re PR rtl-optimization/52060 (Invalid constant simplification in combine with paralle...
[official-gcc.git] / gcc / combine.c
blobd06de04041e8f227e05eb472f14ca7ef7ee305b4
1 /* Optimize by combining instructions for GNU compiler.
2 Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010,
4 2011, 2012 Free Software Foundation, Inc.
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 /* This module is essentially the "combiner" phase of the U. of Arizona
23 Portable Optimizer, but redone to work on our list-structured
24 representation for RTL instead of their string representation.
26 The LOG_LINKS of each insn identify the most recent assignment
27 to each REG used in the insn. It is a list of previous insns,
28 each of which contains a SET for a REG that is used in this insn
29 and not used or set in between. LOG_LINKs never cross basic blocks.
30 They were set up by the preceding pass (lifetime analysis).
32 We try to combine each pair of insns joined by a logical link.
33 We also try to combine triples of insns A, B and C when
34 C has a link back to B and B has a link back to A.
36 LOG_LINKS does not have links for use of the CC0. They don't
37 need to, because the insn that sets the CC0 is always immediately
38 before the insn that tests it. So we always regard a branch
39 insn as having a logical link to the preceding insn. The same is true
40 for an insn explicitly using CC0.
42 We check (with use_crosses_set_p) to avoid combining in such a way
43 as to move a computation to a place where its value would be different.
45 Combination is done by mathematically substituting the previous
46 insn(s) values for the regs they set into the expressions in
47 the later insns that refer to these regs. If the result is a valid insn
48 for our target machine, according to the machine description,
49 we install it, delete the earlier insns, and update the data flow
50 information (LOG_LINKS and REG_NOTES) for what we did.
52 There are a few exceptions where the dataflow information isn't
53 completely updated (however this is only a local issue since it is
54 regenerated before the next pass that uses it):
56 - reg_live_length is not updated
57 - reg_n_refs is not adjusted in the rare case when a register is
58 no longer required in a computation
59 - there are extremely rare cases (see distribute_notes) when a
60 REG_DEAD note is lost
61 - a LOG_LINKS entry that refers to an insn with multiple SETs may be
62 removed because there is no way to know which register it was
63 linking
65 To simplify substitution, we combine only when the earlier insn(s)
66 consist of only a single assignment. To simplify updating afterward,
67 we never combine when a subroutine call appears in the middle.
69 Since we do not represent assignments to CC0 explicitly except when that
70 is all an insn does, there is no LOG_LINKS entry in an insn that uses
71 the condition code for the insn that set the condition code.
72 Fortunately, these two insns must be consecutive.
73 Therefore, every JUMP_INSN is taken to have an implicit logical link
74 to the preceding insn. This is not quite right, since non-jumps can
75 also use the condition code; but in practice such insns would not
76 combine anyway. */
78 #include "config.h"
79 #include "system.h"
80 #include "coretypes.h"
81 #include "tm.h"
82 #include "rtl.h"
83 #include "tree.h"
84 #include "tm_p.h"
85 #include "flags.h"
86 #include "regs.h"
87 #include "hard-reg-set.h"
88 #include "basic-block.h"
89 #include "insn-config.h"
90 #include "function.h"
91 /* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
92 #include "expr.h"
93 #include "insn-attr.h"
94 #include "recog.h"
95 #include "diagnostic-core.h"
96 #include "target.h"
97 #include "optabs.h"
98 #include "insn-codes.h"
99 #include "rtlhooks-def.h"
100 /* Include output.h for dump_file. */
101 #include "output.h"
102 #include "params.h"
103 #include "timevar.h"
104 #include "tree-pass.h"
105 #include "df.h"
106 #include "cgraph.h"
107 #include "obstack.h"
109 /* Number of attempts to combine instructions in this function. */
111 static int combine_attempts;
113 /* Number of attempts that got as far as substitution in this function. */
115 static int combine_merges;
117 /* Number of instructions combined with added SETs in this function. */
119 static int combine_extras;
121 /* Number of instructions combined in this function. */
123 static int combine_successes;
125 /* Totals over entire compilation. */
127 static int total_attempts, total_merges, total_extras, total_successes;
129 /* combine_instructions may try to replace the right hand side of the
130 second instruction with the value of an associated REG_EQUAL note
131 before throwing it at try_combine. That is problematic when there
132 is a REG_DEAD note for a register used in the old right hand side
133 and can cause distribute_notes to do wrong things. This is the
134 second instruction if it has been so modified, null otherwise. */
136 static rtx i2mod;
138 /* When I2MOD is nonnull, this is a copy of the old right hand side. */
140 static rtx i2mod_old_rhs;
142 /* When I2MOD is nonnull, this is a copy of the new right hand side. */
144 static rtx i2mod_new_rhs;
146 typedef struct reg_stat_struct {
147 /* Record last point of death of (hard or pseudo) register n. */
148 rtx last_death;
150 /* Record last point of modification of (hard or pseudo) register n. */
151 rtx last_set;
153 /* The next group of fields allows the recording of the last value assigned
154 to (hard or pseudo) register n. We use this information to see if an
155 operation being processed is redundant given a prior operation performed
156 on the register. For example, an `and' with a constant is redundant if
157 all the zero bits are already known to be turned off.
159 We use an approach similar to that used by cse, but change it in the
160 following ways:
162 (1) We do not want to reinitialize at each label.
163 (2) It is useful, but not critical, to know the actual value assigned
164 to a register. Often just its form is helpful.
166 Therefore, we maintain the following fields:
168 last_set_value the last value assigned
169 last_set_label records the value of label_tick when the
170 register was assigned
171 last_set_table_tick records the value of label_tick when a
172 value using the register is assigned
173 last_set_invalid set to nonzero when it is not valid
174 to use the value of this register in some
175 register's value
177 To understand the usage of these tables, it is important to understand
178 the distinction between the value in last_set_value being valid and
179 the register being validly contained in some other expression in the
180 table.
182 (The next two parameters are out of date).
184 reg_stat[i].last_set_value is valid if it is nonzero, and either
185 reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
187 Register I may validly appear in any expression returned for the value
188 of another register if reg_n_sets[i] is 1. It may also appear in the
189 value for register J if reg_stat[j].last_set_invalid is zero, or
190 reg_stat[i].last_set_label < reg_stat[j].last_set_label.
192 If an expression is found in the table containing a register which may
193 not validly appear in an expression, the register is replaced by
194 something that won't match, (clobber (const_int 0)). */
196 /* Record last value assigned to (hard or pseudo) register n. */
198 rtx last_set_value;
200 /* Record the value of label_tick when an expression involving register n
201 is placed in last_set_value. */
203 int last_set_table_tick;
205 /* Record the value of label_tick when the value for register n is placed in
206 last_set_value. */
208 int last_set_label;
210 /* These fields are maintained in parallel with last_set_value and are
211 used to store the mode in which the register was last set, the bits
212 that were known to be zero when it was last set, and the number of
213 sign bits copies it was known to have when it was last set. */
215 unsigned HOST_WIDE_INT last_set_nonzero_bits;
216 char last_set_sign_bit_copies;
217 ENUM_BITFIELD(machine_mode) last_set_mode : 8;
219 /* Set nonzero if references to register n in expressions should not be
220 used. last_set_invalid is set nonzero when this register is being
221 assigned to and last_set_table_tick == label_tick. */
223 char last_set_invalid;
225 /* Some registers that are set more than once and used in more than one
226 basic block are nevertheless always set in similar ways. For example,
227 a QImode register may be loaded from memory in two places on a machine
228 where byte loads zero extend.
230 We record in the following fields if a register has some leading bits
231 that are always equal to the sign bit, and what we know about the
232 nonzero bits of a register, specifically which bits are known to be
233 zero.
235 If an entry is zero, it means that we don't know anything special. */
237 unsigned char sign_bit_copies;
239 unsigned HOST_WIDE_INT nonzero_bits;
241 /* Record the value of the label_tick when the last truncation
242 happened. The field truncated_to_mode is only valid if
243 truncation_label == label_tick. */
245 int truncation_label;
247 /* Record the last truncation seen for this register. If truncation
248 is not a nop to this mode we might be able to save an explicit
249 truncation if we know that value already contains a truncated
250 value. */
252 ENUM_BITFIELD(machine_mode) truncated_to_mode : 8;
253 } reg_stat_type;
255 DEF_VEC_O(reg_stat_type);
256 DEF_VEC_ALLOC_O(reg_stat_type,heap);
258 static VEC(reg_stat_type,heap) *reg_stat;
260 /* Record the luid of the last insn that invalidated memory
261 (anything that writes memory, and subroutine calls, but not pushes). */
263 static int mem_last_set;
265 /* Record the luid of the last CALL_INSN
266 so we can tell whether a potential combination crosses any calls. */
268 static int last_call_luid;
270 /* When `subst' is called, this is the insn that is being modified
271 (by combining in a previous insn). The PATTERN of this insn
272 is still the old pattern partially modified and it should not be
273 looked at, but this may be used to examine the successors of the insn
274 to judge whether a simplification is valid. */
276 static rtx subst_insn;
278 /* This is the lowest LUID that `subst' is currently dealing with.
279 get_last_value will not return a value if the register was set at or
280 after this LUID. If not for this mechanism, we could get confused if
281 I2 or I1 in try_combine were an insn that used the old value of a register
282 to obtain a new value. In that case, we might erroneously get the
283 new value of the register when we wanted the old one. */
285 static int subst_low_luid;
287 /* This contains any hard registers that are used in newpat; reg_dead_at_p
288 must consider all these registers to be always live. */
290 static HARD_REG_SET newpat_used_regs;
292 /* This is an insn to which a LOG_LINKS entry has been added. If this
293 insn is the earlier than I2 or I3, combine should rescan starting at
294 that location. */
296 static rtx added_links_insn;
298 /* Basic block in which we are performing combines. */
299 static basic_block this_basic_block;
300 static bool optimize_this_for_speed_p;
303 /* Length of the currently allocated uid_insn_cost array. */
305 static int max_uid_known;
307 /* The following array records the insn_rtx_cost for every insn
308 in the instruction stream. */
310 static int *uid_insn_cost;
312 /* The following array records the LOG_LINKS for every insn in the
313 instruction stream as struct insn_link pointers. */
315 struct insn_link {
316 rtx insn;
317 struct insn_link *next;
320 static struct insn_link **uid_log_links;
322 #define INSN_COST(INSN) (uid_insn_cost[INSN_UID (INSN)])
323 #define LOG_LINKS(INSN) (uid_log_links[INSN_UID (INSN)])
325 #define FOR_EACH_LOG_LINK(L, INSN) \
326 for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
328 /* Links for LOG_LINKS are allocated from this obstack. */
330 static struct obstack insn_link_obstack;
332 /* Allocate a link. */
334 static inline struct insn_link *
335 alloc_insn_link (rtx insn, struct insn_link *next)
337 struct insn_link *l
338 = (struct insn_link *) obstack_alloc (&insn_link_obstack,
339 sizeof (struct insn_link));
340 l->insn = insn;
341 l->next = next;
342 return l;
345 /* Incremented for each basic block. */
347 static int label_tick;
349 /* Reset to label_tick for each extended basic block in scanning order. */
351 static int label_tick_ebb_start;
353 /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
354 largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
356 static enum machine_mode nonzero_bits_mode;
358 /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
359 be safely used. It is zero while computing them and after combine has
360 completed. This former test prevents propagating values based on
361 previously set values, which can be incorrect if a variable is modified
362 in a loop. */
364 static int nonzero_sign_valid;
367 /* Record one modification to rtl structure
368 to be undone by storing old_contents into *where. */
370 enum undo_kind { UNDO_RTX, UNDO_INT, UNDO_MODE, UNDO_LINKS };
372 struct undo
374 struct undo *next;
375 enum undo_kind kind;
376 union { rtx r; int i; enum machine_mode m; struct insn_link *l; } old_contents;
377 union { rtx *r; int *i; struct insn_link **l; } where;
380 /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
381 num_undo says how many are currently recorded.
383 other_insn is nonzero if we have modified some other insn in the process
384 of working on subst_insn. It must be verified too. */
386 struct undobuf
388 struct undo *undos;
389 struct undo *frees;
390 rtx other_insn;
393 static struct undobuf undobuf;
395 /* Number of times the pseudo being substituted for
396 was found and replaced. */
398 static int n_occurrences;
400 static rtx reg_nonzero_bits_for_combine (const_rtx, enum machine_mode, const_rtx,
401 enum machine_mode,
402 unsigned HOST_WIDE_INT,
403 unsigned HOST_WIDE_INT *);
404 static rtx reg_num_sign_bit_copies_for_combine (const_rtx, enum machine_mode, const_rtx,
405 enum machine_mode,
406 unsigned int, unsigned int *);
407 static void do_SUBST (rtx *, rtx);
408 static void do_SUBST_INT (int *, int);
409 static void init_reg_last (void);
410 static void setup_incoming_promotions (rtx);
411 static void set_nonzero_bits_and_sign_copies (rtx, const_rtx, void *);
412 static int cant_combine_insn_p (rtx);
413 static int can_combine_p (rtx, rtx, rtx, rtx, rtx, rtx, rtx *, rtx *);
414 static int combinable_i3pat (rtx, rtx *, rtx, rtx, rtx, int, int, rtx *);
415 static int contains_muldiv (rtx);
416 static rtx try_combine (rtx, rtx, rtx, rtx, int *, rtx);
417 static void undo_all (void);
418 static void undo_commit (void);
419 static rtx *find_split_point (rtx *, rtx, bool);
420 static rtx subst (rtx, rtx, rtx, int, int, int);
421 static rtx combine_simplify_rtx (rtx, enum machine_mode, int, int);
422 static rtx simplify_if_then_else (rtx);
423 static rtx simplify_set (rtx);
424 static rtx simplify_logical (rtx);
425 static rtx expand_compound_operation (rtx);
426 static const_rtx expand_field_assignment (const_rtx);
427 static rtx make_extraction (enum machine_mode, rtx, HOST_WIDE_INT,
428 rtx, unsigned HOST_WIDE_INT, int, int, int);
429 static rtx extract_left_shift (rtx, int);
430 static rtx make_compound_operation (rtx, enum rtx_code);
431 static int get_pos_from_mask (unsigned HOST_WIDE_INT,
432 unsigned HOST_WIDE_INT *);
433 static rtx canon_reg_for_combine (rtx, rtx);
434 static rtx force_to_mode (rtx, enum machine_mode,
435 unsigned HOST_WIDE_INT, int);
436 static rtx if_then_else_cond (rtx, rtx *, rtx *);
437 static rtx known_cond (rtx, enum rtx_code, rtx, rtx);
438 static int rtx_equal_for_field_assignment_p (rtx, rtx);
439 static rtx make_field_assignment (rtx);
440 static rtx apply_distributive_law (rtx);
441 static rtx distribute_and_simplify_rtx (rtx, int);
442 static rtx simplify_and_const_int_1 (enum machine_mode, rtx,
443 unsigned HOST_WIDE_INT);
444 static rtx simplify_and_const_int (rtx, enum machine_mode, rtx,
445 unsigned HOST_WIDE_INT);
446 static int merge_outer_ops (enum rtx_code *, HOST_WIDE_INT *, enum rtx_code,
447 HOST_WIDE_INT, enum machine_mode, int *);
448 static rtx simplify_shift_const_1 (enum rtx_code, enum machine_mode, rtx, int);
449 static rtx simplify_shift_const (rtx, enum rtx_code, enum machine_mode, rtx,
450 int);
451 static int recog_for_combine (rtx *, rtx, rtx *);
452 static rtx gen_lowpart_for_combine (enum machine_mode, rtx);
453 static enum rtx_code simplify_compare_const (enum rtx_code, rtx, rtx *);
454 static enum rtx_code simplify_comparison (enum rtx_code, rtx *, rtx *);
455 static void update_table_tick (rtx);
456 static void record_value_for_reg (rtx, rtx, rtx);
457 static void check_promoted_subreg (rtx, rtx);
458 static void record_dead_and_set_regs_1 (rtx, const_rtx, void *);
459 static void record_dead_and_set_regs (rtx);
460 static int get_last_value_validate (rtx *, rtx, int, int);
461 static rtx get_last_value (const_rtx);
462 static int use_crosses_set_p (const_rtx, int);
463 static void reg_dead_at_p_1 (rtx, const_rtx, void *);
464 static int reg_dead_at_p (rtx, rtx);
465 static void move_deaths (rtx, rtx, int, rtx, rtx *);
466 static int reg_bitfield_target_p (rtx, rtx);
467 static void distribute_notes (rtx, rtx, rtx, rtx, rtx, rtx, rtx);
468 static void distribute_links (struct insn_link *);
469 static void mark_used_regs_combine (rtx);
470 static void record_promoted_value (rtx, rtx);
471 static int unmentioned_reg_p_1 (rtx *, void *);
472 static bool unmentioned_reg_p (rtx, rtx);
473 static int record_truncated_value (rtx *, void *);
474 static void record_truncated_values (rtx *, void *);
475 static bool reg_truncated_to_mode (enum machine_mode, const_rtx);
476 static rtx gen_lowpart_or_truncate (enum machine_mode, rtx);
479 /* It is not safe to use ordinary gen_lowpart in combine.
480 See comments in gen_lowpart_for_combine. */
481 #undef RTL_HOOKS_GEN_LOWPART
482 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
484 /* Our implementation of gen_lowpart never emits a new pseudo. */
485 #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
486 #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
488 #undef RTL_HOOKS_REG_NONZERO_REG_BITS
489 #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
491 #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
492 #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
494 #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
495 #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
497 static const struct rtl_hooks combine_rtl_hooks = RTL_HOOKS_INITIALIZER;
500 /* Try to split PATTERN found in INSN. This returns NULL_RTX if
501 PATTERN can not be split. Otherwise, it returns an insn sequence.
502 This is a wrapper around split_insns which ensures that the
503 reg_stat vector is made larger if the splitter creates a new
504 register. */
506 static rtx
507 combine_split_insns (rtx pattern, rtx insn)
509 rtx ret;
510 unsigned int nregs;
512 ret = split_insns (pattern, insn);
513 nregs = max_reg_num ();
514 if (nregs > VEC_length (reg_stat_type, reg_stat))
515 VEC_safe_grow_cleared (reg_stat_type, heap, reg_stat, nregs);
516 return ret;
519 /* This is used by find_single_use to locate an rtx in LOC that
520 contains exactly one use of DEST, which is typically either a REG
521 or CC0. It returns a pointer to the innermost rtx expression
522 containing DEST. Appearances of DEST that are being used to
523 totally replace it are not counted. */
525 static rtx *
526 find_single_use_1 (rtx dest, rtx *loc)
528 rtx x = *loc;
529 enum rtx_code code = GET_CODE (x);
530 rtx *result = NULL;
531 rtx *this_result;
532 int i;
533 const char *fmt;
535 switch (code)
537 case CONST_INT:
538 case CONST:
539 case LABEL_REF:
540 case SYMBOL_REF:
541 case CONST_DOUBLE:
542 case CONST_VECTOR:
543 case CLOBBER:
544 return 0;
546 case SET:
547 /* If the destination is anything other than CC0, PC, a REG or a SUBREG
548 of a REG that occupies all of the REG, the insn uses DEST if
549 it is mentioned in the destination or the source. Otherwise, we
550 need just check the source. */
551 if (GET_CODE (SET_DEST (x)) != CC0
552 && GET_CODE (SET_DEST (x)) != PC
553 && !REG_P (SET_DEST (x))
554 && ! (GET_CODE (SET_DEST (x)) == SUBREG
555 && REG_P (SUBREG_REG (SET_DEST (x)))
556 && (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (SET_DEST (x))))
557 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
558 == ((GET_MODE_SIZE (GET_MODE (SET_DEST (x)))
559 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD))))
560 break;
562 return find_single_use_1 (dest, &SET_SRC (x));
564 case MEM:
565 case SUBREG:
566 return find_single_use_1 (dest, &XEXP (x, 0));
568 default:
569 break;
572 /* If it wasn't one of the common cases above, check each expression and
573 vector of this code. Look for a unique usage of DEST. */
575 fmt = GET_RTX_FORMAT (code);
576 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
578 if (fmt[i] == 'e')
580 if (dest == XEXP (x, i)
581 || (REG_P (dest) && REG_P (XEXP (x, i))
582 && REGNO (dest) == REGNO (XEXP (x, i))))
583 this_result = loc;
584 else
585 this_result = find_single_use_1 (dest, &XEXP (x, i));
587 if (result == NULL)
588 result = this_result;
589 else if (this_result)
590 /* Duplicate usage. */
591 return NULL;
593 else if (fmt[i] == 'E')
595 int j;
597 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
599 if (XVECEXP (x, i, j) == dest
600 || (REG_P (dest)
601 && REG_P (XVECEXP (x, i, j))
602 && REGNO (XVECEXP (x, i, j)) == REGNO (dest)))
603 this_result = loc;
604 else
605 this_result = find_single_use_1 (dest, &XVECEXP (x, i, j));
607 if (result == NULL)
608 result = this_result;
609 else if (this_result)
610 return NULL;
615 return result;
619 /* See if DEST, produced in INSN, is used only a single time in the
620 sequel. If so, return a pointer to the innermost rtx expression in which
621 it is used.
623 If PLOC is nonzero, *PLOC is set to the insn containing the single use.
625 If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
626 care about REG_DEAD notes or LOG_LINKS.
628 Otherwise, we find the single use by finding an insn that has a
629 LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
630 only referenced once in that insn, we know that it must be the first
631 and last insn referencing DEST. */
633 static rtx *
634 find_single_use (rtx dest, rtx insn, rtx *ploc)
636 basic_block bb;
637 rtx next;
638 rtx *result;
639 struct insn_link *link;
641 #ifdef HAVE_cc0
642 if (dest == cc0_rtx)
644 next = NEXT_INSN (insn);
645 if (next == 0
646 || (!NONJUMP_INSN_P (next) && !JUMP_P (next)))
647 return 0;
649 result = find_single_use_1 (dest, &PATTERN (next));
650 if (result && ploc)
651 *ploc = next;
652 return result;
654 #endif
656 if (!REG_P (dest))
657 return 0;
659 bb = BLOCK_FOR_INSN (insn);
660 for (next = NEXT_INSN (insn);
661 next && BLOCK_FOR_INSN (next) == bb;
662 next = NEXT_INSN (next))
663 if (INSN_P (next) && dead_or_set_p (next, dest))
665 FOR_EACH_LOG_LINK (link, next)
666 if (link->insn == insn)
667 break;
669 if (link)
671 result = find_single_use_1 (dest, &PATTERN (next));
672 if (ploc)
673 *ploc = next;
674 return result;
678 return 0;
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. */
687 static void
688 do_SUBST (rtx *into, rtx newval)
690 struct undo *buf;
691 rtx oldval = *into;
693 if (oldval == newval)
694 return;
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 if (GET_MODE_CLASS (GET_MODE (oldval)) == MODE_INT
702 && CONST_INT_P (newval))
704 /* Sanity check that we're replacing oldval with a CONST_INT
705 that is a valid sign-extension for the original mode. */
706 gcc_assert (INTVAL (newval)
707 == trunc_int_for_mode (INTVAL (newval), GET_MODE (oldval)));
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 gcc_assert (!(GET_CODE (oldval) == SUBREG
716 && CONST_INT_P (SUBREG_REG (oldval))));
717 gcc_assert (!(GET_CODE (oldval) == ZERO_EXTEND
718 && CONST_INT_P (XEXP (oldval, 0))));
721 if (undobuf.frees)
722 buf = undobuf.frees, undobuf.frees = buf->next;
723 else
724 buf = XNEW (struct undo);
726 buf->kind = UNDO_RTX;
727 buf->where.r = into;
728 buf->old_contents.r = oldval;
729 *into = newval;
731 buf->next = undobuf.undos, undobuf.undos = buf;
734 #define SUBST(INTO, NEWVAL) do_SUBST(&(INTO), (NEWVAL))
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. */
740 static void
741 do_SUBST_INT (int *into, int newval)
743 struct undo *buf;
744 int oldval = *into;
746 if (oldval == newval)
747 return;
749 if (undobuf.frees)
750 buf = undobuf.frees, undobuf.frees = buf->next;
751 else
752 buf = XNEW (struct undo);
754 buf->kind = UNDO_INT;
755 buf->where.i = into;
756 buf->old_contents.i = oldval;
757 *into = newval;
759 buf->next = undobuf.undos, undobuf.undos = buf;
762 #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT(&(INTO), (NEWVAL))
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. */
769 static void
770 do_SUBST_MODE (rtx *into, enum machine_mode newval)
772 struct undo *buf;
773 enum machine_mode oldval = GET_MODE (*into);
775 if (oldval == newval)
776 return;
778 if (undobuf.frees)
779 buf = undobuf.frees, undobuf.frees = buf->next;
780 else
781 buf = XNEW (struct undo);
783 buf->kind = UNDO_MODE;
784 buf->where.r = into;
785 buf->old_contents.m = oldval;
786 adjust_reg_mode (*into, newval);
788 buf->next = undobuf.undos, undobuf.undos = buf;
791 #define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE(&(INTO), (NEWVAL))
793 #ifndef HAVE_cc0
794 /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
796 static void
797 do_SUBST_LINK (struct insn_link **into, struct insn_link *newval)
799 struct undo *buf;
800 struct insn_link * oldval = *into;
802 if (oldval == newval)
803 return;
805 if (undobuf.frees)
806 buf = undobuf.frees, undobuf.frees = buf->next;
807 else
808 buf = XNEW (struct undo);
810 buf->kind = UNDO_LINKS;
811 buf->where.l = into;
812 buf->old_contents.l = oldval;
813 *into = newval;
815 buf->next = undobuf.undos, undobuf.undos = buf;
818 #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
819 #endif
821 /* Subroutine of try_combine. Determine whether the replacement patterns
822 NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to insn_rtx_cost
823 than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
824 that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
825 undobuf.other_insn may also both be NULL_RTX. Return false if the cost
826 of all the instructions can be estimated and the replacements are more
827 expensive than the original sequence. */
829 static bool
830 combine_validate_cost (rtx i0, rtx i1, rtx i2, rtx i3, rtx newpat,
831 rtx newi2pat, rtx newotherpat)
833 int i0_cost, i1_cost, i2_cost, i3_cost;
834 int new_i2_cost, new_i3_cost;
835 int old_cost, new_cost;
837 /* Lookup the original insn_rtx_costs. */
838 i2_cost = INSN_COST (i2);
839 i3_cost = INSN_COST (i3);
841 if (i1)
843 i1_cost = INSN_COST (i1);
844 if (i0)
846 i0_cost = INSN_COST (i0);
847 old_cost = (i0_cost > 0 && i1_cost > 0 && i2_cost > 0 && i3_cost > 0
848 ? i0_cost + i1_cost + i2_cost + i3_cost : 0);
850 else
852 old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0
853 ? i1_cost + i2_cost + i3_cost : 0);
854 i0_cost = 0;
857 else
859 old_cost = (i2_cost > 0 && i3_cost > 0) ? i2_cost + i3_cost : 0;
860 i1_cost = i0_cost = 0;
863 /* Calculate the replacement insn_rtx_costs. */
864 new_i3_cost = insn_rtx_cost (newpat, optimize_this_for_speed_p);
865 if (newi2pat)
867 new_i2_cost = insn_rtx_cost (newi2pat, optimize_this_for_speed_p);
868 new_cost = (new_i2_cost > 0 && new_i3_cost > 0)
869 ? new_i2_cost + new_i3_cost : 0;
871 else
873 new_cost = new_i3_cost;
874 new_i2_cost = 0;
877 if (undobuf.other_insn)
879 int old_other_cost, new_other_cost;
881 old_other_cost = INSN_COST (undobuf.other_insn);
882 new_other_cost = insn_rtx_cost (newotherpat, optimize_this_for_speed_p);
883 if (old_other_cost > 0 && new_other_cost > 0)
885 old_cost += old_other_cost;
886 new_cost += new_other_cost;
888 else
889 old_cost = 0;
892 /* Disallow this combination if both new_cost and old_cost are greater than
893 zero, and new_cost is greater than old cost. */
894 if (old_cost > 0 && new_cost > old_cost)
896 if (dump_file)
898 if (i0)
900 fprintf (dump_file,
901 "rejecting combination of insns %d, %d, %d and %d\n",
902 INSN_UID (i0), INSN_UID (i1), INSN_UID (i2),
903 INSN_UID (i3));
904 fprintf (dump_file, "original costs %d + %d + %d + %d = %d\n",
905 i0_cost, i1_cost, i2_cost, i3_cost, old_cost);
907 else if (i1)
909 fprintf (dump_file,
910 "rejecting combination of insns %d, %d and %d\n",
911 INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
912 fprintf (dump_file, "original costs %d + %d + %d = %d\n",
913 i1_cost, i2_cost, i3_cost, old_cost);
915 else
917 fprintf (dump_file,
918 "rejecting combination of insns %d and %d\n",
919 INSN_UID (i2), INSN_UID (i3));
920 fprintf (dump_file, "original costs %d + %d = %d\n",
921 i2_cost, i3_cost, old_cost);
924 if (newi2pat)
926 fprintf (dump_file, "replacement costs %d + %d = %d\n",
927 new_i2_cost, new_i3_cost, new_cost);
929 else
930 fprintf (dump_file, "replacement cost %d\n", new_cost);
933 return false;
936 /* Update the uid_insn_cost array with the replacement costs. */
937 INSN_COST (i2) = new_i2_cost;
938 INSN_COST (i3) = new_i3_cost;
939 if (i1)
941 INSN_COST (i1) = 0;
942 if (i0)
943 INSN_COST (i0) = 0;
946 return true;
950 /* Delete any insns that copy a register to itself. */
952 static void
953 delete_noop_moves (void)
955 rtx insn, next;
956 basic_block bb;
958 FOR_EACH_BB (bb)
960 for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
962 next = NEXT_INSN (insn);
963 if (INSN_P (insn) && noop_move_p (insn))
965 if (dump_file)
966 fprintf (dump_file, "deleting noop move %d\n", INSN_UID (insn));
968 delete_insn_and_edges (insn);
975 /* Fill in log links field for all insns. */
977 static void
978 create_log_links (void)
980 basic_block bb;
981 rtx *next_use, insn;
982 df_ref *def_vec, *use_vec;
984 next_use = XCNEWVEC (rtx, max_reg_num ());
986 /* Pass through each block from the end, recording the uses of each
987 register and establishing log links when def is encountered.
988 Note that we do not clear next_use array in order to save time,
989 so we have to test whether the use is in the same basic block as def.
991 There are a few cases below when we do not consider the definition or
992 usage -- these are taken from original flow.c did. Don't ask me why it is
993 done this way; I don't know and if it works, I don't want to know. */
995 FOR_EACH_BB (bb)
997 FOR_BB_INSNS_REVERSE (bb, insn)
999 if (!NONDEBUG_INSN_P (insn))
1000 continue;
1002 /* Log links are created only once. */
1003 gcc_assert (!LOG_LINKS (insn));
1005 for (def_vec = DF_INSN_DEFS (insn); *def_vec; def_vec++)
1007 df_ref def = *def_vec;
1008 int regno = DF_REF_REGNO (def);
1009 rtx use_insn;
1011 if (!next_use[regno])
1012 continue;
1014 /* Do not consider if it is pre/post modification in MEM. */
1015 if (DF_REF_FLAGS (def) & DF_REF_PRE_POST_MODIFY)
1016 continue;
1018 /* Do not make the log link for frame pointer. */
1019 if ((regno == FRAME_POINTER_REGNUM
1020 && (! reload_completed || frame_pointer_needed))
1021 #if !HARD_FRAME_POINTER_IS_FRAME_POINTER
1022 || (regno == HARD_FRAME_POINTER_REGNUM
1023 && (! reload_completed || frame_pointer_needed))
1024 #endif
1025 #if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
1026 || (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
1027 #endif
1029 continue;
1031 use_insn = next_use[regno];
1032 if (BLOCK_FOR_INSN (use_insn) == bb)
1034 /* flow.c claimed:
1036 We don't build a LOG_LINK for hard registers contained
1037 in ASM_OPERANDs. If these registers get replaced,
1038 we might wind up changing the semantics of the insn,
1039 even if reload can make what appear to be valid
1040 assignments later. */
1041 if (regno >= FIRST_PSEUDO_REGISTER
1042 || asm_noperands (PATTERN (use_insn)) < 0)
1044 /* Don't add duplicate links between instructions. */
1045 struct insn_link *links;
1046 FOR_EACH_LOG_LINK (links, use_insn)
1047 if (insn == links->insn)
1048 break;
1050 if (!links)
1051 LOG_LINKS (use_insn)
1052 = alloc_insn_link (insn, LOG_LINKS (use_insn));
1055 next_use[regno] = NULL_RTX;
1058 for (use_vec = DF_INSN_USES (insn); *use_vec; use_vec++)
1060 df_ref use = *use_vec;
1061 int regno = DF_REF_REGNO (use);
1063 /* Do not consider the usage of the stack pointer
1064 by function call. */
1065 if (DF_REF_FLAGS (use) & DF_REF_CALL_STACK_USAGE)
1066 continue;
1068 next_use[regno] = insn;
1073 free (next_use);
1076 /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1077 true if we found a LOG_LINK that proves that A feeds B. This only works
1078 if there are no instructions between A and B which could have a link
1079 depending on A, since in that case we would not record a link for B.
1080 We also check the implicit dependency created by a cc0 setter/user
1081 pair. */
1083 static bool
1084 insn_a_feeds_b (rtx a, rtx b)
1086 struct insn_link *links;
1087 FOR_EACH_LOG_LINK (links, b)
1088 if (links->insn == a)
1089 return true;
1090 #ifdef HAVE_cc0
1091 if (sets_cc0_p (a))
1092 return true;
1093 #endif
1094 return false;
1097 /* Main entry point for combiner. F is the first insn of the function.
1098 NREGS is the first unused pseudo-reg number.
1100 Return nonzero if the combiner has turned an indirect jump
1101 instruction into a direct jump. */
1102 static int
1103 combine_instructions (rtx f, unsigned int nregs)
1105 rtx insn, next;
1106 #ifdef HAVE_cc0
1107 rtx prev;
1108 #endif
1109 struct insn_link *links, *nextlinks;
1110 rtx first;
1111 basic_block last_bb;
1113 int new_direct_jump_p = 0;
1115 for (first = f; first && !INSN_P (first); )
1116 first = NEXT_INSN (first);
1117 if (!first)
1118 return 0;
1120 combine_attempts = 0;
1121 combine_merges = 0;
1122 combine_extras = 0;
1123 combine_successes = 0;
1125 rtl_hooks = combine_rtl_hooks;
1127 VEC_safe_grow_cleared (reg_stat_type, heap, reg_stat, nregs);
1129 init_recog_no_volatile ();
1131 /* Allocate array for insn info. */
1132 max_uid_known = get_max_uid ();
1133 uid_log_links = XCNEWVEC (struct insn_link *, max_uid_known + 1);
1134 uid_insn_cost = XCNEWVEC (int, max_uid_known + 1);
1135 gcc_obstack_init (&insn_link_obstack);
1137 nonzero_bits_mode = mode_for_size (HOST_BITS_PER_WIDE_INT, MODE_INT, 0);
1139 /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1140 problems when, for example, we have j <<= 1 in a loop. */
1142 nonzero_sign_valid = 0;
1143 label_tick = label_tick_ebb_start = 1;
1145 /* Scan all SETs and see if we can deduce anything about what
1146 bits are known to be zero for some registers and how many copies
1147 of the sign bit are known to exist for those registers.
1149 Also set any known values so that we can use it while searching
1150 for what bits are known to be set. */
1152 setup_incoming_promotions (first);
1153 /* Allow the entry block and the first block to fall into the same EBB.
1154 Conceptually the incoming promotions are assigned to the entry block. */
1155 last_bb = ENTRY_BLOCK_PTR;
1157 create_log_links ();
1158 FOR_EACH_BB (this_basic_block)
1160 optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
1161 last_call_luid = 0;
1162 mem_last_set = -1;
1164 label_tick++;
1165 if (!single_pred_p (this_basic_block)
1166 || single_pred (this_basic_block) != last_bb)
1167 label_tick_ebb_start = label_tick;
1168 last_bb = this_basic_block;
1170 FOR_BB_INSNS (this_basic_block, insn)
1171 if (INSN_P (insn) && BLOCK_FOR_INSN (insn))
1173 #ifdef AUTO_INC_DEC
1174 rtx links;
1175 #endif
1177 subst_low_luid = DF_INSN_LUID (insn);
1178 subst_insn = insn;
1180 note_stores (PATTERN (insn), set_nonzero_bits_and_sign_copies,
1181 insn);
1182 record_dead_and_set_regs (insn);
1184 #ifdef AUTO_INC_DEC
1185 for (links = REG_NOTES (insn); links; links = XEXP (links, 1))
1186 if (REG_NOTE_KIND (links) == REG_INC)
1187 set_nonzero_bits_and_sign_copies (XEXP (links, 0), NULL_RTX,
1188 insn);
1189 #endif
1191 /* Record the current insn_rtx_cost of this instruction. */
1192 if (NONJUMP_INSN_P (insn))
1193 INSN_COST (insn) = insn_rtx_cost (PATTERN (insn),
1194 optimize_this_for_speed_p);
1195 if (dump_file)
1196 fprintf(dump_file, "insn_cost %d: %d\n",
1197 INSN_UID (insn), INSN_COST (insn));
1201 nonzero_sign_valid = 1;
1203 /* Now scan all the insns in forward order. */
1204 label_tick = label_tick_ebb_start = 1;
1205 init_reg_last ();
1206 setup_incoming_promotions (first);
1207 last_bb = ENTRY_BLOCK_PTR;
1209 FOR_EACH_BB (this_basic_block)
1211 rtx last_combined_insn = NULL_RTX;
1212 optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
1213 last_call_luid = 0;
1214 mem_last_set = -1;
1216 label_tick++;
1217 if (!single_pred_p (this_basic_block)
1218 || single_pred (this_basic_block) != last_bb)
1219 label_tick_ebb_start = label_tick;
1220 last_bb = this_basic_block;
1222 rtl_profile_for_bb (this_basic_block);
1223 for (insn = BB_HEAD (this_basic_block);
1224 insn != NEXT_INSN (BB_END (this_basic_block));
1225 insn = next ? next : NEXT_INSN (insn))
1227 next = 0;
1228 if (NONDEBUG_INSN_P (insn))
1230 while (last_combined_insn
1231 && INSN_DELETED_P (last_combined_insn))
1232 last_combined_insn = PREV_INSN (last_combined_insn);
1233 if (last_combined_insn == NULL_RTX
1234 || BARRIER_P (last_combined_insn)
1235 || BLOCK_FOR_INSN (last_combined_insn) != this_basic_block
1236 || DF_INSN_LUID (last_combined_insn) <= DF_INSN_LUID (insn))
1237 last_combined_insn = insn;
1239 /* See if we know about function return values before this
1240 insn based upon SUBREG flags. */
1241 check_promoted_subreg (insn, PATTERN (insn));
1243 /* See if we can find hardregs and subreg of pseudos in
1244 narrower modes. This could help turning TRUNCATEs
1245 into SUBREGs. */
1246 note_uses (&PATTERN (insn), record_truncated_values, NULL);
1248 /* Try this insn with each insn it links back to. */
1250 FOR_EACH_LOG_LINK (links, insn)
1251 if ((next = try_combine (insn, links->insn, NULL_RTX,
1252 NULL_RTX, &new_direct_jump_p,
1253 last_combined_insn)) != 0)
1254 goto retry;
1256 /* Try each sequence of three linked insns ending with this one. */
1258 FOR_EACH_LOG_LINK (links, insn)
1260 rtx link = links->insn;
1262 /* If the linked insn has been replaced by a note, then there
1263 is no point in pursuing this chain any further. */
1264 if (NOTE_P (link))
1265 continue;
1267 FOR_EACH_LOG_LINK (nextlinks, link)
1268 if ((next = try_combine (insn, link, nextlinks->insn,
1269 NULL_RTX, &new_direct_jump_p,
1270 last_combined_insn)) != 0)
1271 goto retry;
1274 #ifdef HAVE_cc0
1275 /* Try to combine a jump insn that uses CC0
1276 with a preceding insn that sets CC0, and maybe with its
1277 logical predecessor as well.
1278 This is how we make decrement-and-branch insns.
1279 We need this special code because data flow connections
1280 via CC0 do not get entered in LOG_LINKS. */
1282 if (JUMP_P (insn)
1283 && (prev = prev_nonnote_insn (insn)) != 0
1284 && NONJUMP_INSN_P (prev)
1285 && sets_cc0_p (PATTERN (prev)))
1287 if ((next = try_combine (insn, prev, NULL_RTX, NULL_RTX,
1288 &new_direct_jump_p,
1289 last_combined_insn)) != 0)
1290 goto retry;
1292 FOR_EACH_LOG_LINK (nextlinks, prev)
1293 if ((next = try_combine (insn, prev, nextlinks->insn,
1294 NULL_RTX, &new_direct_jump_p,
1295 last_combined_insn)) != 0)
1296 goto retry;
1299 /* Do the same for an insn that explicitly references CC0. */
1300 if (NONJUMP_INSN_P (insn)
1301 && (prev = prev_nonnote_insn (insn)) != 0
1302 && NONJUMP_INSN_P (prev)
1303 && sets_cc0_p (PATTERN (prev))
1304 && GET_CODE (PATTERN (insn)) == SET
1305 && reg_mentioned_p (cc0_rtx, SET_SRC (PATTERN (insn))))
1307 if ((next = try_combine (insn, prev, NULL_RTX, NULL_RTX,
1308 &new_direct_jump_p,
1309 last_combined_insn)) != 0)
1310 goto retry;
1312 FOR_EACH_LOG_LINK (nextlinks, prev)
1313 if ((next = try_combine (insn, prev, nextlinks->insn,
1314 NULL_RTX, &new_direct_jump_p,
1315 last_combined_insn)) != 0)
1316 goto retry;
1319 /* Finally, see if any of the insns that this insn links to
1320 explicitly references CC0. If so, try this insn, that insn,
1321 and its predecessor if it sets CC0. */
1322 FOR_EACH_LOG_LINK (links, insn)
1323 if (NONJUMP_INSN_P (links->insn)
1324 && GET_CODE (PATTERN (links->insn)) == SET
1325 && reg_mentioned_p (cc0_rtx, SET_SRC (PATTERN (links->insn)))
1326 && (prev = prev_nonnote_insn (links->insn)) != 0
1327 && NONJUMP_INSN_P (prev)
1328 && sets_cc0_p (PATTERN (prev))
1329 && (next = try_combine (insn, links->insn,
1330 prev, NULL_RTX, &new_direct_jump_p,
1331 last_combined_insn)) != 0)
1332 goto retry;
1333 #endif
1335 /* Try combining an insn with two different insns whose results it
1336 uses. */
1337 FOR_EACH_LOG_LINK (links, insn)
1338 for (nextlinks = links->next; nextlinks;
1339 nextlinks = nextlinks->next)
1340 if ((next = try_combine (insn, links->insn,
1341 nextlinks->insn, NULL_RTX,
1342 &new_direct_jump_p,
1343 last_combined_insn)) != 0)
1344 goto retry;
1346 /* Try four-instruction combinations. */
1347 FOR_EACH_LOG_LINK (links, insn)
1349 struct insn_link *next1;
1350 rtx link = links->insn;
1352 /* If the linked insn has been replaced by a note, then there
1353 is no point in pursuing this chain any further. */
1354 if (NOTE_P (link))
1355 continue;
1357 FOR_EACH_LOG_LINK (next1, link)
1359 rtx link1 = next1->insn;
1360 if (NOTE_P (link1))
1361 continue;
1362 /* I0 -> I1 -> I2 -> I3. */
1363 FOR_EACH_LOG_LINK (nextlinks, link1)
1364 if ((next = try_combine (insn, link, link1,
1365 nextlinks->insn,
1366 &new_direct_jump_p,
1367 last_combined_insn)) != 0)
1368 goto retry;
1369 /* I0, I1 -> I2, I2 -> I3. */
1370 for (nextlinks = next1->next; nextlinks;
1371 nextlinks = nextlinks->next)
1372 if ((next = try_combine (insn, link, link1,
1373 nextlinks->insn,
1374 &new_direct_jump_p,
1375 last_combined_insn)) != 0)
1376 goto retry;
1379 for (next1 = links->next; next1; next1 = next1->next)
1381 rtx link1 = next1->insn;
1382 if (NOTE_P (link1))
1383 continue;
1384 /* I0 -> I2; I1, I2 -> I3. */
1385 FOR_EACH_LOG_LINK (nextlinks, link)
1386 if ((next = try_combine (insn, link, link1,
1387 nextlinks->insn,
1388 &new_direct_jump_p,
1389 last_combined_insn)) != 0)
1390 goto retry;
1391 /* I0 -> I1; I1, I2 -> I3. */
1392 FOR_EACH_LOG_LINK (nextlinks, link1)
1393 if ((next = try_combine (insn, link, link1,
1394 nextlinks->insn,
1395 &new_direct_jump_p,
1396 last_combined_insn)) != 0)
1397 goto retry;
1401 /* Try this insn with each REG_EQUAL note it links back to. */
1402 FOR_EACH_LOG_LINK (links, insn)
1404 rtx set, note;
1405 rtx temp = links->insn;
1406 if ((set = single_set (temp)) != 0
1407 && (note = find_reg_equal_equiv_note (temp)) != 0
1408 && (note = XEXP (note, 0), GET_CODE (note)) != EXPR_LIST
1409 /* Avoid using a register that may already been marked
1410 dead by an earlier instruction. */
1411 && ! unmentioned_reg_p (note, SET_SRC (set))
1412 && (GET_MODE (note) == VOIDmode
1413 ? SCALAR_INT_MODE_P (GET_MODE (SET_DEST (set)))
1414 : GET_MODE (SET_DEST (set)) == GET_MODE (note)))
1416 /* Temporarily replace the set's source with the
1417 contents of the REG_EQUAL note. The insn will
1418 be deleted or recognized by try_combine. */
1419 rtx orig = SET_SRC (set);
1420 SET_SRC (set) = note;
1421 i2mod = temp;
1422 i2mod_old_rhs = copy_rtx (orig);
1423 i2mod_new_rhs = copy_rtx (note);
1424 next = try_combine (insn, i2mod, NULL_RTX, NULL_RTX,
1425 &new_direct_jump_p,
1426 last_combined_insn);
1427 i2mod = NULL_RTX;
1428 if (next)
1429 goto retry;
1430 SET_SRC (set) = orig;
1434 if (!NOTE_P (insn))
1435 record_dead_and_set_regs (insn);
1437 retry:
1443 default_rtl_profile ();
1444 clear_bb_flags ();
1445 new_direct_jump_p |= purge_all_dead_edges ();
1446 delete_noop_moves ();
1448 /* Clean up. */
1449 obstack_free (&insn_link_obstack, NULL);
1450 free (uid_log_links);
1451 free (uid_insn_cost);
1452 VEC_free (reg_stat_type, heap, reg_stat);
1455 struct undo *undo, *next;
1456 for (undo = undobuf.frees; undo; undo = next)
1458 next = undo->next;
1459 free (undo);
1461 undobuf.frees = 0;
1464 total_attempts += combine_attempts;
1465 total_merges += combine_merges;
1466 total_extras += combine_extras;
1467 total_successes += combine_successes;
1469 nonzero_sign_valid = 0;
1470 rtl_hooks = general_rtl_hooks;
1472 /* Make recognizer allow volatile MEMs again. */
1473 init_recog ();
1475 return new_direct_jump_p;
1478 /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1480 static void
1481 init_reg_last (void)
1483 unsigned int i;
1484 reg_stat_type *p;
1486 FOR_EACH_VEC_ELT (reg_stat_type, reg_stat, i, p)
1487 memset (p, 0, offsetof (reg_stat_type, sign_bit_copies));
1490 /* Set up any promoted values for incoming argument registers. */
1492 static void
1493 setup_incoming_promotions (rtx first)
1495 tree arg;
1496 bool strictly_local = false;
1498 for (arg = DECL_ARGUMENTS (current_function_decl); arg;
1499 arg = DECL_CHAIN (arg))
1501 rtx x, reg = DECL_INCOMING_RTL (arg);
1502 int uns1, uns3;
1503 enum machine_mode mode1, mode2, mode3, mode4;
1505 /* Only continue if the incoming argument is in a register. */
1506 if (!REG_P (reg))
1507 continue;
1509 /* Determine, if possible, whether all call sites of the current
1510 function lie within the current compilation unit. (This does
1511 take into account the exporting of a function via taking its
1512 address, and so forth.) */
1513 strictly_local = cgraph_local_info (current_function_decl)->local;
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 mode1 = TYPE_MODE (TREE_TYPE (arg));
1518 uns1 = TYPE_UNSIGNED (TREE_TYPE (arg));
1520 /* The mode and signedness of the argument after any source language and
1521 TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1522 mode2 = TYPE_MODE (DECL_ARG_TYPE (arg));
1523 uns3 = TYPE_UNSIGNED (DECL_ARG_TYPE (arg));
1525 /* The mode and signedness of the argument as it is actually passed,
1526 after any TARGET_PROMOTE_FUNCTION_ARGS-driven ABI promotions. */
1527 mode3 = promote_function_mode (DECL_ARG_TYPE (arg), mode2, &uns3,
1528 TREE_TYPE (cfun->decl), 0);
1530 /* The mode of the register in which the argument is being passed. */
1531 mode4 = GET_MODE (reg);
1533 /* Eliminate sign extensions in the callee when:
1534 (a) A mode promotion has occurred; */
1535 if (mode1 == mode3)
1536 continue;
1537 /* (b) The mode of the register is the same as the mode of
1538 the argument as it is passed; */
1539 if (mode3 != mode4)
1540 continue;
1541 /* (c) There's no language level extension; */
1542 if (mode1 == mode2)
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 else if (!strictly_local)
1549 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 else if (uns1)
1555 uns3 = true;
1556 else if (uns3)
1557 continue;
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 x = gen_rtx_CLOBBER (mode1, const0_rtx);
1563 x = gen_rtx_fmt_e ((uns3 ? ZERO_EXTEND : SIGN_EXTEND), mode3, x);
1564 record_value_for_reg (reg, first, x);
1568 /* Called via note_stores. If X is a pseudo that is narrower than
1569 HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1571 If we are setting only a portion of X and we can't figure out what
1572 portion, assume all bits will be used since we don't know what will
1573 be happening.
1575 Similarly, set how many bits of X are known to be copies of the sign bit
1576 at all locations in the function. This is the smallest number implied
1577 by any set of X. */
1579 static void
1580 set_nonzero_bits_and_sign_copies (rtx x, const_rtx set, void *data)
1582 rtx insn = (rtx) data;
1583 unsigned int num;
1585 if (REG_P (x)
1586 && REGNO (x) >= FIRST_PSEUDO_REGISTER
1587 /* If this register is undefined at the start of the file, we can't
1588 say what its contents were. */
1589 && ! REGNO_REG_SET_P
1590 (DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x))
1591 && HWI_COMPUTABLE_MODE_P (GET_MODE (x)))
1593 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
1595 if (set == 0 || GET_CODE (set) == CLOBBER)
1597 rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
1598 rsp->sign_bit_copies = 1;
1599 return;
1602 /* If this register is being initialized using itself, and the
1603 register is uninitialized in this basic block, and there are
1604 no LOG_LINKS which set the register, then part of the
1605 register is uninitialized. In that case we can't assume
1606 anything about the number of nonzero bits.
1608 ??? We could do better if we checked this in
1609 reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1610 could avoid making assumptions about the insn which initially
1611 sets the register, while still using the information in other
1612 insns. We would have to be careful to check every insn
1613 involved in the combination. */
1615 if (insn
1616 && reg_referenced_p (x, PATTERN (insn))
1617 && !REGNO_REG_SET_P (DF_LR_IN (BLOCK_FOR_INSN (insn)),
1618 REGNO (x)))
1620 struct insn_link *link;
1622 FOR_EACH_LOG_LINK (link, insn)
1623 if (dead_or_set_p (link->insn, x))
1624 break;
1625 if (!link)
1627 rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
1628 rsp->sign_bit_copies = 1;
1629 return;
1633 /* If this is a complex assignment, see if we can convert it into a
1634 simple assignment. */
1635 set = expand_field_assignment (set);
1637 /* If this is a simple assignment, or we have a paradoxical SUBREG,
1638 set what we know about X. */
1640 if (SET_DEST (set) == x
1641 || (paradoxical_subreg_p (SET_DEST (set))
1642 && SUBREG_REG (SET_DEST (set)) == x))
1644 rtx src = SET_SRC (set);
1646 #ifdef SHORT_IMMEDIATES_SIGN_EXTEND
1647 /* If X is narrower than a word and SRC is a non-negative
1648 constant that would appear negative in the mode of X,
1649 sign-extend it for use in reg_stat[].nonzero_bits because some
1650 machines (maybe most) will actually do the sign-extension
1651 and this is the conservative approach.
1653 ??? For 2.5, try to tighten up the MD files in this regard
1654 instead of this kludge. */
1656 if (GET_MODE_PRECISION (GET_MODE (x)) < BITS_PER_WORD
1657 && CONST_INT_P (src)
1658 && INTVAL (src) > 0
1659 && val_signbit_known_set_p (GET_MODE (x), INTVAL (src)))
1660 src = GEN_INT (INTVAL (src) | ~GET_MODE_MASK (GET_MODE (x)));
1661 #endif
1663 /* Don't call nonzero_bits if it cannot change anything. */
1664 if (rsp->nonzero_bits != ~(unsigned HOST_WIDE_INT) 0)
1665 rsp->nonzero_bits |= nonzero_bits (src, nonzero_bits_mode);
1666 num = num_sign_bit_copies (SET_SRC (set), GET_MODE (x));
1667 if (rsp->sign_bit_copies == 0
1668 || rsp->sign_bit_copies > num)
1669 rsp->sign_bit_copies = num;
1671 else
1673 rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
1674 rsp->sign_bit_copies = 1;
1679 /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1680 optionally insns that were previously combined into I3 or that will be
1681 combined into the merger of INSN and I3. The order is PRED, PRED2,
1682 INSN, SUCC, SUCC2, I3.
1684 Return 0 if the combination is not allowed for any reason.
1686 If the combination is allowed, *PDEST will be set to the single
1687 destination of INSN and *PSRC to the single source, and this function
1688 will return 1. */
1690 static int
1691 can_combine_p (rtx insn, rtx i3, rtx pred ATTRIBUTE_UNUSED,
1692 rtx pred2 ATTRIBUTE_UNUSED, rtx succ, rtx succ2,
1693 rtx *pdest, rtx *psrc)
1695 int i;
1696 const_rtx set = 0;
1697 rtx src, dest;
1698 rtx p;
1699 #ifdef AUTO_INC_DEC
1700 rtx link;
1701 #endif
1702 bool all_adjacent = true;
1703 int (*is_volatile_p) (const_rtx);
1705 if (succ)
1707 if (succ2)
1709 if (next_active_insn (succ2) != i3)
1710 all_adjacent = false;
1711 if (next_active_insn (succ) != succ2)
1712 all_adjacent = false;
1714 else if (next_active_insn (succ) != i3)
1715 all_adjacent = false;
1716 if (next_active_insn (insn) != succ)
1717 all_adjacent = false;
1719 else if (next_active_insn (insn) != i3)
1720 all_adjacent = false;
1722 /* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
1723 or a PARALLEL consisting of such a SET and CLOBBERs.
1725 If INSN has CLOBBER parallel parts, ignore them for our processing.
1726 By definition, these happen during the execution of the insn. When it
1727 is merged with another insn, all bets are off. If they are, in fact,
1728 needed and aren't also supplied in I3, they may be added by
1729 recog_for_combine. Otherwise, it won't match.
1731 We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1732 note.
1734 Get the source and destination of INSN. If more than one, can't
1735 combine. */
1737 if (GET_CODE (PATTERN (insn)) == SET)
1738 set = PATTERN (insn);
1739 else if (GET_CODE (PATTERN (insn)) == PARALLEL
1740 && GET_CODE (XVECEXP (PATTERN (insn), 0, 0)) == SET)
1742 for (i = 0; i < XVECLEN (PATTERN (insn), 0); i++)
1744 rtx elt = XVECEXP (PATTERN (insn), 0, i);
1746 switch (GET_CODE (elt))
1748 /* This is important to combine floating point insns
1749 for the SH4 port. */
1750 case USE:
1751 /* Combining an isolated USE doesn't make sense.
1752 We depend here on combinable_i3pat to reject them. */
1753 /* The code below this loop only verifies that the inputs of
1754 the SET in INSN do not change. We call reg_set_between_p
1755 to verify that the REG in the USE does not change between
1756 I3 and INSN.
1757 If the USE in INSN was for a pseudo register, the matching
1758 insn pattern will likely match any register; combining this
1759 with any other USE would only be safe if we knew that the
1760 used registers have identical values, or if there was
1761 something to tell them apart, e.g. different modes. For
1762 now, we forgo such complicated tests and simply disallow
1763 combining of USES of pseudo registers with any other USE. */
1764 if (REG_P (XEXP (elt, 0))
1765 && GET_CODE (PATTERN (i3)) == PARALLEL)
1767 rtx i3pat = PATTERN (i3);
1768 int i = XVECLEN (i3pat, 0) - 1;
1769 unsigned int regno = REGNO (XEXP (elt, 0));
1773 rtx i3elt = XVECEXP (i3pat, 0, i);
1775 if (GET_CODE (i3elt) == USE
1776 && REG_P (XEXP (i3elt, 0))
1777 && (REGNO (XEXP (i3elt, 0)) == regno
1778 ? reg_set_between_p (XEXP (elt, 0),
1779 PREV_INSN (insn), i3)
1780 : regno >= FIRST_PSEUDO_REGISTER))
1781 return 0;
1783 while (--i >= 0);
1785 break;
1787 /* We can ignore CLOBBERs. */
1788 case CLOBBER:
1789 break;
1791 case SET:
1792 /* Ignore SETs whose result isn't used but not those that
1793 have side-effects. */
1794 if (find_reg_note (insn, REG_UNUSED, SET_DEST (elt))
1795 && insn_nothrow_p (insn)
1796 && !side_effects_p (elt))
1797 break;
1799 /* If we have already found a SET, this is a second one and
1800 so we cannot combine with this insn. */
1801 if (set)
1802 return 0;
1804 set = elt;
1805 break;
1807 default:
1808 /* Anything else means we can't combine. */
1809 return 0;
1813 if (set == 0
1814 /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1815 so don't do anything with it. */
1816 || GET_CODE (SET_SRC (set)) == ASM_OPERANDS)
1817 return 0;
1819 else
1820 return 0;
1822 if (set == 0)
1823 return 0;
1825 set = expand_field_assignment (set);
1826 src = SET_SRC (set), dest = SET_DEST (set);
1828 /* Don't eliminate a store in the stack pointer. */
1829 if (dest == stack_pointer_rtx
1830 /* Don't combine with an insn that sets a register to itself if it has
1831 a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1832 || (rtx_equal_p (src, dest) && find_reg_note (insn, REG_EQUAL, NULL_RTX))
1833 /* Can't merge an ASM_OPERANDS. */
1834 || GET_CODE (src) == ASM_OPERANDS
1835 /* Can't merge a function call. */
1836 || GET_CODE (src) == CALL
1837 /* Don't eliminate a function call argument. */
1838 || (CALL_P (i3)
1839 && (find_reg_fusage (i3, USE, dest)
1840 || (REG_P (dest)
1841 && REGNO (dest) < FIRST_PSEUDO_REGISTER
1842 && global_regs[REGNO (dest)])))
1843 /* Don't substitute into an incremented register. */
1844 || FIND_REG_INC_NOTE (i3, dest)
1845 || (succ && FIND_REG_INC_NOTE (succ, dest))
1846 || (succ2 && FIND_REG_INC_NOTE (succ2, dest))
1847 /* Don't substitute into a non-local goto, this confuses CFG. */
1848 || (JUMP_P (i3) && find_reg_note (i3, REG_NON_LOCAL_GOTO, NULL_RTX))
1849 /* Make sure that DEST is not used after SUCC but before I3. */
1850 || (!all_adjacent
1851 && ((succ2
1852 && (reg_used_between_p (dest, succ2, i3)
1853 || reg_used_between_p (dest, succ, succ2)))
1854 || (!succ2 && succ && reg_used_between_p (dest, succ, i3))))
1855 /* Make sure that the value that is to be substituted for the register
1856 does not use any registers whose values alter in between. However,
1857 If the insns are adjacent, a use can't cross a set even though we
1858 think it might (this can happen for a sequence of insns each setting
1859 the same destination; last_set of that register might point to
1860 a NOTE). If INSN has a REG_EQUIV note, the register is always
1861 equivalent to the memory so the substitution is valid even if there
1862 are intervening stores. Also, don't move a volatile asm or
1863 UNSPEC_VOLATILE across any other insns. */
1864 || (! all_adjacent
1865 && (((!MEM_P (src)
1866 || ! find_reg_note (insn, REG_EQUIV, src))
1867 && use_crosses_set_p (src, DF_INSN_LUID (insn)))
1868 || (GET_CODE (src) == ASM_OPERANDS && MEM_VOLATILE_P (src))
1869 || GET_CODE (src) == UNSPEC_VOLATILE))
1870 /* Don't combine across a CALL_INSN, because that would possibly
1871 change whether the life span of some REGs crosses calls or not,
1872 and it is a pain to update that information.
1873 Exception: if source is a constant, moving it later can't hurt.
1874 Accept that as a special case. */
1875 || (DF_INSN_LUID (insn) < last_call_luid && ! CONSTANT_P (src)))
1876 return 0;
1878 /* DEST must either be a REG or CC0. */
1879 if (REG_P (dest))
1881 /* If register alignment is being enforced for multi-word items in all
1882 cases except for parameters, it is possible to have a register copy
1883 insn referencing a hard register that is not allowed to contain the
1884 mode being copied and which would not be valid as an operand of most
1885 insns. Eliminate this problem by not combining with such an insn.
1887 Also, on some machines we don't want to extend the life of a hard
1888 register. */
1890 if (REG_P (src)
1891 && ((REGNO (dest) < FIRST_PSEUDO_REGISTER
1892 && ! HARD_REGNO_MODE_OK (REGNO (dest), GET_MODE (dest)))
1893 /* Don't extend the life of a hard register unless it is
1894 user variable (if we have few registers) or it can't
1895 fit into the desired register (meaning something special
1896 is going on).
1897 Also avoid substituting a return register into I3, because
1898 reload can't handle a conflict with constraints of other
1899 inputs. */
1900 || (REGNO (src) < FIRST_PSEUDO_REGISTER
1901 && ! HARD_REGNO_MODE_OK (REGNO (src), GET_MODE (src)))))
1902 return 0;
1904 else if (GET_CODE (dest) != CC0)
1905 return 0;
1908 if (GET_CODE (PATTERN (i3)) == PARALLEL)
1909 for (i = XVECLEN (PATTERN (i3), 0) - 1; i >= 0; i--)
1910 if (GET_CODE (XVECEXP (PATTERN (i3), 0, i)) == CLOBBER)
1912 /* Don't substitute for a register intended as a clobberable
1913 operand. */
1914 rtx reg = XEXP (XVECEXP (PATTERN (i3), 0, i), 0);
1915 if (rtx_equal_p (reg, dest))
1916 return 0;
1918 /* If the clobber represents an earlyclobber operand, we must not
1919 substitute an expression containing the clobbered register.
1920 As we do not analyze the constraint strings here, we have to
1921 make the conservative assumption. However, if the register is
1922 a fixed hard reg, the clobber cannot represent any operand;
1923 we leave it up to the machine description to either accept or
1924 reject use-and-clobber patterns. */
1925 if (!REG_P (reg)
1926 || REGNO (reg) >= FIRST_PSEUDO_REGISTER
1927 || !fixed_regs[REGNO (reg)])
1928 if (reg_overlap_mentioned_p (reg, src))
1929 return 0;
1932 /* If INSN contains anything volatile, or is an `asm' (whether volatile
1933 or not), reject, unless nothing volatile comes between it and I3 */
1935 if (GET_CODE (src) == ASM_OPERANDS || volatile_refs_p (src))
1937 /* Make sure neither succ nor succ2 contains a volatile reference. */
1938 if (succ2 != 0 && volatile_refs_p (PATTERN (succ2)))
1939 return 0;
1940 if (succ != 0 && volatile_refs_p (PATTERN (succ)))
1941 return 0;
1942 /* We'll check insns between INSN and I3 below. */
1945 /* If INSN is an asm, and DEST is a hard register, reject, since it has
1946 to be an explicit register variable, and was chosen for a reason. */
1948 if (GET_CODE (src) == ASM_OPERANDS
1949 && REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER)
1950 return 0;
1952 /* If INSN contains volatile references (specifically volatile MEMs),
1953 we cannot combine across any other volatile references.
1954 Even if INSN doesn't contain volatile references, any intervening
1955 volatile insn might affect machine state. */
1957 is_volatile_p = volatile_refs_p (PATTERN (insn))
1958 ? volatile_refs_p
1959 : volatile_insn_p;
1961 for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
1962 if (INSN_P (p) && p != succ && p != succ2 && is_volatile_p (PATTERN (p)))
1963 return 0;
1965 /* If INSN contains an autoincrement or autodecrement, make sure that
1966 register is not used between there and I3, and not already used in
1967 I3 either. Neither must it be used in PRED or SUCC, if they exist.
1968 Also insist that I3 not be a jump; if it were one
1969 and the incremented register were spilled, we would lose. */
1971 #ifdef AUTO_INC_DEC
1972 for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
1973 if (REG_NOTE_KIND (link) == REG_INC
1974 && (JUMP_P (i3)
1975 || reg_used_between_p (XEXP (link, 0), insn, i3)
1976 || (pred != NULL_RTX
1977 && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred)))
1978 || (pred2 != NULL_RTX
1979 && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred2)))
1980 || (succ != NULL_RTX
1981 && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ)))
1982 || (succ2 != NULL_RTX
1983 && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ2)))
1984 || reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i3))))
1985 return 0;
1986 #endif
1988 #ifdef HAVE_cc0
1989 /* Don't combine an insn that follows a CC0-setting insn.
1990 An insn that uses CC0 must not be separated from the one that sets it.
1991 We do, however, allow I2 to follow a CC0-setting insn if that insn
1992 is passed as I1; in that case it will be deleted also.
1993 We also allow combining in this case if all the insns are adjacent
1994 because that would leave the two CC0 insns adjacent as well.
1995 It would be more logical to test whether CC0 occurs inside I1 or I2,
1996 but that would be much slower, and this ought to be equivalent. */
1998 p = prev_nonnote_insn (insn);
1999 if (p && p != pred && NONJUMP_INSN_P (p) && sets_cc0_p (PATTERN (p))
2000 && ! all_adjacent)
2001 return 0;
2002 #endif
2004 /* If we get here, we have passed all the tests and the combination is
2005 to be allowed. */
2007 *pdest = dest;
2008 *psrc = src;
2010 return 1;
2013 /* LOC is the location within I3 that contains its pattern or the component
2014 of a PARALLEL of the pattern. We validate that it is valid for combining.
2016 One problem is if I3 modifies its output, as opposed to replacing it
2017 entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2018 doing so would produce an insn that is not equivalent to the original insns.
2020 Consider:
2022 (set (reg:DI 101) (reg:DI 100))
2023 (set (subreg:SI (reg:DI 101) 0) <foo>)
2025 This is NOT equivalent to:
2027 (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2028 (set (reg:DI 101) (reg:DI 100))])
2030 Not only does this modify 100 (in which case it might still be valid
2031 if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2033 We can also run into a problem if I2 sets a register that I1
2034 uses and I1 gets directly substituted into I3 (not via I2). In that
2035 case, we would be getting the wrong value of I2DEST into I3, so we
2036 must reject the combination. This case occurs when I2 and I1 both
2037 feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2038 If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2039 of a SET must prevent combination from occurring. The same situation
2040 can occur for I0, in which case I0_NOT_IN_SRC is set.
2042 Before doing the above check, we first try to expand a field assignment
2043 into a set of logical operations.
2045 If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2046 we place a register that is both set and used within I3. If more than one
2047 such register is detected, we fail.
2049 Return 1 if the combination is valid, zero otherwise. */
2051 static int
2052 combinable_i3pat (rtx i3, rtx *loc, rtx i2dest, rtx i1dest, rtx i0dest,
2053 int i1_not_in_src, int i0_not_in_src, rtx *pi3dest_killed)
2055 rtx x = *loc;
2057 if (GET_CODE (x) == SET)
2059 rtx set = x ;
2060 rtx dest = SET_DEST (set);
2061 rtx src = SET_SRC (set);
2062 rtx inner_dest = dest;
2063 rtx subdest;
2065 while (GET_CODE (inner_dest) == STRICT_LOW_PART
2066 || GET_CODE (inner_dest) == SUBREG
2067 || GET_CODE (inner_dest) == ZERO_EXTRACT)
2068 inner_dest = XEXP (inner_dest, 0);
2070 /* Check for the case where I3 modifies its output, as discussed
2071 above. We don't want to prevent pseudos from being combined
2072 into the address of a MEM, so only prevent the combination if
2073 i1 or i2 set the same MEM. */
2074 if ((inner_dest != dest &&
2075 (!MEM_P (inner_dest)
2076 || rtx_equal_p (i2dest, inner_dest)
2077 || (i1dest && rtx_equal_p (i1dest, inner_dest))
2078 || (i0dest && rtx_equal_p (i0dest, inner_dest)))
2079 && (reg_overlap_mentioned_p (i2dest, inner_dest)
2080 || (i1dest && reg_overlap_mentioned_p (i1dest, inner_dest))
2081 || (i0dest && reg_overlap_mentioned_p (i0dest, inner_dest))))
2083 /* This is the same test done in can_combine_p except we can't test
2084 all_adjacent; we don't have to, since this instruction will stay
2085 in place, thus we are not considering increasing the lifetime of
2086 INNER_DEST.
2088 Also, if this insn sets a function argument, combining it with
2089 something that might need a spill could clobber a previous
2090 function argument; the all_adjacent test in can_combine_p also
2091 checks this; here, we do a more specific test for this case. */
2093 || (REG_P (inner_dest)
2094 && REGNO (inner_dest) < FIRST_PSEUDO_REGISTER
2095 && (! HARD_REGNO_MODE_OK (REGNO (inner_dest),
2096 GET_MODE (inner_dest))))
2097 || (i1_not_in_src && reg_overlap_mentioned_p (i1dest, src))
2098 || (i0_not_in_src && reg_overlap_mentioned_p (i0dest, src)))
2099 return 0;
2101 /* If DEST is used in I3, it is being killed in this insn, so
2102 record that for later. We have to consider paradoxical
2103 subregs here, since they kill the whole register, but we
2104 ignore partial subregs, STRICT_LOW_PART, etc.
2105 Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2106 STACK_POINTER_REGNUM, since these are always considered to be
2107 live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2108 subdest = dest;
2109 if (GET_CODE (subdest) == SUBREG
2110 && (GET_MODE_SIZE (GET_MODE (subdest))
2111 >= GET_MODE_SIZE (GET_MODE (SUBREG_REG (subdest)))))
2112 subdest = SUBREG_REG (subdest);
2113 if (pi3dest_killed
2114 && REG_P (subdest)
2115 && reg_referenced_p (subdest, PATTERN (i3))
2116 && REGNO (subdest) != FRAME_POINTER_REGNUM
2117 #if !HARD_FRAME_POINTER_IS_FRAME_POINTER
2118 && REGNO (subdest) != HARD_FRAME_POINTER_REGNUM
2119 #endif
2120 #if ARG_POINTER_REGNUM != FRAME_POINTER_REGNUM
2121 && (REGNO (subdest) != ARG_POINTER_REGNUM
2122 || ! fixed_regs [REGNO (subdest)])
2123 #endif
2124 && REGNO (subdest) != STACK_POINTER_REGNUM)
2126 if (*pi3dest_killed)
2127 return 0;
2129 *pi3dest_killed = subdest;
2133 else if (GET_CODE (x) == PARALLEL)
2135 int i;
2137 for (i = 0; i < XVECLEN (x, 0); i++)
2138 if (! combinable_i3pat (i3, &XVECEXP (x, 0, i), i2dest, i1dest, i0dest,
2139 i1_not_in_src, i0_not_in_src, pi3dest_killed))
2140 return 0;
2143 return 1;
2146 /* Return 1 if X is an arithmetic expression that contains a multiplication
2147 and division. We don't count multiplications by powers of two here. */
2149 static int
2150 contains_muldiv (rtx x)
2152 switch (GET_CODE (x))
2154 case MOD: case DIV: case UMOD: case UDIV:
2155 return 1;
2157 case MULT:
2158 return ! (CONST_INT_P (XEXP (x, 1))
2159 && exact_log2 (UINTVAL (XEXP (x, 1))) >= 0);
2160 default:
2161 if (BINARY_P (x))
2162 return contains_muldiv (XEXP (x, 0))
2163 || contains_muldiv (XEXP (x, 1));
2165 if (UNARY_P (x))
2166 return contains_muldiv (XEXP (x, 0));
2168 return 0;
2172 /* Determine whether INSN can be used in a combination. Return nonzero if
2173 not. This is used in try_combine to detect early some cases where we
2174 can't perform combinations. */
2176 static int
2177 cant_combine_insn_p (rtx insn)
2179 rtx set;
2180 rtx src, dest;
2182 /* If this isn't really an insn, we can't do anything.
2183 This can occur when flow deletes an insn that it has merged into an
2184 auto-increment address. */
2185 if (! INSN_P (insn))
2186 return 1;
2188 /* Never combine loads and stores involving hard regs that are likely
2189 to be spilled. The register allocator can usually handle such
2190 reg-reg moves by tying. If we allow the combiner to make
2191 substitutions of likely-spilled regs, reload might die.
2192 As an exception, we allow combinations involving fixed regs; these are
2193 not available to the register allocator so there's no risk involved. */
2195 set = single_set (insn);
2196 if (! set)
2197 return 0;
2198 src = SET_SRC (set);
2199 dest = SET_DEST (set);
2200 if (GET_CODE (src) == SUBREG)
2201 src = SUBREG_REG (src);
2202 if (GET_CODE (dest) == SUBREG)
2203 dest = SUBREG_REG (dest);
2204 if (REG_P (src) && REG_P (dest)
2205 && ((HARD_REGISTER_P (src)
2206 && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (src))
2207 && targetm.class_likely_spilled_p (REGNO_REG_CLASS (REGNO (src))))
2208 || (HARD_REGISTER_P (dest)
2209 && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (dest))
2210 && targetm.class_likely_spilled_p (REGNO_REG_CLASS (REGNO (dest))))))
2211 return 1;
2213 return 0;
2216 struct likely_spilled_retval_info
2218 unsigned regno, nregs;
2219 unsigned mask;
2222 /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2223 hard registers that are known to be written to / clobbered in full. */
2224 static void
2225 likely_spilled_retval_1 (rtx x, const_rtx set, void *data)
2227 struct likely_spilled_retval_info *const info =
2228 (struct likely_spilled_retval_info *) data;
2229 unsigned regno, nregs;
2230 unsigned new_mask;
2232 if (!REG_P (XEXP (set, 0)))
2233 return;
2234 regno = REGNO (x);
2235 if (regno >= info->regno + info->nregs)
2236 return;
2237 nregs = hard_regno_nregs[regno][GET_MODE (x)];
2238 if (regno + nregs <= info->regno)
2239 return;
2240 new_mask = (2U << (nregs - 1)) - 1;
2241 if (regno < info->regno)
2242 new_mask >>= info->regno - regno;
2243 else
2244 new_mask <<= regno - info->regno;
2245 info->mask &= ~new_mask;
2248 /* Return nonzero iff part of the return value is live during INSN, and
2249 it is likely spilled. This can happen when more than one insn is needed
2250 to copy the return value, e.g. when we consider to combine into the
2251 second copy insn for a complex value. */
2253 static int
2254 likely_spilled_retval_p (rtx insn)
2256 rtx use = BB_END (this_basic_block);
2257 rtx reg, p;
2258 unsigned regno, nregs;
2259 /* We assume here that no machine mode needs more than
2260 32 hard registers when the value overlaps with a register
2261 for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2262 unsigned mask;
2263 struct likely_spilled_retval_info info;
2265 if (!NONJUMP_INSN_P (use) || GET_CODE (PATTERN (use)) != USE || insn == use)
2266 return 0;
2267 reg = XEXP (PATTERN (use), 0);
2268 if (!REG_P (reg) || !targetm.calls.function_value_regno_p (REGNO (reg)))
2269 return 0;
2270 regno = REGNO (reg);
2271 nregs = hard_regno_nregs[regno][GET_MODE (reg)];
2272 if (nregs == 1)
2273 return 0;
2274 mask = (2U << (nregs - 1)) - 1;
2276 /* Disregard parts of the return value that are set later. */
2277 info.regno = regno;
2278 info.nregs = nregs;
2279 info.mask = mask;
2280 for (p = PREV_INSN (use); info.mask && p != insn; p = PREV_INSN (p))
2281 if (INSN_P (p))
2282 note_stores (PATTERN (p), likely_spilled_retval_1, &info);
2283 mask = info.mask;
2285 /* Check if any of the (probably) live return value registers is
2286 likely spilled. */
2287 nregs --;
2290 if ((mask & 1 << nregs)
2291 && targetm.class_likely_spilled_p (REGNO_REG_CLASS (regno + nregs)))
2292 return 1;
2293 } while (nregs--);
2294 return 0;
2297 /* Adjust INSN after we made a change to its destination.
2299 Changing the destination can invalidate notes that say something about
2300 the results of the insn and a LOG_LINK pointing to the insn. */
2302 static void
2303 adjust_for_new_dest (rtx insn)
2305 /* For notes, be conservative and simply remove them. */
2306 remove_reg_equal_equiv_notes (insn);
2308 /* The new insn will have a destination that was previously the destination
2309 of an insn just above it. Call distribute_links to make a LOG_LINK from
2310 the next use of that destination. */
2311 distribute_links (alloc_insn_link (insn, NULL));
2313 df_insn_rescan (insn);
2316 /* Return TRUE if combine can reuse reg X in mode MODE.
2317 ADDED_SETS is nonzero if the original set is still required. */
2318 static bool
2319 can_change_dest_mode (rtx x, int added_sets, enum machine_mode mode)
2321 unsigned int regno;
2323 if (!REG_P(x))
2324 return false;
2326 regno = REGNO (x);
2327 /* Allow hard registers if the new mode is legal, and occupies no more
2328 registers than the old mode. */
2329 if (regno < FIRST_PSEUDO_REGISTER)
2330 return (HARD_REGNO_MODE_OK (regno, mode)
2331 && (hard_regno_nregs[regno][GET_MODE (x)]
2332 >= hard_regno_nregs[regno][mode]));
2334 /* Or a pseudo that is only used once. */
2335 return (REG_N_SETS (regno) == 1 && !added_sets
2336 && !REG_USERVAR_P (x));
2340 /* Check whether X, the destination of a set, refers to part of
2341 the register specified by REG. */
2343 static bool
2344 reg_subword_p (rtx x, rtx reg)
2346 /* Check that reg is an integer mode register. */
2347 if (!REG_P (reg) || GET_MODE_CLASS (GET_MODE (reg)) != MODE_INT)
2348 return false;
2350 if (GET_CODE (x) == STRICT_LOW_PART
2351 || GET_CODE (x) == ZERO_EXTRACT)
2352 x = XEXP (x, 0);
2354 return GET_CODE (x) == SUBREG
2355 && SUBREG_REG (x) == reg
2356 && GET_MODE_CLASS (GET_MODE (x)) == MODE_INT;
2359 #ifdef AUTO_INC_DEC
2360 /* Replace auto-increment addressing modes with explicit operations to access
2361 the same addresses without modifying the corresponding registers. */
2363 static rtx
2364 cleanup_auto_inc_dec (rtx src, enum machine_mode mem_mode)
2366 rtx x = src;
2367 const RTX_CODE code = GET_CODE (x);
2368 int i;
2369 const char *fmt;
2371 switch (code)
2373 case REG:
2374 case CONST_INT:
2375 case CONST_DOUBLE:
2376 case CONST_FIXED:
2377 case CONST_VECTOR:
2378 case SYMBOL_REF:
2379 case CODE_LABEL:
2380 case PC:
2381 case CC0:
2382 case SCRATCH:
2383 /* SCRATCH must be shared because they represent distinct values. */
2384 return x;
2385 case CLOBBER:
2386 if (REG_P (XEXP (x, 0)) && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER)
2387 return x;
2388 break;
2390 case CONST:
2391 if (shared_const_p (x))
2392 return x;
2393 break;
2395 case MEM:
2396 mem_mode = GET_MODE (x);
2397 break;
2399 case PRE_INC:
2400 case PRE_DEC:
2401 gcc_assert (mem_mode != VOIDmode && mem_mode != BLKmode);
2402 return gen_rtx_PLUS (GET_MODE (x),
2403 cleanup_auto_inc_dec (XEXP (x, 0), mem_mode),
2404 GEN_INT (code == PRE_INC
2405 ? GET_MODE_SIZE (mem_mode)
2406 : -GET_MODE_SIZE (mem_mode)));
2408 case POST_INC:
2409 case POST_DEC:
2410 case PRE_MODIFY:
2411 case POST_MODIFY:
2412 return cleanup_auto_inc_dec (code == PRE_MODIFY
2413 ? XEXP (x, 1) : XEXP (x, 0),
2414 mem_mode);
2416 default:
2417 break;
2420 /* Copy the various flags, fields, and other information. We assume
2421 that all fields need copying, and then clear the fields that should
2422 not be copied. That is the sensible default behavior, and forces
2423 us to explicitly document why we are *not* copying a flag. */
2424 x = shallow_copy_rtx (x);
2426 /* We do not copy the USED flag, which is used as a mark bit during
2427 walks over the RTL. */
2428 RTX_FLAG (x, used) = 0;
2430 /* We do not copy FRAME_RELATED for INSNs. */
2431 if (INSN_P (x))
2432 RTX_FLAG (x, frame_related) = 0;
2434 fmt = GET_RTX_FORMAT (code);
2435 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2436 if (fmt[i] == 'e')
2437 XEXP (x, i) = cleanup_auto_inc_dec (XEXP (x, i), mem_mode);
2438 else if (fmt[i] == 'E' || fmt[i] == 'V')
2440 int j;
2441 XVEC (x, i) = rtvec_alloc (XVECLEN (x, i));
2442 for (j = 0; j < XVECLEN (x, i); j++)
2443 XVECEXP (x, i, j)
2444 = cleanup_auto_inc_dec (XVECEXP (src, i, j), mem_mode);
2447 return x;
2449 #endif
2451 /* Auxiliary data structure for propagate_for_debug_stmt. */
2453 struct rtx_subst_pair
2455 rtx to;
2456 bool adjusted;
2459 /* DATA points to an rtx_subst_pair. Return the value that should be
2460 substituted. */
2462 static rtx
2463 propagate_for_debug_subst (rtx from, const_rtx old_rtx, void *data)
2465 struct rtx_subst_pair *pair = (struct rtx_subst_pair *)data;
2467 if (!rtx_equal_p (from, old_rtx))
2468 return NULL_RTX;
2469 if (!pair->adjusted)
2471 pair->adjusted = true;
2472 #ifdef AUTO_INC_DEC
2473 pair->to = cleanup_auto_inc_dec (pair->to, VOIDmode);
2474 #else
2475 pair->to = copy_rtx (pair->to);
2476 #endif
2477 pair->to = make_compound_operation (pair->to, SET);
2478 return pair->to;
2480 return copy_rtx (pair->to);
2483 /* Replace all the occurrences of DEST with SRC in DEBUG_INSNs between INSN
2484 and LAST, not including INSN, but including LAST. Also stop at the end
2485 of THIS_BASIC_BLOCK. */
2487 static void
2488 propagate_for_debug (rtx insn, rtx last, rtx dest, rtx src)
2490 rtx next, loc, end = NEXT_INSN (BB_END (this_basic_block));
2492 struct rtx_subst_pair p;
2493 p.to = src;
2494 p.adjusted = false;
2496 next = NEXT_INSN (insn);
2497 last = NEXT_INSN (last);
2498 while (next != last && next != end)
2500 insn = next;
2501 next = NEXT_INSN (insn);
2502 if (DEBUG_INSN_P (insn))
2504 loc = simplify_replace_fn_rtx (INSN_VAR_LOCATION_LOC (insn),
2505 dest, propagate_for_debug_subst, &p);
2506 if (loc == INSN_VAR_LOCATION_LOC (insn))
2507 continue;
2508 INSN_VAR_LOCATION_LOC (insn) = loc;
2509 df_insn_rescan (insn);
2514 /* Delete the unconditional jump INSN and adjust the CFG correspondingly.
2515 Note that the INSN should be deleted *after* removing dead edges, so
2516 that the kept edge is the fallthrough edge for a (set (pc) (pc))
2517 but not for a (set (pc) (label_ref FOO)). */
2519 static void
2520 update_cfg_for_uncondjump (rtx insn)
2522 basic_block bb = BLOCK_FOR_INSN (insn);
2523 gcc_assert (BB_END (bb) == insn);
2525 purge_dead_edges (bb);
2527 delete_insn (insn);
2528 if (EDGE_COUNT (bb->succs) == 1)
2530 rtx insn;
2532 single_succ_edge (bb)->flags |= EDGE_FALLTHRU;
2534 /* Remove barriers from the footer if there are any. */
2535 for (insn = bb->il.rtl->footer; insn; insn = NEXT_INSN (insn))
2536 if (BARRIER_P (insn))
2538 if (PREV_INSN (insn))
2539 NEXT_INSN (PREV_INSN (insn)) = NEXT_INSN (insn);
2540 else
2541 bb->il.rtl->footer = NEXT_INSN (insn);
2542 if (NEXT_INSN (insn))
2543 PREV_INSN (NEXT_INSN (insn)) = PREV_INSN (insn);
2545 else if (LABEL_P (insn))
2546 break;
2550 /* Try to combine the insns I0, I1 and I2 into I3.
2551 Here I0, I1 and I2 appear earlier than I3.
2552 I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2555 If we are combining more than two insns and the resulting insn is not
2556 recognized, try splitting it into two insns. If that happens, I2 and I3
2557 are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2558 Otherwise, I0, I1 and I2 are pseudo-deleted.
2560 Return 0 if the combination does not work. Then nothing is changed.
2561 If we did the combination, return the insn at which combine should
2562 resume scanning.
2564 Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
2565 new direct jump instruction.
2567 LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2568 been I3 passed to an earlier try_combine within the same basic
2569 block. */
2571 static rtx
2572 try_combine (rtx i3, rtx i2, rtx i1, rtx i0, int *new_direct_jump_p,
2573 rtx last_combined_insn)
2575 /* New patterns for I3 and I2, respectively. */
2576 rtx newpat, newi2pat = 0;
2577 rtvec newpat_vec_with_clobbers = 0;
2578 int substed_i2 = 0, substed_i1 = 0, substed_i0 = 0;
2579 /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2580 dead. */
2581 int added_sets_0, added_sets_1, added_sets_2;
2582 /* Total number of SETs to put into I3. */
2583 int total_sets;
2584 /* Nonzero if I2's or I1's body now appears in I3. */
2585 int i2_is_used = 0, i1_is_used = 0;
2586 /* INSN_CODEs for new I3, new I2, and user of condition code. */
2587 int insn_code_number, i2_code_number = 0, other_code_number = 0;
2588 /* Contains I3 if the destination of I3 is used in its source, which means
2589 that the old life of I3 is being killed. If that usage is placed into
2590 I2 and not in I3, a REG_DEAD note must be made. */
2591 rtx i3dest_killed = 0;
2592 /* SET_DEST and SET_SRC of I2, I1 and I0. */
2593 rtx i2dest = 0, i2src = 0, i1dest = 0, i1src = 0, i0dest = 0, i0src = 0;
2594 /* Copy of SET_SRC of I1 and I0, if needed. */
2595 rtx i1src_copy = 0, i0src_copy = 0, i0src_copy2 = 0;
2596 /* Set if I2DEST was reused as a scratch register. */
2597 bool i2scratch = false;
2598 /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2599 rtx i0pat = 0, i1pat = 0, i2pat = 0;
2600 /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2601 int i2dest_in_i2src = 0, i1dest_in_i1src = 0, i2dest_in_i1src = 0;
2602 int i0dest_in_i0src = 0, i1dest_in_i0src = 0, i2dest_in_i0src = 0;
2603 int i2dest_killed = 0, i1dest_killed = 0, i0dest_killed = 0;
2604 int i1_feeds_i2_n = 0, i0_feeds_i2_n = 0, i0_feeds_i1_n = 0;
2605 /* Notes that must be added to REG_NOTES in I3 and I2. */
2606 rtx new_i3_notes, new_i2_notes;
2607 /* Notes that we substituted I3 into I2 instead of the normal case. */
2608 int i3_subst_into_i2 = 0;
2609 /* Notes that I1, I2 or I3 is a MULT operation. */
2610 int have_mult = 0;
2611 int swap_i2i3 = 0;
2612 int changed_i3_dest = 0;
2614 int maxreg;
2615 rtx temp;
2616 struct insn_link *link;
2617 rtx other_pat = 0;
2618 rtx new_other_notes;
2619 int i;
2621 /* Only try four-insn combinations when there's high likelihood of
2622 success. Look for simple insns, such as loads of constants or
2623 binary operations involving a constant. */
2624 if (i0)
2626 int i;
2627 int ngood = 0;
2628 int nshift = 0;
2630 if (!flag_expensive_optimizations)
2631 return 0;
2633 for (i = 0; i < 4; i++)
2635 rtx insn = i == 0 ? i0 : i == 1 ? i1 : i == 2 ? i2 : i3;
2636 rtx set = single_set (insn);
2637 rtx src;
2638 if (!set)
2639 continue;
2640 src = SET_SRC (set);
2641 if (CONSTANT_P (src))
2643 ngood += 2;
2644 break;
2646 else if (BINARY_P (src) && CONSTANT_P (XEXP (src, 1)))
2647 ngood++;
2648 else if (GET_CODE (src) == ASHIFT || GET_CODE (src) == ASHIFTRT
2649 || GET_CODE (src) == LSHIFTRT)
2650 nshift++;
2652 if (ngood < 2 && nshift < 2)
2653 return 0;
2656 /* Exit early if one of the insns involved can't be used for
2657 combinations. */
2658 if (cant_combine_insn_p (i3)
2659 || cant_combine_insn_p (i2)
2660 || (i1 && cant_combine_insn_p (i1))
2661 || (i0 && cant_combine_insn_p (i0))
2662 || likely_spilled_retval_p (i3))
2663 return 0;
2665 combine_attempts++;
2666 undobuf.other_insn = 0;
2668 /* Reset the hard register usage information. */
2669 CLEAR_HARD_REG_SET (newpat_used_regs);
2671 if (dump_file && (dump_flags & TDF_DETAILS))
2673 if (i0)
2674 fprintf (dump_file, "\nTrying %d, %d, %d -> %d:\n",
2675 INSN_UID (i0), INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
2676 else if (i1)
2677 fprintf (dump_file, "\nTrying %d, %d -> %d:\n",
2678 INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
2679 else
2680 fprintf (dump_file, "\nTrying %d -> %d:\n",
2681 INSN_UID (i2), INSN_UID (i3));
2684 /* If multiple insns feed into one of I2 or I3, they can be in any
2685 order. To simplify the code below, reorder them in sequence. */
2686 if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i2))
2687 temp = i2, i2 = i0, i0 = temp;
2688 if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i1))
2689 temp = i1, i1 = i0, i0 = temp;
2690 if (i1 && DF_INSN_LUID (i1) > DF_INSN_LUID (i2))
2691 temp = i1, i1 = i2, i2 = temp;
2693 added_links_insn = 0;
2695 /* First check for one important special case that the code below will
2696 not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2697 and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2698 we may be able to replace that destination with the destination of I3.
2699 This occurs in the common code where we compute both a quotient and
2700 remainder into a structure, in which case we want to do the computation
2701 directly into the structure to avoid register-register copies.
2703 Note that this case handles both multiple sets in I2 and also cases
2704 where I2 has a number of CLOBBERs inside the PARALLEL.
2706 We make very conservative checks below and only try to handle the
2707 most common cases of this. For example, we only handle the case
2708 where I2 and I3 are adjacent to avoid making difficult register
2709 usage tests. */
2711 if (i1 == 0 && NONJUMP_INSN_P (i3) && GET_CODE (PATTERN (i3)) == SET
2712 && REG_P (SET_SRC (PATTERN (i3)))
2713 && REGNO (SET_SRC (PATTERN (i3))) >= FIRST_PSEUDO_REGISTER
2714 && find_reg_note (i3, REG_DEAD, SET_SRC (PATTERN (i3)))
2715 && GET_CODE (PATTERN (i2)) == PARALLEL
2716 && ! side_effects_p (SET_DEST (PATTERN (i3)))
2717 /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2718 below would need to check what is inside (and reg_overlap_mentioned_p
2719 doesn't support those codes anyway). Don't allow those destinations;
2720 the resulting insn isn't likely to be recognized anyway. */
2721 && GET_CODE (SET_DEST (PATTERN (i3))) != ZERO_EXTRACT
2722 && GET_CODE (SET_DEST (PATTERN (i3))) != STRICT_LOW_PART
2723 && ! reg_overlap_mentioned_p (SET_SRC (PATTERN (i3)),
2724 SET_DEST (PATTERN (i3)))
2725 && next_active_insn (i2) == i3)
2727 rtx p2 = PATTERN (i2);
2729 /* Make sure that the destination of I3,
2730 which we are going to substitute into one output of I2,
2731 is not used within another output of I2. We must avoid making this:
2732 (parallel [(set (mem (reg 69)) ...)
2733 (set (reg 69) ...)])
2734 which is not well-defined as to order of actions.
2735 (Besides, reload can't handle output reloads for this.)
2737 The problem can also happen if the dest of I3 is a memory ref,
2738 if another dest in I2 is an indirect memory ref. */
2739 for (i = 0; i < XVECLEN (p2, 0); i++)
2740 if ((GET_CODE (XVECEXP (p2, 0, i)) == SET
2741 || GET_CODE (XVECEXP (p2, 0, i)) == CLOBBER)
2742 && reg_overlap_mentioned_p (SET_DEST (PATTERN (i3)),
2743 SET_DEST (XVECEXP (p2, 0, i))))
2744 break;
2746 if (i == XVECLEN (p2, 0))
2747 for (i = 0; i < XVECLEN (p2, 0); i++)
2748 if (GET_CODE (XVECEXP (p2, 0, i)) == SET
2749 && SET_DEST (XVECEXP (p2, 0, i)) == SET_SRC (PATTERN (i3)))
2751 combine_merges++;
2753 subst_insn = i3;
2754 subst_low_luid = DF_INSN_LUID (i2);
2756 added_sets_2 = added_sets_1 = added_sets_0 = 0;
2757 i2src = SET_SRC (XVECEXP (p2, 0, i));
2758 i2dest = SET_DEST (XVECEXP (p2, 0, i));
2759 i2dest_killed = dead_or_set_p (i2, i2dest);
2761 /* Replace the dest in I2 with our dest and make the resulting
2762 insn the new pattern for I3. Then skip to where we validate
2763 the pattern. Everything was set up above. */
2764 SUBST (SET_DEST (XVECEXP (p2, 0, i)), SET_DEST (PATTERN (i3)));
2765 newpat = p2;
2766 i3_subst_into_i2 = 1;
2767 goto validate_replacement;
2771 /* If I2 is setting a pseudo to a constant and I3 is setting some
2772 sub-part of it to another constant, merge them by making a new
2773 constant. */
2774 if (i1 == 0
2775 && (temp = single_set (i2)) != 0
2776 && (CONST_INT_P (SET_SRC (temp))
2777 || GET_CODE (SET_SRC (temp)) == CONST_DOUBLE)
2778 && GET_CODE (PATTERN (i3)) == SET
2779 && (CONST_INT_P (SET_SRC (PATTERN (i3)))
2780 || GET_CODE (SET_SRC (PATTERN (i3))) == CONST_DOUBLE)
2781 && reg_subword_p (SET_DEST (PATTERN (i3)), SET_DEST (temp)))
2783 rtx dest = SET_DEST (PATTERN (i3));
2784 int offset = -1;
2785 int width = 0;
2787 if (GET_CODE (dest) == ZERO_EXTRACT)
2789 if (CONST_INT_P (XEXP (dest, 1))
2790 && CONST_INT_P (XEXP (dest, 2)))
2792 width = INTVAL (XEXP (dest, 1));
2793 offset = INTVAL (XEXP (dest, 2));
2794 dest = XEXP (dest, 0);
2795 if (BITS_BIG_ENDIAN)
2796 offset = GET_MODE_PRECISION (GET_MODE (dest)) - width - offset;
2799 else
2801 if (GET_CODE (dest) == STRICT_LOW_PART)
2802 dest = XEXP (dest, 0);
2803 width = GET_MODE_PRECISION (GET_MODE (dest));
2804 offset = 0;
2807 if (offset >= 0)
2809 /* If this is the low part, we're done. */
2810 if (subreg_lowpart_p (dest))
2812 /* Handle the case where inner is twice the size of outer. */
2813 else if (GET_MODE_PRECISION (GET_MODE (SET_DEST (temp)))
2814 == 2 * GET_MODE_PRECISION (GET_MODE (dest)))
2815 offset += GET_MODE_PRECISION (GET_MODE (dest));
2816 /* Otherwise give up for now. */
2817 else
2818 offset = -1;
2821 if (offset >= 0
2822 && (GET_MODE_PRECISION (GET_MODE (SET_DEST (temp)))
2823 <= HOST_BITS_PER_DOUBLE_INT))
2825 double_int m, o, i;
2826 rtx inner = SET_SRC (PATTERN (i3));
2827 rtx outer = SET_SRC (temp);
2829 o = rtx_to_double_int (outer);
2830 i = rtx_to_double_int (inner);
2832 m = double_int_mask (width);
2833 i = double_int_and (i, m);
2834 m = double_int_lshift (m, offset, HOST_BITS_PER_DOUBLE_INT, false);
2835 i = double_int_lshift (i, offset, HOST_BITS_PER_DOUBLE_INT, false);
2836 o = double_int_ior (double_int_and_not (o, m), i);
2838 combine_merges++;
2839 subst_insn = i3;
2840 subst_low_luid = DF_INSN_LUID (i2);
2841 added_sets_2 = added_sets_1 = added_sets_0 = 0;
2842 i2dest = SET_DEST (temp);
2843 i2dest_killed = dead_or_set_p (i2, i2dest);
2845 /* Replace the source in I2 with the new constant and make the
2846 resulting insn the new pattern for I3. Then skip to where we
2847 validate the pattern. Everything was set up above. */
2848 SUBST (SET_SRC (temp),
2849 immed_double_int_const (o, GET_MODE (SET_DEST (temp))));
2851 newpat = PATTERN (i2);
2853 /* The dest of I3 has been replaced with the dest of I2. */
2854 changed_i3_dest = 1;
2855 goto validate_replacement;
2859 #ifndef HAVE_cc0
2860 /* If we have no I1 and I2 looks like:
2861 (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
2862 (set Y OP)])
2863 make up a dummy I1 that is
2864 (set Y OP)
2865 and change I2 to be
2866 (set (reg:CC X) (compare:CC Y (const_int 0)))
2868 (We can ignore any trailing CLOBBERs.)
2870 This undoes a previous combination and allows us to match a branch-and-
2871 decrement insn. */
2873 if (i1 == 0 && GET_CODE (PATTERN (i2)) == PARALLEL
2874 && XVECLEN (PATTERN (i2), 0) >= 2
2875 && GET_CODE (XVECEXP (PATTERN (i2), 0, 0)) == SET
2876 && (GET_MODE_CLASS (GET_MODE (SET_DEST (XVECEXP (PATTERN (i2), 0, 0))))
2877 == MODE_CC)
2878 && GET_CODE (SET_SRC (XVECEXP (PATTERN (i2), 0, 0))) == COMPARE
2879 && XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 1) == const0_rtx
2880 && GET_CODE (XVECEXP (PATTERN (i2), 0, 1)) == SET
2881 && REG_P (SET_DEST (XVECEXP (PATTERN (i2), 0, 1)))
2882 && rtx_equal_p (XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 0),
2883 SET_SRC (XVECEXP (PATTERN (i2), 0, 1))))
2885 for (i = XVECLEN (PATTERN (i2), 0) - 1; i >= 2; i--)
2886 if (GET_CODE (XVECEXP (PATTERN (i2), 0, i)) != CLOBBER)
2887 break;
2889 if (i == 1)
2891 /* We make I1 with the same INSN_UID as I2. This gives it
2892 the same DF_INSN_LUID for value tracking. Our fake I1 will
2893 never appear in the insn stream so giving it the same INSN_UID
2894 as I2 will not cause a problem. */
2896 i1 = gen_rtx_INSN (VOIDmode, INSN_UID (i2), NULL_RTX, i2,
2897 BLOCK_FOR_INSN (i2), XVECEXP (PATTERN (i2), 0, 1),
2898 INSN_LOCATOR (i2), -1, NULL_RTX);
2900 SUBST (PATTERN (i2), XVECEXP (PATTERN (i2), 0, 0));
2901 SUBST (XEXP (SET_SRC (PATTERN (i2)), 0),
2902 SET_DEST (PATTERN (i1)));
2903 SUBST_LINK (LOG_LINKS (i2), alloc_insn_link (i1, LOG_LINKS (i2)));
2906 #endif
2908 /* Verify that I2 and I1 are valid for combining. */
2909 if (! can_combine_p (i2, i3, i0, i1, NULL_RTX, NULL_RTX, &i2dest, &i2src)
2910 || (i1 && ! can_combine_p (i1, i3, i0, NULL_RTX, i2, NULL_RTX,
2911 &i1dest, &i1src))
2912 || (i0 && ! can_combine_p (i0, i3, NULL_RTX, NULL_RTX, i1, i2,
2913 &i0dest, &i0src)))
2915 undo_all ();
2916 return 0;
2919 /* Record whether I2DEST is used in I2SRC and similarly for the other
2920 cases. Knowing this will help in register status updating below. */
2921 i2dest_in_i2src = reg_overlap_mentioned_p (i2dest, i2src);
2922 i1dest_in_i1src = i1 && reg_overlap_mentioned_p (i1dest, i1src);
2923 i2dest_in_i1src = i1 && reg_overlap_mentioned_p (i2dest, i1src);
2924 i0dest_in_i0src = i0 && reg_overlap_mentioned_p (i0dest, i0src);
2925 i1dest_in_i0src = i0 && reg_overlap_mentioned_p (i1dest, i0src);
2926 i2dest_in_i0src = i0 && reg_overlap_mentioned_p (i2dest, i0src);
2927 i2dest_killed = dead_or_set_p (i2, i2dest);
2928 i1dest_killed = i1 && dead_or_set_p (i1, i1dest);
2929 i0dest_killed = i0 && dead_or_set_p (i0, i0dest);
2931 /* For the earlier insns, determine which of the subsequent ones they
2932 feed. */
2933 i1_feeds_i2_n = i1 && insn_a_feeds_b (i1, i2);
2934 i0_feeds_i1_n = i0 && insn_a_feeds_b (i0, i1);
2935 i0_feeds_i2_n = (i0 && (!i0_feeds_i1_n ? insn_a_feeds_b (i0, i2)
2936 : (!reg_overlap_mentioned_p (i1dest, i0dest)
2937 && reg_overlap_mentioned_p (i0dest, i2src))));
2939 /* Ensure that I3's pattern can be the destination of combines. */
2940 if (! combinable_i3pat (i3, &PATTERN (i3), i2dest, i1dest, i0dest,
2941 i1 && i2dest_in_i1src && !i1_feeds_i2_n,
2942 i0 && ((i2dest_in_i0src && !i0_feeds_i2_n)
2943 || (i1dest_in_i0src && !i0_feeds_i1_n)),
2944 &i3dest_killed))
2946 undo_all ();
2947 return 0;
2950 /* See if any of the insns is a MULT operation. Unless one is, we will
2951 reject a combination that is, since it must be slower. Be conservative
2952 here. */
2953 if (GET_CODE (i2src) == MULT
2954 || (i1 != 0 && GET_CODE (i1src) == MULT)
2955 || (i0 != 0 && GET_CODE (i0src) == MULT)
2956 || (GET_CODE (PATTERN (i3)) == SET
2957 && GET_CODE (SET_SRC (PATTERN (i3))) == MULT))
2958 have_mult = 1;
2960 /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
2961 We used to do this EXCEPT in one case: I3 has a post-inc in an
2962 output operand. However, that exception can give rise to insns like
2963 mov r3,(r3)+
2964 which is a famous insn on the PDP-11 where the value of r3 used as the
2965 source was model-dependent. Avoid this sort of thing. */
2967 #if 0
2968 if (!(GET_CODE (PATTERN (i3)) == SET
2969 && REG_P (SET_SRC (PATTERN (i3)))
2970 && MEM_P (SET_DEST (PATTERN (i3)))
2971 && (GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_INC
2972 || GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_DEC)))
2973 /* It's not the exception. */
2974 #endif
2975 #ifdef AUTO_INC_DEC
2977 rtx link;
2978 for (link = REG_NOTES (i3); link; link = XEXP (link, 1))
2979 if (REG_NOTE_KIND (link) == REG_INC
2980 && (reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i2))
2981 || (i1 != 0
2982 && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i1)))))
2984 undo_all ();
2985 return 0;
2988 #endif
2990 /* See if the SETs in I1 or I2 need to be kept around in the merged
2991 instruction: whenever the value set there is still needed past I3.
2992 For the SETs in I2, this is easy: we see if I2DEST dies or is set in I3.
2994 For the SET in I1, we have two cases: If I1 and I2 independently
2995 feed into I3, the set in I1 needs to be kept around if I1DEST dies
2996 or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
2997 in I1 needs to be kept around unless I1DEST dies or is set in either
2998 I2 or I3. The same consideration applies to I0. */
3000 added_sets_2 = !dead_or_set_p (i3, i2dest);
3002 if (i1)
3003 added_sets_1 = !(dead_or_set_p (i3, i1dest)
3004 || (i1_feeds_i2_n && dead_or_set_p (i2, i1dest)));
3005 else
3006 added_sets_1 = 0;
3008 if (i0)
3009 added_sets_0 = !(dead_or_set_p (i3, i0dest)
3010 || (i0_feeds_i2_n && dead_or_set_p (i2, i0dest))
3011 || (i0_feeds_i1_n && dead_or_set_p (i1, i0dest)));
3012 else
3013 added_sets_0 = 0;
3015 /* We are about to copy insns for the case where they need to be kept
3016 around. Check that they can be copied in the merged instruction. */
3018 if (targetm.cannot_copy_insn_p
3019 && ((added_sets_2 && targetm.cannot_copy_insn_p (i2))
3020 || (i1 && added_sets_1 && targetm.cannot_copy_insn_p (i1))
3021 || (i0 && added_sets_0 && targetm.cannot_copy_insn_p (i0))))
3023 undo_all ();
3024 return 0;
3027 /* If the set in I2 needs to be kept around, we must make a copy of
3028 PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3029 PATTERN (I2), we are only substituting for the original I1DEST, not into
3030 an already-substituted copy. This also prevents making self-referential
3031 rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3032 I2DEST. */
3034 if (added_sets_2)
3036 if (GET_CODE (PATTERN (i2)) == PARALLEL)
3037 i2pat = gen_rtx_SET (VOIDmode, i2dest, copy_rtx (i2src));
3038 else
3039 i2pat = copy_rtx (PATTERN (i2));
3042 if (added_sets_1)
3044 if (GET_CODE (PATTERN (i1)) == PARALLEL)
3045 i1pat = gen_rtx_SET (VOIDmode, i1dest, copy_rtx (i1src));
3046 else
3047 i1pat = copy_rtx (PATTERN (i1));
3050 if (added_sets_0)
3052 if (GET_CODE (PATTERN (i0)) == PARALLEL)
3053 i0pat = gen_rtx_SET (VOIDmode, i0dest, copy_rtx (i0src));
3054 else
3055 i0pat = copy_rtx (PATTERN (i0));
3058 combine_merges++;
3060 /* Substitute in the latest insn for the regs set by the earlier ones. */
3062 maxreg = max_reg_num ();
3064 subst_insn = i3;
3066 #ifndef HAVE_cc0
3067 /* Many machines that don't use CC0 have insns that can both perform an
3068 arithmetic operation and set the condition code. These operations will
3069 be represented as a PARALLEL with the first element of the vector
3070 being a COMPARE of an arithmetic operation with the constant zero.
3071 The second element of the vector will set some pseudo to the result
3072 of the same arithmetic operation. If we simplify the COMPARE, we won't
3073 match such a pattern and so will generate an extra insn. Here we test
3074 for this case, where both the comparison and the operation result are
3075 needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3076 I2SRC. Later we will make the PARALLEL that contains I2. */
3078 if (i1 == 0 && added_sets_2 && GET_CODE (PATTERN (i3)) == SET
3079 && GET_CODE (SET_SRC (PATTERN (i3))) == COMPARE
3080 && CONST_INT_P (XEXP (SET_SRC (PATTERN (i3)), 1))
3081 && rtx_equal_p (XEXP (SET_SRC (PATTERN (i3)), 0), i2dest))
3083 rtx newpat_dest;
3084 rtx *cc_use_loc = NULL, cc_use_insn = NULL_RTX;
3085 rtx op0 = i2src, op1 = XEXP (SET_SRC (PATTERN (i3)), 1);
3086 enum machine_mode compare_mode, orig_compare_mode;
3087 enum rtx_code compare_code = UNKNOWN, orig_compare_code = UNKNOWN;
3089 newpat = PATTERN (i3);
3090 newpat_dest = SET_DEST (newpat);
3091 compare_mode = orig_compare_mode = GET_MODE (newpat_dest);
3093 if (undobuf.other_insn == 0
3094 && (cc_use_loc = find_single_use (SET_DEST (newpat), i3,
3095 &cc_use_insn)))
3097 compare_code = orig_compare_code = GET_CODE (*cc_use_loc);
3098 compare_code = simplify_compare_const (compare_code,
3099 op0, &op1);
3100 #ifdef CANONICALIZE_COMPARISON
3101 CANONICALIZE_COMPARISON (compare_code, op0, op1);
3102 #endif
3105 /* Do the rest only if op1 is const0_rtx, which may be the
3106 result of simplification. */
3107 if (op1 == const0_rtx)
3109 /* If a single use of the CC is found, prepare to modify it
3110 when SELECT_CC_MODE returns a new CC-class mode, or when
3111 the above simplify_compare_const() returned a new comparison
3112 operator. undobuf.other_insn is assigned the CC use insn
3113 when modifying it. */
3114 if (cc_use_loc)
3116 #ifdef SELECT_CC_MODE
3117 enum machine_mode new_mode
3118 = SELECT_CC_MODE (compare_code, op0, op1);
3119 if (new_mode != orig_compare_mode
3120 && can_change_dest_mode (SET_DEST (newpat),
3121 added_sets_2, new_mode))
3123 unsigned int regno = REGNO (newpat_dest);
3124 compare_mode = new_mode;
3125 if (regno < FIRST_PSEUDO_REGISTER)
3126 newpat_dest = gen_rtx_REG (compare_mode, regno);
3127 else
3129 SUBST_MODE (regno_reg_rtx[regno], compare_mode);
3130 newpat_dest = regno_reg_rtx[regno];
3133 #endif
3134 /* Cases for modifying the CC-using comparison. */
3135 if (compare_code != orig_compare_code
3136 /* ??? Do we need to verify the zero rtx? */
3137 && XEXP (*cc_use_loc, 1) == const0_rtx)
3139 /* Replace cc_use_loc with entire new RTX. */
3140 SUBST (*cc_use_loc,
3141 gen_rtx_fmt_ee (compare_code, compare_mode,
3142 newpat_dest, const0_rtx));
3143 undobuf.other_insn = cc_use_insn;
3145 else if (compare_mode != orig_compare_mode)
3147 /* Just replace the CC reg with a new mode. */
3148 SUBST (XEXP (*cc_use_loc, 0), newpat_dest);
3149 undobuf.other_insn = cc_use_insn;
3153 /* Now we modify the current newpat:
3154 First, SET_DEST(newpat) is updated if the CC mode has been
3155 altered. For targets without SELECT_CC_MODE, this should be
3156 optimized away. */
3157 if (compare_mode != orig_compare_mode)
3158 SUBST (SET_DEST (newpat), newpat_dest);
3159 /* This is always done to propagate i2src into newpat. */
3160 SUBST (SET_SRC (newpat),
3161 gen_rtx_COMPARE (compare_mode, op0, op1));
3162 /* Create new version of i2pat if needed; the below PARALLEL
3163 creation needs this to work correctly. */
3164 if (! rtx_equal_p (i2src, op0))
3165 i2pat = gen_rtx_SET (VOIDmode, i2dest, op0);
3166 i2_is_used = 1;
3169 #endif
3171 if (i2_is_used == 0)
3173 /* It is possible that the source of I2 or I1 may be performing
3174 an unneeded operation, such as a ZERO_EXTEND of something
3175 that is known to have the high part zero. Handle that case
3176 by letting subst look at the inner insns.
3178 Another way to do this would be to have a function that tries
3179 to simplify a single insn instead of merging two or more
3180 insns. We don't do this because of the potential of infinite
3181 loops and because of the potential extra memory required.
3182 However, doing it the way we are is a bit of a kludge and
3183 doesn't catch all cases.
3185 But only do this if -fexpensive-optimizations since it slows
3186 things down and doesn't usually win.
3188 This is not done in the COMPARE case above because the
3189 unmodified I2PAT is used in the PARALLEL and so a pattern
3190 with a modified I2SRC would not match. */
3192 if (flag_expensive_optimizations)
3194 /* Pass pc_rtx so no substitutions are done, just
3195 simplifications. */
3196 if (i1)
3198 subst_low_luid = DF_INSN_LUID (i1);
3199 i1src = subst (i1src, pc_rtx, pc_rtx, 0, 0, 0);
3202 subst_low_luid = DF_INSN_LUID (i2);
3203 i2src = subst (i2src, pc_rtx, pc_rtx, 0, 0, 0);
3206 n_occurrences = 0; /* `subst' counts here */
3207 subst_low_luid = DF_INSN_LUID (i2);
3209 /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3210 copy of I2SRC each time we substitute it, in order to avoid creating
3211 self-referential RTL when we will be substituting I1SRC for I1DEST
3212 later. Likewise if I0 feeds into I2, either directly or indirectly
3213 through I1, and I0DEST is in I0SRC. */
3214 newpat = subst (PATTERN (i3), i2dest, i2src, 0, 0,
3215 (i1_feeds_i2_n && i1dest_in_i1src)
3216 || ((i0_feeds_i2_n || (i0_feeds_i1_n && i1_feeds_i2_n))
3217 && i0dest_in_i0src));
3218 substed_i2 = 1;
3220 /* Record whether I2's body now appears within I3's body. */
3221 i2_is_used = n_occurrences;
3224 /* If we already got a failure, don't try to do more. Otherwise, try to
3225 substitute I1 if we have it. */
3227 if (i1 && GET_CODE (newpat) != CLOBBER)
3229 /* Check that an autoincrement side-effect on I1 has not been lost.
3230 This happens if I1DEST is mentioned in I2 and dies there, and
3231 has disappeared from the new pattern. */
3232 if ((FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
3233 && i1_feeds_i2_n
3234 && dead_or_set_p (i2, i1dest)
3235 && !reg_overlap_mentioned_p (i1dest, newpat))
3236 /* Before we can do this substitution, we must redo the test done
3237 above (see detailed comments there) that ensures I1DEST isn't
3238 mentioned in any SETs in NEWPAT that are field assignments. */
3239 || !combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX, NULL_RTX,
3240 0, 0, 0))
3242 undo_all ();
3243 return 0;
3246 n_occurrences = 0;
3247 subst_low_luid = DF_INSN_LUID (i1);
3249 /* If the following substitution will modify I1SRC, make a copy of it
3250 for the case where it is substituted for I1DEST in I2PAT later. */
3251 if (added_sets_2 && i1_feeds_i2_n)
3252 i1src_copy = copy_rtx (i1src);
3254 /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3255 copy of I1SRC each time we substitute it, in order to avoid creating
3256 self-referential RTL when we will be substituting I0SRC for I0DEST
3257 later. */
3258 newpat = subst (newpat, i1dest, i1src, 0, 0,
3259 i0_feeds_i1_n && i0dest_in_i0src);
3260 substed_i1 = 1;
3262 /* Record whether I1's body now appears within I3's body. */
3263 i1_is_used = n_occurrences;
3266 /* Likewise for I0 if we have it. */
3268 if (i0 && GET_CODE (newpat) != CLOBBER)
3270 if ((FIND_REG_INC_NOTE (i0, NULL_RTX) != 0
3271 && ((i0_feeds_i2_n && dead_or_set_p (i2, i0dest))
3272 || (i0_feeds_i1_n && dead_or_set_p (i1, i0dest)))
3273 && !reg_overlap_mentioned_p (i0dest, newpat))
3274 || !combinable_i3pat (NULL_RTX, &newpat, i0dest, NULL_RTX, NULL_RTX,
3275 0, 0, 0))
3277 undo_all ();
3278 return 0;
3281 /* If the following substitution will modify I0SRC, make a copy of it
3282 for the case where it is substituted for I0DEST in I1PAT later. */
3283 if (added_sets_1 && i0_feeds_i1_n)
3284 i0src_copy = copy_rtx (i0src);
3285 /* And a copy for I0DEST in I2PAT substitution. */
3286 if (added_sets_2 && ((i0_feeds_i1_n && i1_feeds_i2_n)
3287 || (i0_feeds_i2_n)))
3288 i0src_copy2 = copy_rtx (i0src);
3290 n_occurrences = 0;
3291 subst_low_luid = DF_INSN_LUID (i0);
3292 newpat = subst (newpat, i0dest, i0src, 0, 0, 0);
3293 substed_i0 = 1;
3296 /* Fail if an autoincrement side-effect has been duplicated. Be careful
3297 to count all the ways that I2SRC and I1SRC can be used. */
3298 if ((FIND_REG_INC_NOTE (i2, NULL_RTX) != 0
3299 && i2_is_used + added_sets_2 > 1)
3300 || (i1 != 0 && FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
3301 && (i1_is_used + added_sets_1 + (added_sets_2 && i1_feeds_i2_n)
3302 > 1))
3303 || (i0 != 0 && FIND_REG_INC_NOTE (i0, NULL_RTX) != 0
3304 && (n_occurrences + added_sets_0
3305 + (added_sets_1 && i0_feeds_i1_n)
3306 + (added_sets_2 && i0_feeds_i2_n)
3307 > 1))
3308 /* Fail if we tried to make a new register. */
3309 || max_reg_num () != maxreg
3310 /* Fail if we couldn't do something and have a CLOBBER. */
3311 || GET_CODE (newpat) == CLOBBER
3312 /* Fail if this new pattern is a MULT and we didn't have one before
3313 at the outer level. */
3314 || (GET_CODE (newpat) == SET && GET_CODE (SET_SRC (newpat)) == MULT
3315 && ! have_mult))
3317 undo_all ();
3318 return 0;
3321 /* If the actions of the earlier insns must be kept
3322 in addition to substituting them into the latest one,
3323 we must make a new PARALLEL for the latest insn
3324 to hold additional the SETs. */
3326 if (added_sets_0 || added_sets_1 || added_sets_2)
3328 int extra_sets = added_sets_0 + added_sets_1 + added_sets_2;
3329 combine_extras++;
3331 if (GET_CODE (newpat) == PARALLEL)
3333 rtvec old = XVEC (newpat, 0);
3334 total_sets = XVECLEN (newpat, 0) + extra_sets;
3335 newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
3336 memcpy (XVEC (newpat, 0)->elem, &old->elem[0],
3337 sizeof (old->elem[0]) * old->num_elem);
3339 else
3341 rtx old = newpat;
3342 total_sets = 1 + extra_sets;
3343 newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
3344 XVECEXP (newpat, 0, 0) = old;
3347 if (added_sets_0)
3348 XVECEXP (newpat, 0, --total_sets) = i0pat;
3350 if (added_sets_1)
3352 rtx t = i1pat;
3353 if (i0_feeds_i1_n)
3354 t = subst (t, i0dest, i0src_copy ? i0src_copy : i0src, 0, 0, 0);
3356 XVECEXP (newpat, 0, --total_sets) = t;
3358 if (added_sets_2)
3360 rtx t = i2pat;
3361 if (i1_feeds_i2_n)
3362 t = subst (t, i1dest, i1src_copy ? i1src_copy : i1src, 0, 0,
3363 i0_feeds_i1_n && i0dest_in_i0src);
3364 if ((i0_feeds_i1_n && i1_feeds_i2_n) || i0_feeds_i2_n)
3365 t = subst (t, i0dest, i0src_copy2 ? i0src_copy2 : i0src, 0, 0, 0);
3367 XVECEXP (newpat, 0, --total_sets) = t;
3371 validate_replacement:
3373 /* Note which hard regs this insn has as inputs. */
3374 mark_used_regs_combine (newpat);
3376 /* If recog_for_combine fails, it strips existing clobbers. If we'll
3377 consider splitting this pattern, we might need these clobbers. */
3378 if (i1 && GET_CODE (newpat) == PARALLEL
3379 && GET_CODE (XVECEXP (newpat, 0, XVECLEN (newpat, 0) - 1)) == CLOBBER)
3381 int len = XVECLEN (newpat, 0);
3383 newpat_vec_with_clobbers = rtvec_alloc (len);
3384 for (i = 0; i < len; i++)
3385 RTVEC_ELT (newpat_vec_with_clobbers, i) = XVECEXP (newpat, 0, i);
3388 /* Is the result of combination a valid instruction? */
3389 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3391 /* If the result isn't valid, see if it is a PARALLEL of two SETs where
3392 the second SET's destination is a register that is unused and isn't
3393 marked as an instruction that might trap in an EH region. In that case,
3394 we just need the first SET. This can occur when simplifying a divmod
3395 insn. We *must* test for this case here because the code below that
3396 splits two independent SETs doesn't handle this case correctly when it
3397 updates the register status.
3399 It's pointless doing this if we originally had two sets, one from
3400 i3, and one from i2. Combining then splitting the parallel results
3401 in the original i2 again plus an invalid insn (which we delete).
3402 The net effect is only to move instructions around, which makes
3403 debug info less accurate.
3405 Also check the case where the first SET's destination is unused.
3406 That would not cause incorrect code, but does cause an unneeded
3407 insn to remain. */
3409 if (insn_code_number < 0
3410 && !(added_sets_2 && i1 == 0)
3411 && GET_CODE (newpat) == PARALLEL
3412 && XVECLEN (newpat, 0) == 2
3413 && GET_CODE (XVECEXP (newpat, 0, 0)) == SET
3414 && GET_CODE (XVECEXP (newpat, 0, 1)) == SET
3415 && asm_noperands (newpat) < 0)
3417 rtx set0 = XVECEXP (newpat, 0, 0);
3418 rtx set1 = XVECEXP (newpat, 0, 1);
3420 if (((REG_P (SET_DEST (set1))
3421 && find_reg_note (i3, REG_UNUSED, SET_DEST (set1)))
3422 || (GET_CODE (SET_DEST (set1)) == SUBREG
3423 && find_reg_note (i3, REG_UNUSED, SUBREG_REG (SET_DEST (set1)))))
3424 && insn_nothrow_p (i3)
3425 && !side_effects_p (SET_SRC (set1)))
3427 newpat = set0;
3428 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3431 else if (((REG_P (SET_DEST (set0))
3432 && find_reg_note (i3, REG_UNUSED, SET_DEST (set0)))
3433 || (GET_CODE (SET_DEST (set0)) == SUBREG
3434 && find_reg_note (i3, REG_UNUSED,
3435 SUBREG_REG (SET_DEST (set0)))))
3436 && insn_nothrow_p (i3)
3437 && !side_effects_p (SET_SRC (set0)))
3439 newpat = set1;
3440 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3442 if (insn_code_number >= 0)
3443 changed_i3_dest = 1;
3447 /* If we were combining three insns and the result is a simple SET
3448 with no ASM_OPERANDS that wasn't recognized, try to split it into two
3449 insns. There are two ways to do this. It can be split using a
3450 machine-specific method (like when you have an addition of a large
3451 constant) or by combine in the function find_split_point. */
3453 if (i1 && insn_code_number < 0 && GET_CODE (newpat) == SET
3454 && asm_noperands (newpat) < 0)
3456 rtx parallel, m_split, *split;
3458 /* See if the MD file can split NEWPAT. If it can't, see if letting it
3459 use I2DEST as a scratch register will help. In the latter case,
3460 convert I2DEST to the mode of the source of NEWPAT if we can. */
3462 m_split = combine_split_insns (newpat, i3);
3464 /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3465 inputs of NEWPAT. */
3467 /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3468 possible to try that as a scratch reg. This would require adding
3469 more code to make it work though. */
3471 if (m_split == 0 && ! reg_overlap_mentioned_p (i2dest, newpat))
3473 enum machine_mode new_mode = GET_MODE (SET_DEST (newpat));
3475 /* First try to split using the original register as a
3476 scratch register. */
3477 parallel = gen_rtx_PARALLEL (VOIDmode,
3478 gen_rtvec (2, newpat,
3479 gen_rtx_CLOBBER (VOIDmode,
3480 i2dest)));
3481 m_split = combine_split_insns (parallel, i3);
3483 /* If that didn't work, try changing the mode of I2DEST if
3484 we can. */
3485 if (m_split == 0
3486 && new_mode != GET_MODE (i2dest)
3487 && new_mode != VOIDmode
3488 && can_change_dest_mode (i2dest, added_sets_2, new_mode))
3490 enum machine_mode old_mode = GET_MODE (i2dest);
3491 rtx ni2dest;
3493 if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
3494 ni2dest = gen_rtx_REG (new_mode, REGNO (i2dest));
3495 else
3497 SUBST_MODE (regno_reg_rtx[REGNO (i2dest)], new_mode);
3498 ni2dest = regno_reg_rtx[REGNO (i2dest)];
3501 parallel = (gen_rtx_PARALLEL
3502 (VOIDmode,
3503 gen_rtvec (2, newpat,
3504 gen_rtx_CLOBBER (VOIDmode,
3505 ni2dest))));
3506 m_split = combine_split_insns (parallel, i3);
3508 if (m_split == 0
3509 && REGNO (i2dest) >= FIRST_PSEUDO_REGISTER)
3511 struct undo *buf;
3513 adjust_reg_mode (regno_reg_rtx[REGNO (i2dest)], old_mode);
3514 buf = undobuf.undos;
3515 undobuf.undos = buf->next;
3516 buf->next = undobuf.frees;
3517 undobuf.frees = buf;
3521 i2scratch = m_split != 0;
3524 /* If recog_for_combine has discarded clobbers, try to use them
3525 again for the split. */
3526 if (m_split == 0 && newpat_vec_with_clobbers)
3528 parallel = gen_rtx_PARALLEL (VOIDmode, newpat_vec_with_clobbers);
3529 m_split = combine_split_insns (parallel, i3);
3532 if (m_split && NEXT_INSN (m_split) == NULL_RTX)
3534 m_split = PATTERN (m_split);
3535 insn_code_number = recog_for_combine (&m_split, i3, &new_i3_notes);
3536 if (insn_code_number >= 0)
3537 newpat = m_split;
3539 else if (m_split && NEXT_INSN (NEXT_INSN (m_split)) == NULL_RTX
3540 && (next_nonnote_nondebug_insn (i2) == i3
3541 || ! use_crosses_set_p (PATTERN (m_split), DF_INSN_LUID (i2))))
3543 rtx i2set, i3set;
3544 rtx newi3pat = PATTERN (NEXT_INSN (m_split));
3545 newi2pat = PATTERN (m_split);
3547 i3set = single_set (NEXT_INSN (m_split));
3548 i2set = single_set (m_split);
3550 i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3552 /* If I2 or I3 has multiple SETs, we won't know how to track
3553 register status, so don't use these insns. If I2's destination
3554 is used between I2 and I3, we also can't use these insns. */
3556 if (i2_code_number >= 0 && i2set && i3set
3557 && (next_nonnote_nondebug_insn (i2) == i3
3558 || ! reg_used_between_p (SET_DEST (i2set), i2, i3)))
3559 insn_code_number = recog_for_combine (&newi3pat, i3,
3560 &new_i3_notes);
3561 if (insn_code_number >= 0)
3562 newpat = newi3pat;
3564 /* It is possible that both insns now set the destination of I3.
3565 If so, we must show an extra use of it. */
3567 if (insn_code_number >= 0)
3569 rtx new_i3_dest = SET_DEST (i3set);
3570 rtx new_i2_dest = SET_DEST (i2set);
3572 while (GET_CODE (new_i3_dest) == ZERO_EXTRACT
3573 || GET_CODE (new_i3_dest) == STRICT_LOW_PART
3574 || GET_CODE (new_i3_dest) == SUBREG)
3575 new_i3_dest = XEXP (new_i3_dest, 0);
3577 while (GET_CODE (new_i2_dest) == ZERO_EXTRACT
3578 || GET_CODE (new_i2_dest) == STRICT_LOW_PART
3579 || GET_CODE (new_i2_dest) == SUBREG)
3580 new_i2_dest = XEXP (new_i2_dest, 0);
3582 if (REG_P (new_i3_dest)
3583 && REG_P (new_i2_dest)
3584 && REGNO (new_i3_dest) == REGNO (new_i2_dest))
3585 INC_REG_N_SETS (REGNO (new_i2_dest), 1);
3589 /* If we can split it and use I2DEST, go ahead and see if that
3590 helps things be recognized. Verify that none of the registers
3591 are set between I2 and I3. */
3592 if (insn_code_number < 0
3593 && (split = find_split_point (&newpat, i3, false)) != 0
3594 #ifdef HAVE_cc0
3595 && REG_P (i2dest)
3596 #endif
3597 /* We need I2DEST in the proper mode. If it is a hard register
3598 or the only use of a pseudo, we can change its mode.
3599 Make sure we don't change a hard register to have a mode that
3600 isn't valid for it, or change the number of registers. */
3601 && (GET_MODE (*split) == GET_MODE (i2dest)
3602 || GET_MODE (*split) == VOIDmode
3603 || can_change_dest_mode (i2dest, added_sets_2,
3604 GET_MODE (*split)))
3605 && (next_nonnote_nondebug_insn (i2) == i3
3606 || ! use_crosses_set_p (*split, DF_INSN_LUID (i2)))
3607 /* We can't overwrite I2DEST if its value is still used by
3608 NEWPAT. */
3609 && ! reg_referenced_p (i2dest, newpat))
3611 rtx newdest = i2dest;
3612 enum rtx_code split_code = GET_CODE (*split);
3613 enum machine_mode split_mode = GET_MODE (*split);
3614 bool subst_done = false;
3615 newi2pat = NULL_RTX;
3617 i2scratch = true;
3619 /* *SPLIT may be part of I2SRC, so make sure we have the
3620 original expression around for later debug processing.
3621 We should not need I2SRC any more in other cases. */
3622 if (MAY_HAVE_DEBUG_INSNS)
3623 i2src = copy_rtx (i2src);
3624 else
3625 i2src = NULL;
3627 /* Get NEWDEST as a register in the proper mode. We have already
3628 validated that we can do this. */
3629 if (GET_MODE (i2dest) != split_mode && split_mode != VOIDmode)
3631 if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
3632 newdest = gen_rtx_REG (split_mode, REGNO (i2dest));
3633 else
3635 SUBST_MODE (regno_reg_rtx[REGNO (i2dest)], split_mode);
3636 newdest = regno_reg_rtx[REGNO (i2dest)];
3640 /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3641 an ASHIFT. This can occur if it was inside a PLUS and hence
3642 appeared to be a memory address. This is a kludge. */
3643 if (split_code == MULT
3644 && CONST_INT_P (XEXP (*split, 1))
3645 && INTVAL (XEXP (*split, 1)) > 0
3646 && (i = exact_log2 (UINTVAL (XEXP (*split, 1)))) >= 0)
3648 SUBST (*split, gen_rtx_ASHIFT (split_mode,
3649 XEXP (*split, 0), GEN_INT (i)));
3650 /* Update split_code because we may not have a multiply
3651 anymore. */
3652 split_code = GET_CODE (*split);
3655 #ifdef INSN_SCHEDULING
3656 /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3657 be written as a ZERO_EXTEND. */
3658 if (split_code == SUBREG && MEM_P (SUBREG_REG (*split)))
3660 #ifdef LOAD_EXTEND_OP
3661 /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3662 what it really is. */
3663 if (LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (*split)))
3664 == SIGN_EXTEND)
3665 SUBST (*split, gen_rtx_SIGN_EXTEND (split_mode,
3666 SUBREG_REG (*split)));
3667 else
3668 #endif
3669 SUBST (*split, gen_rtx_ZERO_EXTEND (split_mode,
3670 SUBREG_REG (*split)));
3672 #endif
3674 /* Attempt to split binary operators using arithmetic identities. */
3675 if (BINARY_P (SET_SRC (newpat))
3676 && split_mode == GET_MODE (SET_SRC (newpat))
3677 && ! side_effects_p (SET_SRC (newpat)))
3679 rtx setsrc = SET_SRC (newpat);
3680 enum machine_mode mode = GET_MODE (setsrc);
3681 enum rtx_code code = GET_CODE (setsrc);
3682 rtx src_op0 = XEXP (setsrc, 0);
3683 rtx src_op1 = XEXP (setsrc, 1);
3685 /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3686 if (rtx_equal_p (src_op0, src_op1))
3688 newi2pat = gen_rtx_SET (VOIDmode, newdest, src_op0);
3689 SUBST (XEXP (setsrc, 0), newdest);
3690 SUBST (XEXP (setsrc, 1), newdest);
3691 subst_done = true;
3693 /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3694 else if ((code == PLUS || code == MULT)
3695 && GET_CODE (src_op0) == code
3696 && GET_CODE (XEXP (src_op0, 0)) == code
3697 && (INTEGRAL_MODE_P (mode)
3698 || (FLOAT_MODE_P (mode)
3699 && flag_unsafe_math_optimizations)))
3701 rtx p = XEXP (XEXP (src_op0, 0), 0);
3702 rtx q = XEXP (XEXP (src_op0, 0), 1);
3703 rtx r = XEXP (src_op0, 1);
3704 rtx s = src_op1;
3706 /* Split both "((X op Y) op X) op Y" and
3707 "((X op Y) op Y) op X" as "T op T" where T is
3708 "X op Y". */
3709 if ((rtx_equal_p (p,r) && rtx_equal_p (q,s))
3710 || (rtx_equal_p (p,s) && rtx_equal_p (q,r)))
3712 newi2pat = gen_rtx_SET (VOIDmode, newdest,
3713 XEXP (src_op0, 0));
3714 SUBST (XEXP (setsrc, 0), newdest);
3715 SUBST (XEXP (setsrc, 1), newdest);
3716 subst_done = true;
3718 /* Split "((X op X) op Y) op Y)" as "T op T" where
3719 T is "X op Y". */
3720 else if (rtx_equal_p (p,q) && rtx_equal_p (r,s))
3722 rtx tmp = simplify_gen_binary (code, mode, p, r);
3723 newi2pat = gen_rtx_SET (VOIDmode, newdest, tmp);
3724 SUBST (XEXP (setsrc, 0), newdest);
3725 SUBST (XEXP (setsrc, 1), newdest);
3726 subst_done = true;
3731 if (!subst_done)
3733 newi2pat = gen_rtx_SET (VOIDmode, newdest, *split);
3734 SUBST (*split, newdest);
3737 i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3739 /* recog_for_combine might have added CLOBBERs to newi2pat.
3740 Make sure NEWPAT does not depend on the clobbered regs. */
3741 if (GET_CODE (newi2pat) == PARALLEL)
3742 for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
3743 if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
3745 rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
3746 if (reg_overlap_mentioned_p (reg, newpat))
3748 undo_all ();
3749 return 0;
3753 /* If the split point was a MULT and we didn't have one before,
3754 don't use one now. */
3755 if (i2_code_number >= 0 && ! (split_code == MULT && ! have_mult))
3756 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3760 /* Check for a case where we loaded from memory in a narrow mode and
3761 then sign extended it, but we need both registers. In that case,
3762 we have a PARALLEL with both loads from the same memory location.
3763 We can split this into a load from memory followed by a register-register
3764 copy. This saves at least one insn, more if register allocation can
3765 eliminate the copy.
3767 We cannot do this if the destination of the first assignment is a
3768 condition code register or cc0. We eliminate this case by making sure
3769 the SET_DEST and SET_SRC have the same mode.
3771 We cannot do this if the destination of the second assignment is
3772 a register that we have already assumed is zero-extended. Similarly
3773 for a SUBREG of such a register. */
3775 else if (i1 && insn_code_number < 0 && asm_noperands (newpat) < 0
3776 && GET_CODE (newpat) == PARALLEL
3777 && XVECLEN (newpat, 0) == 2
3778 && GET_CODE (XVECEXP (newpat, 0, 0)) == SET
3779 && GET_CODE (SET_SRC (XVECEXP (newpat, 0, 0))) == SIGN_EXTEND
3780 && (GET_MODE (SET_DEST (XVECEXP (newpat, 0, 0)))
3781 == GET_MODE (SET_SRC (XVECEXP (newpat, 0, 0))))
3782 && GET_CODE (XVECEXP (newpat, 0, 1)) == SET
3783 && rtx_equal_p (SET_SRC (XVECEXP (newpat, 0, 1)),
3784 XEXP (SET_SRC (XVECEXP (newpat, 0, 0)), 0))
3785 && ! use_crosses_set_p (SET_SRC (XVECEXP (newpat, 0, 1)),
3786 DF_INSN_LUID (i2))
3787 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
3788 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
3789 && ! (temp = SET_DEST (XVECEXP (newpat, 0, 1)),
3790 (REG_P (temp)
3791 && VEC_index (reg_stat_type, reg_stat,
3792 REGNO (temp))->nonzero_bits != 0
3793 && GET_MODE_PRECISION (GET_MODE (temp)) < BITS_PER_WORD
3794 && GET_MODE_PRECISION (GET_MODE (temp)) < HOST_BITS_PER_INT
3795 && (VEC_index (reg_stat_type, reg_stat,
3796 REGNO (temp))->nonzero_bits
3797 != GET_MODE_MASK (word_mode))))
3798 && ! (GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) == SUBREG
3799 && (temp = SUBREG_REG (SET_DEST (XVECEXP (newpat, 0, 1))),
3800 (REG_P (temp)
3801 && VEC_index (reg_stat_type, reg_stat,
3802 REGNO (temp))->nonzero_bits != 0
3803 && GET_MODE_PRECISION (GET_MODE (temp)) < BITS_PER_WORD
3804 && GET_MODE_PRECISION (GET_MODE (temp)) < HOST_BITS_PER_INT
3805 && (VEC_index (reg_stat_type, reg_stat,
3806 REGNO (temp))->nonzero_bits
3807 != GET_MODE_MASK (word_mode)))))
3808 && ! reg_overlap_mentioned_p (SET_DEST (XVECEXP (newpat, 0, 1)),
3809 SET_SRC (XVECEXP (newpat, 0, 1)))
3810 && ! find_reg_note (i3, REG_UNUSED,
3811 SET_DEST (XVECEXP (newpat, 0, 0))))
3813 rtx ni2dest;
3815 newi2pat = XVECEXP (newpat, 0, 0);
3816 ni2dest = SET_DEST (XVECEXP (newpat, 0, 0));
3817 newpat = XVECEXP (newpat, 0, 1);
3818 SUBST (SET_SRC (newpat),
3819 gen_lowpart (GET_MODE (SET_SRC (newpat)), ni2dest));
3820 i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3822 if (i2_code_number >= 0)
3823 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3825 if (insn_code_number >= 0)
3826 swap_i2i3 = 1;
3829 /* Similarly, check for a case where we have a PARALLEL of two independent
3830 SETs but we started with three insns. In this case, we can do the sets
3831 as two separate insns. This case occurs when some SET allows two
3832 other insns to combine, but the destination of that SET is still live. */
3834 else if (i1 && insn_code_number < 0 && asm_noperands (newpat) < 0
3835 && GET_CODE (newpat) == PARALLEL
3836 && XVECLEN (newpat, 0) == 2
3837 && GET_CODE (XVECEXP (newpat, 0, 0)) == SET
3838 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != ZERO_EXTRACT
3839 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != STRICT_LOW_PART
3840 && GET_CODE (XVECEXP (newpat, 0, 1)) == SET
3841 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
3842 && GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
3843 && ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 1)),
3844 XVECEXP (newpat, 0, 0))
3845 && ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 0)),
3846 XVECEXP (newpat, 0, 1))
3847 && ! (contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 0)))
3848 && contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 1)))))
3850 /* Normally, it doesn't matter which of the two is done first,
3851 but the one that references cc0 can't be the second, and
3852 one which uses any regs/memory set in between i2 and i3 can't
3853 be first. */
3854 if (!use_crosses_set_p (SET_SRC (XVECEXP (newpat, 0, 1)),
3855 DF_INSN_LUID (i2))
3856 #ifdef HAVE_cc0
3857 && !reg_referenced_p (cc0_rtx, XVECEXP (newpat, 0, 0))
3858 #endif
3861 newi2pat = XVECEXP (newpat, 0, 1);
3862 newpat = XVECEXP (newpat, 0, 0);
3864 else if (!use_crosses_set_p (SET_SRC (XVECEXP (newpat, 0, 0)),
3865 DF_INSN_LUID (i2))
3866 #ifdef HAVE_cc0
3867 && !reg_referenced_p (cc0_rtx, XVECEXP (newpat, 0, 1))
3868 #endif
3871 newi2pat = XVECEXP (newpat, 0, 0);
3872 newpat = XVECEXP (newpat, 0, 1);
3874 else
3876 undo_all ();
3877 return 0;
3880 i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
3882 if (i2_code_number >= 0)
3884 /* recog_for_combine might have added CLOBBERs to newi2pat.
3885 Make sure NEWPAT does not depend on the clobbered regs. */
3886 if (GET_CODE (newi2pat) == PARALLEL)
3888 for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
3889 if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
3891 rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
3892 if (reg_overlap_mentioned_p (reg, newpat))
3894 undo_all ();
3895 return 0;
3900 insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
3904 /* If it still isn't recognized, fail and change things back the way they
3905 were. */
3906 if ((insn_code_number < 0
3907 /* Is the result a reasonable ASM_OPERANDS? */
3908 && (! check_asm_operands (newpat) || added_sets_1 || added_sets_2)))
3910 undo_all ();
3911 return 0;
3914 /* If we had to change another insn, make sure it is valid also. */
3915 if (undobuf.other_insn)
3917 CLEAR_HARD_REG_SET (newpat_used_regs);
3919 other_pat = PATTERN (undobuf.other_insn);
3920 other_code_number = recog_for_combine (&other_pat, undobuf.other_insn,
3921 &new_other_notes);
3923 if (other_code_number < 0 && ! check_asm_operands (other_pat))
3925 undo_all ();
3926 return 0;
3930 #ifdef HAVE_cc0
3931 /* If I2 is the CC0 setter and I3 is the CC0 user then check whether
3932 they are adjacent to each other or not. */
3934 rtx p = prev_nonnote_insn (i3);
3935 if (p && p != i2 && NONJUMP_INSN_P (p) && newi2pat
3936 && sets_cc0_p (newi2pat))
3938 undo_all ();
3939 return 0;
3942 #endif
3944 /* Only allow this combination if insn_rtx_costs reports that the
3945 replacement instructions are cheaper than the originals. */
3946 if (!combine_validate_cost (i0, i1, i2, i3, newpat, newi2pat, other_pat))
3948 undo_all ();
3949 return 0;
3952 if (MAY_HAVE_DEBUG_INSNS)
3954 struct undo *undo;
3956 for (undo = undobuf.undos; undo; undo = undo->next)
3957 if (undo->kind == UNDO_MODE)
3959 rtx reg = *undo->where.r;
3960 enum machine_mode new_mode = GET_MODE (reg);
3961 enum machine_mode old_mode = undo->old_contents.m;
3963 /* Temporarily revert mode back. */
3964 adjust_reg_mode (reg, old_mode);
3966 if (reg == i2dest && i2scratch)
3968 /* If we used i2dest as a scratch register with a
3969 different mode, substitute it for the original
3970 i2src while its original mode is temporarily
3971 restored, and then clear i2scratch so that we don't
3972 do it again later. */
3973 propagate_for_debug (i2, last_combined_insn, reg, i2src);
3974 i2scratch = false;
3975 /* Put back the new mode. */
3976 adjust_reg_mode (reg, new_mode);
3978 else
3980 rtx tempreg = gen_raw_REG (old_mode, REGNO (reg));
3981 rtx first, last;
3983 if (reg == i2dest)
3985 first = i2;
3986 last = last_combined_insn;
3988 else
3990 first = i3;
3991 last = undobuf.other_insn;
3992 gcc_assert (last);
3993 if (DF_INSN_LUID (last)
3994 < DF_INSN_LUID (last_combined_insn))
3995 last = last_combined_insn;
3998 /* We're dealing with a reg that changed mode but not
3999 meaning, so we want to turn it into a subreg for
4000 the new mode. However, because of REG sharing and
4001 because its mode had already changed, we have to do
4002 it in two steps. First, replace any debug uses of
4003 reg, with its original mode temporarily restored,
4004 with this copy we have created; then, replace the
4005 copy with the SUBREG of the original shared reg,
4006 once again changed to the new mode. */
4007 propagate_for_debug (first, last, reg, tempreg);
4008 adjust_reg_mode (reg, new_mode);
4009 propagate_for_debug (first, last, tempreg,
4010 lowpart_subreg (old_mode, reg, new_mode));
4015 /* If we will be able to accept this, we have made a
4016 change to the destination of I3. This requires us to
4017 do a few adjustments. */
4019 if (changed_i3_dest)
4021 PATTERN (i3) = newpat;
4022 adjust_for_new_dest (i3);
4025 /* We now know that we can do this combination. Merge the insns and
4026 update the status of registers and LOG_LINKS. */
4028 if (undobuf.other_insn)
4030 rtx note, next;
4032 PATTERN (undobuf.other_insn) = other_pat;
4034 /* If any of the notes in OTHER_INSN were REG_UNUSED, ensure that they
4035 are still valid. Then add any non-duplicate notes added by
4036 recog_for_combine. */
4037 for (note = REG_NOTES (undobuf.other_insn); note; note = next)
4039 next = XEXP (note, 1);
4041 if (REG_NOTE_KIND (note) == REG_UNUSED
4042 && ! reg_set_p (XEXP (note, 0), PATTERN (undobuf.other_insn)))
4043 remove_note (undobuf.other_insn, note);
4046 distribute_notes (new_other_notes, undobuf.other_insn,
4047 undobuf.other_insn, NULL_RTX, NULL_RTX, NULL_RTX,
4048 NULL_RTX);
4051 if (swap_i2i3)
4053 rtx insn;
4054 struct insn_link *link;
4055 rtx ni2dest;
4057 /* I3 now uses what used to be its destination and which is now
4058 I2's destination. This requires us to do a few adjustments. */
4059 PATTERN (i3) = newpat;
4060 adjust_for_new_dest (i3);
4062 /* We need a LOG_LINK from I3 to I2. But we used to have one,
4063 so we still will.
4065 However, some later insn might be using I2's dest and have
4066 a LOG_LINK pointing at I3. We must remove this link.
4067 The simplest way to remove the link is to point it at I1,
4068 which we know will be a NOTE. */
4070 /* newi2pat is usually a SET here; however, recog_for_combine might
4071 have added some clobbers. */
4072 if (GET_CODE (newi2pat) == PARALLEL)
4073 ni2dest = SET_DEST (XVECEXP (newi2pat, 0, 0));
4074 else
4075 ni2dest = SET_DEST (newi2pat);
4077 for (insn = NEXT_INSN (i3);
4078 insn && (this_basic_block->next_bb == EXIT_BLOCK_PTR
4079 || insn != BB_HEAD (this_basic_block->next_bb));
4080 insn = NEXT_INSN (insn))
4082 if (INSN_P (insn) && reg_referenced_p (ni2dest, PATTERN (insn)))
4084 FOR_EACH_LOG_LINK (link, insn)
4085 if (link->insn == i3)
4086 link->insn = i1;
4088 break;
4094 rtx i3notes, i2notes, i1notes = 0, i0notes = 0;
4095 struct insn_link *i3links, *i2links, *i1links = 0, *i0links = 0;
4096 rtx midnotes = 0;
4097 int from_luid;
4098 /* Compute which registers we expect to eliminate. newi2pat may be setting
4099 either i3dest or i2dest, so we must check it. Also, i1dest may be the
4100 same as i3dest, in which case newi2pat may be setting i1dest. */
4101 rtx elim_i2 = ((newi2pat && reg_set_p (i2dest, newi2pat))
4102 || i2dest_in_i2src || i2dest_in_i1src || i2dest_in_i0src
4103 || !i2dest_killed
4104 ? 0 : i2dest);
4105 rtx elim_i1 = (i1 == 0 || i1dest_in_i1src || i1dest_in_i0src
4106 || (newi2pat && reg_set_p (i1dest, newi2pat))
4107 || !i1dest_killed
4108 ? 0 : i1dest);
4109 rtx elim_i0 = (i0 == 0 || i0dest_in_i0src
4110 || (newi2pat && reg_set_p (i0dest, newi2pat))
4111 || !i0dest_killed
4112 ? 0 : i0dest);
4114 /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4115 clear them. */
4116 i3notes = REG_NOTES (i3), i3links = LOG_LINKS (i3);
4117 i2notes = REG_NOTES (i2), i2links = LOG_LINKS (i2);
4118 if (i1)
4119 i1notes = REG_NOTES (i1), i1links = LOG_LINKS (i1);
4120 if (i0)
4121 i0notes = REG_NOTES (i0), i0links = LOG_LINKS (i0);
4123 /* Ensure that we do not have something that should not be shared but
4124 occurs multiple times in the new insns. Check this by first
4125 resetting all the `used' flags and then copying anything is shared. */
4127 reset_used_flags (i3notes);
4128 reset_used_flags (i2notes);
4129 reset_used_flags (i1notes);
4130 reset_used_flags (i0notes);
4131 reset_used_flags (newpat);
4132 reset_used_flags (newi2pat);
4133 if (undobuf.other_insn)
4134 reset_used_flags (PATTERN (undobuf.other_insn));
4136 i3notes = copy_rtx_if_shared (i3notes);
4137 i2notes = copy_rtx_if_shared (i2notes);
4138 i1notes = copy_rtx_if_shared (i1notes);
4139 i0notes = copy_rtx_if_shared (i0notes);
4140 newpat = copy_rtx_if_shared (newpat);
4141 newi2pat = copy_rtx_if_shared (newi2pat);
4142 if (undobuf.other_insn)
4143 reset_used_flags (PATTERN (undobuf.other_insn));
4145 INSN_CODE (i3) = insn_code_number;
4146 PATTERN (i3) = newpat;
4148 if (CALL_P (i3) && CALL_INSN_FUNCTION_USAGE (i3))
4150 rtx call_usage = CALL_INSN_FUNCTION_USAGE (i3);
4152 reset_used_flags (call_usage);
4153 call_usage = copy_rtx (call_usage);
4155 if (substed_i2)
4157 /* I2SRC must still be meaningful at this point. Some splitting
4158 operations can invalidate I2SRC, but those operations do not
4159 apply to calls. */
4160 gcc_assert (i2src);
4161 replace_rtx (call_usage, i2dest, i2src);
4164 if (substed_i1)
4165 replace_rtx (call_usage, i1dest, i1src);
4166 if (substed_i0)
4167 replace_rtx (call_usage, i0dest, i0src);
4169 CALL_INSN_FUNCTION_USAGE (i3) = call_usage;
4172 if (undobuf.other_insn)
4173 INSN_CODE (undobuf.other_insn) = other_code_number;
4175 /* We had one special case above where I2 had more than one set and
4176 we replaced a destination of one of those sets with the destination
4177 of I3. In that case, we have to update LOG_LINKS of insns later
4178 in this basic block. Note that this (expensive) case is rare.
4180 Also, in this case, we must pretend that all REG_NOTEs for I2
4181 actually came from I3, so that REG_UNUSED notes from I2 will be
4182 properly handled. */
4184 if (i3_subst_into_i2)
4186 for (i = 0; i < XVECLEN (PATTERN (i2), 0); i++)
4187 if ((GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == SET
4188 || GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == CLOBBER)
4189 && REG_P (SET_DEST (XVECEXP (PATTERN (i2), 0, i)))
4190 && SET_DEST (XVECEXP (PATTERN (i2), 0, i)) != i2dest
4191 && ! find_reg_note (i2, REG_UNUSED,
4192 SET_DEST (XVECEXP (PATTERN (i2), 0, i))))
4193 for (temp = NEXT_INSN (i2);
4194 temp && (this_basic_block->next_bb == EXIT_BLOCK_PTR
4195 || BB_HEAD (this_basic_block) != temp);
4196 temp = NEXT_INSN (temp))
4197 if (temp != i3 && INSN_P (temp))
4198 FOR_EACH_LOG_LINK (link, temp)
4199 if (link->insn == i2)
4200 link->insn = i3;
4202 if (i3notes)
4204 rtx link = i3notes;
4205 while (XEXP (link, 1))
4206 link = XEXP (link, 1);
4207 XEXP (link, 1) = i2notes;
4209 else
4210 i3notes = i2notes;
4211 i2notes = 0;
4214 LOG_LINKS (i3) = NULL;
4215 REG_NOTES (i3) = 0;
4216 LOG_LINKS (i2) = NULL;
4217 REG_NOTES (i2) = 0;
4219 if (newi2pat)
4221 if (MAY_HAVE_DEBUG_INSNS && i2scratch)
4222 propagate_for_debug (i2, last_combined_insn, i2dest, i2src);
4223 INSN_CODE (i2) = i2_code_number;
4224 PATTERN (i2) = newi2pat;
4226 else
4228 if (MAY_HAVE_DEBUG_INSNS && i2src)
4229 propagate_for_debug (i2, last_combined_insn, i2dest, i2src);
4230 SET_INSN_DELETED (i2);
4233 if (i1)
4235 LOG_LINKS (i1) = NULL;
4236 REG_NOTES (i1) = 0;
4237 if (MAY_HAVE_DEBUG_INSNS)
4238 propagate_for_debug (i1, last_combined_insn, i1dest, i1src);
4239 SET_INSN_DELETED (i1);
4242 if (i0)
4244 LOG_LINKS (i0) = NULL;
4245 REG_NOTES (i0) = 0;
4246 if (MAY_HAVE_DEBUG_INSNS)
4247 propagate_for_debug (i0, last_combined_insn, i0dest, i0src);
4248 SET_INSN_DELETED (i0);
4251 /* Get death notes for everything that is now used in either I3 or
4252 I2 and used to die in a previous insn. If we built two new
4253 patterns, move from I1 to I2 then I2 to I3 so that we get the
4254 proper movement on registers that I2 modifies. */
4256 if (i0)
4257 from_luid = DF_INSN_LUID (i0);
4258 else if (i1)
4259 from_luid = DF_INSN_LUID (i1);
4260 else
4261 from_luid = DF_INSN_LUID (i2);
4262 if (newi2pat)
4263 move_deaths (newi2pat, NULL_RTX, from_luid, i2, &midnotes);
4264 move_deaths (newpat, newi2pat, from_luid, i3, &midnotes);
4266 /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4267 if (i3notes)
4268 distribute_notes (i3notes, i3, i3, newi2pat ? i2 : NULL_RTX,
4269 elim_i2, elim_i1, elim_i0);
4270 if (i2notes)
4271 distribute_notes (i2notes, i2, i3, newi2pat ? i2 : NULL_RTX,
4272 elim_i2, elim_i1, elim_i0);
4273 if (i1notes)
4274 distribute_notes (i1notes, i1, i3, newi2pat ? i2 : NULL_RTX,
4275 elim_i2, elim_i1, elim_i0);
4276 if (i0notes)
4277 distribute_notes (i0notes, i0, i3, newi2pat ? i2 : NULL_RTX,
4278 elim_i2, elim_i1, elim_i0);
4279 if (midnotes)
4280 distribute_notes (midnotes, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
4281 elim_i2, elim_i1, elim_i0);
4283 /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4284 know these are REG_UNUSED and want them to go to the desired insn,
4285 so we always pass it as i3. */
4287 if (newi2pat && new_i2_notes)
4288 distribute_notes (new_i2_notes, i2, i2, NULL_RTX, NULL_RTX, NULL_RTX,
4289 NULL_RTX);
4291 if (new_i3_notes)
4292 distribute_notes (new_i3_notes, i3, i3, NULL_RTX, NULL_RTX, NULL_RTX,
4293 NULL_RTX);
4295 /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4296 put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4297 I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4298 in that case, it might delete I2. Similarly for I2 and I1.
4299 Show an additional death due to the REG_DEAD note we make here. If
4300 we discard it in distribute_notes, we will decrement it again. */
4302 if (i3dest_killed)
4304 if (newi2pat && reg_set_p (i3dest_killed, newi2pat))
4305 distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
4306 NULL_RTX),
4307 NULL_RTX, i2, NULL_RTX, elim_i2, elim_i1, elim_i0);
4308 else
4309 distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
4310 NULL_RTX),
4311 NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
4312 elim_i2, elim_i1, elim_i0);
4315 if (i2dest_in_i2src)
4317 rtx new_note = alloc_reg_note (REG_DEAD, i2dest, NULL_RTX);
4318 if (newi2pat && reg_set_p (i2dest, newi2pat))
4319 distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
4320 NULL_RTX, NULL_RTX);
4321 else
4322 distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
4323 NULL_RTX, NULL_RTX, NULL_RTX);
4326 if (i1dest_in_i1src)
4328 rtx new_note = alloc_reg_note (REG_DEAD, i1dest, NULL_RTX);
4329 if (newi2pat && reg_set_p (i1dest, newi2pat))
4330 distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
4331 NULL_RTX, NULL_RTX);
4332 else
4333 distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
4334 NULL_RTX, NULL_RTX, NULL_RTX);
4337 if (i0dest_in_i0src)
4339 rtx new_note = alloc_reg_note (REG_DEAD, i0dest, NULL_RTX);
4340 if (newi2pat && reg_set_p (i0dest, newi2pat))
4341 distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
4342 NULL_RTX, NULL_RTX);
4343 else
4344 distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
4345 NULL_RTX, NULL_RTX, NULL_RTX);
4348 distribute_links (i3links);
4349 distribute_links (i2links);
4350 distribute_links (i1links);
4351 distribute_links (i0links);
4353 if (REG_P (i2dest))
4355 struct insn_link *link;
4356 rtx i2_insn = 0, i2_val = 0, set;
4358 /* The insn that used to set this register doesn't exist, and
4359 this life of the register may not exist either. See if one of
4360 I3's links points to an insn that sets I2DEST. If it does,
4361 that is now the last known value for I2DEST. If we don't update
4362 this and I2 set the register to a value that depended on its old
4363 contents, we will get confused. If this insn is used, thing
4364 will be set correctly in combine_instructions. */
4365 FOR_EACH_LOG_LINK (link, i3)
4366 if ((set = single_set (link->insn)) != 0
4367 && rtx_equal_p (i2dest, SET_DEST (set)))
4368 i2_insn = link->insn, i2_val = SET_SRC (set);
4370 record_value_for_reg (i2dest, i2_insn, i2_val);
4372 /* If the reg formerly set in I2 died only once and that was in I3,
4373 zero its use count so it won't make `reload' do any work. */
4374 if (! added_sets_2
4375 && (newi2pat == 0 || ! reg_mentioned_p (i2dest, newi2pat))
4376 && ! i2dest_in_i2src)
4377 INC_REG_N_SETS (REGNO (i2dest), -1);
4380 if (i1 && REG_P (i1dest))
4382 struct insn_link *link;
4383 rtx i1_insn = 0, i1_val = 0, set;
4385 FOR_EACH_LOG_LINK (link, i3)
4386 if ((set = single_set (link->insn)) != 0
4387 && rtx_equal_p (i1dest, SET_DEST (set)))
4388 i1_insn = link->insn, i1_val = SET_SRC (set);
4390 record_value_for_reg (i1dest, i1_insn, i1_val);
4392 if (! added_sets_1 && ! i1dest_in_i1src)
4393 INC_REG_N_SETS (REGNO (i1dest), -1);
4396 if (i0 && REG_P (i0dest))
4398 struct insn_link *link;
4399 rtx i0_insn = 0, i0_val = 0, set;
4401 FOR_EACH_LOG_LINK (link, i3)
4402 if ((set = single_set (link->insn)) != 0
4403 && rtx_equal_p (i0dest, SET_DEST (set)))
4404 i0_insn = link->insn, i0_val = SET_SRC (set);
4406 record_value_for_reg (i0dest, i0_insn, i0_val);
4408 if (! added_sets_0 && ! i0dest_in_i0src)
4409 INC_REG_N_SETS (REGNO (i0dest), -1);
4412 /* Update reg_stat[].nonzero_bits et al for any changes that may have
4413 been made to this insn. The order of
4414 set_nonzero_bits_and_sign_copies() is important. Because newi2pat
4415 can affect nonzero_bits of newpat */
4416 if (newi2pat)
4417 note_stores (newi2pat, set_nonzero_bits_and_sign_copies, NULL);
4418 note_stores (newpat, set_nonzero_bits_and_sign_copies, NULL);
4421 if (undobuf.other_insn != NULL_RTX)
4423 if (dump_file)
4425 fprintf (dump_file, "modifying other_insn ");
4426 dump_insn_slim (dump_file, undobuf.other_insn);
4428 df_insn_rescan (undobuf.other_insn);
4431 if (i0 && !(NOTE_P(i0) && (NOTE_KIND (i0) == NOTE_INSN_DELETED)))
4433 if (dump_file)
4435 fprintf (dump_file, "modifying insn i1 ");
4436 dump_insn_slim (dump_file, i0);
4438 df_insn_rescan (i0);
4441 if (i1 && !(NOTE_P(i1) && (NOTE_KIND (i1) == NOTE_INSN_DELETED)))
4443 if (dump_file)
4445 fprintf (dump_file, "modifying insn i1 ");
4446 dump_insn_slim (dump_file, i1);
4448 df_insn_rescan (i1);
4451 if (i2 && !(NOTE_P(i2) && (NOTE_KIND (i2) == NOTE_INSN_DELETED)))
4453 if (dump_file)
4455 fprintf (dump_file, "modifying insn i2 ");
4456 dump_insn_slim (dump_file, i2);
4458 df_insn_rescan (i2);
4461 if (i3 && !(NOTE_P(i3) && (NOTE_KIND (i3) == NOTE_INSN_DELETED)))
4463 if (dump_file)
4465 fprintf (dump_file, "modifying insn i3 ");
4466 dump_insn_slim (dump_file, i3);
4468 df_insn_rescan (i3);
4471 /* Set new_direct_jump_p if a new return or simple jump instruction
4472 has been created. Adjust the CFG accordingly. */
4474 if (returnjump_p (i3) || any_uncondjump_p (i3))
4476 *new_direct_jump_p = 1;
4477 mark_jump_label (PATTERN (i3), i3, 0);
4478 update_cfg_for_uncondjump (i3);
4481 if (undobuf.other_insn != NULL_RTX
4482 && (returnjump_p (undobuf.other_insn)
4483 || any_uncondjump_p (undobuf.other_insn)))
4485 *new_direct_jump_p = 1;
4486 update_cfg_for_uncondjump (undobuf.other_insn);
4489 /* A noop might also need cleaning up of CFG, if it comes from the
4490 simplification of a jump. */
4491 if (JUMP_P (i3)
4492 && GET_CODE (newpat) == SET
4493 && SET_SRC (newpat) == pc_rtx
4494 && SET_DEST (newpat) == pc_rtx)
4496 *new_direct_jump_p = 1;
4497 update_cfg_for_uncondjump (i3);
4500 if (undobuf.other_insn != NULL_RTX
4501 && JUMP_P (undobuf.other_insn)
4502 && GET_CODE (PATTERN (undobuf.other_insn)) == SET
4503 && SET_SRC (PATTERN (undobuf.other_insn)) == pc_rtx
4504 && SET_DEST (PATTERN (undobuf.other_insn)) == pc_rtx)
4506 *new_direct_jump_p = 1;
4507 update_cfg_for_uncondjump (undobuf.other_insn);
4510 combine_successes++;
4511 undo_commit ();
4513 if (added_links_insn
4514 && (newi2pat == 0 || DF_INSN_LUID (added_links_insn) < DF_INSN_LUID (i2))
4515 && DF_INSN_LUID (added_links_insn) < DF_INSN_LUID (i3))
4516 return added_links_insn;
4517 else
4518 return newi2pat ? i2 : i3;
4521 /* Undo all the modifications recorded in undobuf. */
4523 static void
4524 undo_all (void)
4526 struct undo *undo, *next;
4528 for (undo = undobuf.undos; undo; undo = next)
4530 next = undo->next;
4531 switch (undo->kind)
4533 case UNDO_RTX:
4534 *undo->where.r = undo->old_contents.r;
4535 break;
4536 case UNDO_INT:
4537 *undo->where.i = undo->old_contents.i;
4538 break;
4539 case UNDO_MODE:
4540 adjust_reg_mode (*undo->where.r, undo->old_contents.m);
4541 break;
4542 case UNDO_LINKS:
4543 *undo->where.l = undo->old_contents.l;
4544 break;
4545 default:
4546 gcc_unreachable ();
4549 undo->next = undobuf.frees;
4550 undobuf.frees = undo;
4553 undobuf.undos = 0;
4556 /* We've committed to accepting the changes we made. Move all
4557 of the undos to the free list. */
4559 static void
4560 undo_commit (void)
4562 struct undo *undo, *next;
4564 for (undo = undobuf.undos; undo; undo = next)
4566 next = undo->next;
4567 undo->next = undobuf.frees;
4568 undobuf.frees = undo;
4570 undobuf.undos = 0;
4573 /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4574 where we have an arithmetic expression and return that point. LOC will
4575 be inside INSN.
4577 try_combine will call this function to see if an insn can be split into
4578 two insns. */
4580 static rtx *
4581 find_split_point (rtx *loc, rtx insn, bool set_src)
4583 rtx x = *loc;
4584 enum rtx_code code = GET_CODE (x);
4585 rtx *split;
4586 unsigned HOST_WIDE_INT len = 0;
4587 HOST_WIDE_INT pos = 0;
4588 int unsignedp = 0;
4589 rtx inner = NULL_RTX;
4591 /* First special-case some codes. */
4592 switch (code)
4594 case SUBREG:
4595 #ifdef INSN_SCHEDULING
4596 /* If we are making a paradoxical SUBREG invalid, it becomes a split
4597 point. */
4598 if (MEM_P (SUBREG_REG (x)))
4599 return loc;
4600 #endif
4601 return find_split_point (&SUBREG_REG (x), insn, false);
4603 case MEM:
4604 #ifdef HAVE_lo_sum
4605 /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4606 using LO_SUM and HIGH. */
4607 if (GET_CODE (XEXP (x, 0)) == CONST
4608 || GET_CODE (XEXP (x, 0)) == SYMBOL_REF)
4610 enum machine_mode address_mode
4611 = targetm.addr_space.address_mode (MEM_ADDR_SPACE (x));
4613 SUBST (XEXP (x, 0),
4614 gen_rtx_LO_SUM (address_mode,
4615 gen_rtx_HIGH (address_mode, XEXP (x, 0)),
4616 XEXP (x, 0)));
4617 return &XEXP (XEXP (x, 0), 0);
4619 #endif
4621 /* If we have a PLUS whose second operand is a constant and the
4622 address is not valid, perhaps will can split it up using
4623 the machine-specific way to split large constants. We use
4624 the first pseudo-reg (one of the virtual regs) as a placeholder;
4625 it will not remain in the result. */
4626 if (GET_CODE (XEXP (x, 0)) == PLUS
4627 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
4628 && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
4629 MEM_ADDR_SPACE (x)))
4631 rtx reg = regno_reg_rtx[FIRST_PSEUDO_REGISTER];
4632 rtx seq = combine_split_insns (gen_rtx_SET (VOIDmode, reg,
4633 XEXP (x, 0)),
4634 subst_insn);
4636 /* This should have produced two insns, each of which sets our
4637 placeholder. If the source of the second is a valid address,
4638 we can make put both sources together and make a split point
4639 in the middle. */
4641 if (seq
4642 && NEXT_INSN (seq) != NULL_RTX
4643 && NEXT_INSN (NEXT_INSN (seq)) == NULL_RTX
4644 && NONJUMP_INSN_P (seq)
4645 && GET_CODE (PATTERN (seq)) == SET
4646 && SET_DEST (PATTERN (seq)) == reg
4647 && ! reg_mentioned_p (reg,
4648 SET_SRC (PATTERN (seq)))
4649 && NONJUMP_INSN_P (NEXT_INSN (seq))
4650 && GET_CODE (PATTERN (NEXT_INSN (seq))) == SET
4651 && SET_DEST (PATTERN (NEXT_INSN (seq))) == reg
4652 && memory_address_addr_space_p
4653 (GET_MODE (x), SET_SRC (PATTERN (NEXT_INSN (seq))),
4654 MEM_ADDR_SPACE (x)))
4656 rtx src1 = SET_SRC (PATTERN (seq));
4657 rtx src2 = SET_SRC (PATTERN (NEXT_INSN (seq)));
4659 /* Replace the placeholder in SRC2 with SRC1. If we can
4660 find where in SRC2 it was placed, that can become our
4661 split point and we can replace this address with SRC2.
4662 Just try two obvious places. */
4664 src2 = replace_rtx (src2, reg, src1);
4665 split = 0;
4666 if (XEXP (src2, 0) == src1)
4667 split = &XEXP (src2, 0);
4668 else if (GET_RTX_FORMAT (GET_CODE (XEXP (src2, 0)))[0] == 'e'
4669 && XEXP (XEXP (src2, 0), 0) == src1)
4670 split = &XEXP (XEXP (src2, 0), 0);
4672 if (split)
4674 SUBST (XEXP (x, 0), src2);
4675 return split;
4679 /* If that didn't work, perhaps the first operand is complex and
4680 needs to be computed separately, so make a split point there.
4681 This will occur on machines that just support REG + CONST
4682 and have a constant moved through some previous computation. */
4684 else if (!OBJECT_P (XEXP (XEXP (x, 0), 0))
4685 && ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
4686 && OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
4687 return &XEXP (XEXP (x, 0), 0);
4690 /* If we have a PLUS whose first operand is complex, try computing it
4691 separately by making a split there. */
4692 if (GET_CODE (XEXP (x, 0)) == PLUS
4693 && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
4694 MEM_ADDR_SPACE (x))
4695 && ! OBJECT_P (XEXP (XEXP (x, 0), 0))
4696 && ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
4697 && OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
4698 return &XEXP (XEXP (x, 0), 0);
4699 break;
4701 case SET:
4702 #ifdef HAVE_cc0
4703 /* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
4704 ZERO_EXTRACT, the most likely reason why this doesn't match is that
4705 we need to put the operand into a register. So split at that
4706 point. */
4708 if (SET_DEST (x) == cc0_rtx
4709 && GET_CODE (SET_SRC (x)) != COMPARE
4710 && GET_CODE (SET_SRC (x)) != ZERO_EXTRACT
4711 && !OBJECT_P (SET_SRC (x))
4712 && ! (GET_CODE (SET_SRC (x)) == SUBREG
4713 && OBJECT_P (SUBREG_REG (SET_SRC (x)))))
4714 return &SET_SRC (x);
4715 #endif
4717 /* See if we can split SET_SRC as it stands. */
4718 split = find_split_point (&SET_SRC (x), insn, true);
4719 if (split && split != &SET_SRC (x))
4720 return split;
4722 /* See if we can split SET_DEST as it stands. */
4723 split = find_split_point (&SET_DEST (x), insn, false);
4724 if (split && split != &SET_DEST (x))
4725 return split;
4727 /* See if this is a bitfield assignment with everything constant. If
4728 so, this is an IOR of an AND, so split it into that. */
4729 if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
4730 && HWI_COMPUTABLE_MODE_P (GET_MODE (XEXP (SET_DEST (x), 0)))
4731 && CONST_INT_P (XEXP (SET_DEST (x), 1))
4732 && CONST_INT_P (XEXP (SET_DEST (x), 2))
4733 && CONST_INT_P (SET_SRC (x))
4734 && ((INTVAL (XEXP (SET_DEST (x), 1))
4735 + INTVAL (XEXP (SET_DEST (x), 2)))
4736 <= GET_MODE_PRECISION (GET_MODE (XEXP (SET_DEST (x), 0))))
4737 && ! side_effects_p (XEXP (SET_DEST (x), 0)))
4739 HOST_WIDE_INT pos = INTVAL (XEXP (SET_DEST (x), 2));
4740 unsigned HOST_WIDE_INT len = INTVAL (XEXP (SET_DEST (x), 1));
4741 unsigned HOST_WIDE_INT src = INTVAL (SET_SRC (x));
4742 rtx dest = XEXP (SET_DEST (x), 0);
4743 enum machine_mode mode = GET_MODE (dest);
4744 unsigned HOST_WIDE_INT mask
4745 = ((unsigned HOST_WIDE_INT) 1 << len) - 1;
4746 rtx or_mask;
4748 if (BITS_BIG_ENDIAN)
4749 pos = GET_MODE_PRECISION (mode) - len - pos;
4751 or_mask = gen_int_mode (src << pos, mode);
4752 if (src == mask)
4753 SUBST (SET_SRC (x),
4754 simplify_gen_binary (IOR, mode, dest, or_mask));
4755 else
4757 rtx negmask = gen_int_mode (~(mask << pos), mode);
4758 SUBST (SET_SRC (x),
4759 simplify_gen_binary (IOR, mode,
4760 simplify_gen_binary (AND, mode,
4761 dest, negmask),
4762 or_mask));
4765 SUBST (SET_DEST (x), dest);
4767 split = find_split_point (&SET_SRC (x), insn, true);
4768 if (split && split != &SET_SRC (x))
4769 return split;
4772 /* Otherwise, see if this is an operation that we can split into two.
4773 If so, try to split that. */
4774 code = GET_CODE (SET_SRC (x));
4776 switch (code)
4778 case AND:
4779 /* If we are AND'ing with a large constant that is only a single
4780 bit and the result is only being used in a context where we
4781 need to know if it is zero or nonzero, replace it with a bit
4782 extraction. This will avoid the large constant, which might
4783 have taken more than one insn to make. If the constant were
4784 not a valid argument to the AND but took only one insn to make,
4785 this is no worse, but if it took more than one insn, it will
4786 be better. */
4788 if (CONST_INT_P (XEXP (SET_SRC (x), 1))
4789 && REG_P (XEXP (SET_SRC (x), 0))
4790 && (pos = exact_log2 (UINTVAL (XEXP (SET_SRC (x), 1)))) >= 7
4791 && REG_P (SET_DEST (x))
4792 && (split = find_single_use (SET_DEST (x), insn, (rtx*) 0)) != 0
4793 && (GET_CODE (*split) == EQ || GET_CODE (*split) == NE)
4794 && XEXP (*split, 0) == SET_DEST (x)
4795 && XEXP (*split, 1) == const0_rtx)
4797 rtx extraction = make_extraction (GET_MODE (SET_DEST (x)),
4798 XEXP (SET_SRC (x), 0),
4799 pos, NULL_RTX, 1, 1, 0, 0);
4800 if (extraction != 0)
4802 SUBST (SET_SRC (x), extraction);
4803 return find_split_point (loc, insn, false);
4806 break;
4808 case NE:
4809 /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
4810 is known to be on, this can be converted into a NEG of a shift. */
4811 if (STORE_FLAG_VALUE == -1 && XEXP (SET_SRC (x), 1) == const0_rtx
4812 && GET_MODE (SET_SRC (x)) == GET_MODE (XEXP (SET_SRC (x), 0))
4813 && 1 <= (pos = exact_log2
4814 (nonzero_bits (XEXP (SET_SRC (x), 0),
4815 GET_MODE (XEXP (SET_SRC (x), 0))))))
4817 enum machine_mode mode = GET_MODE (XEXP (SET_SRC (x), 0));
4819 SUBST (SET_SRC (x),
4820 gen_rtx_NEG (mode,
4821 gen_rtx_LSHIFTRT (mode,
4822 XEXP (SET_SRC (x), 0),
4823 GEN_INT (pos))));
4825 split = find_split_point (&SET_SRC (x), insn, true);
4826 if (split && split != &SET_SRC (x))
4827 return split;
4829 break;
4831 case SIGN_EXTEND:
4832 inner = XEXP (SET_SRC (x), 0);
4834 /* We can't optimize if either mode is a partial integer
4835 mode as we don't know how many bits are significant
4836 in those modes. */
4837 if (GET_MODE_CLASS (GET_MODE (inner)) == MODE_PARTIAL_INT
4838 || GET_MODE_CLASS (GET_MODE (SET_SRC (x))) == MODE_PARTIAL_INT)
4839 break;
4841 pos = 0;
4842 len = GET_MODE_PRECISION (GET_MODE (inner));
4843 unsignedp = 0;
4844 break;
4846 case SIGN_EXTRACT:
4847 case ZERO_EXTRACT:
4848 if (CONST_INT_P (XEXP (SET_SRC (x), 1))
4849 && CONST_INT_P (XEXP (SET_SRC (x), 2)))
4851 inner = XEXP (SET_SRC (x), 0);
4852 len = INTVAL (XEXP (SET_SRC (x), 1));
4853 pos = INTVAL (XEXP (SET_SRC (x), 2));
4855 if (BITS_BIG_ENDIAN)
4856 pos = GET_MODE_PRECISION (GET_MODE (inner)) - len - pos;
4857 unsignedp = (code == ZERO_EXTRACT);
4859 break;
4861 default:
4862 break;
4865 if (len && pos >= 0
4866 && pos + len <= GET_MODE_PRECISION (GET_MODE (inner)))
4868 enum machine_mode mode = GET_MODE (SET_SRC (x));
4870 /* For unsigned, we have a choice of a shift followed by an
4871 AND or two shifts. Use two shifts for field sizes where the
4872 constant might be too large. We assume here that we can
4873 always at least get 8-bit constants in an AND insn, which is
4874 true for every current RISC. */
4876 if (unsignedp && len <= 8)
4878 SUBST (SET_SRC (x),
4879 gen_rtx_AND (mode,
4880 gen_rtx_LSHIFTRT
4881 (mode, gen_lowpart (mode, inner),
4882 GEN_INT (pos)),
4883 GEN_INT (((unsigned HOST_WIDE_INT) 1 << len)
4884 - 1)));
4886 split = find_split_point (&SET_SRC (x), insn, true);
4887 if (split && split != &SET_SRC (x))
4888 return split;
4890 else
4892 SUBST (SET_SRC (x),
4893 gen_rtx_fmt_ee
4894 (unsignedp ? LSHIFTRT : ASHIFTRT, mode,
4895 gen_rtx_ASHIFT (mode,
4896 gen_lowpart (mode, inner),
4897 GEN_INT (GET_MODE_PRECISION (mode)
4898 - len - pos)),
4899 GEN_INT (GET_MODE_PRECISION (mode) - len)));
4901 split = find_split_point (&SET_SRC (x), insn, true);
4902 if (split && split != &SET_SRC (x))
4903 return split;
4907 /* See if this is a simple operation with a constant as the second
4908 operand. It might be that this constant is out of range and hence
4909 could be used as a split point. */
4910 if (BINARY_P (SET_SRC (x))
4911 && CONSTANT_P (XEXP (SET_SRC (x), 1))
4912 && (OBJECT_P (XEXP (SET_SRC (x), 0))
4913 || (GET_CODE (XEXP (SET_SRC (x), 0)) == SUBREG
4914 && OBJECT_P (SUBREG_REG (XEXP (SET_SRC (x), 0))))))
4915 return &XEXP (SET_SRC (x), 1);
4917 /* Finally, see if this is a simple operation with its first operand
4918 not in a register. The operation might require this operand in a
4919 register, so return it as a split point. We can always do this
4920 because if the first operand were another operation, we would have
4921 already found it as a split point. */
4922 if ((BINARY_P (SET_SRC (x)) || UNARY_P (SET_SRC (x)))
4923 && ! register_operand (XEXP (SET_SRC (x), 0), VOIDmode))
4924 return &XEXP (SET_SRC (x), 0);
4926 return 0;
4928 case AND:
4929 case IOR:
4930 /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
4931 it is better to write this as (not (ior A B)) so we can split it.
4932 Similarly for IOR. */
4933 if (GET_CODE (XEXP (x, 0)) == NOT && GET_CODE (XEXP (x, 1)) == NOT)
4935 SUBST (*loc,
4936 gen_rtx_NOT (GET_MODE (x),
4937 gen_rtx_fmt_ee (code == IOR ? AND : IOR,
4938 GET_MODE (x),
4939 XEXP (XEXP (x, 0), 0),
4940 XEXP (XEXP (x, 1), 0))));
4941 return find_split_point (loc, insn, set_src);
4944 /* Many RISC machines have a large set of logical insns. If the
4945 second operand is a NOT, put it first so we will try to split the
4946 other operand first. */
4947 if (GET_CODE (XEXP (x, 1)) == NOT)
4949 rtx tem = XEXP (x, 0);
4950 SUBST (XEXP (x, 0), XEXP (x, 1));
4951 SUBST (XEXP (x, 1), tem);
4953 break;
4955 case PLUS:
4956 case MINUS:
4957 /* Canonicalization can produce (minus A (mult B C)), where C is a
4958 constant. It may be better to try splitting (plus (mult B -C) A)
4959 instead if this isn't a multiply by a power of two. */
4960 if (set_src && code == MINUS && GET_CODE (XEXP (x, 1)) == MULT
4961 && GET_CODE (XEXP (XEXP (x, 1), 1)) == CONST_INT
4962 && exact_log2 (INTVAL (XEXP (XEXP (x, 1), 1))) < 0)
4964 enum machine_mode mode = GET_MODE (x);
4965 unsigned HOST_WIDE_INT this_int = INTVAL (XEXP (XEXP (x, 1), 1));
4966 HOST_WIDE_INT other_int = trunc_int_for_mode (-this_int, mode);
4967 SUBST (*loc, gen_rtx_PLUS (mode, gen_rtx_MULT (mode,
4968 XEXP (XEXP (x, 1), 0),
4969 GEN_INT (other_int)),
4970 XEXP (x, 0)));
4971 return find_split_point (loc, insn, set_src);
4974 /* Split at a multiply-accumulate instruction. However if this is
4975 the SET_SRC, we likely do not have such an instruction and it's
4976 worthless to try this split. */
4977 if (!set_src && GET_CODE (XEXP (x, 0)) == MULT)
4978 return loc;
4980 default:
4981 break;
4984 /* Otherwise, select our actions depending on our rtx class. */
4985 switch (GET_RTX_CLASS (code))
4987 case RTX_BITFIELD_OPS: /* This is ZERO_EXTRACT and SIGN_EXTRACT. */
4988 case RTX_TERNARY:
4989 split = find_split_point (&XEXP (x, 2), insn, false);
4990 if (split)
4991 return split;
4992 /* ... fall through ... */
4993 case RTX_BIN_ARITH:
4994 case RTX_COMM_ARITH:
4995 case RTX_COMPARE:
4996 case RTX_COMM_COMPARE:
4997 split = find_split_point (&XEXP (x, 1), insn, false);
4998 if (split)
4999 return split;
5000 /* ... fall through ... */
5001 case RTX_UNARY:
5002 /* Some machines have (and (shift ...) ...) insns. If X is not
5003 an AND, but XEXP (X, 0) is, use it as our split point. */
5004 if (GET_CODE (x) != AND && GET_CODE (XEXP (x, 0)) == AND)
5005 return &XEXP (x, 0);
5007 split = find_split_point (&XEXP (x, 0), insn, false);
5008 if (split)
5009 return split;
5010 return loc;
5012 default:
5013 /* Otherwise, we don't have a split point. */
5014 return 0;
5018 /* Throughout X, replace FROM with TO, and return the result.
5019 The result is TO if X is FROM;
5020 otherwise the result is X, but its contents may have been modified.
5021 If they were modified, a record was made in undobuf so that
5022 undo_all will (among other things) return X to its original state.
5024 If the number of changes necessary is too much to record to undo,
5025 the excess changes are not made, so the result is invalid.
5026 The changes already made can still be undone.
5027 undobuf.num_undo is incremented for such changes, so by testing that
5028 the caller can tell whether the result is valid.
5030 `n_occurrences' is incremented each time FROM is replaced.
5032 IN_DEST is nonzero if we are processing the SET_DEST of a SET.
5034 IN_COND is nonzero if we are at the top level of a condition.
5036 UNIQUE_COPY is nonzero if each substitution must be unique. We do this
5037 by copying if `n_occurrences' is nonzero. */
5039 static rtx
5040 subst (rtx x, rtx from, rtx to, int in_dest, int in_cond, int unique_copy)
5042 enum rtx_code code = GET_CODE (x);
5043 enum machine_mode op0_mode = VOIDmode;
5044 const char *fmt;
5045 int len, i;
5046 rtx new_rtx;
5048 /* Two expressions are equal if they are identical copies of a shared
5049 RTX or if they are both registers with the same register number
5050 and mode. */
5052 #define COMBINE_RTX_EQUAL_P(X,Y) \
5053 ((X) == (Y) \
5054 || (REG_P (X) && REG_P (Y) \
5055 && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5057 if (! in_dest && COMBINE_RTX_EQUAL_P (x, from))
5059 n_occurrences++;
5060 return (unique_copy && n_occurrences > 1 ? copy_rtx (to) : to);
5063 /* If X and FROM are the same register but different modes, they
5064 will not have been seen as equal above. However, the log links code
5065 will make a LOG_LINKS entry for that case. If we do nothing, we
5066 will try to rerecognize our original insn and, when it succeeds,
5067 we will delete the feeding insn, which is incorrect.
5069 So force this insn not to match in this (rare) case. */
5070 if (! in_dest && code == REG && REG_P (from)
5071 && reg_overlap_mentioned_p (x, from))
5072 return gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
5074 /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5075 of which may contain things that can be combined. */
5076 if (code != MEM && code != LO_SUM && OBJECT_P (x))
5077 return x;
5079 /* It is possible to have a subexpression appear twice in the insn.
5080 Suppose that FROM is a register that appears within TO.
5081 Then, after that subexpression has been scanned once by `subst',
5082 the second time it is scanned, TO may be found. If we were
5083 to scan TO here, we would find FROM within it and create a
5084 self-referent rtl structure which is completely wrong. */
5085 if (COMBINE_RTX_EQUAL_P (x, to))
5086 return to;
5088 /* Parallel asm_operands need special attention because all of the
5089 inputs are shared across the arms. Furthermore, unsharing the
5090 rtl results in recognition failures. Failure to handle this case
5091 specially can result in circular rtl.
5093 Solve this by doing a normal pass across the first entry of the
5094 parallel, and only processing the SET_DESTs of the subsequent
5095 entries. Ug. */
5097 if (code == PARALLEL
5098 && GET_CODE (XVECEXP (x, 0, 0)) == SET
5099 && GET_CODE (SET_SRC (XVECEXP (x, 0, 0))) == ASM_OPERANDS)
5101 new_rtx = subst (XVECEXP (x, 0, 0), from, to, 0, 0, unique_copy);
5103 /* If this substitution failed, this whole thing fails. */
5104 if (GET_CODE (new_rtx) == CLOBBER
5105 && XEXP (new_rtx, 0) == const0_rtx)
5106 return new_rtx;
5108 SUBST (XVECEXP (x, 0, 0), new_rtx);
5110 for (i = XVECLEN (x, 0) - 1; i >= 1; i--)
5112 rtx dest = SET_DEST (XVECEXP (x, 0, i));
5114 if (!REG_P (dest)
5115 && GET_CODE (dest) != CC0
5116 && GET_CODE (dest) != PC)
5118 new_rtx = subst (dest, from, to, 0, 0, unique_copy);
5120 /* If this substitution failed, this whole thing fails. */
5121 if (GET_CODE (new_rtx) == CLOBBER
5122 && XEXP (new_rtx, 0) == const0_rtx)
5123 return new_rtx;
5125 SUBST (SET_DEST (XVECEXP (x, 0, i)), new_rtx);
5129 else
5131 len = GET_RTX_LENGTH (code);
5132 fmt = GET_RTX_FORMAT (code);
5134 /* We don't need to process a SET_DEST that is a register, CC0,
5135 or PC, so set up to skip this common case. All other cases
5136 where we want to suppress replacing something inside a
5137 SET_SRC are handled via the IN_DEST operand. */
5138 if (code == SET
5139 && (REG_P (SET_DEST (x))
5140 || GET_CODE (SET_DEST (x)) == CC0
5141 || GET_CODE (SET_DEST (x)) == PC))
5142 fmt = "ie";
5144 /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5145 constant. */
5146 if (fmt[0] == 'e')
5147 op0_mode = GET_MODE (XEXP (x, 0));
5149 for (i = 0; i < len; i++)
5151 if (fmt[i] == 'E')
5153 int j;
5154 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
5156 if (COMBINE_RTX_EQUAL_P (XVECEXP (x, i, j), from))
5158 new_rtx = (unique_copy && n_occurrences
5159 ? copy_rtx (to) : to);
5160 n_occurrences++;
5162 else
5164 new_rtx = subst (XVECEXP (x, i, j), from, to, 0, 0,
5165 unique_copy);
5167 /* If this substitution failed, this whole thing
5168 fails. */
5169 if (GET_CODE (new_rtx) == CLOBBER
5170 && XEXP (new_rtx, 0) == const0_rtx)
5171 return new_rtx;
5174 SUBST (XVECEXP (x, i, j), new_rtx);
5177 else if (fmt[i] == 'e')
5179 /* If this is a register being set, ignore it. */
5180 new_rtx = XEXP (x, i);
5181 if (in_dest
5182 && i == 0
5183 && (((code == SUBREG || code == ZERO_EXTRACT)
5184 && REG_P (new_rtx))
5185 || code == STRICT_LOW_PART))
5188 else if (COMBINE_RTX_EQUAL_P (XEXP (x, i), from))
5190 /* In general, don't install a subreg involving two
5191 modes not tieable. It can worsen register
5192 allocation, and can even make invalid reload
5193 insns, since the reg inside may need to be copied
5194 from in the outside mode, and that may be invalid
5195 if it is an fp reg copied in integer mode.
5197 We allow two exceptions to this: It is valid if
5198 it is inside another SUBREG and the mode of that
5199 SUBREG and the mode of the inside of TO is
5200 tieable and it is valid if X is a SET that copies
5201 FROM to CC0. */
5203 if (GET_CODE (to) == SUBREG
5204 && ! MODES_TIEABLE_P (GET_MODE (to),
5205 GET_MODE (SUBREG_REG (to)))
5206 && ! (code == SUBREG
5207 && MODES_TIEABLE_P (GET_MODE (x),
5208 GET_MODE (SUBREG_REG (to))))
5209 #ifdef HAVE_cc0
5210 && ! (code == SET && i == 1 && XEXP (x, 0) == cc0_rtx)
5211 #endif
5213 return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5215 #ifdef CANNOT_CHANGE_MODE_CLASS
5216 if (code == SUBREG
5217 && REG_P (to)
5218 && REGNO (to) < FIRST_PSEUDO_REGISTER
5219 && REG_CANNOT_CHANGE_MODE_P (REGNO (to),
5220 GET_MODE (to),
5221 GET_MODE (x)))
5222 return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
5223 #endif
5225 new_rtx = (unique_copy && n_occurrences ? copy_rtx (to) : to);
5226 n_occurrences++;
5228 else
5229 /* If we are in a SET_DEST, suppress most cases unless we
5230 have gone inside a MEM, in which case we want to
5231 simplify the address. We assume here that things that
5232 are actually part of the destination have their inner
5233 parts in the first expression. This is true for SUBREG,
5234 STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5235 things aside from REG and MEM that should appear in a
5236 SET_DEST. */
5237 new_rtx = subst (XEXP (x, i), from, to,
5238 (((in_dest
5239 && (code == SUBREG || code == STRICT_LOW_PART
5240 || code == ZERO_EXTRACT))
5241 || code == SET)
5242 && i == 0),
5243 code == IF_THEN_ELSE && i == 0,
5244 unique_copy);
5246 /* If we found that we will have to reject this combination,
5247 indicate that by returning the CLOBBER ourselves, rather than
5248 an expression containing it. This will speed things up as
5249 well as prevent accidents where two CLOBBERs are considered
5250 to be equal, thus producing an incorrect simplification. */
5252 if (GET_CODE (new_rtx) == CLOBBER && XEXP (new_rtx, 0) == const0_rtx)
5253 return new_rtx;
5255 if (GET_CODE (x) == SUBREG
5256 && (CONST_INT_P (new_rtx)
5257 || GET_CODE (new_rtx) == CONST_DOUBLE))
5259 enum machine_mode mode = GET_MODE (x);
5261 x = simplify_subreg (GET_MODE (x), new_rtx,
5262 GET_MODE (SUBREG_REG (x)),
5263 SUBREG_BYTE (x));
5264 if (! x)
5265 x = gen_rtx_CLOBBER (mode, const0_rtx);
5267 else if (CONST_INT_P (new_rtx)
5268 && GET_CODE (x) == ZERO_EXTEND)
5270 x = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
5271 new_rtx, GET_MODE (XEXP (x, 0)));
5272 gcc_assert (x);
5274 else
5275 SUBST (XEXP (x, i), new_rtx);
5280 /* Check if we are loading something from the constant pool via float
5281 extension; in this case we would undo compress_float_constant
5282 optimization and degenerate constant load to an immediate value. */
5283 if (GET_CODE (x) == FLOAT_EXTEND
5284 && MEM_P (XEXP (x, 0))
5285 && MEM_READONLY_P (XEXP (x, 0)))
5287 rtx tmp = avoid_constant_pool_reference (x);
5288 if (x != tmp)
5289 return x;
5292 /* Try to simplify X. If the simplification changed the code, it is likely
5293 that further simplification will help, so loop, but limit the number
5294 of repetitions that will be performed. */
5296 for (i = 0; i < 4; i++)
5298 /* If X is sufficiently simple, don't bother trying to do anything
5299 with it. */
5300 if (code != CONST_INT && code != REG && code != CLOBBER)
5301 x = combine_simplify_rtx (x, op0_mode, in_dest, in_cond);
5303 if (GET_CODE (x) == code)
5304 break;
5306 code = GET_CODE (x);
5308 /* We no longer know the original mode of operand 0 since we
5309 have changed the form of X) */
5310 op0_mode = VOIDmode;
5313 return x;
5316 /* Simplify X, a piece of RTL. We just operate on the expression at the
5317 outer level; call `subst' to simplify recursively. Return the new
5318 expression.
5320 OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
5321 if we are inside a SET_DEST. IN_COND is nonzero if we are at the top level
5322 of a condition. */
5324 static rtx
5325 combine_simplify_rtx (rtx x, enum machine_mode op0_mode, int in_dest,
5326 int in_cond)
5328 enum rtx_code code = GET_CODE (x);
5329 enum machine_mode mode = GET_MODE (x);
5330 rtx temp;
5331 int i;
5333 /* If this is a commutative operation, put a constant last and a complex
5334 expression first. We don't need to do this for comparisons here. */
5335 if (COMMUTATIVE_ARITH_P (x)
5336 && swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1)))
5338 temp = XEXP (x, 0);
5339 SUBST (XEXP (x, 0), XEXP (x, 1));
5340 SUBST (XEXP (x, 1), temp);
5343 /* If this is a simple operation applied to an IF_THEN_ELSE, try
5344 applying it to the arms of the IF_THEN_ELSE. This often simplifies
5345 things. Check for cases where both arms are testing the same
5346 condition.
5348 Don't do anything if all operands are very simple. */
5350 if ((BINARY_P (x)
5351 && ((!OBJECT_P (XEXP (x, 0))
5352 && ! (GET_CODE (XEXP (x, 0)) == SUBREG
5353 && OBJECT_P (SUBREG_REG (XEXP (x, 0)))))
5354 || (!OBJECT_P (XEXP (x, 1))
5355 && ! (GET_CODE (XEXP (x, 1)) == SUBREG
5356 && OBJECT_P (SUBREG_REG (XEXP (x, 1)))))))
5357 || (UNARY_P (x)
5358 && (!OBJECT_P (XEXP (x, 0))
5359 && ! (GET_CODE (XEXP (x, 0)) == SUBREG
5360 && OBJECT_P (SUBREG_REG (XEXP (x, 0)))))))
5362 rtx cond, true_rtx, false_rtx;
5364 cond = if_then_else_cond (x, &true_rtx, &false_rtx);
5365 if (cond != 0
5366 /* If everything is a comparison, what we have is highly unlikely
5367 to be simpler, so don't use it. */
5368 && ! (COMPARISON_P (x)
5369 && (COMPARISON_P (true_rtx) || COMPARISON_P (false_rtx))))
5371 rtx cop1 = const0_rtx;
5372 enum rtx_code cond_code = simplify_comparison (NE, &cond, &cop1);
5374 if (cond_code == NE && COMPARISON_P (cond))
5375 return x;
5377 /* Simplify the alternative arms; this may collapse the true and
5378 false arms to store-flag values. Be careful to use copy_rtx
5379 here since true_rtx or false_rtx might share RTL with x as a
5380 result of the if_then_else_cond call above. */
5381 true_rtx = subst (copy_rtx (true_rtx), pc_rtx, pc_rtx, 0, 0, 0);
5382 false_rtx = subst (copy_rtx (false_rtx), pc_rtx, pc_rtx, 0, 0, 0);
5384 /* If true_rtx and false_rtx are not general_operands, an if_then_else
5385 is unlikely to be simpler. */
5386 if (general_operand (true_rtx, VOIDmode)
5387 && general_operand (false_rtx, VOIDmode))
5389 enum rtx_code reversed;
5391 /* Restarting if we generate a store-flag expression will cause
5392 us to loop. Just drop through in this case. */
5394 /* If the result values are STORE_FLAG_VALUE and zero, we can
5395 just make the comparison operation. */
5396 if (true_rtx == const_true_rtx && false_rtx == const0_rtx)
5397 x = simplify_gen_relational (cond_code, mode, VOIDmode,
5398 cond, cop1);
5399 else if (true_rtx == const0_rtx && false_rtx == const_true_rtx
5400 && ((reversed = reversed_comparison_code_parts
5401 (cond_code, cond, cop1, NULL))
5402 != UNKNOWN))
5403 x = simplify_gen_relational (reversed, mode, VOIDmode,
5404 cond, cop1);
5406 /* Likewise, we can make the negate of a comparison operation
5407 if the result values are - STORE_FLAG_VALUE and zero. */
5408 else if (CONST_INT_P (true_rtx)
5409 && INTVAL (true_rtx) == - STORE_FLAG_VALUE
5410 && false_rtx == const0_rtx)
5411 x = simplify_gen_unary (NEG, mode,
5412 simplify_gen_relational (cond_code,
5413 mode, VOIDmode,
5414 cond, cop1),
5415 mode);
5416 else if (CONST_INT_P (false_rtx)
5417 && INTVAL (false_rtx) == - STORE_FLAG_VALUE
5418 && true_rtx == const0_rtx
5419 && ((reversed = reversed_comparison_code_parts
5420 (cond_code, cond, cop1, NULL))
5421 != UNKNOWN))
5422 x = simplify_gen_unary (NEG, mode,
5423 simplify_gen_relational (reversed,
5424 mode, VOIDmode,
5425 cond, cop1),
5426 mode);
5427 else
5428 return gen_rtx_IF_THEN_ELSE (mode,
5429 simplify_gen_relational (cond_code,
5430 mode,
5431 VOIDmode,
5432 cond,
5433 cop1),
5434 true_rtx, false_rtx);
5436 code = GET_CODE (x);
5437 op0_mode = VOIDmode;
5442 /* Try to fold this expression in case we have constants that weren't
5443 present before. */
5444 temp = 0;
5445 switch (GET_RTX_CLASS (code))
5447 case RTX_UNARY:
5448 if (op0_mode == VOIDmode)
5449 op0_mode = GET_MODE (XEXP (x, 0));
5450 temp = simplify_unary_operation (code, mode, XEXP (x, 0), op0_mode);
5451 break;
5452 case RTX_COMPARE:
5453 case RTX_COMM_COMPARE:
5455 enum machine_mode cmp_mode = GET_MODE (XEXP (x, 0));
5456 if (cmp_mode == VOIDmode)
5458 cmp_mode = GET_MODE (XEXP (x, 1));
5459 if (cmp_mode == VOIDmode)
5460 cmp_mode = op0_mode;
5462 temp = simplify_relational_operation (code, mode, cmp_mode,
5463 XEXP (x, 0), XEXP (x, 1));
5465 break;
5466 case RTX_COMM_ARITH:
5467 case RTX_BIN_ARITH:
5468 temp = simplify_binary_operation (code, mode, XEXP (x, 0), XEXP (x, 1));
5469 break;
5470 case RTX_BITFIELD_OPS:
5471 case RTX_TERNARY:
5472 temp = simplify_ternary_operation (code, mode, op0_mode, XEXP (x, 0),
5473 XEXP (x, 1), XEXP (x, 2));
5474 break;
5475 default:
5476 break;
5479 if (temp)
5481 x = temp;
5482 code = GET_CODE (temp);
5483 op0_mode = VOIDmode;
5484 mode = GET_MODE (temp);
5487 /* First see if we can apply the inverse distributive law. */
5488 if (code == PLUS || code == MINUS
5489 || code == AND || code == IOR || code == XOR)
5491 x = apply_distributive_law (x);
5492 code = GET_CODE (x);
5493 op0_mode = VOIDmode;
5496 /* If CODE is an associative operation not otherwise handled, see if we
5497 can associate some operands. This can win if they are constants or
5498 if they are logically related (i.e. (a & b) & a). */
5499 if ((code == PLUS || code == MINUS || code == MULT || code == DIV
5500 || code == AND || code == IOR || code == XOR
5501 || code == SMAX || code == SMIN || code == UMAX || code == UMIN)
5502 && ((INTEGRAL_MODE_P (mode) && code != DIV)
5503 || (flag_associative_math && FLOAT_MODE_P (mode))))
5505 if (GET_CODE (XEXP (x, 0)) == code)
5507 rtx other = XEXP (XEXP (x, 0), 0);
5508 rtx inner_op0 = XEXP (XEXP (x, 0), 1);
5509 rtx inner_op1 = XEXP (x, 1);
5510 rtx inner;
5512 /* Make sure we pass the constant operand if any as the second
5513 one if this is a commutative operation. */
5514 if (CONSTANT_P (inner_op0) && COMMUTATIVE_ARITH_P (x))
5516 rtx tem = inner_op0;
5517 inner_op0 = inner_op1;
5518 inner_op1 = tem;
5520 inner = simplify_binary_operation (code == MINUS ? PLUS
5521 : code == DIV ? MULT
5522 : code,
5523 mode, inner_op0, inner_op1);
5525 /* For commutative operations, try the other pair if that one
5526 didn't simplify. */
5527 if (inner == 0 && COMMUTATIVE_ARITH_P (x))
5529 other = XEXP (XEXP (x, 0), 1);
5530 inner = simplify_binary_operation (code, mode,
5531 XEXP (XEXP (x, 0), 0),
5532 XEXP (x, 1));
5535 if (inner)
5536 return simplify_gen_binary (code, mode, other, inner);
5540 /* A little bit of algebraic simplification here. */
5541 switch (code)
5543 case MEM:
5544 /* Ensure that our address has any ASHIFTs converted to MULT in case
5545 address-recognizing predicates are called later. */
5546 temp = make_compound_operation (XEXP (x, 0), MEM);
5547 SUBST (XEXP (x, 0), temp);
5548 break;
5550 case SUBREG:
5551 if (op0_mode == VOIDmode)
5552 op0_mode = GET_MODE (SUBREG_REG (x));
5554 /* See if this can be moved to simplify_subreg. */
5555 if (CONSTANT_P (SUBREG_REG (x))
5556 && subreg_lowpart_offset (mode, op0_mode) == SUBREG_BYTE (x)
5557 /* Don't call gen_lowpart if the inner mode
5558 is VOIDmode and we cannot simplify it, as SUBREG without
5559 inner mode is invalid. */
5560 && (GET_MODE (SUBREG_REG (x)) != VOIDmode
5561 || gen_lowpart_common (mode, SUBREG_REG (x))))
5562 return gen_lowpart (mode, SUBREG_REG (x));
5564 if (GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_CC)
5565 break;
5567 rtx temp;
5568 temp = simplify_subreg (mode, SUBREG_REG (x), op0_mode,
5569 SUBREG_BYTE (x));
5570 if (temp)
5571 return temp;
5574 /* Don't change the mode of the MEM if that would change the meaning
5575 of the address. */
5576 if (MEM_P (SUBREG_REG (x))
5577 && (MEM_VOLATILE_P (SUBREG_REG (x))
5578 || mode_dependent_address_p (XEXP (SUBREG_REG (x), 0))))
5579 return gen_rtx_CLOBBER (mode, const0_rtx);
5581 /* Note that we cannot do any narrowing for non-constants since
5582 we might have been counting on using the fact that some bits were
5583 zero. We now do this in the SET. */
5585 break;
5587 case NEG:
5588 temp = expand_compound_operation (XEXP (x, 0));
5590 /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
5591 replaced by (lshiftrt X C). This will convert
5592 (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
5594 if (GET_CODE (temp) == ASHIFTRT
5595 && CONST_INT_P (XEXP (temp, 1))
5596 && INTVAL (XEXP (temp, 1)) == GET_MODE_PRECISION (mode) - 1)
5597 return simplify_shift_const (NULL_RTX, LSHIFTRT, mode, XEXP (temp, 0),
5598 INTVAL (XEXP (temp, 1)));
5600 /* If X has only a single bit that might be nonzero, say, bit I, convert
5601 (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
5602 MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
5603 (sign_extract X 1 Y). But only do this if TEMP isn't a register
5604 or a SUBREG of one since we'd be making the expression more
5605 complex if it was just a register. */
5607 if (!REG_P (temp)
5608 && ! (GET_CODE (temp) == SUBREG
5609 && REG_P (SUBREG_REG (temp)))
5610 && (i = exact_log2 (nonzero_bits (temp, mode))) >= 0)
5612 rtx temp1 = simplify_shift_const
5613 (NULL_RTX, ASHIFTRT, mode,
5614 simplify_shift_const (NULL_RTX, ASHIFT, mode, temp,
5615 GET_MODE_PRECISION (mode) - 1 - i),
5616 GET_MODE_PRECISION (mode) - 1 - i);
5618 /* If all we did was surround TEMP with the two shifts, we
5619 haven't improved anything, so don't use it. Otherwise,
5620 we are better off with TEMP1. */
5621 if (GET_CODE (temp1) != ASHIFTRT
5622 || GET_CODE (XEXP (temp1, 0)) != ASHIFT
5623 || XEXP (XEXP (temp1, 0), 0) != temp)
5624 return temp1;
5626 break;
5628 case TRUNCATE:
5629 /* We can't handle truncation to a partial integer mode here
5630 because we don't know the real bitsize of the partial
5631 integer mode. */
5632 if (GET_MODE_CLASS (mode) == MODE_PARTIAL_INT)
5633 break;
5635 if (HWI_COMPUTABLE_MODE_P (mode))
5636 SUBST (XEXP (x, 0),
5637 force_to_mode (XEXP (x, 0), GET_MODE (XEXP (x, 0)),
5638 GET_MODE_MASK (mode), 0));
5640 /* We can truncate a constant value and return it. */
5641 if (CONST_INT_P (XEXP (x, 0)))
5642 return gen_int_mode (INTVAL (XEXP (x, 0)), mode);
5644 /* Similarly to what we do in simplify-rtx.c, a truncate of a register
5645 whose value is a comparison can be replaced with a subreg if
5646 STORE_FLAG_VALUE permits. */
5647 if (HWI_COMPUTABLE_MODE_P (mode)
5648 && (STORE_FLAG_VALUE & ~GET_MODE_MASK (mode)) == 0
5649 && (temp = get_last_value (XEXP (x, 0)))
5650 && COMPARISON_P (temp))
5651 return gen_lowpart (mode, XEXP (x, 0));
5652 break;
5654 case CONST:
5655 /* (const (const X)) can become (const X). Do it this way rather than
5656 returning the inner CONST since CONST can be shared with a
5657 REG_EQUAL note. */
5658 if (GET_CODE (XEXP (x, 0)) == CONST)
5659 SUBST (XEXP (x, 0), XEXP (XEXP (x, 0), 0));
5660 break;
5662 #ifdef HAVE_lo_sum
5663 case LO_SUM:
5664 /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
5665 can add in an offset. find_split_point will split this address up
5666 again if it doesn't match. */
5667 if (GET_CODE (XEXP (x, 0)) == HIGH
5668 && rtx_equal_p (XEXP (XEXP (x, 0), 0), XEXP (x, 1)))
5669 return XEXP (x, 1);
5670 break;
5671 #endif
5673 case PLUS:
5674 /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
5675 when c is (const_int (pow2 + 1) / 2) is a sign extension of a
5676 bit-field and can be replaced by either a sign_extend or a
5677 sign_extract. The `and' may be a zero_extend and the two
5678 <c>, -<c> constants may be reversed. */
5679 if (GET_CODE (XEXP (x, 0)) == XOR
5680 && CONST_INT_P (XEXP (x, 1))
5681 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
5682 && INTVAL (XEXP (x, 1)) == -INTVAL (XEXP (XEXP (x, 0), 1))
5683 && ((i = exact_log2 (UINTVAL (XEXP (XEXP (x, 0), 1)))) >= 0
5684 || (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0)
5685 && HWI_COMPUTABLE_MODE_P (mode)
5686 && ((GET_CODE (XEXP (XEXP (x, 0), 0)) == AND
5687 && CONST_INT_P (XEXP (XEXP (XEXP (x, 0), 0), 1))
5688 && (UINTVAL (XEXP (XEXP (XEXP (x, 0), 0), 1))
5689 == ((unsigned HOST_WIDE_INT) 1 << (i + 1)) - 1))
5690 || (GET_CODE (XEXP (XEXP (x, 0), 0)) == ZERO_EXTEND
5691 && (GET_MODE_PRECISION (GET_MODE (XEXP (XEXP (XEXP (x, 0), 0), 0)))
5692 == (unsigned int) i + 1))))
5693 return simplify_shift_const
5694 (NULL_RTX, ASHIFTRT, mode,
5695 simplify_shift_const (NULL_RTX, ASHIFT, mode,
5696 XEXP (XEXP (XEXP (x, 0), 0), 0),
5697 GET_MODE_PRECISION (mode) - (i + 1)),
5698 GET_MODE_PRECISION (mode) - (i + 1));
5700 /* If only the low-order bit of X is possibly nonzero, (plus x -1)
5701 can become (ashiftrt (ashift (xor x 1) C) C) where C is
5702 the bitsize of the mode - 1. This allows simplification of
5703 "a = (b & 8) == 0;" */
5704 if (XEXP (x, 1) == constm1_rtx
5705 && !REG_P (XEXP (x, 0))
5706 && ! (GET_CODE (XEXP (x, 0)) == SUBREG
5707 && REG_P (SUBREG_REG (XEXP (x, 0))))
5708 && nonzero_bits (XEXP (x, 0), mode) == 1)
5709 return simplify_shift_const (NULL_RTX, ASHIFTRT, mode,
5710 simplify_shift_const (NULL_RTX, ASHIFT, mode,
5711 gen_rtx_XOR (mode, XEXP (x, 0), const1_rtx),
5712 GET_MODE_PRECISION (mode) - 1),
5713 GET_MODE_PRECISION (mode) - 1);
5715 /* If we are adding two things that have no bits in common, convert
5716 the addition into an IOR. This will often be further simplified,
5717 for example in cases like ((a & 1) + (a & 2)), which can
5718 become a & 3. */
5720 if (HWI_COMPUTABLE_MODE_P (mode)
5721 && (nonzero_bits (XEXP (x, 0), mode)
5722 & nonzero_bits (XEXP (x, 1), mode)) == 0)
5724 /* Try to simplify the expression further. */
5725 rtx tor = simplify_gen_binary (IOR, mode, XEXP (x, 0), XEXP (x, 1));
5726 temp = combine_simplify_rtx (tor, VOIDmode, in_dest, 0);
5728 /* If we could, great. If not, do not go ahead with the IOR
5729 replacement, since PLUS appears in many special purpose
5730 address arithmetic instructions. */
5731 if (GET_CODE (temp) != CLOBBER
5732 && (GET_CODE (temp) != IOR
5733 || ((XEXP (temp, 0) != XEXP (x, 0)
5734 || XEXP (temp, 1) != XEXP (x, 1))
5735 && (XEXP (temp, 0) != XEXP (x, 1)
5736 || XEXP (temp, 1) != XEXP (x, 0)))))
5737 return temp;
5739 break;
5741 case MINUS:
5742 /* (minus <foo> (and <foo> (const_int -pow2))) becomes
5743 (and <foo> (const_int pow2-1)) */
5744 if (GET_CODE (XEXP (x, 1)) == AND
5745 && CONST_INT_P (XEXP (XEXP (x, 1), 1))
5746 && exact_log2 (-UINTVAL (XEXP (XEXP (x, 1), 1))) >= 0
5747 && rtx_equal_p (XEXP (XEXP (x, 1), 0), XEXP (x, 0)))
5748 return simplify_and_const_int (NULL_RTX, mode, XEXP (x, 0),
5749 -INTVAL (XEXP (XEXP (x, 1), 1)) - 1);
5750 break;
5752 case MULT:
5753 /* If we have (mult (plus A B) C), apply the distributive law and then
5754 the inverse distributive law to see if things simplify. This
5755 occurs mostly in addresses, often when unrolling loops. */
5757 if (GET_CODE (XEXP (x, 0)) == PLUS)
5759 rtx result = distribute_and_simplify_rtx (x, 0);
5760 if (result)
5761 return result;
5764 /* Try simplify a*(b/c) as (a*b)/c. */
5765 if (FLOAT_MODE_P (mode) && flag_associative_math
5766 && GET_CODE (XEXP (x, 0)) == DIV)
5768 rtx tem = simplify_binary_operation (MULT, mode,
5769 XEXP (XEXP (x, 0), 0),
5770 XEXP (x, 1));
5771 if (tem)
5772 return simplify_gen_binary (DIV, mode, tem, XEXP (XEXP (x, 0), 1));
5774 break;
5776 case UDIV:
5777 /* If this is a divide by a power of two, treat it as a shift if
5778 its first operand is a shift. */
5779 if (CONST_INT_P (XEXP (x, 1))
5780 && (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0
5781 && (GET_CODE (XEXP (x, 0)) == ASHIFT
5782 || GET_CODE (XEXP (x, 0)) == LSHIFTRT
5783 || GET_CODE (XEXP (x, 0)) == ASHIFTRT
5784 || GET_CODE (XEXP (x, 0)) == ROTATE
5785 || GET_CODE (XEXP (x, 0)) == ROTATERT))
5786 return simplify_shift_const (NULL_RTX, LSHIFTRT, mode, XEXP (x, 0), i);
5787 break;
5789 case EQ: case NE:
5790 case GT: case GTU: case GE: case GEU:
5791 case LT: case LTU: case LE: case LEU:
5792 case UNEQ: case LTGT:
5793 case UNGT: case UNGE:
5794 case UNLT: case UNLE:
5795 case UNORDERED: case ORDERED:
5796 /* If the first operand is a condition code, we can't do anything
5797 with it. */
5798 if (GET_CODE (XEXP (x, 0)) == COMPARE
5799 || (GET_MODE_CLASS (GET_MODE (XEXP (x, 0))) != MODE_CC
5800 && ! CC0_P (XEXP (x, 0))))
5802 rtx op0 = XEXP (x, 0);
5803 rtx op1 = XEXP (x, 1);
5804 enum rtx_code new_code;
5806 if (GET_CODE (op0) == COMPARE)
5807 op1 = XEXP (op0, 1), op0 = XEXP (op0, 0);
5809 /* Simplify our comparison, if possible. */
5810 new_code = simplify_comparison (code, &op0, &op1);
5812 /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
5813 if only the low-order bit is possibly nonzero in X (such as when
5814 X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
5815 (xor X 1) or (minus 1 X); we use the former. Finally, if X is
5816 known to be either 0 or -1, NE becomes a NEG and EQ becomes
5817 (plus X 1).
5819 Remove any ZERO_EXTRACT we made when thinking this was a
5820 comparison. It may now be simpler to use, e.g., an AND. If a
5821 ZERO_EXTRACT is indeed appropriate, it will be placed back by
5822 the call to make_compound_operation in the SET case.
5824 Don't apply these optimizations if the caller would
5825 prefer a comparison rather than a value.
5826 E.g., for the condition in an IF_THEN_ELSE most targets need
5827 an explicit comparison. */
5829 if (in_cond)
5832 else if (STORE_FLAG_VALUE == 1
5833 && new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
5834 && op1 == const0_rtx
5835 && mode == GET_MODE (op0)
5836 && nonzero_bits (op0, mode) == 1)
5837 return gen_lowpart (mode,
5838 expand_compound_operation (op0));
5840 else if (STORE_FLAG_VALUE == 1
5841 && new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
5842 && op1 == const0_rtx
5843 && mode == GET_MODE (op0)
5844 && (num_sign_bit_copies (op0, mode)
5845 == GET_MODE_PRECISION (mode)))
5847 op0 = expand_compound_operation (op0);
5848 return simplify_gen_unary (NEG, mode,
5849 gen_lowpart (mode, op0),
5850 mode);
5853 else if (STORE_FLAG_VALUE == 1
5854 && new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
5855 && op1 == const0_rtx
5856 && mode == GET_MODE (op0)
5857 && nonzero_bits (op0, mode) == 1)
5859 op0 = expand_compound_operation (op0);
5860 return simplify_gen_binary (XOR, mode,
5861 gen_lowpart (mode, op0),
5862 const1_rtx);
5865 else if (STORE_FLAG_VALUE == 1
5866 && new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
5867 && op1 == const0_rtx
5868 && mode == GET_MODE (op0)
5869 && (num_sign_bit_copies (op0, mode)
5870 == GET_MODE_PRECISION (mode)))
5872 op0 = expand_compound_operation (op0);
5873 return plus_constant (gen_lowpart (mode, op0), 1);
5876 /* If STORE_FLAG_VALUE is -1, we have cases similar to
5877 those above. */
5878 if (in_cond)
5881 else if (STORE_FLAG_VALUE == -1
5882 && new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
5883 && op1 == const0_rtx
5884 && (num_sign_bit_copies (op0, mode)
5885 == GET_MODE_PRECISION (mode)))
5886 return gen_lowpart (mode,
5887 expand_compound_operation (op0));
5889 else if (STORE_FLAG_VALUE == -1
5890 && new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
5891 && op1 == const0_rtx
5892 && mode == GET_MODE (op0)
5893 && nonzero_bits (op0, mode) == 1)
5895 op0 = expand_compound_operation (op0);
5896 return simplify_gen_unary (NEG, mode,
5897 gen_lowpart (mode, op0),
5898 mode);
5901 else if (STORE_FLAG_VALUE == -1
5902 && new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
5903 && op1 == const0_rtx
5904 && mode == GET_MODE (op0)
5905 && (num_sign_bit_copies (op0, mode)
5906 == GET_MODE_PRECISION (mode)))
5908 op0 = expand_compound_operation (op0);
5909 return simplify_gen_unary (NOT, mode,
5910 gen_lowpart (mode, op0),
5911 mode);
5914 /* If X is 0/1, (eq X 0) is X-1. */
5915 else if (STORE_FLAG_VALUE == -1
5916 && new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
5917 && op1 == const0_rtx
5918 && mode == GET_MODE (op0)
5919 && nonzero_bits (op0, mode) == 1)
5921 op0 = expand_compound_operation (op0);
5922 return plus_constant (gen_lowpart (mode, op0), -1);
5925 /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
5926 one bit that might be nonzero, we can convert (ne x 0) to
5927 (ashift x c) where C puts the bit in the sign bit. Remove any
5928 AND with STORE_FLAG_VALUE when we are done, since we are only
5929 going to test the sign bit. */
5930 if (new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
5931 && HWI_COMPUTABLE_MODE_P (mode)
5932 && val_signbit_p (mode, STORE_FLAG_VALUE)
5933 && op1 == const0_rtx
5934 && mode == GET_MODE (op0)
5935 && (i = exact_log2 (nonzero_bits (op0, mode))) >= 0)
5937 x = simplify_shift_const (NULL_RTX, ASHIFT, mode,
5938 expand_compound_operation (op0),
5939 GET_MODE_PRECISION (mode) - 1 - i);
5940 if (GET_CODE (x) == AND && XEXP (x, 1) == const_true_rtx)
5941 return XEXP (x, 0);
5942 else
5943 return x;
5946 /* If the code changed, return a whole new comparison. */
5947 if (new_code != code)
5948 return gen_rtx_fmt_ee (new_code, mode, op0, op1);
5950 /* Otherwise, keep this operation, but maybe change its operands.
5951 This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
5952 SUBST (XEXP (x, 0), op0);
5953 SUBST (XEXP (x, 1), op1);
5955 break;
5957 case IF_THEN_ELSE:
5958 return simplify_if_then_else (x);
5960 case ZERO_EXTRACT:
5961 case SIGN_EXTRACT:
5962 case ZERO_EXTEND:
5963 case SIGN_EXTEND:
5964 /* If we are processing SET_DEST, we are done. */
5965 if (in_dest)
5966 return x;
5968 return expand_compound_operation (x);
5970 case SET:
5971 return simplify_set (x);
5973 case AND:
5974 case IOR:
5975 return simplify_logical (x);
5977 case ASHIFT:
5978 case LSHIFTRT:
5979 case ASHIFTRT:
5980 case ROTATE:
5981 case ROTATERT:
5982 /* If this is a shift by a constant amount, simplify it. */
5983 if (CONST_INT_P (XEXP (x, 1)))
5984 return simplify_shift_const (x, code, mode, XEXP (x, 0),
5985 INTVAL (XEXP (x, 1)));
5987 else if (SHIFT_COUNT_TRUNCATED && !REG_P (XEXP (x, 1)))
5988 SUBST (XEXP (x, 1),
5989 force_to_mode (XEXP (x, 1), GET_MODE (XEXP (x, 1)),
5990 ((unsigned HOST_WIDE_INT) 1
5991 << exact_log2 (GET_MODE_BITSIZE (GET_MODE (x))))
5992 - 1,
5993 0));
5994 break;
5996 default:
5997 break;
6000 return x;
6003 /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6005 static rtx
6006 simplify_if_then_else (rtx x)
6008 enum machine_mode mode = GET_MODE (x);
6009 rtx cond = XEXP (x, 0);
6010 rtx true_rtx = XEXP (x, 1);
6011 rtx false_rtx = XEXP (x, 2);
6012 enum rtx_code true_code = GET_CODE (cond);
6013 int comparison_p = COMPARISON_P (cond);
6014 rtx temp;
6015 int i;
6016 enum rtx_code false_code;
6017 rtx reversed;
6019 /* Simplify storing of the truth value. */
6020 if (comparison_p && true_rtx == const_true_rtx && false_rtx == const0_rtx)
6021 return simplify_gen_relational (true_code, mode, VOIDmode,
6022 XEXP (cond, 0), XEXP (cond, 1));
6024 /* Also when the truth value has to be reversed. */
6025 if (comparison_p
6026 && true_rtx == const0_rtx && false_rtx == const_true_rtx
6027 && (reversed = reversed_comparison (cond, mode)))
6028 return reversed;
6030 /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6031 in it is being compared against certain values. Get the true and false
6032 comparisons and see if that says anything about the value of each arm. */
6034 if (comparison_p
6035 && ((false_code = reversed_comparison_code (cond, NULL))
6036 != UNKNOWN)
6037 && REG_P (XEXP (cond, 0)))
6039 HOST_WIDE_INT nzb;
6040 rtx from = XEXP (cond, 0);
6041 rtx true_val = XEXP (cond, 1);
6042 rtx false_val = true_val;
6043 int swapped = 0;
6045 /* If FALSE_CODE is EQ, swap the codes and arms. */
6047 if (false_code == EQ)
6049 swapped = 1, true_code = EQ, false_code = NE;
6050 temp = true_rtx, true_rtx = false_rtx, false_rtx = temp;
6053 /* If we are comparing against zero and the expression being tested has
6054 only a single bit that might be nonzero, that is its value when it is
6055 not equal to zero. Similarly if it is known to be -1 or 0. */
6057 if (true_code == EQ && true_val == const0_rtx
6058 && exact_log2 (nzb = nonzero_bits (from, GET_MODE (from))) >= 0)
6060 false_code = EQ;
6061 false_val = gen_int_mode (nzb, GET_MODE (from));
6063 else if (true_code == EQ && true_val == const0_rtx
6064 && (num_sign_bit_copies (from, GET_MODE (from))
6065 == GET_MODE_PRECISION (GET_MODE (from))))
6067 false_code = EQ;
6068 false_val = constm1_rtx;
6071 /* Now simplify an arm if we know the value of the register in the
6072 branch and it is used in the arm. Be careful due to the potential
6073 of locally-shared RTL. */
6075 if (reg_mentioned_p (from, true_rtx))
6076 true_rtx = subst (known_cond (copy_rtx (true_rtx), true_code,
6077 from, true_val),
6078 pc_rtx, pc_rtx, 0, 0, 0);
6079 if (reg_mentioned_p (from, false_rtx))
6080 false_rtx = subst (known_cond (copy_rtx (false_rtx), false_code,
6081 from, false_val),
6082 pc_rtx, pc_rtx, 0, 0, 0);
6084 SUBST (XEXP (x, 1), swapped ? false_rtx : true_rtx);
6085 SUBST (XEXP (x, 2), swapped ? true_rtx : false_rtx);
6087 true_rtx = XEXP (x, 1);
6088 false_rtx = XEXP (x, 2);
6089 true_code = GET_CODE (cond);
6092 /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6093 reversed, do so to avoid needing two sets of patterns for
6094 subtract-and-branch insns. Similarly if we have a constant in the true
6095 arm, the false arm is the same as the first operand of the comparison, or
6096 the false arm is more complicated than the true arm. */
6098 if (comparison_p
6099 && reversed_comparison_code (cond, NULL) != UNKNOWN
6100 && (true_rtx == pc_rtx
6101 || (CONSTANT_P (true_rtx)
6102 && !CONST_INT_P (false_rtx) && false_rtx != pc_rtx)
6103 || true_rtx == const0_rtx
6104 || (OBJECT_P (true_rtx) && !OBJECT_P (false_rtx))
6105 || (GET_CODE (true_rtx) == SUBREG && OBJECT_P (SUBREG_REG (true_rtx))
6106 && !OBJECT_P (false_rtx))
6107 || reg_mentioned_p (true_rtx, false_rtx)
6108 || rtx_equal_p (false_rtx, XEXP (cond, 0))))
6110 true_code = reversed_comparison_code (cond, NULL);
6111 SUBST (XEXP (x, 0), reversed_comparison (cond, GET_MODE (cond)));
6112 SUBST (XEXP (x, 1), false_rtx);
6113 SUBST (XEXP (x, 2), true_rtx);
6115 temp = true_rtx, true_rtx = false_rtx, false_rtx = temp;
6116 cond = XEXP (x, 0);
6118 /* It is possible that the conditional has been simplified out. */
6119 true_code = GET_CODE (cond);
6120 comparison_p = COMPARISON_P (cond);
6123 /* If the two arms are identical, we don't need the comparison. */
6125 if (rtx_equal_p (true_rtx, false_rtx) && ! side_effects_p (cond))
6126 return true_rtx;
6128 /* Convert a == b ? b : a to "a". */
6129 if (true_code == EQ && ! side_effects_p (cond)
6130 && !HONOR_NANS (mode)
6131 && rtx_equal_p (XEXP (cond, 0), false_rtx)
6132 && rtx_equal_p (XEXP (cond, 1), true_rtx))
6133 return false_rtx;
6134 else if (true_code == NE && ! side_effects_p (cond)
6135 && !HONOR_NANS (mode)
6136 && rtx_equal_p (XEXP (cond, 0), true_rtx)
6137 && rtx_equal_p (XEXP (cond, 1), false_rtx))
6138 return true_rtx;
6140 /* Look for cases where we have (abs x) or (neg (abs X)). */
6142 if (GET_MODE_CLASS (mode) == MODE_INT
6143 && comparison_p
6144 && XEXP (cond, 1) == const0_rtx
6145 && GET_CODE (false_rtx) == NEG
6146 && rtx_equal_p (true_rtx, XEXP (false_rtx, 0))
6147 && rtx_equal_p (true_rtx, XEXP (cond, 0))
6148 && ! side_effects_p (true_rtx))
6149 switch (true_code)
6151 case GT:
6152 case GE:
6153 return simplify_gen_unary (ABS, mode, true_rtx, mode);
6154 case LT:
6155 case LE:
6156 return
6157 simplify_gen_unary (NEG, mode,
6158 simplify_gen_unary (ABS, mode, true_rtx, mode),
6159 mode);
6160 default:
6161 break;
6164 /* Look for MIN or MAX. */
6166 if ((! FLOAT_MODE_P (mode) || flag_unsafe_math_optimizations)
6167 && comparison_p
6168 && rtx_equal_p (XEXP (cond, 0), true_rtx)
6169 && rtx_equal_p (XEXP (cond, 1), false_rtx)
6170 && ! side_effects_p (cond))
6171 switch (true_code)
6173 case GE:
6174 case GT:
6175 return simplify_gen_binary (SMAX, mode, true_rtx, false_rtx);
6176 case LE:
6177 case LT:
6178 return simplify_gen_binary (SMIN, mode, true_rtx, false_rtx);
6179 case GEU:
6180 case GTU:
6181 return simplify_gen_binary (UMAX, mode, true_rtx, false_rtx);
6182 case LEU:
6183 case LTU:
6184 return simplify_gen_binary (UMIN, mode, true_rtx, false_rtx);
6185 default:
6186 break;
6189 /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6190 second operand is zero, this can be done as (OP Z (mult COND C2)) where
6191 C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6192 SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6193 We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6194 neither 1 or -1, but it isn't worth checking for. */
6196 if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
6197 && comparison_p
6198 && GET_MODE_CLASS (mode) == MODE_INT
6199 && ! side_effects_p (x))
6201 rtx t = make_compound_operation (true_rtx, SET);
6202 rtx f = make_compound_operation (false_rtx, SET);
6203 rtx cond_op0 = XEXP (cond, 0);
6204 rtx cond_op1 = XEXP (cond, 1);
6205 enum rtx_code op = UNKNOWN, extend_op = UNKNOWN;
6206 enum machine_mode m = mode;
6207 rtx z = 0, c1 = NULL_RTX;
6209 if ((GET_CODE (t) == PLUS || GET_CODE (t) == MINUS
6210 || GET_CODE (t) == IOR || GET_CODE (t) == XOR
6211 || GET_CODE (t) == ASHIFT
6212 || GET_CODE (t) == LSHIFTRT || GET_CODE (t) == ASHIFTRT)
6213 && rtx_equal_p (XEXP (t, 0), f))
6214 c1 = XEXP (t, 1), op = GET_CODE (t), z = f;
6216 /* If an identity-zero op is commutative, check whether there
6217 would be a match if we swapped the operands. */
6218 else if ((GET_CODE (t) == PLUS || GET_CODE (t) == IOR
6219 || GET_CODE (t) == XOR)
6220 && rtx_equal_p (XEXP (t, 1), f))
6221 c1 = XEXP (t, 0), op = GET_CODE (t), z = f;
6222 else if (GET_CODE (t) == SIGN_EXTEND
6223 && (GET_CODE (XEXP (t, 0)) == PLUS
6224 || GET_CODE (XEXP (t, 0)) == MINUS
6225 || GET_CODE (XEXP (t, 0)) == IOR
6226 || GET_CODE (XEXP (t, 0)) == XOR
6227 || GET_CODE (XEXP (t, 0)) == ASHIFT
6228 || GET_CODE (XEXP (t, 0)) == LSHIFTRT
6229 || GET_CODE (XEXP (t, 0)) == ASHIFTRT)
6230 && GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
6231 && subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
6232 && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
6233 && (num_sign_bit_copies (f, GET_MODE (f))
6234 > (unsigned int)
6235 (GET_MODE_PRECISION (mode)
6236 - GET_MODE_PRECISION (GET_MODE (XEXP (XEXP (t, 0), 0))))))
6238 c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
6239 extend_op = SIGN_EXTEND;
6240 m = GET_MODE (XEXP (t, 0));
6242 else if (GET_CODE (t) == SIGN_EXTEND
6243 && (GET_CODE (XEXP (t, 0)) == PLUS
6244 || GET_CODE (XEXP (t, 0)) == IOR
6245 || GET_CODE (XEXP (t, 0)) == XOR)
6246 && GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
6247 && subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
6248 && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
6249 && (num_sign_bit_copies (f, GET_MODE (f))
6250 > (unsigned int)
6251 (GET_MODE_PRECISION (mode)
6252 - GET_MODE_PRECISION (GET_MODE (XEXP (XEXP (t, 0), 1))))))
6254 c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
6255 extend_op = SIGN_EXTEND;
6256 m = GET_MODE (XEXP (t, 0));
6258 else if (GET_CODE (t) == ZERO_EXTEND
6259 && (GET_CODE (XEXP (t, 0)) == PLUS
6260 || GET_CODE (XEXP (t, 0)) == MINUS
6261 || GET_CODE (XEXP (t, 0)) == IOR
6262 || GET_CODE (XEXP (t, 0)) == XOR
6263 || GET_CODE (XEXP (t, 0)) == ASHIFT
6264 || GET_CODE (XEXP (t, 0)) == LSHIFTRT
6265 || GET_CODE (XEXP (t, 0)) == ASHIFTRT)
6266 && GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
6267 && HWI_COMPUTABLE_MODE_P (mode)
6268 && subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
6269 && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
6270 && ((nonzero_bits (f, GET_MODE (f))
6271 & ~GET_MODE_MASK (GET_MODE (XEXP (XEXP (t, 0), 0))))
6272 == 0))
6274 c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
6275 extend_op = ZERO_EXTEND;
6276 m = GET_MODE (XEXP (t, 0));
6278 else if (GET_CODE (t) == ZERO_EXTEND
6279 && (GET_CODE (XEXP (t, 0)) == PLUS
6280 || GET_CODE (XEXP (t, 0)) == IOR
6281 || GET_CODE (XEXP (t, 0)) == XOR)
6282 && GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
6283 && HWI_COMPUTABLE_MODE_P (mode)
6284 && subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
6285 && rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
6286 && ((nonzero_bits (f, GET_MODE (f))
6287 & ~GET_MODE_MASK (GET_MODE (XEXP (XEXP (t, 0), 1))))
6288 == 0))
6290 c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
6291 extend_op = ZERO_EXTEND;
6292 m = GET_MODE (XEXP (t, 0));
6295 if (z)
6297 temp = subst (simplify_gen_relational (true_code, m, VOIDmode,
6298 cond_op0, cond_op1),
6299 pc_rtx, pc_rtx, 0, 0, 0);
6300 temp = simplify_gen_binary (MULT, m, temp,
6301 simplify_gen_binary (MULT, m, c1,
6302 const_true_rtx));
6303 temp = subst (temp, pc_rtx, pc_rtx, 0, 0, 0);
6304 temp = simplify_gen_binary (op, m, gen_lowpart (m, z), temp);
6306 if (extend_op != UNKNOWN)
6307 temp = simplify_gen_unary (extend_op, mode, temp, m);
6309 return temp;
6313 /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6314 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6315 negation of a single bit, we can convert this operation to a shift. We
6316 can actually do this more generally, but it doesn't seem worth it. */
6318 if (true_code == NE && XEXP (cond, 1) == const0_rtx
6319 && false_rtx == const0_rtx && CONST_INT_P (true_rtx)
6320 && ((1 == nonzero_bits (XEXP (cond, 0), mode)
6321 && (i = exact_log2 (UINTVAL (true_rtx))) >= 0)
6322 || ((num_sign_bit_copies (XEXP (cond, 0), mode)
6323 == GET_MODE_PRECISION (mode))
6324 && (i = exact_log2 (-UINTVAL (true_rtx))) >= 0)))
6325 return
6326 simplify_shift_const (NULL_RTX, ASHIFT, mode,
6327 gen_lowpart (mode, XEXP (cond, 0)), i);
6329 /* (IF_THEN_ELSE (NE REG 0) (0) (8)) is REG for nonzero_bits (REG) == 8. */
6330 if (true_code == NE && XEXP (cond, 1) == const0_rtx
6331 && false_rtx == const0_rtx && CONST_INT_P (true_rtx)
6332 && GET_MODE (XEXP (cond, 0)) == mode
6333 && (UINTVAL (true_rtx) & GET_MODE_MASK (mode))
6334 == nonzero_bits (XEXP (cond, 0), mode)
6335 && (i = exact_log2 (UINTVAL (true_rtx) & GET_MODE_MASK (mode))) >= 0)
6336 return XEXP (cond, 0);
6338 return x;
6341 /* Simplify X, a SET expression. Return the new expression. */
6343 static rtx
6344 simplify_set (rtx x)
6346 rtx src = SET_SRC (x);
6347 rtx dest = SET_DEST (x);
6348 enum machine_mode mode
6349 = GET_MODE (src) != VOIDmode ? GET_MODE (src) : GET_MODE (dest);
6350 rtx other_insn;
6351 rtx *cc_use;
6353 /* (set (pc) (return)) gets written as (return). */
6354 if (GET_CODE (dest) == PC && ANY_RETURN_P (src))
6355 return src;
6357 /* Now that we know for sure which bits of SRC we are using, see if we can
6358 simplify the expression for the object knowing that we only need the
6359 low-order bits. */
6361 if (GET_MODE_CLASS (mode) == MODE_INT && HWI_COMPUTABLE_MODE_P (mode))
6363 src = force_to_mode (src, mode, ~(unsigned HOST_WIDE_INT) 0, 0);
6364 SUBST (SET_SRC (x), src);
6367 /* If we are setting CC0 or if the source is a COMPARE, look for the use of
6368 the comparison result and try to simplify it unless we already have used
6369 undobuf.other_insn. */
6370 if ((GET_MODE_CLASS (mode) == MODE_CC
6371 || GET_CODE (src) == COMPARE
6372 || CC0_P (dest))
6373 && (cc_use = find_single_use (dest, subst_insn, &other_insn)) != 0
6374 && (undobuf.other_insn == 0 || other_insn == undobuf.other_insn)
6375 && COMPARISON_P (*cc_use)
6376 && rtx_equal_p (XEXP (*cc_use, 0), dest))
6378 enum rtx_code old_code = GET_CODE (*cc_use);
6379 enum rtx_code new_code;
6380 rtx op0, op1, tmp;
6381 int other_changed = 0;
6382 rtx inner_compare = NULL_RTX;
6383 enum machine_mode compare_mode = GET_MODE (dest);
6385 if (GET_CODE (src) == COMPARE)
6387 op0 = XEXP (src, 0), op1 = XEXP (src, 1);
6388 if (GET_CODE (op0) == COMPARE && op1 == const0_rtx)
6390 inner_compare = op0;
6391 op0 = XEXP (inner_compare, 0), op1 = XEXP (inner_compare, 1);
6394 else
6395 op0 = src, op1 = CONST0_RTX (GET_MODE (src));
6397 tmp = simplify_relational_operation (old_code, compare_mode, VOIDmode,
6398 op0, op1);
6399 if (!tmp)
6400 new_code = old_code;
6401 else if (!CONSTANT_P (tmp))
6403 new_code = GET_CODE (tmp);
6404 op0 = XEXP (tmp, 0);
6405 op1 = XEXP (tmp, 1);
6407 else
6409 rtx pat = PATTERN (other_insn);
6410 undobuf.other_insn = other_insn;
6411 SUBST (*cc_use, tmp);
6413 /* Attempt to simplify CC user. */
6414 if (GET_CODE (pat) == SET)
6416 rtx new_rtx = simplify_rtx (SET_SRC (pat));
6417 if (new_rtx != NULL_RTX)
6418 SUBST (SET_SRC (pat), new_rtx);
6421 /* Convert X into a no-op move. */
6422 SUBST (SET_DEST (x), pc_rtx);
6423 SUBST (SET_SRC (x), pc_rtx);
6424 return x;
6427 /* Simplify our comparison, if possible. */
6428 new_code = simplify_comparison (new_code, &op0, &op1);
6430 #ifdef SELECT_CC_MODE
6431 /* If this machine has CC modes other than CCmode, check to see if we
6432 need to use a different CC mode here. */
6433 if (GET_MODE_CLASS (GET_MODE (op0)) == MODE_CC)
6434 compare_mode = GET_MODE (op0);
6435 else if (inner_compare
6436 && GET_MODE_CLASS (GET_MODE (inner_compare)) == MODE_CC
6437 && new_code == old_code
6438 && op0 == XEXP (inner_compare, 0)
6439 && op1 == XEXP (inner_compare, 1))
6440 compare_mode = GET_MODE (inner_compare);
6441 else
6442 compare_mode = SELECT_CC_MODE (new_code, op0, op1);
6444 #ifndef HAVE_cc0
6445 /* If the mode changed, we have to change SET_DEST, the mode in the
6446 compare, and the mode in the place SET_DEST is used. If SET_DEST is
6447 a hard register, just build new versions with the proper mode. If it
6448 is a pseudo, we lose unless it is only time we set the pseudo, in
6449 which case we can safely change its mode. */
6450 if (compare_mode != GET_MODE (dest))
6452 if (can_change_dest_mode (dest, 0, compare_mode))
6454 unsigned int regno = REGNO (dest);
6455 rtx new_dest;
6457 if (regno < FIRST_PSEUDO_REGISTER)
6458 new_dest = gen_rtx_REG (compare_mode, regno);
6459 else
6461 SUBST_MODE (regno_reg_rtx[regno], compare_mode);
6462 new_dest = regno_reg_rtx[regno];
6465 SUBST (SET_DEST (x), new_dest);
6466 SUBST (XEXP (*cc_use, 0), new_dest);
6467 other_changed = 1;
6469 dest = new_dest;
6472 #endif /* cc0 */
6473 #endif /* SELECT_CC_MODE */
6475 /* If the code changed, we have to build a new comparison in
6476 undobuf.other_insn. */
6477 if (new_code != old_code)
6479 int other_changed_previously = other_changed;
6480 unsigned HOST_WIDE_INT mask;
6481 rtx old_cc_use = *cc_use;
6483 SUBST (*cc_use, gen_rtx_fmt_ee (new_code, GET_MODE (*cc_use),
6484 dest, const0_rtx));
6485 other_changed = 1;
6487 /* If the only change we made was to change an EQ into an NE or
6488 vice versa, OP0 has only one bit that might be nonzero, and OP1
6489 is zero, check if changing the user of the condition code will
6490 produce a valid insn. If it won't, we can keep the original code
6491 in that insn by surrounding our operation with an XOR. */
6493 if (((old_code == NE && new_code == EQ)
6494 || (old_code == EQ && new_code == NE))
6495 && ! other_changed_previously && op1 == const0_rtx
6496 && HWI_COMPUTABLE_MODE_P (GET_MODE (op0))
6497 && exact_log2 (mask = nonzero_bits (op0, GET_MODE (op0))) >= 0)
6499 rtx pat = PATTERN (other_insn), note = 0;
6501 if ((recog_for_combine (&pat, other_insn, &note) < 0
6502 && ! check_asm_operands (pat)))
6504 *cc_use = old_cc_use;
6505 other_changed = 0;
6507 op0 = simplify_gen_binary (XOR, GET_MODE (op0),
6508 op0, GEN_INT (mask));
6513 if (other_changed)
6514 undobuf.other_insn = other_insn;
6516 /* Otherwise, if we didn't previously have a COMPARE in the
6517 correct mode, we need one. */
6518 if (GET_CODE (src) != COMPARE || GET_MODE (src) != compare_mode)
6520 SUBST (SET_SRC (x), gen_rtx_COMPARE (compare_mode, op0, op1));
6521 src = SET_SRC (x);
6523 else if (GET_MODE (op0) == compare_mode && op1 == const0_rtx)
6525 SUBST (SET_SRC (x), op0);
6526 src = SET_SRC (x);
6528 /* Otherwise, update the COMPARE if needed. */
6529 else if (XEXP (src, 0) != op0 || XEXP (src, 1) != op1)
6531 SUBST (SET_SRC (x), gen_rtx_COMPARE (compare_mode, op0, op1));
6532 src = SET_SRC (x);
6535 else
6537 /* Get SET_SRC in a form where we have placed back any
6538 compound expressions. Then do the checks below. */
6539 src = make_compound_operation (src, SET);
6540 SUBST (SET_SRC (x), src);
6543 /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
6544 and X being a REG or (subreg (reg)), we may be able to convert this to
6545 (set (subreg:m2 x) (op)).
6547 We can always do this if M1 is narrower than M2 because that means that
6548 we only care about the low bits of the result.
6550 However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
6551 perform a narrower operation than requested since the high-order bits will
6552 be undefined. On machine where it is defined, this transformation is safe
6553 as long as M1 and M2 have the same number of words. */
6555 if (GET_CODE (src) == SUBREG && subreg_lowpart_p (src)
6556 && !OBJECT_P (SUBREG_REG (src))
6557 && (((GET_MODE_SIZE (GET_MODE (src)) + (UNITS_PER_WORD - 1))
6558 / UNITS_PER_WORD)
6559 == ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
6560 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD))
6561 #ifndef WORD_REGISTER_OPERATIONS
6562 && (GET_MODE_SIZE (GET_MODE (src))
6563 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (src))))
6564 #endif
6565 #ifdef CANNOT_CHANGE_MODE_CLASS
6566 && ! (REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER
6567 && REG_CANNOT_CHANGE_MODE_P (REGNO (dest),
6568 GET_MODE (SUBREG_REG (src)),
6569 GET_MODE (src)))
6570 #endif
6571 && (REG_P (dest)
6572 || (GET_CODE (dest) == SUBREG
6573 && REG_P (SUBREG_REG (dest)))))
6575 SUBST (SET_DEST (x),
6576 gen_lowpart (GET_MODE (SUBREG_REG (src)),
6577 dest));
6578 SUBST (SET_SRC (x), SUBREG_REG (src));
6580 src = SET_SRC (x), dest = SET_DEST (x);
6583 #ifdef HAVE_cc0
6584 /* If we have (set (cc0) (subreg ...)), we try to remove the subreg
6585 in SRC. */
6586 if (dest == cc0_rtx
6587 && GET_CODE (src) == SUBREG
6588 && subreg_lowpart_p (src)
6589 && (GET_MODE_PRECISION (GET_MODE (src))
6590 < GET_MODE_PRECISION (GET_MODE (SUBREG_REG (src)))))
6592 rtx inner = SUBREG_REG (src);
6593 enum machine_mode inner_mode = GET_MODE (inner);
6595 /* Here we make sure that we don't have a sign bit on. */
6596 if (val_signbit_known_clear_p (GET_MODE (src),
6597 nonzero_bits (inner, inner_mode)))
6599 SUBST (SET_SRC (x), inner);
6600 src = SET_SRC (x);
6603 #endif
6605 #ifdef LOAD_EXTEND_OP
6606 /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
6607 would require a paradoxical subreg. Replace the subreg with a
6608 zero_extend to avoid the reload that would otherwise be required. */
6610 if (GET_CODE (src) == SUBREG && subreg_lowpart_p (src)
6611 && INTEGRAL_MODE_P (GET_MODE (SUBREG_REG (src)))
6612 && LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (src))) != UNKNOWN
6613 && SUBREG_BYTE (src) == 0
6614 && paradoxical_subreg_p (src)
6615 && MEM_P (SUBREG_REG (src)))
6617 SUBST (SET_SRC (x),
6618 gen_rtx_fmt_e (LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (src))),
6619 GET_MODE (src), SUBREG_REG (src)));
6621 src = SET_SRC (x);
6623 #endif
6625 /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
6626 are comparing an item known to be 0 or -1 against 0, use a logical
6627 operation instead. Check for one of the arms being an IOR of the other
6628 arm with some value. We compute three terms to be IOR'ed together. In
6629 practice, at most two will be nonzero. Then we do the IOR's. */
6631 if (GET_CODE (dest) != PC
6632 && GET_CODE (src) == IF_THEN_ELSE
6633 && GET_MODE_CLASS (GET_MODE (src)) == MODE_INT
6634 && (GET_CODE (XEXP (src, 0)) == EQ || GET_CODE (XEXP (src, 0)) == NE)
6635 && XEXP (XEXP (src, 0), 1) == const0_rtx
6636 && GET_MODE (src) == GET_MODE (XEXP (XEXP (src, 0), 0))
6637 #ifdef HAVE_conditional_move
6638 && ! can_conditionally_move_p (GET_MODE (src))
6639 #endif
6640 && (num_sign_bit_copies (XEXP (XEXP (src, 0), 0),
6641 GET_MODE (XEXP (XEXP (src, 0), 0)))
6642 == GET_MODE_PRECISION (GET_MODE (XEXP (XEXP (src, 0), 0))))
6643 && ! side_effects_p (src))
6645 rtx true_rtx = (GET_CODE (XEXP (src, 0)) == NE
6646 ? XEXP (src, 1) : XEXP (src, 2));
6647 rtx false_rtx = (GET_CODE (XEXP (src, 0)) == NE
6648 ? XEXP (src, 2) : XEXP (src, 1));
6649 rtx term1 = const0_rtx, term2, term3;
6651 if (GET_CODE (true_rtx) == IOR
6652 && rtx_equal_p (XEXP (true_rtx, 0), false_rtx))
6653 term1 = false_rtx, true_rtx = XEXP (true_rtx, 1), false_rtx = const0_rtx;
6654 else if (GET_CODE (true_rtx) == IOR
6655 && rtx_equal_p (XEXP (true_rtx, 1), false_rtx))
6656 term1 = false_rtx, true_rtx = XEXP (true_rtx, 0), false_rtx = const0_rtx;
6657 else if (GET_CODE (false_rtx) == IOR
6658 && rtx_equal_p (XEXP (false_rtx, 0), true_rtx))
6659 term1 = true_rtx, false_rtx = XEXP (false_rtx, 1), true_rtx = const0_rtx;
6660 else if (GET_CODE (false_rtx) == IOR
6661 && rtx_equal_p (XEXP (false_rtx, 1), true_rtx))
6662 term1 = true_rtx, false_rtx = XEXP (false_rtx, 0), true_rtx = const0_rtx;
6664 term2 = simplify_gen_binary (AND, GET_MODE (src),
6665 XEXP (XEXP (src, 0), 0), true_rtx);
6666 term3 = simplify_gen_binary (AND, GET_MODE (src),
6667 simplify_gen_unary (NOT, GET_MODE (src),
6668 XEXP (XEXP (src, 0), 0),
6669 GET_MODE (src)),
6670 false_rtx);
6672 SUBST (SET_SRC (x),
6673 simplify_gen_binary (IOR, GET_MODE (src),
6674 simplify_gen_binary (IOR, GET_MODE (src),
6675 term1, term2),
6676 term3));
6678 src = SET_SRC (x);
6681 /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
6682 whole thing fail. */
6683 if (GET_CODE (src) == CLOBBER && XEXP (src, 0) == const0_rtx)
6684 return src;
6685 else if (GET_CODE (dest) == CLOBBER && XEXP (dest, 0) == const0_rtx)
6686 return dest;
6687 else
6688 /* Convert this into a field assignment operation, if possible. */
6689 return make_field_assignment (x);
6692 /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
6693 result. */
6695 static rtx
6696 simplify_logical (rtx x)
6698 enum machine_mode mode = GET_MODE (x);
6699 rtx op0 = XEXP (x, 0);
6700 rtx op1 = XEXP (x, 1);
6702 switch (GET_CODE (x))
6704 case AND:
6705 /* We can call simplify_and_const_int only if we don't lose
6706 any (sign) bits when converting INTVAL (op1) to
6707 "unsigned HOST_WIDE_INT". */
6708 if (CONST_INT_P (op1)
6709 && (HWI_COMPUTABLE_MODE_P (mode)
6710 || INTVAL (op1) > 0))
6712 x = simplify_and_const_int (x, mode, op0, INTVAL (op1));
6713 if (GET_CODE (x) != AND)
6714 return x;
6716 op0 = XEXP (x, 0);
6717 op1 = XEXP (x, 1);
6720 /* If we have any of (and (ior A B) C) or (and (xor A B) C),
6721 apply the distributive law and then the inverse distributive
6722 law to see if things simplify. */
6723 if (GET_CODE (op0) == IOR || GET_CODE (op0) == XOR)
6725 rtx result = distribute_and_simplify_rtx (x, 0);
6726 if (result)
6727 return result;
6729 if (GET_CODE (op1) == IOR || GET_CODE (op1) == XOR)
6731 rtx result = distribute_and_simplify_rtx (x, 1);
6732 if (result)
6733 return result;
6735 break;
6737 case IOR:
6738 /* If we have (ior (and A B) C), apply the distributive law and then
6739 the inverse distributive law to see if things simplify. */
6741 if (GET_CODE (op0) == AND)
6743 rtx result = distribute_and_simplify_rtx (x, 0);
6744 if (result)
6745 return result;
6748 if (GET_CODE (op1) == AND)
6750 rtx result = distribute_and_simplify_rtx (x, 1);
6751 if (result)
6752 return result;
6754 break;
6756 default:
6757 gcc_unreachable ();
6760 return x;
6763 /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
6764 operations" because they can be replaced with two more basic operations.
6765 ZERO_EXTEND is also considered "compound" because it can be replaced with
6766 an AND operation, which is simpler, though only one operation.
6768 The function expand_compound_operation is called with an rtx expression
6769 and will convert it to the appropriate shifts and AND operations,
6770 simplifying at each stage.
6772 The function make_compound_operation is called to convert an expression
6773 consisting of shifts and ANDs into the equivalent compound expression.
6774 It is the inverse of this function, loosely speaking. */
6776 static rtx
6777 expand_compound_operation (rtx x)
6779 unsigned HOST_WIDE_INT pos = 0, len;
6780 int unsignedp = 0;
6781 unsigned int modewidth;
6782 rtx tem;
6784 switch (GET_CODE (x))
6786 case ZERO_EXTEND:
6787 unsignedp = 1;
6788 case SIGN_EXTEND:
6789 /* We can't necessarily use a const_int for a multiword mode;
6790 it depends on implicitly extending the value.
6791 Since we don't know the right way to extend it,
6792 we can't tell whether the implicit way is right.
6794 Even for a mode that is no wider than a const_int,
6795 we can't win, because we need to sign extend one of its bits through
6796 the rest of it, and we don't know which bit. */
6797 if (CONST_INT_P (XEXP (x, 0)))
6798 return x;
6800 /* Return if (subreg:MODE FROM 0) is not a safe replacement for
6801 (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
6802 because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
6803 reloaded. If not for that, MEM's would very rarely be safe.
6805 Reject MODEs bigger than a word, because we might not be able
6806 to reference a two-register group starting with an arbitrary register
6807 (and currently gen_lowpart might crash for a SUBREG). */
6809 if (GET_MODE_SIZE (GET_MODE (XEXP (x, 0))) > UNITS_PER_WORD)
6810 return x;
6812 /* Reject MODEs that aren't scalar integers because turning vector
6813 or complex modes into shifts causes problems. */
6815 if (! SCALAR_INT_MODE_P (GET_MODE (XEXP (x, 0))))
6816 return x;
6818 len = GET_MODE_PRECISION (GET_MODE (XEXP (x, 0)));
6819 /* If the inner object has VOIDmode (the only way this can happen
6820 is if it is an ASM_OPERANDS), we can't do anything since we don't
6821 know how much masking to do. */
6822 if (len == 0)
6823 return x;
6825 break;
6827 case ZERO_EXTRACT:
6828 unsignedp = 1;
6830 /* ... fall through ... */
6832 case SIGN_EXTRACT:
6833 /* If the operand is a CLOBBER, just return it. */
6834 if (GET_CODE (XEXP (x, 0)) == CLOBBER)
6835 return XEXP (x, 0);
6837 if (!CONST_INT_P (XEXP (x, 1))
6838 || !CONST_INT_P (XEXP (x, 2))
6839 || GET_MODE (XEXP (x, 0)) == VOIDmode)
6840 return x;
6842 /* Reject MODEs that aren't scalar integers because turning vector
6843 or complex modes into shifts causes problems. */
6845 if (! SCALAR_INT_MODE_P (GET_MODE (XEXP (x, 0))))
6846 return x;
6848 len = INTVAL (XEXP (x, 1));
6849 pos = INTVAL (XEXP (x, 2));
6851 /* This should stay within the object being extracted, fail otherwise. */
6852 if (len + pos > GET_MODE_PRECISION (GET_MODE (XEXP (x, 0))))
6853 return x;
6855 if (BITS_BIG_ENDIAN)
6856 pos = GET_MODE_PRECISION (GET_MODE (XEXP (x, 0))) - len - pos;
6858 break;
6860 default:
6861 return x;
6863 /* Convert sign extension to zero extension, if we know that the high
6864 bit is not set, as this is easier to optimize. It will be converted
6865 back to cheaper alternative in make_extraction. */
6866 if (GET_CODE (x) == SIGN_EXTEND
6867 && (HWI_COMPUTABLE_MODE_P (GET_MODE (x))
6868 && ((nonzero_bits (XEXP (x, 0), GET_MODE (XEXP (x, 0)))
6869 & ~(((unsigned HOST_WIDE_INT)
6870 GET_MODE_MASK (GET_MODE (XEXP (x, 0))))
6871 >> 1))
6872 == 0)))
6874 rtx temp = gen_rtx_ZERO_EXTEND (GET_MODE (x), XEXP (x, 0));
6875 rtx temp2 = expand_compound_operation (temp);
6877 /* Make sure this is a profitable operation. */
6878 if (set_src_cost (x, optimize_this_for_speed_p)
6879 > set_src_cost (temp2, optimize_this_for_speed_p))
6880 return temp2;
6881 else if (set_src_cost (x, optimize_this_for_speed_p)
6882 > set_src_cost (temp, optimize_this_for_speed_p))
6883 return temp;
6884 else
6885 return x;
6888 /* We can optimize some special cases of ZERO_EXTEND. */
6889 if (GET_CODE (x) == ZERO_EXTEND)
6891 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
6892 know that the last value didn't have any inappropriate bits
6893 set. */
6894 if (GET_CODE (XEXP (x, 0)) == TRUNCATE
6895 && GET_MODE (XEXP (XEXP (x, 0), 0)) == GET_MODE (x)
6896 && HWI_COMPUTABLE_MODE_P (GET_MODE (x))
6897 && (nonzero_bits (XEXP (XEXP (x, 0), 0), GET_MODE (x))
6898 & ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
6899 return XEXP (XEXP (x, 0), 0);
6901 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
6902 if (GET_CODE (XEXP (x, 0)) == SUBREG
6903 && GET_MODE (SUBREG_REG (XEXP (x, 0))) == GET_MODE (x)
6904 && subreg_lowpart_p (XEXP (x, 0))
6905 && HWI_COMPUTABLE_MODE_P (GET_MODE (x))
6906 && (nonzero_bits (SUBREG_REG (XEXP (x, 0)), GET_MODE (x))
6907 & ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
6908 return SUBREG_REG (XEXP (x, 0));
6910 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
6911 is a comparison and STORE_FLAG_VALUE permits. This is like
6912 the first case, but it works even when GET_MODE (x) is larger
6913 than HOST_WIDE_INT. */
6914 if (GET_CODE (XEXP (x, 0)) == TRUNCATE
6915 && GET_MODE (XEXP (XEXP (x, 0), 0)) == GET_MODE (x)
6916 && COMPARISON_P (XEXP (XEXP (x, 0), 0))
6917 && (GET_MODE_PRECISION (GET_MODE (XEXP (x, 0)))
6918 <= HOST_BITS_PER_WIDE_INT)
6919 && (STORE_FLAG_VALUE & ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
6920 return XEXP (XEXP (x, 0), 0);
6922 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
6923 if (GET_CODE (XEXP (x, 0)) == SUBREG
6924 && GET_MODE (SUBREG_REG (XEXP (x, 0))) == GET_MODE (x)
6925 && subreg_lowpart_p (XEXP (x, 0))
6926 && COMPARISON_P (SUBREG_REG (XEXP (x, 0)))
6927 && (GET_MODE_PRECISION (GET_MODE (XEXP (x, 0)))
6928 <= HOST_BITS_PER_WIDE_INT)
6929 && (STORE_FLAG_VALUE & ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
6930 return SUBREG_REG (XEXP (x, 0));
6934 /* If we reach here, we want to return a pair of shifts. The inner
6935 shift is a left shift of BITSIZE - POS - LEN bits. The outer
6936 shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
6937 logical depending on the value of UNSIGNEDP.
6939 If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
6940 converted into an AND of a shift.
6942 We must check for the case where the left shift would have a negative
6943 count. This can happen in a case like (x >> 31) & 255 on machines
6944 that can't shift by a constant. On those machines, we would first
6945 combine the shift with the AND to produce a variable-position
6946 extraction. Then the constant of 31 would be substituted in
6947 to produce such a position. */
6949 modewidth = GET_MODE_PRECISION (GET_MODE (x));
6950 if (modewidth >= pos + len)
6952 enum machine_mode mode = GET_MODE (x);
6953 tem = gen_lowpart (mode, XEXP (x, 0));
6954 if (!tem || GET_CODE (tem) == CLOBBER)
6955 return x;
6956 tem = simplify_shift_const (NULL_RTX, ASHIFT, mode,
6957 tem, modewidth - pos - len);
6958 tem = simplify_shift_const (NULL_RTX, unsignedp ? LSHIFTRT : ASHIFTRT,
6959 mode, tem, modewidth - len);
6961 else if (unsignedp && len < HOST_BITS_PER_WIDE_INT)
6962 tem = simplify_and_const_int (NULL_RTX, GET_MODE (x),
6963 simplify_shift_const (NULL_RTX, LSHIFTRT,
6964 GET_MODE (x),
6965 XEXP (x, 0), pos),
6966 ((unsigned HOST_WIDE_INT) 1 << len) - 1);
6967 else
6968 /* Any other cases we can't handle. */
6969 return x;
6971 /* If we couldn't do this for some reason, return the original
6972 expression. */
6973 if (GET_CODE (tem) == CLOBBER)
6974 return x;
6976 return tem;
6979 /* X is a SET which contains an assignment of one object into
6980 a part of another (such as a bit-field assignment, STRICT_LOW_PART,
6981 or certain SUBREGS). If possible, convert it into a series of
6982 logical operations.
6984 We half-heartedly support variable positions, but do not at all
6985 support variable lengths. */
6987 static const_rtx
6988 expand_field_assignment (const_rtx x)
6990 rtx inner;
6991 rtx pos; /* Always counts from low bit. */
6992 int len;
6993 rtx mask, cleared, masked;
6994 enum machine_mode compute_mode;
6996 /* Loop until we find something we can't simplify. */
6997 while (1)
6999 if (GET_CODE (SET_DEST (x)) == STRICT_LOW_PART
7000 && GET_CODE (XEXP (SET_DEST (x), 0)) == SUBREG)
7002 inner = SUBREG_REG (XEXP (SET_DEST (x), 0));
7003 len = GET_MODE_PRECISION (GET_MODE (XEXP (SET_DEST (x), 0)));
7004 pos = GEN_INT (subreg_lsb (XEXP (SET_DEST (x), 0)));
7006 else if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
7007 && CONST_INT_P (XEXP (SET_DEST (x), 1)))
7009 inner = XEXP (SET_DEST (x), 0);
7010 len = INTVAL (XEXP (SET_DEST (x), 1));
7011 pos = XEXP (SET_DEST (x), 2);
7013 /* A constant position should stay within the width of INNER. */
7014 if (CONST_INT_P (pos)
7015 && INTVAL (pos) + len > GET_MODE_PRECISION (GET_MODE (inner)))
7016 break;
7018 if (BITS_BIG_ENDIAN)
7020 if (CONST_INT_P (pos))
7021 pos = GEN_INT (GET_MODE_PRECISION (GET_MODE (inner)) - len
7022 - INTVAL (pos));
7023 else if (GET_CODE (pos) == MINUS
7024 && CONST_INT_P (XEXP (pos, 1))
7025 && (INTVAL (XEXP (pos, 1))
7026 == GET_MODE_PRECISION (GET_MODE (inner)) - len))
7027 /* If position is ADJUST - X, new position is X. */
7028 pos = XEXP (pos, 0);
7029 else
7030 pos = simplify_gen_binary (MINUS, GET_MODE (pos),
7031 GEN_INT (GET_MODE_PRECISION (
7032 GET_MODE (inner))
7033 - len),
7034 pos);
7038 /* A SUBREG between two modes that occupy the same numbers of words
7039 can be done by moving the SUBREG to the source. */
7040 else if (GET_CODE (SET_DEST (x)) == SUBREG
7041 /* We need SUBREGs to compute nonzero_bits properly. */
7042 && nonzero_sign_valid
7043 && (((GET_MODE_SIZE (GET_MODE (SET_DEST (x)))
7044 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
7045 == ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (SET_DEST (x))))
7046 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)))
7048 x = gen_rtx_SET (VOIDmode, SUBREG_REG (SET_DEST (x)),
7049 gen_lowpart
7050 (GET_MODE (SUBREG_REG (SET_DEST (x))),
7051 SET_SRC (x)));
7052 continue;
7054 else
7055 break;
7057 while (GET_CODE (inner) == SUBREG && subreg_lowpart_p (inner))
7058 inner = SUBREG_REG (inner);
7060 compute_mode = GET_MODE (inner);
7062 /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7063 if (! SCALAR_INT_MODE_P (compute_mode))
7065 enum machine_mode imode;
7067 /* Don't do anything for vector or complex integral types. */
7068 if (! FLOAT_MODE_P (compute_mode))
7069 break;
7071 /* Try to find an integral mode to pun with. */
7072 imode = mode_for_size (GET_MODE_BITSIZE (compute_mode), MODE_INT, 0);
7073 if (imode == BLKmode)
7074 break;
7076 compute_mode = imode;
7077 inner = gen_lowpart (imode, inner);
7080 /* Compute a mask of LEN bits, if we can do this on the host machine. */
7081 if (len >= HOST_BITS_PER_WIDE_INT)
7082 break;
7084 /* Now compute the equivalent expression. Make a copy of INNER
7085 for the SET_DEST in case it is a MEM into which we will substitute;
7086 we don't want shared RTL in that case. */
7087 mask = GEN_INT (((unsigned HOST_WIDE_INT) 1 << len) - 1);
7088 cleared = simplify_gen_binary (AND, compute_mode,
7089 simplify_gen_unary (NOT, compute_mode,
7090 simplify_gen_binary (ASHIFT,
7091 compute_mode,
7092 mask, pos),
7093 compute_mode),
7094 inner);
7095 masked = simplify_gen_binary (ASHIFT, compute_mode,
7096 simplify_gen_binary (
7097 AND, compute_mode,
7098 gen_lowpart (compute_mode, SET_SRC (x)),
7099 mask),
7100 pos);
7102 x = gen_rtx_SET (VOIDmode, copy_rtx (inner),
7103 simplify_gen_binary (IOR, compute_mode,
7104 cleared, masked));
7107 return x;
7110 /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7111 it is an RTX that represents a variable starting position; otherwise,
7112 POS is the (constant) starting bit position (counted from the LSB).
7114 UNSIGNEDP is nonzero for an unsigned reference and zero for a
7115 signed reference.
7117 IN_DEST is nonzero if this is a reference in the destination of a
7118 SET. This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7119 a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7120 be used.
7122 IN_COMPARE is nonzero if we are in a COMPARE. This means that a
7123 ZERO_EXTRACT should be built even for bits starting at bit 0.
7125 MODE is the desired mode of the result (if IN_DEST == 0).
7127 The result is an RTX for the extraction or NULL_RTX if the target
7128 can't handle it. */
7130 static rtx
7131 make_extraction (enum machine_mode mode, rtx inner, HOST_WIDE_INT pos,
7132 rtx pos_rtx, unsigned HOST_WIDE_INT len, int unsignedp,
7133 int in_dest, int in_compare)
7135 /* This mode describes the size of the storage area
7136 to fetch the overall value from. Within that, we
7137 ignore the POS lowest bits, etc. */
7138 enum machine_mode is_mode = GET_MODE (inner);
7139 enum machine_mode inner_mode;
7140 enum machine_mode wanted_inner_mode;
7141 enum machine_mode wanted_inner_reg_mode = word_mode;
7142 enum machine_mode pos_mode = word_mode;
7143 enum machine_mode extraction_mode = word_mode;
7144 enum machine_mode tmode = mode_for_size (len, MODE_INT, 1);
7145 rtx new_rtx = 0;
7146 rtx orig_pos_rtx = pos_rtx;
7147 HOST_WIDE_INT orig_pos;
7149 if (GET_CODE (inner) == SUBREG && subreg_lowpart_p (inner))
7151 /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7152 consider just the QI as the memory to extract from.
7153 The subreg adds or removes high bits; its mode is
7154 irrelevant to the meaning of this extraction,
7155 since POS and LEN count from the lsb. */
7156 if (MEM_P (SUBREG_REG (inner)))
7157 is_mode = GET_MODE (SUBREG_REG (inner));
7158 inner = SUBREG_REG (inner);
7160 else if (GET_CODE (inner) == ASHIFT
7161 && CONST_INT_P (XEXP (inner, 1))
7162 && pos_rtx == 0 && pos == 0
7163 && len > UINTVAL (XEXP (inner, 1)))
7165 /* We're extracting the least significant bits of an rtx
7166 (ashift X (const_int C)), where LEN > C. Extract the
7167 least significant (LEN - C) bits of X, giving an rtx
7168 whose mode is MODE, then shift it left C times. */
7169 new_rtx = make_extraction (mode, XEXP (inner, 0),
7170 0, 0, len - INTVAL (XEXP (inner, 1)),
7171 unsignedp, in_dest, in_compare);
7172 if (new_rtx != 0)
7173 return gen_rtx_ASHIFT (mode, new_rtx, XEXP (inner, 1));
7176 inner_mode = GET_MODE (inner);
7178 if (pos_rtx && CONST_INT_P (pos_rtx))
7179 pos = INTVAL (pos_rtx), pos_rtx = 0;
7181 /* See if this can be done without an extraction. We never can if the
7182 width of the field is not the same as that of some integer mode. For
7183 registers, we can only avoid the extraction if the position is at the
7184 low-order bit and this is either not in the destination or we have the
7185 appropriate STRICT_LOW_PART operation available.
7187 For MEM, we can avoid an extract if the field starts on an appropriate
7188 boundary and we can change the mode of the memory reference. */
7190 if (tmode != BLKmode
7191 && ((pos_rtx == 0 && (pos % BITS_PER_WORD) == 0
7192 && !MEM_P (inner)
7193 && (inner_mode == tmode
7194 || !REG_P (inner)
7195 || TRULY_NOOP_TRUNCATION_MODES_P (tmode, inner_mode)
7196 || reg_truncated_to_mode (tmode, inner))
7197 && (! in_dest
7198 || (REG_P (inner)
7199 && have_insn_for (STRICT_LOW_PART, tmode))))
7200 || (MEM_P (inner) && pos_rtx == 0
7201 && (pos
7202 % (STRICT_ALIGNMENT ? GET_MODE_ALIGNMENT (tmode)
7203 : BITS_PER_UNIT)) == 0
7204 /* We can't do this if we are widening INNER_MODE (it
7205 may not be aligned, for one thing). */
7206 && GET_MODE_PRECISION (inner_mode) >= GET_MODE_PRECISION (tmode)
7207 && (inner_mode == tmode
7208 || (! mode_dependent_address_p (XEXP (inner, 0))
7209 && ! MEM_VOLATILE_P (inner))))))
7211 /* If INNER is a MEM, make a new MEM that encompasses just the desired
7212 field. If the original and current mode are the same, we need not
7213 adjust the offset. Otherwise, we do if bytes big endian.
7215 If INNER is not a MEM, get a piece consisting of just the field
7216 of interest (in this case POS % BITS_PER_WORD must be 0). */
7218 if (MEM_P (inner))
7220 HOST_WIDE_INT offset;
7222 /* POS counts from lsb, but make OFFSET count in memory order. */
7223 if (BYTES_BIG_ENDIAN)
7224 offset = (GET_MODE_PRECISION (is_mode) - len - pos) / BITS_PER_UNIT;
7225 else
7226 offset = pos / BITS_PER_UNIT;
7228 new_rtx = adjust_address_nv (inner, tmode, offset);
7230 else if (REG_P (inner))
7232 if (tmode != inner_mode)
7234 /* We can't call gen_lowpart in a DEST since we
7235 always want a SUBREG (see below) and it would sometimes
7236 return a new hard register. */
7237 if (pos || in_dest)
7239 HOST_WIDE_INT final_word = pos / BITS_PER_WORD;
7241 if (WORDS_BIG_ENDIAN
7242 && GET_MODE_SIZE (inner_mode) > UNITS_PER_WORD)
7243 final_word = ((GET_MODE_SIZE (inner_mode)
7244 - GET_MODE_SIZE (tmode))
7245 / UNITS_PER_WORD) - final_word;
7247 final_word *= UNITS_PER_WORD;
7248 if (BYTES_BIG_ENDIAN &&
7249 GET_MODE_SIZE (inner_mode) > GET_MODE_SIZE (tmode))
7250 final_word += (GET_MODE_SIZE (inner_mode)
7251 - GET_MODE_SIZE (tmode)) % UNITS_PER_WORD;
7253 /* Avoid creating invalid subregs, for example when
7254 simplifying (x>>32)&255. */
7255 if (!validate_subreg (tmode, inner_mode, inner, final_word))
7256 return NULL_RTX;
7258 new_rtx = gen_rtx_SUBREG (tmode, inner, final_word);
7260 else
7261 new_rtx = gen_lowpart (tmode, inner);
7263 else
7264 new_rtx = inner;
7266 else
7267 new_rtx = force_to_mode (inner, tmode,
7268 len >= HOST_BITS_PER_WIDE_INT
7269 ? ~(unsigned HOST_WIDE_INT) 0
7270 : ((unsigned HOST_WIDE_INT) 1 << len) - 1,
7273 /* If this extraction is going into the destination of a SET,
7274 make a STRICT_LOW_PART unless we made a MEM. */
7276 if (in_dest)
7277 return (MEM_P (new_rtx) ? new_rtx
7278 : (GET_CODE (new_rtx) != SUBREG
7279 ? gen_rtx_CLOBBER (tmode, const0_rtx)
7280 : gen_rtx_STRICT_LOW_PART (VOIDmode, new_rtx)));
7282 if (mode == tmode)
7283 return new_rtx;
7285 if (CONST_INT_P (new_rtx)
7286 || GET_CODE (new_rtx) == CONST_DOUBLE)
7287 return simplify_unary_operation (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
7288 mode, new_rtx, tmode);
7290 /* If we know that no extraneous bits are set, and that the high
7291 bit is not set, convert the extraction to the cheaper of
7292 sign and zero extension, that are equivalent in these cases. */
7293 if (flag_expensive_optimizations
7294 && (HWI_COMPUTABLE_MODE_P (tmode)
7295 && ((nonzero_bits (new_rtx, tmode)
7296 & ~(((unsigned HOST_WIDE_INT)GET_MODE_MASK (tmode)) >> 1))
7297 == 0)))
7299 rtx temp = gen_rtx_ZERO_EXTEND (mode, new_rtx);
7300 rtx temp1 = gen_rtx_SIGN_EXTEND (mode, new_rtx);
7302 /* Prefer ZERO_EXTENSION, since it gives more information to
7303 backends. */
7304 if (set_src_cost (temp, optimize_this_for_speed_p)
7305 <= set_src_cost (temp1, optimize_this_for_speed_p))
7306 return temp;
7307 return temp1;
7310 /* Otherwise, sign- or zero-extend unless we already are in the
7311 proper mode. */
7313 return (gen_rtx_fmt_e (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
7314 mode, new_rtx));
7317 /* Unless this is a COMPARE or we have a funny memory reference,
7318 don't do anything with zero-extending field extracts starting at
7319 the low-order bit since they are simple AND operations. */
7320 if (pos_rtx == 0 && pos == 0 && ! in_dest
7321 && ! in_compare && unsignedp)
7322 return 0;
7324 /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7325 if the position is not a constant and the length is not 1. In all
7326 other cases, we would only be going outside our object in cases when
7327 an original shift would have been undefined. */
7328 if (MEM_P (inner)
7329 && ((pos_rtx == 0 && pos + len > GET_MODE_PRECISION (is_mode))
7330 || (pos_rtx != 0 && len != 1)))
7331 return 0;
7333 /* Get the mode to use should INNER not be a MEM, the mode for the position,
7334 and the mode for the result. */
7335 if (in_dest && mode_for_extraction (EP_insv, -1) != MAX_MACHINE_MODE)
7337 wanted_inner_reg_mode = mode_for_extraction (EP_insv, 0);
7338 pos_mode = mode_for_extraction (EP_insv, 2);
7339 extraction_mode = mode_for_extraction (EP_insv, 3);
7342 if (! in_dest && unsignedp
7343 && mode_for_extraction (EP_extzv, -1) != MAX_MACHINE_MODE)
7345 wanted_inner_reg_mode = mode_for_extraction (EP_extzv, 1);
7346 pos_mode = mode_for_extraction (EP_extzv, 3);
7347 extraction_mode = mode_for_extraction (EP_extzv, 0);
7350 if (! in_dest && ! unsignedp
7351 && mode_for_extraction (EP_extv, -1) != MAX_MACHINE_MODE)
7353 wanted_inner_reg_mode = mode_for_extraction (EP_extv, 1);
7354 pos_mode = mode_for_extraction (EP_extv, 3);
7355 extraction_mode = mode_for_extraction (EP_extv, 0);
7358 /* Never narrow an object, since that might not be safe. */
7360 if (mode != VOIDmode
7361 && GET_MODE_SIZE (extraction_mode) < GET_MODE_SIZE (mode))
7362 extraction_mode = mode;
7364 if (pos_rtx && GET_MODE (pos_rtx) != VOIDmode
7365 && GET_MODE_SIZE (pos_mode) < GET_MODE_SIZE (GET_MODE (pos_rtx)))
7366 pos_mode = GET_MODE (pos_rtx);
7368 /* If this is not from memory, the desired mode is the preferred mode
7369 for an extraction pattern's first input operand, or word_mode if there
7370 is none. */
7371 if (!MEM_P (inner))
7372 wanted_inner_mode = wanted_inner_reg_mode;
7373 else
7375 /* Be careful not to go beyond the extracted object and maintain the
7376 natural alignment of the memory. */
7377 wanted_inner_mode = smallest_mode_for_size (len, MODE_INT);
7378 while (pos % GET_MODE_BITSIZE (wanted_inner_mode) + len
7379 > GET_MODE_BITSIZE (wanted_inner_mode))
7381 wanted_inner_mode = GET_MODE_WIDER_MODE (wanted_inner_mode);
7382 gcc_assert (wanted_inner_mode != VOIDmode);
7385 /* If we have to change the mode of memory and cannot, the desired mode
7386 is EXTRACTION_MODE. */
7387 if (inner_mode != wanted_inner_mode
7388 && (mode_dependent_address_p (XEXP (inner, 0))
7389 || MEM_VOLATILE_P (inner)
7390 || pos_rtx))
7391 wanted_inner_mode = extraction_mode;
7394 orig_pos = pos;
7396 if (BITS_BIG_ENDIAN)
7398 /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7399 BITS_BIG_ENDIAN style. If position is constant, compute new
7400 position. Otherwise, build subtraction.
7401 Note that POS is relative to the mode of the original argument.
7402 If it's a MEM we need to recompute POS relative to that.
7403 However, if we're extracting from (or inserting into) a register,
7404 we want to recompute POS relative to wanted_inner_mode. */
7405 int width = (MEM_P (inner)
7406 ? GET_MODE_BITSIZE (is_mode)
7407 : GET_MODE_BITSIZE (wanted_inner_mode));
7409 if (pos_rtx == 0)
7410 pos = width - len - pos;
7411 else
7412 pos_rtx
7413 = gen_rtx_MINUS (GET_MODE (pos_rtx), GEN_INT (width - len), pos_rtx);
7414 /* POS may be less than 0 now, but we check for that below.
7415 Note that it can only be less than 0 if !MEM_P (inner). */
7418 /* If INNER has a wider mode, and this is a constant extraction, try to
7419 make it smaller and adjust the byte to point to the byte containing
7420 the value. */
7421 if (wanted_inner_mode != VOIDmode
7422 && inner_mode != wanted_inner_mode
7423 && ! pos_rtx
7424 && GET_MODE_SIZE (wanted_inner_mode) < GET_MODE_SIZE (is_mode)
7425 && MEM_P (inner)
7426 && ! mode_dependent_address_p (XEXP (inner, 0))
7427 && ! MEM_VOLATILE_P (inner))
7429 int offset = 0;
7431 /* The computations below will be correct if the machine is big
7432 endian in both bits and bytes or little endian in bits and bytes.
7433 If it is mixed, we must adjust. */
7435 /* If bytes are big endian and we had a paradoxical SUBREG, we must
7436 adjust OFFSET to compensate. */
7437 if (BYTES_BIG_ENDIAN
7438 && GET_MODE_SIZE (inner_mode) < GET_MODE_SIZE (is_mode))
7439 offset -= GET_MODE_SIZE (is_mode) - GET_MODE_SIZE (inner_mode);
7441 /* We can now move to the desired byte. */
7442 offset += (pos / GET_MODE_BITSIZE (wanted_inner_mode))
7443 * GET_MODE_SIZE (wanted_inner_mode);
7444 pos %= GET_MODE_BITSIZE (wanted_inner_mode);
7446 if (BYTES_BIG_ENDIAN != BITS_BIG_ENDIAN
7447 && is_mode != wanted_inner_mode)
7448 offset = (GET_MODE_SIZE (is_mode)
7449 - GET_MODE_SIZE (wanted_inner_mode) - offset);
7451 inner = adjust_address_nv (inner, wanted_inner_mode, offset);
7454 /* If INNER is not memory, get it into the proper mode. If we are changing
7455 its mode, POS must be a constant and smaller than the size of the new
7456 mode. */
7457 else if (!MEM_P (inner))
7459 /* On the LHS, don't create paradoxical subregs implicitely truncating
7460 the register unless TRULY_NOOP_TRUNCATION. */
7461 if (in_dest
7462 && !TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (inner),
7463 wanted_inner_mode))
7464 return NULL_RTX;
7466 if (GET_MODE (inner) != wanted_inner_mode
7467 && (pos_rtx != 0
7468 || orig_pos + len > GET_MODE_BITSIZE (wanted_inner_mode)))
7469 return NULL_RTX;
7471 if (orig_pos < 0)
7472 return NULL_RTX;
7474 inner = force_to_mode (inner, wanted_inner_mode,
7475 pos_rtx
7476 || len + orig_pos >= HOST_BITS_PER_WIDE_INT
7477 ? ~(unsigned HOST_WIDE_INT) 0
7478 : ((((unsigned HOST_WIDE_INT) 1 << len) - 1)
7479 << orig_pos),
7483 /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7484 have to zero extend. Otherwise, we can just use a SUBREG. */
7485 if (pos_rtx != 0
7486 && GET_MODE_SIZE (pos_mode) > GET_MODE_SIZE (GET_MODE (pos_rtx)))
7488 rtx temp = gen_rtx_ZERO_EXTEND (pos_mode, pos_rtx);
7490 /* If we know that no extraneous bits are set, and that the high
7491 bit is not set, convert extraction to cheaper one - either
7492 SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7493 cases. */
7494 if (flag_expensive_optimizations
7495 && (HWI_COMPUTABLE_MODE_P (GET_MODE (pos_rtx))
7496 && ((nonzero_bits (pos_rtx, GET_MODE (pos_rtx))
7497 & ~(((unsigned HOST_WIDE_INT)
7498 GET_MODE_MASK (GET_MODE (pos_rtx)))
7499 >> 1))
7500 == 0)))
7502 rtx temp1 = gen_rtx_SIGN_EXTEND (pos_mode, pos_rtx);
7504 /* Prefer ZERO_EXTENSION, since it gives more information to
7505 backends. */
7506 if (set_src_cost (temp1, optimize_this_for_speed_p)
7507 < set_src_cost (temp, optimize_this_for_speed_p))
7508 temp = temp1;
7510 pos_rtx = temp;
7512 else if (pos_rtx != 0
7513 && GET_MODE_SIZE (pos_mode) < GET_MODE_SIZE (GET_MODE (pos_rtx)))
7514 pos_rtx = gen_lowpart (pos_mode, pos_rtx);
7516 /* Make POS_RTX unless we already have it and it is correct. If we don't
7517 have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7518 be a CONST_INT. */
7519 if (pos_rtx == 0 && orig_pos_rtx != 0 && INTVAL (orig_pos_rtx) == pos)
7520 pos_rtx = orig_pos_rtx;
7522 else if (pos_rtx == 0)
7523 pos_rtx = GEN_INT (pos);
7525 /* Make the required operation. See if we can use existing rtx. */
7526 new_rtx = gen_rtx_fmt_eee (unsignedp ? ZERO_EXTRACT : SIGN_EXTRACT,
7527 extraction_mode, inner, GEN_INT (len), pos_rtx);
7528 if (! in_dest)
7529 new_rtx = gen_lowpart (mode, new_rtx);
7531 return new_rtx;
7534 /* See if X contains an ASHIFT of COUNT or more bits that can be commuted
7535 with any other operations in X. Return X without that shift if so. */
7537 static rtx
7538 extract_left_shift (rtx x, int count)
7540 enum rtx_code code = GET_CODE (x);
7541 enum machine_mode mode = GET_MODE (x);
7542 rtx tem;
7544 switch (code)
7546 case ASHIFT:
7547 /* This is the shift itself. If it is wide enough, we will return
7548 either the value being shifted if the shift count is equal to
7549 COUNT or a shift for the difference. */
7550 if (CONST_INT_P (XEXP (x, 1))
7551 && INTVAL (XEXP (x, 1)) >= count)
7552 return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (x, 0),
7553 INTVAL (XEXP (x, 1)) - count);
7554 break;
7556 case NEG: case NOT:
7557 if ((tem = extract_left_shift (XEXP (x, 0), count)) != 0)
7558 return simplify_gen_unary (code, mode, tem, mode);
7560 break;
7562 case PLUS: case IOR: case XOR: case AND:
7563 /* If we can safely shift this constant and we find the inner shift,
7564 make a new operation. */
7565 if (CONST_INT_P (XEXP (x, 1))
7566 && (UINTVAL (XEXP (x, 1))
7567 & ((((unsigned HOST_WIDE_INT) 1 << count)) - 1)) == 0
7568 && (tem = extract_left_shift (XEXP (x, 0), count)) != 0)
7569 return simplify_gen_binary (code, mode, tem,
7570 GEN_INT (INTVAL (XEXP (x, 1)) >> count));
7572 break;
7574 default:
7575 break;
7578 return 0;
7581 /* Look at the expression rooted at X. Look for expressions
7582 equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
7583 Form these expressions.
7585 Return the new rtx, usually just X.
7587 Also, for machines like the VAX that don't have logical shift insns,
7588 try to convert logical to arithmetic shift operations in cases where
7589 they are equivalent. This undoes the canonicalizations to logical
7590 shifts done elsewhere.
7592 We try, as much as possible, to re-use rtl expressions to save memory.
7594 IN_CODE says what kind of expression we are processing. Normally, it is
7595 SET. In a memory address (inside a MEM, PLUS or minus, the latter two
7596 being kludges), it is MEM. When processing the arguments of a comparison
7597 or a COMPARE against zero, it is COMPARE. */
7599 static rtx
7600 make_compound_operation (rtx x, enum rtx_code in_code)
7602 enum rtx_code code = GET_CODE (x);
7603 enum machine_mode mode = GET_MODE (x);
7604 int mode_width = GET_MODE_PRECISION (mode);
7605 rtx rhs, lhs;
7606 enum rtx_code next_code;
7607 int i, j;
7608 rtx new_rtx = 0;
7609 rtx tem;
7610 const char *fmt;
7612 /* Select the code to be used in recursive calls. Once we are inside an
7613 address, we stay there. If we have a comparison, set to COMPARE,
7614 but once inside, go back to our default of SET. */
7616 next_code = (code == MEM ? MEM
7617 : ((code == PLUS || code == MINUS)
7618 && SCALAR_INT_MODE_P (mode)) ? MEM
7619 : ((code == COMPARE || COMPARISON_P (x))
7620 && XEXP (x, 1) == const0_rtx) ? COMPARE
7621 : in_code == COMPARE ? SET : in_code);
7623 /* Process depending on the code of this operation. If NEW is set
7624 nonzero, it will be returned. */
7626 switch (code)
7628 case ASHIFT:
7629 /* Convert shifts by constants into multiplications if inside
7630 an address. */
7631 if (in_code == MEM && CONST_INT_P (XEXP (x, 1))
7632 && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
7633 && INTVAL (XEXP (x, 1)) >= 0
7634 && SCALAR_INT_MODE_P (mode))
7636 HOST_WIDE_INT count = INTVAL (XEXP (x, 1));
7637 HOST_WIDE_INT multval = (HOST_WIDE_INT) 1 << count;
7639 new_rtx = make_compound_operation (XEXP (x, 0), next_code);
7640 if (GET_CODE (new_rtx) == NEG)
7642 new_rtx = XEXP (new_rtx, 0);
7643 multval = -multval;
7645 multval = trunc_int_for_mode (multval, mode);
7646 new_rtx = gen_rtx_MULT (mode, new_rtx, GEN_INT (multval));
7648 break;
7650 case PLUS:
7651 lhs = XEXP (x, 0);
7652 rhs = XEXP (x, 1);
7653 lhs = make_compound_operation (lhs, next_code);
7654 rhs = make_compound_operation (rhs, next_code);
7655 if (GET_CODE (lhs) == MULT && GET_CODE (XEXP (lhs, 0)) == NEG
7656 && SCALAR_INT_MODE_P (mode))
7658 tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (lhs, 0), 0),
7659 XEXP (lhs, 1));
7660 new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
7662 else if (GET_CODE (lhs) == MULT
7663 && (CONST_INT_P (XEXP (lhs, 1)) && INTVAL (XEXP (lhs, 1)) < 0))
7665 tem = simplify_gen_binary (MULT, mode, XEXP (lhs, 0),
7666 simplify_gen_unary (NEG, mode,
7667 XEXP (lhs, 1),
7668 mode));
7669 new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
7671 else
7673 SUBST (XEXP (x, 0), lhs);
7674 SUBST (XEXP (x, 1), rhs);
7675 goto maybe_swap;
7677 x = gen_lowpart (mode, new_rtx);
7678 goto maybe_swap;
7680 case MINUS:
7681 lhs = XEXP (x, 0);
7682 rhs = XEXP (x, 1);
7683 lhs = make_compound_operation (lhs, next_code);
7684 rhs = make_compound_operation (rhs, next_code);
7685 if (GET_CODE (rhs) == MULT && GET_CODE (XEXP (rhs, 0)) == NEG
7686 && SCALAR_INT_MODE_P (mode))
7688 tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (rhs, 0), 0),
7689 XEXP (rhs, 1));
7690 new_rtx = simplify_gen_binary (PLUS, mode, tem, lhs);
7692 else if (GET_CODE (rhs) == MULT
7693 && (CONST_INT_P (XEXP (rhs, 1)) && INTVAL (XEXP (rhs, 1)) < 0))
7695 tem = simplify_gen_binary (MULT, mode, XEXP (rhs, 0),
7696 simplify_gen_unary (NEG, mode,
7697 XEXP (rhs, 1),
7698 mode));
7699 new_rtx = simplify_gen_binary (PLUS, mode, tem, lhs);
7701 else
7703 SUBST (XEXP (x, 0), lhs);
7704 SUBST (XEXP (x, 1), rhs);
7705 return x;
7707 return gen_lowpart (mode, new_rtx);
7709 case AND:
7710 /* If the second operand is not a constant, we can't do anything
7711 with it. */
7712 if (!CONST_INT_P (XEXP (x, 1)))
7713 break;
7715 /* If the constant is a power of two minus one and the first operand
7716 is a logical right shift, make an extraction. */
7717 if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
7718 && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
7720 new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
7721 new_rtx = make_extraction (mode, new_rtx, 0, XEXP (XEXP (x, 0), 1), i, 1,
7722 0, in_code == COMPARE);
7725 /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
7726 else if (GET_CODE (XEXP (x, 0)) == SUBREG
7727 && subreg_lowpart_p (XEXP (x, 0))
7728 && GET_CODE (SUBREG_REG (XEXP (x, 0))) == LSHIFTRT
7729 && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
7731 new_rtx = make_compound_operation (XEXP (SUBREG_REG (XEXP (x, 0)), 0),
7732 next_code);
7733 new_rtx = make_extraction (GET_MODE (SUBREG_REG (XEXP (x, 0))), new_rtx, 0,
7734 XEXP (SUBREG_REG (XEXP (x, 0)), 1), i, 1,
7735 0, in_code == COMPARE);
7737 /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
7738 else if ((GET_CODE (XEXP (x, 0)) == XOR
7739 || GET_CODE (XEXP (x, 0)) == IOR)
7740 && GET_CODE (XEXP (XEXP (x, 0), 0)) == LSHIFTRT
7741 && GET_CODE (XEXP (XEXP (x, 0), 1)) == LSHIFTRT
7742 && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
7744 /* Apply the distributive law, and then try to make extractions. */
7745 new_rtx = gen_rtx_fmt_ee (GET_CODE (XEXP (x, 0)), mode,
7746 gen_rtx_AND (mode, XEXP (XEXP (x, 0), 0),
7747 XEXP (x, 1)),
7748 gen_rtx_AND (mode, XEXP (XEXP (x, 0), 1),
7749 XEXP (x, 1)));
7750 new_rtx = make_compound_operation (new_rtx, in_code);
7753 /* If we are have (and (rotate X C) M) and C is larger than the number
7754 of bits in M, this is an extraction. */
7756 else if (GET_CODE (XEXP (x, 0)) == ROTATE
7757 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
7758 && (i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0
7759 && i <= INTVAL (XEXP (XEXP (x, 0), 1)))
7761 new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
7762 new_rtx = make_extraction (mode, new_rtx,
7763 (GET_MODE_PRECISION (mode)
7764 - INTVAL (XEXP (XEXP (x, 0), 1))),
7765 NULL_RTX, i, 1, 0, in_code == COMPARE);
7768 /* On machines without logical shifts, if the operand of the AND is
7769 a logical shift and our mask turns off all the propagated sign
7770 bits, we can replace the logical shift with an arithmetic shift. */
7771 else if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
7772 && !have_insn_for (LSHIFTRT, mode)
7773 && have_insn_for (ASHIFTRT, mode)
7774 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
7775 && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
7776 && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
7777 && mode_width <= HOST_BITS_PER_WIDE_INT)
7779 unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
7781 mask >>= INTVAL (XEXP (XEXP (x, 0), 1));
7782 if ((INTVAL (XEXP (x, 1)) & ~mask) == 0)
7783 SUBST (XEXP (x, 0),
7784 gen_rtx_ASHIFTRT (mode,
7785 make_compound_operation
7786 (XEXP (XEXP (x, 0), 0), next_code),
7787 XEXP (XEXP (x, 0), 1)));
7790 /* If the constant is one less than a power of two, this might be
7791 representable by an extraction even if no shift is present.
7792 If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
7793 we are in a COMPARE. */
7794 else if ((i = exact_log2 (UINTVAL (XEXP (x, 1)) + 1)) >= 0)
7795 new_rtx = make_extraction (mode,
7796 make_compound_operation (XEXP (x, 0),
7797 next_code),
7798 0, NULL_RTX, i, 1, 0, in_code == COMPARE);
7800 /* If we are in a comparison and this is an AND with a power of two,
7801 convert this into the appropriate bit extract. */
7802 else if (in_code == COMPARE
7803 && (i = exact_log2 (UINTVAL (XEXP (x, 1)))) >= 0)
7804 new_rtx = make_extraction (mode,
7805 make_compound_operation (XEXP (x, 0),
7806 next_code),
7807 i, NULL_RTX, 1, 1, 0, 1);
7809 break;
7811 case LSHIFTRT:
7812 /* If the sign bit is known to be zero, replace this with an
7813 arithmetic shift. */
7814 if (have_insn_for (ASHIFTRT, mode)
7815 && ! have_insn_for (LSHIFTRT, mode)
7816 && mode_width <= HOST_BITS_PER_WIDE_INT
7817 && (nonzero_bits (XEXP (x, 0), mode) & (1 << (mode_width - 1))) == 0)
7819 new_rtx = gen_rtx_ASHIFTRT (mode,
7820 make_compound_operation (XEXP (x, 0),
7821 next_code),
7822 XEXP (x, 1));
7823 break;
7826 /* ... fall through ... */
7828 case ASHIFTRT:
7829 lhs = XEXP (x, 0);
7830 rhs = XEXP (x, 1);
7832 /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
7833 this is a SIGN_EXTRACT. */
7834 if (CONST_INT_P (rhs)
7835 && GET_CODE (lhs) == ASHIFT
7836 && CONST_INT_P (XEXP (lhs, 1))
7837 && INTVAL (rhs) >= INTVAL (XEXP (lhs, 1))
7838 && INTVAL (XEXP (lhs, 1)) >= 0
7839 && INTVAL (rhs) < mode_width)
7841 new_rtx = make_compound_operation (XEXP (lhs, 0), next_code);
7842 new_rtx = make_extraction (mode, new_rtx,
7843 INTVAL (rhs) - INTVAL (XEXP (lhs, 1)),
7844 NULL_RTX, mode_width - INTVAL (rhs),
7845 code == LSHIFTRT, 0, in_code == COMPARE);
7846 break;
7849 /* See if we have operations between an ASHIFTRT and an ASHIFT.
7850 If so, try to merge the shifts into a SIGN_EXTEND. We could
7851 also do this for some cases of SIGN_EXTRACT, but it doesn't
7852 seem worth the effort; the case checked for occurs on Alpha. */
7854 if (!OBJECT_P (lhs)
7855 && ! (GET_CODE (lhs) == SUBREG
7856 && (OBJECT_P (SUBREG_REG (lhs))))
7857 && CONST_INT_P (rhs)
7858 && INTVAL (rhs) < HOST_BITS_PER_WIDE_INT
7859 && INTVAL (rhs) < mode_width
7860 && (new_rtx = extract_left_shift (lhs, INTVAL (rhs))) != 0)
7861 new_rtx = make_extraction (mode, make_compound_operation (new_rtx, next_code),
7862 0, NULL_RTX, mode_width - INTVAL (rhs),
7863 code == LSHIFTRT, 0, in_code == COMPARE);
7865 break;
7867 case SUBREG:
7868 /* Call ourselves recursively on the inner expression. If we are
7869 narrowing the object and it has a different RTL code from
7870 what it originally did, do this SUBREG as a force_to_mode. */
7872 rtx inner = SUBREG_REG (x), simplified;
7874 tem = make_compound_operation (inner, in_code);
7876 simplified
7877 = simplify_subreg (mode, tem, GET_MODE (inner), SUBREG_BYTE (x));
7878 if (simplified)
7879 tem = simplified;
7881 if (GET_CODE (tem) != GET_CODE (inner)
7882 && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (inner))
7883 && subreg_lowpart_p (x))
7885 rtx newer
7886 = force_to_mode (tem, mode, ~(unsigned HOST_WIDE_INT) 0, 0);
7888 /* If we have something other than a SUBREG, we might have
7889 done an expansion, so rerun ourselves. */
7890 if (GET_CODE (newer) != SUBREG)
7891 newer = make_compound_operation (newer, in_code);
7893 /* force_to_mode can expand compounds. If it just re-expanded the
7894 compound, use gen_lowpart to convert to the desired mode. */
7895 if (rtx_equal_p (newer, x)
7896 /* Likewise if it re-expanded the compound only partially.
7897 This happens for SUBREG of ZERO_EXTRACT if they extract
7898 the same number of bits. */
7899 || (GET_CODE (newer) == SUBREG
7900 && (GET_CODE (SUBREG_REG (newer)) == LSHIFTRT
7901 || GET_CODE (SUBREG_REG (newer)) == ASHIFTRT)
7902 && GET_CODE (inner) == AND
7903 && rtx_equal_p (SUBREG_REG (newer), XEXP (inner, 0))))
7904 return gen_lowpart (GET_MODE (x), tem);
7906 return newer;
7909 if (simplified)
7910 return tem;
7912 break;
7914 default:
7915 break;
7918 if (new_rtx)
7920 x = gen_lowpart (mode, new_rtx);
7921 code = GET_CODE (x);
7924 /* Now recursively process each operand of this operation. We need to
7925 handle ZERO_EXTEND specially so that we don't lose track of the
7926 inner mode. */
7927 if (GET_CODE (x) == ZERO_EXTEND)
7929 new_rtx = make_compound_operation (XEXP (x, 0), next_code);
7930 tem = simplify_const_unary_operation (ZERO_EXTEND, GET_MODE (x),
7931 new_rtx, GET_MODE (XEXP (x, 0)));
7932 if (tem)
7933 return tem;
7934 SUBST (XEXP (x, 0), new_rtx);
7935 return x;
7938 fmt = GET_RTX_FORMAT (code);
7939 for (i = 0; i < GET_RTX_LENGTH (code); i++)
7940 if (fmt[i] == 'e')
7942 new_rtx = make_compound_operation (XEXP (x, i), next_code);
7943 SUBST (XEXP (x, i), new_rtx);
7945 else if (fmt[i] == 'E')
7946 for (j = 0; j < XVECLEN (x, i); j++)
7948 new_rtx = make_compound_operation (XVECEXP (x, i, j), next_code);
7949 SUBST (XVECEXP (x, i, j), new_rtx);
7952 maybe_swap:
7953 /* If this is a commutative operation, the changes to the operands
7954 may have made it noncanonical. */
7955 if (COMMUTATIVE_ARITH_P (x)
7956 && swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1)))
7958 tem = XEXP (x, 0);
7959 SUBST (XEXP (x, 0), XEXP (x, 1));
7960 SUBST (XEXP (x, 1), tem);
7963 return x;
7966 /* Given M see if it is a value that would select a field of bits
7967 within an item, but not the entire word. Return -1 if not.
7968 Otherwise, return the starting position of the field, where 0 is the
7969 low-order bit.
7971 *PLEN is set to the length of the field. */
7973 static int
7974 get_pos_from_mask (unsigned HOST_WIDE_INT m, unsigned HOST_WIDE_INT *plen)
7976 /* Get the bit number of the first 1 bit from the right, -1 if none. */
7977 int pos = m ? ctz_hwi (m) : -1;
7978 int len = 0;
7980 if (pos >= 0)
7981 /* Now shift off the low-order zero bits and see if we have a
7982 power of two minus 1. */
7983 len = exact_log2 ((m >> pos) + 1);
7985 if (len <= 0)
7986 pos = -1;
7988 *plen = len;
7989 return pos;
7992 /* If X refers to a register that equals REG in value, replace these
7993 references with REG. */
7994 static rtx
7995 canon_reg_for_combine (rtx x, rtx reg)
7997 rtx op0, op1, op2;
7998 const char *fmt;
7999 int i;
8000 bool copied;
8002 enum rtx_code code = GET_CODE (x);
8003 switch (GET_RTX_CLASS (code))
8005 case RTX_UNARY:
8006 op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8007 if (op0 != XEXP (x, 0))
8008 return simplify_gen_unary (GET_CODE (x), GET_MODE (x), op0,
8009 GET_MODE (reg));
8010 break;
8012 case RTX_BIN_ARITH:
8013 case RTX_COMM_ARITH:
8014 op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8015 op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8016 if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
8017 return simplify_gen_binary (GET_CODE (x), GET_MODE (x), op0, op1);
8018 break;
8020 case RTX_COMPARE:
8021 case RTX_COMM_COMPARE:
8022 op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8023 op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8024 if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
8025 return simplify_gen_relational (GET_CODE (x), GET_MODE (x),
8026 GET_MODE (op0), op0, op1);
8027 break;
8029 case RTX_TERNARY:
8030 case RTX_BITFIELD_OPS:
8031 op0 = canon_reg_for_combine (XEXP (x, 0), reg);
8032 op1 = canon_reg_for_combine (XEXP (x, 1), reg);
8033 op2 = canon_reg_for_combine (XEXP (x, 2), reg);
8034 if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1) || op2 != XEXP (x, 2))
8035 return simplify_gen_ternary (GET_CODE (x), GET_MODE (x),
8036 GET_MODE (op0), op0, op1, op2);
8038 case RTX_OBJ:
8039 if (REG_P (x))
8041 if (rtx_equal_p (get_last_value (reg), x)
8042 || rtx_equal_p (reg, get_last_value (x)))
8043 return reg;
8044 else
8045 break;
8048 /* fall through */
8050 default:
8051 fmt = GET_RTX_FORMAT (code);
8052 copied = false;
8053 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
8054 if (fmt[i] == 'e')
8056 rtx op = canon_reg_for_combine (XEXP (x, i), reg);
8057 if (op != XEXP (x, i))
8059 if (!copied)
8061 copied = true;
8062 x = copy_rtx (x);
8064 XEXP (x, i) = op;
8067 else if (fmt[i] == 'E')
8069 int j;
8070 for (j = 0; j < XVECLEN (x, i); j++)
8072 rtx op = canon_reg_for_combine (XVECEXP (x, i, j), reg);
8073 if (op != XVECEXP (x, i, j))
8075 if (!copied)
8077 copied = true;
8078 x = copy_rtx (x);
8080 XVECEXP (x, i, j) = op;
8085 break;
8088 return x;
8091 /* Return X converted to MODE. If the value is already truncated to
8092 MODE we can just return a subreg even though in the general case we
8093 would need an explicit truncation. */
8095 static rtx
8096 gen_lowpart_or_truncate (enum machine_mode mode, rtx x)
8098 if (!CONST_INT_P (x)
8099 && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (x))
8100 && !TRULY_NOOP_TRUNCATION_MODES_P (mode, GET_MODE (x))
8101 && !(REG_P (x) && reg_truncated_to_mode (mode, x)))
8103 /* Bit-cast X into an integer mode. */
8104 if (!SCALAR_INT_MODE_P (GET_MODE (x)))
8105 x = gen_lowpart (int_mode_for_mode (GET_MODE (x)), x);
8106 x = simplify_gen_unary (TRUNCATE, int_mode_for_mode (mode),
8107 x, GET_MODE (x));
8110 return gen_lowpart (mode, x);
8113 /* See if X can be simplified knowing that we will only refer to it in
8114 MODE and will only refer to those bits that are nonzero in MASK.
8115 If other bits are being computed or if masking operations are done
8116 that select a superset of the bits in MASK, they can sometimes be
8117 ignored.
8119 Return a possibly simplified expression, but always convert X to
8120 MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8122 If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
8123 are all off in X. This is used when X will be complemented, by either
8124 NOT, NEG, or XOR. */
8126 static rtx
8127 force_to_mode (rtx x, enum machine_mode mode, unsigned HOST_WIDE_INT mask,
8128 int just_select)
8130 enum rtx_code code = GET_CODE (x);
8131 int next_select = just_select || code == XOR || code == NOT || code == NEG;
8132 enum machine_mode op_mode;
8133 unsigned HOST_WIDE_INT fuller_mask, nonzero;
8134 rtx op0, op1, temp;
8136 /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8137 code below will do the wrong thing since the mode of such an
8138 expression is VOIDmode.
8140 Also do nothing if X is a CLOBBER; this can happen if X was
8141 the return value from a call to gen_lowpart. */
8142 if (code == CALL || code == ASM_OPERANDS || code == CLOBBER)
8143 return x;
8145 /* We want to perform the operation is its present mode unless we know
8146 that the operation is valid in MODE, in which case we do the operation
8147 in MODE. */
8148 op_mode = ((GET_MODE_CLASS (mode) == GET_MODE_CLASS (GET_MODE (x))
8149 && have_insn_for (code, mode))
8150 ? mode : GET_MODE (x));
8152 /* It is not valid to do a right-shift in a narrower mode
8153 than the one it came in with. */
8154 if ((code == LSHIFTRT || code == ASHIFTRT)
8155 && GET_MODE_PRECISION (mode) < GET_MODE_PRECISION (GET_MODE (x)))
8156 op_mode = GET_MODE (x);
8158 /* Truncate MASK to fit OP_MODE. */
8159 if (op_mode)
8160 mask &= GET_MODE_MASK (op_mode);
8162 /* When we have an arithmetic operation, or a shift whose count we
8163 do not know, we need to assume that all bits up to the highest-order
8164 bit in MASK will be needed. This is how we form such a mask. */
8165 if (mask & ((unsigned HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1)))
8166 fuller_mask = ~(unsigned HOST_WIDE_INT) 0;
8167 else
8168 fuller_mask = (((unsigned HOST_WIDE_INT) 1 << (floor_log2 (mask) + 1))
8169 - 1);
8171 /* Determine what bits of X are guaranteed to be (non)zero. */
8172 nonzero = nonzero_bits (x, mode);
8174 /* If none of the bits in X are needed, return a zero. */
8175 if (!just_select && (nonzero & mask) == 0 && !side_effects_p (x))
8176 x = const0_rtx;
8178 /* If X is a CONST_INT, return a new one. Do this here since the
8179 test below will fail. */
8180 if (CONST_INT_P (x))
8182 if (SCALAR_INT_MODE_P (mode))
8183 return gen_int_mode (INTVAL (x) & mask, mode);
8184 else
8186 x = GEN_INT (INTVAL (x) & mask);
8187 return gen_lowpart_common (mode, x);
8191 /* If X is narrower than MODE and we want all the bits in X's mode, just
8192 get X in the proper mode. */
8193 if (GET_MODE_SIZE (GET_MODE (x)) < GET_MODE_SIZE (mode)
8194 && (GET_MODE_MASK (GET_MODE (x)) & ~mask) == 0)
8195 return gen_lowpart (mode, x);
8197 /* We can ignore the effect of a SUBREG if it narrows the mode or
8198 if the constant masks to zero all the bits the mode doesn't have. */
8199 if (GET_CODE (x) == SUBREG
8200 && subreg_lowpart_p (x)
8201 && ((GET_MODE_SIZE (GET_MODE (x))
8202 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
8203 || (0 == (mask
8204 & GET_MODE_MASK (GET_MODE (x))
8205 & ~GET_MODE_MASK (GET_MODE (SUBREG_REG (x)))))))
8206 return force_to_mode (SUBREG_REG (x), mode, mask, next_select);
8208 /* The arithmetic simplifications here only work for scalar integer modes. */
8209 if (!SCALAR_INT_MODE_P (mode) || !SCALAR_INT_MODE_P (GET_MODE (x)))
8210 return gen_lowpart_or_truncate (mode, x);
8212 switch (code)
8214 case CLOBBER:
8215 /* If X is a (clobber (const_int)), return it since we know we are
8216 generating something that won't match. */
8217 return x;
8219 case SIGN_EXTEND:
8220 case ZERO_EXTEND:
8221 case ZERO_EXTRACT:
8222 case SIGN_EXTRACT:
8223 x = expand_compound_operation (x);
8224 if (GET_CODE (x) != code)
8225 return force_to_mode (x, mode, mask, next_select);
8226 break;
8228 case TRUNCATE:
8229 /* Similarly for a truncate. */
8230 return force_to_mode (XEXP (x, 0), mode, mask, next_select);
8232 case AND:
8233 /* If this is an AND with a constant, convert it into an AND
8234 whose constant is the AND of that constant with MASK. If it
8235 remains an AND of MASK, delete it since it is redundant. */
8237 if (CONST_INT_P (XEXP (x, 1)))
8239 x = simplify_and_const_int (x, op_mode, XEXP (x, 0),
8240 mask & INTVAL (XEXP (x, 1)));
8242 /* If X is still an AND, see if it is an AND with a mask that
8243 is just some low-order bits. If so, and it is MASK, we don't
8244 need it. */
8246 if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
8247 && ((INTVAL (XEXP (x, 1)) & GET_MODE_MASK (GET_MODE (x)))
8248 == mask))
8249 x = XEXP (x, 0);
8251 /* If it remains an AND, try making another AND with the bits
8252 in the mode mask that aren't in MASK turned on. If the
8253 constant in the AND is wide enough, this might make a
8254 cheaper constant. */
8256 if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
8257 && GET_MODE_MASK (GET_MODE (x)) != mask
8258 && HWI_COMPUTABLE_MODE_P (GET_MODE (x)))
8260 unsigned HOST_WIDE_INT cval
8261 = UINTVAL (XEXP (x, 1))
8262 | (GET_MODE_MASK (GET_MODE (x)) & ~mask);
8263 int width = GET_MODE_PRECISION (GET_MODE (x));
8264 rtx y;
8266 /* If MODE is narrower than HOST_WIDE_INT and CVAL is a negative
8267 number, sign extend it. */
8268 if (width > 0 && width < HOST_BITS_PER_WIDE_INT
8269 && (cval & ((unsigned HOST_WIDE_INT) 1 << (width - 1))) != 0)
8270 cval |= (unsigned HOST_WIDE_INT) -1 << width;
8272 y = simplify_gen_binary (AND, GET_MODE (x),
8273 XEXP (x, 0), GEN_INT (cval));
8274 if (set_src_cost (y, optimize_this_for_speed_p)
8275 < set_src_cost (x, optimize_this_for_speed_p))
8276 x = y;
8279 break;
8282 goto binop;
8284 case PLUS:
8285 /* In (and (plus FOO C1) M), if M is a mask that just turns off
8286 low-order bits (as in an alignment operation) and FOO is already
8287 aligned to that boundary, mask C1 to that boundary as well.
8288 This may eliminate that PLUS and, later, the AND. */
8291 unsigned int width = GET_MODE_PRECISION (mode);
8292 unsigned HOST_WIDE_INT smask = mask;
8294 /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8295 number, sign extend it. */
8297 if (width < HOST_BITS_PER_WIDE_INT
8298 && (smask & ((unsigned HOST_WIDE_INT) 1 << (width - 1))) != 0)
8299 smask |= (unsigned HOST_WIDE_INT) (-1) << width;
8301 if (CONST_INT_P (XEXP (x, 1))
8302 && exact_log2 (- smask) >= 0
8303 && (nonzero_bits (XEXP (x, 0), mode) & ~smask) == 0
8304 && (INTVAL (XEXP (x, 1)) & ~smask) != 0)
8305 return force_to_mode (plus_constant (XEXP (x, 0),
8306 (INTVAL (XEXP (x, 1)) & smask)),
8307 mode, smask, next_select);
8310 /* ... fall through ... */
8312 case MULT:
8313 /* For PLUS, MINUS and MULT, we need any bits less significant than the
8314 most significant bit in MASK since carries from those bits will
8315 affect the bits we are interested in. */
8316 mask = fuller_mask;
8317 goto binop;
8319 case MINUS:
8320 /* If X is (minus C Y) where C's least set bit is larger than any bit
8321 in the mask, then we may replace with (neg Y). */
8322 if (CONST_INT_P (XEXP (x, 0))
8323 && (((unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 0))
8324 & -INTVAL (XEXP (x, 0))))
8325 > mask))
8327 x = simplify_gen_unary (NEG, GET_MODE (x), XEXP (x, 1),
8328 GET_MODE (x));
8329 return force_to_mode (x, mode, mask, next_select);
8332 /* Similarly, if C contains every bit in the fuller_mask, then we may
8333 replace with (not Y). */
8334 if (CONST_INT_P (XEXP (x, 0))
8335 && ((UINTVAL (XEXP (x, 0)) | fuller_mask) == UINTVAL (XEXP (x, 0))))
8337 x = simplify_gen_unary (NOT, GET_MODE (x),
8338 XEXP (x, 1), GET_MODE (x));
8339 return force_to_mode (x, mode, mask, next_select);
8342 mask = fuller_mask;
8343 goto binop;
8345 case IOR:
8346 case XOR:
8347 /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8348 LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8349 operation which may be a bitfield extraction. Ensure that the
8350 constant we form is not wider than the mode of X. */
8352 if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
8353 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
8354 && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
8355 && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
8356 && CONST_INT_P (XEXP (x, 1))
8357 && ((INTVAL (XEXP (XEXP (x, 0), 1))
8358 + floor_log2 (INTVAL (XEXP (x, 1))))
8359 < GET_MODE_PRECISION (GET_MODE (x)))
8360 && (UINTVAL (XEXP (x, 1))
8361 & ~nonzero_bits (XEXP (x, 0), GET_MODE (x))) == 0)
8363 temp = GEN_INT ((INTVAL (XEXP (x, 1)) & mask)
8364 << INTVAL (XEXP (XEXP (x, 0), 1)));
8365 temp = simplify_gen_binary (GET_CODE (x), GET_MODE (x),
8366 XEXP (XEXP (x, 0), 0), temp);
8367 x = simplify_gen_binary (LSHIFTRT, GET_MODE (x), temp,
8368 XEXP (XEXP (x, 0), 1));
8369 return force_to_mode (x, mode, mask, next_select);
8372 binop:
8373 /* For most binary operations, just propagate into the operation and
8374 change the mode if we have an operation of that mode. */
8376 op0 = force_to_mode (XEXP (x, 0), mode, mask, next_select);
8377 op1 = force_to_mode (XEXP (x, 1), mode, mask, next_select);
8379 /* If we ended up truncating both operands, truncate the result of the
8380 operation instead. */
8381 if (GET_CODE (op0) == TRUNCATE
8382 && GET_CODE (op1) == TRUNCATE)
8384 op0 = XEXP (op0, 0);
8385 op1 = XEXP (op1, 0);
8388 op0 = gen_lowpart_or_truncate (op_mode, op0);
8389 op1 = gen_lowpart_or_truncate (op_mode, op1);
8391 if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
8392 x = simplify_gen_binary (code, op_mode, op0, op1);
8393 break;
8395 case ASHIFT:
8396 /* For left shifts, do the same, but just for the first operand.
8397 However, we cannot do anything with shifts where we cannot
8398 guarantee that the counts are smaller than the size of the mode
8399 because such a count will have a different meaning in a
8400 wider mode. */
8402 if (! (CONST_INT_P (XEXP (x, 1))
8403 && INTVAL (XEXP (x, 1)) >= 0
8404 && INTVAL (XEXP (x, 1)) < GET_MODE_PRECISION (mode))
8405 && ! (GET_MODE (XEXP (x, 1)) != VOIDmode
8406 && (nonzero_bits (XEXP (x, 1), GET_MODE (XEXP (x, 1)))
8407 < (unsigned HOST_WIDE_INT) GET_MODE_PRECISION (mode))))
8408 break;
8410 /* If the shift count is a constant and we can do arithmetic in
8411 the mode of the shift, refine which bits we need. Otherwise, use the
8412 conservative form of the mask. */
8413 if (CONST_INT_P (XEXP (x, 1))
8414 && INTVAL (XEXP (x, 1)) >= 0
8415 && INTVAL (XEXP (x, 1)) < GET_MODE_PRECISION (op_mode)
8416 && HWI_COMPUTABLE_MODE_P (op_mode))
8417 mask >>= INTVAL (XEXP (x, 1));
8418 else
8419 mask = fuller_mask;
8421 op0 = gen_lowpart_or_truncate (op_mode,
8422 force_to_mode (XEXP (x, 0), op_mode,
8423 mask, next_select));
8425 if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0))
8426 x = simplify_gen_binary (code, op_mode, op0, XEXP (x, 1));
8427 break;
8429 case LSHIFTRT:
8430 /* Here we can only do something if the shift count is a constant,
8431 this shift constant is valid for the host, and we can do arithmetic
8432 in OP_MODE. */
8434 if (CONST_INT_P (XEXP (x, 1))
8435 && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
8436 && HWI_COMPUTABLE_MODE_P (op_mode))
8438 rtx inner = XEXP (x, 0);
8439 unsigned HOST_WIDE_INT inner_mask;
8441 /* Select the mask of the bits we need for the shift operand. */
8442 inner_mask = mask << INTVAL (XEXP (x, 1));
8444 /* We can only change the mode of the shift if we can do arithmetic
8445 in the mode of the shift and INNER_MASK is no wider than the
8446 width of X's mode. */
8447 if ((inner_mask & ~GET_MODE_MASK (GET_MODE (x))) != 0)
8448 op_mode = GET_MODE (x);
8450 inner = force_to_mode (inner, op_mode, inner_mask, next_select);
8452 if (GET_MODE (x) != op_mode || inner != XEXP (x, 0))
8453 x = simplify_gen_binary (LSHIFTRT, op_mode, inner, XEXP (x, 1));
8456 /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
8457 shift and AND produces only copies of the sign bit (C2 is one less
8458 than a power of two), we can do this with just a shift. */
8460 if (GET_CODE (x) == LSHIFTRT
8461 && CONST_INT_P (XEXP (x, 1))
8462 /* The shift puts one of the sign bit copies in the least significant
8463 bit. */
8464 && ((INTVAL (XEXP (x, 1))
8465 + num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0))))
8466 >= GET_MODE_PRECISION (GET_MODE (x)))
8467 && exact_log2 (mask + 1) >= 0
8468 /* Number of bits left after the shift must be more than the mask
8469 needs. */
8470 && ((INTVAL (XEXP (x, 1)) + exact_log2 (mask + 1))
8471 <= GET_MODE_PRECISION (GET_MODE (x)))
8472 /* Must be more sign bit copies than the mask needs. */
8473 && ((int) num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0)))
8474 >= exact_log2 (mask + 1)))
8475 x = simplify_gen_binary (LSHIFTRT, GET_MODE (x), XEXP (x, 0),
8476 GEN_INT (GET_MODE_PRECISION (GET_MODE (x))
8477 - exact_log2 (mask + 1)));
8479 goto shiftrt;
8481 case ASHIFTRT:
8482 /* If we are just looking for the sign bit, we don't need this shift at
8483 all, even if it has a variable count. */
8484 if (val_signbit_p (GET_MODE (x), mask))
8485 return force_to_mode (XEXP (x, 0), mode, mask, next_select);
8487 /* If this is a shift by a constant, get a mask that contains those bits
8488 that are not copies of the sign bit. We then have two cases: If
8489 MASK only includes those bits, this can be a logical shift, which may
8490 allow simplifications. If MASK is a single-bit field not within
8491 those bits, we are requesting a copy of the sign bit and hence can
8492 shift the sign bit to the appropriate location. */
8494 if (CONST_INT_P (XEXP (x, 1)) && INTVAL (XEXP (x, 1)) >= 0
8495 && INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT)
8497 int i;
8499 /* If the considered data is wider than HOST_WIDE_INT, we can't
8500 represent a mask for all its bits in a single scalar.
8501 But we only care about the lower bits, so calculate these. */
8503 if (GET_MODE_PRECISION (GET_MODE (x)) > HOST_BITS_PER_WIDE_INT)
8505 nonzero = ~(unsigned HOST_WIDE_INT) 0;
8507 /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
8508 is the number of bits a full-width mask would have set.
8509 We need only shift if these are fewer than nonzero can
8510 hold. If not, we must keep all bits set in nonzero. */
8512 if (GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
8513 < HOST_BITS_PER_WIDE_INT)
8514 nonzero >>= INTVAL (XEXP (x, 1))
8515 + HOST_BITS_PER_WIDE_INT
8516 - GET_MODE_PRECISION (GET_MODE (x)) ;
8518 else
8520 nonzero = GET_MODE_MASK (GET_MODE (x));
8521 nonzero >>= INTVAL (XEXP (x, 1));
8524 if ((mask & ~nonzero) == 0)
8526 x = simplify_shift_const (NULL_RTX, LSHIFTRT, GET_MODE (x),
8527 XEXP (x, 0), INTVAL (XEXP (x, 1)));
8528 if (GET_CODE (x) != ASHIFTRT)
8529 return force_to_mode (x, mode, mask, next_select);
8532 else if ((i = exact_log2 (mask)) >= 0)
8534 x = simplify_shift_const
8535 (NULL_RTX, LSHIFTRT, GET_MODE (x), XEXP (x, 0),
8536 GET_MODE_PRECISION (GET_MODE (x)) - 1 - i);
8538 if (GET_CODE (x) != ASHIFTRT)
8539 return force_to_mode (x, mode, mask, next_select);
8543 /* If MASK is 1, convert this to an LSHIFTRT. This can be done
8544 even if the shift count isn't a constant. */
8545 if (mask == 1)
8546 x = simplify_gen_binary (LSHIFTRT, GET_MODE (x),
8547 XEXP (x, 0), XEXP (x, 1));
8549 shiftrt:
8551 /* If this is a zero- or sign-extension operation that just affects bits
8552 we don't care about, remove it. Be sure the call above returned
8553 something that is still a shift. */
8555 if ((GET_CODE (x) == LSHIFTRT || GET_CODE (x) == ASHIFTRT)
8556 && CONST_INT_P (XEXP (x, 1))
8557 && INTVAL (XEXP (x, 1)) >= 0
8558 && (INTVAL (XEXP (x, 1))
8559 <= GET_MODE_PRECISION (GET_MODE (x)) - (floor_log2 (mask) + 1))
8560 && GET_CODE (XEXP (x, 0)) == ASHIFT
8561 && XEXP (XEXP (x, 0), 1) == XEXP (x, 1))
8562 return force_to_mode (XEXP (XEXP (x, 0), 0), mode, mask,
8563 next_select);
8565 break;
8567 case ROTATE:
8568 case ROTATERT:
8569 /* If the shift count is constant and we can do computations
8570 in the mode of X, compute where the bits we care about are.
8571 Otherwise, we can't do anything. Don't change the mode of
8572 the shift or propagate MODE into the shift, though. */
8573 if (CONST_INT_P (XEXP (x, 1))
8574 && INTVAL (XEXP (x, 1)) >= 0)
8576 temp = simplify_binary_operation (code == ROTATE ? ROTATERT : ROTATE,
8577 GET_MODE (x), GEN_INT (mask),
8578 XEXP (x, 1));
8579 if (temp && CONST_INT_P (temp))
8580 SUBST (XEXP (x, 0),
8581 force_to_mode (XEXP (x, 0), GET_MODE (x),
8582 INTVAL (temp), next_select));
8584 break;
8586 case NEG:
8587 /* If we just want the low-order bit, the NEG isn't needed since it
8588 won't change the low-order bit. */
8589 if (mask == 1)
8590 return force_to_mode (XEXP (x, 0), mode, mask, just_select);
8592 /* We need any bits less significant than the most significant bit in
8593 MASK since carries from those bits will affect the bits we are
8594 interested in. */
8595 mask = fuller_mask;
8596 goto unop;
8598 case NOT:
8599 /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
8600 same as the XOR case above. Ensure that the constant we form is not
8601 wider than the mode of X. */
8603 if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
8604 && CONST_INT_P (XEXP (XEXP (x, 0), 1))
8605 && INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
8606 && (INTVAL (XEXP (XEXP (x, 0), 1)) + floor_log2 (mask)
8607 < GET_MODE_PRECISION (GET_MODE (x)))
8608 && INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT)
8610 temp = gen_int_mode (mask << INTVAL (XEXP (XEXP (x, 0), 1)),
8611 GET_MODE (x));
8612 temp = simplify_gen_binary (XOR, GET_MODE (x),
8613 XEXP (XEXP (x, 0), 0), temp);
8614 x = simplify_gen_binary (LSHIFTRT, GET_MODE (x),
8615 temp, XEXP (XEXP (x, 0), 1));
8617 return force_to_mode (x, mode, mask, next_select);
8620 /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
8621 use the full mask inside the NOT. */
8622 mask = fuller_mask;
8624 unop:
8625 op0 = gen_lowpart_or_truncate (op_mode,
8626 force_to_mode (XEXP (x, 0), mode, mask,
8627 next_select));
8628 if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0))
8629 x = simplify_gen_unary (code, op_mode, op0, op_mode);
8630 break;
8632 case NE:
8633 /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
8634 in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
8635 which is equal to STORE_FLAG_VALUE. */
8636 if ((mask & ~STORE_FLAG_VALUE) == 0
8637 && XEXP (x, 1) == const0_rtx
8638 && GET_MODE (XEXP (x, 0)) == mode
8639 && exact_log2 (nonzero_bits (XEXP (x, 0), mode)) >= 0
8640 && (nonzero_bits (XEXP (x, 0), mode)
8641 == (unsigned HOST_WIDE_INT) STORE_FLAG_VALUE))
8642 return force_to_mode (XEXP (x, 0), mode, mask, next_select);
8644 break;
8646 case IF_THEN_ELSE:
8647 /* We have no way of knowing if the IF_THEN_ELSE can itself be
8648 written in a narrower mode. We play it safe and do not do so. */
8650 SUBST (XEXP (x, 1),
8651 gen_lowpart_or_truncate (GET_MODE (x),
8652 force_to_mode (XEXP (x, 1), mode,
8653 mask, next_select)));
8654 SUBST (XEXP (x, 2),
8655 gen_lowpart_or_truncate (GET_MODE (x),
8656 force_to_mode (XEXP (x, 2), mode,
8657 mask, next_select)));
8658 break;
8660 default:
8661 break;
8664 /* Ensure we return a value of the proper mode. */
8665 return gen_lowpart_or_truncate (mode, x);
8668 /* Return nonzero if X is an expression that has one of two values depending on
8669 whether some other value is zero or nonzero. In that case, we return the
8670 value that is being tested, *PTRUE is set to the value if the rtx being
8671 returned has a nonzero value, and *PFALSE is set to the other alternative.
8673 If we return zero, we set *PTRUE and *PFALSE to X. */
8675 static rtx
8676 if_then_else_cond (rtx x, rtx *ptrue, rtx *pfalse)
8678 enum machine_mode mode = GET_MODE (x);
8679 enum rtx_code code = GET_CODE (x);
8680 rtx cond0, cond1, true0, true1, false0, false1;
8681 unsigned HOST_WIDE_INT nz;
8683 /* If we are comparing a value against zero, we are done. */
8684 if ((code == NE || code == EQ)
8685 && XEXP (x, 1) == const0_rtx)
8687 *ptrue = (code == NE) ? const_true_rtx : const0_rtx;
8688 *pfalse = (code == NE) ? const0_rtx : const_true_rtx;
8689 return XEXP (x, 0);
8692 /* If this is a unary operation whose operand has one of two values, apply
8693 our opcode to compute those values. */
8694 else if (UNARY_P (x)
8695 && (cond0 = if_then_else_cond (XEXP (x, 0), &true0, &false0)) != 0)
8697 *ptrue = simplify_gen_unary (code, mode, true0, GET_MODE (XEXP (x, 0)));
8698 *pfalse = simplify_gen_unary (code, mode, false0,
8699 GET_MODE (XEXP (x, 0)));
8700 return cond0;
8703 /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
8704 make can't possibly match and would suppress other optimizations. */
8705 else if (code == COMPARE)
8708 /* If this is a binary operation, see if either side has only one of two
8709 values. If either one does or if both do and they are conditional on
8710 the same value, compute the new true and false values. */
8711 else if (BINARY_P (x))
8713 cond0 = if_then_else_cond (XEXP (x, 0), &true0, &false0);
8714 cond1 = if_then_else_cond (XEXP (x, 1), &true1, &false1);
8716 if ((cond0 != 0 || cond1 != 0)
8717 && ! (cond0 != 0 && cond1 != 0 && ! rtx_equal_p (cond0, cond1)))
8719 /* If if_then_else_cond returned zero, then true/false are the
8720 same rtl. We must copy one of them to prevent invalid rtl
8721 sharing. */
8722 if (cond0 == 0)
8723 true0 = copy_rtx (true0);
8724 else if (cond1 == 0)
8725 true1 = copy_rtx (true1);
8727 if (COMPARISON_P (x))
8729 *ptrue = simplify_gen_relational (code, mode, VOIDmode,
8730 true0, true1);
8731 *pfalse = simplify_gen_relational (code, mode, VOIDmode,
8732 false0, false1);
8734 else
8736 *ptrue = simplify_gen_binary (code, mode, true0, true1);
8737 *pfalse = simplify_gen_binary (code, mode, false0, false1);
8740 return cond0 ? cond0 : cond1;
8743 /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
8744 operands is zero when the other is nonzero, and vice-versa,
8745 and STORE_FLAG_VALUE is 1 or -1. */
8747 if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
8748 && (code == PLUS || code == IOR || code == XOR || code == MINUS
8749 || code == UMAX)
8750 && GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
8752 rtx op0 = XEXP (XEXP (x, 0), 1);
8753 rtx op1 = XEXP (XEXP (x, 1), 1);
8755 cond0 = XEXP (XEXP (x, 0), 0);
8756 cond1 = XEXP (XEXP (x, 1), 0);
8758 if (COMPARISON_P (cond0)
8759 && COMPARISON_P (cond1)
8760 && ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
8761 && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
8762 && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
8763 || ((swap_condition (GET_CODE (cond0))
8764 == reversed_comparison_code (cond1, NULL))
8765 && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
8766 && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
8767 && ! side_effects_p (x))
8769 *ptrue = simplify_gen_binary (MULT, mode, op0, const_true_rtx);
8770 *pfalse = simplify_gen_binary (MULT, mode,
8771 (code == MINUS
8772 ? simplify_gen_unary (NEG, mode,
8773 op1, mode)
8774 : op1),
8775 const_true_rtx);
8776 return cond0;
8780 /* Similarly for MULT, AND and UMIN, except that for these the result
8781 is always zero. */
8782 if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
8783 && (code == MULT || code == AND || code == UMIN)
8784 && GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
8786 cond0 = XEXP (XEXP (x, 0), 0);
8787 cond1 = XEXP (XEXP (x, 1), 0);
8789 if (COMPARISON_P (cond0)
8790 && COMPARISON_P (cond1)
8791 && ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
8792 && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
8793 && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
8794 || ((swap_condition (GET_CODE (cond0))
8795 == reversed_comparison_code (cond1, NULL))
8796 && rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
8797 && rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
8798 && ! side_effects_p (x))
8800 *ptrue = *pfalse = const0_rtx;
8801 return cond0;
8806 else if (code == IF_THEN_ELSE)
8808 /* If we have IF_THEN_ELSE already, extract the condition and
8809 canonicalize it if it is NE or EQ. */
8810 cond0 = XEXP (x, 0);
8811 *ptrue = XEXP (x, 1), *pfalse = XEXP (x, 2);
8812 if (GET_CODE (cond0) == NE && XEXP (cond0, 1) == const0_rtx)
8813 return XEXP (cond0, 0);
8814 else if (GET_CODE (cond0) == EQ && XEXP (cond0, 1) == const0_rtx)
8816 *ptrue = XEXP (x, 2), *pfalse = XEXP (x, 1);
8817 return XEXP (cond0, 0);
8819 else
8820 return cond0;
8823 /* If X is a SUBREG, we can narrow both the true and false values
8824 if the inner expression, if there is a condition. */
8825 else if (code == SUBREG
8826 && 0 != (cond0 = if_then_else_cond (SUBREG_REG (x),
8827 &true0, &false0)))
8829 true0 = simplify_gen_subreg (mode, true0,
8830 GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
8831 false0 = simplify_gen_subreg (mode, false0,
8832 GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
8833 if (true0 && false0)
8835 *ptrue = true0;
8836 *pfalse = false0;
8837 return cond0;
8841 /* If X is a constant, this isn't special and will cause confusions
8842 if we treat it as such. Likewise if it is equivalent to a constant. */
8843 else if (CONSTANT_P (x)
8844 || ((cond0 = get_last_value (x)) != 0 && CONSTANT_P (cond0)))
8847 /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
8848 will be least confusing to the rest of the compiler. */
8849 else if (mode == BImode)
8851 *ptrue = GEN_INT (STORE_FLAG_VALUE), *pfalse = const0_rtx;
8852 return x;
8855 /* If X is known to be either 0 or -1, those are the true and
8856 false values when testing X. */
8857 else if (x == constm1_rtx || x == const0_rtx
8858 || (mode != VOIDmode
8859 && num_sign_bit_copies (x, mode) == GET_MODE_PRECISION (mode)))
8861 *ptrue = constm1_rtx, *pfalse = const0_rtx;
8862 return x;
8865 /* Likewise for 0 or a single bit. */
8866 else if (HWI_COMPUTABLE_MODE_P (mode)
8867 && exact_log2 (nz = nonzero_bits (x, mode)) >= 0)
8869 *ptrue = gen_int_mode (nz, mode), *pfalse = const0_rtx;
8870 return x;
8873 /* Otherwise fail; show no condition with true and false values the same. */
8874 *ptrue = *pfalse = x;
8875 return 0;
8878 /* Return the value of expression X given the fact that condition COND
8879 is known to be true when applied to REG as its first operand and VAL
8880 as its second. X is known to not be shared and so can be modified in
8881 place.
8883 We only handle the simplest cases, and specifically those cases that
8884 arise with IF_THEN_ELSE expressions. */
8886 static rtx
8887 known_cond (rtx x, enum rtx_code cond, rtx reg, rtx val)
8889 enum rtx_code code = GET_CODE (x);
8890 rtx temp;
8891 const char *fmt;
8892 int i, j;
8894 if (side_effects_p (x))
8895 return x;
8897 /* If either operand of the condition is a floating point value,
8898 then we have to avoid collapsing an EQ comparison. */
8899 if (cond == EQ
8900 && rtx_equal_p (x, reg)
8901 && ! FLOAT_MODE_P (GET_MODE (x))
8902 && ! FLOAT_MODE_P (GET_MODE (val)))
8903 return val;
8905 if (cond == UNEQ && rtx_equal_p (x, reg))
8906 return val;
8908 /* If X is (abs REG) and we know something about REG's relationship
8909 with zero, we may be able to simplify this. */
8911 if (code == ABS && rtx_equal_p (XEXP (x, 0), reg) && val == const0_rtx)
8912 switch (cond)
8914 case GE: case GT: case EQ:
8915 return XEXP (x, 0);
8916 case LT: case LE:
8917 return simplify_gen_unary (NEG, GET_MODE (XEXP (x, 0)),
8918 XEXP (x, 0),
8919 GET_MODE (XEXP (x, 0)));
8920 default:
8921 break;
8924 /* The only other cases we handle are MIN, MAX, and comparisons if the
8925 operands are the same as REG and VAL. */
8927 else if (COMPARISON_P (x) || COMMUTATIVE_ARITH_P (x))
8929 if (rtx_equal_p (XEXP (x, 0), val))
8930 cond = swap_condition (cond), temp = val, val = reg, reg = temp;
8932 if (rtx_equal_p (XEXP (x, 0), reg) && rtx_equal_p (XEXP (x, 1), val))
8934 if (COMPARISON_P (x))
8936 if (comparison_dominates_p (cond, code))
8937 return const_true_rtx;
8939 code = reversed_comparison_code (x, NULL);
8940 if (code != UNKNOWN
8941 && comparison_dominates_p (cond, code))
8942 return const0_rtx;
8943 else
8944 return x;
8946 else if (code == SMAX || code == SMIN
8947 || code == UMIN || code == UMAX)
8949 int unsignedp = (code == UMIN || code == UMAX);
8951 /* Do not reverse the condition when it is NE or EQ.
8952 This is because we cannot conclude anything about
8953 the value of 'SMAX (x, y)' when x is not equal to y,
8954 but we can when x equals y. */
8955 if ((code == SMAX || code == UMAX)
8956 && ! (cond == EQ || cond == NE))
8957 cond = reverse_condition (cond);
8959 switch (cond)
8961 case GE: case GT:
8962 return unsignedp ? x : XEXP (x, 1);
8963 case LE: case LT:
8964 return unsignedp ? x : XEXP (x, 0);
8965 case GEU: case GTU:
8966 return unsignedp ? XEXP (x, 1) : x;
8967 case LEU: case LTU:
8968 return unsignedp ? XEXP (x, 0) : x;
8969 default:
8970 break;
8975 else if (code == SUBREG)
8977 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (x));
8978 rtx new_rtx, r = known_cond (SUBREG_REG (x), cond, reg, val);
8980 if (SUBREG_REG (x) != r)
8982 /* We must simplify subreg here, before we lose track of the
8983 original inner_mode. */
8984 new_rtx = simplify_subreg (GET_MODE (x), r,
8985 inner_mode, SUBREG_BYTE (x));
8986 if (new_rtx)
8987 return new_rtx;
8988 else
8989 SUBST (SUBREG_REG (x), r);
8992 return x;
8994 /* We don't have to handle SIGN_EXTEND here, because even in the
8995 case of replacing something with a modeless CONST_INT, a
8996 CONST_INT is already (supposed to be) a valid sign extension for
8997 its narrower mode, which implies it's already properly
8998 sign-extended for the wider mode. Now, for ZERO_EXTEND, the
8999 story is different. */
9000 else if (code == ZERO_EXTEND)
9002 enum machine_mode inner_mode = GET_MODE (XEXP (x, 0));
9003 rtx new_rtx, r = known_cond (XEXP (x, 0), cond, reg, val);
9005 if (XEXP (x, 0) != r)
9007 /* We must simplify the zero_extend here, before we lose
9008 track of the original inner_mode. */
9009 new_rtx = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
9010 r, inner_mode);
9011 if (new_rtx)
9012 return new_rtx;
9013 else
9014 SUBST (XEXP (x, 0), r);
9017 return x;
9020 fmt = GET_RTX_FORMAT (code);
9021 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
9023 if (fmt[i] == 'e')
9024 SUBST (XEXP (x, i), known_cond (XEXP (x, i), cond, reg, val));
9025 else if (fmt[i] == 'E')
9026 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
9027 SUBST (XVECEXP (x, i, j), known_cond (XVECEXP (x, i, j),
9028 cond, reg, val));
9031 return x;
9034 /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9035 assignment as a field assignment. */
9037 static int
9038 rtx_equal_for_field_assignment_p (rtx x, rtx y)
9040 if (x == y || rtx_equal_p (x, y))
9041 return 1;
9043 if (x == 0 || y == 0 || GET_MODE (x) != GET_MODE (y))
9044 return 0;
9046 /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9047 Note that all SUBREGs of MEM are paradoxical; otherwise they
9048 would have been rewritten. */
9049 if (MEM_P (x) && GET_CODE (y) == SUBREG
9050 && MEM_P (SUBREG_REG (y))
9051 && rtx_equal_p (SUBREG_REG (y),
9052 gen_lowpart (GET_MODE (SUBREG_REG (y)), x)))
9053 return 1;
9055 if (MEM_P (y) && GET_CODE (x) == SUBREG
9056 && MEM_P (SUBREG_REG (x))
9057 && rtx_equal_p (SUBREG_REG (x),
9058 gen_lowpart (GET_MODE (SUBREG_REG (x)), y)))
9059 return 1;
9061 /* We used to see if get_last_value of X and Y were the same but that's
9062 not correct. In one direction, we'll cause the assignment to have
9063 the wrong destination and in the case, we'll import a register into this
9064 insn that might have already have been dead. So fail if none of the
9065 above cases are true. */
9066 return 0;
9069 /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9070 Return that assignment if so.
9072 We only handle the most common cases. */
9074 static rtx
9075 make_field_assignment (rtx x)
9077 rtx dest = SET_DEST (x);
9078 rtx src = SET_SRC (x);
9079 rtx assign;
9080 rtx rhs, lhs;
9081 HOST_WIDE_INT c1;
9082 HOST_WIDE_INT pos;
9083 unsigned HOST_WIDE_INT len;
9084 rtx other;
9085 enum machine_mode mode;
9087 /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9088 a clear of a one-bit field. We will have changed it to
9089 (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9090 for a SUBREG. */
9092 if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == ROTATE
9093 && CONST_INT_P (XEXP (XEXP (src, 0), 0))
9094 && INTVAL (XEXP (XEXP (src, 0), 0)) == -2
9095 && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9097 assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
9098 1, 1, 1, 0);
9099 if (assign != 0)
9100 return gen_rtx_SET (VOIDmode, assign, const0_rtx);
9101 return x;
9104 if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == SUBREG
9105 && subreg_lowpart_p (XEXP (src, 0))
9106 && (GET_MODE_SIZE (GET_MODE (XEXP (src, 0)))
9107 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (XEXP (src, 0)))))
9108 && GET_CODE (SUBREG_REG (XEXP (src, 0))) == ROTATE
9109 && CONST_INT_P (XEXP (SUBREG_REG (XEXP (src, 0)), 0))
9110 && INTVAL (XEXP (SUBREG_REG (XEXP (src, 0)), 0)) == -2
9111 && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9113 assign = make_extraction (VOIDmode, dest, 0,
9114 XEXP (SUBREG_REG (XEXP (src, 0)), 1),
9115 1, 1, 1, 0);
9116 if (assign != 0)
9117 return gen_rtx_SET (VOIDmode, assign, const0_rtx);
9118 return x;
9121 /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9122 one-bit field. */
9123 if (GET_CODE (src) == IOR && GET_CODE (XEXP (src, 0)) == ASHIFT
9124 && XEXP (XEXP (src, 0), 0) == const1_rtx
9125 && rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
9127 assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
9128 1, 1, 1, 0);
9129 if (assign != 0)
9130 return gen_rtx_SET (VOIDmode, assign, const1_rtx);
9131 return x;
9134 /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9135 SRC is an AND with all bits of that field set, then we can discard
9136 the AND. */
9137 if (GET_CODE (dest) == ZERO_EXTRACT
9138 && CONST_INT_P (XEXP (dest, 1))
9139 && GET_CODE (src) == AND
9140 && CONST_INT_P (XEXP (src, 1)))
9142 HOST_WIDE_INT width = INTVAL (XEXP (dest, 1));
9143 unsigned HOST_WIDE_INT and_mask = INTVAL (XEXP (src, 1));
9144 unsigned HOST_WIDE_INT ze_mask;
9146 if (width >= HOST_BITS_PER_WIDE_INT)
9147 ze_mask = -1;
9148 else
9149 ze_mask = ((unsigned HOST_WIDE_INT)1 << width) - 1;
9151 /* Complete overlap. We can remove the source AND. */
9152 if ((and_mask & ze_mask) == ze_mask)
9153 return gen_rtx_SET (VOIDmode, dest, XEXP (src, 0));
9155 /* Partial overlap. We can reduce the source AND. */
9156 if ((and_mask & ze_mask) != and_mask)
9158 mode = GET_MODE (src);
9159 src = gen_rtx_AND (mode, XEXP (src, 0),
9160 gen_int_mode (and_mask & ze_mask, mode));
9161 return gen_rtx_SET (VOIDmode, dest, src);
9165 /* The other case we handle is assignments into a constant-position
9166 field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9167 a mask that has all one bits except for a group of zero bits and
9168 OTHER is known to have zeros where C1 has ones, this is such an
9169 assignment. Compute the position and length from C1. Shift OTHER
9170 to the appropriate position, force it to the required mode, and
9171 make the extraction. Check for the AND in both operands. */
9173 if (GET_CODE (src) != IOR && GET_CODE (src) != XOR)
9174 return x;
9176 rhs = expand_compound_operation (XEXP (src, 0));
9177 lhs = expand_compound_operation (XEXP (src, 1));
9179 if (GET_CODE (rhs) == AND
9180 && CONST_INT_P (XEXP (rhs, 1))
9181 && rtx_equal_for_field_assignment_p (XEXP (rhs, 0), dest))
9182 c1 = INTVAL (XEXP (rhs, 1)), other = lhs;
9183 else if (GET_CODE (lhs) == AND
9184 && CONST_INT_P (XEXP (lhs, 1))
9185 && rtx_equal_for_field_assignment_p (XEXP (lhs, 0), dest))
9186 c1 = INTVAL (XEXP (lhs, 1)), other = rhs;
9187 else
9188 return x;
9190 pos = get_pos_from_mask ((~c1) & GET_MODE_MASK (GET_MODE (dest)), &len);
9191 if (pos < 0 || pos + len > GET_MODE_PRECISION (GET_MODE (dest))
9192 || GET_MODE_PRECISION (GET_MODE (dest)) > HOST_BITS_PER_WIDE_INT
9193 || (c1 & nonzero_bits (other, GET_MODE (dest))) != 0)
9194 return x;
9196 assign = make_extraction (VOIDmode, dest, pos, NULL_RTX, len, 1, 1, 0);
9197 if (assign == 0)
9198 return x;
9200 /* The mode to use for the source is the mode of the assignment, or of
9201 what is inside a possible STRICT_LOW_PART. */
9202 mode = (GET_CODE (assign) == STRICT_LOW_PART
9203 ? GET_MODE (XEXP (assign, 0)) : GET_MODE (assign));
9205 /* Shift OTHER right POS places and make it the source, restricting it
9206 to the proper length and mode. */
9208 src = canon_reg_for_combine (simplify_shift_const (NULL_RTX, LSHIFTRT,
9209 GET_MODE (src),
9210 other, pos),
9211 dest);
9212 src = force_to_mode (src, mode,
9213 GET_MODE_PRECISION (mode) >= HOST_BITS_PER_WIDE_INT
9214 ? ~(unsigned HOST_WIDE_INT) 0
9215 : ((unsigned HOST_WIDE_INT) 1 << len) - 1,
9218 /* If SRC is masked by an AND that does not make a difference in
9219 the value being stored, strip it. */
9220 if (GET_CODE (assign) == ZERO_EXTRACT
9221 && CONST_INT_P (XEXP (assign, 1))
9222 && INTVAL (XEXP (assign, 1)) < HOST_BITS_PER_WIDE_INT
9223 && GET_CODE (src) == AND
9224 && CONST_INT_P (XEXP (src, 1))
9225 && UINTVAL (XEXP (src, 1))
9226 == ((unsigned HOST_WIDE_INT) 1 << INTVAL (XEXP (assign, 1))) - 1)
9227 src = XEXP (src, 0);
9229 return gen_rtx_SET (VOIDmode, assign, src);
9232 /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9233 if so. */
9235 static rtx
9236 apply_distributive_law (rtx x)
9238 enum rtx_code code = GET_CODE (x);
9239 enum rtx_code inner_code;
9240 rtx lhs, rhs, other;
9241 rtx tem;
9243 /* Distributivity is not true for floating point as it can change the
9244 value. So we don't do it unless -funsafe-math-optimizations. */
9245 if (FLOAT_MODE_P (GET_MODE (x))
9246 && ! flag_unsafe_math_optimizations)
9247 return x;
9249 /* The outer operation can only be one of the following: */
9250 if (code != IOR && code != AND && code != XOR
9251 && code != PLUS && code != MINUS)
9252 return x;
9254 lhs = XEXP (x, 0);
9255 rhs = XEXP (x, 1);
9257 /* If either operand is a primitive we can't do anything, so get out
9258 fast. */
9259 if (OBJECT_P (lhs) || OBJECT_P (rhs))
9260 return x;
9262 lhs = expand_compound_operation (lhs);
9263 rhs = expand_compound_operation (rhs);
9264 inner_code = GET_CODE (lhs);
9265 if (inner_code != GET_CODE (rhs))
9266 return x;
9268 /* See if the inner and outer operations distribute. */
9269 switch (inner_code)
9271 case LSHIFTRT:
9272 case ASHIFTRT:
9273 case AND:
9274 case IOR:
9275 /* These all distribute except over PLUS. */
9276 if (code == PLUS || code == MINUS)
9277 return x;
9278 break;
9280 case MULT:
9281 if (code != PLUS && code != MINUS)
9282 return x;
9283 break;
9285 case ASHIFT:
9286 /* This is also a multiply, so it distributes over everything. */
9287 break;
9289 case SUBREG:
9290 /* Non-paradoxical SUBREGs distributes over all operations,
9291 provided the inner modes and byte offsets are the same, this
9292 is an extraction of a low-order part, we don't convert an fp
9293 operation to int or vice versa, this is not a vector mode,
9294 and we would not be converting a single-word operation into a
9295 multi-word operation. The latter test is not required, but
9296 it prevents generating unneeded multi-word operations. Some
9297 of the previous tests are redundant given the latter test,
9298 but are retained because they are required for correctness.
9300 We produce the result slightly differently in this case. */
9302 if (GET_MODE (SUBREG_REG (lhs)) != GET_MODE (SUBREG_REG (rhs))
9303 || SUBREG_BYTE (lhs) != SUBREG_BYTE (rhs)
9304 || ! subreg_lowpart_p (lhs)
9305 || (GET_MODE_CLASS (GET_MODE (lhs))
9306 != GET_MODE_CLASS (GET_MODE (SUBREG_REG (lhs))))
9307 || paradoxical_subreg_p (lhs)
9308 || VECTOR_MODE_P (GET_MODE (lhs))
9309 || GET_MODE_SIZE (GET_MODE (SUBREG_REG (lhs))) > UNITS_PER_WORD
9310 /* Result might need to be truncated. Don't change mode if
9311 explicit truncation is needed. */
9312 || !TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (x),
9313 GET_MODE (SUBREG_REG (lhs))))
9314 return x;
9316 tem = simplify_gen_binary (code, GET_MODE (SUBREG_REG (lhs)),
9317 SUBREG_REG (lhs), SUBREG_REG (rhs));
9318 return gen_lowpart (GET_MODE (x), tem);
9320 default:
9321 return x;
9324 /* Set LHS and RHS to the inner operands (A and B in the example
9325 above) and set OTHER to the common operand (C in the example).
9326 There is only one way to do this unless the inner operation is
9327 commutative. */
9328 if (COMMUTATIVE_ARITH_P (lhs)
9329 && rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 0)))
9330 other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 1);
9331 else if (COMMUTATIVE_ARITH_P (lhs)
9332 && rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 1)))
9333 other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 0);
9334 else if (COMMUTATIVE_ARITH_P (lhs)
9335 && rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 0)))
9336 other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 1);
9337 else if (rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 1)))
9338 other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 0);
9339 else
9340 return x;
9342 /* Form the new inner operation, seeing if it simplifies first. */
9343 tem = simplify_gen_binary (code, GET_MODE (x), lhs, rhs);
9345 /* There is one exception to the general way of distributing:
9346 (a | c) ^ (b | c) -> (a ^ b) & ~c */
9347 if (code == XOR && inner_code == IOR)
9349 inner_code = AND;
9350 other = simplify_gen_unary (NOT, GET_MODE (x), other, GET_MODE (x));
9353 /* We may be able to continuing distributing the result, so call
9354 ourselves recursively on the inner operation before forming the
9355 outer operation, which we return. */
9356 return simplify_gen_binary (inner_code, GET_MODE (x),
9357 apply_distributive_law (tem), other);
9360 /* See if X is of the form (* (+ A B) C), and if so convert to
9361 (+ (* A C) (* B C)) and try to simplify.
9363 Most of the time, this results in no change. However, if some of
9364 the operands are the same or inverses of each other, simplifications
9365 will result.
9367 For example, (and (ior A B) (not B)) can occur as the result of
9368 expanding a bit field assignment. When we apply the distributive
9369 law to this, we get (ior (and (A (not B))) (and (B (not B)))),
9370 which then simplifies to (and (A (not B))).
9372 Note that no checks happen on the validity of applying the inverse
9373 distributive law. This is pointless since we can do it in the
9374 few places where this routine is called.
9376 N is the index of the term that is decomposed (the arithmetic operation,
9377 i.e. (+ A B) in the first example above). !N is the index of the term that
9378 is distributed, i.e. of C in the first example above. */
9379 static rtx
9380 distribute_and_simplify_rtx (rtx x, int n)
9382 enum machine_mode mode;
9383 enum rtx_code outer_code, inner_code;
9384 rtx decomposed, distributed, inner_op0, inner_op1, new_op0, new_op1, tmp;
9386 /* Distributivity is not true for floating point as it can change the
9387 value. So we don't do it unless -funsafe-math-optimizations. */
9388 if (FLOAT_MODE_P (GET_MODE (x))
9389 && ! flag_unsafe_math_optimizations)
9390 return NULL_RTX;
9392 decomposed = XEXP (x, n);
9393 if (!ARITHMETIC_P (decomposed))
9394 return NULL_RTX;
9396 mode = GET_MODE (x);
9397 outer_code = GET_CODE (x);
9398 distributed = XEXP (x, !n);
9400 inner_code = GET_CODE (decomposed);
9401 inner_op0 = XEXP (decomposed, 0);
9402 inner_op1 = XEXP (decomposed, 1);
9404 /* Special case (and (xor B C) (not A)), which is equivalent to
9405 (xor (ior A B) (ior A C)) */
9406 if (outer_code == AND && inner_code == XOR && GET_CODE (distributed) == NOT)
9408 distributed = XEXP (distributed, 0);
9409 outer_code = IOR;
9412 if (n == 0)
9414 /* Distribute the second term. */
9415 new_op0 = simplify_gen_binary (outer_code, mode, inner_op0, distributed);
9416 new_op1 = simplify_gen_binary (outer_code, mode, inner_op1, distributed);
9418 else
9420 /* Distribute the first term. */
9421 new_op0 = simplify_gen_binary (outer_code, mode, distributed, inner_op0);
9422 new_op1 = simplify_gen_binary (outer_code, mode, distributed, inner_op1);
9425 tmp = apply_distributive_law (simplify_gen_binary (inner_code, mode,
9426 new_op0, new_op1));
9427 if (GET_CODE (tmp) != outer_code
9428 && (set_src_cost (tmp, optimize_this_for_speed_p)
9429 < set_src_cost (x, optimize_this_for_speed_p)))
9430 return tmp;
9432 return NULL_RTX;
9435 /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
9436 in MODE. Return an equivalent form, if different from (and VAROP
9437 (const_int CONSTOP)). Otherwise, return NULL_RTX. */
9439 static rtx
9440 simplify_and_const_int_1 (enum machine_mode mode, rtx varop,
9441 unsigned HOST_WIDE_INT constop)
9443 unsigned HOST_WIDE_INT nonzero;
9444 unsigned HOST_WIDE_INT orig_constop;
9445 rtx orig_varop;
9446 int i;
9448 orig_varop = varop;
9449 orig_constop = constop;
9450 if (GET_CODE (varop) == CLOBBER)
9451 return NULL_RTX;
9453 /* Simplify VAROP knowing that we will be only looking at some of the
9454 bits in it.
9456 Note by passing in CONSTOP, we guarantee that the bits not set in
9457 CONSTOP are not significant and will never be examined. We must
9458 ensure that is the case by explicitly masking out those bits
9459 before returning. */
9460 varop = force_to_mode (varop, mode, constop, 0);
9462 /* If VAROP is a CLOBBER, we will fail so return it. */
9463 if (GET_CODE (varop) == CLOBBER)
9464 return varop;
9466 /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
9467 to VAROP and return the new constant. */
9468 if (CONST_INT_P (varop))
9469 return gen_int_mode (INTVAL (varop) & constop, mode);
9471 /* See what bits may be nonzero in VAROP. Unlike the general case of
9472 a call to nonzero_bits, here we don't care about bits outside
9473 MODE. */
9475 nonzero = nonzero_bits (varop, mode) & GET_MODE_MASK (mode);
9477 /* Turn off all bits in the constant that are known to already be zero.
9478 Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
9479 which is tested below. */
9481 constop &= nonzero;
9483 /* If we don't have any bits left, return zero. */
9484 if (constop == 0)
9485 return const0_rtx;
9487 /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
9488 a power of two, we can replace this with an ASHIFT. */
9489 if (GET_CODE (varop) == NEG && nonzero_bits (XEXP (varop, 0), mode) == 1
9490 && (i = exact_log2 (constop)) >= 0)
9491 return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (varop, 0), i);
9493 /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
9494 or XOR, then try to apply the distributive law. This may eliminate
9495 operations if either branch can be simplified because of the AND.
9496 It may also make some cases more complex, but those cases probably
9497 won't match a pattern either with or without this. */
9499 if (GET_CODE (varop) == IOR || GET_CODE (varop) == XOR)
9500 return
9501 gen_lowpart
9502 (mode,
9503 apply_distributive_law
9504 (simplify_gen_binary (GET_CODE (varop), GET_MODE (varop),
9505 simplify_and_const_int (NULL_RTX,
9506 GET_MODE (varop),
9507 XEXP (varop, 0),
9508 constop),
9509 simplify_and_const_int (NULL_RTX,
9510 GET_MODE (varop),
9511 XEXP (varop, 1),
9512 constop))));
9514 /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
9515 the AND and see if one of the operands simplifies to zero. If so, we
9516 may eliminate it. */
9518 if (GET_CODE (varop) == PLUS
9519 && exact_log2 (constop + 1) >= 0)
9521 rtx o0, o1;
9523 o0 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 0), constop);
9524 o1 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 1), constop);
9525 if (o0 == const0_rtx)
9526 return o1;
9527 if (o1 == const0_rtx)
9528 return o0;
9531 /* Make a SUBREG if necessary. If we can't make it, fail. */
9532 varop = gen_lowpart (mode, varop);
9533 if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
9534 return NULL_RTX;
9536 /* If we are only masking insignificant bits, return VAROP. */
9537 if (constop == nonzero)
9538 return varop;
9540 if (varop == orig_varop && constop == orig_constop)
9541 return NULL_RTX;
9543 /* Otherwise, return an AND. */
9544 return simplify_gen_binary (AND, mode, varop, gen_int_mode (constop, mode));
9548 /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
9549 in MODE.
9551 Return an equivalent form, if different from X. Otherwise, return X. If
9552 X is zero, we are to always construct the equivalent form. */
9554 static rtx
9555 simplify_and_const_int (rtx x, enum machine_mode mode, rtx varop,
9556 unsigned HOST_WIDE_INT constop)
9558 rtx tem = simplify_and_const_int_1 (mode, varop, constop);
9559 if (tem)
9560 return tem;
9562 if (!x)
9563 x = simplify_gen_binary (AND, GET_MODE (varop), varop,
9564 gen_int_mode (constop, mode));
9565 if (GET_MODE (x) != mode)
9566 x = gen_lowpart (mode, x);
9567 return x;
9570 /* Given a REG, X, compute which bits in X can be nonzero.
9571 We don't care about bits outside of those defined in MODE.
9573 For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
9574 a shift, AND, or zero_extract, we can do better. */
9576 static rtx
9577 reg_nonzero_bits_for_combine (const_rtx x, enum machine_mode mode,
9578 const_rtx known_x ATTRIBUTE_UNUSED,
9579 enum machine_mode known_mode ATTRIBUTE_UNUSED,
9580 unsigned HOST_WIDE_INT known_ret ATTRIBUTE_UNUSED,
9581 unsigned HOST_WIDE_INT *nonzero)
9583 rtx tem;
9584 reg_stat_type *rsp;
9586 /* If X is a register whose nonzero bits value is current, use it.
9587 Otherwise, if X is a register whose value we can find, use that
9588 value. Otherwise, use the previously-computed global nonzero bits
9589 for this register. */
9591 rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
9592 if (rsp->last_set_value != 0
9593 && (rsp->last_set_mode == mode
9594 || (GET_MODE_CLASS (rsp->last_set_mode) == MODE_INT
9595 && GET_MODE_CLASS (mode) == MODE_INT))
9596 && ((rsp->last_set_label >= label_tick_ebb_start
9597 && rsp->last_set_label < label_tick)
9598 || (rsp->last_set_label == label_tick
9599 && DF_INSN_LUID (rsp->last_set) < subst_low_luid)
9600 || (REGNO (x) >= FIRST_PSEUDO_REGISTER
9601 && REG_N_SETS (REGNO (x)) == 1
9602 && !REGNO_REG_SET_P
9603 (DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x)))))
9605 *nonzero &= rsp->last_set_nonzero_bits;
9606 return NULL;
9609 tem = get_last_value (x);
9611 if (tem)
9613 #ifdef SHORT_IMMEDIATES_SIGN_EXTEND
9614 /* If X is narrower than MODE and TEM is a non-negative
9615 constant that would appear negative in the mode of X,
9616 sign-extend it for use in reg_nonzero_bits because some
9617 machines (maybe most) will actually do the sign-extension
9618 and this is the conservative approach.
9620 ??? For 2.5, try to tighten up the MD files in this regard
9621 instead of this kludge. */
9623 if (GET_MODE_PRECISION (GET_MODE (x)) < GET_MODE_PRECISION (mode)
9624 && CONST_INT_P (tem)
9625 && INTVAL (tem) > 0
9626 && val_signbit_known_set_p (GET_MODE (x), INTVAL (tem)))
9627 tem = GEN_INT (INTVAL (tem) | ~GET_MODE_MASK (GET_MODE (x)));
9628 #endif
9629 return tem;
9631 else if (nonzero_sign_valid && rsp->nonzero_bits)
9633 unsigned HOST_WIDE_INT mask = rsp->nonzero_bits;
9635 if (GET_MODE_PRECISION (GET_MODE (x)) < GET_MODE_PRECISION (mode))
9636 /* We don't know anything about the upper bits. */
9637 mask |= GET_MODE_MASK (mode) ^ GET_MODE_MASK (GET_MODE (x));
9638 *nonzero &= mask;
9641 return NULL;
9644 /* Return the number of bits at the high-order end of X that are known to
9645 be equal to the sign bit. X will be used in mode MODE; if MODE is
9646 VOIDmode, X will be used in its own mode. The returned value will always
9647 be between 1 and the number of bits in MODE. */
9649 static rtx
9650 reg_num_sign_bit_copies_for_combine (const_rtx x, enum machine_mode mode,
9651 const_rtx known_x ATTRIBUTE_UNUSED,
9652 enum machine_mode known_mode
9653 ATTRIBUTE_UNUSED,
9654 unsigned int known_ret ATTRIBUTE_UNUSED,
9655 unsigned int *result)
9657 rtx tem;
9658 reg_stat_type *rsp;
9660 rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
9661 if (rsp->last_set_value != 0
9662 && rsp->last_set_mode == mode
9663 && ((rsp->last_set_label >= label_tick_ebb_start
9664 && rsp->last_set_label < label_tick)
9665 || (rsp->last_set_label == label_tick
9666 && DF_INSN_LUID (rsp->last_set) < subst_low_luid)
9667 || (REGNO (x) >= FIRST_PSEUDO_REGISTER
9668 && REG_N_SETS (REGNO (x)) == 1
9669 && !REGNO_REG_SET_P
9670 (DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x)))))
9672 *result = rsp->last_set_sign_bit_copies;
9673 return NULL;
9676 tem = get_last_value (x);
9677 if (tem != 0)
9678 return tem;
9680 if (nonzero_sign_valid && rsp->sign_bit_copies != 0
9681 && GET_MODE_PRECISION (GET_MODE (x)) == GET_MODE_PRECISION (mode))
9682 *result = rsp->sign_bit_copies;
9684 return NULL;
9687 /* Return the number of "extended" bits there are in X, when interpreted
9688 as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
9689 unsigned quantities, this is the number of high-order zero bits.
9690 For signed quantities, this is the number of copies of the sign bit
9691 minus 1. In both case, this function returns the number of "spare"
9692 bits. For example, if two quantities for which this function returns
9693 at least 1 are added, the addition is known not to overflow.
9695 This function will always return 0 unless called during combine, which
9696 implies that it must be called from a define_split. */
9698 unsigned int
9699 extended_count (const_rtx x, enum machine_mode mode, int unsignedp)
9701 if (nonzero_sign_valid == 0)
9702 return 0;
9704 return (unsignedp
9705 ? (HWI_COMPUTABLE_MODE_P (mode)
9706 ? (unsigned int) (GET_MODE_PRECISION (mode) - 1
9707 - floor_log2 (nonzero_bits (x, mode)))
9708 : 0)
9709 : num_sign_bit_copies (x, mode) - 1);
9712 /* This function is called from `simplify_shift_const' to merge two
9713 outer operations. Specifically, we have already found that we need
9714 to perform operation *POP0 with constant *PCONST0 at the outermost
9715 position. We would now like to also perform OP1 with constant CONST1
9716 (with *POP0 being done last).
9718 Return 1 if we can do the operation and update *POP0 and *PCONST0 with
9719 the resulting operation. *PCOMP_P is set to 1 if we would need to
9720 complement the innermost operand, otherwise it is unchanged.
9722 MODE is the mode in which the operation will be done. No bits outside
9723 the width of this mode matter. It is assumed that the width of this mode
9724 is smaller than or equal to HOST_BITS_PER_WIDE_INT.
9726 If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
9727 IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
9728 result is simply *PCONST0.
9730 If the resulting operation cannot be expressed as one operation, we
9731 return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
9733 static int
9734 merge_outer_ops (enum rtx_code *pop0, HOST_WIDE_INT *pconst0, enum rtx_code op1, HOST_WIDE_INT const1, enum machine_mode mode, int *pcomp_p)
9736 enum rtx_code op0 = *pop0;
9737 HOST_WIDE_INT const0 = *pconst0;
9739 const0 &= GET_MODE_MASK (mode);
9740 const1 &= GET_MODE_MASK (mode);
9742 /* If OP0 is an AND, clear unimportant bits in CONST1. */
9743 if (op0 == AND)
9744 const1 &= const0;
9746 /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
9747 if OP0 is SET. */
9749 if (op1 == UNKNOWN || op0 == SET)
9750 return 1;
9752 else if (op0 == UNKNOWN)
9753 op0 = op1, const0 = const1;
9755 else if (op0 == op1)
9757 switch (op0)
9759 case AND:
9760 const0 &= const1;
9761 break;
9762 case IOR:
9763 const0 |= const1;
9764 break;
9765 case XOR:
9766 const0 ^= const1;
9767 break;
9768 case PLUS:
9769 const0 += const1;
9770 break;
9771 case NEG:
9772 op0 = UNKNOWN;
9773 break;
9774 default:
9775 break;
9779 /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
9780 else if (op0 == PLUS || op1 == PLUS || op0 == NEG || op1 == NEG)
9781 return 0;
9783 /* If the two constants aren't the same, we can't do anything. The
9784 remaining six cases can all be done. */
9785 else if (const0 != const1)
9786 return 0;
9788 else
9789 switch (op0)
9791 case IOR:
9792 if (op1 == AND)
9793 /* (a & b) | b == b */
9794 op0 = SET;
9795 else /* op1 == XOR */
9796 /* (a ^ b) | b == a | b */
9798 break;
9800 case XOR:
9801 if (op1 == AND)
9802 /* (a & b) ^ b == (~a) & b */
9803 op0 = AND, *pcomp_p = 1;
9804 else /* op1 == IOR */
9805 /* (a | b) ^ b == a & ~b */
9806 op0 = AND, const0 = ~const0;
9807 break;
9809 case AND:
9810 if (op1 == IOR)
9811 /* (a | b) & b == b */
9812 op0 = SET;
9813 else /* op1 == XOR */
9814 /* (a ^ b) & b) == (~a) & b */
9815 *pcomp_p = 1;
9816 break;
9817 default:
9818 break;
9821 /* Check for NO-OP cases. */
9822 const0 &= GET_MODE_MASK (mode);
9823 if (const0 == 0
9824 && (op0 == IOR || op0 == XOR || op0 == PLUS))
9825 op0 = UNKNOWN;
9826 else if (const0 == 0 && op0 == AND)
9827 op0 = SET;
9828 else if ((unsigned HOST_WIDE_INT) const0 == GET_MODE_MASK (mode)
9829 && op0 == AND)
9830 op0 = UNKNOWN;
9832 *pop0 = op0;
9834 /* ??? Slightly redundant with the above mask, but not entirely.
9835 Moving this above means we'd have to sign-extend the mode mask
9836 for the final test. */
9837 if (op0 != UNKNOWN && op0 != NEG)
9838 *pconst0 = trunc_int_for_mode (const0, mode);
9840 return 1;
9843 /* A helper to simplify_shift_const_1 to determine the mode we can perform
9844 the shift in. The original shift operation CODE is performed on OP in
9845 ORIG_MODE. Return the wider mode MODE if we can perform the operation
9846 in that mode. Return ORIG_MODE otherwise. We can also assume that the
9847 result of the shift is subject to operation OUTER_CODE with operand
9848 OUTER_CONST. */
9850 static enum machine_mode
9851 try_widen_shift_mode (enum rtx_code code, rtx op, int count,
9852 enum machine_mode orig_mode, enum machine_mode mode,
9853 enum rtx_code outer_code, HOST_WIDE_INT outer_const)
9855 if (orig_mode == mode)
9856 return mode;
9857 gcc_assert (GET_MODE_PRECISION (mode) > GET_MODE_PRECISION (orig_mode));
9859 /* In general we can't perform in wider mode for right shift and rotate. */
9860 switch (code)
9862 case ASHIFTRT:
9863 /* We can still widen if the bits brought in from the left are identical
9864 to the sign bit of ORIG_MODE. */
9865 if (num_sign_bit_copies (op, mode)
9866 > (unsigned) (GET_MODE_PRECISION (mode)
9867 - GET_MODE_PRECISION (orig_mode)))
9868 return mode;
9869 return orig_mode;
9871 case LSHIFTRT:
9872 /* Similarly here but with zero bits. */
9873 if (HWI_COMPUTABLE_MODE_P (mode)
9874 && (nonzero_bits (op, mode) & ~GET_MODE_MASK (orig_mode)) == 0)
9875 return mode;
9877 /* We can also widen if the bits brought in will be masked off. This
9878 operation is performed in ORIG_MODE. */
9879 if (outer_code == AND)
9881 int care_bits = low_bitmask_len (orig_mode, outer_const);
9883 if (care_bits >= 0
9884 && GET_MODE_PRECISION (orig_mode) - care_bits >= count)
9885 return mode;
9887 /* fall through */
9889 case ROTATE:
9890 return orig_mode;
9892 case ROTATERT:
9893 gcc_unreachable ();
9895 default:
9896 return mode;
9900 /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
9901 of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
9902 if we cannot simplify it. Otherwise, return a simplified value.
9904 The shift is normally computed in the widest mode we find in VAROP, as
9905 long as it isn't a different number of words than RESULT_MODE. Exceptions
9906 are ASHIFTRT and ROTATE, which are always done in their original mode. */
9908 static rtx
9909 simplify_shift_const_1 (enum rtx_code code, enum machine_mode result_mode,
9910 rtx varop, int orig_count)
9912 enum rtx_code orig_code = code;
9913 rtx orig_varop = varop;
9914 int count;
9915 enum machine_mode mode = result_mode;
9916 enum machine_mode shift_mode, tmode;
9917 unsigned int mode_words
9918 = (GET_MODE_SIZE (mode) + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD;
9919 /* We form (outer_op (code varop count) (outer_const)). */
9920 enum rtx_code outer_op = UNKNOWN;
9921 HOST_WIDE_INT outer_const = 0;
9922 int complement_p = 0;
9923 rtx new_rtx, x;
9925 /* Make sure and truncate the "natural" shift on the way in. We don't
9926 want to do this inside the loop as it makes it more difficult to
9927 combine shifts. */
9928 if (SHIFT_COUNT_TRUNCATED)
9929 orig_count &= GET_MODE_BITSIZE (mode) - 1;
9931 /* If we were given an invalid count, don't do anything except exactly
9932 what was requested. */
9934 if (orig_count < 0 || orig_count >= (int) GET_MODE_PRECISION (mode))
9935 return NULL_RTX;
9937 count = orig_count;
9939 /* Unless one of the branches of the `if' in this loop does a `continue',
9940 we will `break' the loop after the `if'. */
9942 while (count != 0)
9944 /* If we have an operand of (clobber (const_int 0)), fail. */
9945 if (GET_CODE (varop) == CLOBBER)
9946 return NULL_RTX;
9948 /* Convert ROTATERT to ROTATE. */
9949 if (code == ROTATERT)
9951 unsigned int bitsize = GET_MODE_PRECISION (result_mode);
9952 code = ROTATE;
9953 if (VECTOR_MODE_P (result_mode))
9954 count = bitsize / GET_MODE_NUNITS (result_mode) - count;
9955 else
9956 count = bitsize - count;
9959 shift_mode = try_widen_shift_mode (code, varop, count, result_mode,
9960 mode, outer_op, outer_const);
9962 /* Handle cases where the count is greater than the size of the mode
9963 minus 1. For ASHIFT, use the size minus one as the count (this can
9964 occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
9965 take the count modulo the size. For other shifts, the result is
9966 zero.
9968 Since these shifts are being produced by the compiler by combining
9969 multiple operations, each of which are defined, we know what the
9970 result is supposed to be. */
9972 if (count > (GET_MODE_PRECISION (shift_mode) - 1))
9974 if (code == ASHIFTRT)
9975 count = GET_MODE_PRECISION (shift_mode) - 1;
9976 else if (code == ROTATE || code == ROTATERT)
9977 count %= GET_MODE_PRECISION (shift_mode);
9978 else
9980 /* We can't simply return zero because there may be an
9981 outer op. */
9982 varop = const0_rtx;
9983 count = 0;
9984 break;
9988 /* If we discovered we had to complement VAROP, leave. Making a NOT
9989 here would cause an infinite loop. */
9990 if (complement_p)
9991 break;
9993 /* An arithmetic right shift of a quantity known to be -1 or 0
9994 is a no-op. */
9995 if (code == ASHIFTRT
9996 && (num_sign_bit_copies (varop, shift_mode)
9997 == GET_MODE_PRECISION (shift_mode)))
9999 count = 0;
10000 break;
10003 /* If we are doing an arithmetic right shift and discarding all but
10004 the sign bit copies, this is equivalent to doing a shift by the
10005 bitsize minus one. Convert it into that shift because it will often
10006 allow other simplifications. */
10008 if (code == ASHIFTRT
10009 && (count + num_sign_bit_copies (varop, shift_mode)
10010 >= GET_MODE_PRECISION (shift_mode)))
10011 count = GET_MODE_PRECISION (shift_mode) - 1;
10013 /* We simplify the tests below and elsewhere by converting
10014 ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10015 `make_compound_operation' will convert it to an ASHIFTRT for
10016 those machines (such as VAX) that don't have an LSHIFTRT. */
10017 if (code == ASHIFTRT
10018 && val_signbit_known_clear_p (shift_mode,
10019 nonzero_bits (varop, shift_mode)))
10020 code = LSHIFTRT;
10022 if (((code == LSHIFTRT
10023 && HWI_COMPUTABLE_MODE_P (shift_mode)
10024 && !(nonzero_bits (varop, shift_mode) >> count))
10025 || (code == ASHIFT
10026 && HWI_COMPUTABLE_MODE_P (shift_mode)
10027 && !((nonzero_bits (varop, shift_mode) << count)
10028 & GET_MODE_MASK (shift_mode))))
10029 && !side_effects_p (varop))
10030 varop = const0_rtx;
10032 switch (GET_CODE (varop))
10034 case SIGN_EXTEND:
10035 case ZERO_EXTEND:
10036 case SIGN_EXTRACT:
10037 case ZERO_EXTRACT:
10038 new_rtx = expand_compound_operation (varop);
10039 if (new_rtx != varop)
10041 varop = new_rtx;
10042 continue;
10044 break;
10046 case MEM:
10047 /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10048 minus the width of a smaller mode, we can do this with a
10049 SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10050 if ((code == ASHIFTRT || code == LSHIFTRT)
10051 && ! mode_dependent_address_p (XEXP (varop, 0))
10052 && ! MEM_VOLATILE_P (varop)
10053 && (tmode = mode_for_size (GET_MODE_BITSIZE (mode) - count,
10054 MODE_INT, 1)) != BLKmode)
10056 new_rtx = adjust_address_nv (varop, tmode,
10057 BYTES_BIG_ENDIAN ? 0
10058 : count / BITS_PER_UNIT);
10060 varop = gen_rtx_fmt_e (code == ASHIFTRT ? SIGN_EXTEND
10061 : ZERO_EXTEND, mode, new_rtx);
10062 count = 0;
10063 continue;
10065 break;
10067 case SUBREG:
10068 /* If VAROP is a SUBREG, strip it as long as the inner operand has
10069 the same number of words as what we've seen so far. Then store
10070 the widest mode in MODE. */
10071 if (subreg_lowpart_p (varop)
10072 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (varop)))
10073 > GET_MODE_SIZE (GET_MODE (varop)))
10074 && (unsigned int) ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (varop)))
10075 + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
10076 == mode_words
10077 && GET_MODE_CLASS (GET_MODE (varop)) == MODE_INT
10078 && GET_MODE_CLASS (GET_MODE (SUBREG_REG (varop))) == MODE_INT)
10080 varop = SUBREG_REG (varop);
10081 if (GET_MODE_SIZE (GET_MODE (varop)) > GET_MODE_SIZE (mode))
10082 mode = GET_MODE (varop);
10083 continue;
10085 break;
10087 case MULT:
10088 /* Some machines use MULT instead of ASHIFT because MULT
10089 is cheaper. But it is still better on those machines to
10090 merge two shifts into one. */
10091 if (CONST_INT_P (XEXP (varop, 1))
10092 && exact_log2 (UINTVAL (XEXP (varop, 1))) >= 0)
10094 varop
10095 = simplify_gen_binary (ASHIFT, GET_MODE (varop),
10096 XEXP (varop, 0),
10097 GEN_INT (exact_log2 (
10098 UINTVAL (XEXP (varop, 1)))));
10099 continue;
10101 break;
10103 case UDIV:
10104 /* Similar, for when divides are cheaper. */
10105 if (CONST_INT_P (XEXP (varop, 1))
10106 && exact_log2 (UINTVAL (XEXP (varop, 1))) >= 0)
10108 varop
10109 = simplify_gen_binary (LSHIFTRT, GET_MODE (varop),
10110 XEXP (varop, 0),
10111 GEN_INT (exact_log2 (
10112 UINTVAL (XEXP (varop, 1)))));
10113 continue;
10115 break;
10117 case ASHIFTRT:
10118 /* If we are extracting just the sign bit of an arithmetic
10119 right shift, that shift is not needed. However, the sign
10120 bit of a wider mode may be different from what would be
10121 interpreted as the sign bit in a narrower mode, so, if
10122 the result is narrower, don't discard the shift. */
10123 if (code == LSHIFTRT
10124 && count == (GET_MODE_BITSIZE (result_mode) - 1)
10125 && (GET_MODE_BITSIZE (result_mode)
10126 >= GET_MODE_BITSIZE (GET_MODE (varop))))
10128 varop = XEXP (varop, 0);
10129 continue;
10132 /* ... fall through ... */
10134 case LSHIFTRT:
10135 case ASHIFT:
10136 case ROTATE:
10137 /* Here we have two nested shifts. The result is usually the
10138 AND of a new shift with a mask. We compute the result below. */
10139 if (CONST_INT_P (XEXP (varop, 1))
10140 && INTVAL (XEXP (varop, 1)) >= 0
10141 && INTVAL (XEXP (varop, 1)) < GET_MODE_PRECISION (GET_MODE (varop))
10142 && HWI_COMPUTABLE_MODE_P (result_mode)
10143 && HWI_COMPUTABLE_MODE_P (mode)
10144 && !VECTOR_MODE_P (result_mode))
10146 enum rtx_code first_code = GET_CODE (varop);
10147 unsigned int first_count = INTVAL (XEXP (varop, 1));
10148 unsigned HOST_WIDE_INT mask;
10149 rtx mask_rtx;
10151 /* We have one common special case. We can't do any merging if
10152 the inner code is an ASHIFTRT of a smaller mode. However, if
10153 we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10154 with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10155 we can convert it to
10156 (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10157 This simplifies certain SIGN_EXTEND operations. */
10158 if (code == ASHIFT && first_code == ASHIFTRT
10159 && count == (GET_MODE_PRECISION (result_mode)
10160 - GET_MODE_PRECISION (GET_MODE (varop))))
10162 /* C3 has the low-order C1 bits zero. */
10164 mask = GET_MODE_MASK (mode)
10165 & ~(((unsigned HOST_WIDE_INT) 1 << first_count) - 1);
10167 varop = simplify_and_const_int (NULL_RTX, result_mode,
10168 XEXP (varop, 0), mask);
10169 varop = simplify_shift_const (NULL_RTX, ASHIFT, result_mode,
10170 varop, count);
10171 count = first_count;
10172 code = ASHIFTRT;
10173 continue;
10176 /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10177 than C1 high-order bits equal to the sign bit, we can convert
10178 this to either an ASHIFT or an ASHIFTRT depending on the
10179 two counts.
10181 We cannot do this if VAROP's mode is not SHIFT_MODE. */
10183 if (code == ASHIFTRT && first_code == ASHIFT
10184 && GET_MODE (varop) == shift_mode
10185 && (num_sign_bit_copies (XEXP (varop, 0), shift_mode)
10186 > first_count))
10188 varop = XEXP (varop, 0);
10189 count -= first_count;
10190 if (count < 0)
10192 count = -count;
10193 code = ASHIFT;
10196 continue;
10199 /* There are some cases we can't do. If CODE is ASHIFTRT,
10200 we can only do this if FIRST_CODE is also ASHIFTRT.
10202 We can't do the case when CODE is ROTATE and FIRST_CODE is
10203 ASHIFTRT.
10205 If the mode of this shift is not the mode of the outer shift,
10206 we can't do this if either shift is a right shift or ROTATE.
10208 Finally, we can't do any of these if the mode is too wide
10209 unless the codes are the same.
10211 Handle the case where the shift codes are the same
10212 first. */
10214 if (code == first_code)
10216 if (GET_MODE (varop) != result_mode
10217 && (code == ASHIFTRT || code == LSHIFTRT
10218 || code == ROTATE))
10219 break;
10221 count += first_count;
10222 varop = XEXP (varop, 0);
10223 continue;
10226 if (code == ASHIFTRT
10227 || (code == ROTATE && first_code == ASHIFTRT)
10228 || GET_MODE_PRECISION (mode) > HOST_BITS_PER_WIDE_INT
10229 || (GET_MODE (varop) != result_mode
10230 && (first_code == ASHIFTRT || first_code == LSHIFTRT
10231 || first_code == ROTATE
10232 || code == ROTATE)))
10233 break;
10235 /* To compute the mask to apply after the shift, shift the
10236 nonzero bits of the inner shift the same way the
10237 outer shift will. */
10239 mask_rtx = GEN_INT (nonzero_bits (varop, GET_MODE (varop)));
10241 mask_rtx
10242 = simplify_const_binary_operation (code, result_mode, mask_rtx,
10243 GEN_INT (count));
10245 /* Give up if we can't compute an outer operation to use. */
10246 if (mask_rtx == 0
10247 || !CONST_INT_P (mask_rtx)
10248 || ! merge_outer_ops (&outer_op, &outer_const, AND,
10249 INTVAL (mask_rtx),
10250 result_mode, &complement_p))
10251 break;
10253 /* If the shifts are in the same direction, we add the
10254 counts. Otherwise, we subtract them. */
10255 if ((code == ASHIFTRT || code == LSHIFTRT)
10256 == (first_code == ASHIFTRT || first_code == LSHIFTRT))
10257 count += first_count;
10258 else
10259 count -= first_count;
10261 /* If COUNT is positive, the new shift is usually CODE,
10262 except for the two exceptions below, in which case it is
10263 FIRST_CODE. If the count is negative, FIRST_CODE should
10264 always be used */
10265 if (count > 0
10266 && ((first_code == ROTATE && code == ASHIFT)
10267 || (first_code == ASHIFTRT && code == LSHIFTRT)))
10268 code = first_code;
10269 else if (count < 0)
10270 code = first_code, count = -count;
10272 varop = XEXP (varop, 0);
10273 continue;
10276 /* If we have (A << B << C) for any shift, we can convert this to
10277 (A << C << B). This wins if A is a constant. Only try this if
10278 B is not a constant. */
10280 else if (GET_CODE (varop) == code
10281 && CONST_INT_P (XEXP (varop, 0))
10282 && !CONST_INT_P (XEXP (varop, 1)))
10284 rtx new_rtx = simplify_const_binary_operation (code, mode,
10285 XEXP (varop, 0),
10286 GEN_INT (count));
10287 varop = gen_rtx_fmt_ee (code, mode, new_rtx, XEXP (varop, 1));
10288 count = 0;
10289 continue;
10291 break;
10293 case NOT:
10294 if (VECTOR_MODE_P (mode))
10295 break;
10297 /* Make this fit the case below. */
10298 varop = gen_rtx_XOR (mode, XEXP (varop, 0),
10299 GEN_INT (GET_MODE_MASK (mode)));
10300 continue;
10302 case IOR:
10303 case AND:
10304 case XOR:
10305 /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
10306 with C the size of VAROP - 1 and the shift is logical if
10307 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10308 we have an (le X 0) operation. If we have an arithmetic shift
10309 and STORE_FLAG_VALUE is 1 or we have a logical shift with
10310 STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
10312 if (GET_CODE (varop) == IOR && GET_CODE (XEXP (varop, 0)) == PLUS
10313 && XEXP (XEXP (varop, 0), 1) == constm1_rtx
10314 && (STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
10315 && (code == LSHIFTRT || code == ASHIFTRT)
10316 && count == (GET_MODE_PRECISION (GET_MODE (varop)) - 1)
10317 && rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
10319 count = 0;
10320 varop = gen_rtx_LE (GET_MODE (varop), XEXP (varop, 1),
10321 const0_rtx);
10323 if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
10324 varop = gen_rtx_NEG (GET_MODE (varop), varop);
10326 continue;
10329 /* If we have (shift (logical)), move the logical to the outside
10330 to allow it to possibly combine with another logical and the
10331 shift to combine with another shift. This also canonicalizes to
10332 what a ZERO_EXTRACT looks like. Also, some machines have
10333 (and (shift)) insns. */
10335 if (CONST_INT_P (XEXP (varop, 1))
10336 /* We can't do this if we have (ashiftrt (xor)) and the
10337 constant has its sign bit set in shift_mode. */
10338 && !(code == ASHIFTRT && GET_CODE (varop) == XOR
10339 && 0 > trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
10340 shift_mode))
10341 && (new_rtx = simplify_const_binary_operation (code, result_mode,
10342 XEXP (varop, 1),
10343 GEN_INT (count))) != 0
10344 && CONST_INT_P (new_rtx)
10345 && merge_outer_ops (&outer_op, &outer_const, GET_CODE (varop),
10346 INTVAL (new_rtx), result_mode, &complement_p))
10348 varop = XEXP (varop, 0);
10349 continue;
10352 /* If we can't do that, try to simplify the shift in each arm of the
10353 logical expression, make a new logical expression, and apply
10354 the inverse distributive law. This also can't be done
10355 for some (ashiftrt (xor)). */
10356 if (CONST_INT_P (XEXP (varop, 1))
10357 && !(code == ASHIFTRT && GET_CODE (varop) == XOR
10358 && 0 > trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
10359 shift_mode)))
10361 rtx lhs = simplify_shift_const (NULL_RTX, code, shift_mode,
10362 XEXP (varop, 0), count);
10363 rtx rhs = simplify_shift_const (NULL_RTX, code, shift_mode,
10364 XEXP (varop, 1), count);
10366 varop = simplify_gen_binary (GET_CODE (varop), shift_mode,
10367 lhs, rhs);
10368 varop = apply_distributive_law (varop);
10370 count = 0;
10371 continue;
10373 break;
10375 case EQ:
10376 /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
10377 says that the sign bit can be tested, FOO has mode MODE, C is
10378 GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
10379 that may be nonzero. */
10380 if (code == LSHIFTRT
10381 && XEXP (varop, 1) == const0_rtx
10382 && GET_MODE (XEXP (varop, 0)) == result_mode
10383 && count == (GET_MODE_PRECISION (result_mode) - 1)
10384 && HWI_COMPUTABLE_MODE_P (result_mode)
10385 && STORE_FLAG_VALUE == -1
10386 && nonzero_bits (XEXP (varop, 0), result_mode) == 1
10387 && merge_outer_ops (&outer_op, &outer_const, XOR, 1, result_mode,
10388 &complement_p))
10390 varop = XEXP (varop, 0);
10391 count = 0;
10392 continue;
10394 break;
10396 case NEG:
10397 /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
10398 than the number of bits in the mode is equivalent to A. */
10399 if (code == LSHIFTRT
10400 && count == (GET_MODE_PRECISION (result_mode) - 1)
10401 && nonzero_bits (XEXP (varop, 0), result_mode) == 1)
10403 varop = XEXP (varop, 0);
10404 count = 0;
10405 continue;
10408 /* NEG commutes with ASHIFT since it is multiplication. Move the
10409 NEG outside to allow shifts to combine. */
10410 if (code == ASHIFT
10411 && merge_outer_ops (&outer_op, &outer_const, NEG, 0, result_mode,
10412 &complement_p))
10414 varop = XEXP (varop, 0);
10415 continue;
10417 break;
10419 case PLUS:
10420 /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
10421 is one less than the number of bits in the mode is
10422 equivalent to (xor A 1). */
10423 if (code == LSHIFTRT
10424 && count == (GET_MODE_PRECISION (result_mode) - 1)
10425 && XEXP (varop, 1) == constm1_rtx
10426 && nonzero_bits (XEXP (varop, 0), result_mode) == 1
10427 && merge_outer_ops (&outer_op, &outer_const, XOR, 1, result_mode,
10428 &complement_p))
10430 count = 0;
10431 varop = XEXP (varop, 0);
10432 continue;
10435 /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
10436 that might be nonzero in BAR are those being shifted out and those
10437 bits are known zero in FOO, we can replace the PLUS with FOO.
10438 Similarly in the other operand order. This code occurs when
10439 we are computing the size of a variable-size array. */
10441 if ((code == ASHIFTRT || code == LSHIFTRT)
10442 && count < HOST_BITS_PER_WIDE_INT
10443 && nonzero_bits (XEXP (varop, 1), result_mode) >> count == 0
10444 && (nonzero_bits (XEXP (varop, 1), result_mode)
10445 & nonzero_bits (XEXP (varop, 0), result_mode)) == 0)
10447 varop = XEXP (varop, 0);
10448 continue;
10450 else if ((code == ASHIFTRT || code == LSHIFTRT)
10451 && count < HOST_BITS_PER_WIDE_INT
10452 && HWI_COMPUTABLE_MODE_P (result_mode)
10453 && 0 == (nonzero_bits (XEXP (varop, 0), result_mode)
10454 >> count)
10455 && 0 == (nonzero_bits (XEXP (varop, 0), result_mode)
10456 & nonzero_bits (XEXP (varop, 1),
10457 result_mode)))
10459 varop = XEXP (varop, 1);
10460 continue;
10463 /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
10464 if (code == ASHIFT
10465 && CONST_INT_P (XEXP (varop, 1))
10466 && (new_rtx = simplify_const_binary_operation (ASHIFT, result_mode,
10467 XEXP (varop, 1),
10468 GEN_INT (count))) != 0
10469 && CONST_INT_P (new_rtx)
10470 && merge_outer_ops (&outer_op, &outer_const, PLUS,
10471 INTVAL (new_rtx), result_mode, &complement_p))
10473 varop = XEXP (varop, 0);
10474 continue;
10477 /* Check for 'PLUS signbit', which is the canonical form of 'XOR
10478 signbit', and attempt to change the PLUS to an XOR and move it to
10479 the outer operation as is done above in the AND/IOR/XOR case
10480 leg for shift(logical). See details in logical handling above
10481 for reasoning in doing so. */
10482 if (code == LSHIFTRT
10483 && CONST_INT_P (XEXP (varop, 1))
10484 && mode_signbit_p (result_mode, XEXP (varop, 1))
10485 && (new_rtx = simplify_const_binary_operation (code, result_mode,
10486 XEXP (varop, 1),
10487 GEN_INT (count))) != 0
10488 && CONST_INT_P (new_rtx)
10489 && merge_outer_ops (&outer_op, &outer_const, XOR,
10490 INTVAL (new_rtx), result_mode, &complement_p))
10492 varop = XEXP (varop, 0);
10493 continue;
10496 break;
10498 case MINUS:
10499 /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
10500 with C the size of VAROP - 1 and the shift is logical if
10501 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10502 we have a (gt X 0) operation. If the shift is arithmetic with
10503 STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
10504 we have a (neg (gt X 0)) operation. */
10506 if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
10507 && GET_CODE (XEXP (varop, 0)) == ASHIFTRT
10508 && count == (GET_MODE_PRECISION (GET_MODE (varop)) - 1)
10509 && (code == LSHIFTRT || code == ASHIFTRT)
10510 && CONST_INT_P (XEXP (XEXP (varop, 0), 1))
10511 && INTVAL (XEXP (XEXP (varop, 0), 1)) == count
10512 && rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
10514 count = 0;
10515 varop = gen_rtx_GT (GET_MODE (varop), XEXP (varop, 1),
10516 const0_rtx);
10518 if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
10519 varop = gen_rtx_NEG (GET_MODE (varop), varop);
10521 continue;
10523 break;
10525 case TRUNCATE:
10526 /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
10527 if the truncate does not affect the value. */
10528 if (code == LSHIFTRT
10529 && GET_CODE (XEXP (varop, 0)) == LSHIFTRT
10530 && CONST_INT_P (XEXP (XEXP (varop, 0), 1))
10531 && (INTVAL (XEXP (XEXP (varop, 0), 1))
10532 >= (GET_MODE_PRECISION (GET_MODE (XEXP (varop, 0)))
10533 - GET_MODE_PRECISION (GET_MODE (varop)))))
10535 rtx varop_inner = XEXP (varop, 0);
10537 varop_inner
10538 = gen_rtx_LSHIFTRT (GET_MODE (varop_inner),
10539 XEXP (varop_inner, 0),
10540 GEN_INT
10541 (count + INTVAL (XEXP (varop_inner, 1))));
10542 varop = gen_rtx_TRUNCATE (GET_MODE (varop), varop_inner);
10543 count = 0;
10544 continue;
10546 break;
10548 default:
10549 break;
10552 break;
10555 shift_mode = try_widen_shift_mode (code, varop, count, result_mode, mode,
10556 outer_op, outer_const);
10558 /* We have now finished analyzing the shift. The result should be
10559 a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
10560 OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
10561 to the result of the shift. OUTER_CONST is the relevant constant,
10562 but we must turn off all bits turned off in the shift. */
10564 if (outer_op == UNKNOWN
10565 && orig_code == code && orig_count == count
10566 && varop == orig_varop
10567 && shift_mode == GET_MODE (varop))
10568 return NULL_RTX;
10570 /* Make a SUBREG if necessary. If we can't make it, fail. */
10571 varop = gen_lowpart (shift_mode, varop);
10572 if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
10573 return NULL_RTX;
10575 /* If we have an outer operation and we just made a shift, it is
10576 possible that we could have simplified the shift were it not
10577 for the outer operation. So try to do the simplification
10578 recursively. */
10580 if (outer_op != UNKNOWN)
10581 x = simplify_shift_const_1 (code, shift_mode, varop, count);
10582 else
10583 x = NULL_RTX;
10585 if (x == NULL_RTX)
10586 x = simplify_gen_binary (code, shift_mode, varop, GEN_INT (count));
10588 /* If we were doing an LSHIFTRT in a wider mode than it was originally,
10589 turn off all the bits that the shift would have turned off. */
10590 if (orig_code == LSHIFTRT && result_mode != shift_mode)
10591 x = simplify_and_const_int (NULL_RTX, shift_mode, x,
10592 GET_MODE_MASK (result_mode) >> orig_count);
10594 /* Do the remainder of the processing in RESULT_MODE. */
10595 x = gen_lowpart_or_truncate (result_mode, x);
10597 /* If COMPLEMENT_P is set, we have to complement X before doing the outer
10598 operation. */
10599 if (complement_p)
10600 x = simplify_gen_unary (NOT, result_mode, x, result_mode);
10602 if (outer_op != UNKNOWN)
10604 if (GET_RTX_CLASS (outer_op) != RTX_UNARY
10605 && GET_MODE_PRECISION (result_mode) < HOST_BITS_PER_WIDE_INT)
10606 outer_const = trunc_int_for_mode (outer_const, result_mode);
10608 if (outer_op == AND)
10609 x = simplify_and_const_int (NULL_RTX, result_mode, x, outer_const);
10610 else if (outer_op == SET)
10612 /* This means that we have determined that the result is
10613 equivalent to a constant. This should be rare. */
10614 if (!side_effects_p (x))
10615 x = GEN_INT (outer_const);
10617 else if (GET_RTX_CLASS (outer_op) == RTX_UNARY)
10618 x = simplify_gen_unary (outer_op, result_mode, x, result_mode);
10619 else
10620 x = simplify_gen_binary (outer_op, result_mode, x,
10621 GEN_INT (outer_const));
10624 return x;
10627 /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
10628 The result of the shift is RESULT_MODE. If we cannot simplify it,
10629 return X or, if it is NULL, synthesize the expression with
10630 simplify_gen_binary. Otherwise, return a simplified value.
10632 The shift is normally computed in the widest mode we find in VAROP, as
10633 long as it isn't a different number of words than RESULT_MODE. Exceptions
10634 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10636 static rtx
10637 simplify_shift_const (rtx x, enum rtx_code code, enum machine_mode result_mode,
10638 rtx varop, int count)
10640 rtx tem = simplify_shift_const_1 (code, result_mode, varop, count);
10641 if (tem)
10642 return tem;
10644 if (!x)
10645 x = simplify_gen_binary (code, GET_MODE (varop), varop, GEN_INT (count));
10646 if (GET_MODE (x) != result_mode)
10647 x = gen_lowpart (result_mode, x);
10648 return x;
10652 /* Like recog, but we receive the address of a pointer to a new pattern.
10653 We try to match the rtx that the pointer points to.
10654 If that fails, we may try to modify or replace the pattern,
10655 storing the replacement into the same pointer object.
10657 Modifications include deletion or addition of CLOBBERs.
10659 PNOTES is a pointer to a location where any REG_UNUSED notes added for
10660 the CLOBBERs are placed.
10662 The value is the final insn code from the pattern ultimately matched,
10663 or -1. */
10665 static int
10666 recog_for_combine (rtx *pnewpat, rtx insn, rtx *pnotes)
10668 rtx pat = *pnewpat;
10669 int insn_code_number;
10670 int num_clobbers_to_add = 0;
10671 int i;
10672 rtx notes = 0;
10673 rtx old_notes, old_pat;
10675 /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
10676 we use to indicate that something didn't match. If we find such a
10677 thing, force rejection. */
10678 if (GET_CODE (pat) == PARALLEL)
10679 for (i = XVECLEN (pat, 0) - 1; i >= 0; i--)
10680 if (GET_CODE (XVECEXP (pat, 0, i)) == CLOBBER
10681 && XEXP (XVECEXP (pat, 0, i), 0) == const0_rtx)
10682 return -1;
10684 old_pat = PATTERN (insn);
10685 old_notes = REG_NOTES (insn);
10686 PATTERN (insn) = pat;
10687 REG_NOTES (insn) = 0;
10689 insn_code_number = recog (pat, insn, &num_clobbers_to_add);
10690 if (dump_file && (dump_flags & TDF_DETAILS))
10692 if (insn_code_number < 0)
10693 fputs ("Failed to match this instruction:\n", dump_file);
10694 else
10695 fputs ("Successfully matched this instruction:\n", dump_file);
10696 print_rtl_single (dump_file, pat);
10699 /* If it isn't, there is the possibility that we previously had an insn
10700 that clobbered some register as a side effect, but the combined
10701 insn doesn't need to do that. So try once more without the clobbers
10702 unless this represents an ASM insn. */
10704 if (insn_code_number < 0 && ! check_asm_operands (pat)
10705 && GET_CODE (pat) == PARALLEL)
10707 int pos;
10709 for (pos = 0, i = 0; i < XVECLEN (pat, 0); i++)
10710 if (GET_CODE (XVECEXP (pat, 0, i)) != CLOBBER)
10712 if (i != pos)
10713 SUBST (XVECEXP (pat, 0, pos), XVECEXP (pat, 0, i));
10714 pos++;
10717 SUBST_INT (XVECLEN (pat, 0), pos);
10719 if (pos == 1)
10720 pat = XVECEXP (pat, 0, 0);
10722 PATTERN (insn) = pat;
10723 insn_code_number = recog (pat, insn, &num_clobbers_to_add);
10724 if (dump_file && (dump_flags & TDF_DETAILS))
10726 if (insn_code_number < 0)
10727 fputs ("Failed to match this instruction:\n", dump_file);
10728 else
10729 fputs ("Successfully matched this instruction:\n", dump_file);
10730 print_rtl_single (dump_file, pat);
10733 PATTERN (insn) = old_pat;
10734 REG_NOTES (insn) = old_notes;
10736 /* Recognize all noop sets, these will be killed by followup pass. */
10737 if (insn_code_number < 0 && GET_CODE (pat) == SET && set_noop_p (pat))
10738 insn_code_number = NOOP_MOVE_INSN_CODE, num_clobbers_to_add = 0;
10740 /* If we had any clobbers to add, make a new pattern than contains
10741 them. Then check to make sure that all of them are dead. */
10742 if (num_clobbers_to_add)
10744 rtx newpat = gen_rtx_PARALLEL (VOIDmode,
10745 rtvec_alloc (GET_CODE (pat) == PARALLEL
10746 ? (XVECLEN (pat, 0)
10747 + num_clobbers_to_add)
10748 : num_clobbers_to_add + 1));
10750 if (GET_CODE (pat) == PARALLEL)
10751 for (i = 0; i < XVECLEN (pat, 0); i++)
10752 XVECEXP (newpat, 0, i) = XVECEXP (pat, 0, i);
10753 else
10754 XVECEXP (newpat, 0, 0) = pat;
10756 add_clobbers (newpat, insn_code_number);
10758 for (i = XVECLEN (newpat, 0) - num_clobbers_to_add;
10759 i < XVECLEN (newpat, 0); i++)
10761 if (REG_P (XEXP (XVECEXP (newpat, 0, i), 0))
10762 && ! reg_dead_at_p (XEXP (XVECEXP (newpat, 0, i), 0), insn))
10763 return -1;
10764 if (GET_CODE (XEXP (XVECEXP (newpat, 0, i), 0)) != SCRATCH)
10766 gcc_assert (REG_P (XEXP (XVECEXP (newpat, 0, i), 0)));
10767 notes = alloc_reg_note (REG_UNUSED,
10768 XEXP (XVECEXP (newpat, 0, i), 0), notes);
10771 pat = newpat;
10774 *pnewpat = pat;
10775 *pnotes = notes;
10777 return insn_code_number;
10780 /* Like gen_lowpart_general but for use by combine. In combine it
10781 is not possible to create any new pseudoregs. However, it is
10782 safe to create invalid memory addresses, because combine will
10783 try to recognize them and all they will do is make the combine
10784 attempt fail.
10786 If for some reason this cannot do its job, an rtx
10787 (clobber (const_int 0)) is returned.
10788 An insn containing that will not be recognized. */
10790 static rtx
10791 gen_lowpart_for_combine (enum machine_mode omode, rtx x)
10793 enum machine_mode imode = GET_MODE (x);
10794 unsigned int osize = GET_MODE_SIZE (omode);
10795 unsigned int isize = GET_MODE_SIZE (imode);
10796 rtx result;
10798 if (omode == imode)
10799 return x;
10801 /* Return identity if this is a CONST or symbolic reference. */
10802 if (omode == Pmode
10803 && (GET_CODE (x) == CONST
10804 || GET_CODE (x) == SYMBOL_REF
10805 || GET_CODE (x) == LABEL_REF))
10806 return x;
10808 /* We can only support MODE being wider than a word if X is a
10809 constant integer or has a mode the same size. */
10810 if (GET_MODE_SIZE (omode) > UNITS_PER_WORD
10811 && ! ((imode == VOIDmode
10812 && (CONST_INT_P (x)
10813 || GET_CODE (x) == CONST_DOUBLE))
10814 || isize == osize))
10815 goto fail;
10817 /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
10818 won't know what to do. So we will strip off the SUBREG here and
10819 process normally. */
10820 if (GET_CODE (x) == SUBREG && MEM_P (SUBREG_REG (x)))
10822 x = SUBREG_REG (x);
10824 /* For use in case we fall down into the address adjustments
10825 further below, we need to adjust the known mode and size of
10826 x; imode and isize, since we just adjusted x. */
10827 imode = GET_MODE (x);
10829 if (imode == omode)
10830 return x;
10832 isize = GET_MODE_SIZE (imode);
10835 result = gen_lowpart_common (omode, x);
10837 if (result)
10838 return result;
10840 if (MEM_P (x))
10842 int offset = 0;
10844 /* Refuse to work on a volatile memory ref or one with a mode-dependent
10845 address. */
10846 if (MEM_VOLATILE_P (x) || mode_dependent_address_p (XEXP (x, 0)))
10847 goto fail;
10849 /* If we want to refer to something bigger than the original memref,
10850 generate a paradoxical subreg instead. That will force a reload
10851 of the original memref X. */
10852 if (isize < osize)
10853 return gen_rtx_SUBREG (omode, x, 0);
10855 if (WORDS_BIG_ENDIAN)
10856 offset = MAX (isize, UNITS_PER_WORD) - MAX (osize, UNITS_PER_WORD);
10858 /* Adjust the address so that the address-after-the-data is
10859 unchanged. */
10860 if (BYTES_BIG_ENDIAN)
10861 offset -= MIN (UNITS_PER_WORD, osize) - MIN (UNITS_PER_WORD, isize);
10863 return adjust_address_nv (x, omode, offset);
10866 /* If X is a comparison operator, rewrite it in a new mode. This
10867 probably won't match, but may allow further simplifications. */
10868 else if (COMPARISON_P (x))
10869 return gen_rtx_fmt_ee (GET_CODE (x), omode, XEXP (x, 0), XEXP (x, 1));
10871 /* If we couldn't simplify X any other way, just enclose it in a
10872 SUBREG. Normally, this SUBREG won't match, but some patterns may
10873 include an explicit SUBREG or we may simplify it further in combine. */
10874 else
10876 int offset = 0;
10877 rtx res;
10879 offset = subreg_lowpart_offset (omode, imode);
10880 if (imode == VOIDmode)
10882 imode = int_mode_for_mode (omode);
10883 x = gen_lowpart_common (imode, x);
10884 if (x == NULL)
10885 goto fail;
10887 res = simplify_gen_subreg (omode, x, imode, offset);
10888 if (res)
10889 return res;
10892 fail:
10893 return gen_rtx_CLOBBER (omode, const0_rtx);
10896 /* Try to simplify a comparison between OP0 and a constant OP1,
10897 where CODE is the comparison code that will be tested, into a
10898 (CODE OP0 const0_rtx) form.
10900 The result is a possibly different comparison code to use.
10901 *POP1 may be updated. */
10903 static enum rtx_code
10904 simplify_compare_const (enum rtx_code code, rtx op0, rtx *pop1)
10906 enum machine_mode mode = GET_MODE (op0);
10907 unsigned int mode_width = GET_MODE_PRECISION (mode);
10908 HOST_WIDE_INT const_op = INTVAL (*pop1);
10910 /* Get the constant we are comparing against and turn off all bits
10911 not on in our mode. */
10912 if (mode != VOIDmode)
10913 const_op = trunc_int_for_mode (const_op, mode);
10915 /* If we are comparing against a constant power of two and the value
10916 being compared can only have that single bit nonzero (e.g., it was
10917 `and'ed with that bit), we can replace this with a comparison
10918 with zero. */
10919 if (const_op
10920 && (code == EQ || code == NE || code == GE || code == GEU
10921 || code == LT || code == LTU)
10922 && mode_width <= HOST_BITS_PER_WIDE_INT
10923 && exact_log2 (const_op) >= 0
10924 && nonzero_bits (op0, mode) == (unsigned HOST_WIDE_INT) const_op)
10926 code = (code == EQ || code == GE || code == GEU ? NE : EQ);
10927 const_op = 0;
10930 /* Similarly, if we are comparing a value known to be either -1 or
10931 0 with -1, change it to the opposite comparison against zero. */
10932 if (const_op == -1
10933 && (code == EQ || code == NE || code == GT || code == LE
10934 || code == GEU || code == LTU)
10935 && num_sign_bit_copies (op0, mode) == mode_width)
10937 code = (code == EQ || code == LE || code == GEU ? NE : EQ);
10938 const_op = 0;
10941 /* Do some canonicalizations based on the comparison code. We prefer
10942 comparisons against zero and then prefer equality comparisons.
10943 If we can reduce the size of a constant, we will do that too. */
10944 switch (code)
10946 case LT:
10947 /* < C is equivalent to <= (C - 1) */
10948 if (const_op > 0)
10950 const_op -= 1;
10951 code = LE;
10952 /* ... fall through to LE case below. */
10954 else
10955 break;
10957 case LE:
10958 /* <= C is equivalent to < (C + 1); we do this for C < 0 */
10959 if (const_op < 0)
10961 const_op += 1;
10962 code = LT;
10965 /* If we are doing a <= 0 comparison on a value known to have
10966 a zero sign bit, we can replace this with == 0. */
10967 else if (const_op == 0
10968 && mode_width <= HOST_BITS_PER_WIDE_INT
10969 && (nonzero_bits (op0, mode)
10970 & ((unsigned HOST_WIDE_INT) 1 << (mode_width - 1)))
10971 == 0)
10972 code = EQ;
10973 break;
10975 case GE:
10976 /* >= C is equivalent to > (C - 1). */
10977 if (const_op > 0)
10979 const_op -= 1;
10980 code = GT;
10981 /* ... fall through to GT below. */
10983 else
10984 break;
10986 case GT:
10987 /* > C is equivalent to >= (C + 1); we do this for C < 0. */
10988 if (const_op < 0)
10990 const_op += 1;
10991 code = GE;
10994 /* If we are doing a > 0 comparison on a value known to have
10995 a zero sign bit, we can replace this with != 0. */
10996 else if (const_op == 0
10997 && mode_width <= HOST_BITS_PER_WIDE_INT
10998 && (nonzero_bits (op0, mode)
10999 & ((unsigned HOST_WIDE_INT) 1 << (mode_width - 1)))
11000 == 0)
11001 code = NE;
11002 break;
11004 case LTU:
11005 /* < C is equivalent to <= (C - 1). */
11006 if (const_op > 0)
11008 const_op -= 1;
11009 code = LEU;
11010 /* ... fall through ... */
11012 /* (unsigned) < 0x80000000 is equivalent to >= 0. */
11013 else if (mode_width <= HOST_BITS_PER_WIDE_INT
11014 && (unsigned HOST_WIDE_INT) const_op
11015 == (unsigned HOST_WIDE_INT) 1 << (mode_width - 1))
11017 const_op = 0;
11018 code = GE;
11019 break;
11021 else
11022 break;
11024 case LEU:
11025 /* unsigned <= 0 is equivalent to == 0 */
11026 if (const_op == 0)
11027 code = EQ;
11028 /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
11029 else if (mode_width <= HOST_BITS_PER_WIDE_INT
11030 && (unsigned HOST_WIDE_INT) const_op
11031 == ((unsigned HOST_WIDE_INT) 1 << (mode_width - 1)) - 1)
11033 const_op = 0;
11034 code = GE;
11036 break;
11038 case GEU:
11039 /* >= C is equivalent to > (C - 1). */
11040 if (const_op > 1)
11042 const_op -= 1;
11043 code = GTU;
11044 /* ... fall through ... */
11047 /* (unsigned) >= 0x80000000 is equivalent to < 0. */
11048 else if (mode_width <= HOST_BITS_PER_WIDE_INT
11049 && (unsigned HOST_WIDE_INT) const_op
11050 == (unsigned HOST_WIDE_INT) 1 << (mode_width - 1))
11052 const_op = 0;
11053 code = LT;
11054 break;
11056 else
11057 break;
11059 case GTU:
11060 /* unsigned > 0 is equivalent to != 0 */
11061 if (const_op == 0)
11062 code = NE;
11063 /* (unsigned) > 0x7fffffff is equivalent to < 0. */
11064 else if (mode_width <= HOST_BITS_PER_WIDE_INT
11065 && (unsigned HOST_WIDE_INT) const_op
11066 == ((unsigned HOST_WIDE_INT) 1 << (mode_width - 1)) - 1)
11068 const_op = 0;
11069 code = LT;
11071 break;
11073 default:
11074 break;
11077 *pop1 = GEN_INT (const_op);
11078 return code;
11081 /* Simplify a comparison between *POP0 and *POP1 where CODE is the
11082 comparison code that will be tested.
11084 The result is a possibly different comparison code to use. *POP0 and
11085 *POP1 may be updated.
11087 It is possible that we might detect that a comparison is either always
11088 true or always false. However, we do not perform general constant
11089 folding in combine, so this knowledge isn't useful. Such tautologies
11090 should have been detected earlier. Hence we ignore all such cases. */
11092 static enum rtx_code
11093 simplify_comparison (enum rtx_code code, rtx *pop0, rtx *pop1)
11095 rtx op0 = *pop0;
11096 rtx op1 = *pop1;
11097 rtx tem, tem1;
11098 int i;
11099 enum machine_mode mode, tmode;
11101 /* Try a few ways of applying the same transformation to both operands. */
11102 while (1)
11104 #ifndef WORD_REGISTER_OPERATIONS
11105 /* The test below this one won't handle SIGN_EXTENDs on these machines,
11106 so check specially. */
11107 if (code != GTU && code != GEU && code != LTU && code != LEU
11108 && GET_CODE (op0) == ASHIFTRT && GET_CODE (op1) == ASHIFTRT
11109 && GET_CODE (XEXP (op0, 0)) == ASHIFT
11110 && GET_CODE (XEXP (op1, 0)) == ASHIFT
11111 && GET_CODE (XEXP (XEXP (op0, 0), 0)) == SUBREG
11112 && GET_CODE (XEXP (XEXP (op1, 0), 0)) == SUBREG
11113 && (GET_MODE (SUBREG_REG (XEXP (XEXP (op0, 0), 0)))
11114 == GET_MODE (SUBREG_REG (XEXP (XEXP (op1, 0), 0))))
11115 && CONST_INT_P (XEXP (op0, 1))
11116 && XEXP (op0, 1) == XEXP (op1, 1)
11117 && XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
11118 && XEXP (op0, 1) == XEXP (XEXP (op1, 0), 1)
11119 && (INTVAL (XEXP (op0, 1))
11120 == (GET_MODE_PRECISION (GET_MODE (op0))
11121 - (GET_MODE_PRECISION
11122 (GET_MODE (SUBREG_REG (XEXP (XEXP (op0, 0), 0))))))))
11124 op0 = SUBREG_REG (XEXP (XEXP (op0, 0), 0));
11125 op1 = SUBREG_REG (XEXP (XEXP (op1, 0), 0));
11127 #endif
11129 /* If both operands are the same constant shift, see if we can ignore the
11130 shift. We can if the shift is a rotate or if the bits shifted out of
11131 this shift are known to be zero for both inputs and if the type of
11132 comparison is compatible with the shift. */
11133 if (GET_CODE (op0) == GET_CODE (op1)
11134 && HWI_COMPUTABLE_MODE_P (GET_MODE(op0))
11135 && ((GET_CODE (op0) == ROTATE && (code == NE || code == EQ))
11136 || ((GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFT)
11137 && (code != GT && code != LT && code != GE && code != LE))
11138 || (GET_CODE (op0) == ASHIFTRT
11139 && (code != GTU && code != LTU
11140 && code != GEU && code != LEU)))
11141 && CONST_INT_P (XEXP (op0, 1))
11142 && INTVAL (XEXP (op0, 1)) >= 0
11143 && INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
11144 && XEXP (op0, 1) == XEXP (op1, 1))
11146 enum machine_mode mode = GET_MODE (op0);
11147 unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
11148 int shift_count = INTVAL (XEXP (op0, 1));
11150 if (GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFTRT)
11151 mask &= (mask >> shift_count) << shift_count;
11152 else if (GET_CODE (op0) == ASHIFT)
11153 mask = (mask & (mask << shift_count)) >> shift_count;
11155 if ((nonzero_bits (XEXP (op0, 0), mode) & ~mask) == 0
11156 && (nonzero_bits (XEXP (op1, 0), mode) & ~mask) == 0)
11157 op0 = XEXP (op0, 0), op1 = XEXP (op1, 0);
11158 else
11159 break;
11162 /* If both operands are AND's of a paradoxical SUBREG by constant, the
11163 SUBREGs are of the same mode, and, in both cases, the AND would
11164 be redundant if the comparison was done in the narrower mode,
11165 do the comparison in the narrower mode (e.g., we are AND'ing with 1
11166 and the operand's possibly nonzero bits are 0xffffff01; in that case
11167 if we only care about QImode, we don't need the AND). This case
11168 occurs if the output mode of an scc insn is not SImode and
11169 STORE_FLAG_VALUE == 1 (e.g., the 386).
11171 Similarly, check for a case where the AND's are ZERO_EXTEND
11172 operations from some narrower mode even though a SUBREG is not
11173 present. */
11175 else if (GET_CODE (op0) == AND && GET_CODE (op1) == AND
11176 && CONST_INT_P (XEXP (op0, 1))
11177 && CONST_INT_P (XEXP (op1, 1)))
11179 rtx inner_op0 = XEXP (op0, 0);
11180 rtx inner_op1 = XEXP (op1, 0);
11181 HOST_WIDE_INT c0 = INTVAL (XEXP (op0, 1));
11182 HOST_WIDE_INT c1 = INTVAL (XEXP (op1, 1));
11183 int changed = 0;
11185 if (paradoxical_subreg_p (inner_op0)
11186 && GET_CODE (inner_op1) == SUBREG
11187 && (GET_MODE (SUBREG_REG (inner_op0))
11188 == GET_MODE (SUBREG_REG (inner_op1)))
11189 && (GET_MODE_PRECISION (GET_MODE (SUBREG_REG (inner_op0)))
11190 <= HOST_BITS_PER_WIDE_INT)
11191 && (0 == ((~c0) & nonzero_bits (SUBREG_REG (inner_op0),
11192 GET_MODE (SUBREG_REG (inner_op0)))))
11193 && (0 == ((~c1) & nonzero_bits (SUBREG_REG (inner_op1),
11194 GET_MODE (SUBREG_REG (inner_op1))))))
11196 op0 = SUBREG_REG (inner_op0);
11197 op1 = SUBREG_REG (inner_op1);
11199 /* The resulting comparison is always unsigned since we masked
11200 off the original sign bit. */
11201 code = unsigned_condition (code);
11203 changed = 1;
11206 else if (c0 == c1)
11207 for (tmode = GET_CLASS_NARROWEST_MODE
11208 (GET_MODE_CLASS (GET_MODE (op0)));
11209 tmode != GET_MODE (op0); tmode = GET_MODE_WIDER_MODE (tmode))
11210 if ((unsigned HOST_WIDE_INT) c0 == GET_MODE_MASK (tmode))
11212 op0 = gen_lowpart (tmode, inner_op0);
11213 op1 = gen_lowpart (tmode, inner_op1);
11214 code = unsigned_condition (code);
11215 changed = 1;
11216 break;
11219 if (! changed)
11220 break;
11223 /* If both operands are NOT, we can strip off the outer operation
11224 and adjust the comparison code for swapped operands; similarly for
11225 NEG, except that this must be an equality comparison. */
11226 else if ((GET_CODE (op0) == NOT && GET_CODE (op1) == NOT)
11227 || (GET_CODE (op0) == NEG && GET_CODE (op1) == NEG
11228 && (code == EQ || code == NE)))
11229 op0 = XEXP (op0, 0), op1 = XEXP (op1, 0), code = swap_condition (code);
11231 else
11232 break;
11235 /* If the first operand is a constant, swap the operands and adjust the
11236 comparison code appropriately, but don't do this if the second operand
11237 is already a constant integer. */
11238 if (swap_commutative_operands_p (op0, op1))
11240 tem = op0, op0 = op1, op1 = tem;
11241 code = swap_condition (code);
11244 /* We now enter a loop during which we will try to simplify the comparison.
11245 For the most part, we only are concerned with comparisons with zero,
11246 but some things may really be comparisons with zero but not start
11247 out looking that way. */
11249 while (CONST_INT_P (op1))
11251 enum machine_mode mode = GET_MODE (op0);
11252 unsigned int mode_width = GET_MODE_PRECISION (mode);
11253 unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
11254 int equality_comparison_p;
11255 int sign_bit_comparison_p;
11256 int unsigned_comparison_p;
11257 HOST_WIDE_INT const_op;
11259 /* We only want to handle integral modes. This catches VOIDmode,
11260 CCmode, and the floating-point modes. An exception is that we
11261 can handle VOIDmode if OP0 is a COMPARE or a comparison
11262 operation. */
11264 if (GET_MODE_CLASS (mode) != MODE_INT
11265 && ! (mode == VOIDmode
11266 && (GET_CODE (op0) == COMPARE || COMPARISON_P (op0))))
11267 break;
11269 /* Try to simplify the compare to constant, possibly changing the
11270 comparison op, and/or changing op1 to zero. */
11271 code = simplify_compare_const (code, op0, &op1);
11272 const_op = INTVAL (op1);
11274 /* Compute some predicates to simplify code below. */
11276 equality_comparison_p = (code == EQ || code == NE);
11277 sign_bit_comparison_p = ((code == LT || code == GE) && const_op == 0);
11278 unsigned_comparison_p = (code == LTU || code == LEU || code == GTU
11279 || code == GEU);
11281 /* If this is a sign bit comparison and we can do arithmetic in
11282 MODE, say that we will only be needing the sign bit of OP0. */
11283 if (sign_bit_comparison_p && HWI_COMPUTABLE_MODE_P (mode))
11284 op0 = force_to_mode (op0, mode,
11285 (unsigned HOST_WIDE_INT) 1
11286 << (GET_MODE_PRECISION (mode) - 1),
11289 /* Now try cases based on the opcode of OP0. If none of the cases
11290 does a "continue", we exit this loop immediately after the
11291 switch. */
11293 switch (GET_CODE (op0))
11295 case ZERO_EXTRACT:
11296 /* If we are extracting a single bit from a variable position in
11297 a constant that has only a single bit set and are comparing it
11298 with zero, we can convert this into an equality comparison
11299 between the position and the location of the single bit. */
11300 /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
11301 have already reduced the shift count modulo the word size. */
11302 if (!SHIFT_COUNT_TRUNCATED
11303 && CONST_INT_P (XEXP (op0, 0))
11304 && XEXP (op0, 1) == const1_rtx
11305 && equality_comparison_p && const_op == 0
11306 && (i = exact_log2 (UINTVAL (XEXP (op0, 0)))) >= 0)
11308 if (BITS_BIG_ENDIAN)
11310 enum machine_mode new_mode
11311 = mode_for_extraction (EP_extzv, 1);
11312 if (new_mode == MAX_MACHINE_MODE)
11313 i = BITS_PER_WORD - 1 - i;
11314 else
11316 mode = new_mode;
11317 i = (GET_MODE_PRECISION (mode) - 1 - i);
11321 op0 = XEXP (op0, 2);
11322 op1 = GEN_INT (i);
11323 const_op = i;
11325 /* Result is nonzero iff shift count is equal to I. */
11326 code = reverse_condition (code);
11327 continue;
11330 /* ... fall through ... */
11332 case SIGN_EXTRACT:
11333 tem = expand_compound_operation (op0);
11334 if (tem != op0)
11336 op0 = tem;
11337 continue;
11339 break;
11341 case NOT:
11342 /* If testing for equality, we can take the NOT of the constant. */
11343 if (equality_comparison_p
11344 && (tem = simplify_unary_operation (NOT, mode, op1, mode)) != 0)
11346 op0 = XEXP (op0, 0);
11347 op1 = tem;
11348 continue;
11351 /* If just looking at the sign bit, reverse the sense of the
11352 comparison. */
11353 if (sign_bit_comparison_p)
11355 op0 = XEXP (op0, 0);
11356 code = (code == GE ? LT : GE);
11357 continue;
11359 break;
11361 case NEG:
11362 /* If testing for equality, we can take the NEG of the constant. */
11363 if (equality_comparison_p
11364 && (tem = simplify_unary_operation (NEG, mode, op1, mode)) != 0)
11366 op0 = XEXP (op0, 0);
11367 op1 = tem;
11368 continue;
11371 /* The remaining cases only apply to comparisons with zero. */
11372 if (const_op != 0)
11373 break;
11375 /* When X is ABS or is known positive,
11376 (neg X) is < 0 if and only if X != 0. */
11378 if (sign_bit_comparison_p
11379 && (GET_CODE (XEXP (op0, 0)) == ABS
11380 || (mode_width <= HOST_BITS_PER_WIDE_INT
11381 && (nonzero_bits (XEXP (op0, 0), mode)
11382 & ((unsigned HOST_WIDE_INT) 1 << (mode_width - 1)))
11383 == 0)))
11385 op0 = XEXP (op0, 0);
11386 code = (code == LT ? NE : EQ);
11387 continue;
11390 /* If we have NEG of something whose two high-order bits are the
11391 same, we know that "(-a) < 0" is equivalent to "a > 0". */
11392 if (num_sign_bit_copies (op0, mode) >= 2)
11394 op0 = XEXP (op0, 0);
11395 code = swap_condition (code);
11396 continue;
11398 break;
11400 case ROTATE:
11401 /* If we are testing equality and our count is a constant, we
11402 can perform the inverse operation on our RHS. */
11403 if (equality_comparison_p && CONST_INT_P (XEXP (op0, 1))
11404 && (tem = simplify_binary_operation (ROTATERT, mode,
11405 op1, XEXP (op0, 1))) != 0)
11407 op0 = XEXP (op0, 0);
11408 op1 = tem;
11409 continue;
11412 /* If we are doing a < 0 or >= 0 comparison, it means we are testing
11413 a particular bit. Convert it to an AND of a constant of that
11414 bit. This will be converted into a ZERO_EXTRACT. */
11415 if (const_op == 0 && sign_bit_comparison_p
11416 && CONST_INT_P (XEXP (op0, 1))
11417 && mode_width <= HOST_BITS_PER_WIDE_INT)
11419 op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
11420 ((unsigned HOST_WIDE_INT) 1
11421 << (mode_width - 1
11422 - INTVAL (XEXP (op0, 1)))));
11423 code = (code == LT ? NE : EQ);
11424 continue;
11427 /* Fall through. */
11429 case ABS:
11430 /* ABS is ignorable inside an equality comparison with zero. */
11431 if (const_op == 0 && equality_comparison_p)
11433 op0 = XEXP (op0, 0);
11434 continue;
11436 break;
11438 case SIGN_EXTEND:
11439 /* Can simplify (compare (zero/sign_extend FOO) CONST) to
11440 (compare FOO CONST) if CONST fits in FOO's mode and we
11441 are either testing inequality or have an unsigned
11442 comparison with ZERO_EXTEND or a signed comparison with
11443 SIGN_EXTEND. But don't do it if we don't have a compare
11444 insn of the given mode, since we'd have to revert it
11445 later on, and then we wouldn't know whether to sign- or
11446 zero-extend. */
11447 mode = GET_MODE (XEXP (op0, 0));
11448 if (GET_MODE_CLASS (mode) == MODE_INT
11449 && ! unsigned_comparison_p
11450 && HWI_COMPUTABLE_MODE_P (mode)
11451 && trunc_int_for_mode (const_op, mode) == const_op
11452 && have_insn_for (COMPARE, mode))
11454 op0 = XEXP (op0, 0);
11455 continue;
11457 break;
11459 case SUBREG:
11460 /* Check for the case where we are comparing A - C1 with C2, that is
11462 (subreg:MODE (plus (A) (-C1))) op (C2)
11464 with C1 a constant, and try to lift the SUBREG, i.e. to do the
11465 comparison in the wider mode. One of the following two conditions
11466 must be true in order for this to be valid:
11468 1. The mode extension results in the same bit pattern being added
11469 on both sides and the comparison is equality or unsigned. As
11470 C2 has been truncated to fit in MODE, the pattern can only be
11471 all 0s or all 1s.
11473 2. The mode extension results in the sign bit being copied on
11474 each side.
11476 The difficulty here is that we have predicates for A but not for
11477 (A - C1) so we need to check that C1 is within proper bounds so
11478 as to perturbate A as little as possible. */
11480 if (mode_width <= HOST_BITS_PER_WIDE_INT
11481 && subreg_lowpart_p (op0)
11482 && GET_MODE_PRECISION (GET_MODE (SUBREG_REG (op0))) > mode_width
11483 && GET_CODE (SUBREG_REG (op0)) == PLUS
11484 && CONST_INT_P (XEXP (SUBREG_REG (op0), 1)))
11486 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
11487 rtx a = XEXP (SUBREG_REG (op0), 0);
11488 HOST_WIDE_INT c1 = -INTVAL (XEXP (SUBREG_REG (op0), 1));
11490 if ((c1 > 0
11491 && (unsigned HOST_WIDE_INT) c1
11492 < (unsigned HOST_WIDE_INT) 1 << (mode_width - 1)
11493 && (equality_comparison_p || unsigned_comparison_p)
11494 /* (A - C1) zero-extends if it is positive and sign-extends
11495 if it is negative, C2 both zero- and sign-extends. */
11496 && ((0 == (nonzero_bits (a, inner_mode)
11497 & ~GET_MODE_MASK (mode))
11498 && const_op >= 0)
11499 /* (A - C1) sign-extends if it is positive and 1-extends
11500 if it is negative, C2 both sign- and 1-extends. */
11501 || (num_sign_bit_copies (a, inner_mode)
11502 > (unsigned int) (GET_MODE_PRECISION (inner_mode)
11503 - mode_width)
11504 && const_op < 0)))
11505 || ((unsigned HOST_WIDE_INT) c1
11506 < (unsigned HOST_WIDE_INT) 1 << (mode_width - 2)
11507 /* (A - C1) always sign-extends, like C2. */
11508 && num_sign_bit_copies (a, inner_mode)
11509 > (unsigned int) (GET_MODE_PRECISION (inner_mode)
11510 - (mode_width - 1))))
11512 op0 = SUBREG_REG (op0);
11513 continue;
11517 /* If the inner mode is narrower and we are extracting the low part,
11518 we can treat the SUBREG as if it were a ZERO_EXTEND. */
11519 if (subreg_lowpart_p (op0)
11520 && GET_MODE_PRECISION (GET_MODE (SUBREG_REG (op0))) < mode_width)
11521 /* Fall through */ ;
11522 else
11523 break;
11525 /* ... fall through ... */
11527 case ZERO_EXTEND:
11528 mode = GET_MODE (XEXP (op0, 0));
11529 if (GET_MODE_CLASS (mode) == MODE_INT
11530 && (unsigned_comparison_p || equality_comparison_p)
11531 && HWI_COMPUTABLE_MODE_P (mode)
11532 && (unsigned HOST_WIDE_INT) const_op <= GET_MODE_MASK (mode)
11533 && const_op >= 0
11534 && have_insn_for (COMPARE, mode))
11536 op0 = XEXP (op0, 0);
11537 continue;
11539 break;
11541 case PLUS:
11542 /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
11543 this for equality comparisons due to pathological cases involving
11544 overflows. */
11545 if (equality_comparison_p
11546 && 0 != (tem = simplify_binary_operation (MINUS, mode,
11547 op1, XEXP (op0, 1))))
11549 op0 = XEXP (op0, 0);
11550 op1 = tem;
11551 continue;
11554 /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
11555 if (const_op == 0 && XEXP (op0, 1) == constm1_rtx
11556 && GET_CODE (XEXP (op0, 0)) == ABS && sign_bit_comparison_p)
11558 op0 = XEXP (XEXP (op0, 0), 0);
11559 code = (code == LT ? EQ : NE);
11560 continue;
11562 break;
11564 case MINUS:
11565 /* We used to optimize signed comparisons against zero, but that
11566 was incorrect. Unsigned comparisons against zero (GTU, LEU)
11567 arrive here as equality comparisons, or (GEU, LTU) are
11568 optimized away. No need to special-case them. */
11570 /* (eq (minus A B) C) -> (eq A (plus B C)) or
11571 (eq B (minus A C)), whichever simplifies. We can only do
11572 this for equality comparisons due to pathological cases involving
11573 overflows. */
11574 if (equality_comparison_p
11575 && 0 != (tem = simplify_binary_operation (PLUS, mode,
11576 XEXP (op0, 1), op1)))
11578 op0 = XEXP (op0, 0);
11579 op1 = tem;
11580 continue;
11583 if (equality_comparison_p
11584 && 0 != (tem = simplify_binary_operation (MINUS, mode,
11585 XEXP (op0, 0), op1)))
11587 op0 = XEXP (op0, 1);
11588 op1 = tem;
11589 continue;
11592 /* The sign bit of (minus (ashiftrt X C) X), where C is the number
11593 of bits in X minus 1, is one iff X > 0. */
11594 if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == ASHIFTRT
11595 && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
11596 && UINTVAL (XEXP (XEXP (op0, 0), 1)) == mode_width - 1
11597 && rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
11599 op0 = XEXP (op0, 1);
11600 code = (code == GE ? LE : GT);
11601 continue;
11603 break;
11605 case XOR:
11606 /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
11607 if C is zero or B is a constant. */
11608 if (equality_comparison_p
11609 && 0 != (tem = simplify_binary_operation (XOR, mode,
11610 XEXP (op0, 1), op1)))
11612 op0 = XEXP (op0, 0);
11613 op1 = tem;
11614 continue;
11616 break;
11618 case EQ: case NE:
11619 case UNEQ: case LTGT:
11620 case LT: case LTU: case UNLT: case LE: case LEU: case UNLE:
11621 case GT: case GTU: case UNGT: case GE: case GEU: case UNGE:
11622 case UNORDERED: case ORDERED:
11623 /* We can't do anything if OP0 is a condition code value, rather
11624 than an actual data value. */
11625 if (const_op != 0
11626 || CC0_P (XEXP (op0, 0))
11627 || GET_MODE_CLASS (GET_MODE (XEXP (op0, 0))) == MODE_CC)
11628 break;
11630 /* Get the two operands being compared. */
11631 if (GET_CODE (XEXP (op0, 0)) == COMPARE)
11632 tem = XEXP (XEXP (op0, 0), 0), tem1 = XEXP (XEXP (op0, 0), 1);
11633 else
11634 tem = XEXP (op0, 0), tem1 = XEXP (op0, 1);
11636 /* Check for the cases where we simply want the result of the
11637 earlier test or the opposite of that result. */
11638 if (code == NE || code == EQ
11639 || (val_signbit_known_set_p (GET_MODE (op0), STORE_FLAG_VALUE)
11640 && (code == LT || code == GE)))
11642 enum rtx_code new_code;
11643 if (code == LT || code == NE)
11644 new_code = GET_CODE (op0);
11645 else
11646 new_code = reversed_comparison_code (op0, NULL);
11648 if (new_code != UNKNOWN)
11650 code = new_code;
11651 op0 = tem;
11652 op1 = tem1;
11653 continue;
11656 break;
11658 case IOR:
11659 /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
11660 iff X <= 0. */
11661 if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == PLUS
11662 && XEXP (XEXP (op0, 0), 1) == constm1_rtx
11663 && rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
11665 op0 = XEXP (op0, 1);
11666 code = (code == GE ? GT : LE);
11667 continue;
11669 break;
11671 case AND:
11672 /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
11673 will be converted to a ZERO_EXTRACT later. */
11674 if (const_op == 0 && equality_comparison_p
11675 && GET_CODE (XEXP (op0, 0)) == ASHIFT
11676 && XEXP (XEXP (op0, 0), 0) == const1_rtx)
11678 op0 = gen_rtx_LSHIFTRT (mode, XEXP (op0, 1),
11679 XEXP (XEXP (op0, 0), 1));
11680 op0 = simplify_and_const_int (NULL_RTX, mode, op0, 1);
11681 continue;
11684 /* If we are comparing (and (lshiftrt X C1) C2) for equality with
11685 zero and X is a comparison and C1 and C2 describe only bits set
11686 in STORE_FLAG_VALUE, we can compare with X. */
11687 if (const_op == 0 && equality_comparison_p
11688 && mode_width <= HOST_BITS_PER_WIDE_INT
11689 && CONST_INT_P (XEXP (op0, 1))
11690 && GET_CODE (XEXP (op0, 0)) == LSHIFTRT
11691 && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
11692 && INTVAL (XEXP (XEXP (op0, 0), 1)) >= 0
11693 && INTVAL (XEXP (XEXP (op0, 0), 1)) < HOST_BITS_PER_WIDE_INT)
11695 mask = ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
11696 << INTVAL (XEXP (XEXP (op0, 0), 1)));
11697 if ((~STORE_FLAG_VALUE & mask) == 0
11698 && (COMPARISON_P (XEXP (XEXP (op0, 0), 0))
11699 || ((tem = get_last_value (XEXP (XEXP (op0, 0), 0))) != 0
11700 && COMPARISON_P (tem))))
11702 op0 = XEXP (XEXP (op0, 0), 0);
11703 continue;
11707 /* If we are doing an equality comparison of an AND of a bit equal
11708 to the sign bit, replace this with a LT or GE comparison of
11709 the underlying value. */
11710 if (equality_comparison_p
11711 && const_op == 0
11712 && CONST_INT_P (XEXP (op0, 1))
11713 && mode_width <= HOST_BITS_PER_WIDE_INT
11714 && ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
11715 == (unsigned HOST_WIDE_INT) 1 << (mode_width - 1)))
11717 op0 = XEXP (op0, 0);
11718 code = (code == EQ ? GE : LT);
11719 continue;
11722 /* If this AND operation is really a ZERO_EXTEND from a narrower
11723 mode, the constant fits within that mode, and this is either an
11724 equality or unsigned comparison, try to do this comparison in
11725 the narrower mode.
11727 Note that in:
11729 (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
11730 -> (ne:DI (reg:SI 4) (const_int 0))
11732 unless TRULY_NOOP_TRUNCATION allows it or the register is
11733 known to hold a value of the required mode the
11734 transformation is invalid. */
11735 if ((equality_comparison_p || unsigned_comparison_p)
11736 && CONST_INT_P (XEXP (op0, 1))
11737 && (i = exact_log2 ((UINTVAL (XEXP (op0, 1))
11738 & GET_MODE_MASK (mode))
11739 + 1)) >= 0
11740 && const_op >> i == 0
11741 && (tmode = mode_for_size (i, MODE_INT, 1)) != BLKmode
11742 && (TRULY_NOOP_TRUNCATION_MODES_P (tmode, GET_MODE (op0))
11743 || (REG_P (XEXP (op0, 0))
11744 && reg_truncated_to_mode (tmode, XEXP (op0, 0)))))
11746 op0 = gen_lowpart (tmode, XEXP (op0, 0));
11747 continue;
11750 /* If this is (and:M1 (subreg:M2 X 0) (const_int C1)) where C1
11751 fits in both M1 and M2 and the SUBREG is either paradoxical
11752 or represents the low part, permute the SUBREG and the AND
11753 and try again. */
11754 if (GET_CODE (XEXP (op0, 0)) == SUBREG)
11756 unsigned HOST_WIDE_INT c1;
11757 tmode = GET_MODE (SUBREG_REG (XEXP (op0, 0)));
11758 /* Require an integral mode, to avoid creating something like
11759 (AND:SF ...). */
11760 if (SCALAR_INT_MODE_P (tmode)
11761 /* It is unsafe to commute the AND into the SUBREG if the
11762 SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
11763 not defined. As originally written the upper bits
11764 have a defined value due to the AND operation.
11765 However, if we commute the AND inside the SUBREG then
11766 they no longer have defined values and the meaning of
11767 the code has been changed. */
11768 && (0
11769 #ifdef WORD_REGISTER_OPERATIONS
11770 || (mode_width > GET_MODE_PRECISION (tmode)
11771 && mode_width <= BITS_PER_WORD)
11772 #endif
11773 || (mode_width <= GET_MODE_PRECISION (tmode)
11774 && subreg_lowpart_p (XEXP (op0, 0))))
11775 && CONST_INT_P (XEXP (op0, 1))
11776 && mode_width <= HOST_BITS_PER_WIDE_INT
11777 && HWI_COMPUTABLE_MODE_P (tmode)
11778 && ((c1 = INTVAL (XEXP (op0, 1))) & ~mask) == 0
11779 && (c1 & ~GET_MODE_MASK (tmode)) == 0
11780 && c1 != mask
11781 && c1 != GET_MODE_MASK (tmode))
11783 op0 = simplify_gen_binary (AND, tmode,
11784 SUBREG_REG (XEXP (op0, 0)),
11785 gen_int_mode (c1, tmode));
11786 op0 = gen_lowpart (mode, op0);
11787 continue;
11791 /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
11792 if (const_op == 0 && equality_comparison_p
11793 && XEXP (op0, 1) == const1_rtx
11794 && GET_CODE (XEXP (op0, 0)) == NOT)
11796 op0 = simplify_and_const_int (NULL_RTX, mode,
11797 XEXP (XEXP (op0, 0), 0), 1);
11798 code = (code == NE ? EQ : NE);
11799 continue;
11802 /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
11803 (eq (and (lshiftrt X) 1) 0).
11804 Also handle the case where (not X) is expressed using xor. */
11805 if (const_op == 0 && equality_comparison_p
11806 && XEXP (op0, 1) == const1_rtx
11807 && GET_CODE (XEXP (op0, 0)) == LSHIFTRT)
11809 rtx shift_op = XEXP (XEXP (op0, 0), 0);
11810 rtx shift_count = XEXP (XEXP (op0, 0), 1);
11812 if (GET_CODE (shift_op) == NOT
11813 || (GET_CODE (shift_op) == XOR
11814 && CONST_INT_P (XEXP (shift_op, 1))
11815 && CONST_INT_P (shift_count)
11816 && HWI_COMPUTABLE_MODE_P (mode)
11817 && (UINTVAL (XEXP (shift_op, 1))
11818 == (unsigned HOST_WIDE_INT) 1
11819 << INTVAL (shift_count))))
11822 = gen_rtx_LSHIFTRT (mode, XEXP (shift_op, 0), shift_count);
11823 op0 = simplify_and_const_int (NULL_RTX, mode, op0, 1);
11824 code = (code == NE ? EQ : NE);
11825 continue;
11828 break;
11830 case ASHIFT:
11831 /* If we have (compare (ashift FOO N) (const_int C)) and
11832 the high order N bits of FOO (N+1 if an inequality comparison)
11833 are known to be zero, we can do this by comparing FOO with C
11834 shifted right N bits so long as the low-order N bits of C are
11835 zero. */
11836 if (CONST_INT_P (XEXP (op0, 1))
11837 && INTVAL (XEXP (op0, 1)) >= 0
11838 && ((INTVAL (XEXP (op0, 1)) + ! equality_comparison_p)
11839 < HOST_BITS_PER_WIDE_INT)
11840 && (((unsigned HOST_WIDE_INT) const_op
11841 & (((unsigned HOST_WIDE_INT) 1 << INTVAL (XEXP (op0, 1)))
11842 - 1)) == 0)
11843 && mode_width <= HOST_BITS_PER_WIDE_INT
11844 && (nonzero_bits (XEXP (op0, 0), mode)
11845 & ~(mask >> (INTVAL (XEXP (op0, 1))
11846 + ! equality_comparison_p))) == 0)
11848 /* We must perform a logical shift, not an arithmetic one,
11849 as we want the top N bits of C to be zero. */
11850 unsigned HOST_WIDE_INT temp = const_op & GET_MODE_MASK (mode);
11852 temp >>= INTVAL (XEXP (op0, 1));
11853 op1 = gen_int_mode (temp, mode);
11854 op0 = XEXP (op0, 0);
11855 continue;
11858 /* If we are doing a sign bit comparison, it means we are testing
11859 a particular bit. Convert it to the appropriate AND. */
11860 if (sign_bit_comparison_p && CONST_INT_P (XEXP (op0, 1))
11861 && mode_width <= HOST_BITS_PER_WIDE_INT)
11863 op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
11864 ((unsigned HOST_WIDE_INT) 1
11865 << (mode_width - 1
11866 - INTVAL (XEXP (op0, 1)))));
11867 code = (code == LT ? NE : EQ);
11868 continue;
11871 /* If this an equality comparison with zero and we are shifting
11872 the low bit to the sign bit, we can convert this to an AND of the
11873 low-order bit. */
11874 if (const_op == 0 && equality_comparison_p
11875 && CONST_INT_P (XEXP (op0, 1))
11876 && UINTVAL (XEXP (op0, 1)) == mode_width - 1)
11878 op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0), 1);
11879 continue;
11881 break;
11883 case ASHIFTRT:
11884 /* If this is an equality comparison with zero, we can do this
11885 as a logical shift, which might be much simpler. */
11886 if (equality_comparison_p && const_op == 0
11887 && CONST_INT_P (XEXP (op0, 1)))
11889 op0 = simplify_shift_const (NULL_RTX, LSHIFTRT, mode,
11890 XEXP (op0, 0),
11891 INTVAL (XEXP (op0, 1)));
11892 continue;
11895 /* If OP0 is a sign extension and CODE is not an unsigned comparison,
11896 do the comparison in a narrower mode. */
11897 if (! unsigned_comparison_p
11898 && CONST_INT_P (XEXP (op0, 1))
11899 && GET_CODE (XEXP (op0, 0)) == ASHIFT
11900 && XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
11901 && (tmode = mode_for_size (mode_width - INTVAL (XEXP (op0, 1)),
11902 MODE_INT, 1)) != BLKmode
11903 && (((unsigned HOST_WIDE_INT) const_op
11904 + (GET_MODE_MASK (tmode) >> 1) + 1)
11905 <= GET_MODE_MASK (tmode)))
11907 op0 = gen_lowpart (tmode, XEXP (XEXP (op0, 0), 0));
11908 continue;
11911 /* Likewise if OP0 is a PLUS of a sign extension with a
11912 constant, which is usually represented with the PLUS
11913 between the shifts. */
11914 if (! unsigned_comparison_p
11915 && CONST_INT_P (XEXP (op0, 1))
11916 && GET_CODE (XEXP (op0, 0)) == PLUS
11917 && CONST_INT_P (XEXP (XEXP (op0, 0), 1))
11918 && GET_CODE (XEXP (XEXP (op0, 0), 0)) == ASHIFT
11919 && XEXP (op0, 1) == XEXP (XEXP (XEXP (op0, 0), 0), 1)
11920 && (tmode = mode_for_size (mode_width - INTVAL (XEXP (op0, 1)),
11921 MODE_INT, 1)) != BLKmode
11922 && (((unsigned HOST_WIDE_INT) const_op
11923 + (GET_MODE_MASK (tmode) >> 1) + 1)
11924 <= GET_MODE_MASK (tmode)))
11926 rtx inner = XEXP (XEXP (XEXP (op0, 0), 0), 0);
11927 rtx add_const = XEXP (XEXP (op0, 0), 1);
11928 rtx new_const = simplify_gen_binary (ASHIFTRT, GET_MODE (op0),
11929 add_const, XEXP (op0, 1));
11931 op0 = simplify_gen_binary (PLUS, tmode,
11932 gen_lowpart (tmode, inner),
11933 new_const);
11934 continue;
11937 /* ... fall through ... */
11938 case LSHIFTRT:
11939 /* If we have (compare (xshiftrt FOO N) (const_int C)) and
11940 the low order N bits of FOO are known to be zero, we can do this
11941 by comparing FOO with C shifted left N bits so long as no
11942 overflow occurs. Even if the low order N bits of FOO aren't known
11943 to be zero, if the comparison is >= or < we can use the same
11944 optimization and for > or <= by setting all the low
11945 order N bits in the comparison constant. */
11946 if (CONST_INT_P (XEXP (op0, 1))
11947 && INTVAL (XEXP (op0, 1)) > 0
11948 && INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
11949 && mode_width <= HOST_BITS_PER_WIDE_INT
11950 && (((unsigned HOST_WIDE_INT) const_op
11951 + (GET_CODE (op0) != LSHIFTRT
11952 ? ((GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1)) >> 1)
11953 + 1)
11954 : 0))
11955 <= GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1))))
11957 unsigned HOST_WIDE_INT low_bits
11958 = (nonzero_bits (XEXP (op0, 0), mode)
11959 & (((unsigned HOST_WIDE_INT) 1
11960 << INTVAL (XEXP (op0, 1))) - 1));
11961 if (low_bits == 0 || !equality_comparison_p)
11963 /* If the shift was logical, then we must make the condition
11964 unsigned. */
11965 if (GET_CODE (op0) == LSHIFTRT)
11966 code = unsigned_condition (code);
11968 const_op <<= INTVAL (XEXP (op0, 1));
11969 if (low_bits != 0
11970 && (code == GT || code == GTU
11971 || code == LE || code == LEU))
11972 const_op
11973 |= (((HOST_WIDE_INT) 1 << INTVAL (XEXP (op0, 1))) - 1);
11974 op1 = GEN_INT (const_op);
11975 op0 = XEXP (op0, 0);
11976 continue;
11980 /* If we are using this shift to extract just the sign bit, we
11981 can replace this with an LT or GE comparison. */
11982 if (const_op == 0
11983 && (equality_comparison_p || sign_bit_comparison_p)
11984 && CONST_INT_P (XEXP (op0, 1))
11985 && UINTVAL (XEXP (op0, 1)) == mode_width - 1)
11987 op0 = XEXP (op0, 0);
11988 code = (code == NE || code == GT ? LT : GE);
11989 continue;
11991 break;
11993 default:
11994 break;
11997 break;
12000 /* Now make any compound operations involved in this comparison. Then,
12001 check for an outmost SUBREG on OP0 that is not doing anything or is
12002 paradoxical. The latter transformation must only be performed when
12003 it is known that the "extra" bits will be the same in op0 and op1 or
12004 that they don't matter. There are three cases to consider:
12006 1. SUBREG_REG (op0) is a register. In this case the bits are don't
12007 care bits and we can assume they have any convenient value. So
12008 making the transformation is safe.
12010 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not defined.
12011 In this case the upper bits of op0 are undefined. We should not make
12012 the simplification in that case as we do not know the contents of
12013 those bits.
12015 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is defined and not
12016 UNKNOWN. In that case we know those bits are zeros or ones. We must
12017 also be sure that they are the same as the upper bits of op1.
12019 We can never remove a SUBREG for a non-equality comparison because
12020 the sign bit is in a different place in the underlying object. */
12022 op0 = make_compound_operation (op0, op1 == const0_rtx ? COMPARE : SET);
12023 op1 = make_compound_operation (op1, SET);
12025 if (GET_CODE (op0) == SUBREG && subreg_lowpart_p (op0)
12026 && GET_MODE_CLASS (GET_MODE (op0)) == MODE_INT
12027 && GET_MODE_CLASS (GET_MODE (SUBREG_REG (op0))) == MODE_INT
12028 && (code == NE || code == EQ))
12030 if (paradoxical_subreg_p (op0))
12032 /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
12033 implemented. */
12034 if (REG_P (SUBREG_REG (op0)))
12036 op0 = SUBREG_REG (op0);
12037 op1 = gen_lowpart (GET_MODE (op0), op1);
12040 else if ((GET_MODE_PRECISION (GET_MODE (SUBREG_REG (op0)))
12041 <= HOST_BITS_PER_WIDE_INT)
12042 && (nonzero_bits (SUBREG_REG (op0),
12043 GET_MODE (SUBREG_REG (op0)))
12044 & ~GET_MODE_MASK (GET_MODE (op0))) == 0)
12046 tem = gen_lowpart (GET_MODE (SUBREG_REG (op0)), op1);
12048 if ((nonzero_bits (tem, GET_MODE (SUBREG_REG (op0)))
12049 & ~GET_MODE_MASK (GET_MODE (op0))) == 0)
12050 op0 = SUBREG_REG (op0), op1 = tem;
12054 /* We now do the opposite procedure: Some machines don't have compare
12055 insns in all modes. If OP0's mode is an integer mode smaller than a
12056 word and we can't do a compare in that mode, see if there is a larger
12057 mode for which we can do the compare. There are a number of cases in
12058 which we can use the wider mode. */
12060 mode = GET_MODE (op0);
12061 if (mode != VOIDmode && GET_MODE_CLASS (mode) == MODE_INT
12062 && GET_MODE_SIZE (mode) < UNITS_PER_WORD
12063 && ! have_insn_for (COMPARE, mode))
12064 for (tmode = GET_MODE_WIDER_MODE (mode);
12065 (tmode != VOIDmode && HWI_COMPUTABLE_MODE_P (tmode));
12066 tmode = GET_MODE_WIDER_MODE (tmode))
12067 if (have_insn_for (COMPARE, tmode))
12069 int zero_extended;
12071 /* If this is a test for negative, we can make an explicit
12072 test of the sign bit. Test this first so we can use
12073 a paradoxical subreg to extend OP0. */
12075 if (op1 == const0_rtx && (code == LT || code == GE)
12076 && HWI_COMPUTABLE_MODE_P (mode))
12078 op0 = simplify_gen_binary (AND, tmode,
12079 gen_lowpart (tmode, op0),
12080 GEN_INT ((unsigned HOST_WIDE_INT) 1
12081 << (GET_MODE_BITSIZE (mode)
12082 - 1)));
12083 code = (code == LT) ? NE : EQ;
12084 break;
12087 /* If the only nonzero bits in OP0 and OP1 are those in the
12088 narrower mode and this is an equality or unsigned comparison,
12089 we can use the wider mode. Similarly for sign-extended
12090 values, in which case it is true for all comparisons. */
12091 zero_extended = ((code == EQ || code == NE
12092 || code == GEU || code == GTU
12093 || code == LEU || code == LTU)
12094 && (nonzero_bits (op0, tmode)
12095 & ~GET_MODE_MASK (mode)) == 0
12096 && ((CONST_INT_P (op1)
12097 || (nonzero_bits (op1, tmode)
12098 & ~GET_MODE_MASK (mode)) == 0)));
12100 if (zero_extended
12101 || ((num_sign_bit_copies (op0, tmode)
12102 > (unsigned int) (GET_MODE_PRECISION (tmode)
12103 - GET_MODE_PRECISION (mode)))
12104 && (num_sign_bit_copies (op1, tmode)
12105 > (unsigned int) (GET_MODE_PRECISION (tmode)
12106 - GET_MODE_PRECISION (mode)))))
12108 /* If OP0 is an AND and we don't have an AND in MODE either,
12109 make a new AND in the proper mode. */
12110 if (GET_CODE (op0) == AND
12111 && !have_insn_for (AND, mode))
12112 op0 = simplify_gen_binary (AND, tmode,
12113 gen_lowpart (tmode,
12114 XEXP (op0, 0)),
12115 gen_lowpart (tmode,
12116 XEXP (op0, 1)));
12117 else
12119 if (zero_extended)
12121 op0 = simplify_gen_unary (ZERO_EXTEND, tmode, op0, mode);
12122 op1 = simplify_gen_unary (ZERO_EXTEND, tmode, op1, mode);
12124 else
12126 op0 = simplify_gen_unary (SIGN_EXTEND, tmode, op0, mode);
12127 op1 = simplify_gen_unary (SIGN_EXTEND, tmode, op1, mode);
12129 break;
12134 #ifdef CANONICALIZE_COMPARISON
12135 /* If this machine only supports a subset of valid comparisons, see if we
12136 can convert an unsupported one into a supported one. */
12137 CANONICALIZE_COMPARISON (code, op0, op1);
12138 #endif
12140 *pop0 = op0;
12141 *pop1 = op1;
12143 return code;
12146 /* Utility function for record_value_for_reg. Count number of
12147 rtxs in X. */
12148 static int
12149 count_rtxs (rtx x)
12151 enum rtx_code code = GET_CODE (x);
12152 const char *fmt;
12153 int i, j, ret = 1;
12155 if (GET_RTX_CLASS (code) == '2'
12156 || GET_RTX_CLASS (code) == 'c')
12158 rtx x0 = XEXP (x, 0);
12159 rtx x1 = XEXP (x, 1);
12161 if (x0 == x1)
12162 return 1 + 2 * count_rtxs (x0);
12164 if ((GET_RTX_CLASS (GET_CODE (x1)) == '2'
12165 || GET_RTX_CLASS (GET_CODE (x1)) == 'c')
12166 && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
12167 return 2 + 2 * count_rtxs (x0)
12168 + count_rtxs (x == XEXP (x1, 0)
12169 ? XEXP (x1, 1) : XEXP (x1, 0));
12171 if ((GET_RTX_CLASS (GET_CODE (x0)) == '2'
12172 || GET_RTX_CLASS (GET_CODE (x0)) == 'c')
12173 && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
12174 return 2 + 2 * count_rtxs (x1)
12175 + count_rtxs (x == XEXP (x0, 0)
12176 ? XEXP (x0, 1) : XEXP (x0, 0));
12179 fmt = GET_RTX_FORMAT (code);
12180 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
12181 if (fmt[i] == 'e')
12182 ret += count_rtxs (XEXP (x, i));
12183 else if (fmt[i] == 'E')
12184 for (j = 0; j < XVECLEN (x, i); j++)
12185 ret += count_rtxs (XVECEXP (x, i, j));
12187 return ret;
12190 /* Utility function for following routine. Called when X is part of a value
12191 being stored into last_set_value. Sets last_set_table_tick
12192 for each register mentioned. Similar to mention_regs in cse.c */
12194 static void
12195 update_table_tick (rtx x)
12197 enum rtx_code code = GET_CODE (x);
12198 const char *fmt = GET_RTX_FORMAT (code);
12199 int i, j;
12201 if (code == REG)
12203 unsigned int regno = REGNO (x);
12204 unsigned int endregno = END_REGNO (x);
12205 unsigned int r;
12207 for (r = regno; r < endregno; r++)
12209 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, r);
12210 rsp->last_set_table_tick = label_tick;
12213 return;
12216 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
12217 if (fmt[i] == 'e')
12219 /* Check for identical subexpressions. If x contains
12220 identical subexpression we only have to traverse one of
12221 them. */
12222 if (i == 0 && ARITHMETIC_P (x))
12224 /* Note that at this point x1 has already been
12225 processed. */
12226 rtx x0 = XEXP (x, 0);
12227 rtx x1 = XEXP (x, 1);
12229 /* If x0 and x1 are identical then there is no need to
12230 process x0. */
12231 if (x0 == x1)
12232 break;
12234 /* If x0 is identical to a subexpression of x1 then while
12235 processing x1, x0 has already been processed. Thus we
12236 are done with x. */
12237 if (ARITHMETIC_P (x1)
12238 && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
12239 break;
12241 /* If x1 is identical to a subexpression of x0 then we
12242 still have to process the rest of x0. */
12243 if (ARITHMETIC_P (x0)
12244 && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
12246 update_table_tick (XEXP (x0, x1 == XEXP (x0, 0) ? 1 : 0));
12247 break;
12251 update_table_tick (XEXP (x, i));
12253 else if (fmt[i] == 'E')
12254 for (j = 0; j < XVECLEN (x, i); j++)
12255 update_table_tick (XVECEXP (x, i, j));
12258 /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
12259 are saying that the register is clobbered and we no longer know its
12260 value. If INSN is zero, don't update reg_stat[].last_set; this is
12261 only permitted with VALUE also zero and is used to invalidate the
12262 register. */
12264 static void
12265 record_value_for_reg (rtx reg, rtx insn, rtx value)
12267 unsigned int regno = REGNO (reg);
12268 unsigned int endregno = END_REGNO (reg);
12269 unsigned int i;
12270 reg_stat_type *rsp;
12272 /* If VALUE contains REG and we have a previous value for REG, substitute
12273 the previous value. */
12274 if (value && insn && reg_overlap_mentioned_p (reg, value))
12276 rtx tem;
12278 /* Set things up so get_last_value is allowed to see anything set up to
12279 our insn. */
12280 subst_low_luid = DF_INSN_LUID (insn);
12281 tem = get_last_value (reg);
12283 /* If TEM is simply a binary operation with two CLOBBERs as operands,
12284 it isn't going to be useful and will take a lot of time to process,
12285 so just use the CLOBBER. */
12287 if (tem)
12289 if (ARITHMETIC_P (tem)
12290 && GET_CODE (XEXP (tem, 0)) == CLOBBER
12291 && GET_CODE (XEXP (tem, 1)) == CLOBBER)
12292 tem = XEXP (tem, 0);
12293 else if (count_occurrences (value, reg, 1) >= 2)
12295 /* If there are two or more occurrences of REG in VALUE,
12296 prevent the value from growing too much. */
12297 if (count_rtxs (tem) > MAX_LAST_VALUE_RTL)
12298 tem = gen_rtx_CLOBBER (GET_MODE (tem), const0_rtx);
12301 value = replace_rtx (copy_rtx (value), reg, tem);
12305 /* For each register modified, show we don't know its value, that
12306 we don't know about its bitwise content, that its value has been
12307 updated, and that we don't know the location of the death of the
12308 register. */
12309 for (i = regno; i < endregno; i++)
12311 rsp = VEC_index (reg_stat_type, reg_stat, i);
12313 if (insn)
12314 rsp->last_set = insn;
12316 rsp->last_set_value = 0;
12317 rsp->last_set_mode = VOIDmode;
12318 rsp->last_set_nonzero_bits = 0;
12319 rsp->last_set_sign_bit_copies = 0;
12320 rsp->last_death = 0;
12321 rsp->truncated_to_mode = VOIDmode;
12324 /* Mark registers that are being referenced in this value. */
12325 if (value)
12326 update_table_tick (value);
12328 /* Now update the status of each register being set.
12329 If someone is using this register in this block, set this register
12330 to invalid since we will get confused between the two lives in this
12331 basic block. This makes using this register always invalid. In cse, we
12332 scan the table to invalidate all entries using this register, but this
12333 is too much work for us. */
12335 for (i = regno; i < endregno; i++)
12337 rsp = VEC_index (reg_stat_type, reg_stat, i);
12338 rsp->last_set_label = label_tick;
12339 if (!insn
12340 || (value && rsp->last_set_table_tick >= label_tick_ebb_start))
12341 rsp->last_set_invalid = 1;
12342 else
12343 rsp->last_set_invalid = 0;
12346 /* The value being assigned might refer to X (like in "x++;"). In that
12347 case, we must replace it with (clobber (const_int 0)) to prevent
12348 infinite loops. */
12349 rsp = VEC_index (reg_stat_type, reg_stat, regno);
12350 if (value && !get_last_value_validate (&value, insn, label_tick, 0))
12352 value = copy_rtx (value);
12353 if (!get_last_value_validate (&value, insn, label_tick, 1))
12354 value = 0;
12357 /* For the main register being modified, update the value, the mode, the
12358 nonzero bits, and the number of sign bit copies. */
12360 rsp->last_set_value = value;
12362 if (value)
12364 enum machine_mode mode = GET_MODE (reg);
12365 subst_low_luid = DF_INSN_LUID (insn);
12366 rsp->last_set_mode = mode;
12367 if (GET_MODE_CLASS (mode) == MODE_INT
12368 && HWI_COMPUTABLE_MODE_P (mode))
12369 mode = nonzero_bits_mode;
12370 rsp->last_set_nonzero_bits = nonzero_bits (value, mode);
12371 rsp->last_set_sign_bit_copies
12372 = num_sign_bit_copies (value, GET_MODE (reg));
12376 /* Called via note_stores from record_dead_and_set_regs to handle one
12377 SET or CLOBBER in an insn. DATA is the instruction in which the
12378 set is occurring. */
12380 static void
12381 record_dead_and_set_regs_1 (rtx dest, const_rtx setter, void *data)
12383 rtx record_dead_insn = (rtx) data;
12385 if (GET_CODE (dest) == SUBREG)
12386 dest = SUBREG_REG (dest);
12388 if (!record_dead_insn)
12390 if (REG_P (dest))
12391 record_value_for_reg (dest, NULL_RTX, NULL_RTX);
12392 return;
12395 if (REG_P (dest))
12397 /* If we are setting the whole register, we know its value. Otherwise
12398 show that we don't know the value. We can handle SUBREG in
12399 some cases. */
12400 if (GET_CODE (setter) == SET && dest == SET_DEST (setter))
12401 record_value_for_reg (dest, record_dead_insn, SET_SRC (setter));
12402 else if (GET_CODE (setter) == SET
12403 && GET_CODE (SET_DEST (setter)) == SUBREG
12404 && SUBREG_REG (SET_DEST (setter)) == dest
12405 && GET_MODE_PRECISION (GET_MODE (dest)) <= BITS_PER_WORD
12406 && subreg_lowpart_p (SET_DEST (setter)))
12407 record_value_for_reg (dest, record_dead_insn,
12408 gen_lowpart (GET_MODE (dest),
12409 SET_SRC (setter)));
12410 else
12411 record_value_for_reg (dest, record_dead_insn, NULL_RTX);
12413 else if (MEM_P (dest)
12414 /* Ignore pushes, they clobber nothing. */
12415 && ! push_operand (dest, GET_MODE (dest)))
12416 mem_last_set = DF_INSN_LUID (record_dead_insn);
12419 /* Update the records of when each REG was most recently set or killed
12420 for the things done by INSN. This is the last thing done in processing
12421 INSN in the combiner loop.
12423 We update reg_stat[], in particular fields last_set, last_set_value,
12424 last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
12425 last_death, and also the similar information mem_last_set (which insn
12426 most recently modified memory) and last_call_luid (which insn was the
12427 most recent subroutine call). */
12429 static void
12430 record_dead_and_set_regs (rtx insn)
12432 rtx link;
12433 unsigned int i;
12435 for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
12437 if (REG_NOTE_KIND (link) == REG_DEAD
12438 && REG_P (XEXP (link, 0)))
12440 unsigned int regno = REGNO (XEXP (link, 0));
12441 unsigned int endregno = END_REGNO (XEXP (link, 0));
12443 for (i = regno; i < endregno; i++)
12445 reg_stat_type *rsp;
12447 rsp = VEC_index (reg_stat_type, reg_stat, i);
12448 rsp->last_death = insn;
12451 else if (REG_NOTE_KIND (link) == REG_INC)
12452 record_value_for_reg (XEXP (link, 0), insn, NULL_RTX);
12455 if (CALL_P (insn))
12457 for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
12458 if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
12460 reg_stat_type *rsp;
12462 rsp = VEC_index (reg_stat_type, reg_stat, i);
12463 rsp->last_set_invalid = 1;
12464 rsp->last_set = insn;
12465 rsp->last_set_value = 0;
12466 rsp->last_set_mode = VOIDmode;
12467 rsp->last_set_nonzero_bits = 0;
12468 rsp->last_set_sign_bit_copies = 0;
12469 rsp->last_death = 0;
12470 rsp->truncated_to_mode = VOIDmode;
12473 last_call_luid = mem_last_set = DF_INSN_LUID (insn);
12475 /* We can't combine into a call pattern. Remember, though, that
12476 the return value register is set at this LUID. We could
12477 still replace a register with the return value from the
12478 wrong subroutine call! */
12479 note_stores (PATTERN (insn), record_dead_and_set_regs_1, NULL_RTX);
12481 else
12482 note_stores (PATTERN (insn), record_dead_and_set_regs_1, insn);
12485 /* If a SUBREG has the promoted bit set, it is in fact a property of the
12486 register present in the SUBREG, so for each such SUBREG go back and
12487 adjust nonzero and sign bit information of the registers that are
12488 known to have some zero/sign bits set.
12490 This is needed because when combine blows the SUBREGs away, the
12491 information on zero/sign bits is lost and further combines can be
12492 missed because of that. */
12494 static void
12495 record_promoted_value (rtx insn, rtx subreg)
12497 struct insn_link *links;
12498 rtx set;
12499 unsigned int regno = REGNO (SUBREG_REG (subreg));
12500 enum machine_mode mode = GET_MODE (subreg);
12502 if (GET_MODE_PRECISION (mode) > HOST_BITS_PER_WIDE_INT)
12503 return;
12505 for (links = LOG_LINKS (insn); links;)
12507 reg_stat_type *rsp;
12509 insn = links->insn;
12510 set = single_set (insn);
12512 if (! set || !REG_P (SET_DEST (set))
12513 || REGNO (SET_DEST (set)) != regno
12514 || GET_MODE (SET_DEST (set)) != GET_MODE (SUBREG_REG (subreg)))
12516 links = links->next;
12517 continue;
12520 rsp = VEC_index (reg_stat_type, reg_stat, regno);
12521 if (rsp->last_set == insn)
12523 if (SUBREG_PROMOTED_UNSIGNED_P (subreg) > 0)
12524 rsp->last_set_nonzero_bits &= GET_MODE_MASK (mode);
12527 if (REG_P (SET_SRC (set)))
12529 regno = REGNO (SET_SRC (set));
12530 links = LOG_LINKS (insn);
12532 else
12533 break;
12537 /* Check if X, a register, is known to contain a value already
12538 truncated to MODE. In this case we can use a subreg to refer to
12539 the truncated value even though in the generic case we would need
12540 an explicit truncation. */
12542 static bool
12543 reg_truncated_to_mode (enum machine_mode mode, const_rtx x)
12545 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
12546 enum machine_mode truncated = rsp->truncated_to_mode;
12548 if (truncated == 0
12549 || rsp->truncation_label < label_tick_ebb_start)
12550 return false;
12551 if (GET_MODE_SIZE (truncated) <= GET_MODE_SIZE (mode))
12552 return true;
12553 if (TRULY_NOOP_TRUNCATION_MODES_P (mode, truncated))
12554 return true;
12555 return false;
12558 /* Callback for for_each_rtx. If *P is a hard reg or a subreg record the mode
12559 that the register is accessed in. For non-TRULY_NOOP_TRUNCATION targets we
12560 might be able to turn a truncate into a subreg using this information.
12561 Return -1 if traversing *P is complete or 0 otherwise. */
12563 static int
12564 record_truncated_value (rtx *p, void *data ATTRIBUTE_UNUSED)
12566 rtx x = *p;
12567 enum machine_mode truncated_mode;
12568 reg_stat_type *rsp;
12570 if (GET_CODE (x) == SUBREG && REG_P (SUBREG_REG (x)))
12572 enum machine_mode original_mode = GET_MODE (SUBREG_REG (x));
12573 truncated_mode = GET_MODE (x);
12575 if (GET_MODE_SIZE (original_mode) <= GET_MODE_SIZE (truncated_mode))
12576 return -1;
12578 if (TRULY_NOOP_TRUNCATION_MODES_P (truncated_mode, original_mode))
12579 return -1;
12581 x = SUBREG_REG (x);
12583 /* ??? For hard-regs we now record everything. We might be able to
12584 optimize this using last_set_mode. */
12585 else if (REG_P (x) && REGNO (x) < FIRST_PSEUDO_REGISTER)
12586 truncated_mode = GET_MODE (x);
12587 else
12588 return 0;
12590 rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
12591 if (rsp->truncated_to_mode == 0
12592 || rsp->truncation_label < label_tick_ebb_start
12593 || (GET_MODE_SIZE (truncated_mode)
12594 < GET_MODE_SIZE (rsp->truncated_to_mode)))
12596 rsp->truncated_to_mode = truncated_mode;
12597 rsp->truncation_label = label_tick;
12600 return -1;
12603 /* Callback for note_uses. Find hardregs and subregs of pseudos and
12604 the modes they are used in. This can help truning TRUNCATEs into
12605 SUBREGs. */
12607 static void
12608 record_truncated_values (rtx *x, void *data ATTRIBUTE_UNUSED)
12610 for_each_rtx (x, record_truncated_value, NULL);
12613 /* Scan X for promoted SUBREGs. For each one found,
12614 note what it implies to the registers used in it. */
12616 static void
12617 check_promoted_subreg (rtx insn, rtx x)
12619 if (GET_CODE (x) == SUBREG
12620 && SUBREG_PROMOTED_VAR_P (x)
12621 && REG_P (SUBREG_REG (x)))
12622 record_promoted_value (insn, x);
12623 else
12625 const char *format = GET_RTX_FORMAT (GET_CODE (x));
12626 int i, j;
12628 for (i = 0; i < GET_RTX_LENGTH (GET_CODE (x)); i++)
12629 switch (format[i])
12631 case 'e':
12632 check_promoted_subreg (insn, XEXP (x, i));
12633 break;
12634 case 'V':
12635 case 'E':
12636 if (XVEC (x, i) != 0)
12637 for (j = 0; j < XVECLEN (x, i); j++)
12638 check_promoted_subreg (insn, XVECEXP (x, i, j));
12639 break;
12644 /* Verify that all the registers and memory references mentioned in *LOC are
12645 still valid. *LOC was part of a value set in INSN when label_tick was
12646 equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
12647 the invalid references with (clobber (const_int 0)) and return 1. This
12648 replacement is useful because we often can get useful information about
12649 the form of a value (e.g., if it was produced by a shift that always
12650 produces -1 or 0) even though we don't know exactly what registers it
12651 was produced from. */
12653 static int
12654 get_last_value_validate (rtx *loc, rtx insn, int tick, int replace)
12656 rtx x = *loc;
12657 const char *fmt = GET_RTX_FORMAT (GET_CODE (x));
12658 int len = GET_RTX_LENGTH (GET_CODE (x));
12659 int i, j;
12661 if (REG_P (x))
12663 unsigned int regno = REGNO (x);
12664 unsigned int endregno = END_REGNO (x);
12665 unsigned int j;
12667 for (j = regno; j < endregno; j++)
12669 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, j);
12670 if (rsp->last_set_invalid
12671 /* If this is a pseudo-register that was only set once and not
12672 live at the beginning of the function, it is always valid. */
12673 || (! (regno >= FIRST_PSEUDO_REGISTER
12674 && REG_N_SETS (regno) == 1
12675 && (!REGNO_REG_SET_P
12676 (DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), regno)))
12677 && rsp->last_set_label > tick))
12679 if (replace)
12680 *loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
12681 return replace;
12685 return 1;
12687 /* If this is a memory reference, make sure that there were no stores after
12688 it that might have clobbered the value. We don't have alias info, so we
12689 assume any store invalidates it. Moreover, we only have local UIDs, so
12690 we also assume that there were stores in the intervening basic blocks. */
12691 else if (MEM_P (x) && !MEM_READONLY_P (x)
12692 && (tick != label_tick || DF_INSN_LUID (insn) <= mem_last_set))
12694 if (replace)
12695 *loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
12696 return replace;
12699 for (i = 0; i < len; i++)
12701 if (fmt[i] == 'e')
12703 /* Check for identical subexpressions. If x contains
12704 identical subexpression we only have to traverse one of
12705 them. */
12706 if (i == 1 && ARITHMETIC_P (x))
12708 /* Note that at this point x0 has already been checked
12709 and found valid. */
12710 rtx x0 = XEXP (x, 0);
12711 rtx x1 = XEXP (x, 1);
12713 /* If x0 and x1 are identical then x is also valid. */
12714 if (x0 == x1)
12715 return 1;
12717 /* If x1 is identical to a subexpression of x0 then
12718 while checking x0, x1 has already been checked. Thus
12719 it is valid and so as x. */
12720 if (ARITHMETIC_P (x0)
12721 && (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
12722 return 1;
12724 /* If x0 is identical to a subexpression of x1 then x is
12725 valid iff the rest of x1 is valid. */
12726 if (ARITHMETIC_P (x1)
12727 && (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
12728 return
12729 get_last_value_validate (&XEXP (x1,
12730 x0 == XEXP (x1, 0) ? 1 : 0),
12731 insn, tick, replace);
12734 if (get_last_value_validate (&XEXP (x, i), insn, tick,
12735 replace) == 0)
12736 return 0;
12738 else if (fmt[i] == 'E')
12739 for (j = 0; j < XVECLEN (x, i); j++)
12740 if (get_last_value_validate (&XVECEXP (x, i, j),
12741 insn, tick, replace) == 0)
12742 return 0;
12745 /* If we haven't found a reason for it to be invalid, it is valid. */
12746 return 1;
12749 /* Get the last value assigned to X, if known. Some registers
12750 in the value may be replaced with (clobber (const_int 0)) if their value
12751 is known longer known reliably. */
12753 static rtx
12754 get_last_value (const_rtx x)
12756 unsigned int regno;
12757 rtx value;
12758 reg_stat_type *rsp;
12760 /* If this is a non-paradoxical SUBREG, get the value of its operand and
12761 then convert it to the desired mode. If this is a paradoxical SUBREG,
12762 we cannot predict what values the "extra" bits might have. */
12763 if (GET_CODE (x) == SUBREG
12764 && subreg_lowpart_p (x)
12765 && !paradoxical_subreg_p (x)
12766 && (value = get_last_value (SUBREG_REG (x))) != 0)
12767 return gen_lowpart (GET_MODE (x), value);
12769 if (!REG_P (x))
12770 return 0;
12772 regno = REGNO (x);
12773 rsp = VEC_index (reg_stat_type, reg_stat, regno);
12774 value = rsp->last_set_value;
12776 /* If we don't have a value, or if it isn't for this basic block and
12777 it's either a hard register, set more than once, or it's a live
12778 at the beginning of the function, return 0.
12780 Because if it's not live at the beginning of the function then the reg
12781 is always set before being used (is never used without being set).
12782 And, if it's set only once, and it's always set before use, then all
12783 uses must have the same last value, even if it's not from this basic
12784 block. */
12786 if (value == 0
12787 || (rsp->last_set_label < label_tick_ebb_start
12788 && (regno < FIRST_PSEUDO_REGISTER
12789 || REG_N_SETS (regno) != 1
12790 || REGNO_REG_SET_P
12791 (DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), regno))))
12792 return 0;
12794 /* If the value was set in a later insn than the ones we are processing,
12795 we can't use it even if the register was only set once. */
12796 if (rsp->last_set_label == label_tick
12797 && DF_INSN_LUID (rsp->last_set) >= subst_low_luid)
12798 return 0;
12800 /* If the value has all its registers valid, return it. */
12801 if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 0))
12802 return value;
12804 /* Otherwise, make a copy and replace any invalid register with
12805 (clobber (const_int 0)). If that fails for some reason, return 0. */
12807 value = copy_rtx (value);
12808 if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 1))
12809 return value;
12811 return 0;
12814 /* Return nonzero if expression X refers to a REG or to memory
12815 that is set in an instruction more recent than FROM_LUID. */
12817 static int
12818 use_crosses_set_p (const_rtx x, int from_luid)
12820 const char *fmt;
12821 int i;
12822 enum rtx_code code = GET_CODE (x);
12824 if (code == REG)
12826 unsigned int regno = REGNO (x);
12827 unsigned endreg = END_REGNO (x);
12829 #ifdef PUSH_ROUNDING
12830 /* Don't allow uses of the stack pointer to be moved,
12831 because we don't know whether the move crosses a push insn. */
12832 if (regno == STACK_POINTER_REGNUM && PUSH_ARGS)
12833 return 1;
12834 #endif
12835 for (; regno < endreg; regno++)
12837 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, regno);
12838 if (rsp->last_set
12839 && rsp->last_set_label == label_tick
12840 && DF_INSN_LUID (rsp->last_set) > from_luid)
12841 return 1;
12843 return 0;
12846 if (code == MEM && mem_last_set > from_luid)
12847 return 1;
12849 fmt = GET_RTX_FORMAT (code);
12851 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
12853 if (fmt[i] == 'E')
12855 int j;
12856 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
12857 if (use_crosses_set_p (XVECEXP (x, i, j), from_luid))
12858 return 1;
12860 else if (fmt[i] == 'e'
12861 && use_crosses_set_p (XEXP (x, i), from_luid))
12862 return 1;
12864 return 0;
12867 /* Define three variables used for communication between the following
12868 routines. */
12870 static unsigned int reg_dead_regno, reg_dead_endregno;
12871 static int reg_dead_flag;
12873 /* Function called via note_stores from reg_dead_at_p.
12875 If DEST is within [reg_dead_regno, reg_dead_endregno), set
12876 reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
12878 static void
12879 reg_dead_at_p_1 (rtx dest, const_rtx x, void *data ATTRIBUTE_UNUSED)
12881 unsigned int regno, endregno;
12883 if (!REG_P (dest))
12884 return;
12886 regno = REGNO (dest);
12887 endregno = END_REGNO (dest);
12888 if (reg_dead_endregno > regno && reg_dead_regno < endregno)
12889 reg_dead_flag = (GET_CODE (x) == CLOBBER) ? 1 : -1;
12892 /* Return nonzero if REG is known to be dead at INSN.
12894 We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
12895 referencing REG, it is dead. If we hit a SET referencing REG, it is
12896 live. Otherwise, see if it is live or dead at the start of the basic
12897 block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
12898 must be assumed to be always live. */
12900 static int
12901 reg_dead_at_p (rtx reg, rtx insn)
12903 basic_block block;
12904 unsigned int i;
12906 /* Set variables for reg_dead_at_p_1. */
12907 reg_dead_regno = REGNO (reg);
12908 reg_dead_endregno = END_REGNO (reg);
12910 reg_dead_flag = 0;
12912 /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
12913 we allow the machine description to decide whether use-and-clobber
12914 patterns are OK. */
12915 if (reg_dead_regno < FIRST_PSEUDO_REGISTER)
12917 for (i = reg_dead_regno; i < reg_dead_endregno; i++)
12918 if (!fixed_regs[i] && TEST_HARD_REG_BIT (newpat_used_regs, i))
12919 return 0;
12922 /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
12923 beginning of basic block. */
12924 block = BLOCK_FOR_INSN (insn);
12925 for (;;)
12927 if (INSN_P (insn))
12929 note_stores (PATTERN (insn), reg_dead_at_p_1, NULL);
12930 if (reg_dead_flag)
12931 return reg_dead_flag == 1 ? 1 : 0;
12933 if (find_regno_note (insn, REG_DEAD, reg_dead_regno))
12934 return 1;
12937 if (insn == BB_HEAD (block))
12938 break;
12940 insn = PREV_INSN (insn);
12943 /* Look at live-in sets for the basic block that we were in. */
12944 for (i = reg_dead_regno; i < reg_dead_endregno; i++)
12945 if (REGNO_REG_SET_P (df_get_live_in (block), i))
12946 return 0;
12948 return 1;
12951 /* Note hard registers in X that are used. */
12953 static void
12954 mark_used_regs_combine (rtx x)
12956 RTX_CODE code = GET_CODE (x);
12957 unsigned int regno;
12958 int i;
12960 switch (code)
12962 case LABEL_REF:
12963 case SYMBOL_REF:
12964 case CONST_INT:
12965 case CONST:
12966 case CONST_DOUBLE:
12967 case CONST_VECTOR:
12968 case PC:
12969 case ADDR_VEC:
12970 case ADDR_DIFF_VEC:
12971 case ASM_INPUT:
12972 #ifdef HAVE_cc0
12973 /* CC0 must die in the insn after it is set, so we don't need to take
12974 special note of it here. */
12975 case CC0:
12976 #endif
12977 return;
12979 case CLOBBER:
12980 /* If we are clobbering a MEM, mark any hard registers inside the
12981 address as used. */
12982 if (MEM_P (XEXP (x, 0)))
12983 mark_used_regs_combine (XEXP (XEXP (x, 0), 0));
12984 return;
12986 case REG:
12987 regno = REGNO (x);
12988 /* A hard reg in a wide mode may really be multiple registers.
12989 If so, mark all of them just like the first. */
12990 if (regno < FIRST_PSEUDO_REGISTER)
12992 /* None of this applies to the stack, frame or arg pointers. */
12993 if (regno == STACK_POINTER_REGNUM
12994 #if !HARD_FRAME_POINTER_IS_FRAME_POINTER
12995 || regno == HARD_FRAME_POINTER_REGNUM
12996 #endif
12997 #if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
12998 || (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
12999 #endif
13000 || regno == FRAME_POINTER_REGNUM)
13001 return;
13003 add_to_hard_reg_set (&newpat_used_regs, GET_MODE (x), regno);
13005 return;
13007 case SET:
13009 /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
13010 the address. */
13011 rtx testreg = SET_DEST (x);
13013 while (GET_CODE (testreg) == SUBREG
13014 || GET_CODE (testreg) == ZERO_EXTRACT
13015 || GET_CODE (testreg) == STRICT_LOW_PART)
13016 testreg = XEXP (testreg, 0);
13018 if (MEM_P (testreg))
13019 mark_used_regs_combine (XEXP (testreg, 0));
13021 mark_used_regs_combine (SET_SRC (x));
13023 return;
13025 default:
13026 break;
13029 /* Recursively scan the operands of this expression. */
13032 const char *fmt = GET_RTX_FORMAT (code);
13034 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
13036 if (fmt[i] == 'e')
13037 mark_used_regs_combine (XEXP (x, i));
13038 else if (fmt[i] == 'E')
13040 int j;
13042 for (j = 0; j < XVECLEN (x, i); j++)
13043 mark_used_regs_combine (XVECEXP (x, i, j));
13049 /* Remove register number REGNO from the dead registers list of INSN.
13051 Return the note used to record the death, if there was one. */
13054 remove_death (unsigned int regno, rtx insn)
13056 rtx note = find_regno_note (insn, REG_DEAD, regno);
13058 if (note)
13059 remove_note (insn, note);
13061 return note;
13064 /* For each register (hardware or pseudo) used within expression X, if its
13065 death is in an instruction with luid between FROM_LUID (inclusive) and
13066 TO_INSN (exclusive), put a REG_DEAD note for that register in the
13067 list headed by PNOTES.
13069 That said, don't move registers killed by maybe_kill_insn.
13071 This is done when X is being merged by combination into TO_INSN. These
13072 notes will then be distributed as needed. */
13074 static void
13075 move_deaths (rtx x, rtx maybe_kill_insn, int from_luid, rtx to_insn,
13076 rtx *pnotes)
13078 const char *fmt;
13079 int len, i;
13080 enum rtx_code code = GET_CODE (x);
13082 if (code == REG)
13084 unsigned int regno = REGNO (x);
13085 rtx where_dead = VEC_index (reg_stat_type, reg_stat, regno)->last_death;
13087 /* Don't move the register if it gets killed in between from and to. */
13088 if (maybe_kill_insn && reg_set_p (x, maybe_kill_insn)
13089 && ! reg_referenced_p (x, maybe_kill_insn))
13090 return;
13092 if (where_dead
13093 && BLOCK_FOR_INSN (where_dead) == BLOCK_FOR_INSN (to_insn)
13094 && DF_INSN_LUID (where_dead) >= from_luid
13095 && DF_INSN_LUID (where_dead) < DF_INSN_LUID (to_insn))
13097 rtx note = remove_death (regno, where_dead);
13099 /* It is possible for the call above to return 0. This can occur
13100 when last_death points to I2 or I1 that we combined with.
13101 In that case make a new note.
13103 We must also check for the case where X is a hard register
13104 and NOTE is a death note for a range of hard registers
13105 including X. In that case, we must put REG_DEAD notes for
13106 the remaining registers in place of NOTE. */
13108 if (note != 0 && regno < FIRST_PSEUDO_REGISTER
13109 && (GET_MODE_SIZE (GET_MODE (XEXP (note, 0)))
13110 > GET_MODE_SIZE (GET_MODE (x))))
13112 unsigned int deadregno = REGNO (XEXP (note, 0));
13113 unsigned int deadend = END_HARD_REGNO (XEXP (note, 0));
13114 unsigned int ourend = END_HARD_REGNO (x);
13115 unsigned int i;
13117 for (i = deadregno; i < deadend; i++)
13118 if (i < regno || i >= ourend)
13119 add_reg_note (where_dead, REG_DEAD, regno_reg_rtx[i]);
13122 /* If we didn't find any note, or if we found a REG_DEAD note that
13123 covers only part of the given reg, and we have a multi-reg hard
13124 register, then to be safe we must check for REG_DEAD notes
13125 for each register other than the first. They could have
13126 their own REG_DEAD notes lying around. */
13127 else if ((note == 0
13128 || (note != 0
13129 && (GET_MODE_SIZE (GET_MODE (XEXP (note, 0)))
13130 < GET_MODE_SIZE (GET_MODE (x)))))
13131 && regno < FIRST_PSEUDO_REGISTER
13132 && hard_regno_nregs[regno][GET_MODE (x)] > 1)
13134 unsigned int ourend = END_HARD_REGNO (x);
13135 unsigned int i, offset;
13136 rtx oldnotes = 0;
13138 if (note)
13139 offset = hard_regno_nregs[regno][GET_MODE (XEXP (note, 0))];
13140 else
13141 offset = 1;
13143 for (i = regno + offset; i < ourend; i++)
13144 move_deaths (regno_reg_rtx[i],
13145 maybe_kill_insn, from_luid, to_insn, &oldnotes);
13148 if (note != 0 && GET_MODE (XEXP (note, 0)) == GET_MODE (x))
13150 XEXP (note, 1) = *pnotes;
13151 *pnotes = note;
13153 else
13154 *pnotes = alloc_reg_note (REG_DEAD, x, *pnotes);
13157 return;
13160 else if (GET_CODE (x) == SET)
13162 rtx dest = SET_DEST (x);
13164 move_deaths (SET_SRC (x), maybe_kill_insn, from_luid, to_insn, pnotes);
13166 /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
13167 that accesses one word of a multi-word item, some
13168 piece of everything register in the expression is used by
13169 this insn, so remove any old death. */
13170 /* ??? So why do we test for equality of the sizes? */
13172 if (GET_CODE (dest) == ZERO_EXTRACT
13173 || GET_CODE (dest) == STRICT_LOW_PART
13174 || (GET_CODE (dest) == SUBREG
13175 && (((GET_MODE_SIZE (GET_MODE (dest))
13176 + UNITS_PER_WORD - 1) / UNITS_PER_WORD)
13177 == ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest)))
13178 + UNITS_PER_WORD - 1) / UNITS_PER_WORD))))
13180 move_deaths (dest, maybe_kill_insn, from_luid, to_insn, pnotes);
13181 return;
13184 /* If this is some other SUBREG, we know it replaces the entire
13185 value, so use that as the destination. */
13186 if (GET_CODE (dest) == SUBREG)
13187 dest = SUBREG_REG (dest);
13189 /* If this is a MEM, adjust deaths of anything used in the address.
13190 For a REG (the only other possibility), the entire value is
13191 being replaced so the old value is not used in this insn. */
13193 if (MEM_P (dest))
13194 move_deaths (XEXP (dest, 0), maybe_kill_insn, from_luid,
13195 to_insn, pnotes);
13196 return;
13199 else if (GET_CODE (x) == CLOBBER)
13200 return;
13202 len = GET_RTX_LENGTH (code);
13203 fmt = GET_RTX_FORMAT (code);
13205 for (i = 0; i < len; i++)
13207 if (fmt[i] == 'E')
13209 int j;
13210 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
13211 move_deaths (XVECEXP (x, i, j), maybe_kill_insn, from_luid,
13212 to_insn, pnotes);
13214 else if (fmt[i] == 'e')
13215 move_deaths (XEXP (x, i), maybe_kill_insn, from_luid, to_insn, pnotes);
13219 /* Return 1 if X is the target of a bit-field assignment in BODY, the
13220 pattern of an insn. X must be a REG. */
13222 static int
13223 reg_bitfield_target_p (rtx x, rtx body)
13225 int i;
13227 if (GET_CODE (body) == SET)
13229 rtx dest = SET_DEST (body);
13230 rtx target;
13231 unsigned int regno, tregno, endregno, endtregno;
13233 if (GET_CODE (dest) == ZERO_EXTRACT)
13234 target = XEXP (dest, 0);
13235 else if (GET_CODE (dest) == STRICT_LOW_PART)
13236 target = SUBREG_REG (XEXP (dest, 0));
13237 else
13238 return 0;
13240 if (GET_CODE (target) == SUBREG)
13241 target = SUBREG_REG (target);
13243 if (!REG_P (target))
13244 return 0;
13246 tregno = REGNO (target), regno = REGNO (x);
13247 if (tregno >= FIRST_PSEUDO_REGISTER || regno >= FIRST_PSEUDO_REGISTER)
13248 return target == x;
13250 endtregno = end_hard_regno (GET_MODE (target), tregno);
13251 endregno = end_hard_regno (GET_MODE (x), regno);
13253 return endregno > tregno && regno < endtregno;
13256 else if (GET_CODE (body) == PARALLEL)
13257 for (i = XVECLEN (body, 0) - 1; i >= 0; i--)
13258 if (reg_bitfield_target_p (x, XVECEXP (body, 0, i)))
13259 return 1;
13261 return 0;
13264 /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
13265 as appropriate. I3 and I2 are the insns resulting from the combination
13266 insns including FROM (I2 may be zero).
13268 ELIM_I2 and ELIM_I1 are either zero or registers that we know will
13269 not need REG_DEAD notes because they are being substituted for. This
13270 saves searching in the most common cases.
13272 Each note in the list is either ignored or placed on some insns, depending
13273 on the type of note. */
13275 static void
13276 distribute_notes (rtx notes, rtx from_insn, rtx i3, rtx i2, rtx elim_i2,
13277 rtx elim_i1, rtx elim_i0)
13279 rtx note, next_note;
13280 rtx tem;
13282 for (note = notes; note; note = next_note)
13284 rtx place = 0, place2 = 0;
13286 next_note = XEXP (note, 1);
13287 switch (REG_NOTE_KIND (note))
13289 case REG_BR_PROB:
13290 case REG_BR_PRED:
13291 /* Doesn't matter much where we put this, as long as it's somewhere.
13292 It is preferable to keep these notes on branches, which is most
13293 likely to be i3. */
13294 place = i3;
13295 break;
13297 case REG_NON_LOCAL_GOTO:
13298 if (JUMP_P (i3))
13299 place = i3;
13300 else
13302 gcc_assert (i2 && JUMP_P (i2));
13303 place = i2;
13305 break;
13307 case REG_EH_REGION:
13308 /* These notes must remain with the call or trapping instruction. */
13309 if (CALL_P (i3))
13310 place = i3;
13311 else if (i2 && CALL_P (i2))
13312 place = i2;
13313 else
13315 gcc_assert (cfun->can_throw_non_call_exceptions);
13316 if (may_trap_p (i3))
13317 place = i3;
13318 else if (i2 && may_trap_p (i2))
13319 place = i2;
13320 /* ??? Otherwise assume we've combined things such that we
13321 can now prove that the instructions can't trap. Drop the
13322 note in this case. */
13324 break;
13326 case REG_ARGS_SIZE:
13327 /* ??? How to distribute between i3-i1. Assume i3 contains the
13328 entire adjustment. Assert i3 contains at least some adjust. */
13329 if (!noop_move_p (i3))
13331 int old_size, args_size = INTVAL (XEXP (note, 0));
13332 /* fixup_args_size_notes looks at REG_NORETURN note,
13333 so ensure the note is placed there first. */
13334 if (CALL_P (i3))
13336 rtx *np;
13337 for (np = &next_note; *np; np = &XEXP (*np, 1))
13338 if (REG_NOTE_KIND (*np) == REG_NORETURN)
13340 rtx n = *np;
13341 *np = XEXP (n, 1);
13342 XEXP (n, 1) = REG_NOTES (i3);
13343 REG_NOTES (i3) = n;
13344 break;
13347 old_size = fixup_args_size_notes (PREV_INSN (i3), i3, args_size);
13348 /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
13349 REG_ARGS_SIZE note to all noreturn calls, allow that here. */
13350 gcc_assert (old_size != args_size
13351 || (CALL_P (i3)
13352 && !ACCUMULATE_OUTGOING_ARGS
13353 && find_reg_note (i3, REG_NORETURN, NULL_RTX)));
13355 break;
13357 case REG_NORETURN:
13358 case REG_SETJMP:
13359 case REG_TM:
13360 /* These notes must remain with the call. It should not be
13361 possible for both I2 and I3 to be a call. */
13362 if (CALL_P (i3))
13363 place = i3;
13364 else
13366 gcc_assert (i2 && CALL_P (i2));
13367 place = i2;
13369 break;
13371 case REG_UNUSED:
13372 /* Any clobbers for i3 may still exist, and so we must process
13373 REG_UNUSED notes from that insn.
13375 Any clobbers from i2 or i1 can only exist if they were added by
13376 recog_for_combine. In that case, recog_for_combine created the
13377 necessary REG_UNUSED notes. Trying to keep any original
13378 REG_UNUSED notes from these insns can cause incorrect output
13379 if it is for the same register as the original i3 dest.
13380 In that case, we will notice that the register is set in i3,
13381 and then add a REG_UNUSED note for the destination of i3, which
13382 is wrong. However, it is possible to have REG_UNUSED notes from
13383 i2 or i1 for register which were both used and clobbered, so
13384 we keep notes from i2 or i1 if they will turn into REG_DEAD
13385 notes. */
13387 /* If this register is set or clobbered in I3, put the note there
13388 unless there is one already. */
13389 if (reg_set_p (XEXP (note, 0), PATTERN (i3)))
13391 if (from_insn != i3)
13392 break;
13394 if (! (REG_P (XEXP (note, 0))
13395 ? find_regno_note (i3, REG_UNUSED, REGNO (XEXP (note, 0)))
13396 : find_reg_note (i3, REG_UNUSED, XEXP (note, 0))))
13397 place = i3;
13399 /* Otherwise, if this register is used by I3, then this register
13400 now dies here, so we must put a REG_DEAD note here unless there
13401 is one already. */
13402 else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3))
13403 && ! (REG_P (XEXP (note, 0))
13404 ? find_regno_note (i3, REG_DEAD,
13405 REGNO (XEXP (note, 0)))
13406 : find_reg_note (i3, REG_DEAD, XEXP (note, 0))))
13408 PUT_REG_NOTE_KIND (note, REG_DEAD);
13409 place = i3;
13411 break;
13413 case REG_EQUAL:
13414 case REG_EQUIV:
13415 case REG_NOALIAS:
13416 /* These notes say something about results of an insn. We can
13417 only support them if they used to be on I3 in which case they
13418 remain on I3. Otherwise they are ignored.
13420 If the note refers to an expression that is not a constant, we
13421 must also ignore the note since we cannot tell whether the
13422 equivalence is still true. It might be possible to do
13423 slightly better than this (we only have a problem if I2DEST
13424 or I1DEST is present in the expression), but it doesn't
13425 seem worth the trouble. */
13427 if (from_insn == i3
13428 && (XEXP (note, 0) == 0 || CONSTANT_P (XEXP (note, 0))))
13429 place = i3;
13430 break;
13432 case REG_INC:
13433 /* These notes say something about how a register is used. They must
13434 be present on any use of the register in I2 or I3. */
13435 if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3)))
13436 place = i3;
13438 if (i2 && reg_mentioned_p (XEXP (note, 0), PATTERN (i2)))
13440 if (place)
13441 place2 = i2;
13442 else
13443 place = i2;
13445 break;
13447 case REG_LABEL_TARGET:
13448 case REG_LABEL_OPERAND:
13449 /* This can show up in several ways -- either directly in the
13450 pattern, or hidden off in the constant pool with (or without?)
13451 a REG_EQUAL note. */
13452 /* ??? Ignore the without-reg_equal-note problem for now. */
13453 if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3))
13454 || ((tem = find_reg_note (i3, REG_EQUAL, NULL_RTX))
13455 && GET_CODE (XEXP (tem, 0)) == LABEL_REF
13456 && XEXP (XEXP (tem, 0), 0) == XEXP (note, 0)))
13457 place = i3;
13459 if (i2
13460 && (reg_mentioned_p (XEXP (note, 0), PATTERN (i2))
13461 || ((tem = find_reg_note (i2, REG_EQUAL, NULL_RTX))
13462 && GET_CODE (XEXP (tem, 0)) == LABEL_REF
13463 && XEXP (XEXP (tem, 0), 0) == XEXP (note, 0))))
13465 if (place)
13466 place2 = i2;
13467 else
13468 place = i2;
13471 /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
13472 as a JUMP_LABEL or decrement LABEL_NUSES if it's already
13473 there. */
13474 if (place && JUMP_P (place)
13475 && REG_NOTE_KIND (note) == REG_LABEL_TARGET
13476 && (JUMP_LABEL (place) == NULL
13477 || JUMP_LABEL (place) == XEXP (note, 0)))
13479 rtx label = JUMP_LABEL (place);
13481 if (!label)
13482 JUMP_LABEL (place) = XEXP (note, 0);
13483 else if (LABEL_P (label))
13484 LABEL_NUSES (label)--;
13487 if (place2 && JUMP_P (place2)
13488 && REG_NOTE_KIND (note) == REG_LABEL_TARGET
13489 && (JUMP_LABEL (place2) == NULL
13490 || JUMP_LABEL (place2) == XEXP (note, 0)))
13492 rtx label = JUMP_LABEL (place2);
13494 if (!label)
13495 JUMP_LABEL (place2) = XEXP (note, 0);
13496 else if (LABEL_P (label))
13497 LABEL_NUSES (label)--;
13498 place2 = 0;
13500 break;
13502 case REG_NONNEG:
13503 /* This note says something about the value of a register prior
13504 to the execution of an insn. It is too much trouble to see
13505 if the note is still correct in all situations. It is better
13506 to simply delete it. */
13507 break;
13509 case REG_DEAD:
13510 /* If we replaced the right hand side of FROM_INSN with a
13511 REG_EQUAL note, the original use of the dying register
13512 will not have been combined into I3 and I2. In such cases,
13513 FROM_INSN is guaranteed to be the first of the combined
13514 instructions, so we simply need to search back before
13515 FROM_INSN for the previous use or set of this register,
13516 then alter the notes there appropriately.
13518 If the register is used as an input in I3, it dies there.
13519 Similarly for I2, if it is nonzero and adjacent to I3.
13521 If the register is not used as an input in either I3 or I2
13522 and it is not one of the registers we were supposed to eliminate,
13523 there are two possibilities. We might have a non-adjacent I2
13524 or we might have somehow eliminated an additional register
13525 from a computation. For example, we might have had A & B where
13526 we discover that B will always be zero. In this case we will
13527 eliminate the reference to A.
13529 In both cases, we must search to see if we can find a previous
13530 use of A and put the death note there. */
13532 if (from_insn
13533 && from_insn == i2mod
13534 && !reg_overlap_mentioned_p (XEXP (note, 0), i2mod_new_rhs))
13535 tem = from_insn;
13536 else
13538 if (from_insn
13539 && CALL_P (from_insn)
13540 && find_reg_fusage (from_insn, USE, XEXP (note, 0)))
13541 place = from_insn;
13542 else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3)))
13543 place = i3;
13544 else if (i2 != 0 && next_nonnote_nondebug_insn (i2) == i3
13545 && reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
13546 place = i2;
13547 else if ((rtx_equal_p (XEXP (note, 0), elim_i2)
13548 && !(i2mod
13549 && reg_overlap_mentioned_p (XEXP (note, 0),
13550 i2mod_old_rhs)))
13551 || rtx_equal_p (XEXP (note, 0), elim_i1)
13552 || rtx_equal_p (XEXP (note, 0), elim_i0))
13553 break;
13554 tem = i3;
13557 if (place == 0)
13559 basic_block bb = this_basic_block;
13561 for (tem = PREV_INSN (tem); place == 0; tem = PREV_INSN (tem))
13563 if (!NONDEBUG_INSN_P (tem))
13565 if (tem == BB_HEAD (bb))
13566 break;
13567 continue;
13570 /* If the register is being set at TEM, see if that is all
13571 TEM is doing. If so, delete TEM. Otherwise, make this
13572 into a REG_UNUSED note instead. Don't delete sets to
13573 global register vars. */
13574 if ((REGNO (XEXP (note, 0)) >= FIRST_PSEUDO_REGISTER
13575 || !global_regs[REGNO (XEXP (note, 0))])
13576 && reg_set_p (XEXP (note, 0), PATTERN (tem)))
13578 rtx set = single_set (tem);
13579 rtx inner_dest = 0;
13580 #ifdef HAVE_cc0
13581 rtx cc0_setter = NULL_RTX;
13582 #endif
13584 if (set != 0)
13585 for (inner_dest = SET_DEST (set);
13586 (GET_CODE (inner_dest) == STRICT_LOW_PART
13587 || GET_CODE (inner_dest) == SUBREG
13588 || GET_CODE (inner_dest) == ZERO_EXTRACT);
13589 inner_dest = XEXP (inner_dest, 0))
13592 /* Verify that it was the set, and not a clobber that
13593 modified the register.
13595 CC0 targets must be careful to maintain setter/user
13596 pairs. If we cannot delete the setter due to side
13597 effects, mark the user with an UNUSED note instead
13598 of deleting it. */
13600 if (set != 0 && ! side_effects_p (SET_SRC (set))
13601 && rtx_equal_p (XEXP (note, 0), inner_dest)
13602 #ifdef HAVE_cc0
13603 && (! reg_mentioned_p (cc0_rtx, SET_SRC (set))
13604 || ((cc0_setter = prev_cc0_setter (tem)) != NULL
13605 && sets_cc0_p (PATTERN (cc0_setter)) > 0))
13606 #endif
13609 /* Move the notes and links of TEM elsewhere.
13610 This might delete other dead insns recursively.
13611 First set the pattern to something that won't use
13612 any register. */
13613 rtx old_notes = REG_NOTES (tem);
13615 PATTERN (tem) = pc_rtx;
13616 REG_NOTES (tem) = NULL;
13618 distribute_notes (old_notes, tem, tem, NULL_RTX,
13619 NULL_RTX, NULL_RTX, NULL_RTX);
13620 distribute_links (LOG_LINKS (tem));
13622 SET_INSN_DELETED (tem);
13623 if (tem == i2)
13624 i2 = NULL_RTX;
13626 #ifdef HAVE_cc0
13627 /* Delete the setter too. */
13628 if (cc0_setter)
13630 PATTERN (cc0_setter) = pc_rtx;
13631 old_notes = REG_NOTES (cc0_setter);
13632 REG_NOTES (cc0_setter) = NULL;
13634 distribute_notes (old_notes, cc0_setter,
13635 cc0_setter, NULL_RTX,
13636 NULL_RTX, NULL_RTX, NULL_RTX);
13637 distribute_links (LOG_LINKS (cc0_setter));
13639 SET_INSN_DELETED (cc0_setter);
13640 if (cc0_setter == i2)
13641 i2 = NULL_RTX;
13643 #endif
13645 else
13647 PUT_REG_NOTE_KIND (note, REG_UNUSED);
13649 /* If there isn't already a REG_UNUSED note, put one
13650 here. Do not place a REG_DEAD note, even if
13651 the register is also used here; that would not
13652 match the algorithm used in lifetime analysis
13653 and can cause the consistency check in the
13654 scheduler to fail. */
13655 if (! find_regno_note (tem, REG_UNUSED,
13656 REGNO (XEXP (note, 0))))
13657 place = tem;
13658 break;
13661 else if (reg_referenced_p (XEXP (note, 0), PATTERN (tem))
13662 || (CALL_P (tem)
13663 && find_reg_fusage (tem, USE, XEXP (note, 0))))
13665 place = tem;
13667 /* If we are doing a 3->2 combination, and we have a
13668 register which formerly died in i3 and was not used
13669 by i2, which now no longer dies in i3 and is used in
13670 i2 but does not die in i2, and place is between i2
13671 and i3, then we may need to move a link from place to
13672 i2. */
13673 if (i2 && DF_INSN_LUID (place) > DF_INSN_LUID (i2)
13674 && from_insn
13675 && DF_INSN_LUID (from_insn) > DF_INSN_LUID (i2)
13676 && reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
13678 struct insn_link *links = LOG_LINKS (place);
13679 LOG_LINKS (place) = NULL;
13680 distribute_links (links);
13682 break;
13685 if (tem == BB_HEAD (bb))
13686 break;
13691 /* If the register is set or already dead at PLACE, we needn't do
13692 anything with this note if it is still a REG_DEAD note.
13693 We check here if it is set at all, not if is it totally replaced,
13694 which is what `dead_or_set_p' checks, so also check for it being
13695 set partially. */
13697 if (place && REG_NOTE_KIND (note) == REG_DEAD)
13699 unsigned int regno = REGNO (XEXP (note, 0));
13700 reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, regno);
13702 if (dead_or_set_p (place, XEXP (note, 0))
13703 || reg_bitfield_target_p (XEXP (note, 0), PATTERN (place)))
13705 /* Unless the register previously died in PLACE, clear
13706 last_death. [I no longer understand why this is
13707 being done.] */
13708 if (rsp->last_death != place)
13709 rsp->last_death = 0;
13710 place = 0;
13712 else
13713 rsp->last_death = place;
13715 /* If this is a death note for a hard reg that is occupying
13716 multiple registers, ensure that we are still using all
13717 parts of the object. If we find a piece of the object
13718 that is unused, we must arrange for an appropriate REG_DEAD
13719 note to be added for it. However, we can't just emit a USE
13720 and tag the note to it, since the register might actually
13721 be dead; so we recourse, and the recursive call then finds
13722 the previous insn that used this register. */
13724 if (place && regno < FIRST_PSEUDO_REGISTER
13725 && hard_regno_nregs[regno][GET_MODE (XEXP (note, 0))] > 1)
13727 unsigned int endregno = END_HARD_REGNO (XEXP (note, 0));
13728 int all_used = 1;
13729 unsigned int i;
13731 for (i = regno; i < endregno; i++)
13732 if ((! refers_to_regno_p (i, i + 1, PATTERN (place), 0)
13733 && ! find_regno_fusage (place, USE, i))
13734 || dead_or_set_regno_p (place, i))
13735 all_used = 0;
13737 if (! all_used)
13739 /* Put only REG_DEAD notes for pieces that are
13740 not already dead or set. */
13742 for (i = regno; i < endregno;
13743 i += hard_regno_nregs[i][reg_raw_mode[i]])
13745 rtx piece = regno_reg_rtx[i];
13746 basic_block bb = this_basic_block;
13748 if (! dead_or_set_p (place, piece)
13749 && ! reg_bitfield_target_p (piece,
13750 PATTERN (place)))
13752 rtx new_note = alloc_reg_note (REG_DEAD, piece,
13753 NULL_RTX);
13755 distribute_notes (new_note, place, place,
13756 NULL_RTX, NULL_RTX, NULL_RTX,
13757 NULL_RTX);
13759 else if (! refers_to_regno_p (i, i + 1,
13760 PATTERN (place), 0)
13761 && ! find_regno_fusage (place, USE, i))
13762 for (tem = PREV_INSN (place); ;
13763 tem = PREV_INSN (tem))
13765 if (!NONDEBUG_INSN_P (tem))
13767 if (tem == BB_HEAD (bb))
13768 break;
13769 continue;
13771 if (dead_or_set_p (tem, piece)
13772 || reg_bitfield_target_p (piece,
13773 PATTERN (tem)))
13775 add_reg_note (tem, REG_UNUSED, piece);
13776 break;
13782 place = 0;
13786 break;
13788 default:
13789 /* Any other notes should not be present at this point in the
13790 compilation. */
13791 gcc_unreachable ();
13794 if (place)
13796 XEXP (note, 1) = REG_NOTES (place);
13797 REG_NOTES (place) = note;
13800 if (place2)
13801 add_reg_note (place2, REG_NOTE_KIND (note), XEXP (note, 0));
13805 /* Similarly to above, distribute the LOG_LINKS that used to be present on
13806 I3, I2, and I1 to new locations. This is also called to add a link
13807 pointing at I3 when I3's destination is changed. */
13809 static void
13810 distribute_links (struct insn_link *links)
13812 struct insn_link *link, *next_link;
13814 for (link = links; link; link = next_link)
13816 rtx place = 0;
13817 rtx insn;
13818 rtx set, reg;
13820 next_link = link->next;
13822 /* If the insn that this link points to is a NOTE or isn't a single
13823 set, ignore it. In the latter case, it isn't clear what we
13824 can do other than ignore the link, since we can't tell which
13825 register it was for. Such links wouldn't be used by combine
13826 anyway.
13828 It is not possible for the destination of the target of the link to
13829 have been changed by combine. The only potential of this is if we
13830 replace I3, I2, and I1 by I3 and I2. But in that case the
13831 destination of I2 also remains unchanged. */
13833 if (NOTE_P (link->insn)
13834 || (set = single_set (link->insn)) == 0)
13835 continue;
13837 reg = SET_DEST (set);
13838 while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
13839 || GET_CODE (reg) == STRICT_LOW_PART)
13840 reg = XEXP (reg, 0);
13842 /* A LOG_LINK is defined as being placed on the first insn that uses
13843 a register and points to the insn that sets the register. Start
13844 searching at the next insn after the target of the link and stop
13845 when we reach a set of the register or the end of the basic block.
13847 Note that this correctly handles the link that used to point from
13848 I3 to I2. Also note that not much searching is typically done here
13849 since most links don't point very far away. */
13851 for (insn = NEXT_INSN (link->insn);
13852 (insn && (this_basic_block->next_bb == EXIT_BLOCK_PTR
13853 || BB_HEAD (this_basic_block->next_bb) != insn));
13854 insn = NEXT_INSN (insn))
13855 if (DEBUG_INSN_P (insn))
13856 continue;
13857 else if (INSN_P (insn) && reg_overlap_mentioned_p (reg, PATTERN (insn)))
13859 if (reg_referenced_p (reg, PATTERN (insn)))
13860 place = insn;
13861 break;
13863 else if (CALL_P (insn)
13864 && find_reg_fusage (insn, USE, reg))
13866 place = insn;
13867 break;
13869 else if (INSN_P (insn) && reg_set_p (reg, insn))
13870 break;
13872 /* If we found a place to put the link, place it there unless there
13873 is already a link to the same insn as LINK at that point. */
13875 if (place)
13877 struct insn_link *link2;
13879 FOR_EACH_LOG_LINK (link2, place)
13880 if (link2->insn == link->insn)
13881 break;
13883 if (link2 == NULL)
13885 link->next = LOG_LINKS (place);
13886 LOG_LINKS (place) = link;
13888 /* Set added_links_insn to the earliest insn we added a
13889 link to. */
13890 if (added_links_insn == 0
13891 || DF_INSN_LUID (added_links_insn) > DF_INSN_LUID (place))
13892 added_links_insn = place;
13898 /* Subroutine of unmentioned_reg_p and callback from for_each_rtx.
13899 Check whether the expression pointer to by LOC is a register or
13900 memory, and if so return 1 if it isn't mentioned in the rtx EXPR.
13901 Otherwise return zero. */
13903 static int
13904 unmentioned_reg_p_1 (rtx *loc, void *expr)
13906 rtx x = *loc;
13908 if (x != NULL_RTX
13909 && (REG_P (x) || MEM_P (x))
13910 && ! reg_mentioned_p (x, (rtx) expr))
13911 return 1;
13912 return 0;
13915 /* Check for any register or memory mentioned in EQUIV that is not
13916 mentioned in EXPR. This is used to restrict EQUIV to "specializations"
13917 of EXPR where some registers may have been replaced by constants. */
13919 static bool
13920 unmentioned_reg_p (rtx equiv, rtx expr)
13922 return for_each_rtx (&equiv, unmentioned_reg_p_1, expr);
13925 void
13926 dump_combine_stats (FILE *file)
13928 fprintf
13929 (file,
13930 ";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
13931 combine_attempts, combine_merges, combine_extras, combine_successes);
13934 void
13935 dump_combine_total_stats (FILE *file)
13937 fprintf
13938 (file,
13939 "\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
13940 total_attempts, total_merges, total_extras, total_successes);
13943 static bool
13944 gate_handle_combine (void)
13946 return (optimize > 0);
13949 /* Try combining insns through substitution. */
13950 static unsigned int
13951 rest_of_handle_combine (void)
13953 int rebuild_jump_labels_after_combine;
13955 df_set_flags (DF_LR_RUN_DCE + DF_DEFER_INSN_RESCAN);
13956 df_note_add_problem ();
13957 df_analyze ();
13959 regstat_init_n_sets_and_refs ();
13961 rebuild_jump_labels_after_combine
13962 = combine_instructions (get_insns (), max_reg_num ());
13964 /* Combining insns may have turned an indirect jump into a
13965 direct jump. Rebuild the JUMP_LABEL fields of jumping
13966 instructions. */
13967 if (rebuild_jump_labels_after_combine)
13969 timevar_push (TV_JUMP);
13970 rebuild_jump_labels (get_insns ());
13971 cleanup_cfg (0);
13972 timevar_pop (TV_JUMP);
13975 regstat_free_n_sets_and_refs ();
13976 return 0;
13979 struct rtl_opt_pass pass_combine =
13982 RTL_PASS,
13983 "combine", /* name */
13984 gate_handle_combine, /* gate */
13985 rest_of_handle_combine, /* execute */
13986 NULL, /* sub */
13987 NULL, /* next */
13988 0, /* static_pass_number */
13989 TV_COMBINE, /* tv_id */
13990 PROP_cfglayout, /* properties_required */
13991 0, /* properties_provided */
13992 0, /* properties_destroyed */
13993 0, /* todo_flags_start */
13994 TODO_df_finish | TODO_verify_rtl_sharing |
13995 TODO_ggc_collect, /* todo_flags_finish */