : if (SUBREG_PROMOTED_UNSIGNED_P (subreg))
13707 5309 : rsp->last_set_nonzero_bits &= GET_MODE_MASK (mode);
13708 : }
13709 :
13710 5309 : if (REG_P (SET_SRC (set)))
13711 : {
13712 136 : regno = REGNO (SET_SRC (set));
13713 136 : links = LOG_LINKS (insn);
13714 : }
13715 : else
13716 : break;
13717 : }
13718 : }
13719 :
13720 : /* Check if X, a register, is known to contain a value already
13721 : truncated to MODE. In this case we can use a subreg to refer to
13722 : the truncated value even though in the generic case we would need
13723 : an explicit truncation. */
13724 :
13725 : static bool
13726 0 : reg_truncated_to_mode (machine_mode mode, const_rtx x)
13727 : {
13728 0 : reg_stat_type *rsp = ®_stat[REGNO (x)];
13729 0 : machine_mode truncated = rsp->truncated_to_mode;
13730 :
13731 0 : if (truncated == 0
13732 0 : || rsp->truncation_label < label_tick_ebb_start)
13733 : return false;
13734 0 : if (!partial_subreg_p (mode, truncated))
13735 : return true;
13736 0 : if (TRULY_NOOP_TRUNCATION_MODES_P (mode, truncated))
13737 : return true;
13738 : return false;
13739 : }
13740 :
13741 : /* If X is a hard reg or a subreg record the mode that the register is
13742 : accessed in. For non-TARGET_TRULY_NOOP_TRUNCATION targets we might be
13743 : able to turn a truncate into a subreg using this information. Return true
13744 : if traversing X is complete. */
13745 :
13746 : static bool
13747 197900725 : record_truncated_value (rtx x)
13748 : {
13749 197900725 : machine_mode truncated_mode;
13750 197900725 : reg_stat_type *rsp;
13751 :
13752 197900725 : if (GET_CODE (x) == SUBREG && REG_P (SUBREG_REG (x)))
13753 : {
13754 1770502 : machine_mode original_mode = GET_MODE (SUBREG_REG (x));
13755 1770502 : truncated_mode = GET_MODE (x);
13756 :
13757 1770502 : if (!partial_subreg_p (truncated_mode, original_mode))
13758 : return true;
13759 :
13760 1057297 : truncated_mode = GET_MODE (x);
13761 1057297 : if (TRULY_NOOP_TRUNCATION_MODES_P (truncated_mode, original_mode))
13762 : return true;
13763 :
13764 0 : x = SUBREG_REG (x);
13765 0 : }
13766 : /* ??? For hard-regs we now record everything. We might be able to
13767 : optimize this using last_set_mode. */
13768 196130223 : else if (REG_P (x) && REGNO (x) < FIRST_PSEUDO_REGISTER)
13769 20775440 : truncated_mode = GET_MODE (x);
13770 : else
13771 : return false;
13772 :
13773 20775440 : rsp = ®_stat[REGNO (x)];
13774 20775440 : if (rsp->truncated_to_mode == 0
13775 9613444 : || rsp->truncation_label < label_tick_ebb_start
13776 29152631 : || partial_subreg_p (truncated_mode, rsp->truncated_to_mode))
13777 : {
13778 12398842 : rsp->truncated_to_mode = truncated_mode;
13779 12398842 : rsp->truncation_label = label_tick;
13780 : }
13781 :
13782 : return true;
13783 : }
13784 :
13785 : /* Callback for note_uses. Find hardregs and subregs of pseudos and
13786 : the modes they are used in. This can help turning TRUNCATEs into
13787 : SUBREGs. */
13788 :
13789 : static void
13790 75531186 : record_truncated_values (rtx *loc, void *data ATTRIBUTE_UNUSED)
13791 : {
13792 75531186 : subrtx_var_iterator::array_type array;
13793 273431911 : FOR_EACH_SUBRTX_VAR (iter, array, *loc, NONCONST)
13794 197900725 : if (record_truncated_value (*iter))
13795 22545942 : iter.skip_subrtxes ();
13796 75531186 : }
13797 :
13798 : /* Scan X for promoted SUBREGs. For each one found,
13799 : note what it implies to the registers used in it. */
13800 :
13801 : static void
13802 359783947 : check_promoted_subreg (rtx_insn *insn, rtx x)
13803 : {
13804 359783947 : if (GET_CODE (x) == SUBREG
13805 2134183 : && SUBREG_PROMOTED_VAR_P (x)
13806 359789780 : && REG_P (SUBREG_REG (x)))
13807 5833 : record_promoted_value (insn, x);
13808 : else
13809 : {
13810 359778114 : const char *format = GET_RTX_FORMAT (GET_CODE (x));
13811 359778114 : int i, j;
13812 :
13813 865825135 : for (i = 0; i < GET_RTX_LENGTH (GET_CODE (x)); i++)
13814 506047021 : switch (format[i])
13815 : {
13816 269366018 : case 'e':
13817 269366018 : check_promoted_subreg (insn, XEXP (x, i));
13818 269366018 : break;
13819 11702849 : case 'V':
13820 11702849 : case 'E':
13821 11702849 : if (XVEC (x, i) != 0)
13822 36199189 : for (j = 0; j < XVECLEN (x, i); j++)
13823 24496340 : check_promoted_subreg (insn, XVECEXP (x, i, j));
13824 : break;
13825 : }
13826 : }
13827 359783947 : }
13828 :
13829 : /* Verify that all the registers and memory references mentioned in *LOC are
13830 : still valid. *LOC was part of a value set in INSN when label_tick was
13831 : equal to TICK. Return false if some are not. If REPLACE is true, replace
13832 : the invalid references with (clobber (const_int 0)) and return true. This
13833 : replacement is useful because we often can get useful information about
13834 : the form of a value (e.g., if it was produced by a shift that always
13835 : produces -1 or 0) even though we don't know exactly what registers it
13836 : was produced from. */
13837 :
13838 : static bool
13839 488314386 : get_last_value_validate (rtx *loc, rtx_insn *insn, int tick, bool replace)
13840 : {
13841 488314386 : rtx x = *loc;
13842 488314386 : const char *fmt = GET_RTX_FORMAT (GET_CODE (x));
13843 488314386 : int len = GET_RTX_LENGTH (GET_CODE (x));
13844 488314386 : int i, j;
13845 :
13846 488314386 : if (REG_P (x))
13847 : {
13848 156020820 : unsigned int regno = REGNO (x);
13849 156020820 : unsigned int endregno = END_REGNO (x);
13850 156020820 : unsigned int j;
13851 :
13852 288162701 : for (j = regno; j < endregno; j++)
13853 : {
13854 156046828 : reg_stat_type *rsp = ®_stat[j];
13855 156046828 : if (rsp->last_set_invalid
13856 : /* If this is a pseudo-register that was only set once and not
13857 : live at the beginning of the function, it is always valid. */
13858 257951591 : || (! (regno >= FIRST_PSEUDO_REGISTER
13859 117541734 : && regno < reg_n_sets_max
13860 117507185 : && REG_N_SETS (regno) == 1
13861 203809526 : && (!REGNO_REG_SET_P
13862 : (DF_LR_IN (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb),
13863 : regno)))
13864 30538525 : && rsp->last_set_label > tick))
13865 : {
13866 23904947 : if (replace)
13867 12350329 : *loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
13868 23904947 : return replace;
13869 : }
13870 : }
13871 :
13872 : return true;
13873 : }
13874 : /* If this is a memory reference, make sure that there were no stores after
13875 : it that might have clobbered the value. We don't have alias info, so we
13876 : assume any store invalidates it. Moreover, we only have local UIDs, so
13877 : we also assume that there were stores in the intervening basic blocks. */
13878 34242082 : else if (MEM_P (x) && !MEM_READONLY_P (x)
13879 364453113 : && (tick != label_tick || DF_INSN_LUID (insn) <= mem_last_set))
13880 : {
13881 7465518 : if (replace)
13882 3735411 : *loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
13883 7465518 : return replace;
13884 : }
13885 :
13886 811262789 : for (i = 0; i < len; i++)
13887 : {
13888 498162500 : if (fmt[i] == 'e')
13889 : {
13890 : /* Check for identical subexpressions. If x contains
13891 : identical subexpression we only have to traverse one of
13892 : them. */
13893 309410962 : if (i == 1 && ARITHMETIC_P (x))
13894 : {
13895 : /* Note that at this point x0 has already been checked
13896 : and found valid. */
13897 114681607 : rtx x0 = XEXP (x, 0);
13898 114681607 : rtx x1 = XEXP (x, 1);
13899 :
13900 : /* If x0 and x1 are identical then x is also valid. */
13901 114681607 : if (x0 == x1)
13902 : return true;
13903 :
13904 : /* If x1 is identical to a subexpression of x0 then
13905 : while checking x0, x1 has already been checked. Thus
13906 : it is valid and so as x. */
13907 114302200 : if (ARITHMETIC_P (x0)
13908 32539128 : && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
13909 : return true;
13910 :
13911 : /* If x0 is identical to a subexpression of x1 then x is
13912 : valid iff the rest of x1 is valid. */
13913 112378968 : if (ARITHMETIC_P (x1)
13914 1262088 : && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
13915 490 : return
13916 530 : get_last_value_validate (&XEXP (x1,
13917 : x0 == XEXP (x1, 0) ? 1 : 0),
13918 490 : insn, tick, replace);
13919 : }
13920 :
13921 307107833 : if (!get_last_value_validate (&XEXP (x, i), insn, tick, replace))
13922 : return false;
13923 : }
13924 188751538 : else if (fmt[i] == 'E')
13925 29204254 : for (j = 0; j < XVECLEN (x, i); j++)
13926 23152833 : if (!get_last_value_validate (&XVECEXP (x, i, j),
13927 : insn, tick, replace))
13928 : return false;
13929 : }
13930 :
13931 : /* If we haven't found a reason for it to be invalid, it is valid. */
13932 : return true;
13933 : }
13934 :
13935 : /* Get the last value assigned to X, if known. Some registers
13936 : in the value may be replaced with (clobber (const_int 0)) if their value
13937 : is known longer known reliably. */
13938 :
13939 : static rtx
13940 222959720 : get_last_value (const_rtx x)
13941 : {
13942 222959720 : unsigned int regno;
13943 222959720 : rtx value;
13944 222959720 : reg_stat_type *rsp;
13945 :
13946 : /* If this is a non-paradoxical SUBREG, get the value of its operand and
13947 : then convert it to the desired mode. If this is a paradoxical SUBREG,
13948 : we cannot predict what values the "extra" bits might have. */
13949 222959720 : if (GET_CODE (x) == SUBREG
13950 12104966 : && subreg_lowpart_p (x)
13951 11611875 : && !paradoxical_subreg_p (x)
13952 229694594 : && (value = get_last_value (SUBREG_REG (x))) != 0)
13953 3429925 : return gen_lowpart (GET_MODE (x), value);
13954 :
13955 219529795 : if (!REG_P (x))
13956 : return 0;
13957 :
13958 191335343 : regno = REGNO (x);
13959 191335343 : rsp = ®_stat[regno];
13960 191335343 : value = rsp->last_set_value;
13961 :
13962 : /* If we don't have a value, or if it isn't for this basic block and
13963 : it's either a hard register, set more than once, or it's a live
13964 : at the beginning of the function, return 0.
13965 :
13966 : Because if it's not live at the beginning of the function then the reg
13967 : is always set before being used (is never used without being set).
13968 : And, if it's set only once, and it's always set before use, then all
13969 : uses must have the same last value, even if it's not from this basic
13970 : block. */
13971 :
13972 191335343 : if (value == 0
13973 191335343 : || (rsp->last_set_label < label_tick_ebb_start
13974 78187375 : && (regno < FIRST_PSEUDO_REGISTER
13975 77315569 : || regno >= reg_n_sets_max
13976 77315569 : || REG_N_SETS (regno) != 1
13977 16433434 : || REGNO_REG_SET_P
13978 : (DF_LR_IN (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb), regno))))
13979 114088450 : return 0;
13980 :
13981 : /* If the value was set in a later insn than the ones we are processing,
13982 : we can't use it even if the register was only set once. */
13983 77246893 : if (rsp->last_set_label == label_tick
13984 77246893 : && DF_INSN_LUID (rsp->last_set) >= subst_low_luid)
13985 : return 0;
13986 :
13987 : /* If fewer bits were set than what we are asked for now, we cannot use
13988 : the value. */
13989 57135993 : if (maybe_lt (GET_MODE_PRECISION (rsp->last_set_mode),
13990 57135993 : GET_MODE_PRECISION (GET_MODE (x))))
13991 : return 0;
13992 :
13993 : /* If the value has all its registers valid, return it. */
13994 57134571 : if (get_last_value_validate (&value, rsp->last_set,
13995 : rsp->last_set_label, false))
13996 52933636 : return value;
13997 :
13998 : /* Otherwise, make a copy and replace any invalid register with
13999 : (clobber (const_int 0)). If that fails for some reason, return 0. */
14000 :
14001 4200935 : value = copy_rtx (value);
14002 4200935 : if (get_last_value_validate (&value, rsp->last_set,
14003 : rsp->last_set_label, true))
14004 4200935 : return value;
14005 :
14006 : return 0;
14007 : }
14008 :
14009 : /* Define three variables used for communication between the following
14010 : routines. */
14011 :
14012 : static unsigned int reg_dead_regno, reg_dead_endregno;
14013 : static int reg_dead_flag;
14014 : rtx reg_dead_reg;
14015 :
14016 : /* Function called via note_stores from reg_dead_at_p.
14017 :
14018 : If DEST is within [reg_dead_regno, reg_dead_endregno), set
14019 : reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
14020 :
14021 : static void
14022 614654 : reg_dead_at_p_1 (rtx dest, const_rtx x, void *data ATTRIBUTE_UNUSED)
14023 : {
14024 614654 : unsigned int regno, endregno;
14025 :
14026 614654 : if (!REG_P (dest))
14027 : return;
14028 :
14029 563575 : regno = REGNO (dest);
14030 563575 : endregno = END_REGNO (dest);
14031 563575 : if (reg_dead_endregno > regno && reg_dead_regno < endregno)
14032 285638 : reg_dead_flag = (GET_CODE (x) == CLOBBER) ? 1 : -1;
14033 : }
14034 :
14035 : /* Return true if REG is known to be dead at INSN.
14036 :
14037 : We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
14038 : referencing REG, it is dead. If we hit a SET referencing REG, it is
14039 : live. Otherwise, see if it is live or dead at the start of the basic
14040 : block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
14041 : must be assumed to be always live. */
14042 :
14043 : static bool
14044 1553328 : reg_dead_at_p (rtx reg, rtx_insn *insn)
14045 : {
14046 1553328 : basic_block block;
14047 1553328 : unsigned int i;
14048 :
14049 : /* Set variables for reg_dead_at_p_1. */
14050 1553328 : reg_dead_regno = REGNO (reg);
14051 1553328 : reg_dead_endregno = END_REGNO (reg);
14052 1553328 : reg_dead_reg = reg;
14053 :
14054 1553328 : reg_dead_flag = 0;
14055 :
14056 : /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
14057 : we allow the machine description to decide whether use-and-clobber
14058 : patterns are OK. */
14059 1553328 : if (reg_dead_regno < FIRST_PSEUDO_REGISTER)
14060 : {
14061 3106656 : for (i = reg_dead_regno; i < reg_dead_endregno; i++)
14062 1553328 : if (!fixed_regs[i] && TEST_HARD_REG_BIT (newpat_used_regs, i))
14063 : return false;
14064 : }
14065 :
14066 : /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
14067 : beginning of basic block. */
14068 1553328 : block = BLOCK_FOR_INSN (insn);
14069 747557 : for (;;)
14070 : {
14071 2300885 : if (INSN_P (insn))
14072 : {
14073 2148451 : if (find_regno_note (insn, REG_UNUSED, reg_dead_regno))
14074 : return true;
14075 :
14076 795787 : note_stores (insn, reg_dead_at_p_1, NULL);
14077 795787 : if (reg_dead_flag)
14078 142819 : return reg_dead_flag == 1 ? 1 : 0;
14079 :
14080 652968 : if (find_regno_note (insn, REG_DEAD, reg_dead_regno))
14081 : return true;
14082 : }
14083 :
14084 777093 : if (insn == BB_HEAD (block))
14085 : break;
14086 :
14087 747557 : insn = PREV_INSN (insn);
14088 : }
14089 :
14090 : /* Look at live-in sets for the basic block that we were in. */
14091 59072 : for (i = reg_dead_regno; i < reg_dead_endregno; i++)
14092 29536 : if (REGNO_REG_SET_P (df_get_live_in (block), i))
14093 : return false;
14094 :
14095 : return true;
14096 : }
14097 :
14098 : /* Note hard registers in X that are used. */
14099 :
14100 : static void
14101 286610034 : mark_used_regs_combine (rtx x)
14102 : {
14103 331303997 : RTX_CODE code = GET_CODE (x);
14104 331303997 : unsigned int regno;
14105 331303997 : int i;
14106 :
14107 331303997 : switch (code)
14108 : {
14109 : case LABEL_REF:
14110 : case SYMBOL_REF:
14111 : case CONST:
14112 : CASE_CONST_ANY:
14113 : case PC:
14114 : case ADDR_VEC:
14115 : case ADDR_DIFF_VEC:
14116 : case ASM_INPUT:
14117 : return;
14118 :
14119 7210416 : case CLOBBER:
14120 : /* If we are clobbering a MEM, mark any hard registers inside the
14121 : address as used. */
14122 7210416 : if (MEM_P (XEXP (x, 0)))
14123 5574 : mark_used_regs_combine (XEXP (XEXP (x, 0), 0));
14124 : return;
14125 :
14126 75759746 : case REG:
14127 75759746 : regno = REGNO (x);
14128 : /* A hard reg in a wide mode may really be multiple registers.
14129 : If so, mark all of them just like the first. */
14130 75759746 : if (regno < FIRST_PSEUDO_REGISTER)
14131 : {
14132 : /* None of this applies to the stack, frame or arg pointers. */
14133 8992084 : if (regno == STACK_POINTER_REGNUM
14134 8992084 : || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
14135 : && regno == HARD_FRAME_POINTER_REGNUM)
14136 8059660 : || (FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
14137 1093785 : && regno == ARG_POINTER_REGNUM && fixed_regs[regno])
14138 6965875 : || regno == FRAME_POINTER_REGNUM)
14139 : return;
14140 :
14141 1657879 : add_to_hard_reg_set (&newpat_used_regs, GET_MODE (x), regno);
14142 : }
14143 : return;
14144 :
14145 44688389 : case SET:
14146 44688389 : {
14147 : /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
14148 : the address. */
14149 44688389 : rtx testreg = SET_DEST (x);
14150 :
14151 44688389 : while (GET_CODE (testreg) == SUBREG
14152 44704330 : || GET_CODE (testreg) == ZERO_EXTRACT
14153 89725032 : || GET_CODE (testreg) == STRICT_LOW_PART)
14154 340650 : testreg = XEXP (testreg, 0);
14155 :
14156 44688389 : if (MEM_P (testreg))
14157 4850812 : mark_used_regs_combine (XEXP (testreg, 0));
14158 :
14159 44688389 : mark_used_regs_combine (SET_SRC (x));
14160 : }
14161 44688389 : return;
14162 :
14163 132318220 : default:
14164 132318220 : break;
14165 : }
14166 :
14167 : /* Recursively scan the operands of this expression. */
14168 :
14169 132318220 : {
14170 132318220 : const char *fmt = GET_RTX_FORMAT (code);
14171 :
14172 383810651 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
14173 : {
14174 251492431 : if (fmt[i] == 'e')
14175 204110050 : mark_used_regs_combine (XEXP (x, i));
14176 47382381 : else if (fmt[i] == 'E')
14177 : {
14178 : int j;
14179 :
14180 64859596 : for (j = 0; j < XVECLEN (x, i); j++)
14181 44942314 : mark_used_regs_combine (XVECEXP (x, i, j));
14182 : }
14183 : }
14184 : }
14185 : }
14186 :
14187 : /* Remove register number REGNO from the dead registers list of INSN.
14188 :
14189 : Return the note used to record the death, if there was one. */
14190 :
14191 : rtx
14192 3102828 : remove_death (unsigned int regno, rtx_insn *insn)
14193 : {
14194 3102828 : rtx note = find_regno_note (insn, REG_DEAD, regno);
14195 :
14196 3102828 : if (note)
14197 466077 : remove_note (insn, note);
14198 :
14199 3102828 : return note;
14200 : }
14201 :
14202 : /* For each register (hardware or pseudo) used within expression X, if its
14203 : death is in an instruction with luid between FROM_LUID (inclusive) and
14204 : TO_INSN (exclusive), put a REG_DEAD note for that register in the
14205 : list headed by PNOTES.
14206 :
14207 : That said, don't move registers killed by maybe_kill_insn.
14208 :
14209 : This is done when X is being merged by combination into TO_INSN. These
14210 : notes will then be distributed as needed. */
14211 :
14212 : static void
14213 24424901 : move_deaths (rtx x, rtx maybe_kill_insn, int from_luid, rtx_insn *to_insn,
14214 : rtx *pnotes)
14215 : {
14216 24960562 : const char *fmt;
14217 24960562 : int len, i;
14218 24960562 : enum rtx_code code = GET_CODE (x);
14219 :
14220 24960562 : if (code == REG)
14221 : {
14222 6142981 : unsigned int regno = REGNO (x);
14223 6142981 : rtx_insn *where_dead = reg_stat[regno].last_death;
14224 :
14225 : /* If we do not know where the register died, it may still die between
14226 : FROM_LUID and TO_INSN. If so, find it. This is PR83304. */
14227 6142981 : if (!where_dead || DF_INSN_LUID (where_dead) >= DF_INSN_LUID (to_insn))
14228 : {
14229 3312784 : rtx_insn *insn = prev_real_nondebug_insn (to_insn);
14230 3312784 : while (insn
14231 4890140 : && BLOCK_FOR_INSN (insn) == BLOCK_FOR_INSN (to_insn)
14232 9042521 : && DF_INSN_LUID (insn) >= from_luid)
14233 : {
14234 2193826 : if (dead_or_set_regno_p (insn, regno))
14235 : {
14236 584758 : if (find_regno_note (insn, REG_DEAD, regno))
14237 6142981 : where_dead = insn;
14238 : break;
14239 : }
14240 :
14241 1609068 : insn = prev_real_nondebug_insn (insn);
14242 : }
14243 : }
14244 :
14245 : /* Don't move the register if it gets killed in between from and to. */
14246 148944 : if (maybe_kill_insn && reg_set_p (x, maybe_kill_insn)
14247 6185831 : && ! reg_referenced_p (x, maybe_kill_insn))
14248 : return;
14249 :
14250 6100131 : if (where_dead
14251 3191975 : && BLOCK_FOR_INSN (where_dead) == BLOCK_FOR_INSN (to_insn)
14252 3013482 : && DF_INSN_LUID (where_dead) >= from_luid
14253 9113392 : && DF_INSN_LUID (where_dead) < DF_INSN_LUID (to_insn))
14254 : {
14255 2727062 : rtx note = remove_death (regno, where_dead);
14256 :
14257 : /* It is possible for the call above to return 0. This can occur
14258 : when last_death points to I2 or I1 that we combined with.
14259 : In that case make a new note.
14260 :
14261 : We must also check for the case where X is a hard register
14262 : and NOTE is a death note for a range of hard registers
14263 : including X. In that case, we must put REG_DEAD notes for
14264 : the remaining registers in place of NOTE. */
14265 :
14266 2727062 : if (note != 0 && regno < FIRST_PSEUDO_REGISTER
14267 2727062 : && partial_subreg_p (GET_MODE (x), GET_MODE (XEXP (note, 0))))
14268 : {
14269 0 : unsigned int deadregno = REGNO (XEXP (note, 0));
14270 0 : unsigned int deadend = END_REGNO (XEXP (note, 0));
14271 0 : unsigned int ourend = END_REGNO (x);
14272 0 : unsigned int i;
14273 :
14274 0 : for (i = deadregno; i < deadend; i++)
14275 0 : if (i < regno || i >= ourend)
14276 0 : add_reg_note (where_dead, REG_DEAD, regno_reg_rtx[i]);
14277 : }
14278 :
14279 : /* If we didn't find any note, or if we found a REG_DEAD note that
14280 : covers only part of the given reg, and we have a multi-reg hard
14281 : register, then to be safe we must check for REG_DEAD notes
14282 : for each register other than the first. They could have
14283 : their own REG_DEAD notes lying around. */
14284 2727062 : else if ((note == 0
14285 : || (note != 0
14286 90358 : && partial_subreg_p (GET_MODE (XEXP (note, 0)),
14287 90358 : GET_MODE (x))))
14288 2636704 : && regno < FIRST_PSEUDO_REGISTER
14289 3031855 : && REG_NREGS (x) > 1)
14290 : {
14291 0 : unsigned int ourend = END_REGNO (x);
14292 0 : unsigned int i, offset;
14293 0 : rtx oldnotes = 0;
14294 :
14295 0 : if (note)
14296 0 : offset = hard_regno_nregs (regno, GET_MODE (XEXP (note, 0)));
14297 : else
14298 : offset = 1;
14299 :
14300 0 : for (i = regno + offset; i < ourend; i++)
14301 0 : move_deaths (regno_reg_rtx[i],
14302 : maybe_kill_insn, from_luid, to_insn, &oldnotes);
14303 : }
14304 :
14305 2727062 : if (note != 0 && GET_MODE (XEXP (note, 0)) == GET_MODE (x))
14306 : {
14307 90334 : XEXP (note, 1) = *pnotes;
14308 90334 : *pnotes = note;
14309 : }
14310 : else
14311 2636728 : *pnotes = alloc_reg_note (REG_DEAD, x, *pnotes);
14312 : }
14313 :
14314 6100131 : return;
14315 : }
14316 :
14317 18817581 : else if (GET_CODE (x) == SET)
14318 : {
14319 4190391 : rtx dest = SET_DEST (x);
14320 :
14321 4190391 : move_deaths (SET_SRC (x), maybe_kill_insn, from_luid, to_insn, pnotes);
14322 :
14323 : /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
14324 : that accesses one word of a multi-word item, some
14325 : piece of everything register in the expression is used by
14326 : this insn, so remove any old death. */
14327 : /* ??? So why do we test for equality of the sizes? */
14328 :
14329 4190391 : if (GET_CODE (dest) == ZERO_EXTRACT
14330 4189964 : || GET_CODE (dest) == STRICT_LOW_PART
14331 8378614 : || (GET_CODE (dest) == SUBREG
14332 74957 : && !read_modify_subreg_p (dest)))
14333 : {
14334 61920 : move_deaths (dest, maybe_kill_insn, from_luid, to_insn, pnotes);
14335 61920 : return;
14336 : }
14337 :
14338 : /* If this is some other SUBREG, we know it replaces the entire
14339 : value, so use that as the destination. */
14340 4128471 : if (GET_CODE (dest) == SUBREG)
14341 15205 : dest = SUBREG_REG (dest);
14342 :
14343 : /* If this is a MEM, adjust deaths of anything used in the address.
14344 : For a REG (the only other possibility), the entire value is
14345 : being replaced so the old value is not used in this insn. */
14346 :
14347 4128471 : if (MEM_P (dest))
14348 473741 : move_deaths (XEXP (dest, 0), maybe_kill_insn, from_luid,
14349 : to_insn, pnotes);
14350 : return;
14351 : }
14352 :
14353 14627190 : else if (GET_CODE (x) == CLOBBER)
14354 : return;
14355 :
14356 14028875 : len = GET_RTX_LENGTH (code);
14357 14028875 : fmt = GET_RTX_FORMAT (code);
14358 :
14359 36542256 : for (i = 0; i < len; i++)
14360 : {
14361 22513381 : if (fmt[i] == 'E')
14362 : {
14363 982223 : int j;
14364 3488689 : for (j = XVECLEN (x, i) - 1; j >= 0; j--)
14365 2506466 : move_deaths (XVECEXP (x, i, j), maybe_kill_insn, from_luid,
14366 : to_insn, pnotes);
14367 : }
14368 21531158 : else if (fmt[i] == 'e')
14369 13601977 : move_deaths (XEXP (x, i), maybe_kill_insn, from_luid, to_insn, pnotes);
14370 : }
14371 : }
14372 :
14373 : /* Return true if X is the target of a bit-field assignment in BODY, the
14374 : pattern of an insn. X must be a REG. */
14375 :
14376 : static bool
14377 4725399 : reg_bitfield_target_p (rtx x, rtx body)
14378 : {
14379 4725399 : int i;
14380 :
14381 4725399 : if (GET_CODE (body) == SET)
14382 : {
14383 3459471 : rtx dest = SET_DEST (body);
14384 3459471 : rtx target;
14385 3459471 : unsigned int regno, tregno, endregno, endtregno;
14386 :
14387 3459471 : if (GET_CODE (dest) == ZERO_EXTRACT)
14388 434 : target = XEXP (dest, 0);
14389 3459037 : else if (GET_CODE (dest) == STRICT_LOW_PART)
14390 1990 : target = SUBREG_REG (XEXP (dest, 0));
14391 : else
14392 : return false;
14393 :
14394 2424 : if (GET_CODE (target) == SUBREG)
14395 227 : target = SUBREG_REG (target);
14396 :
14397 2424 : if (!REG_P (target))
14398 : return false;
14399 :
14400 2345 : tregno = REGNO (target), regno = REGNO (x);
14401 2345 : if (tregno >= FIRST_PSEUDO_REGISTER || regno >= FIRST_PSEUDO_REGISTER)
14402 2335 : return target == x;
14403 :
14404 10 : endtregno = end_hard_regno (GET_MODE (target), tregno);
14405 10 : endregno = end_hard_regno (GET_MODE (x), regno);
14406 :
14407 10 : return endregno > tregno && regno < endtregno;
14408 : }
14409 :
14410 1265928 : else if (GET_CODE (body) == PARALLEL)
14411 1887568 : for (i = XVECLEN (body, 0) - 1; i >= 0; i--)
14412 1269030 : if (reg_bitfield_target_p (x, XVECEXP (body, 0, i)))
14413 : return true;
14414 :
14415 : return false;
14416 : }
14417 :
14418 : /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
14419 : as appropriate. I3 and I2 are the insns resulting from the combination
14420 : insns including FROM (I2 may be zero).
14421 :
14422 : ELIM_I2 and ELIM_I1 are either zero or registers that we know will
14423 : not need REG_DEAD notes because they are being substituted for. This
14424 : saves searching in the most common cases.
14425 :
14426 : Each note in the list is either ignored or placed on some insns, depending
14427 : on the type of note. */
14428 :
14429 : static void
14430 9833396 : distribute_notes (rtx notes, rtx_insn *from_insn, rtx_insn *i3, rtx_insn *i2,
14431 : rtx elim_i2, rtx elim_i1, rtx elim_i0)
14432 : {
14433 9833396 : rtx note, next_note;
14434 9833396 : rtx tem_note;
14435 9833396 : rtx_insn *tem_insn;
14436 :
14437 22884086 : for (note = notes; note; note = next_note)
14438 : {
14439 13050690 : rtx_insn *place = 0, *place2 = 0;
14440 :
14441 13050690 : next_note = XEXP (note, 1);
14442 13050690 : switch (REG_NOTE_KIND (note))
14443 : {
14444 : case REG_BR_PROB:
14445 : case REG_BR_PRED:
14446 : /* Doesn't matter much where we put this, as long as it's somewhere.
14447 : It is preferable to keep these notes on branches, which is most
14448 : likely to be i3. */
14449 : place = i3;
14450 : break;
14451 :
14452 0 : case REG_NON_LOCAL_GOTO:
14453 0 : if (JUMP_P (i3))
14454 : place = i3;
14455 : else
14456 : {
14457 0 : gcc_assert (i2 && JUMP_P (i2));
14458 : place = i2;
14459 : }
14460 : break;
14461 :
14462 22151 : case REG_EH_REGION:
14463 22151 : {
14464 : /* The landing pad handling needs to be kept in sync with the
14465 : prerequisite checking in try_combine. */
14466 22151 : int lp_nr = INTVAL (XEXP (note, 0));
14467 : /* A REG_EH_REGION note transfering control can only ever come
14468 : from i3. */
14469 22151 : if (lp_nr > 0)
14470 12626 : gcc_assert (from_insn == i3);
14471 : /* We are making sure there is a single effective REG_EH_REGION
14472 : note and it's valid to put it on i3. */
14473 22151 : if (!insn_could_throw_p (from_insn)
14474 22151 : && !(lp_nr == INT_MIN && can_nonlocal_goto (from_insn)))
14475 : /* Throw away stray notes on insns that can never throw or
14476 : make a nonlocal goto. */
14477 : ;
14478 : else
14479 : {
14480 22070 : if (CALL_P (i3))
14481 : place = i3;
14482 : else
14483 : {
14484 2087 : gcc_assert (cfun->can_throw_non_call_exceptions);
14485 : /* If i3 can still trap preserve the note, otherwise we've
14486 : combined things such that we can now prove that the
14487 : instructions can't trap. Drop the note in this case. */
14488 2087 : if (may_trap_p (i3))
14489 : place = i3;
14490 : }
14491 : }
14492 : break;
14493 : }
14494 :
14495 126085 : case REG_ARGS_SIZE:
14496 : /* ??? How to distribute between i3-i1. Assume i3 contains the
14497 : entire adjustment. Assert i3 contains at least some adjust. */
14498 126085 : if (!noop_move_p (i3))
14499 : {
14500 126084 : poly_int64 old_size, args_size = get_args_size (note);
14501 : /* fixup_args_size_notes looks at REG_NORETURN note,
14502 : so ensure the note is placed there first. */
14503 126084 : if (CALL_P (i3))
14504 : {
14505 : rtx *np;
14506 1629 : for (np = &next_note; *np; np = &XEXP (*np, 1))
14507 20 : if (REG_NOTE_KIND (*np) == REG_NORETURN)
14508 : {
14509 9 : rtx n = *np;
14510 9 : *np = XEXP (n, 1);
14511 9 : XEXP (n, 1) = REG_NOTES (i3);
14512 9 : REG_NOTES (i3) = n;
14513 9 : break;
14514 : }
14515 : }
14516 126084 : old_size = fixup_args_size_notes (PREV_INSN (i3), i3, args_size);
14517 : /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
14518 : REG_ARGS_SIZE note to all noreturn calls, allow that here. */
14519 126084 : gcc_assert (maybe_ne (old_size, args_size)
14520 : || (CALL_P (i3)
14521 : && !ACCUMULATE_OUTGOING_ARGS
14522 : && find_reg_note (i3, REG_NORETURN, NULL_RTX)));
14523 : }
14524 : break;
14525 :
14526 83281 : case REG_NORETURN:
14527 83281 : case REG_SETJMP:
14528 83281 : case REG_TM:
14529 83281 : case REG_CALL_DECL:
14530 83281 : case REG_UNTYPED_CALL:
14531 83281 : case REG_CALL_NOCF_CHECK:
14532 : /* These notes must remain with the call. It should not be
14533 : possible for both I2 and I3 to be a call. */
14534 83281 : if (CALL_P (i3))
14535 : place = i3;
14536 : else
14537 : {
14538 0 : gcc_assert (i2 && CALL_P (i2));
14539 : place = i2;
14540 : }
14541 : break;
14542 :
14543 2000000 : case REG_UNUSED:
14544 : /* Any clobbers for i3 may still exist, and so we must process
14545 : REG_UNUSED notes from that insn.
14546 :
14547 : Any clobbers from i2 or i1 can only exist if they were added by
14548 : recog_for_combine. In that case, recog_for_combine created the
14549 : necessary REG_UNUSED notes. Trying to keep any original
14550 : REG_UNUSED notes from these insns can cause incorrect output
14551 : if it is for the same register as the original i3 dest.
14552 : In that case, we will notice that the register is set in i3,
14553 : and then add a REG_UNUSED note for the destination of i3, which
14554 : is wrong. However, it is possible to have REG_UNUSED notes from
14555 : i2 or i1 for register which were both used and clobbered, so
14556 : we keep notes from i2 or i1 if they will turn into REG_DEAD
14557 : notes. */
14558 :
14559 : /* If this register is set or clobbered between FROM_INSN and I3,
14560 : we should not create a note for it. */
14561 2000000 : if (reg_set_between_p (XEXP (note, 0), from_insn, i3))
14562 : break;
14563 :
14564 : /* If this register is set or clobbered in I3, put the note there
14565 : unless there is one already. */
14566 1917924 : if (reg_set_p (XEXP (note, 0), PATTERN (i3)))
14567 : {
14568 1113226 : if (from_insn != i3)
14569 : break;
14570 :
14571 622366 : if (! (REG_P (XEXP (note, 0))
14572 622366 : ? find_regno_note (i3, REG_UNUSED, REGNO (XEXP (note, 0)))
14573 0 : : find_reg_note (i3, REG_UNUSED, XEXP (note, 0))))
14574 : place = i3;
14575 : }
14576 : /* Otherwise, if this register is used by I3, then this register
14577 : now dies here, so we must put a REG_DEAD note here unless there
14578 : is one already. */
14579 804698 : else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3)))
14580 : {
14581 7680 : if (! (REG_P (XEXP (note, 0))
14582 7680 : ? find_regno_note (i3, REG_DEAD, REGNO (XEXP (note, 0)))
14583 0 : : find_reg_note (i3, REG_DEAD, XEXP (note, 0))))
14584 : {
14585 7443 : PUT_REG_NOTE_KIND (note, REG_DEAD);
14586 7443 : place = i3;
14587 : }
14588 : }
14589 :
14590 : /* A SET or CLOBBER of the REG_UNUSED reg has been removed,
14591 : but we can't tell which at this point. We must reset any
14592 : expectations we had about the value that was previously
14593 : stored in the reg. ??? Ideally, we'd adjust REG_N_SETS
14594 : and, if appropriate, restore its previous value, but we
14595 : don't have enough information for that at this point. */
14596 : else
14597 : {
14598 797018 : record_value_for_reg (XEXP (note, 0), NULL, NULL_RTX);
14599 :
14600 : /* Otherwise, if this register is now referenced in i2
14601 : then the register used to be modified in one of the
14602 : original insns. If it was i3 (say, in an unused
14603 : parallel), it's now completely gone, so the note can
14604 : be discarded. But if it was modified in i2, i1 or i0
14605 : and we still reference it in i2, then we're
14606 : referencing the previous value, and since the
14607 : register was modified and REG_UNUSED, we know that
14608 : the previous value is now dead. So, if we only
14609 : reference the register in i2, we change the note to
14610 : REG_DEAD, to reflect the previous value. However, if
14611 : we're also setting or clobbering the register as
14612 : scratch, we know (because the register was not
14613 : referenced in i3) that it's unused, just as it was
14614 : unused before, and we place the note in i2. */
14615 19761 : if (from_insn != i3 && i2 && INSN_P (i2)
14616 816779 : && reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
14617 : {
14618 7 : if (!reg_set_p (XEXP (note, 0), PATTERN (i2)))
14619 7 : PUT_REG_NOTE_KIND (note, REG_DEAD);
14620 7 : if (! (REG_P (XEXP (note, 0))
14621 7 : ? find_regno_note (i2, REG_NOTE_KIND (note),
14622 7 : REGNO (XEXP (note, 0)))
14623 0 : : find_reg_note (i2, REG_NOTE_KIND (note),
14624 : XEXP (note, 0))))
14625 : place = i2;
14626 : }
14627 : }
14628 :
14629 : break;
14630 :
14631 392145 : case REG_EQUAL:
14632 392145 : case REG_EQUIV:
14633 392145 : case REG_NOALIAS:
14634 : /* These notes say something about results of an insn. We can
14635 : only support them if they used to be on I3 in which case they
14636 : remain on I3. Otherwise they are ignored.
14637 :
14638 : If the note refers to an expression that is not a constant, we
14639 : must also ignore the note since we cannot tell whether the
14640 : equivalence is still true. It might be possible to do
14641 : slightly better than this (we only have a problem if I2DEST
14642 : or I1DEST is present in the expression), but it doesn't
14643 : seem worth the trouble. */
14644 :
14645 392145 : if (from_insn == i3
14646 182058 : && (XEXP (note, 0) == 0 || CONSTANT_P (XEXP (note, 0))))
14647 : place = i3;
14648 : break;
14649 :
14650 0 : case REG_INC:
14651 : /* These notes say something about how a register is used. They must
14652 : be present on any use of the register in I2 or I3. */
14653 0 : if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3)))
14654 0 : place = i3;
14655 :
14656 0 : if (i2 && reg_mentioned_p (XEXP (note, 0), PATTERN (i2)))
14657 : {
14658 0 : if (place)
14659 : place2 = i2;
14660 : else
14661 : place = i2;
14662 : }
14663 : break;
14664 :
14665 7791 : case REG_LABEL_TARGET:
14666 7791 : case REG_LABEL_OPERAND:
14667 : /* This can show up in several ways -- either directly in the
14668 : pattern, or hidden off in the constant pool with (or without?)
14669 : a REG_EQUAL note. */
14670 : /* ??? Ignore the without-reg_equal-note problem for now. */
14671 7791 : if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3))
14672 7791 : || ((tem_note = find_reg_note (i3, REG_EQUAL, NULL_RTX))
14673 0 : && GET_CODE (XEXP (tem_note, 0)) == LABEL_REF
14674 0 : && label_ref_label (XEXP (tem_note, 0)) == XEXP (note, 0)))
14675 : place = i3;
14676 :
14677 7791 : if (i2
14678 7791 : && (reg_mentioned_p (XEXP (note, 0), PATTERN (i2))
14679 0 : || ((tem_note = find_reg_note (i2, REG_EQUAL, NULL_RTX))
14680 0 : && GET_CODE (XEXP (tem_note, 0)) == LABEL_REF
14681 0 : && label_ref_label (XEXP (tem_note, 0)) == XEXP (note, 0))))
14682 : {
14683 0 : if (place)
14684 : place2 = i2;
14685 : else
14686 : place = i2;
14687 : }
14688 :
14689 : /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
14690 : as a JUMP_LABEL or decrement LABEL_NUSES if it's already
14691 : there. */
14692 7791 : if (place && JUMP_P (place)
14693 6758 : && REG_NOTE_KIND (note) == REG_LABEL_TARGET
14694 0 : && (JUMP_LABEL (place) == NULL
14695 0 : || JUMP_LABEL (place) == XEXP (note, 0)))
14696 : {
14697 0 : rtx label = JUMP_LABEL (place);
14698 :
14699 0 : if (!label)
14700 0 : JUMP_LABEL (place) = XEXP (note, 0);
14701 0 : else if (LABEL_P (label))
14702 0 : LABEL_NUSES (label)--;
14703 : }
14704 :
14705 7791 : if (place2 && JUMP_P (place2)
14706 0 : && REG_NOTE_KIND (note) == REG_LABEL_TARGET
14707 0 : && (JUMP_LABEL (place2) == NULL
14708 0 : || JUMP_LABEL (place2) == XEXP (note, 0)))
14709 : {
14710 0 : rtx label = JUMP_LABEL (place2);
14711 :
14712 0 : if (!label)
14713 0 : JUMP_LABEL (place2) = XEXP (note, 0);
14714 0 : else if (LABEL_P (label))
14715 0 : LABEL_NUSES (label)--;
14716 : place2 = 0;
14717 : }
14718 : break;
14719 :
14720 : case REG_NONNEG:
14721 : /* This note says something about the value of a register prior
14722 : to the execution of an insn. It is too much trouble to see
14723 : if the note is still correct in all situations. It is better
14724 : to simply delete it. */
14725 : break;
14726 :
14727 10379601 : case REG_DEAD:
14728 : /* If we replaced the right hand side of FROM_INSN with a
14729 : REG_EQUAL note, the original use of the dying register
14730 : will not have been combined into I3 and I2. In such cases,
14731 : FROM_INSN is guaranteed to be the first of the combined
14732 : instructions, so we simply need to search back before
14733 : FROM_INSN for the previous use or set of this register,
14734 : then alter the notes there appropriately.
14735 :
14736 : If the register is used as an input in I3, it dies there.
14737 : Similarly for I2, if it is nonzero and adjacent to I3.
14738 :
14739 : If the register is not used as an input in either I3 or I2
14740 : and it is not one of the registers we were supposed to eliminate,
14741 : there are two possibilities. We might have a non-adjacent I2
14742 : or we might have somehow eliminated an additional register
14743 : from a computation. For example, we might have had A & B where
14744 : we discover that B will always be zero. In this case we will
14745 : eliminate the reference to A.
14746 :
14747 : In both cases, we must search to see if we can find a previous
14748 : use of A and put the death note there. */
14749 :
14750 10379601 : if (from_insn
14751 7281382 : && from_insn == i2mod
14752 10381399 : && !reg_overlap_mentioned_p (XEXP (note, 0), i2mod_new_rhs))
14753 : tem_insn = from_insn;
14754 : else
14755 : {
14756 10378106 : if (from_insn
14757 7279887 : && CALL_P (from_insn)
14758 10614204 : && find_reg_fusage (from_insn, USE, XEXP (note, 0)))
14759 : place = from_insn;
14760 10225282 : else if (i2 && reg_set_p (XEXP (note, 0), PATTERN (i2)))
14761 : {
14762 : /* If the new I2 sets the same register that is marked
14763 : dead in the note, we do not in general know where to
14764 : put the note. One important case we _can_ handle is
14765 : when the note comes from I3. */
14766 38850 : if (from_insn == i3)
14767 : place = i3;
14768 : else
14769 : break;
14770 : }
14771 10186432 : else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3)))
14772 : place = i3;
14773 106354 : else if (i2 != 0 && next_nonnote_nondebug_insn (i2) == i3
14774 4072851 : && reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
14775 : place = i2;
14776 3924103 : else if ((rtx_equal_p (XEXP (note, 0), elim_i2)
14777 3812310 : && !(i2mod
14778 32764 : && reg_overlap_mentioned_p (XEXP (note, 0),
14779 : i2mod_old_rhs)))
14780 144567 : || rtx_equal_p (XEXP (note, 0), elim_i1)
14781 3978271 : || rtx_equal_p (XEXP (note, 0), elim_i0))
14782 : break;
14783 237104 : tem_insn = i3;
14784 : }
14785 :
14786 237104 : if (place == 0)
14787 : {
14788 50664 : basic_block bb = this_basic_block;
14789 :
14790 2384073 : for (tem_insn = PREV_INSN (tem_insn); place == 0; tem_insn = PREV_INSN (tem_insn))
14791 : {
14792 2384073 : if (!NONDEBUG_INSN_P (tem_insn))
14793 : {
14794 1758953 : if (tem_insn == BB_HEAD (bb))
14795 : break;
14796 1724498 : continue;
14797 : }
14798 :
14799 : /* If the register is being set at TEM_INSN, see if that is all
14800 : TEM_INSN is doing. If so, delete TEM_INSN. Otherwise, make this
14801 : into a REG_UNUSED note instead. Don't delete sets to
14802 : global register vars. */
14803 625120 : if ((REGNO (XEXP (note, 0)) >= FIRST_PSEUDO_REGISTER
14804 1477 : || !global_regs[REGNO (XEXP (note, 0))])
14805 626597 : && reg_set_p (XEXP (note, 0), PATTERN (tem_insn)))
14806 : {
14807 15079 : rtx set = single_set (tem_insn);
14808 15079 : rtx inner_dest = 0;
14809 :
14810 15079 : if (set != 0)
14811 12205 : for (inner_dest = SET_DEST (set);
14812 12214 : (GET_CODE (inner_dest) == STRICT_LOW_PART
14813 12214 : || GET_CODE (inner_dest) == SUBREG
14814 12214 : || GET_CODE (inner_dest) == ZERO_EXTRACT);
14815 9 : inner_dest = XEXP (inner_dest, 0))
14816 : ;
14817 :
14818 : /* Verify that it was the set, and not a clobber that
14819 : modified the register.
14820 :
14821 : If we cannot delete the setter due to side
14822 : effects, mark the user with an UNUSED note instead
14823 : of deleting it. */
14824 :
14825 12205 : if (set != 0 && ! side_effects_p (SET_SRC (set))
14826 11833 : && rtx_equal_p (XEXP (note, 0), inner_dest))
14827 : {
14828 : /* Move the notes and links of TEM_INSN elsewhere.
14829 : This might delete other dead insns recursively.
14830 : First set the pattern to something that won't use
14831 : any register. */
14832 11730 : rtx old_notes = REG_NOTES (tem_insn);
14833 :
14834 11730 : PATTERN (tem_insn) = pc_rtx;
14835 11730 : REG_NOTES (tem_insn) = NULL;
14836 :
14837 11730 : distribute_notes (old_notes, tem_insn, tem_insn, NULL,
14838 : NULL_RTX, NULL_RTX, NULL_RTX);
14839 11730 : distribute_links (LOG_LINKS (tem_insn));
14840 :
14841 11730 : unsigned int regno = REGNO (XEXP (note, 0));
14842 11730 : reg_stat_type *rsp = ®_stat[regno];
14843 11730 : if (rsp->last_set == tem_insn)
14844 10354 : record_value_for_reg (XEXP (note, 0), NULL, NULL_RTX);
14845 :
14846 11730 : SET_INSN_DELETED (tem_insn);
14847 11730 : if (tem_insn == i2)
14848 608911 : i2 = NULL;
14849 : }
14850 : else
14851 : {
14852 3349 : PUT_REG_NOTE_KIND (note, REG_UNUSED);
14853 :
14854 : /* If there isn't already a REG_UNUSED note, put one
14855 : here. Do not place a REG_DEAD note, even if
14856 : the register is also used here; that would not
14857 : match the algorithm used in lifetime analysis
14858 : and can cause the consistency check in the
14859 : scheduler to fail. */
14860 3349 : if (! find_regno_note (tem_insn, REG_UNUSED,
14861 3349 : REGNO (XEXP (note, 0))))
14862 1824 : place = tem_insn;
14863 : break;
14864 : }
14865 : }
14866 610041 : else if (reg_referenced_p (XEXP (note, 0), PATTERN (tem_insn))
14867 610041 : || (CALL_P (tem_insn)
14868 16093 : && find_reg_fusage (tem_insn, USE, XEXP (note, 0))))
14869 : {
14870 12860 : place = tem_insn;
14871 :
14872 : /* If we are doing a 3->2 combination, and we have a
14873 : register which formerly died in i3 and was not used
14874 : by i2, which now no longer dies in i3 and is used in
14875 : i2 but does not die in i2, and place is between i2
14876 : and i3, then we may need to move a link from place to
14877 : i2. */
14878 3776 : if (i2 && DF_INSN_LUID (place) > DF_INSN_LUID (i2)
14879 78 : && from_insn
14880 78 : && DF_INSN_LUID (from_insn) > DF_INSN_LUID (i2)
14881 12938 : && reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
14882 : {
14883 78 : struct insn_link *links = LOG_LINKS (place);
14884 78 : LOG_LINKS (place) = NULL;
14885 78 : distribute_links (links);
14886 : }
14887 : break;
14888 : }
14889 :
14890 608911 : if (tem_insn == BB_HEAD (bb))
14891 : break;
14892 : }
14893 :
14894 : }
14895 :
14896 : /* If the register is set or already dead at PLACE, we needn't do
14897 : anything with this note if it is still a REG_DEAD note.
14898 : We check here if it is set at all, not if is it totally replaced,
14899 : which is what `dead_or_set_p' checks, so also check for it being
14900 : set partially. */
14901 :
14902 6501627 : if (place && REG_NOTE_KIND (note) == REG_DEAD)
14903 : {
14904 6463823 : unsigned int regno = REGNO (XEXP (note, 0));
14905 6463823 : reg_stat_type *rsp = ®_stat[regno];
14906 :
14907 6463823 : if (dead_or_set_p (place, XEXP (note, 0))
14908 6463823 : || reg_bitfield_target_p (XEXP (note, 0), PATTERN (place)))
14909 : {
14910 : /* Unless the register previously died in PLACE, clear
14911 : last_death. [I no longer understand why this is
14912 : being done.] */
14913 3007480 : if (rsp->last_death != place)
14914 569876 : rsp->last_death = 0;
14915 : place = 0;
14916 : }
14917 : else
14918 3456343 : rsp->last_death = place;
14919 :
14920 : /* If this is a death note for a hard reg that is occupying
14921 : multiple registers, ensure that we are still using all
14922 : parts of the object. If we find a piece of the object
14923 : that is unused, we must arrange for an appropriate REG_DEAD
14924 : note to be added for it. However, we can't just emit a USE
14925 : and tag the note to it, since the register might actually
14926 : be dead; so we recurse, and the recursive call then finds
14927 : the previous insn that used this register. */
14928 :
14929 3456343 : if (place && REG_NREGS (XEXP (note, 0)) > 1)
14930 : {
14931 777 : unsigned int endregno = END_REGNO (XEXP (note, 0));
14932 777 : bool all_used = true;
14933 777 : unsigned int i;
14934 :
14935 2331 : for (i = regno; i < endregno; i++)
14936 1554 : if ((! refers_to_regno_p (i, PATTERN (place))
14937 1554 : && ! find_regno_fusage (place, USE, i))
14938 3108 : || dead_or_set_regno_p (place, i))
14939 : {
14940 : all_used = false;
14941 : break;
14942 : }
14943 :
14944 777 : if (! all_used)
14945 : {
14946 : /* Put only REG_DEAD notes for pieces that are
14947 : not already dead or set. */
14948 :
14949 0 : for (i = regno; i < endregno;
14950 0 : i += hard_regno_nregs (i, reg_raw_mode[i]))
14951 : {
14952 0 : rtx piece = regno_reg_rtx[i];
14953 0 : basic_block bb = this_basic_block;
14954 :
14955 0 : if (! dead_or_set_p (place, piece)
14956 0 : && ! reg_bitfield_target_p (piece,
14957 0 : PATTERN (place)))
14958 : {
14959 0 : rtx new_note = alloc_reg_note (REG_DEAD, piece,
14960 : NULL_RTX);
14961 :
14962 0 : distribute_notes (new_note, place, place,
14963 : NULL, NULL_RTX, NULL_RTX,
14964 : NULL_RTX);
14965 : }
14966 0 : else if (! refers_to_regno_p (i, PATTERN (place))
14967 0 : && ! find_regno_fusage (place, USE, i))
14968 0 : for (tem_insn = PREV_INSN (place); ;
14969 0 : tem_insn = PREV_INSN (tem_insn))
14970 : {
14971 0 : if (!NONDEBUG_INSN_P (tem_insn))
14972 : {
14973 0 : if (tem_insn == BB_HEAD (bb))
14974 : break;
14975 0 : continue;
14976 : }
14977 0 : if (dead_or_set_p (tem_insn, piece)
14978 0 : || reg_bitfield_target_p (piece,
14979 0 : PATTERN (tem_insn)))
14980 : {
14981 0 : add_reg_note (tem_insn, REG_UNUSED, piece);
14982 0 : break;
14983 : }
14984 : }
14985 : }
14986 :
14987 : place = 0;
14988 : }
14989 : }
14990 : }
14991 : break;
14992 :
14993 0 : default:
14994 : /* Any other notes should not be present at this point in the
14995 : compilation. */
14996 0 : gcc_unreachable ();
14997 : }
14998 :
14999 4244558 : if (place)
15000 : {
15001 4221657 : XEXP (note, 1) = REG_NOTES (place);
15002 4221657 : REG_NOTES (place) = note;
15003 :
15004 : /* Set added_notes_insn to the earliest insn we added a note to. */
15005 4221657 : if (added_notes_insn == 0
15006 4221657 : || DF_INSN_LUID (added_notes_insn) > DF_INSN_LUID (place))
15007 2809779 : added_notes_insn = place;
15008 : }
15009 :
15010 13050690 : if (place2)
15011 : {
15012 0 : add_shallow_copy_of_reg_note (place2, note);
15013 :
15014 : /* Set added_notes_insn to the earliest insn we added a note to. */
15015 0 : if (added_notes_insn == 0
15016 0 : || DF_INSN_LUID (added_notes_insn) > DF_INSN_LUID (place2))
15017 0 : added_notes_insn = place2;
15018 : }
15019 : }
15020 9833396 : }
15021 :
15022 : /* Similarly to above, distribute the LOG_LINKS that used to be present on
15023 : I3, I2, and I1 to new locations. This is also called to add a link
15024 : pointing at I3 when I3's destination is changed.
15025 :
15026 : If START is nonnull and an insn, we know that the next location for each
15027 : link is no earlier than START. LIMIT is the maximum number of nondebug
15028 : instructions that can be scanned when looking for the next use of a
15029 : definition. */
15030 :
15031 : static void
15032 16025632 : distribute_links (struct insn_link *links, rtx_insn *start, int limit)
15033 : {
15034 16025632 : struct insn_link *link, *next_link;
15035 :
15036 23581101 : for (link = links; link; link = next_link)
15037 : {
15038 7555469 : rtx_insn *place = 0;
15039 7555469 : rtx_insn *insn;
15040 7555469 : rtx set, reg;
15041 :
15042 7555469 : next_link = link->next;
15043 :
15044 : /* If the insn that this link points to is a NOTE, ignore it. */
15045 7555469 : if (NOTE_P (link->insn))
15046 4032814 : continue;
15047 :
15048 3522655 : set = 0;
15049 3522655 : rtx pat = PATTERN (link->insn);
15050 3522655 : if (GET_CODE (pat) == SET)
15051 : set = pat;
15052 627321 : else if (GET_CODE (pat) == PARALLEL)
15053 : {
15054 : int i;
15055 741910 : for (i = 0; i < XVECLEN (pat, 0); i++)
15056 : {
15057 737730 : set = XVECEXP (pat, 0, i);
15058 737730 : if (GET_CODE (set) != SET)
15059 4189 : continue;
15060 :
15061 733541 : reg = SET_DEST (set);
15062 733541 : while (GET_CODE (reg) == ZERO_EXTRACT
15063 742051 : || GET_CODE (reg) == STRICT_LOW_PART
15064 1484022 : || GET_CODE (reg) == SUBREG)
15065 8517 : reg = XEXP (reg, 0);
15066 :
15067 733541 : if (!REG_P (reg))
15068 44126 : continue;
15069 :
15070 689415 : if (REGNO (reg) == link->regno)
15071 : break;
15072 : }
15073 625532 : if (i == XVECLEN (pat, 0))
15074 4180 : continue;
15075 : }
15076 : else
15077 1789 : continue;
15078 :
15079 3516686 : reg = SET_DEST (set);
15080 :
15081 3516686 : while (GET_CODE (reg) == ZERO_EXTRACT
15082 3539061 : || GET_CODE (reg) == STRICT_LOW_PART
15083 7078252 : || GET_CODE (reg) == SUBREG)
15084 22934 : reg = XEXP (reg, 0);
15085 :
15086 3516686 : if (reg == pc_rtx)
15087 543 : continue;
15088 :
15089 : /* A LOG_LINK is defined as being placed on the first insn that uses
15090 : a register and points to the insn that sets the register. Start
15091 : searching at the next insn after the target of the link and stop
15092 : when we reach a set of the register or the end of the basic block.
15093 :
15094 : Note that this correctly handles the link that used to point from
15095 : I3 to I2. Also note that not much searching is typically done here
15096 : since most links don't point very far away. */
15097 :
15098 3516143 : int count = 0;
15099 3516143 : insn = start;
15100 3516143 : if (!insn || NOTE_P (insn))
15101 3466041 : insn = NEXT_INSN (link->insn);
15102 : else
15103 50102 : count = link->insn_count;
15104 11154842 : for (;
15105 14670985 : (insn && (this_basic_block->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)
15106 10119618 : || BB_HEAD (this_basic_block->next_bb) != insn));
15107 11154842 : insn = NEXT_INSN (insn))
15108 14630741 : if (DEBUG_INSN_P (insn))
15109 2810120 : continue;
15110 11820621 : else if (INSN_P (insn) && reg_overlap_mentioned_p (reg, PATTERN (insn)))
15111 : {
15112 3320482 : if (reg_referenced_p (reg, PATTERN (insn)))
15113 3320482 : place = insn;
15114 : break;
15115 : }
15116 8500139 : else if (CALL_P (insn)
15117 8500139 : && find_reg_fusage (insn, USE, reg))
15118 : {
15119 : place = insn;
15120 : break;
15121 : }
15122 8344821 : else if (INSN_P (insn) && reg_set_p (reg, insn))
15123 : break;
15124 8344722 : else if (count >= limit)
15125 : break;
15126 : else
15127 8344722 : count += 1;
15128 3516143 : link->insn_count = count;
15129 :
15130 : /* If we found a place to put the link, place it there unless there
15131 : is already a link to the same insn as LINK at that point. */
15132 :
15133 3516143 : if (place)
15134 : {
15135 3475800 : struct insn_link *link2;
15136 :
15137 4480848 : FOR_EACH_LOG_LINK (link2, place)
15138 1023373 : if (link2->insn == link->insn && link2->regno == link->regno)
15139 : break;
15140 :
15141 3475800 : if (link2 == NULL)
15142 : {
15143 3457475 : link->next = LOG_LINKS (place);
15144 3457475 : LOG_LINKS (place) = link;
15145 :
15146 : /* Set added_links_insn to the earliest insn we added a
15147 : link to. */
15148 3457475 : if (added_links_insn == 0
15149 3457475 : || DF_INSN_LUID (added_links_insn) > DF_INSN_LUID (place))
15150 2741491 : added_links_insn = place;
15151 : }
15152 : }
15153 : }
15154 16025632 : }
15155 :
15156 : /* Check for any register or memory mentioned in EQUIV that is not
15157 : mentioned in EXPR. This is used to restrict EQUIV to "specializations"
15158 : of EXPR where some registers may have been replaced by constants. */
15159 :
15160 : static bool
15161 2545606 : unmentioned_reg_p (rtx equiv, rtx expr)
15162 : {
15163 2545606 : subrtx_iterator::array_type array;
15164 6765635 : FOR_EACH_SUBRTX (iter, array, equiv, NONCONST)
15165 : {
15166 5534467 : const_rtx x = *iter;
15167 3794578 : if ((REG_P (x) || MEM_P (x))
15168 5923305 : && !reg_mentioned_p (x, expr))
15169 1314438 : return true;
15170 : }
15171 1231168 : return false;
15172 2545606 : }
15173 :
15174 : /* Make pseudo-to-pseudo copies after every hard-reg-to-pseudo-copy, because
15175 : the reg-to-reg copy can usefully combine with later instructions, but we
15176 : do not want to combine the hard reg into later instructions, for that
15177 : restricts register allocation. */
15178 : static void
15179 1043685 : make_more_copies (void)
15180 : {
15181 1043685 : basic_block bb;
15182 :
15183 11447172 : FOR_EACH_BB_FN (bb, cfun)
15184 : {
15185 10403487 : rtx_insn *insn;
15186 :
15187 134604718 : FOR_BB_INSNS (bb, insn)
15188 : {
15189 124201231 : if (!NONDEBUG_INSN_P (insn))
15190 65385897 : continue;
15191 :
15192 58815334 : rtx set = single_set (insn);
15193 58815334 : if (!set)
15194 3973455 : continue;
15195 :
15196 54841879 : rtx dest = SET_DEST (set);
15197 54841879 : if (!(REG_P (dest) && !HARD_REGISTER_P (dest)))
15198 31357994 : continue;
15199 :
15200 23483885 : rtx src = SET_SRC (set);
15201 23483885 : if (!(REG_P (src) && HARD_REGISTER_P (src)))
15202 20505842 : continue;
15203 2978043 : if (TEST_HARD_REG_BIT (fixed_reg_set, REGNO (src)))
15204 9523 : continue;
15205 :
15206 2968520 : rtx new_reg = gen_reg_rtx (GET_MODE (dest));
15207 :
15208 : /* The "original" pseudo copies have important attributes
15209 : attached, like pointerness. We want that for these copies
15210 : too, for use by insn recognition and later passes. */
15211 2968520 : set_reg_attrs_from_value (new_reg, dest);
15212 :
15213 2968520 : rtx_insn *new_insn = gen_move_insn (new_reg, src);
15214 2968520 : SET_SRC (set) = new_reg;
15215 2968520 : emit_insn_before (new_insn, insn);
15216 2968520 : df_insn_rescan (insn);
15217 : }
15218 : }
15219 1043685 : }
15220 :
15221 : /* Try combining insns through substitution. */
15222 : static void
15223 1043685 : rest_of_handle_combine (void)
15224 : {
15225 1043685 : make_more_copies ();
15226 :
15227 1043685 : df_set_flags (DF_LR_RUN_DCE + DF_DEFER_INSN_RESCAN);
15228 1043685 : df_note_add_problem ();
15229 1043685 : df_analyze ();
15230 :
15231 1043685 : regstat_init_n_sets_and_refs ();
15232 1043685 : reg_n_sets_max = max_reg_num ();
15233 :
15234 1043685 : bool rebuild_jump_labels_after_combine
15235 1043685 : = combine_instructions (get_insns (), max_reg_num ());
15236 :
15237 : /* Combining insns may have turned an indirect jump into a
15238 : direct jump. Rebuild the JUMP_LABEL fields of jumping
15239 : instructions. */
15240 1043685 : if (rebuild_jump_labels_after_combine)
15241 : {
15242 2497 : if (dom_info_available_p (CDI_DOMINATORS))
15243 0 : free_dominance_info (CDI_DOMINATORS);
15244 2497 : timevar_push (TV_JUMP);
15245 2497 : rebuild_jump_labels (get_insns ());
15246 2497 : cleanup_cfg (0);
15247 2497 : timevar_pop (TV_JUMP);
15248 : }
15249 :
15250 1043685 : regstat_free_n_sets_and_refs ();
15251 1043685 : }
15252 :
15253 : namespace {
15254 :
15255 : const pass_data pass_data_combine =
15256 : {
15257 : RTL_PASS, /* type */
15258 : "combine", /* name */
15259 : OPTGROUP_NONE, /* optinfo_flags */
15260 : TV_COMBINE, /* tv_id */
15261 : PROP_cfglayout, /* properties_required */
15262 : 0, /* properties_provided */
15263 : 0, /* properties_destroyed */
15264 : 0, /* todo_flags_start */
15265 : TODO_df_finish, /* todo_flags_finish */
15266 : };
15267 :
15268 : class pass_combine : public rtl_opt_pass
15269 : {
15270 : public:
15271 285722 : pass_combine (gcc::context *ctxt)
15272 571444 : : rtl_opt_pass (pass_data_combine, ctxt)
15273 : {}
15274 :
15275 : /* opt_pass methods: */
15276 1471370 : bool gate (function *) final override { return (optimize > 0); }
15277 1043685 : unsigned int execute (function *) final override
15278 : {
15279 1043685 : rest_of_handle_combine ();
15280 1043685 : return 0;
15281 : }
15282 :
15283 : }; // class pass_combine
15284 :
15285 : } // anon namespace
15286 :
15287 : rtl_opt_pass *
15288 285722 : make_pass_combine (gcc::context *ctxt)
15289 : {
15290 285722 : return new pass_combine (ctxt);
15291 : }
|
