Rebase.
[official-gcc.git] / gcc / cse.c
blob16d2fc1c753745bd58cd73ea827cb33c1c8d96d4
1 /* Common subexpression elimination for GNU compiler.
2 Copyright (C) 1987-2014 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #include "config.h"
21 #include "system.h"
22 #include "coretypes.h"
23 #include "tm.h"
24 #include "rtl.h"
25 #include "tm_p.h"
26 #include "hard-reg-set.h"
27 #include "regs.h"
28 #include "basic-block.h"
29 #include "flags.h"
30 #include "insn-config.h"
31 #include "recog.h"
32 #include "function.h"
33 #include "expr.h"
34 #include "diagnostic-core.h"
35 #include "toplev.h"
36 #include "ggc.h"
37 #include "except.h"
38 #include "target.h"
39 #include "params.h"
40 #include "rtlhooks-def.h"
41 #include "tree-pass.h"
42 #include "df.h"
43 #include "dbgcnt.h"
44 #include "hash-set.h"
46 /* The basic idea of common subexpression elimination is to go
47 through the code, keeping a record of expressions that would
48 have the same value at the current scan point, and replacing
49 expressions encountered with the cheapest equivalent expression.
51 It is too complicated to keep track of the different possibilities
52 when control paths merge in this code; so, at each label, we forget all
53 that is known and start fresh. This can be described as processing each
54 extended basic block separately. We have a separate pass to perform
55 global CSE.
57 Note CSE can turn a conditional or computed jump into a nop or
58 an unconditional jump. When this occurs we arrange to run the jump
59 optimizer after CSE to delete the unreachable code.
61 We use two data structures to record the equivalent expressions:
62 a hash table for most expressions, and a vector of "quantity
63 numbers" to record equivalent (pseudo) registers.
65 The use of the special data structure for registers is desirable
66 because it is faster. It is possible because registers references
67 contain a fairly small number, the register number, taken from
68 a contiguously allocated series, and two register references are
69 identical if they have the same number. General expressions
70 do not have any such thing, so the only way to retrieve the
71 information recorded on an expression other than a register
72 is to keep it in a hash table.
74 Registers and "quantity numbers":
76 At the start of each basic block, all of the (hardware and pseudo)
77 registers used in the function are given distinct quantity
78 numbers to indicate their contents. During scan, when the code
79 copies one register into another, we copy the quantity number.
80 When a register is loaded in any other way, we allocate a new
81 quantity number to describe the value generated by this operation.
82 `REG_QTY (N)' records what quantity register N is currently thought
83 of as containing.
85 All real quantity numbers are greater than or equal to zero.
86 If register N has not been assigned a quantity, `REG_QTY (N)' will
87 equal -N - 1, which is always negative.
89 Quantity numbers below zero do not exist and none of the `qty_table'
90 entries should be referenced with a negative index.
92 We also maintain a bidirectional chain of registers for each
93 quantity number. The `qty_table` members `first_reg' and `last_reg',
94 and `reg_eqv_table' members `next' and `prev' hold these chains.
96 The first register in a chain is the one whose lifespan is least local.
97 Among equals, it is the one that was seen first.
98 We replace any equivalent register with that one.
100 If two registers have the same quantity number, it must be true that
101 REG expressions with qty_table `mode' must be in the hash table for both
102 registers and must be in the same class.
104 The converse is not true. Since hard registers may be referenced in
105 any mode, two REG expressions might be equivalent in the hash table
106 but not have the same quantity number if the quantity number of one
107 of the registers is not the same mode as those expressions.
109 Constants and quantity numbers
111 When a quantity has a known constant value, that value is stored
112 in the appropriate qty_table `const_rtx'. This is in addition to
113 putting the constant in the hash table as is usual for non-regs.
115 Whether a reg or a constant is preferred is determined by the configuration
116 macro CONST_COSTS and will often depend on the constant value. In any
117 event, expressions containing constants can be simplified, by fold_rtx.
119 When a quantity has a known nearly constant value (such as an address
120 of a stack slot), that value is stored in the appropriate qty_table
121 `const_rtx'.
123 Integer constants don't have a machine mode. However, cse
124 determines the intended machine mode from the destination
125 of the instruction that moves the constant. The machine mode
126 is recorded in the hash table along with the actual RTL
127 constant expression so that different modes are kept separate.
129 Other expressions:
131 To record known equivalences among expressions in general
132 we use a hash table called `table'. It has a fixed number of buckets
133 that contain chains of `struct table_elt' elements for expressions.
134 These chains connect the elements whose expressions have the same
135 hash codes.
137 Other chains through the same elements connect the elements which
138 currently have equivalent values.
140 Register references in an expression are canonicalized before hashing
141 the expression. This is done using `reg_qty' and qty_table `first_reg'.
142 The hash code of a register reference is computed using the quantity
143 number, not the register number.
145 When the value of an expression changes, it is necessary to remove from the
146 hash table not just that expression but all expressions whose values
147 could be different as a result.
149 1. If the value changing is in memory, except in special cases
150 ANYTHING referring to memory could be changed. That is because
151 nobody knows where a pointer does not point.
152 The function `invalidate_memory' removes what is necessary.
154 The special cases are when the address is constant or is
155 a constant plus a fixed register such as the frame pointer
156 or a static chain pointer. When such addresses are stored in,
157 we can tell exactly which other such addresses must be invalidated
158 due to overlap. `invalidate' does this.
159 All expressions that refer to non-constant
160 memory addresses are also invalidated. `invalidate_memory' does this.
162 2. If the value changing is a register, all expressions
163 containing references to that register, and only those,
164 must be removed.
166 Because searching the entire hash table for expressions that contain
167 a register is very slow, we try to figure out when it isn't necessary.
168 Precisely, this is necessary only when expressions have been
169 entered in the hash table using this register, and then the value has
170 changed, and then another expression wants to be added to refer to
171 the register's new value. This sequence of circumstances is rare
172 within any one basic block.
174 `REG_TICK' and `REG_IN_TABLE', accessors for members of
175 cse_reg_info, are used to detect this case. REG_TICK (i) is
176 incremented whenever a value is stored in register i.
177 REG_IN_TABLE (i) holds -1 if no references to register i have been
178 entered in the table; otherwise, it contains the value REG_TICK (i)
179 had when the references were entered. If we want to enter a
180 reference and REG_IN_TABLE (i) != REG_TICK (i), we must scan and
181 remove old references. Until we want to enter a new entry, the
182 mere fact that the two vectors don't match makes the entries be
183 ignored if anyone tries to match them.
185 Registers themselves are entered in the hash table as well as in
186 the equivalent-register chains. However, `REG_TICK' and
187 `REG_IN_TABLE' do not apply to expressions which are simple
188 register references. These expressions are removed from the table
189 immediately when they become invalid, and this can be done even if
190 we do not immediately search for all the expressions that refer to
191 the register.
193 A CLOBBER rtx in an instruction invalidates its operand for further
194 reuse. A CLOBBER or SET rtx whose operand is a MEM:BLK
195 invalidates everything that resides in memory.
197 Related expressions:
199 Constant expressions that differ only by an additive integer
200 are called related. When a constant expression is put in
201 the table, the related expression with no constant term
202 is also entered. These are made to point at each other
203 so that it is possible to find out if there exists any
204 register equivalent to an expression related to a given expression. */
206 /* Length of qty_table vector. We know in advance we will not need
207 a quantity number this big. */
209 static int max_qty;
211 /* Next quantity number to be allocated.
212 This is 1 + the largest number needed so far. */
214 static int next_qty;
216 /* Per-qty information tracking.
218 `first_reg' and `last_reg' track the head and tail of the
219 chain of registers which currently contain this quantity.
221 `mode' contains the machine mode of this quantity.
223 `const_rtx' holds the rtx of the constant value of this
224 quantity, if known. A summations of the frame/arg pointer
225 and a constant can also be entered here. When this holds
226 a known value, `const_insn' is the insn which stored the
227 constant value.
229 `comparison_{code,const,qty}' are used to track when a
230 comparison between a quantity and some constant or register has
231 been passed. In such a case, we know the results of the comparison
232 in case we see it again. These members record a comparison that
233 is known to be true. `comparison_code' holds the rtx code of such
234 a comparison, else it is set to UNKNOWN and the other two
235 comparison members are undefined. `comparison_const' holds
236 the constant being compared against, or zero if the comparison
237 is not against a constant. `comparison_qty' holds the quantity
238 being compared against when the result is known. If the comparison
239 is not with a register, `comparison_qty' is -1. */
241 struct qty_table_elem
243 rtx const_rtx;
244 rtx const_insn;
245 rtx comparison_const;
246 int comparison_qty;
247 unsigned int first_reg, last_reg;
248 /* The sizes of these fields should match the sizes of the
249 code and mode fields of struct rtx_def (see rtl.h). */
250 ENUM_BITFIELD(rtx_code) comparison_code : 16;
251 ENUM_BITFIELD(machine_mode) mode : 8;
254 /* The table of all qtys, indexed by qty number. */
255 static struct qty_table_elem *qty_table;
257 /* Structure used to pass arguments via for_each_rtx to function
258 cse_change_cc_mode. */
259 struct change_cc_mode_args
261 rtx insn;
262 rtx newreg;
265 #ifdef HAVE_cc0
266 /* For machines that have a CC0, we do not record its value in the hash
267 table since its use is guaranteed to be the insn immediately following
268 its definition and any other insn is presumed to invalidate it.
270 Instead, we store below the current and last value assigned to CC0.
271 If it should happen to be a constant, it is stored in preference
272 to the actual assigned value. In case it is a constant, we store
273 the mode in which the constant should be interpreted. */
275 static rtx this_insn_cc0, prev_insn_cc0;
276 static enum machine_mode this_insn_cc0_mode, prev_insn_cc0_mode;
277 #endif
279 /* Insn being scanned. */
281 static rtx this_insn;
282 static bool optimize_this_for_speed_p;
284 /* Index by register number, gives the number of the next (or
285 previous) register in the chain of registers sharing the same
286 value.
288 Or -1 if this register is at the end of the chain.
290 If REG_QTY (N) == -N - 1, reg_eqv_table[N].next is undefined. */
292 /* Per-register equivalence chain. */
293 struct reg_eqv_elem
295 int next, prev;
298 /* The table of all register equivalence chains. */
299 static struct reg_eqv_elem *reg_eqv_table;
301 struct cse_reg_info
303 /* The timestamp at which this register is initialized. */
304 unsigned int timestamp;
306 /* The quantity number of the register's current contents. */
307 int reg_qty;
309 /* The number of times the register has been altered in the current
310 basic block. */
311 int reg_tick;
313 /* The REG_TICK value at which rtx's containing this register are
314 valid in the hash table. If this does not equal the current
315 reg_tick value, such expressions existing in the hash table are
316 invalid. */
317 int reg_in_table;
319 /* The SUBREG that was set when REG_TICK was last incremented. Set
320 to -1 if the last store was to the whole register, not a subreg. */
321 unsigned int subreg_ticked;
324 /* A table of cse_reg_info indexed by register numbers. */
325 static struct cse_reg_info *cse_reg_info_table;
327 /* The size of the above table. */
328 static unsigned int cse_reg_info_table_size;
330 /* The index of the first entry that has not been initialized. */
331 static unsigned int cse_reg_info_table_first_uninitialized;
333 /* The timestamp at the beginning of the current run of
334 cse_extended_basic_block. We increment this variable at the beginning of
335 the current run of cse_extended_basic_block. The timestamp field of a
336 cse_reg_info entry matches the value of this variable if and only
337 if the entry has been initialized during the current run of
338 cse_extended_basic_block. */
339 static unsigned int cse_reg_info_timestamp;
341 /* A HARD_REG_SET containing all the hard registers for which there is
342 currently a REG expression in the hash table. Note the difference
343 from the above variables, which indicate if the REG is mentioned in some
344 expression in the table. */
346 static HARD_REG_SET hard_regs_in_table;
348 /* True if CSE has altered the CFG. */
349 static bool cse_cfg_altered;
351 /* True if CSE has altered conditional jump insns in such a way
352 that jump optimization should be redone. */
353 static bool cse_jumps_altered;
355 /* True if we put a LABEL_REF into the hash table for an INSN
356 without a REG_LABEL_OPERAND, we have to rerun jump after CSE
357 to put in the note. */
358 static bool recorded_label_ref;
360 /* canon_hash stores 1 in do_not_record
361 if it notices a reference to CC0, PC, or some other volatile
362 subexpression. */
364 static int do_not_record;
366 /* canon_hash stores 1 in hash_arg_in_memory
367 if it notices a reference to memory within the expression being hashed. */
369 static int hash_arg_in_memory;
371 /* The hash table contains buckets which are chains of `struct table_elt's,
372 each recording one expression's information.
373 That expression is in the `exp' field.
375 The canon_exp field contains a canonical (from the point of view of
376 alias analysis) version of the `exp' field.
378 Those elements with the same hash code are chained in both directions
379 through the `next_same_hash' and `prev_same_hash' fields.
381 Each set of expressions with equivalent values
382 are on a two-way chain through the `next_same_value'
383 and `prev_same_value' fields, and all point with
384 the `first_same_value' field at the first element in
385 that chain. The chain is in order of increasing cost.
386 Each element's cost value is in its `cost' field.
388 The `in_memory' field is nonzero for elements that
389 involve any reference to memory. These elements are removed
390 whenever a write is done to an unidentified location in memory.
391 To be safe, we assume that a memory address is unidentified unless
392 the address is either a symbol constant or a constant plus
393 the frame pointer or argument pointer.
395 The `related_value' field is used to connect related expressions
396 (that differ by adding an integer).
397 The related expressions are chained in a circular fashion.
398 `related_value' is zero for expressions for which this
399 chain is not useful.
401 The `cost' field stores the cost of this element's expression.
402 The `regcost' field stores the value returned by approx_reg_cost for
403 this element's expression.
405 The `is_const' flag is set if the element is a constant (including
406 a fixed address).
408 The `flag' field is used as a temporary during some search routines.
410 The `mode' field is usually the same as GET_MODE (`exp'), but
411 if `exp' is a CONST_INT and has no machine mode then the `mode'
412 field is the mode it was being used as. Each constant is
413 recorded separately for each mode it is used with. */
415 struct table_elt
417 rtx exp;
418 rtx canon_exp;
419 struct table_elt *next_same_hash;
420 struct table_elt *prev_same_hash;
421 struct table_elt *next_same_value;
422 struct table_elt *prev_same_value;
423 struct table_elt *first_same_value;
424 struct table_elt *related_value;
425 int cost;
426 int regcost;
427 /* The size of this field should match the size
428 of the mode field of struct rtx_def (see rtl.h). */
429 ENUM_BITFIELD(machine_mode) mode : 8;
430 char in_memory;
431 char is_const;
432 char flag;
435 /* We don't want a lot of buckets, because we rarely have very many
436 things stored in the hash table, and a lot of buckets slows
437 down a lot of loops that happen frequently. */
438 #define HASH_SHIFT 5
439 #define HASH_SIZE (1 << HASH_SHIFT)
440 #define HASH_MASK (HASH_SIZE - 1)
442 /* Compute hash code of X in mode M. Special-case case where X is a pseudo
443 register (hard registers may require `do_not_record' to be set). */
445 #define HASH(X, M) \
446 ((REG_P (X) && REGNO (X) >= FIRST_PSEUDO_REGISTER \
447 ? (((unsigned) REG << 7) + (unsigned) REG_QTY (REGNO (X))) \
448 : canon_hash (X, M)) & HASH_MASK)
450 /* Like HASH, but without side-effects. */
451 #define SAFE_HASH(X, M) \
452 ((REG_P (X) && REGNO (X) >= FIRST_PSEUDO_REGISTER \
453 ? (((unsigned) REG << 7) + (unsigned) REG_QTY (REGNO (X))) \
454 : safe_hash (X, M)) & HASH_MASK)
456 /* Determine whether register number N is considered a fixed register for the
457 purpose of approximating register costs.
458 It is desirable to replace other regs with fixed regs, to reduce need for
459 non-fixed hard regs.
460 A reg wins if it is either the frame pointer or designated as fixed. */
461 #define FIXED_REGNO_P(N) \
462 ((N) == FRAME_POINTER_REGNUM || (N) == HARD_FRAME_POINTER_REGNUM \
463 || fixed_regs[N] || global_regs[N])
465 /* Compute cost of X, as stored in the `cost' field of a table_elt. Fixed
466 hard registers and pointers into the frame are the cheapest with a cost
467 of 0. Next come pseudos with a cost of one and other hard registers with
468 a cost of 2. Aside from these special cases, call `rtx_cost'. */
470 #define CHEAP_REGNO(N) \
471 (REGNO_PTR_FRAME_P (N) \
472 || (HARD_REGISTER_NUM_P (N) \
473 && FIXED_REGNO_P (N) && REGNO_REG_CLASS (N) != NO_REGS))
475 #define COST(X) (REG_P (X) ? 0 : notreg_cost (X, SET, 1))
476 #define COST_IN(X, OUTER, OPNO) (REG_P (X) ? 0 : notreg_cost (X, OUTER, OPNO))
478 /* Get the number of times this register has been updated in this
479 basic block. */
481 #define REG_TICK(N) (get_cse_reg_info (N)->reg_tick)
483 /* Get the point at which REG was recorded in the table. */
485 #define REG_IN_TABLE(N) (get_cse_reg_info (N)->reg_in_table)
487 /* Get the SUBREG set at the last increment to REG_TICK (-1 if not a
488 SUBREG). */
490 #define SUBREG_TICKED(N) (get_cse_reg_info (N)->subreg_ticked)
492 /* Get the quantity number for REG. */
494 #define REG_QTY(N) (get_cse_reg_info (N)->reg_qty)
496 /* Determine if the quantity number for register X represents a valid index
497 into the qty_table. */
499 #define REGNO_QTY_VALID_P(N) (REG_QTY (N) >= 0)
501 /* Compare table_elt X and Y and return true iff X is cheaper than Y. */
503 #define CHEAPER(X, Y) \
504 (preferable ((X)->cost, (X)->regcost, (Y)->cost, (Y)->regcost) < 0)
506 static struct table_elt *table[HASH_SIZE];
508 /* Chain of `struct table_elt's made so far for this function
509 but currently removed from the table. */
511 static struct table_elt *free_element_chain;
513 /* Set to the cost of a constant pool reference if one was found for a
514 symbolic constant. If this was found, it means we should try to
515 convert constants into constant pool entries if they don't fit in
516 the insn. */
518 static int constant_pool_entries_cost;
519 static int constant_pool_entries_regcost;
521 /* Trace a patch through the CFG. */
523 struct branch_path
525 /* The basic block for this path entry. */
526 basic_block bb;
529 /* This data describes a block that will be processed by
530 cse_extended_basic_block. */
532 struct cse_basic_block_data
534 /* Total number of SETs in block. */
535 int nsets;
536 /* Size of current branch path, if any. */
537 int path_size;
538 /* Current path, indicating which basic_blocks will be processed. */
539 struct branch_path *path;
543 /* Pointers to the live in/live out bitmaps for the boundaries of the
544 current EBB. */
545 static bitmap cse_ebb_live_in, cse_ebb_live_out;
547 /* A simple bitmap to track which basic blocks have been visited
548 already as part of an already processed extended basic block. */
549 static sbitmap cse_visited_basic_blocks;
551 static bool fixed_base_plus_p (rtx x);
552 static int notreg_cost (rtx, enum rtx_code, int);
553 static int approx_reg_cost_1 (rtx *, void *);
554 static int approx_reg_cost (rtx);
555 static int preferable (int, int, int, int);
556 static void new_basic_block (void);
557 static void make_new_qty (unsigned int, enum machine_mode);
558 static void make_regs_eqv (unsigned int, unsigned int);
559 static void delete_reg_equiv (unsigned int);
560 static int mention_regs (rtx);
561 static int insert_regs (rtx, struct table_elt *, int);
562 static void remove_from_table (struct table_elt *, unsigned);
563 static void remove_pseudo_from_table (rtx, unsigned);
564 static struct table_elt *lookup (rtx, unsigned, enum machine_mode);
565 static struct table_elt *lookup_for_remove (rtx, unsigned, enum machine_mode);
566 static rtx lookup_as_function (rtx, enum rtx_code);
567 static struct table_elt *insert_with_costs (rtx, struct table_elt *, unsigned,
568 enum machine_mode, int, int);
569 static struct table_elt *insert (rtx, struct table_elt *, unsigned,
570 enum machine_mode);
571 static void merge_equiv_classes (struct table_elt *, struct table_elt *);
572 static void invalidate (rtx, enum machine_mode);
573 static void remove_invalid_refs (unsigned int);
574 static void remove_invalid_subreg_refs (unsigned int, unsigned int,
575 enum machine_mode);
576 static void rehash_using_reg (rtx);
577 static void invalidate_memory (void);
578 static void invalidate_for_call (void);
579 static rtx use_related_value (rtx, struct table_elt *);
581 static inline unsigned canon_hash (rtx, enum machine_mode);
582 static inline unsigned safe_hash (rtx, enum machine_mode);
583 static inline unsigned hash_rtx_string (const char *);
585 static rtx canon_reg (rtx, rtx);
586 static enum rtx_code find_comparison_args (enum rtx_code, rtx *, rtx *,
587 enum machine_mode *,
588 enum machine_mode *);
589 static rtx fold_rtx (rtx, rtx);
590 static rtx equiv_constant (rtx);
591 static void record_jump_equiv (rtx, bool);
592 static void record_jump_cond (enum rtx_code, enum machine_mode, rtx, rtx,
593 int);
594 static void cse_insn (rtx);
595 static void cse_prescan_path (struct cse_basic_block_data *);
596 static void invalidate_from_clobbers (rtx);
597 static void invalidate_from_sets_and_clobbers (rtx);
598 static rtx cse_process_notes (rtx, rtx, bool *);
599 static void cse_extended_basic_block (struct cse_basic_block_data *);
600 static int check_for_label_ref (rtx *, void *);
601 extern void dump_class (struct table_elt*);
602 static void get_cse_reg_info_1 (unsigned int regno);
603 static struct cse_reg_info * get_cse_reg_info (unsigned int regno);
604 static int check_dependence (rtx *, void *);
606 static void flush_hash_table (void);
607 static bool insn_live_p (rtx, int *);
608 static bool set_live_p (rtx, rtx, int *);
609 static int cse_change_cc_mode (rtx *, void *);
610 static void cse_change_cc_mode_insn (rtx, rtx);
611 static void cse_change_cc_mode_insns (rtx, rtx, rtx);
612 static enum machine_mode cse_cc_succs (basic_block, basic_block, rtx, rtx,
613 bool);
616 #undef RTL_HOOKS_GEN_LOWPART
617 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_if_possible
619 static const struct rtl_hooks cse_rtl_hooks = RTL_HOOKS_INITIALIZER;
621 /* Nonzero if X has the form (PLUS frame-pointer integer). */
623 static bool
624 fixed_base_plus_p (rtx x)
626 switch (GET_CODE (x))
628 case REG:
629 if (x == frame_pointer_rtx || x == hard_frame_pointer_rtx)
630 return true;
631 if (x == arg_pointer_rtx && fixed_regs[ARG_POINTER_REGNUM])
632 return true;
633 return false;
635 case PLUS:
636 if (!CONST_INT_P (XEXP (x, 1)))
637 return false;
638 return fixed_base_plus_p (XEXP (x, 0));
640 default:
641 return false;
645 /* Dump the expressions in the equivalence class indicated by CLASSP.
646 This function is used only for debugging. */
647 DEBUG_FUNCTION void
648 dump_class (struct table_elt *classp)
650 struct table_elt *elt;
652 fprintf (stderr, "Equivalence chain for ");
653 print_rtl (stderr, classp->exp);
654 fprintf (stderr, ": \n");
656 for (elt = classp->first_same_value; elt; elt = elt->next_same_value)
658 print_rtl (stderr, elt->exp);
659 fprintf (stderr, "\n");
663 /* Subroutine of approx_reg_cost; called through for_each_rtx. */
665 static int
666 approx_reg_cost_1 (rtx *xp, void *data)
668 rtx x = *xp;
669 int *cost_p = (int *) data;
671 if (x && REG_P (x))
673 unsigned int regno = REGNO (x);
675 if (! CHEAP_REGNO (regno))
677 if (regno < FIRST_PSEUDO_REGISTER)
679 if (targetm.small_register_classes_for_mode_p (GET_MODE (x)))
680 return 1;
681 *cost_p += 2;
683 else
684 *cost_p += 1;
688 return 0;
691 /* Return an estimate of the cost of the registers used in an rtx.
692 This is mostly the number of different REG expressions in the rtx;
693 however for some exceptions like fixed registers we use a cost of
694 0. If any other hard register reference occurs, return MAX_COST. */
696 static int
697 approx_reg_cost (rtx x)
699 int cost = 0;
701 if (for_each_rtx (&x, approx_reg_cost_1, (void *) &cost))
702 return MAX_COST;
704 return cost;
707 /* Return a negative value if an rtx A, whose costs are given by COST_A
708 and REGCOST_A, is more desirable than an rtx B.
709 Return a positive value if A is less desirable, or 0 if the two are
710 equally good. */
711 static int
712 preferable (int cost_a, int regcost_a, int cost_b, int regcost_b)
714 /* First, get rid of cases involving expressions that are entirely
715 unwanted. */
716 if (cost_a != cost_b)
718 if (cost_a == MAX_COST)
719 return 1;
720 if (cost_b == MAX_COST)
721 return -1;
724 /* Avoid extending lifetimes of hardregs. */
725 if (regcost_a != regcost_b)
727 if (regcost_a == MAX_COST)
728 return 1;
729 if (regcost_b == MAX_COST)
730 return -1;
733 /* Normal operation costs take precedence. */
734 if (cost_a != cost_b)
735 return cost_a - cost_b;
736 /* Only if these are identical consider effects on register pressure. */
737 if (regcost_a != regcost_b)
738 return regcost_a - regcost_b;
739 return 0;
742 /* Internal function, to compute cost when X is not a register; called
743 from COST macro to keep it simple. */
745 static int
746 notreg_cost (rtx x, enum rtx_code outer, int opno)
748 return ((GET_CODE (x) == SUBREG
749 && REG_P (SUBREG_REG (x))
750 && GET_MODE_CLASS (GET_MODE (x)) == MODE_INT
751 && GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_INT
752 && (GET_MODE_SIZE (GET_MODE (x))
753 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
754 && subreg_lowpart_p (x)
755 && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (x),
756 GET_MODE (SUBREG_REG (x))))
758 : rtx_cost (x, outer, opno, optimize_this_for_speed_p) * 2);
762 /* Initialize CSE_REG_INFO_TABLE. */
764 static void
765 init_cse_reg_info (unsigned int nregs)
767 /* Do we need to grow the table? */
768 if (nregs > cse_reg_info_table_size)
770 unsigned int new_size;
772 if (cse_reg_info_table_size < 2048)
774 /* Compute a new size that is a power of 2 and no smaller
775 than the large of NREGS and 64. */
776 new_size = (cse_reg_info_table_size
777 ? cse_reg_info_table_size : 64);
779 while (new_size < nregs)
780 new_size *= 2;
782 else
784 /* If we need a big table, allocate just enough to hold
785 NREGS registers. */
786 new_size = nregs;
789 /* Reallocate the table with NEW_SIZE entries. */
790 free (cse_reg_info_table);
791 cse_reg_info_table = XNEWVEC (struct cse_reg_info, new_size);
792 cse_reg_info_table_size = new_size;
793 cse_reg_info_table_first_uninitialized = 0;
796 /* Do we have all of the first NREGS entries initialized? */
797 if (cse_reg_info_table_first_uninitialized < nregs)
799 unsigned int old_timestamp = cse_reg_info_timestamp - 1;
800 unsigned int i;
802 /* Put the old timestamp on newly allocated entries so that they
803 will all be considered out of date. We do not touch those
804 entries beyond the first NREGS entries to be nice to the
805 virtual memory. */
806 for (i = cse_reg_info_table_first_uninitialized; i < nregs; i++)
807 cse_reg_info_table[i].timestamp = old_timestamp;
809 cse_reg_info_table_first_uninitialized = nregs;
813 /* Given REGNO, initialize the cse_reg_info entry for REGNO. */
815 static void
816 get_cse_reg_info_1 (unsigned int regno)
818 /* Set TIMESTAMP field to CSE_REG_INFO_TIMESTAMP so that this
819 entry will be considered to have been initialized. */
820 cse_reg_info_table[regno].timestamp = cse_reg_info_timestamp;
822 /* Initialize the rest of the entry. */
823 cse_reg_info_table[regno].reg_tick = 1;
824 cse_reg_info_table[regno].reg_in_table = -1;
825 cse_reg_info_table[regno].subreg_ticked = -1;
826 cse_reg_info_table[regno].reg_qty = -regno - 1;
829 /* Find a cse_reg_info entry for REGNO. */
831 static inline struct cse_reg_info *
832 get_cse_reg_info (unsigned int regno)
834 struct cse_reg_info *p = &cse_reg_info_table[regno];
836 /* If this entry has not been initialized, go ahead and initialize
837 it. */
838 if (p->timestamp != cse_reg_info_timestamp)
839 get_cse_reg_info_1 (regno);
841 return p;
844 /* Clear the hash table and initialize each register with its own quantity,
845 for a new basic block. */
847 static void
848 new_basic_block (void)
850 int i;
852 next_qty = 0;
854 /* Invalidate cse_reg_info_table. */
855 cse_reg_info_timestamp++;
857 /* Clear out hash table state for this pass. */
858 CLEAR_HARD_REG_SET (hard_regs_in_table);
860 /* The per-quantity values used to be initialized here, but it is
861 much faster to initialize each as it is made in `make_new_qty'. */
863 for (i = 0; i < HASH_SIZE; i++)
865 struct table_elt *first;
867 first = table[i];
868 if (first != NULL)
870 struct table_elt *last = first;
872 table[i] = NULL;
874 while (last->next_same_hash != NULL)
875 last = last->next_same_hash;
877 /* Now relink this hash entire chain into
878 the free element list. */
880 last->next_same_hash = free_element_chain;
881 free_element_chain = first;
885 #ifdef HAVE_cc0
886 prev_insn_cc0 = 0;
887 #endif
890 /* Say that register REG contains a quantity in mode MODE not in any
891 register before and initialize that quantity. */
893 static void
894 make_new_qty (unsigned int reg, enum machine_mode mode)
896 int q;
897 struct qty_table_elem *ent;
898 struct reg_eqv_elem *eqv;
900 gcc_assert (next_qty < max_qty);
902 q = REG_QTY (reg) = next_qty++;
903 ent = &qty_table[q];
904 ent->first_reg = reg;
905 ent->last_reg = reg;
906 ent->mode = mode;
907 ent->const_rtx = ent->const_insn = NULL_RTX;
908 ent->comparison_code = UNKNOWN;
910 eqv = &reg_eqv_table[reg];
911 eqv->next = eqv->prev = -1;
914 /* Make reg NEW equivalent to reg OLD.
915 OLD is not changing; NEW is. */
917 static void
918 make_regs_eqv (unsigned int new_reg, unsigned int old_reg)
920 unsigned int lastr, firstr;
921 int q = REG_QTY (old_reg);
922 struct qty_table_elem *ent;
924 ent = &qty_table[q];
926 /* Nothing should become eqv until it has a "non-invalid" qty number. */
927 gcc_assert (REGNO_QTY_VALID_P (old_reg));
929 REG_QTY (new_reg) = q;
930 firstr = ent->first_reg;
931 lastr = ent->last_reg;
933 /* Prefer fixed hard registers to anything. Prefer pseudo regs to other
934 hard regs. Among pseudos, if NEW will live longer than any other reg
935 of the same qty, and that is beyond the current basic block,
936 make it the new canonical replacement for this qty. */
937 if (! (firstr < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (firstr))
938 /* Certain fixed registers might be of the class NO_REGS. This means
939 that not only can they not be allocated by the compiler, but
940 they cannot be used in substitutions or canonicalizations
941 either. */
942 && (new_reg >= FIRST_PSEUDO_REGISTER || REGNO_REG_CLASS (new_reg) != NO_REGS)
943 && ((new_reg < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (new_reg))
944 || (new_reg >= FIRST_PSEUDO_REGISTER
945 && (firstr < FIRST_PSEUDO_REGISTER
946 || (bitmap_bit_p (cse_ebb_live_out, new_reg)
947 && !bitmap_bit_p (cse_ebb_live_out, firstr))
948 || (bitmap_bit_p (cse_ebb_live_in, new_reg)
949 && !bitmap_bit_p (cse_ebb_live_in, firstr))))))
951 reg_eqv_table[firstr].prev = new_reg;
952 reg_eqv_table[new_reg].next = firstr;
953 reg_eqv_table[new_reg].prev = -1;
954 ent->first_reg = new_reg;
956 else
958 /* If NEW is a hard reg (known to be non-fixed), insert at end.
959 Otherwise, insert before any non-fixed hard regs that are at the
960 end. Registers of class NO_REGS cannot be used as an
961 equivalent for anything. */
962 while (lastr < FIRST_PSEUDO_REGISTER && reg_eqv_table[lastr].prev >= 0
963 && (REGNO_REG_CLASS (lastr) == NO_REGS || ! FIXED_REGNO_P (lastr))
964 && new_reg >= FIRST_PSEUDO_REGISTER)
965 lastr = reg_eqv_table[lastr].prev;
966 reg_eqv_table[new_reg].next = reg_eqv_table[lastr].next;
967 if (reg_eqv_table[lastr].next >= 0)
968 reg_eqv_table[reg_eqv_table[lastr].next].prev = new_reg;
969 else
970 qty_table[q].last_reg = new_reg;
971 reg_eqv_table[lastr].next = new_reg;
972 reg_eqv_table[new_reg].prev = lastr;
976 /* Remove REG from its equivalence class. */
978 static void
979 delete_reg_equiv (unsigned int reg)
981 struct qty_table_elem *ent;
982 int q = REG_QTY (reg);
983 int p, n;
985 /* If invalid, do nothing. */
986 if (! REGNO_QTY_VALID_P (reg))
987 return;
989 ent = &qty_table[q];
991 p = reg_eqv_table[reg].prev;
992 n = reg_eqv_table[reg].next;
994 if (n != -1)
995 reg_eqv_table[n].prev = p;
996 else
997 ent->last_reg = p;
998 if (p != -1)
999 reg_eqv_table[p].next = n;
1000 else
1001 ent->first_reg = n;
1003 REG_QTY (reg) = -reg - 1;
1006 /* Remove any invalid expressions from the hash table
1007 that refer to any of the registers contained in expression X.
1009 Make sure that newly inserted references to those registers
1010 as subexpressions will be considered valid.
1012 mention_regs is not called when a register itself
1013 is being stored in the table.
1015 Return 1 if we have done something that may have changed the hash code
1016 of X. */
1018 static int
1019 mention_regs (rtx x)
1021 enum rtx_code code;
1022 int i, j;
1023 const char *fmt;
1024 int changed = 0;
1026 if (x == 0)
1027 return 0;
1029 code = GET_CODE (x);
1030 if (code == REG)
1032 unsigned int regno = REGNO (x);
1033 unsigned int endregno = END_REGNO (x);
1034 unsigned int i;
1036 for (i = regno; i < endregno; i++)
1038 if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
1039 remove_invalid_refs (i);
1041 REG_IN_TABLE (i) = REG_TICK (i);
1042 SUBREG_TICKED (i) = -1;
1045 return 0;
1048 /* If this is a SUBREG, we don't want to discard other SUBREGs of the same
1049 pseudo if they don't use overlapping words. We handle only pseudos
1050 here for simplicity. */
1051 if (code == SUBREG && REG_P (SUBREG_REG (x))
1052 && REGNO (SUBREG_REG (x)) >= FIRST_PSEUDO_REGISTER)
1054 unsigned int i = REGNO (SUBREG_REG (x));
1056 if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
1058 /* If REG_IN_TABLE (i) differs from REG_TICK (i) by one, and
1059 the last store to this register really stored into this
1060 subreg, then remove the memory of this subreg.
1061 Otherwise, remove any memory of the entire register and
1062 all its subregs from the table. */
1063 if (REG_TICK (i) - REG_IN_TABLE (i) > 1
1064 || SUBREG_TICKED (i) != REGNO (SUBREG_REG (x)))
1065 remove_invalid_refs (i);
1066 else
1067 remove_invalid_subreg_refs (i, SUBREG_BYTE (x), GET_MODE (x));
1070 REG_IN_TABLE (i) = REG_TICK (i);
1071 SUBREG_TICKED (i) = REGNO (SUBREG_REG (x));
1072 return 0;
1075 /* If X is a comparison or a COMPARE and either operand is a register
1076 that does not have a quantity, give it one. This is so that a later
1077 call to record_jump_equiv won't cause X to be assigned a different
1078 hash code and not found in the table after that call.
1080 It is not necessary to do this here, since rehash_using_reg can
1081 fix up the table later, but doing this here eliminates the need to
1082 call that expensive function in the most common case where the only
1083 use of the register is in the comparison. */
1085 if (code == COMPARE || COMPARISON_P (x))
1087 if (REG_P (XEXP (x, 0))
1088 && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 0))))
1089 if (insert_regs (XEXP (x, 0), NULL, 0))
1091 rehash_using_reg (XEXP (x, 0));
1092 changed = 1;
1095 if (REG_P (XEXP (x, 1))
1096 && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 1))))
1097 if (insert_regs (XEXP (x, 1), NULL, 0))
1099 rehash_using_reg (XEXP (x, 1));
1100 changed = 1;
1104 fmt = GET_RTX_FORMAT (code);
1105 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
1106 if (fmt[i] == 'e')
1107 changed |= mention_regs (XEXP (x, i));
1108 else if (fmt[i] == 'E')
1109 for (j = 0; j < XVECLEN (x, i); j++)
1110 changed |= mention_regs (XVECEXP (x, i, j));
1112 return changed;
1115 /* Update the register quantities for inserting X into the hash table
1116 with a value equivalent to CLASSP.
1117 (If the class does not contain a REG, it is irrelevant.)
1118 If MODIFIED is nonzero, X is a destination; it is being modified.
1119 Note that delete_reg_equiv should be called on a register
1120 before insert_regs is done on that register with MODIFIED != 0.
1122 Nonzero value means that elements of reg_qty have changed
1123 so X's hash code may be different. */
1125 static int
1126 insert_regs (rtx x, struct table_elt *classp, int modified)
1128 if (REG_P (x))
1130 unsigned int regno = REGNO (x);
1131 int qty_valid;
1133 /* If REGNO is in the equivalence table already but is of the
1134 wrong mode for that equivalence, don't do anything here. */
1136 qty_valid = REGNO_QTY_VALID_P (regno);
1137 if (qty_valid)
1139 struct qty_table_elem *ent = &qty_table[REG_QTY (regno)];
1141 if (ent->mode != GET_MODE (x))
1142 return 0;
1145 if (modified || ! qty_valid)
1147 if (classp)
1148 for (classp = classp->first_same_value;
1149 classp != 0;
1150 classp = classp->next_same_value)
1151 if (REG_P (classp->exp)
1152 && GET_MODE (classp->exp) == GET_MODE (x))
1154 unsigned c_regno = REGNO (classp->exp);
1156 gcc_assert (REGNO_QTY_VALID_P (c_regno));
1158 /* Suppose that 5 is hard reg and 100 and 101 are
1159 pseudos. Consider
1161 (set (reg:si 100) (reg:si 5))
1162 (set (reg:si 5) (reg:si 100))
1163 (set (reg:di 101) (reg:di 5))
1165 We would now set REG_QTY (101) = REG_QTY (5), but the
1166 entry for 5 is in SImode. When we use this later in
1167 copy propagation, we get the register in wrong mode. */
1168 if (qty_table[REG_QTY (c_regno)].mode != GET_MODE (x))
1169 continue;
1171 make_regs_eqv (regno, c_regno);
1172 return 1;
1175 /* Mention_regs for a SUBREG checks if REG_TICK is exactly one larger
1176 than REG_IN_TABLE to find out if there was only a single preceding
1177 invalidation - for the SUBREG - or another one, which would be
1178 for the full register. However, if we find here that REG_TICK
1179 indicates that the register is invalid, it means that it has
1180 been invalidated in a separate operation. The SUBREG might be used
1181 now (then this is a recursive call), or we might use the full REG
1182 now and a SUBREG of it later. So bump up REG_TICK so that
1183 mention_regs will do the right thing. */
1184 if (! modified
1185 && REG_IN_TABLE (regno) >= 0
1186 && REG_TICK (regno) == REG_IN_TABLE (regno) + 1)
1187 REG_TICK (regno)++;
1188 make_new_qty (regno, GET_MODE (x));
1189 return 1;
1192 return 0;
1195 /* If X is a SUBREG, we will likely be inserting the inner register in the
1196 table. If that register doesn't have an assigned quantity number at
1197 this point but does later, the insertion that we will be doing now will
1198 not be accessible because its hash code will have changed. So assign
1199 a quantity number now. */
1201 else if (GET_CODE (x) == SUBREG && REG_P (SUBREG_REG (x))
1202 && ! REGNO_QTY_VALID_P (REGNO (SUBREG_REG (x))))
1204 insert_regs (SUBREG_REG (x), NULL, 0);
1205 mention_regs (x);
1206 return 1;
1208 else
1209 return mention_regs (x);
1213 /* Compute upper and lower anchors for CST. Also compute the offset of CST
1214 from these anchors/bases such that *_BASE + *_OFFS = CST. Return false iff
1215 CST is equal to an anchor. */
1217 static bool
1218 compute_const_anchors (rtx cst,
1219 HOST_WIDE_INT *lower_base, HOST_WIDE_INT *lower_offs,
1220 HOST_WIDE_INT *upper_base, HOST_WIDE_INT *upper_offs)
1222 HOST_WIDE_INT n = INTVAL (cst);
1224 *lower_base = n & ~(targetm.const_anchor - 1);
1225 if (*lower_base == n)
1226 return false;
1228 *upper_base =
1229 (n + (targetm.const_anchor - 1)) & ~(targetm.const_anchor - 1);
1230 *upper_offs = n - *upper_base;
1231 *lower_offs = n - *lower_base;
1232 return true;
1235 /* Insert the equivalence between ANCHOR and (REG + OFF) in mode MODE. */
1237 static void
1238 insert_const_anchor (HOST_WIDE_INT anchor, rtx reg, HOST_WIDE_INT offs,
1239 enum machine_mode mode)
1241 struct table_elt *elt;
1242 unsigned hash;
1243 rtx anchor_exp;
1244 rtx exp;
1246 anchor_exp = GEN_INT (anchor);
1247 hash = HASH (anchor_exp, mode);
1248 elt = lookup (anchor_exp, hash, mode);
1249 if (!elt)
1250 elt = insert (anchor_exp, NULL, hash, mode);
1252 exp = plus_constant (mode, reg, offs);
1253 /* REG has just been inserted and the hash codes recomputed. */
1254 mention_regs (exp);
1255 hash = HASH (exp, mode);
1257 /* Use the cost of the register rather than the whole expression. When
1258 looking up constant anchors we will further offset the corresponding
1259 expression therefore it does not make sense to prefer REGs over
1260 reg-immediate additions. Prefer instead the oldest expression. Also
1261 don't prefer pseudos over hard regs so that we derive constants in
1262 argument registers from other argument registers rather than from the
1263 original pseudo that was used to synthesize the constant. */
1264 insert_with_costs (exp, elt, hash, mode, COST (reg), 1);
1267 /* The constant CST is equivalent to the register REG. Create
1268 equivalences between the two anchors of CST and the corresponding
1269 register-offset expressions using REG. */
1271 static void
1272 insert_const_anchors (rtx reg, rtx cst, enum machine_mode mode)
1274 HOST_WIDE_INT lower_base, lower_offs, upper_base, upper_offs;
1276 if (!compute_const_anchors (cst, &lower_base, &lower_offs,
1277 &upper_base, &upper_offs))
1278 return;
1280 /* Ignore anchors of value 0. Constants accessible from zero are
1281 simple. */
1282 if (lower_base != 0)
1283 insert_const_anchor (lower_base, reg, -lower_offs, mode);
1285 if (upper_base != 0)
1286 insert_const_anchor (upper_base, reg, -upper_offs, mode);
1289 /* We need to express ANCHOR_ELT->exp + OFFS. Walk the equivalence list of
1290 ANCHOR_ELT and see if offsetting any of the entries by OFFS would create a
1291 valid expression. Return the cheapest and oldest of such expressions. In
1292 *OLD, return how old the resulting expression is compared to the other
1293 equivalent expressions. */
1295 static rtx
1296 find_reg_offset_for_const (struct table_elt *anchor_elt, HOST_WIDE_INT offs,
1297 unsigned *old)
1299 struct table_elt *elt;
1300 unsigned idx;
1301 struct table_elt *match_elt;
1302 rtx match;
1304 /* Find the cheapest and *oldest* expression to maximize the chance of
1305 reusing the same pseudo. */
1307 match_elt = NULL;
1308 match = NULL_RTX;
1309 for (elt = anchor_elt->first_same_value, idx = 0;
1310 elt;
1311 elt = elt->next_same_value, idx++)
1313 if (match_elt && CHEAPER (match_elt, elt))
1314 return match;
1316 if (REG_P (elt->exp)
1317 || (GET_CODE (elt->exp) == PLUS
1318 && REG_P (XEXP (elt->exp, 0))
1319 && GET_CODE (XEXP (elt->exp, 1)) == CONST_INT))
1321 rtx x;
1323 /* Ignore expressions that are no longer valid. */
1324 if (!REG_P (elt->exp) && !exp_equiv_p (elt->exp, elt->exp, 1, false))
1325 continue;
1327 x = plus_constant (GET_MODE (elt->exp), elt->exp, offs);
1328 if (REG_P (x)
1329 || (GET_CODE (x) == PLUS
1330 && IN_RANGE (INTVAL (XEXP (x, 1)),
1331 -targetm.const_anchor,
1332 targetm.const_anchor - 1)))
1334 match = x;
1335 match_elt = elt;
1336 *old = idx;
1341 return match;
1344 /* Try to express the constant SRC_CONST using a register+offset expression
1345 derived from a constant anchor. Return it if successful or NULL_RTX,
1346 otherwise. */
1348 static rtx
1349 try_const_anchors (rtx src_const, enum machine_mode mode)
1351 struct table_elt *lower_elt, *upper_elt;
1352 HOST_WIDE_INT lower_base, lower_offs, upper_base, upper_offs;
1353 rtx lower_anchor_rtx, upper_anchor_rtx;
1354 rtx lower_exp = NULL_RTX, upper_exp = NULL_RTX;
1355 unsigned lower_old, upper_old;
1357 /* CONST_INT is used for CC modes, but we should leave those alone. */
1358 if (GET_MODE_CLASS (mode) == MODE_CC)
1359 return NULL_RTX;
1361 gcc_assert (SCALAR_INT_MODE_P (mode));
1362 if (!compute_const_anchors (src_const, &lower_base, &lower_offs,
1363 &upper_base, &upper_offs))
1364 return NULL_RTX;
1366 lower_anchor_rtx = GEN_INT (lower_base);
1367 upper_anchor_rtx = GEN_INT (upper_base);
1368 lower_elt = lookup (lower_anchor_rtx, HASH (lower_anchor_rtx, mode), mode);
1369 upper_elt = lookup (upper_anchor_rtx, HASH (upper_anchor_rtx, mode), mode);
1371 if (lower_elt)
1372 lower_exp = find_reg_offset_for_const (lower_elt, lower_offs, &lower_old);
1373 if (upper_elt)
1374 upper_exp = find_reg_offset_for_const (upper_elt, upper_offs, &upper_old);
1376 if (!lower_exp)
1377 return upper_exp;
1378 if (!upper_exp)
1379 return lower_exp;
1381 /* Return the older expression. */
1382 return (upper_old > lower_old ? upper_exp : lower_exp);
1385 /* Look in or update the hash table. */
1387 /* Remove table element ELT from use in the table.
1388 HASH is its hash code, made using the HASH macro.
1389 It's an argument because often that is known in advance
1390 and we save much time not recomputing it. */
1392 static void
1393 remove_from_table (struct table_elt *elt, unsigned int hash)
1395 if (elt == 0)
1396 return;
1398 /* Mark this element as removed. See cse_insn. */
1399 elt->first_same_value = 0;
1401 /* Remove the table element from its equivalence class. */
1404 struct table_elt *prev = elt->prev_same_value;
1405 struct table_elt *next = elt->next_same_value;
1407 if (next)
1408 next->prev_same_value = prev;
1410 if (prev)
1411 prev->next_same_value = next;
1412 else
1414 struct table_elt *newfirst = next;
1415 while (next)
1417 next->first_same_value = newfirst;
1418 next = next->next_same_value;
1423 /* Remove the table element from its hash bucket. */
1426 struct table_elt *prev = elt->prev_same_hash;
1427 struct table_elt *next = elt->next_same_hash;
1429 if (next)
1430 next->prev_same_hash = prev;
1432 if (prev)
1433 prev->next_same_hash = next;
1434 else if (table[hash] == elt)
1435 table[hash] = next;
1436 else
1438 /* This entry is not in the proper hash bucket. This can happen
1439 when two classes were merged by `merge_equiv_classes'. Search
1440 for the hash bucket that it heads. This happens only very
1441 rarely, so the cost is acceptable. */
1442 for (hash = 0; hash < HASH_SIZE; hash++)
1443 if (table[hash] == elt)
1444 table[hash] = next;
1448 /* Remove the table element from its related-value circular chain. */
1450 if (elt->related_value != 0 && elt->related_value != elt)
1452 struct table_elt *p = elt->related_value;
1454 while (p->related_value != elt)
1455 p = p->related_value;
1456 p->related_value = elt->related_value;
1457 if (p->related_value == p)
1458 p->related_value = 0;
1461 /* Now add it to the free element chain. */
1462 elt->next_same_hash = free_element_chain;
1463 free_element_chain = elt;
1466 /* Same as above, but X is a pseudo-register. */
1468 static void
1469 remove_pseudo_from_table (rtx x, unsigned int hash)
1471 struct table_elt *elt;
1473 /* Because a pseudo-register can be referenced in more than one
1474 mode, we might have to remove more than one table entry. */
1475 while ((elt = lookup_for_remove (x, hash, VOIDmode)))
1476 remove_from_table (elt, hash);
1479 /* Look up X in the hash table and return its table element,
1480 or 0 if X is not in the table.
1482 MODE is the machine-mode of X, or if X is an integer constant
1483 with VOIDmode then MODE is the mode with which X will be used.
1485 Here we are satisfied to find an expression whose tree structure
1486 looks like X. */
1488 static struct table_elt *
1489 lookup (rtx x, unsigned int hash, enum machine_mode mode)
1491 struct table_elt *p;
1493 for (p = table[hash]; p; p = p->next_same_hash)
1494 if (mode == p->mode && ((x == p->exp && REG_P (x))
1495 || exp_equiv_p (x, p->exp, !REG_P (x), false)))
1496 return p;
1498 return 0;
1501 /* Like `lookup' but don't care whether the table element uses invalid regs.
1502 Also ignore discrepancies in the machine mode of a register. */
1504 static struct table_elt *
1505 lookup_for_remove (rtx x, unsigned int hash, enum machine_mode mode)
1507 struct table_elt *p;
1509 if (REG_P (x))
1511 unsigned int regno = REGNO (x);
1513 /* Don't check the machine mode when comparing registers;
1514 invalidating (REG:SI 0) also invalidates (REG:DF 0). */
1515 for (p = table[hash]; p; p = p->next_same_hash)
1516 if (REG_P (p->exp)
1517 && REGNO (p->exp) == regno)
1518 return p;
1520 else
1522 for (p = table[hash]; p; p = p->next_same_hash)
1523 if (mode == p->mode
1524 && (x == p->exp || exp_equiv_p (x, p->exp, 0, false)))
1525 return p;
1528 return 0;
1531 /* Look for an expression equivalent to X and with code CODE.
1532 If one is found, return that expression. */
1534 static rtx
1535 lookup_as_function (rtx x, enum rtx_code code)
1537 struct table_elt *p
1538 = lookup (x, SAFE_HASH (x, VOIDmode), GET_MODE (x));
1540 if (p == 0)
1541 return 0;
1543 for (p = p->first_same_value; p; p = p->next_same_value)
1544 if (GET_CODE (p->exp) == code
1545 /* Make sure this is a valid entry in the table. */
1546 && exp_equiv_p (p->exp, p->exp, 1, false))
1547 return p->exp;
1549 return 0;
1552 /* Insert X in the hash table, assuming HASH is its hash code and
1553 CLASSP is an element of the class it should go in (or 0 if a new
1554 class should be made). COST is the code of X and reg_cost is the
1555 cost of registers in X. It is inserted at the proper position to
1556 keep the class in the order cheapest first.
1558 MODE is the machine-mode of X, or if X is an integer constant
1559 with VOIDmode then MODE is the mode with which X will be used.
1561 For elements of equal cheapness, the most recent one
1562 goes in front, except that the first element in the list
1563 remains first unless a cheaper element is added. The order of
1564 pseudo-registers does not matter, as canon_reg will be called to
1565 find the cheapest when a register is retrieved from the table.
1567 The in_memory field in the hash table element is set to 0.
1568 The caller must set it nonzero if appropriate.
1570 You should call insert_regs (X, CLASSP, MODIFY) before calling here,
1571 and if insert_regs returns a nonzero value
1572 you must then recompute its hash code before calling here.
1574 If necessary, update table showing constant values of quantities. */
1576 static struct table_elt *
1577 insert_with_costs (rtx x, struct table_elt *classp, unsigned int hash,
1578 enum machine_mode mode, int cost, int reg_cost)
1580 struct table_elt *elt;
1582 /* If X is a register and we haven't made a quantity for it,
1583 something is wrong. */
1584 gcc_assert (!REG_P (x) || REGNO_QTY_VALID_P (REGNO (x)));
1586 /* If X is a hard register, show it is being put in the table. */
1587 if (REG_P (x) && REGNO (x) < FIRST_PSEUDO_REGISTER)
1588 add_to_hard_reg_set (&hard_regs_in_table, GET_MODE (x), REGNO (x));
1590 /* Put an element for X into the right hash bucket. */
1592 elt = free_element_chain;
1593 if (elt)
1594 free_element_chain = elt->next_same_hash;
1595 else
1596 elt = XNEW (struct table_elt);
1598 elt->exp = x;
1599 elt->canon_exp = NULL_RTX;
1600 elt->cost = cost;
1601 elt->regcost = reg_cost;
1602 elt->next_same_value = 0;
1603 elt->prev_same_value = 0;
1604 elt->next_same_hash = table[hash];
1605 elt->prev_same_hash = 0;
1606 elt->related_value = 0;
1607 elt->in_memory = 0;
1608 elt->mode = mode;
1609 elt->is_const = (CONSTANT_P (x) || fixed_base_plus_p (x));
1611 if (table[hash])
1612 table[hash]->prev_same_hash = elt;
1613 table[hash] = elt;
1615 /* Put it into the proper value-class. */
1616 if (classp)
1618 classp = classp->first_same_value;
1619 if (CHEAPER (elt, classp))
1620 /* Insert at the head of the class. */
1622 struct table_elt *p;
1623 elt->next_same_value = classp;
1624 classp->prev_same_value = elt;
1625 elt->first_same_value = elt;
1627 for (p = classp; p; p = p->next_same_value)
1628 p->first_same_value = elt;
1630 else
1632 /* Insert not at head of the class. */
1633 /* Put it after the last element cheaper than X. */
1634 struct table_elt *p, *next;
1636 for (p = classp;
1637 (next = p->next_same_value) && CHEAPER (next, elt);
1638 p = next)
1641 /* Put it after P and before NEXT. */
1642 elt->next_same_value = next;
1643 if (next)
1644 next->prev_same_value = elt;
1646 elt->prev_same_value = p;
1647 p->next_same_value = elt;
1648 elt->first_same_value = classp;
1651 else
1652 elt->first_same_value = elt;
1654 /* If this is a constant being set equivalent to a register or a register
1655 being set equivalent to a constant, note the constant equivalence.
1657 If this is a constant, it cannot be equivalent to a different constant,
1658 and a constant is the only thing that can be cheaper than a register. So
1659 we know the register is the head of the class (before the constant was
1660 inserted).
1662 If this is a register that is not already known equivalent to a
1663 constant, we must check the entire class.
1665 If this is a register that is already known equivalent to an insn,
1666 update the qtys `const_insn' to show that `this_insn' is the latest
1667 insn making that quantity equivalent to the constant. */
1669 if (elt->is_const && classp && REG_P (classp->exp)
1670 && !REG_P (x))
1672 int exp_q = REG_QTY (REGNO (classp->exp));
1673 struct qty_table_elem *exp_ent = &qty_table[exp_q];
1675 exp_ent->const_rtx = gen_lowpart (exp_ent->mode, x);
1676 exp_ent->const_insn = this_insn;
1679 else if (REG_P (x)
1680 && classp
1681 && ! qty_table[REG_QTY (REGNO (x))].const_rtx
1682 && ! elt->is_const)
1684 struct table_elt *p;
1686 for (p = classp; p != 0; p = p->next_same_value)
1688 if (p->is_const && !REG_P (p->exp))
1690 int x_q = REG_QTY (REGNO (x));
1691 struct qty_table_elem *x_ent = &qty_table[x_q];
1693 x_ent->const_rtx
1694 = gen_lowpart (GET_MODE (x), p->exp);
1695 x_ent->const_insn = this_insn;
1696 break;
1701 else if (REG_P (x)
1702 && qty_table[REG_QTY (REGNO (x))].const_rtx
1703 && GET_MODE (x) == qty_table[REG_QTY (REGNO (x))].mode)
1704 qty_table[REG_QTY (REGNO (x))].const_insn = this_insn;
1706 /* If this is a constant with symbolic value,
1707 and it has a term with an explicit integer value,
1708 link it up with related expressions. */
1709 if (GET_CODE (x) == CONST)
1711 rtx subexp = get_related_value (x);
1712 unsigned subhash;
1713 struct table_elt *subelt, *subelt_prev;
1715 if (subexp != 0)
1717 /* Get the integer-free subexpression in the hash table. */
1718 subhash = SAFE_HASH (subexp, mode);
1719 subelt = lookup (subexp, subhash, mode);
1720 if (subelt == 0)
1721 subelt = insert (subexp, NULL, subhash, mode);
1722 /* Initialize SUBELT's circular chain if it has none. */
1723 if (subelt->related_value == 0)
1724 subelt->related_value = subelt;
1725 /* Find the element in the circular chain that precedes SUBELT. */
1726 subelt_prev = subelt;
1727 while (subelt_prev->related_value != subelt)
1728 subelt_prev = subelt_prev->related_value;
1729 /* Put new ELT into SUBELT's circular chain just before SUBELT.
1730 This way the element that follows SUBELT is the oldest one. */
1731 elt->related_value = subelt_prev->related_value;
1732 subelt_prev->related_value = elt;
1736 return elt;
1739 /* Wrap insert_with_costs by passing the default costs. */
1741 static struct table_elt *
1742 insert (rtx x, struct table_elt *classp, unsigned int hash,
1743 enum machine_mode mode)
1745 return
1746 insert_with_costs (x, classp, hash, mode, COST (x), approx_reg_cost (x));
1750 /* Given two equivalence classes, CLASS1 and CLASS2, put all the entries from
1751 CLASS2 into CLASS1. This is done when we have reached an insn which makes
1752 the two classes equivalent.
1754 CLASS1 will be the surviving class; CLASS2 should not be used after this
1755 call.
1757 Any invalid entries in CLASS2 will not be copied. */
1759 static void
1760 merge_equiv_classes (struct table_elt *class1, struct table_elt *class2)
1762 struct table_elt *elt, *next, *new_elt;
1764 /* Ensure we start with the head of the classes. */
1765 class1 = class1->first_same_value;
1766 class2 = class2->first_same_value;
1768 /* If they were already equal, forget it. */
1769 if (class1 == class2)
1770 return;
1772 for (elt = class2; elt; elt = next)
1774 unsigned int hash;
1775 rtx exp = elt->exp;
1776 enum machine_mode mode = elt->mode;
1778 next = elt->next_same_value;
1780 /* Remove old entry, make a new one in CLASS1's class.
1781 Don't do this for invalid entries as we cannot find their
1782 hash code (it also isn't necessary). */
1783 if (REG_P (exp) || exp_equiv_p (exp, exp, 1, false))
1785 bool need_rehash = false;
1787 hash_arg_in_memory = 0;
1788 hash = HASH (exp, mode);
1790 if (REG_P (exp))
1792 need_rehash = REGNO_QTY_VALID_P (REGNO (exp));
1793 delete_reg_equiv (REGNO (exp));
1796 if (REG_P (exp) && REGNO (exp) >= FIRST_PSEUDO_REGISTER)
1797 remove_pseudo_from_table (exp, hash);
1798 else
1799 remove_from_table (elt, hash);
1801 if (insert_regs (exp, class1, 0) || need_rehash)
1803 rehash_using_reg (exp);
1804 hash = HASH (exp, mode);
1806 new_elt = insert (exp, class1, hash, mode);
1807 new_elt->in_memory = hash_arg_in_memory;
1812 /* Flush the entire hash table. */
1814 static void
1815 flush_hash_table (void)
1817 int i;
1818 struct table_elt *p;
1820 for (i = 0; i < HASH_SIZE; i++)
1821 for (p = table[i]; p; p = table[i])
1823 /* Note that invalidate can remove elements
1824 after P in the current hash chain. */
1825 if (REG_P (p->exp))
1826 invalidate (p->exp, VOIDmode);
1827 else
1828 remove_from_table (p, i);
1832 /* Function called for each rtx to check whether an anti dependence exist. */
1833 struct check_dependence_data
1835 enum machine_mode mode;
1836 rtx exp;
1837 rtx addr;
1840 static int
1841 check_dependence (rtx *x, void *data)
1843 struct check_dependence_data *d = (struct check_dependence_data *) data;
1844 if (*x && MEM_P (*x))
1845 return canon_anti_dependence (*x, true, d->exp, d->mode, d->addr);
1846 else
1847 return 0;
1850 /* Remove from the hash table, or mark as invalid, all expressions whose
1851 values could be altered by storing in X. X is a register, a subreg, or
1852 a memory reference with nonvarying address (because, when a memory
1853 reference with a varying address is stored in, all memory references are
1854 removed by invalidate_memory so specific invalidation is superfluous).
1855 FULL_MODE, if not VOIDmode, indicates that this much should be
1856 invalidated instead of just the amount indicated by the mode of X. This
1857 is only used for bitfield stores into memory.
1859 A nonvarying address may be just a register or just a symbol reference,
1860 or it may be either of those plus a numeric offset. */
1862 static void
1863 invalidate (rtx x, enum machine_mode full_mode)
1865 int i;
1866 struct table_elt *p;
1867 rtx addr;
1869 switch (GET_CODE (x))
1871 case REG:
1873 /* If X is a register, dependencies on its contents are recorded
1874 through the qty number mechanism. Just change the qty number of
1875 the register, mark it as invalid for expressions that refer to it,
1876 and remove it itself. */
1877 unsigned int regno = REGNO (x);
1878 unsigned int hash = HASH (x, GET_MODE (x));
1880 /* Remove REGNO from any quantity list it might be on and indicate
1881 that its value might have changed. If it is a pseudo, remove its
1882 entry from the hash table.
1884 For a hard register, we do the first two actions above for any
1885 additional hard registers corresponding to X. Then, if any of these
1886 registers are in the table, we must remove any REG entries that
1887 overlap these registers. */
1889 delete_reg_equiv (regno);
1890 REG_TICK (regno)++;
1891 SUBREG_TICKED (regno) = -1;
1893 if (regno >= FIRST_PSEUDO_REGISTER)
1894 remove_pseudo_from_table (x, hash);
1895 else
1897 HOST_WIDE_INT in_table
1898 = TEST_HARD_REG_BIT (hard_regs_in_table, regno);
1899 unsigned int endregno = END_HARD_REGNO (x);
1900 unsigned int tregno, tendregno, rn;
1901 struct table_elt *p, *next;
1903 CLEAR_HARD_REG_BIT (hard_regs_in_table, regno);
1905 for (rn = regno + 1; rn < endregno; rn++)
1907 in_table |= TEST_HARD_REG_BIT (hard_regs_in_table, rn);
1908 CLEAR_HARD_REG_BIT (hard_regs_in_table, rn);
1909 delete_reg_equiv (rn);
1910 REG_TICK (rn)++;
1911 SUBREG_TICKED (rn) = -1;
1914 if (in_table)
1915 for (hash = 0; hash < HASH_SIZE; hash++)
1916 for (p = table[hash]; p; p = next)
1918 next = p->next_same_hash;
1920 if (!REG_P (p->exp)
1921 || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
1922 continue;
1924 tregno = REGNO (p->exp);
1925 tendregno = END_HARD_REGNO (p->exp);
1926 if (tendregno > regno && tregno < endregno)
1927 remove_from_table (p, hash);
1931 return;
1933 case SUBREG:
1934 invalidate (SUBREG_REG (x), VOIDmode);
1935 return;
1937 case PARALLEL:
1938 for (i = XVECLEN (x, 0) - 1; i >= 0; --i)
1939 invalidate (XVECEXP (x, 0, i), VOIDmode);
1940 return;
1942 case EXPR_LIST:
1943 /* This is part of a disjoint return value; extract the location in
1944 question ignoring the offset. */
1945 invalidate (XEXP (x, 0), VOIDmode);
1946 return;
1948 case MEM:
1949 addr = canon_rtx (get_addr (XEXP (x, 0)));
1950 /* Calculate the canonical version of X here so that
1951 true_dependence doesn't generate new RTL for X on each call. */
1952 x = canon_rtx (x);
1954 /* Remove all hash table elements that refer to overlapping pieces of
1955 memory. */
1956 if (full_mode == VOIDmode)
1957 full_mode = GET_MODE (x);
1959 for (i = 0; i < HASH_SIZE; i++)
1961 struct table_elt *next;
1963 for (p = table[i]; p; p = next)
1965 next = p->next_same_hash;
1966 if (p->in_memory)
1968 struct check_dependence_data d;
1970 /* Just canonicalize the expression once;
1971 otherwise each time we call invalidate
1972 true_dependence will canonicalize the
1973 expression again. */
1974 if (!p->canon_exp)
1975 p->canon_exp = canon_rtx (p->exp);
1976 d.exp = x;
1977 d.addr = addr;
1978 d.mode = full_mode;
1979 if (for_each_rtx (&p->canon_exp, check_dependence, &d))
1980 remove_from_table (p, i);
1984 return;
1986 default:
1987 gcc_unreachable ();
1991 /* Remove all expressions that refer to register REGNO,
1992 since they are already invalid, and we are about to
1993 mark that register valid again and don't want the old
1994 expressions to reappear as valid. */
1996 static void
1997 remove_invalid_refs (unsigned int regno)
1999 unsigned int i;
2000 struct table_elt *p, *next;
2002 for (i = 0; i < HASH_SIZE; i++)
2003 for (p = table[i]; p; p = next)
2005 next = p->next_same_hash;
2006 if (!REG_P (p->exp)
2007 && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
2008 remove_from_table (p, i);
2012 /* Likewise for a subreg with subreg_reg REGNO, subreg_byte OFFSET,
2013 and mode MODE. */
2014 static void
2015 remove_invalid_subreg_refs (unsigned int regno, unsigned int offset,
2016 enum machine_mode mode)
2018 unsigned int i;
2019 struct table_elt *p, *next;
2020 unsigned int end = offset + (GET_MODE_SIZE (mode) - 1);
2022 for (i = 0; i < HASH_SIZE; i++)
2023 for (p = table[i]; p; p = next)
2025 rtx exp = p->exp;
2026 next = p->next_same_hash;
2028 if (!REG_P (exp)
2029 && (GET_CODE (exp) != SUBREG
2030 || !REG_P (SUBREG_REG (exp))
2031 || REGNO (SUBREG_REG (exp)) != regno
2032 || (((SUBREG_BYTE (exp)
2033 + (GET_MODE_SIZE (GET_MODE (exp)) - 1)) >= offset)
2034 && SUBREG_BYTE (exp) <= end))
2035 && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
2036 remove_from_table (p, i);
2040 /* Recompute the hash codes of any valid entries in the hash table that
2041 reference X, if X is a register, or SUBREG_REG (X) if X is a SUBREG.
2043 This is called when we make a jump equivalence. */
2045 static void
2046 rehash_using_reg (rtx x)
2048 unsigned int i;
2049 struct table_elt *p, *next;
2050 unsigned hash;
2052 if (GET_CODE (x) == SUBREG)
2053 x = SUBREG_REG (x);
2055 /* If X is not a register or if the register is known not to be in any
2056 valid entries in the table, we have no work to do. */
2058 if (!REG_P (x)
2059 || REG_IN_TABLE (REGNO (x)) < 0
2060 || REG_IN_TABLE (REGNO (x)) != REG_TICK (REGNO (x)))
2061 return;
2063 /* Scan all hash chains looking for valid entries that mention X.
2064 If we find one and it is in the wrong hash chain, move it. */
2066 for (i = 0; i < HASH_SIZE; i++)
2067 for (p = table[i]; p; p = next)
2069 next = p->next_same_hash;
2070 if (reg_mentioned_p (x, p->exp)
2071 && exp_equiv_p (p->exp, p->exp, 1, false)
2072 && i != (hash = SAFE_HASH (p->exp, p->mode)))
2074 if (p->next_same_hash)
2075 p->next_same_hash->prev_same_hash = p->prev_same_hash;
2077 if (p->prev_same_hash)
2078 p->prev_same_hash->next_same_hash = p->next_same_hash;
2079 else
2080 table[i] = p->next_same_hash;
2082 p->next_same_hash = table[hash];
2083 p->prev_same_hash = 0;
2084 if (table[hash])
2085 table[hash]->prev_same_hash = p;
2086 table[hash] = p;
2091 /* Remove from the hash table any expression that is a call-clobbered
2092 register. Also update their TICK values. */
2094 static void
2095 invalidate_for_call (void)
2097 unsigned int regno, endregno;
2098 unsigned int i;
2099 unsigned hash;
2100 struct table_elt *p, *next;
2101 int in_table = 0;
2102 hard_reg_set_iterator hrsi;
2104 /* Go through all the hard registers. For each that is clobbered in
2105 a CALL_INSN, remove the register from quantity chains and update
2106 reg_tick if defined. Also see if any of these registers is currently
2107 in the table. */
2108 EXECUTE_IF_SET_IN_HARD_REG_SET (regs_invalidated_by_call, 0, regno, hrsi)
2110 delete_reg_equiv (regno);
2111 if (REG_TICK (regno) >= 0)
2113 REG_TICK (regno)++;
2114 SUBREG_TICKED (regno) = -1;
2116 in_table |= (TEST_HARD_REG_BIT (hard_regs_in_table, regno) != 0);
2119 /* In the case where we have no call-clobbered hard registers in the
2120 table, we are done. Otherwise, scan the table and remove any
2121 entry that overlaps a call-clobbered register. */
2123 if (in_table)
2124 for (hash = 0; hash < HASH_SIZE; hash++)
2125 for (p = table[hash]; p; p = next)
2127 next = p->next_same_hash;
2129 if (!REG_P (p->exp)
2130 || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
2131 continue;
2133 regno = REGNO (p->exp);
2134 endregno = END_HARD_REGNO (p->exp);
2136 for (i = regno; i < endregno; i++)
2137 if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
2139 remove_from_table (p, hash);
2140 break;
2145 /* Given an expression X of type CONST,
2146 and ELT which is its table entry (or 0 if it
2147 is not in the hash table),
2148 return an alternate expression for X as a register plus integer.
2149 If none can be found, return 0. */
2151 static rtx
2152 use_related_value (rtx x, struct table_elt *elt)
2154 struct table_elt *relt = 0;
2155 struct table_elt *p, *q;
2156 HOST_WIDE_INT offset;
2158 /* First, is there anything related known?
2159 If we have a table element, we can tell from that.
2160 Otherwise, must look it up. */
2162 if (elt != 0 && elt->related_value != 0)
2163 relt = elt;
2164 else if (elt == 0 && GET_CODE (x) == CONST)
2166 rtx subexp = get_related_value (x);
2167 if (subexp != 0)
2168 relt = lookup (subexp,
2169 SAFE_HASH (subexp, GET_MODE (subexp)),
2170 GET_MODE (subexp));
2173 if (relt == 0)
2174 return 0;
2176 /* Search all related table entries for one that has an
2177 equivalent register. */
2179 p = relt;
2180 while (1)
2182 /* This loop is strange in that it is executed in two different cases.
2183 The first is when X is already in the table. Then it is searching
2184 the RELATED_VALUE list of X's class (RELT). The second case is when
2185 X is not in the table. Then RELT points to a class for the related
2186 value.
2188 Ensure that, whatever case we are in, that we ignore classes that have
2189 the same value as X. */
2191 if (rtx_equal_p (x, p->exp))
2192 q = 0;
2193 else
2194 for (q = p->first_same_value; q; q = q->next_same_value)
2195 if (REG_P (q->exp))
2196 break;
2198 if (q)
2199 break;
2201 p = p->related_value;
2203 /* We went all the way around, so there is nothing to be found.
2204 Alternatively, perhaps RELT was in the table for some other reason
2205 and it has no related values recorded. */
2206 if (p == relt || p == 0)
2207 break;
2210 if (q == 0)
2211 return 0;
2213 offset = (get_integer_term (x) - get_integer_term (p->exp));
2214 /* Note: OFFSET may be 0 if P->xexp and X are related by commutativity. */
2215 return plus_constant (q->mode, q->exp, offset);
2219 /* Hash a string. Just add its bytes up. */
2220 static inline unsigned
2221 hash_rtx_string (const char *ps)
2223 unsigned hash = 0;
2224 const unsigned char *p = (const unsigned char *) ps;
2226 if (p)
2227 while (*p)
2228 hash += *p++;
2230 return hash;
2233 /* Same as hash_rtx, but call CB on each rtx if it is not NULL.
2234 When the callback returns true, we continue with the new rtx. */
2236 unsigned
2237 hash_rtx_cb (const_rtx x, enum machine_mode mode,
2238 int *do_not_record_p, int *hash_arg_in_memory_p,
2239 bool have_reg_qty, hash_rtx_callback_function cb)
2241 int i, j;
2242 unsigned hash = 0;
2243 enum rtx_code code;
2244 const char *fmt;
2245 enum machine_mode newmode;
2246 rtx newx;
2248 /* Used to turn recursion into iteration. We can't rely on GCC's
2249 tail-recursion elimination since we need to keep accumulating values
2250 in HASH. */
2251 repeat:
2252 if (x == 0)
2253 return hash;
2255 /* Invoke the callback first. */
2256 if (cb != NULL
2257 && ((*cb) (x, mode, &newx, &newmode)))
2259 hash += hash_rtx_cb (newx, newmode, do_not_record_p,
2260 hash_arg_in_memory_p, have_reg_qty, cb);
2261 return hash;
2264 code = GET_CODE (x);
2265 switch (code)
2267 case REG:
2269 unsigned int regno = REGNO (x);
2271 if (do_not_record_p && !reload_completed)
2273 /* On some machines, we can't record any non-fixed hard register,
2274 because extending its life will cause reload problems. We
2275 consider ap, fp, sp, gp to be fixed for this purpose.
2277 We also consider CCmode registers to be fixed for this purpose;
2278 failure to do so leads to failure to simplify 0<100 type of
2279 conditionals.
2281 On all machines, we can't record any global registers.
2282 Nor should we record any register that is in a small
2283 class, as defined by TARGET_CLASS_LIKELY_SPILLED_P. */
2284 bool record;
2286 if (regno >= FIRST_PSEUDO_REGISTER)
2287 record = true;
2288 else if (x == frame_pointer_rtx
2289 || x == hard_frame_pointer_rtx
2290 || x == arg_pointer_rtx
2291 || x == stack_pointer_rtx
2292 || x == pic_offset_table_rtx)
2293 record = true;
2294 else if (global_regs[regno])
2295 record = false;
2296 else if (fixed_regs[regno])
2297 record = true;
2298 else if (GET_MODE_CLASS (GET_MODE (x)) == MODE_CC)
2299 record = true;
2300 else if (targetm.small_register_classes_for_mode_p (GET_MODE (x)))
2301 record = false;
2302 else if (targetm.class_likely_spilled_p (REGNO_REG_CLASS (regno)))
2303 record = false;
2304 else
2305 record = true;
2307 if (!record)
2309 *do_not_record_p = 1;
2310 return 0;
2314 hash += ((unsigned int) REG << 7);
2315 hash += (have_reg_qty ? (unsigned) REG_QTY (regno) : regno);
2316 return hash;
2319 /* We handle SUBREG of a REG specially because the underlying
2320 reg changes its hash value with every value change; we don't
2321 want to have to forget unrelated subregs when one subreg changes. */
2322 case SUBREG:
2324 if (REG_P (SUBREG_REG (x)))
2326 hash += (((unsigned int) SUBREG << 7)
2327 + REGNO (SUBREG_REG (x))
2328 + (SUBREG_BYTE (x) / UNITS_PER_WORD));
2329 return hash;
2331 break;
2334 case CONST_INT:
2335 hash += (((unsigned int) CONST_INT << 7) + (unsigned int) mode
2336 + (unsigned int) INTVAL (x));
2337 return hash;
2339 case CONST_WIDE_INT:
2340 for (i = 0; i < CONST_WIDE_INT_NUNITS (x); i++)
2341 hash += CONST_WIDE_INT_ELT (x, i);
2342 return hash;
2344 case CONST_DOUBLE:
2345 /* This is like the general case, except that it only counts
2346 the integers representing the constant. */
2347 hash += (unsigned int) code + (unsigned int) GET_MODE (x);
2348 if (TARGET_SUPPORTS_WIDE_INT == 0 && GET_MODE (x) == VOIDmode)
2349 hash += ((unsigned int) CONST_DOUBLE_LOW (x)
2350 + (unsigned int) CONST_DOUBLE_HIGH (x));
2351 else
2352 hash += real_hash (CONST_DOUBLE_REAL_VALUE (x));
2353 return hash;
2355 case CONST_FIXED:
2356 hash += (unsigned int) code + (unsigned int) GET_MODE (x);
2357 hash += fixed_hash (CONST_FIXED_VALUE (x));
2358 return hash;
2360 case CONST_VECTOR:
2362 int units;
2363 rtx elt;
2365 units = CONST_VECTOR_NUNITS (x);
2367 for (i = 0; i < units; ++i)
2369 elt = CONST_VECTOR_ELT (x, i);
2370 hash += hash_rtx_cb (elt, GET_MODE (elt),
2371 do_not_record_p, hash_arg_in_memory_p,
2372 have_reg_qty, cb);
2375 return hash;
2378 /* Assume there is only one rtx object for any given label. */
2379 case LABEL_REF:
2380 /* We don't hash on the address of the CODE_LABEL to avoid bootstrap
2381 differences and differences between each stage's debugging dumps. */
2382 hash += (((unsigned int) LABEL_REF << 7)
2383 + CODE_LABEL_NUMBER (XEXP (x, 0)));
2384 return hash;
2386 case SYMBOL_REF:
2388 /* Don't hash on the symbol's address to avoid bootstrap differences.
2389 Different hash values may cause expressions to be recorded in
2390 different orders and thus different registers to be used in the
2391 final assembler. This also avoids differences in the dump files
2392 between various stages. */
2393 unsigned int h = 0;
2394 const unsigned char *p = (const unsigned char *) XSTR (x, 0);
2396 while (*p)
2397 h += (h << 7) + *p++; /* ??? revisit */
2399 hash += ((unsigned int) SYMBOL_REF << 7) + h;
2400 return hash;
2403 case MEM:
2404 /* We don't record if marked volatile or if BLKmode since we don't
2405 know the size of the move. */
2406 if (do_not_record_p && (MEM_VOLATILE_P (x) || GET_MODE (x) == BLKmode))
2408 *do_not_record_p = 1;
2409 return 0;
2411 if (hash_arg_in_memory_p && !MEM_READONLY_P (x))
2412 *hash_arg_in_memory_p = 1;
2414 /* Now that we have already found this special case,
2415 might as well speed it up as much as possible. */
2416 hash += (unsigned) MEM;
2417 x = XEXP (x, 0);
2418 goto repeat;
2420 case USE:
2421 /* A USE that mentions non-volatile memory needs special
2422 handling since the MEM may be BLKmode which normally
2423 prevents an entry from being made. Pure calls are
2424 marked by a USE which mentions BLKmode memory.
2425 See calls.c:emit_call_1. */
2426 if (MEM_P (XEXP (x, 0))
2427 && ! MEM_VOLATILE_P (XEXP (x, 0)))
2429 hash += (unsigned) USE;
2430 x = XEXP (x, 0);
2432 if (hash_arg_in_memory_p && !MEM_READONLY_P (x))
2433 *hash_arg_in_memory_p = 1;
2435 /* Now that we have already found this special case,
2436 might as well speed it up as much as possible. */
2437 hash += (unsigned) MEM;
2438 x = XEXP (x, 0);
2439 goto repeat;
2441 break;
2443 case PRE_DEC:
2444 case PRE_INC:
2445 case POST_DEC:
2446 case POST_INC:
2447 case PRE_MODIFY:
2448 case POST_MODIFY:
2449 case PC:
2450 case CC0:
2451 case CALL:
2452 case UNSPEC_VOLATILE:
2453 if (do_not_record_p) {
2454 *do_not_record_p = 1;
2455 return 0;
2457 else
2458 return hash;
2459 break;
2461 case ASM_OPERANDS:
2462 if (do_not_record_p && MEM_VOLATILE_P (x))
2464 *do_not_record_p = 1;
2465 return 0;
2467 else
2469 /* We don't want to take the filename and line into account. */
2470 hash += (unsigned) code + (unsigned) GET_MODE (x)
2471 + hash_rtx_string (ASM_OPERANDS_TEMPLATE (x))
2472 + hash_rtx_string (ASM_OPERANDS_OUTPUT_CONSTRAINT (x))
2473 + (unsigned) ASM_OPERANDS_OUTPUT_IDX (x);
2475 if (ASM_OPERANDS_INPUT_LENGTH (x))
2477 for (i = 1; i < ASM_OPERANDS_INPUT_LENGTH (x); i++)
2479 hash += (hash_rtx_cb (ASM_OPERANDS_INPUT (x, i),
2480 GET_MODE (ASM_OPERANDS_INPUT (x, i)),
2481 do_not_record_p, hash_arg_in_memory_p,
2482 have_reg_qty, cb)
2483 + hash_rtx_string
2484 (ASM_OPERANDS_INPUT_CONSTRAINT (x, i)));
2487 hash += hash_rtx_string (ASM_OPERANDS_INPUT_CONSTRAINT (x, 0));
2488 x = ASM_OPERANDS_INPUT (x, 0);
2489 mode = GET_MODE (x);
2490 goto repeat;
2493 return hash;
2495 break;
2497 default:
2498 break;
2501 i = GET_RTX_LENGTH (code) - 1;
2502 hash += (unsigned) code + (unsigned) GET_MODE (x);
2503 fmt = GET_RTX_FORMAT (code);
2504 for (; i >= 0; i--)
2506 switch (fmt[i])
2508 case 'e':
2509 /* If we are about to do the last recursive call
2510 needed at this level, change it into iteration.
2511 This function is called enough to be worth it. */
2512 if (i == 0)
2514 x = XEXP (x, i);
2515 goto repeat;
2518 hash += hash_rtx_cb (XEXP (x, i), VOIDmode, do_not_record_p,
2519 hash_arg_in_memory_p,
2520 have_reg_qty, cb);
2521 break;
2523 case 'E':
2524 for (j = 0; j < XVECLEN (x, i); j++)
2525 hash += hash_rtx_cb (XVECEXP (x, i, j), VOIDmode, do_not_record_p,
2526 hash_arg_in_memory_p,
2527 have_reg_qty, cb);
2528 break;
2530 case 's':
2531 hash += hash_rtx_string (XSTR (x, i));
2532 break;
2534 case 'i':
2535 hash += (unsigned int) XINT (x, i);
2536 break;
2538 case '0': case 't':
2539 /* Unused. */
2540 break;
2542 default:
2543 gcc_unreachable ();
2547 return hash;
2550 /* Hash an rtx. We are careful to make sure the value is never negative.
2551 Equivalent registers hash identically.
2552 MODE is used in hashing for CONST_INTs only;
2553 otherwise the mode of X is used.
2555 Store 1 in DO_NOT_RECORD_P if any subexpression is volatile.
2557 If HASH_ARG_IN_MEMORY_P is not NULL, store 1 in it if X contains
2558 a MEM rtx which does not have the MEM_READONLY_P flag set.
2560 Note that cse_insn knows that the hash code of a MEM expression
2561 is just (int) MEM plus the hash code of the address. */
2563 unsigned
2564 hash_rtx (const_rtx x, enum machine_mode mode, int *do_not_record_p,
2565 int *hash_arg_in_memory_p, bool have_reg_qty)
2567 return hash_rtx_cb (x, mode, do_not_record_p,
2568 hash_arg_in_memory_p, have_reg_qty, NULL);
2571 /* Hash an rtx X for cse via hash_rtx.
2572 Stores 1 in do_not_record if any subexpression is volatile.
2573 Stores 1 in hash_arg_in_memory if X contains a mem rtx which
2574 does not have the MEM_READONLY_P flag set. */
2576 static inline unsigned
2577 canon_hash (rtx x, enum machine_mode mode)
2579 return hash_rtx (x, mode, &do_not_record, &hash_arg_in_memory, true);
2582 /* Like canon_hash but with no side effects, i.e. do_not_record
2583 and hash_arg_in_memory are not changed. */
2585 static inline unsigned
2586 safe_hash (rtx x, enum machine_mode mode)
2588 int dummy_do_not_record;
2589 return hash_rtx (x, mode, &dummy_do_not_record, NULL, true);
2592 /* Return 1 iff X and Y would canonicalize into the same thing,
2593 without actually constructing the canonicalization of either one.
2594 If VALIDATE is nonzero,
2595 we assume X is an expression being processed from the rtl
2596 and Y was found in the hash table. We check register refs
2597 in Y for being marked as valid.
2599 If FOR_GCSE is true, we compare X and Y for equivalence for GCSE. */
2602 exp_equiv_p (const_rtx x, const_rtx y, int validate, bool for_gcse)
2604 int i, j;
2605 enum rtx_code code;
2606 const char *fmt;
2608 /* Note: it is incorrect to assume an expression is equivalent to itself
2609 if VALIDATE is nonzero. */
2610 if (x == y && !validate)
2611 return 1;
2613 if (x == 0 || y == 0)
2614 return x == y;
2616 code = GET_CODE (x);
2617 if (code != GET_CODE (y))
2618 return 0;
2620 /* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent. */
2621 if (GET_MODE (x) != GET_MODE (y))
2622 return 0;
2624 /* MEMs referring to different address space are not equivalent. */
2625 if (code == MEM && MEM_ADDR_SPACE (x) != MEM_ADDR_SPACE (y))
2626 return 0;
2628 switch (code)
2630 case PC:
2631 case CC0:
2632 CASE_CONST_UNIQUE:
2633 return x == y;
2635 case LABEL_REF:
2636 return XEXP (x, 0) == XEXP (y, 0);
2638 case SYMBOL_REF:
2639 return XSTR (x, 0) == XSTR (y, 0);
2641 case REG:
2642 if (for_gcse)
2643 return REGNO (x) == REGNO (y);
2644 else
2646 unsigned int regno = REGNO (y);
2647 unsigned int i;
2648 unsigned int endregno = END_REGNO (y);
2650 /* If the quantities are not the same, the expressions are not
2651 equivalent. If there are and we are not to validate, they
2652 are equivalent. Otherwise, ensure all regs are up-to-date. */
2654 if (REG_QTY (REGNO (x)) != REG_QTY (regno))
2655 return 0;
2657 if (! validate)
2658 return 1;
2660 for (i = regno; i < endregno; i++)
2661 if (REG_IN_TABLE (i) != REG_TICK (i))
2662 return 0;
2664 return 1;
2667 case MEM:
2668 if (for_gcse)
2670 /* A volatile mem should not be considered equivalent to any
2671 other. */
2672 if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
2673 return 0;
2675 /* Can't merge two expressions in different alias sets, since we
2676 can decide that the expression is transparent in a block when
2677 it isn't, due to it being set with the different alias set.
2679 Also, can't merge two expressions with different MEM_ATTRS.
2680 They could e.g. be two different entities allocated into the
2681 same space on the stack (see e.g. PR25130). In that case, the
2682 MEM addresses can be the same, even though the two MEMs are
2683 absolutely not equivalent.
2685 But because really all MEM attributes should be the same for
2686 equivalent MEMs, we just use the invariant that MEMs that have
2687 the same attributes share the same mem_attrs data structure. */
2688 if (!mem_attrs_eq_p (MEM_ATTRS (x), MEM_ATTRS (y)))
2689 return 0;
2691 /* If we are handling exceptions, we cannot consider two expressions
2692 with different trapping status as equivalent, because simple_mem
2693 might accept one and reject the other. */
2694 if (cfun->can_throw_non_call_exceptions
2695 && (MEM_NOTRAP_P (x) != MEM_NOTRAP_P (y)))
2696 return 0;
2698 break;
2700 /* For commutative operations, check both orders. */
2701 case PLUS:
2702 case MULT:
2703 case AND:
2704 case IOR:
2705 case XOR:
2706 case NE:
2707 case EQ:
2708 return ((exp_equiv_p (XEXP (x, 0), XEXP (y, 0),
2709 validate, for_gcse)
2710 && exp_equiv_p (XEXP (x, 1), XEXP (y, 1),
2711 validate, for_gcse))
2712 || (exp_equiv_p (XEXP (x, 0), XEXP (y, 1),
2713 validate, for_gcse)
2714 && exp_equiv_p (XEXP (x, 1), XEXP (y, 0),
2715 validate, for_gcse)));
2717 case ASM_OPERANDS:
2718 /* We don't use the generic code below because we want to
2719 disregard filename and line numbers. */
2721 /* A volatile asm isn't equivalent to any other. */
2722 if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
2723 return 0;
2725 if (GET_MODE (x) != GET_MODE (y)
2726 || strcmp (ASM_OPERANDS_TEMPLATE (x), ASM_OPERANDS_TEMPLATE (y))
2727 || strcmp (ASM_OPERANDS_OUTPUT_CONSTRAINT (x),
2728 ASM_OPERANDS_OUTPUT_CONSTRAINT (y))
2729 || ASM_OPERANDS_OUTPUT_IDX (x) != ASM_OPERANDS_OUTPUT_IDX (y)
2730 || ASM_OPERANDS_INPUT_LENGTH (x) != ASM_OPERANDS_INPUT_LENGTH (y))
2731 return 0;
2733 if (ASM_OPERANDS_INPUT_LENGTH (x))
2735 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
2736 if (! exp_equiv_p (ASM_OPERANDS_INPUT (x, i),
2737 ASM_OPERANDS_INPUT (y, i),
2738 validate, for_gcse)
2739 || strcmp (ASM_OPERANDS_INPUT_CONSTRAINT (x, i),
2740 ASM_OPERANDS_INPUT_CONSTRAINT (y, i)))
2741 return 0;
2744 return 1;
2746 default:
2747 break;
2750 /* Compare the elements. If any pair of corresponding elements
2751 fail to match, return 0 for the whole thing. */
2753 fmt = GET_RTX_FORMAT (code);
2754 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2756 switch (fmt[i])
2758 case 'e':
2759 if (! exp_equiv_p (XEXP (x, i), XEXP (y, i),
2760 validate, for_gcse))
2761 return 0;
2762 break;
2764 case 'E':
2765 if (XVECLEN (x, i) != XVECLEN (y, i))
2766 return 0;
2767 for (j = 0; j < XVECLEN (x, i); j++)
2768 if (! exp_equiv_p (XVECEXP (x, i, j), XVECEXP (y, i, j),
2769 validate, for_gcse))
2770 return 0;
2771 break;
2773 case 's':
2774 if (strcmp (XSTR (x, i), XSTR (y, i)))
2775 return 0;
2776 break;
2778 case 'i':
2779 if (XINT (x, i) != XINT (y, i))
2780 return 0;
2781 break;
2783 case 'w':
2784 if (XWINT (x, i) != XWINT (y, i))
2785 return 0;
2786 break;
2788 case '0':
2789 case 't':
2790 break;
2792 default:
2793 gcc_unreachable ();
2797 return 1;
2800 /* Subroutine of canon_reg. Pass *XLOC through canon_reg, and validate
2801 the result if necessary. INSN is as for canon_reg. */
2803 static void
2804 validate_canon_reg (rtx *xloc, rtx insn)
2806 if (*xloc)
2808 rtx new_rtx = canon_reg (*xloc, insn);
2810 /* If replacing pseudo with hard reg or vice versa, ensure the
2811 insn remains valid. Likewise if the insn has MATCH_DUPs. */
2812 gcc_assert (insn && new_rtx);
2813 validate_change (insn, xloc, new_rtx, 1);
2817 /* Canonicalize an expression:
2818 replace each register reference inside it
2819 with the "oldest" equivalent register.
2821 If INSN is nonzero validate_change is used to ensure that INSN remains valid
2822 after we make our substitution. The calls are made with IN_GROUP nonzero
2823 so apply_change_group must be called upon the outermost return from this
2824 function (unless INSN is zero). The result of apply_change_group can
2825 generally be discarded since the changes we are making are optional. */
2827 static rtx
2828 canon_reg (rtx x, rtx insn)
2830 int i;
2831 enum rtx_code code;
2832 const char *fmt;
2834 if (x == 0)
2835 return x;
2837 code = GET_CODE (x);
2838 switch (code)
2840 case PC:
2841 case CC0:
2842 case CONST:
2843 CASE_CONST_ANY:
2844 case SYMBOL_REF:
2845 case LABEL_REF:
2846 case ADDR_VEC:
2847 case ADDR_DIFF_VEC:
2848 return x;
2850 case REG:
2852 int first;
2853 int q;
2854 struct qty_table_elem *ent;
2856 /* Never replace a hard reg, because hard regs can appear
2857 in more than one machine mode, and we must preserve the mode
2858 of each occurrence. Also, some hard regs appear in
2859 MEMs that are shared and mustn't be altered. Don't try to
2860 replace any reg that maps to a reg of class NO_REGS. */
2861 if (REGNO (x) < FIRST_PSEUDO_REGISTER
2862 || ! REGNO_QTY_VALID_P (REGNO (x)))
2863 return x;
2865 q = REG_QTY (REGNO (x));
2866 ent = &qty_table[q];
2867 first = ent->first_reg;
2868 return (first >= FIRST_PSEUDO_REGISTER ? regno_reg_rtx[first]
2869 : REGNO_REG_CLASS (first) == NO_REGS ? x
2870 : gen_rtx_REG (ent->mode, first));
2873 default:
2874 break;
2877 fmt = GET_RTX_FORMAT (code);
2878 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2880 int j;
2882 if (fmt[i] == 'e')
2883 validate_canon_reg (&XEXP (x, i), insn);
2884 else if (fmt[i] == 'E')
2885 for (j = 0; j < XVECLEN (x, i); j++)
2886 validate_canon_reg (&XVECEXP (x, i, j), insn);
2889 return x;
2892 /* Given an operation (CODE, *PARG1, *PARG2), where code is a comparison
2893 operation (EQ, NE, GT, etc.), follow it back through the hash table and
2894 what values are being compared.
2896 *PARG1 and *PARG2 are updated to contain the rtx representing the values
2897 actually being compared. For example, if *PARG1 was (cc0) and *PARG2
2898 was (const_int 0), *PARG1 and *PARG2 will be set to the objects that were
2899 compared to produce cc0.
2901 The return value is the comparison operator and is either the code of
2902 A or the code corresponding to the inverse of the comparison. */
2904 static enum rtx_code
2905 find_comparison_args (enum rtx_code code, rtx *parg1, rtx *parg2,
2906 enum machine_mode *pmode1, enum machine_mode *pmode2)
2908 rtx arg1, arg2;
2909 hash_set<rtx> *visited = NULL;
2910 /* Set nonzero when we find something of interest. */
2911 rtx x = NULL;
2913 arg1 = *parg1, arg2 = *parg2;
2915 /* If ARG2 is const0_rtx, see what ARG1 is equivalent to. */
2917 while (arg2 == CONST0_RTX (GET_MODE (arg1)))
2919 int reverse_code = 0;
2920 struct table_elt *p = 0;
2922 /* Remember state from previous iteration. */
2923 if (x)
2925 if (!visited)
2926 visited = new hash_set<rtx>;
2927 visited->add (x);
2928 x = 0;
2931 /* If arg1 is a COMPARE, extract the comparison arguments from it.
2932 On machines with CC0, this is the only case that can occur, since
2933 fold_rtx will return the COMPARE or item being compared with zero
2934 when given CC0. */
2936 if (GET_CODE (arg1) == COMPARE && arg2 == const0_rtx)
2937 x = arg1;
2939 /* If ARG1 is a comparison operator and CODE is testing for
2940 STORE_FLAG_VALUE, get the inner arguments. */
2942 else if (COMPARISON_P (arg1))
2944 #ifdef FLOAT_STORE_FLAG_VALUE
2945 REAL_VALUE_TYPE fsfv;
2946 #endif
2948 if (code == NE
2949 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
2950 && code == LT && STORE_FLAG_VALUE == -1)
2951 #ifdef FLOAT_STORE_FLAG_VALUE
2952 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
2953 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
2954 REAL_VALUE_NEGATIVE (fsfv)))
2955 #endif
2957 x = arg1;
2958 else if (code == EQ
2959 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
2960 && code == GE && STORE_FLAG_VALUE == -1)
2961 #ifdef FLOAT_STORE_FLAG_VALUE
2962 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
2963 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
2964 REAL_VALUE_NEGATIVE (fsfv)))
2965 #endif
2967 x = arg1, reverse_code = 1;
2970 /* ??? We could also check for
2972 (ne (and (eq (...) (const_int 1))) (const_int 0))
2974 and related forms, but let's wait until we see them occurring. */
2976 if (x == 0)
2977 /* Look up ARG1 in the hash table and see if it has an equivalence
2978 that lets us see what is being compared. */
2979 p = lookup (arg1, SAFE_HASH (arg1, GET_MODE (arg1)), GET_MODE (arg1));
2980 if (p)
2982 p = p->first_same_value;
2984 /* If what we compare is already known to be constant, that is as
2985 good as it gets.
2986 We need to break the loop in this case, because otherwise we
2987 can have an infinite loop when looking at a reg that is known
2988 to be a constant which is the same as a comparison of a reg
2989 against zero which appears later in the insn stream, which in
2990 turn is constant and the same as the comparison of the first reg
2991 against zero... */
2992 if (p->is_const)
2993 break;
2996 for (; p; p = p->next_same_value)
2998 enum machine_mode inner_mode = GET_MODE (p->exp);
2999 #ifdef FLOAT_STORE_FLAG_VALUE
3000 REAL_VALUE_TYPE fsfv;
3001 #endif
3003 /* If the entry isn't valid, skip it. */
3004 if (! exp_equiv_p (p->exp, p->exp, 1, false))
3005 continue;
3007 /* If it's a comparison we've used before, skip it. */
3008 if (visited && visited->contains (p->exp))
3009 continue;
3011 if (GET_CODE (p->exp) == COMPARE
3012 /* Another possibility is that this machine has a compare insn
3013 that includes the comparison code. In that case, ARG1 would
3014 be equivalent to a comparison operation that would set ARG1 to
3015 either STORE_FLAG_VALUE or zero. If this is an NE operation,
3016 ORIG_CODE is the actual comparison being done; if it is an EQ,
3017 we must reverse ORIG_CODE. On machine with a negative value
3018 for STORE_FLAG_VALUE, also look at LT and GE operations. */
3019 || ((code == NE
3020 || (code == LT
3021 && val_signbit_known_set_p (inner_mode,
3022 STORE_FLAG_VALUE))
3023 #ifdef FLOAT_STORE_FLAG_VALUE
3024 || (code == LT
3025 && SCALAR_FLOAT_MODE_P (inner_mode)
3026 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3027 REAL_VALUE_NEGATIVE (fsfv)))
3028 #endif
3030 && COMPARISON_P (p->exp)))
3032 x = p->exp;
3033 break;
3035 else if ((code == EQ
3036 || (code == GE
3037 && val_signbit_known_set_p (inner_mode,
3038 STORE_FLAG_VALUE))
3039 #ifdef FLOAT_STORE_FLAG_VALUE
3040 || (code == GE
3041 && SCALAR_FLOAT_MODE_P (inner_mode)
3042 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3043 REAL_VALUE_NEGATIVE (fsfv)))
3044 #endif
3046 && COMPARISON_P (p->exp))
3048 reverse_code = 1;
3049 x = p->exp;
3050 break;
3053 /* If this non-trapping address, e.g. fp + constant, the
3054 equivalent is a better operand since it may let us predict
3055 the value of the comparison. */
3056 else if (!rtx_addr_can_trap_p (p->exp))
3058 arg1 = p->exp;
3059 continue;
3063 /* If we didn't find a useful equivalence for ARG1, we are done.
3064 Otherwise, set up for the next iteration. */
3065 if (x == 0)
3066 break;
3068 /* If we need to reverse the comparison, make sure that that is
3069 possible -- we can't necessarily infer the value of GE from LT
3070 with floating-point operands. */
3071 if (reverse_code)
3073 enum rtx_code reversed = reversed_comparison_code (x, NULL_RTX);
3074 if (reversed == UNKNOWN)
3075 break;
3076 else
3077 code = reversed;
3079 else if (COMPARISON_P (x))
3080 code = GET_CODE (x);
3081 arg1 = XEXP (x, 0), arg2 = XEXP (x, 1);
3084 /* Return our results. Return the modes from before fold_rtx
3085 because fold_rtx might produce const_int, and then it's too late. */
3086 *pmode1 = GET_MODE (arg1), *pmode2 = GET_MODE (arg2);
3087 *parg1 = fold_rtx (arg1, 0), *parg2 = fold_rtx (arg2, 0);
3089 if (visited)
3090 delete visited;
3091 return code;
3094 /* If X is a nontrivial arithmetic operation on an argument for which
3095 a constant value can be determined, return the result of operating
3096 on that value, as a constant. Otherwise, return X, possibly with
3097 one or more operands changed to a forward-propagated constant.
3099 If X is a register whose contents are known, we do NOT return
3100 those contents here; equiv_constant is called to perform that task.
3101 For SUBREGs and MEMs, we do that both here and in equiv_constant.
3103 INSN is the insn that we may be modifying. If it is 0, make a copy
3104 of X before modifying it. */
3106 static rtx
3107 fold_rtx (rtx x, rtx insn)
3109 enum rtx_code code;
3110 enum machine_mode mode;
3111 const char *fmt;
3112 int i;
3113 rtx new_rtx = 0;
3114 int changed = 0;
3116 /* Operands of X. */
3117 rtx folded_arg0;
3118 rtx folded_arg1;
3120 /* Constant equivalents of first three operands of X;
3121 0 when no such equivalent is known. */
3122 rtx const_arg0;
3123 rtx const_arg1;
3124 rtx const_arg2;
3126 /* The mode of the first operand of X. We need this for sign and zero
3127 extends. */
3128 enum machine_mode mode_arg0;
3130 if (x == 0)
3131 return x;
3133 /* Try to perform some initial simplifications on X. */
3134 code = GET_CODE (x);
3135 switch (code)
3137 case MEM:
3138 case SUBREG:
3139 if ((new_rtx = equiv_constant (x)) != NULL_RTX)
3140 return new_rtx;
3141 return x;
3143 case CONST:
3144 CASE_CONST_ANY:
3145 case SYMBOL_REF:
3146 case LABEL_REF:
3147 case REG:
3148 case PC:
3149 /* No use simplifying an EXPR_LIST
3150 since they are used only for lists of args
3151 in a function call's REG_EQUAL note. */
3152 case EXPR_LIST:
3153 return x;
3155 #ifdef HAVE_cc0
3156 case CC0:
3157 return prev_insn_cc0;
3158 #endif
3160 case ASM_OPERANDS:
3161 if (insn)
3163 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
3164 validate_change (insn, &ASM_OPERANDS_INPUT (x, i),
3165 fold_rtx (ASM_OPERANDS_INPUT (x, i), insn), 0);
3167 return x;
3169 #ifdef NO_FUNCTION_CSE
3170 case CALL:
3171 if (CONSTANT_P (XEXP (XEXP (x, 0), 0)))
3172 return x;
3173 break;
3174 #endif
3176 /* Anything else goes through the loop below. */
3177 default:
3178 break;
3181 mode = GET_MODE (x);
3182 const_arg0 = 0;
3183 const_arg1 = 0;
3184 const_arg2 = 0;
3185 mode_arg0 = VOIDmode;
3187 /* Try folding our operands.
3188 Then see which ones have constant values known. */
3190 fmt = GET_RTX_FORMAT (code);
3191 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
3192 if (fmt[i] == 'e')
3194 rtx folded_arg = XEXP (x, i), const_arg;
3195 enum machine_mode mode_arg = GET_MODE (folded_arg);
3197 switch (GET_CODE (folded_arg))
3199 case MEM:
3200 case REG:
3201 case SUBREG:
3202 const_arg = equiv_constant (folded_arg);
3203 break;
3205 case CONST:
3206 CASE_CONST_ANY:
3207 case SYMBOL_REF:
3208 case LABEL_REF:
3209 const_arg = folded_arg;
3210 break;
3212 #ifdef HAVE_cc0
3213 case CC0:
3214 /* The cc0-user and cc0-setter may be in different blocks if
3215 the cc0-setter potentially traps. In that case PREV_INSN_CC0
3216 will have been cleared as we exited the block with the
3217 setter.
3219 While we could potentially track cc0 in this case, it just
3220 doesn't seem to be worth it given that cc0 targets are not
3221 terribly common or important these days and trapping math
3222 is rarely used. The combination of those two conditions
3223 necessary to trip this situation is exceedingly rare in the
3224 real world. */
3225 if (!prev_insn_cc0)
3227 const_arg = NULL_RTX;
3229 else
3231 folded_arg = prev_insn_cc0;
3232 mode_arg = prev_insn_cc0_mode;
3233 const_arg = equiv_constant (folded_arg);
3235 break;
3236 #endif
3238 default:
3239 folded_arg = fold_rtx (folded_arg, insn);
3240 const_arg = equiv_constant (folded_arg);
3241 break;
3244 /* For the first three operands, see if the operand
3245 is constant or equivalent to a constant. */
3246 switch (i)
3248 case 0:
3249 folded_arg0 = folded_arg;
3250 const_arg0 = const_arg;
3251 mode_arg0 = mode_arg;
3252 break;
3253 case 1:
3254 folded_arg1 = folded_arg;
3255 const_arg1 = const_arg;
3256 break;
3257 case 2:
3258 const_arg2 = const_arg;
3259 break;
3262 /* Pick the least expensive of the argument and an equivalent constant
3263 argument. */
3264 if (const_arg != 0
3265 && const_arg != folded_arg
3266 && COST_IN (const_arg, code, i) <= COST_IN (folded_arg, code, i)
3268 /* It's not safe to substitute the operand of a conversion
3269 operator with a constant, as the conversion's identity
3270 depends upon the mode of its operand. This optimization
3271 is handled by the call to simplify_unary_operation. */
3272 && (GET_RTX_CLASS (code) != RTX_UNARY
3273 || GET_MODE (const_arg) == mode_arg0
3274 || (code != ZERO_EXTEND
3275 && code != SIGN_EXTEND
3276 && code != TRUNCATE
3277 && code != FLOAT_TRUNCATE
3278 && code != FLOAT_EXTEND
3279 && code != FLOAT
3280 && code != FIX
3281 && code != UNSIGNED_FLOAT
3282 && code != UNSIGNED_FIX)))
3283 folded_arg = const_arg;
3285 if (folded_arg == XEXP (x, i))
3286 continue;
3288 if (insn == NULL_RTX && !changed)
3289 x = copy_rtx (x);
3290 changed = 1;
3291 validate_unshare_change (insn, &XEXP (x, i), folded_arg, 1);
3294 if (changed)
3296 /* Canonicalize X if necessary, and keep const_argN and folded_argN
3297 consistent with the order in X. */
3298 if (canonicalize_change_group (insn, x))
3300 rtx tem;
3301 tem = const_arg0, const_arg0 = const_arg1, const_arg1 = tem;
3302 tem = folded_arg0, folded_arg0 = folded_arg1, folded_arg1 = tem;
3305 apply_change_group ();
3308 /* If X is an arithmetic operation, see if we can simplify it. */
3310 switch (GET_RTX_CLASS (code))
3312 case RTX_UNARY:
3314 /* We can't simplify extension ops unless we know the
3315 original mode. */
3316 if ((code == ZERO_EXTEND || code == SIGN_EXTEND)
3317 && mode_arg0 == VOIDmode)
3318 break;
3320 new_rtx = simplify_unary_operation (code, mode,
3321 const_arg0 ? const_arg0 : folded_arg0,
3322 mode_arg0);
3324 break;
3326 case RTX_COMPARE:
3327 case RTX_COMM_COMPARE:
3328 /* See what items are actually being compared and set FOLDED_ARG[01]
3329 to those values and CODE to the actual comparison code. If any are
3330 constant, set CONST_ARG0 and CONST_ARG1 appropriately. We needn't
3331 do anything if both operands are already known to be constant. */
3333 /* ??? Vector mode comparisons are not supported yet. */
3334 if (VECTOR_MODE_P (mode))
3335 break;
3337 if (const_arg0 == 0 || const_arg1 == 0)
3339 struct table_elt *p0, *p1;
3340 rtx true_rtx, false_rtx;
3341 enum machine_mode mode_arg1;
3343 if (SCALAR_FLOAT_MODE_P (mode))
3345 #ifdef FLOAT_STORE_FLAG_VALUE
3346 true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
3347 (FLOAT_STORE_FLAG_VALUE (mode), mode));
3348 #else
3349 true_rtx = NULL_RTX;
3350 #endif
3351 false_rtx = CONST0_RTX (mode);
3353 else
3355 true_rtx = const_true_rtx;
3356 false_rtx = const0_rtx;
3359 code = find_comparison_args (code, &folded_arg0, &folded_arg1,
3360 &mode_arg0, &mode_arg1);
3362 /* If the mode is VOIDmode or a MODE_CC mode, we don't know
3363 what kinds of things are being compared, so we can't do
3364 anything with this comparison. */
3366 if (mode_arg0 == VOIDmode || GET_MODE_CLASS (mode_arg0) == MODE_CC)
3367 break;
3369 const_arg0 = equiv_constant (folded_arg0);
3370 const_arg1 = equiv_constant (folded_arg1);
3372 /* If we do not now have two constants being compared, see
3373 if we can nevertheless deduce some things about the
3374 comparison. */
3375 if (const_arg0 == 0 || const_arg1 == 0)
3377 if (const_arg1 != NULL)
3379 rtx cheapest_simplification;
3380 int cheapest_cost;
3381 rtx simp_result;
3382 struct table_elt *p;
3384 /* See if we can find an equivalent of folded_arg0
3385 that gets us a cheaper expression, possibly a
3386 constant through simplifications. */
3387 p = lookup (folded_arg0, SAFE_HASH (folded_arg0, mode_arg0),
3388 mode_arg0);
3390 if (p != NULL)
3392 cheapest_simplification = x;
3393 cheapest_cost = COST (x);
3395 for (p = p->first_same_value; p != NULL; p = p->next_same_value)
3397 int cost;
3399 /* If the entry isn't valid, skip it. */
3400 if (! exp_equiv_p (p->exp, p->exp, 1, false))
3401 continue;
3403 /* Try to simplify using this equivalence. */
3404 simp_result
3405 = simplify_relational_operation (code, mode,
3406 mode_arg0,
3407 p->exp,
3408 const_arg1);
3410 if (simp_result == NULL)
3411 continue;
3413 cost = COST (simp_result);
3414 if (cost < cheapest_cost)
3416 cheapest_cost = cost;
3417 cheapest_simplification = simp_result;
3421 /* If we have a cheaper expression now, use that
3422 and try folding it further, from the top. */
3423 if (cheapest_simplification != x)
3424 return fold_rtx (copy_rtx (cheapest_simplification),
3425 insn);
3429 /* See if the two operands are the same. */
3431 if ((REG_P (folded_arg0)
3432 && REG_P (folded_arg1)
3433 && (REG_QTY (REGNO (folded_arg0))
3434 == REG_QTY (REGNO (folded_arg1))))
3435 || ((p0 = lookup (folded_arg0,
3436 SAFE_HASH (folded_arg0, mode_arg0),
3437 mode_arg0))
3438 && (p1 = lookup (folded_arg1,
3439 SAFE_HASH (folded_arg1, mode_arg0),
3440 mode_arg0))
3441 && p0->first_same_value == p1->first_same_value))
3442 folded_arg1 = folded_arg0;
3444 /* If FOLDED_ARG0 is a register, see if the comparison we are
3445 doing now is either the same as we did before or the reverse
3446 (we only check the reverse if not floating-point). */
3447 else if (REG_P (folded_arg0))
3449 int qty = REG_QTY (REGNO (folded_arg0));
3451 if (REGNO_QTY_VALID_P (REGNO (folded_arg0)))
3453 struct qty_table_elem *ent = &qty_table[qty];
3455 if ((comparison_dominates_p (ent->comparison_code, code)
3456 || (! FLOAT_MODE_P (mode_arg0)
3457 && comparison_dominates_p (ent->comparison_code,
3458 reverse_condition (code))))
3459 && (rtx_equal_p (ent->comparison_const, folded_arg1)
3460 || (const_arg1
3461 && rtx_equal_p (ent->comparison_const,
3462 const_arg1))
3463 || (REG_P (folded_arg1)
3464 && (REG_QTY (REGNO (folded_arg1)) == ent->comparison_qty))))
3466 if (comparison_dominates_p (ent->comparison_code, code))
3468 if (true_rtx)
3469 return true_rtx;
3470 else
3471 break;
3473 else
3474 return false_rtx;
3481 /* If we are comparing against zero, see if the first operand is
3482 equivalent to an IOR with a constant. If so, we may be able to
3483 determine the result of this comparison. */
3484 if (const_arg1 == const0_rtx && !const_arg0)
3486 rtx y = lookup_as_function (folded_arg0, IOR);
3487 rtx inner_const;
3489 if (y != 0
3490 && (inner_const = equiv_constant (XEXP (y, 1))) != 0
3491 && CONST_INT_P (inner_const)
3492 && INTVAL (inner_const) != 0)
3493 folded_arg0 = gen_rtx_IOR (mode_arg0, XEXP (y, 0), inner_const);
3497 rtx op0 = const_arg0 ? const_arg0 : copy_rtx (folded_arg0);
3498 rtx op1 = const_arg1 ? const_arg1 : copy_rtx (folded_arg1);
3499 new_rtx = simplify_relational_operation (code, mode, mode_arg0,
3500 op0, op1);
3502 break;
3504 case RTX_BIN_ARITH:
3505 case RTX_COMM_ARITH:
3506 switch (code)
3508 case PLUS:
3509 /* If the second operand is a LABEL_REF, see if the first is a MINUS
3510 with that LABEL_REF as its second operand. If so, the result is
3511 the first operand of that MINUS. This handles switches with an
3512 ADDR_DIFF_VEC table. */
3513 if (const_arg1 && GET_CODE (const_arg1) == LABEL_REF)
3515 rtx y
3516 = GET_CODE (folded_arg0) == MINUS ? folded_arg0
3517 : lookup_as_function (folded_arg0, MINUS);
3519 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
3520 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg1, 0))
3521 return XEXP (y, 0);
3523 /* Now try for a CONST of a MINUS like the above. */
3524 if ((y = (GET_CODE (folded_arg0) == CONST ? folded_arg0
3525 : lookup_as_function (folded_arg0, CONST))) != 0
3526 && GET_CODE (XEXP (y, 0)) == MINUS
3527 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
3528 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg1, 0))
3529 return XEXP (XEXP (y, 0), 0);
3532 /* Likewise if the operands are in the other order. */
3533 if (const_arg0 && GET_CODE (const_arg0) == LABEL_REF)
3535 rtx y
3536 = GET_CODE (folded_arg1) == MINUS ? folded_arg1
3537 : lookup_as_function (folded_arg1, MINUS);
3539 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
3540 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg0, 0))
3541 return XEXP (y, 0);
3543 /* Now try for a CONST of a MINUS like the above. */
3544 if ((y = (GET_CODE (folded_arg1) == CONST ? folded_arg1
3545 : lookup_as_function (folded_arg1, CONST))) != 0
3546 && GET_CODE (XEXP (y, 0)) == MINUS
3547 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
3548 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg0, 0))
3549 return XEXP (XEXP (y, 0), 0);
3552 /* If second operand is a register equivalent to a negative
3553 CONST_INT, see if we can find a register equivalent to the
3554 positive constant. Make a MINUS if so. Don't do this for
3555 a non-negative constant since we might then alternate between
3556 choosing positive and negative constants. Having the positive
3557 constant previously-used is the more common case. Be sure
3558 the resulting constant is non-negative; if const_arg1 were
3559 the smallest negative number this would overflow: depending
3560 on the mode, this would either just be the same value (and
3561 hence not save anything) or be incorrect. */
3562 if (const_arg1 != 0 && CONST_INT_P (const_arg1)
3563 && INTVAL (const_arg1) < 0
3564 /* This used to test
3566 -INTVAL (const_arg1) >= 0
3568 But The Sun V5.0 compilers mis-compiled that test. So
3569 instead we test for the problematic value in a more direct
3570 manner and hope the Sun compilers get it correct. */
3571 && INTVAL (const_arg1) !=
3572 ((HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1))
3573 && REG_P (folded_arg1))
3575 rtx new_const = GEN_INT (-INTVAL (const_arg1));
3576 struct table_elt *p
3577 = lookup (new_const, SAFE_HASH (new_const, mode), mode);
3579 if (p)
3580 for (p = p->first_same_value; p; p = p->next_same_value)
3581 if (REG_P (p->exp))
3582 return simplify_gen_binary (MINUS, mode, folded_arg0,
3583 canon_reg (p->exp, NULL_RTX));
3585 goto from_plus;
3587 case MINUS:
3588 /* If we have (MINUS Y C), see if Y is known to be (PLUS Z C2).
3589 If so, produce (PLUS Z C2-C). */
3590 if (const_arg1 != 0 && CONST_INT_P (const_arg1))
3592 rtx y = lookup_as_function (XEXP (x, 0), PLUS);
3593 if (y && CONST_INT_P (XEXP (y, 1)))
3594 return fold_rtx (plus_constant (mode, copy_rtx (y),
3595 -INTVAL (const_arg1)),
3596 NULL_RTX);
3599 /* Fall through. */
3601 from_plus:
3602 case SMIN: case SMAX: case UMIN: case UMAX:
3603 case IOR: case AND: case XOR:
3604 case MULT:
3605 case ASHIFT: case LSHIFTRT: case ASHIFTRT:
3606 /* If we have (<op> <reg> <const_int>) for an associative OP and REG
3607 is known to be of similar form, we may be able to replace the
3608 operation with a combined operation. This may eliminate the
3609 intermediate operation if every use is simplified in this way.
3610 Note that the similar optimization done by combine.c only works
3611 if the intermediate operation's result has only one reference. */
3613 if (REG_P (folded_arg0)
3614 && const_arg1 && CONST_INT_P (const_arg1))
3616 int is_shift
3617 = (code == ASHIFT || code == ASHIFTRT || code == LSHIFTRT);
3618 rtx y, inner_const, new_const;
3619 rtx canon_const_arg1 = const_arg1;
3620 enum rtx_code associate_code;
3622 if (is_shift
3623 && (INTVAL (const_arg1) >= GET_MODE_PRECISION (mode)
3624 || INTVAL (const_arg1) < 0))
3626 if (SHIFT_COUNT_TRUNCATED)
3627 canon_const_arg1 = GEN_INT (INTVAL (const_arg1)
3628 & (GET_MODE_BITSIZE (mode)
3629 - 1));
3630 else
3631 break;
3634 y = lookup_as_function (folded_arg0, code);
3635 if (y == 0)
3636 break;
3638 /* If we have compiled a statement like
3639 "if (x == (x & mask1))", and now are looking at
3640 "x & mask2", we will have a case where the first operand
3641 of Y is the same as our first operand. Unless we detect
3642 this case, an infinite loop will result. */
3643 if (XEXP (y, 0) == folded_arg0)
3644 break;
3646 inner_const = equiv_constant (fold_rtx (XEXP (y, 1), 0));
3647 if (!inner_const || !CONST_INT_P (inner_const))
3648 break;
3650 /* Don't associate these operations if they are a PLUS with the
3651 same constant and it is a power of two. These might be doable
3652 with a pre- or post-increment. Similarly for two subtracts of
3653 identical powers of two with post decrement. */
3655 if (code == PLUS && const_arg1 == inner_const
3656 && ((HAVE_PRE_INCREMENT
3657 && exact_log2 (INTVAL (const_arg1)) >= 0)
3658 || (HAVE_POST_INCREMENT
3659 && exact_log2 (INTVAL (const_arg1)) >= 0)
3660 || (HAVE_PRE_DECREMENT
3661 && exact_log2 (- INTVAL (const_arg1)) >= 0)
3662 || (HAVE_POST_DECREMENT
3663 && exact_log2 (- INTVAL (const_arg1)) >= 0)))
3664 break;
3666 /* ??? Vector mode shifts by scalar
3667 shift operand are not supported yet. */
3668 if (is_shift && VECTOR_MODE_P (mode))
3669 break;
3671 if (is_shift
3672 && (INTVAL (inner_const) >= GET_MODE_PRECISION (mode)
3673 || INTVAL (inner_const) < 0))
3675 if (SHIFT_COUNT_TRUNCATED)
3676 inner_const = GEN_INT (INTVAL (inner_const)
3677 & (GET_MODE_BITSIZE (mode) - 1));
3678 else
3679 break;
3682 /* Compute the code used to compose the constants. For example,
3683 A-C1-C2 is A-(C1 + C2), so if CODE == MINUS, we want PLUS. */
3685 associate_code = (is_shift || code == MINUS ? PLUS : code);
3687 new_const = simplify_binary_operation (associate_code, mode,
3688 canon_const_arg1,
3689 inner_const);
3691 if (new_const == 0)
3692 break;
3694 /* If we are associating shift operations, don't let this
3695 produce a shift of the size of the object or larger.
3696 This could occur when we follow a sign-extend by a right
3697 shift on a machine that does a sign-extend as a pair
3698 of shifts. */
3700 if (is_shift
3701 && CONST_INT_P (new_const)
3702 && INTVAL (new_const) >= GET_MODE_PRECISION (mode))
3704 /* As an exception, we can turn an ASHIFTRT of this
3705 form into a shift of the number of bits - 1. */
3706 if (code == ASHIFTRT)
3707 new_const = GEN_INT (GET_MODE_BITSIZE (mode) - 1);
3708 else if (!side_effects_p (XEXP (y, 0)))
3709 return CONST0_RTX (mode);
3710 else
3711 break;
3714 y = copy_rtx (XEXP (y, 0));
3716 /* If Y contains our first operand (the most common way this
3717 can happen is if Y is a MEM), we would do into an infinite
3718 loop if we tried to fold it. So don't in that case. */
3720 if (! reg_mentioned_p (folded_arg0, y))
3721 y = fold_rtx (y, insn);
3723 return simplify_gen_binary (code, mode, y, new_const);
3725 break;
3727 case DIV: case UDIV:
3728 /* ??? The associative optimization performed immediately above is
3729 also possible for DIV and UDIV using associate_code of MULT.
3730 However, we would need extra code to verify that the
3731 multiplication does not overflow, that is, there is no overflow
3732 in the calculation of new_const. */
3733 break;
3735 default:
3736 break;
3739 new_rtx = simplify_binary_operation (code, mode,
3740 const_arg0 ? const_arg0 : folded_arg0,
3741 const_arg1 ? const_arg1 : folded_arg1);
3742 break;
3744 case RTX_OBJ:
3745 /* (lo_sum (high X) X) is simply X. */
3746 if (code == LO_SUM && const_arg0 != 0
3747 && GET_CODE (const_arg0) == HIGH
3748 && rtx_equal_p (XEXP (const_arg0, 0), const_arg1))
3749 return const_arg1;
3750 break;
3752 case RTX_TERNARY:
3753 case RTX_BITFIELD_OPS:
3754 new_rtx = simplify_ternary_operation (code, mode, mode_arg0,
3755 const_arg0 ? const_arg0 : folded_arg0,
3756 const_arg1 ? const_arg1 : folded_arg1,
3757 const_arg2 ? const_arg2 : XEXP (x, 2));
3758 break;
3760 default:
3761 break;
3764 return new_rtx ? new_rtx : x;
3767 /* Return a constant value currently equivalent to X.
3768 Return 0 if we don't know one. */
3770 static rtx
3771 equiv_constant (rtx x)
3773 if (REG_P (x)
3774 && REGNO_QTY_VALID_P (REGNO (x)))
3776 int x_q = REG_QTY (REGNO (x));
3777 struct qty_table_elem *x_ent = &qty_table[x_q];
3779 if (x_ent->const_rtx)
3780 x = gen_lowpart (GET_MODE (x), x_ent->const_rtx);
3783 if (x == 0 || CONSTANT_P (x))
3784 return x;
3786 if (GET_CODE (x) == SUBREG)
3788 enum machine_mode mode = GET_MODE (x);
3789 enum machine_mode imode = GET_MODE (SUBREG_REG (x));
3790 rtx new_rtx;
3792 /* See if we previously assigned a constant value to this SUBREG. */
3793 if ((new_rtx = lookup_as_function (x, CONST_INT)) != 0
3794 || (new_rtx = lookup_as_function (x, CONST_WIDE_INT)) != 0
3795 || (new_rtx = lookup_as_function (x, CONST_DOUBLE)) != 0
3796 || (new_rtx = lookup_as_function (x, CONST_FIXED)) != 0)
3797 return new_rtx;
3799 /* If we didn't and if doing so makes sense, see if we previously
3800 assigned a constant value to the enclosing word mode SUBREG. */
3801 if (GET_MODE_SIZE (mode) < GET_MODE_SIZE (word_mode)
3802 && GET_MODE_SIZE (word_mode) < GET_MODE_SIZE (imode))
3804 int byte = SUBREG_BYTE (x) - subreg_lowpart_offset (mode, word_mode);
3805 if (byte >= 0 && (byte % UNITS_PER_WORD) == 0)
3807 rtx y = gen_rtx_SUBREG (word_mode, SUBREG_REG (x), byte);
3808 new_rtx = lookup_as_function (y, CONST_INT);
3809 if (new_rtx)
3810 return gen_lowpart (mode, new_rtx);
3814 /* Otherwise see if we already have a constant for the inner REG,
3815 and if that is enough to calculate an equivalent constant for
3816 the subreg. Note that the upper bits of paradoxical subregs
3817 are undefined, so they cannot be said to equal anything. */
3818 if (REG_P (SUBREG_REG (x))
3819 && GET_MODE_SIZE (mode) <= GET_MODE_SIZE (imode)
3820 && (new_rtx = equiv_constant (SUBREG_REG (x))) != 0)
3821 return simplify_subreg (mode, new_rtx, imode, SUBREG_BYTE (x));
3823 return 0;
3826 /* If X is a MEM, see if it is a constant-pool reference, or look it up in
3827 the hash table in case its value was seen before. */
3829 if (MEM_P (x))
3831 struct table_elt *elt;
3833 x = avoid_constant_pool_reference (x);
3834 if (CONSTANT_P (x))
3835 return x;
3837 elt = lookup (x, SAFE_HASH (x, GET_MODE (x)), GET_MODE (x));
3838 if (elt == 0)
3839 return 0;
3841 for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
3842 if (elt->is_const && CONSTANT_P (elt->exp))
3843 return elt->exp;
3846 return 0;
3849 /* Given INSN, a jump insn, TAKEN indicates if we are following the
3850 "taken" branch.
3852 In certain cases, this can cause us to add an equivalence. For example,
3853 if we are following the taken case of
3854 if (i == 2)
3855 we can add the fact that `i' and '2' are now equivalent.
3857 In any case, we can record that this comparison was passed. If the same
3858 comparison is seen later, we will know its value. */
3860 static void
3861 record_jump_equiv (rtx insn, bool taken)
3863 int cond_known_true;
3864 rtx op0, op1;
3865 rtx set;
3866 enum machine_mode mode, mode0, mode1;
3867 int reversed_nonequality = 0;
3868 enum rtx_code code;
3870 /* Ensure this is the right kind of insn. */
3871 gcc_assert (any_condjump_p (insn));
3873 set = pc_set (insn);
3875 /* See if this jump condition is known true or false. */
3876 if (taken)
3877 cond_known_true = (XEXP (SET_SRC (set), 2) == pc_rtx);
3878 else
3879 cond_known_true = (XEXP (SET_SRC (set), 1) == pc_rtx);
3881 /* Get the type of comparison being done and the operands being compared.
3882 If we had to reverse a non-equality condition, record that fact so we
3883 know that it isn't valid for floating-point. */
3884 code = GET_CODE (XEXP (SET_SRC (set), 0));
3885 op0 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 0), insn);
3886 op1 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 1), insn);
3888 code = find_comparison_args (code, &op0, &op1, &mode0, &mode1);
3889 if (! cond_known_true)
3891 code = reversed_comparison_code_parts (code, op0, op1, insn);
3893 /* Don't remember if we can't find the inverse. */
3894 if (code == UNKNOWN)
3895 return;
3898 /* The mode is the mode of the non-constant. */
3899 mode = mode0;
3900 if (mode1 != VOIDmode)
3901 mode = mode1;
3903 record_jump_cond (code, mode, op0, op1, reversed_nonequality);
3906 /* Yet another form of subreg creation. In this case, we want something in
3907 MODE, and we should assume OP has MODE iff it is naturally modeless. */
3909 static rtx
3910 record_jump_cond_subreg (enum machine_mode mode, rtx op)
3912 enum machine_mode op_mode = GET_MODE (op);
3913 if (op_mode == mode || op_mode == VOIDmode)
3914 return op;
3915 return lowpart_subreg (mode, op, op_mode);
3918 /* We know that comparison CODE applied to OP0 and OP1 in MODE is true.
3919 REVERSED_NONEQUALITY is nonzero if CODE had to be swapped.
3920 Make any useful entries we can with that information. Called from
3921 above function and called recursively. */
3923 static void
3924 record_jump_cond (enum rtx_code code, enum machine_mode mode, rtx op0,
3925 rtx op1, int reversed_nonequality)
3927 unsigned op0_hash, op1_hash;
3928 int op0_in_memory, op1_in_memory;
3929 struct table_elt *op0_elt, *op1_elt;
3931 /* If OP0 and OP1 are known equal, and either is a paradoxical SUBREG,
3932 we know that they are also equal in the smaller mode (this is also
3933 true for all smaller modes whether or not there is a SUBREG, but
3934 is not worth testing for with no SUBREG). */
3936 /* Note that GET_MODE (op0) may not equal MODE. */
3937 if (code == EQ && paradoxical_subreg_p (op0))
3939 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
3940 rtx tem = record_jump_cond_subreg (inner_mode, op1);
3941 if (tem)
3942 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
3943 reversed_nonequality);
3946 if (code == EQ && paradoxical_subreg_p (op1))
3948 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
3949 rtx tem = record_jump_cond_subreg (inner_mode, op0);
3950 if (tem)
3951 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
3952 reversed_nonequality);
3955 /* Similarly, if this is an NE comparison, and either is a SUBREG
3956 making a smaller mode, we know the whole thing is also NE. */
3958 /* Note that GET_MODE (op0) may not equal MODE;
3959 if we test MODE instead, we can get an infinite recursion
3960 alternating between two modes each wider than MODE. */
3962 if (code == NE && GET_CODE (op0) == SUBREG
3963 && subreg_lowpart_p (op0)
3964 && (GET_MODE_SIZE (GET_MODE (op0))
3965 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
3967 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
3968 rtx tem = record_jump_cond_subreg (inner_mode, op1);
3969 if (tem)
3970 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
3971 reversed_nonequality);
3974 if (code == NE && GET_CODE (op1) == SUBREG
3975 && subreg_lowpart_p (op1)
3976 && (GET_MODE_SIZE (GET_MODE (op1))
3977 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
3979 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
3980 rtx tem = record_jump_cond_subreg (inner_mode, op0);
3981 if (tem)
3982 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
3983 reversed_nonequality);
3986 /* Hash both operands. */
3988 do_not_record = 0;
3989 hash_arg_in_memory = 0;
3990 op0_hash = HASH (op0, mode);
3991 op0_in_memory = hash_arg_in_memory;
3993 if (do_not_record)
3994 return;
3996 do_not_record = 0;
3997 hash_arg_in_memory = 0;
3998 op1_hash = HASH (op1, mode);
3999 op1_in_memory = hash_arg_in_memory;
4001 if (do_not_record)
4002 return;
4004 /* Look up both operands. */
4005 op0_elt = lookup (op0, op0_hash, mode);
4006 op1_elt = lookup (op1, op1_hash, mode);
4008 /* If both operands are already equivalent or if they are not in the
4009 table but are identical, do nothing. */
4010 if ((op0_elt != 0 && op1_elt != 0
4011 && op0_elt->first_same_value == op1_elt->first_same_value)
4012 || op0 == op1 || rtx_equal_p (op0, op1))
4013 return;
4015 /* If we aren't setting two things equal all we can do is save this
4016 comparison. Similarly if this is floating-point. In the latter
4017 case, OP1 might be zero and both -0.0 and 0.0 are equal to it.
4018 If we record the equality, we might inadvertently delete code
4019 whose intent was to change -0 to +0. */
4021 if (code != EQ || FLOAT_MODE_P (GET_MODE (op0)))
4023 struct qty_table_elem *ent;
4024 int qty;
4026 /* If we reversed a floating-point comparison, if OP0 is not a
4027 register, or if OP1 is neither a register or constant, we can't
4028 do anything. */
4030 if (!REG_P (op1))
4031 op1 = equiv_constant (op1);
4033 if ((reversed_nonequality && FLOAT_MODE_P (mode))
4034 || !REG_P (op0) || op1 == 0)
4035 return;
4037 /* Put OP0 in the hash table if it isn't already. This gives it a
4038 new quantity number. */
4039 if (op0_elt == 0)
4041 if (insert_regs (op0, NULL, 0))
4043 rehash_using_reg (op0);
4044 op0_hash = HASH (op0, mode);
4046 /* If OP0 is contained in OP1, this changes its hash code
4047 as well. Faster to rehash than to check, except
4048 for the simple case of a constant. */
4049 if (! CONSTANT_P (op1))
4050 op1_hash = HASH (op1,mode);
4053 op0_elt = insert (op0, NULL, op0_hash, mode);
4054 op0_elt->in_memory = op0_in_memory;
4057 qty = REG_QTY (REGNO (op0));
4058 ent = &qty_table[qty];
4060 ent->comparison_code = code;
4061 if (REG_P (op1))
4063 /* Look it up again--in case op0 and op1 are the same. */
4064 op1_elt = lookup (op1, op1_hash, mode);
4066 /* Put OP1 in the hash table so it gets a new quantity number. */
4067 if (op1_elt == 0)
4069 if (insert_regs (op1, NULL, 0))
4071 rehash_using_reg (op1);
4072 op1_hash = HASH (op1, mode);
4075 op1_elt = insert (op1, NULL, op1_hash, mode);
4076 op1_elt->in_memory = op1_in_memory;
4079 ent->comparison_const = NULL_RTX;
4080 ent->comparison_qty = REG_QTY (REGNO (op1));
4082 else
4084 ent->comparison_const = op1;
4085 ent->comparison_qty = -1;
4088 return;
4091 /* If either side is still missing an equivalence, make it now,
4092 then merge the equivalences. */
4094 if (op0_elt == 0)
4096 if (insert_regs (op0, NULL, 0))
4098 rehash_using_reg (op0);
4099 op0_hash = HASH (op0, mode);
4102 op0_elt = insert (op0, NULL, op0_hash, mode);
4103 op0_elt->in_memory = op0_in_memory;
4106 if (op1_elt == 0)
4108 if (insert_regs (op1, NULL, 0))
4110 rehash_using_reg (op1);
4111 op1_hash = HASH (op1, mode);
4114 op1_elt = insert (op1, NULL, op1_hash, mode);
4115 op1_elt->in_memory = op1_in_memory;
4118 merge_equiv_classes (op0_elt, op1_elt);
4121 /* CSE processing for one instruction.
4123 Most "true" common subexpressions are mostly optimized away in GIMPLE,
4124 but the few that "leak through" are cleaned up by cse_insn, and complex
4125 addressing modes are often formed here.
4127 The main function is cse_insn, and between here and that function
4128 a couple of helper functions is defined to keep the size of cse_insn
4129 within reasonable proportions.
4131 Data is shared between the main and helper functions via STRUCT SET,
4132 that contains all data related for every set in the instruction that
4133 is being processed.
4135 Note that cse_main processes all sets in the instruction. Most
4136 passes in GCC only process simple SET insns or single_set insns, but
4137 CSE processes insns with multiple sets as well. */
4139 /* Data on one SET contained in the instruction. */
4141 struct set
4143 /* The SET rtx itself. */
4144 rtx rtl;
4145 /* The SET_SRC of the rtx (the original value, if it is changing). */
4146 rtx src;
4147 /* The hash-table element for the SET_SRC of the SET. */
4148 struct table_elt *src_elt;
4149 /* Hash value for the SET_SRC. */
4150 unsigned src_hash;
4151 /* Hash value for the SET_DEST. */
4152 unsigned dest_hash;
4153 /* The SET_DEST, with SUBREG, etc., stripped. */
4154 rtx inner_dest;
4155 /* Nonzero if the SET_SRC is in memory. */
4156 char src_in_memory;
4157 /* Nonzero if the SET_SRC contains something
4158 whose value cannot be predicted and understood. */
4159 char src_volatile;
4160 /* Original machine mode, in case it becomes a CONST_INT.
4161 The size of this field should match the size of the mode
4162 field of struct rtx_def (see rtl.h). */
4163 ENUM_BITFIELD(machine_mode) mode : 8;
4164 /* A constant equivalent for SET_SRC, if any. */
4165 rtx src_const;
4166 /* Hash value of constant equivalent for SET_SRC. */
4167 unsigned src_const_hash;
4168 /* Table entry for constant equivalent for SET_SRC, if any. */
4169 struct table_elt *src_const_elt;
4170 /* Table entry for the destination address. */
4171 struct table_elt *dest_addr_elt;
4174 /* Special handling for (set REG0 REG1) where REG0 is the
4175 "cheapest", cheaper than REG1. After cse, REG1 will probably not
4176 be used in the sequel, so (if easily done) change this insn to
4177 (set REG1 REG0) and replace REG1 with REG0 in the previous insn
4178 that computed their value. Then REG1 will become a dead store
4179 and won't cloud the situation for later optimizations.
4181 Do not make this change if REG1 is a hard register, because it will
4182 then be used in the sequel and we may be changing a two-operand insn
4183 into a three-operand insn.
4185 This is the last transformation that cse_insn will try to do. */
4187 static void
4188 try_back_substitute_reg (rtx set, rtx insn)
4190 rtx dest = SET_DEST (set);
4191 rtx src = SET_SRC (set);
4193 if (REG_P (dest)
4194 && REG_P (src) && ! HARD_REGISTER_P (src)
4195 && REGNO_QTY_VALID_P (REGNO (src)))
4197 int src_q = REG_QTY (REGNO (src));
4198 struct qty_table_elem *src_ent = &qty_table[src_q];
4200 if (src_ent->first_reg == REGNO (dest))
4202 /* Scan for the previous nonnote insn, but stop at a basic
4203 block boundary. */
4204 rtx prev = insn;
4205 rtx bb_head = BB_HEAD (BLOCK_FOR_INSN (insn));
4208 prev = PREV_INSN (prev);
4210 while (prev != bb_head && (NOTE_P (prev) || DEBUG_INSN_P (prev)));
4212 /* Do not swap the registers around if the previous instruction
4213 attaches a REG_EQUIV note to REG1.
4215 ??? It's not entirely clear whether we can transfer a REG_EQUIV
4216 from the pseudo that originally shadowed an incoming argument
4217 to another register. Some uses of REG_EQUIV might rely on it
4218 being attached to REG1 rather than REG2.
4220 This section previously turned the REG_EQUIV into a REG_EQUAL
4221 note. We cannot do that because REG_EQUIV may provide an
4222 uninitialized stack slot when REG_PARM_STACK_SPACE is used. */
4223 if (NONJUMP_INSN_P (prev)
4224 && GET_CODE (PATTERN (prev)) == SET
4225 && SET_DEST (PATTERN (prev)) == src
4226 && ! find_reg_note (prev, REG_EQUIV, NULL_RTX))
4228 rtx note;
4230 validate_change (prev, &SET_DEST (PATTERN (prev)), dest, 1);
4231 validate_change (insn, &SET_DEST (set), src, 1);
4232 validate_change (insn, &SET_SRC (set), dest, 1);
4233 apply_change_group ();
4235 /* If INSN has a REG_EQUAL note, and this note mentions
4236 REG0, then we must delete it, because the value in
4237 REG0 has changed. If the note's value is REG1, we must
4238 also delete it because that is now this insn's dest. */
4239 note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
4240 if (note != 0
4241 && (reg_mentioned_p (dest, XEXP (note, 0))
4242 || rtx_equal_p (src, XEXP (note, 0))))
4243 remove_note (insn, note);
4249 /* Record all the SETs in this instruction into SETS_PTR,
4250 and return the number of recorded sets. */
4251 static int
4252 find_sets_in_insn (rtx insn, struct set **psets)
4254 struct set *sets = *psets;
4255 int n_sets = 0;
4256 rtx x = PATTERN (insn);
4258 if (GET_CODE (x) == SET)
4260 /* Ignore SETs that are unconditional jumps.
4261 They never need cse processing, so this does not hurt.
4262 The reason is not efficiency but rather
4263 so that we can test at the end for instructions
4264 that have been simplified to unconditional jumps
4265 and not be misled by unchanged instructions
4266 that were unconditional jumps to begin with. */
4267 if (SET_DEST (x) == pc_rtx
4268 && GET_CODE (SET_SRC (x)) == LABEL_REF)
4270 /* Don't count call-insns, (set (reg 0) (call ...)), as a set.
4271 The hard function value register is used only once, to copy to
4272 someplace else, so it isn't worth cse'ing. */
4273 else if (GET_CODE (SET_SRC (x)) == CALL)
4275 else
4276 sets[n_sets++].rtl = x;
4278 else if (GET_CODE (x) == PARALLEL)
4280 int i, lim = XVECLEN (x, 0);
4282 /* Go over the epressions of the PARALLEL in forward order, to
4283 put them in the same order in the SETS array. */
4284 for (i = 0; i < lim; i++)
4286 rtx y = XVECEXP (x, 0, i);
4287 if (GET_CODE (y) == SET)
4289 /* As above, we ignore unconditional jumps and call-insns and
4290 ignore the result of apply_change_group. */
4291 if (SET_DEST (y) == pc_rtx
4292 && GET_CODE (SET_SRC (y)) == LABEL_REF)
4294 else if (GET_CODE (SET_SRC (y)) == CALL)
4296 else
4297 sets[n_sets++].rtl = y;
4302 return n_sets;
4305 /* Where possible, substitute every register reference in the N_SETS
4306 number of SETS in INSN with the the canonical register.
4308 Register canonicalization propagatest the earliest register (i.e.
4309 one that is set before INSN) with the same value. This is a very
4310 useful, simple form of CSE, to clean up warts from expanding GIMPLE
4311 to RTL. For instance, a CONST for an address is usually expanded
4312 multiple times to loads into different registers, thus creating many
4313 subexpressions of the form:
4315 (set (reg1) (some_const))
4316 (set (mem (... reg1 ...) (thing)))
4317 (set (reg2) (some_const))
4318 (set (mem (... reg2 ...) (thing)))
4320 After canonicalizing, the code takes the following form:
4322 (set (reg1) (some_const))
4323 (set (mem (... reg1 ...) (thing)))
4324 (set (reg2) (some_const))
4325 (set (mem (... reg1 ...) (thing)))
4327 The set to reg2 is now trivially dead, and the memory reference (or
4328 address, or whatever) may be a candidate for further CSEing.
4330 In this function, the result of apply_change_group can be ignored;
4331 see canon_reg. */
4333 static void
4334 canonicalize_insn (rtx insn, struct set **psets, int n_sets)
4336 struct set *sets = *psets;
4337 rtx tem;
4338 rtx x = PATTERN (insn);
4339 int i;
4341 if (CALL_P (insn))
4343 for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
4344 if (GET_CODE (XEXP (tem, 0)) != SET)
4345 XEXP (tem, 0) = canon_reg (XEXP (tem, 0), insn);
4348 if (GET_CODE (x) == SET && GET_CODE (SET_SRC (x)) == CALL)
4350 canon_reg (SET_SRC (x), insn);
4351 apply_change_group ();
4352 fold_rtx (SET_SRC (x), insn);
4354 else if (GET_CODE (x) == CLOBBER)
4356 /* If we clobber memory, canon the address.
4357 This does nothing when a register is clobbered
4358 because we have already invalidated the reg. */
4359 if (MEM_P (XEXP (x, 0)))
4360 canon_reg (XEXP (x, 0), insn);
4362 else if (GET_CODE (x) == USE
4363 && ! (REG_P (XEXP (x, 0))
4364 && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER))
4365 /* Canonicalize a USE of a pseudo register or memory location. */
4366 canon_reg (x, insn);
4367 else if (GET_CODE (x) == ASM_OPERANDS)
4369 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
4371 rtx input = ASM_OPERANDS_INPUT (x, i);
4372 if (!(REG_P (input) && REGNO (input) < FIRST_PSEUDO_REGISTER))
4374 input = canon_reg (input, insn);
4375 validate_change (insn, &ASM_OPERANDS_INPUT (x, i), input, 1);
4379 else if (GET_CODE (x) == CALL)
4381 canon_reg (x, insn);
4382 apply_change_group ();
4383 fold_rtx (x, insn);
4385 else if (DEBUG_INSN_P (insn))
4386 canon_reg (PATTERN (insn), insn);
4387 else if (GET_CODE (x) == PARALLEL)
4389 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
4391 rtx y = XVECEXP (x, 0, i);
4392 if (GET_CODE (y) == SET && GET_CODE (SET_SRC (y)) == CALL)
4394 canon_reg (SET_SRC (y), insn);
4395 apply_change_group ();
4396 fold_rtx (SET_SRC (y), insn);
4398 else if (GET_CODE (y) == CLOBBER)
4400 if (MEM_P (XEXP (y, 0)))
4401 canon_reg (XEXP (y, 0), insn);
4403 else if (GET_CODE (y) == USE
4404 && ! (REG_P (XEXP (y, 0))
4405 && REGNO (XEXP (y, 0)) < FIRST_PSEUDO_REGISTER))
4406 canon_reg (y, insn);
4407 else if (GET_CODE (y) == CALL)
4409 canon_reg (y, insn);
4410 apply_change_group ();
4411 fold_rtx (y, insn);
4416 if (n_sets == 1 && REG_NOTES (insn) != 0
4417 && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0)
4419 /* We potentially will process this insn many times. Therefore,
4420 drop the REG_EQUAL note if it is equal to the SET_SRC of the
4421 unique set in INSN.
4423 Do not do so if the REG_EQUAL note is for a STRICT_LOW_PART,
4424 because cse_insn handles those specially. */
4425 if (GET_CODE (SET_DEST (sets[0].rtl)) != STRICT_LOW_PART
4426 && rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl)))
4427 remove_note (insn, tem);
4428 else
4430 canon_reg (XEXP (tem, 0), insn);
4431 apply_change_group ();
4432 XEXP (tem, 0) = fold_rtx (XEXP (tem, 0), insn);
4433 df_notes_rescan (insn);
4437 /* Canonicalize sources and addresses of destinations.
4438 We do this in a separate pass to avoid problems when a MATCH_DUP is
4439 present in the insn pattern. In that case, we want to ensure that
4440 we don't break the duplicate nature of the pattern. So we will replace
4441 both operands at the same time. Otherwise, we would fail to find an
4442 equivalent substitution in the loop calling validate_change below.
4444 We used to suppress canonicalization of DEST if it appears in SRC,
4445 but we don't do this any more. */
4447 for (i = 0; i < n_sets; i++)
4449 rtx dest = SET_DEST (sets[i].rtl);
4450 rtx src = SET_SRC (sets[i].rtl);
4451 rtx new_rtx = canon_reg (src, insn);
4453 validate_change (insn, &SET_SRC (sets[i].rtl), new_rtx, 1);
4455 if (GET_CODE (dest) == ZERO_EXTRACT)
4457 validate_change (insn, &XEXP (dest, 1),
4458 canon_reg (XEXP (dest, 1), insn), 1);
4459 validate_change (insn, &XEXP (dest, 2),
4460 canon_reg (XEXP (dest, 2), insn), 1);
4463 while (GET_CODE (dest) == SUBREG
4464 || GET_CODE (dest) == ZERO_EXTRACT
4465 || GET_CODE (dest) == STRICT_LOW_PART)
4466 dest = XEXP (dest, 0);
4468 if (MEM_P (dest))
4469 canon_reg (dest, insn);
4472 /* Now that we have done all the replacements, we can apply the change
4473 group and see if they all work. Note that this will cause some
4474 canonicalizations that would have worked individually not to be applied
4475 because some other canonicalization didn't work, but this should not
4476 occur often.
4478 The result of apply_change_group can be ignored; see canon_reg. */
4480 apply_change_group ();
4483 /* Main function of CSE.
4484 First simplify sources and addresses of all assignments
4485 in the instruction, using previously-computed equivalents values.
4486 Then install the new sources and destinations in the table
4487 of available values. */
4489 static void
4490 cse_insn (rtx insn)
4492 rtx x = PATTERN (insn);
4493 int i;
4494 rtx tem;
4495 int n_sets = 0;
4497 rtx src_eqv = 0;
4498 struct table_elt *src_eqv_elt = 0;
4499 int src_eqv_volatile = 0;
4500 int src_eqv_in_memory = 0;
4501 unsigned src_eqv_hash = 0;
4503 struct set *sets = (struct set *) 0;
4505 if (GET_CODE (x) == SET)
4506 sets = XALLOCA (struct set);
4507 else if (GET_CODE (x) == PARALLEL)
4508 sets = XALLOCAVEC (struct set, XVECLEN (x, 0));
4510 this_insn = insn;
4511 #ifdef HAVE_cc0
4512 /* Records what this insn does to set CC0. */
4513 this_insn_cc0 = 0;
4514 this_insn_cc0_mode = VOIDmode;
4515 #endif
4517 /* Find all regs explicitly clobbered in this insn,
4518 to ensure they are not replaced with any other regs
4519 elsewhere in this insn. */
4520 invalidate_from_sets_and_clobbers (insn);
4522 /* Record all the SETs in this instruction. */
4523 n_sets = find_sets_in_insn (insn, &sets);
4525 /* Substitute the canonical register where possible. */
4526 canonicalize_insn (insn, &sets, n_sets);
4528 /* If this insn has a REG_EQUAL note, store the equivalent value in SRC_EQV,
4529 if different, or if the DEST is a STRICT_LOW_PART. The latter condition
4530 is necessary because SRC_EQV is handled specially for this case, and if
4531 it isn't set, then there will be no equivalence for the destination. */
4532 if (n_sets == 1 && REG_NOTES (insn) != 0
4533 && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0
4534 && (! rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl))
4535 || GET_CODE (SET_DEST (sets[0].rtl)) == STRICT_LOW_PART))
4536 src_eqv = copy_rtx (XEXP (tem, 0));
4538 /* Set sets[i].src_elt to the class each source belongs to.
4539 Detect assignments from or to volatile things
4540 and set set[i] to zero so they will be ignored
4541 in the rest of this function.
4543 Nothing in this loop changes the hash table or the register chains. */
4545 for (i = 0; i < n_sets; i++)
4547 bool repeat = false;
4548 rtx src, dest;
4549 rtx src_folded;
4550 struct table_elt *elt = 0, *p;
4551 enum machine_mode mode;
4552 rtx src_eqv_here;
4553 rtx src_const = 0;
4554 rtx src_related = 0;
4555 bool src_related_is_const_anchor = false;
4556 struct table_elt *src_const_elt = 0;
4557 int src_cost = MAX_COST;
4558 int src_eqv_cost = MAX_COST;
4559 int src_folded_cost = MAX_COST;
4560 int src_related_cost = MAX_COST;
4561 int src_elt_cost = MAX_COST;
4562 int src_regcost = MAX_COST;
4563 int src_eqv_regcost = MAX_COST;
4564 int src_folded_regcost = MAX_COST;
4565 int src_related_regcost = MAX_COST;
4566 int src_elt_regcost = MAX_COST;
4567 /* Set nonzero if we need to call force_const_mem on with the
4568 contents of src_folded before using it. */
4569 int src_folded_force_flag = 0;
4571 dest = SET_DEST (sets[i].rtl);
4572 src = SET_SRC (sets[i].rtl);
4574 /* If SRC is a constant that has no machine mode,
4575 hash it with the destination's machine mode.
4576 This way we can keep different modes separate. */
4578 mode = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
4579 sets[i].mode = mode;
4581 if (src_eqv)
4583 enum machine_mode eqvmode = mode;
4584 if (GET_CODE (dest) == STRICT_LOW_PART)
4585 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
4586 do_not_record = 0;
4587 hash_arg_in_memory = 0;
4588 src_eqv_hash = HASH (src_eqv, eqvmode);
4590 /* Find the equivalence class for the equivalent expression. */
4592 if (!do_not_record)
4593 src_eqv_elt = lookup (src_eqv, src_eqv_hash, eqvmode);
4595 src_eqv_volatile = do_not_record;
4596 src_eqv_in_memory = hash_arg_in_memory;
4599 /* If this is a STRICT_LOW_PART assignment, src_eqv corresponds to the
4600 value of the INNER register, not the destination. So it is not
4601 a valid substitution for the source. But save it for later. */
4602 if (GET_CODE (dest) == STRICT_LOW_PART)
4603 src_eqv_here = 0;
4604 else
4605 src_eqv_here = src_eqv;
4607 /* Simplify and foldable subexpressions in SRC. Then get the fully-
4608 simplified result, which may not necessarily be valid. */
4609 src_folded = fold_rtx (src, insn);
4611 #if 0
4612 /* ??? This caused bad code to be generated for the m68k port with -O2.
4613 Suppose src is (CONST_INT -1), and that after truncation src_folded
4614 is (CONST_INT 3). Suppose src_folded is then used for src_const.
4615 At the end we will add src and src_const to the same equivalence
4616 class. We now have 3 and -1 on the same equivalence class. This
4617 causes later instructions to be mis-optimized. */
4618 /* If storing a constant in a bitfield, pre-truncate the constant
4619 so we will be able to record it later. */
4620 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
4622 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
4624 if (CONST_INT_P (src)
4625 && CONST_INT_P (width)
4626 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
4627 && (INTVAL (src) & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
4628 src_folded
4629 = GEN_INT (INTVAL (src) & (((HOST_WIDE_INT) 1
4630 << INTVAL (width)) - 1));
4632 #endif
4634 /* Compute SRC's hash code, and also notice if it
4635 should not be recorded at all. In that case,
4636 prevent any further processing of this assignment. */
4637 do_not_record = 0;
4638 hash_arg_in_memory = 0;
4640 sets[i].src = src;
4641 sets[i].src_hash = HASH (src, mode);
4642 sets[i].src_volatile = do_not_record;
4643 sets[i].src_in_memory = hash_arg_in_memory;
4645 /* If SRC is a MEM, there is a REG_EQUIV note for SRC, and DEST is
4646 a pseudo, do not record SRC. Using SRC as a replacement for
4647 anything else will be incorrect in that situation. Note that
4648 this usually occurs only for stack slots, in which case all the
4649 RTL would be referring to SRC, so we don't lose any optimization
4650 opportunities by not having SRC in the hash table. */
4652 if (MEM_P (src)
4653 && find_reg_note (insn, REG_EQUIV, NULL_RTX) != 0
4654 && REG_P (dest)
4655 && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
4656 sets[i].src_volatile = 1;
4658 /* Also do not record result of a non-volatile inline asm with
4659 more than one result or with clobbers, we do not want CSE to
4660 break the inline asm apart. */
4661 else if (GET_CODE (src) == ASM_OPERANDS
4662 && GET_CODE (x) == PARALLEL)
4663 sets[i].src_volatile = 1;
4665 #if 0
4666 /* It is no longer clear why we used to do this, but it doesn't
4667 appear to still be needed. So let's try without it since this
4668 code hurts cse'ing widened ops. */
4669 /* If source is a paradoxical subreg (such as QI treated as an SI),
4670 treat it as volatile. It may do the work of an SI in one context
4671 where the extra bits are not being used, but cannot replace an SI
4672 in general. */
4673 if (paradoxical_subreg_p (src))
4674 sets[i].src_volatile = 1;
4675 #endif
4677 /* Locate all possible equivalent forms for SRC. Try to replace
4678 SRC in the insn with each cheaper equivalent.
4680 We have the following types of equivalents: SRC itself, a folded
4681 version, a value given in a REG_EQUAL note, or a value related
4682 to a constant.
4684 Each of these equivalents may be part of an additional class
4685 of equivalents (if more than one is in the table, they must be in
4686 the same class; we check for this).
4688 If the source is volatile, we don't do any table lookups.
4690 We note any constant equivalent for possible later use in a
4691 REG_NOTE. */
4693 if (!sets[i].src_volatile)
4694 elt = lookup (src, sets[i].src_hash, mode);
4696 sets[i].src_elt = elt;
4698 if (elt && src_eqv_here && src_eqv_elt)
4700 if (elt->first_same_value != src_eqv_elt->first_same_value)
4702 /* The REG_EQUAL is indicating that two formerly distinct
4703 classes are now equivalent. So merge them. */
4704 merge_equiv_classes (elt, src_eqv_elt);
4705 src_eqv_hash = HASH (src_eqv, elt->mode);
4706 src_eqv_elt = lookup (src_eqv, src_eqv_hash, elt->mode);
4709 src_eqv_here = 0;
4712 else if (src_eqv_elt)
4713 elt = src_eqv_elt;
4715 /* Try to find a constant somewhere and record it in `src_const'.
4716 Record its table element, if any, in `src_const_elt'. Look in
4717 any known equivalences first. (If the constant is not in the
4718 table, also set `sets[i].src_const_hash'). */
4719 if (elt)
4720 for (p = elt->first_same_value; p; p = p->next_same_value)
4721 if (p->is_const)
4723 src_const = p->exp;
4724 src_const_elt = elt;
4725 break;
4728 if (src_const == 0
4729 && (CONSTANT_P (src_folded)
4730 /* Consider (minus (label_ref L1) (label_ref L2)) as
4731 "constant" here so we will record it. This allows us
4732 to fold switch statements when an ADDR_DIFF_VEC is used. */
4733 || (GET_CODE (src_folded) == MINUS
4734 && GET_CODE (XEXP (src_folded, 0)) == LABEL_REF
4735 && GET_CODE (XEXP (src_folded, 1)) == LABEL_REF)))
4736 src_const = src_folded, src_const_elt = elt;
4737 else if (src_const == 0 && src_eqv_here && CONSTANT_P (src_eqv_here))
4738 src_const = src_eqv_here, src_const_elt = src_eqv_elt;
4740 /* If we don't know if the constant is in the table, get its
4741 hash code and look it up. */
4742 if (src_const && src_const_elt == 0)
4744 sets[i].src_const_hash = HASH (src_const, mode);
4745 src_const_elt = lookup (src_const, sets[i].src_const_hash, mode);
4748 sets[i].src_const = src_const;
4749 sets[i].src_const_elt = src_const_elt;
4751 /* If the constant and our source are both in the table, mark them as
4752 equivalent. Otherwise, if a constant is in the table but the source
4753 isn't, set ELT to it. */
4754 if (src_const_elt && elt
4755 && src_const_elt->first_same_value != elt->first_same_value)
4756 merge_equiv_classes (elt, src_const_elt);
4757 else if (src_const_elt && elt == 0)
4758 elt = src_const_elt;
4760 /* See if there is a register linearly related to a constant
4761 equivalent of SRC. */
4762 if (src_const
4763 && (GET_CODE (src_const) == CONST
4764 || (src_const_elt && src_const_elt->related_value != 0)))
4766 src_related = use_related_value (src_const, src_const_elt);
4767 if (src_related)
4769 struct table_elt *src_related_elt
4770 = lookup (src_related, HASH (src_related, mode), mode);
4771 if (src_related_elt && elt)
4773 if (elt->first_same_value
4774 != src_related_elt->first_same_value)
4775 /* This can occur when we previously saw a CONST
4776 involving a SYMBOL_REF and then see the SYMBOL_REF
4777 twice. Merge the involved classes. */
4778 merge_equiv_classes (elt, src_related_elt);
4780 src_related = 0;
4781 src_related_elt = 0;
4783 else if (src_related_elt && elt == 0)
4784 elt = src_related_elt;
4788 /* See if we have a CONST_INT that is already in a register in a
4789 wider mode. */
4791 if (src_const && src_related == 0 && CONST_INT_P (src_const)
4792 && GET_MODE_CLASS (mode) == MODE_INT
4793 && GET_MODE_PRECISION (mode) < BITS_PER_WORD)
4795 enum machine_mode wider_mode;
4797 for (wider_mode = GET_MODE_WIDER_MODE (mode);
4798 wider_mode != VOIDmode
4799 && GET_MODE_PRECISION (wider_mode) <= BITS_PER_WORD
4800 && src_related == 0;
4801 wider_mode = GET_MODE_WIDER_MODE (wider_mode))
4803 struct table_elt *const_elt
4804 = lookup (src_const, HASH (src_const, wider_mode), wider_mode);
4806 if (const_elt == 0)
4807 continue;
4809 for (const_elt = const_elt->first_same_value;
4810 const_elt; const_elt = const_elt->next_same_value)
4811 if (REG_P (const_elt->exp))
4813 src_related = gen_lowpart (mode, const_elt->exp);
4814 break;
4819 /* Another possibility is that we have an AND with a constant in
4820 a mode narrower than a word. If so, it might have been generated
4821 as part of an "if" which would narrow the AND. If we already
4822 have done the AND in a wider mode, we can use a SUBREG of that
4823 value. */
4825 if (flag_expensive_optimizations && ! src_related
4826 && GET_CODE (src) == AND && CONST_INT_P (XEXP (src, 1))
4827 && GET_MODE_SIZE (mode) < UNITS_PER_WORD)
4829 enum machine_mode tmode;
4830 rtx new_and = gen_rtx_AND (VOIDmode, NULL_RTX, XEXP (src, 1));
4832 for (tmode = GET_MODE_WIDER_MODE (mode);
4833 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
4834 tmode = GET_MODE_WIDER_MODE (tmode))
4836 rtx inner = gen_lowpart (tmode, XEXP (src, 0));
4837 struct table_elt *larger_elt;
4839 if (inner)
4841 PUT_MODE (new_and, tmode);
4842 XEXP (new_and, 0) = inner;
4843 larger_elt = lookup (new_and, HASH (new_and, tmode), tmode);
4844 if (larger_elt == 0)
4845 continue;
4847 for (larger_elt = larger_elt->first_same_value;
4848 larger_elt; larger_elt = larger_elt->next_same_value)
4849 if (REG_P (larger_elt->exp))
4851 src_related
4852 = gen_lowpart (mode, larger_elt->exp);
4853 break;
4856 if (src_related)
4857 break;
4862 #ifdef LOAD_EXTEND_OP
4863 /* See if a MEM has already been loaded with a widening operation;
4864 if it has, we can use a subreg of that. Many CISC machines
4865 also have such operations, but this is only likely to be
4866 beneficial on these machines. */
4868 if (flag_expensive_optimizations && src_related == 0
4869 && (GET_MODE_SIZE (mode) < UNITS_PER_WORD)
4870 && GET_MODE_CLASS (mode) == MODE_INT
4871 && MEM_P (src) && ! do_not_record
4872 && LOAD_EXTEND_OP (mode) != UNKNOWN)
4874 struct rtx_def memory_extend_buf;
4875 rtx memory_extend_rtx = &memory_extend_buf;
4876 enum machine_mode tmode;
4878 /* Set what we are trying to extend and the operation it might
4879 have been extended with. */
4880 memset (memory_extend_rtx, 0, sizeof (*memory_extend_rtx));
4881 PUT_CODE (memory_extend_rtx, LOAD_EXTEND_OP (mode));
4882 XEXP (memory_extend_rtx, 0) = src;
4884 for (tmode = GET_MODE_WIDER_MODE (mode);
4885 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
4886 tmode = GET_MODE_WIDER_MODE (tmode))
4888 struct table_elt *larger_elt;
4890 PUT_MODE (memory_extend_rtx, tmode);
4891 larger_elt = lookup (memory_extend_rtx,
4892 HASH (memory_extend_rtx, tmode), tmode);
4893 if (larger_elt == 0)
4894 continue;
4896 for (larger_elt = larger_elt->first_same_value;
4897 larger_elt; larger_elt = larger_elt->next_same_value)
4898 if (REG_P (larger_elt->exp))
4900 src_related = gen_lowpart (mode, larger_elt->exp);
4901 break;
4904 if (src_related)
4905 break;
4908 #endif /* LOAD_EXTEND_OP */
4910 /* Try to express the constant using a register+offset expression
4911 derived from a constant anchor. */
4913 if (targetm.const_anchor
4914 && !src_related
4915 && src_const
4916 && GET_CODE (src_const) == CONST_INT)
4918 src_related = try_const_anchors (src_const, mode);
4919 src_related_is_const_anchor = src_related != NULL_RTX;
4923 if (src == src_folded)
4924 src_folded = 0;
4926 /* At this point, ELT, if nonzero, points to a class of expressions
4927 equivalent to the source of this SET and SRC, SRC_EQV, SRC_FOLDED,
4928 and SRC_RELATED, if nonzero, each contain additional equivalent
4929 expressions. Prune these latter expressions by deleting expressions
4930 already in the equivalence class.
4932 Check for an equivalent identical to the destination. If found,
4933 this is the preferred equivalent since it will likely lead to
4934 elimination of the insn. Indicate this by placing it in
4935 `src_related'. */
4937 if (elt)
4938 elt = elt->first_same_value;
4939 for (p = elt; p; p = p->next_same_value)
4941 enum rtx_code code = GET_CODE (p->exp);
4943 /* If the expression is not valid, ignore it. Then we do not
4944 have to check for validity below. In most cases, we can use
4945 `rtx_equal_p', since canonicalization has already been done. */
4946 if (code != REG && ! exp_equiv_p (p->exp, p->exp, 1, false))
4947 continue;
4949 /* Also skip paradoxical subregs, unless that's what we're
4950 looking for. */
4951 if (paradoxical_subreg_p (p->exp)
4952 && ! (src != 0
4953 && GET_CODE (src) == SUBREG
4954 && GET_MODE (src) == GET_MODE (p->exp)
4955 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
4956 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))))
4957 continue;
4959 if (src && GET_CODE (src) == code && rtx_equal_p (src, p->exp))
4960 src = 0;
4961 else if (src_folded && GET_CODE (src_folded) == code
4962 && rtx_equal_p (src_folded, p->exp))
4963 src_folded = 0;
4964 else if (src_eqv_here && GET_CODE (src_eqv_here) == code
4965 && rtx_equal_p (src_eqv_here, p->exp))
4966 src_eqv_here = 0;
4967 else if (src_related && GET_CODE (src_related) == code
4968 && rtx_equal_p (src_related, p->exp))
4969 src_related = 0;
4971 /* This is the same as the destination of the insns, we want
4972 to prefer it. Copy it to src_related. The code below will
4973 then give it a negative cost. */
4974 if (GET_CODE (dest) == code && rtx_equal_p (p->exp, dest))
4975 src_related = dest;
4978 /* Find the cheapest valid equivalent, trying all the available
4979 possibilities. Prefer items not in the hash table to ones
4980 that are when they are equal cost. Note that we can never
4981 worsen an insn as the current contents will also succeed.
4982 If we find an equivalent identical to the destination, use it as best,
4983 since this insn will probably be eliminated in that case. */
4984 if (src)
4986 if (rtx_equal_p (src, dest))
4987 src_cost = src_regcost = -1;
4988 else
4990 src_cost = COST (src);
4991 src_regcost = approx_reg_cost (src);
4995 if (src_eqv_here)
4997 if (rtx_equal_p (src_eqv_here, dest))
4998 src_eqv_cost = src_eqv_regcost = -1;
4999 else
5001 src_eqv_cost = COST (src_eqv_here);
5002 src_eqv_regcost = approx_reg_cost (src_eqv_here);
5006 if (src_folded)
5008 if (rtx_equal_p (src_folded, dest))
5009 src_folded_cost = src_folded_regcost = -1;
5010 else
5012 src_folded_cost = COST (src_folded);
5013 src_folded_regcost = approx_reg_cost (src_folded);
5017 if (src_related)
5019 if (rtx_equal_p (src_related, dest))
5020 src_related_cost = src_related_regcost = -1;
5021 else
5023 src_related_cost = COST (src_related);
5024 src_related_regcost = approx_reg_cost (src_related);
5026 /* If a const-anchor is used to synthesize a constant that
5027 normally requires multiple instructions then slightly prefer
5028 it over the original sequence. These instructions are likely
5029 to become redundant now. We can't compare against the cost
5030 of src_eqv_here because, on MIPS for example, multi-insn
5031 constants have zero cost; they are assumed to be hoisted from
5032 loops. */
5033 if (src_related_is_const_anchor
5034 && src_related_cost == src_cost
5035 && src_eqv_here)
5036 src_related_cost--;
5040 /* If this was an indirect jump insn, a known label will really be
5041 cheaper even though it looks more expensive. */
5042 if (dest == pc_rtx && src_const && GET_CODE (src_const) == LABEL_REF)
5043 src_folded = src_const, src_folded_cost = src_folded_regcost = -1;
5045 /* Terminate loop when replacement made. This must terminate since
5046 the current contents will be tested and will always be valid. */
5047 while (1)
5049 rtx trial;
5051 /* Skip invalid entries. */
5052 while (elt && !REG_P (elt->exp)
5053 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
5054 elt = elt->next_same_value;
5056 /* A paradoxical subreg would be bad here: it'll be the right
5057 size, but later may be adjusted so that the upper bits aren't
5058 what we want. So reject it. */
5059 if (elt != 0
5060 && paradoxical_subreg_p (elt->exp)
5061 /* It is okay, though, if the rtx we're trying to match
5062 will ignore any of the bits we can't predict. */
5063 && ! (src != 0
5064 && GET_CODE (src) == SUBREG
5065 && GET_MODE (src) == GET_MODE (elt->exp)
5066 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
5067 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))))
5069 elt = elt->next_same_value;
5070 continue;
5073 if (elt)
5075 src_elt_cost = elt->cost;
5076 src_elt_regcost = elt->regcost;
5079 /* Find cheapest and skip it for the next time. For items
5080 of equal cost, use this order:
5081 src_folded, src, src_eqv, src_related and hash table entry. */
5082 if (src_folded
5083 && preferable (src_folded_cost, src_folded_regcost,
5084 src_cost, src_regcost) <= 0
5085 && preferable (src_folded_cost, src_folded_regcost,
5086 src_eqv_cost, src_eqv_regcost) <= 0
5087 && preferable (src_folded_cost, src_folded_regcost,
5088 src_related_cost, src_related_regcost) <= 0
5089 && preferable (src_folded_cost, src_folded_regcost,
5090 src_elt_cost, src_elt_regcost) <= 0)
5092 trial = src_folded, src_folded_cost = MAX_COST;
5093 if (src_folded_force_flag)
5095 rtx forced = force_const_mem (mode, trial);
5096 if (forced)
5097 trial = forced;
5100 else if (src
5101 && preferable (src_cost, src_regcost,
5102 src_eqv_cost, src_eqv_regcost) <= 0
5103 && preferable (src_cost, src_regcost,
5104 src_related_cost, src_related_regcost) <= 0
5105 && preferable (src_cost, src_regcost,
5106 src_elt_cost, src_elt_regcost) <= 0)
5107 trial = src, src_cost = MAX_COST;
5108 else if (src_eqv_here
5109 && preferable (src_eqv_cost, src_eqv_regcost,
5110 src_related_cost, src_related_regcost) <= 0
5111 && preferable (src_eqv_cost, src_eqv_regcost,
5112 src_elt_cost, src_elt_regcost) <= 0)
5113 trial = src_eqv_here, src_eqv_cost = MAX_COST;
5114 else if (src_related
5115 && preferable (src_related_cost, src_related_regcost,
5116 src_elt_cost, src_elt_regcost) <= 0)
5117 trial = src_related, src_related_cost = MAX_COST;
5118 else
5120 trial = elt->exp;
5121 elt = elt->next_same_value;
5122 src_elt_cost = MAX_COST;
5125 /* Avoid creation of overlapping memory moves. */
5126 if (MEM_P (trial) && MEM_P (SET_DEST (sets[i].rtl)))
5128 rtx src, dest;
5130 /* BLKmode moves are not handled by cse anyway. */
5131 if (GET_MODE (trial) == BLKmode)
5132 break;
5134 src = canon_rtx (trial);
5135 dest = canon_rtx (SET_DEST (sets[i].rtl));
5137 if (!MEM_P (src) || !MEM_P (dest)
5138 || !nonoverlapping_memrefs_p (src, dest, false))
5139 break;
5142 /* Try to optimize
5143 (set (reg:M N) (const_int A))
5144 (set (reg:M2 O) (const_int B))
5145 (set (zero_extract:M2 (reg:M N) (const_int C) (const_int D))
5146 (reg:M2 O)). */
5147 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT
5148 && CONST_INT_P (trial)
5149 && CONST_INT_P (XEXP (SET_DEST (sets[i].rtl), 1))
5150 && CONST_INT_P (XEXP (SET_DEST (sets[i].rtl), 2))
5151 && REG_P (XEXP (SET_DEST (sets[i].rtl), 0))
5152 && (GET_MODE_PRECISION (GET_MODE (SET_DEST (sets[i].rtl)))
5153 >= INTVAL (XEXP (SET_DEST (sets[i].rtl), 1)))
5154 && ((unsigned) INTVAL (XEXP (SET_DEST (sets[i].rtl), 1))
5155 + (unsigned) INTVAL (XEXP (SET_DEST (sets[i].rtl), 2))
5156 <= HOST_BITS_PER_WIDE_INT))
5158 rtx dest_reg = XEXP (SET_DEST (sets[i].rtl), 0);
5159 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5160 rtx pos = XEXP (SET_DEST (sets[i].rtl), 2);
5161 unsigned int dest_hash = HASH (dest_reg, GET_MODE (dest_reg));
5162 struct table_elt *dest_elt
5163 = lookup (dest_reg, dest_hash, GET_MODE (dest_reg));
5164 rtx dest_cst = NULL;
5166 if (dest_elt)
5167 for (p = dest_elt->first_same_value; p; p = p->next_same_value)
5168 if (p->is_const && CONST_INT_P (p->exp))
5170 dest_cst = p->exp;
5171 break;
5173 if (dest_cst)
5175 HOST_WIDE_INT val = INTVAL (dest_cst);
5176 HOST_WIDE_INT mask;
5177 unsigned int shift;
5178 if (BITS_BIG_ENDIAN)
5179 shift = GET_MODE_PRECISION (GET_MODE (dest_reg))
5180 - INTVAL (pos) - INTVAL (width);
5181 else
5182 shift = INTVAL (pos);
5183 if (INTVAL (width) == HOST_BITS_PER_WIDE_INT)
5184 mask = ~(HOST_WIDE_INT) 0;
5185 else
5186 mask = ((HOST_WIDE_INT) 1 << INTVAL (width)) - 1;
5187 val &= ~(mask << shift);
5188 val |= (INTVAL (trial) & mask) << shift;
5189 val = trunc_int_for_mode (val, GET_MODE (dest_reg));
5190 validate_unshare_change (insn, &SET_DEST (sets[i].rtl),
5191 dest_reg, 1);
5192 validate_unshare_change (insn, &SET_SRC (sets[i].rtl),
5193 GEN_INT (val), 1);
5194 if (apply_change_group ())
5196 rtx note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
5197 if (note)
5199 remove_note (insn, note);
5200 df_notes_rescan (insn);
5202 src_eqv = NULL_RTX;
5203 src_eqv_elt = NULL;
5204 src_eqv_volatile = 0;
5205 src_eqv_in_memory = 0;
5206 src_eqv_hash = 0;
5207 repeat = true;
5208 break;
5213 /* We don't normally have an insn matching (set (pc) (pc)), so
5214 check for this separately here. We will delete such an
5215 insn below.
5217 For other cases such as a table jump or conditional jump
5218 where we know the ultimate target, go ahead and replace the
5219 operand. While that may not make a valid insn, we will
5220 reemit the jump below (and also insert any necessary
5221 barriers). */
5222 if (n_sets == 1 && dest == pc_rtx
5223 && (trial == pc_rtx
5224 || (GET_CODE (trial) == LABEL_REF
5225 && ! condjump_p (insn))))
5227 /* Don't substitute non-local labels, this confuses CFG. */
5228 if (GET_CODE (trial) == LABEL_REF
5229 && LABEL_REF_NONLOCAL_P (trial))
5230 continue;
5232 SET_SRC (sets[i].rtl) = trial;
5233 cse_jumps_altered = true;
5234 break;
5237 /* Reject certain invalid forms of CONST that we create. */
5238 else if (CONSTANT_P (trial)
5239 && GET_CODE (trial) == CONST
5240 /* Reject cases that will cause decode_rtx_const to
5241 die. On the alpha when simplifying a switch, we
5242 get (const (truncate (minus (label_ref)
5243 (label_ref)))). */
5244 && (GET_CODE (XEXP (trial, 0)) == TRUNCATE
5245 /* Likewise on IA-64, except without the
5246 truncate. */
5247 || (GET_CODE (XEXP (trial, 0)) == MINUS
5248 && GET_CODE (XEXP (XEXP (trial, 0), 0)) == LABEL_REF
5249 && GET_CODE (XEXP (XEXP (trial, 0), 1)) == LABEL_REF)))
5250 /* Do nothing for this case. */
5253 /* Look for a substitution that makes a valid insn. */
5254 else if (validate_unshare_change
5255 (insn, &SET_SRC (sets[i].rtl), trial, 0))
5257 rtx new_rtx = canon_reg (SET_SRC (sets[i].rtl), insn);
5259 /* The result of apply_change_group can be ignored; see
5260 canon_reg. */
5262 validate_change (insn, &SET_SRC (sets[i].rtl), new_rtx, 1);
5263 apply_change_group ();
5265 break;
5268 /* If we previously found constant pool entries for
5269 constants and this is a constant, try making a
5270 pool entry. Put it in src_folded unless we already have done
5271 this since that is where it likely came from. */
5273 else if (constant_pool_entries_cost
5274 && CONSTANT_P (trial)
5275 && (src_folded == 0
5276 || (!MEM_P (src_folded)
5277 && ! src_folded_force_flag))
5278 && GET_MODE_CLASS (mode) != MODE_CC
5279 && mode != VOIDmode)
5281 src_folded_force_flag = 1;
5282 src_folded = trial;
5283 src_folded_cost = constant_pool_entries_cost;
5284 src_folded_regcost = constant_pool_entries_regcost;
5288 /* If we changed the insn too much, handle this set from scratch. */
5289 if (repeat)
5291 i--;
5292 continue;
5295 src = SET_SRC (sets[i].rtl);
5297 /* In general, it is good to have a SET with SET_SRC == SET_DEST.
5298 However, there is an important exception: If both are registers
5299 that are not the head of their equivalence class, replace SET_SRC
5300 with the head of the class. If we do not do this, we will have
5301 both registers live over a portion of the basic block. This way,
5302 their lifetimes will likely abut instead of overlapping. */
5303 if (REG_P (dest)
5304 && REGNO_QTY_VALID_P (REGNO (dest)))
5306 int dest_q = REG_QTY (REGNO (dest));
5307 struct qty_table_elem *dest_ent = &qty_table[dest_q];
5309 if (dest_ent->mode == GET_MODE (dest)
5310 && dest_ent->first_reg != REGNO (dest)
5311 && REG_P (src) && REGNO (src) == REGNO (dest)
5312 /* Don't do this if the original insn had a hard reg as
5313 SET_SRC or SET_DEST. */
5314 && (!REG_P (sets[i].src)
5315 || REGNO (sets[i].src) >= FIRST_PSEUDO_REGISTER)
5316 && (!REG_P (dest) || REGNO (dest) >= FIRST_PSEUDO_REGISTER))
5317 /* We can't call canon_reg here because it won't do anything if
5318 SRC is a hard register. */
5320 int src_q = REG_QTY (REGNO (src));
5321 struct qty_table_elem *src_ent = &qty_table[src_q];
5322 int first = src_ent->first_reg;
5323 rtx new_src
5324 = (first >= FIRST_PSEUDO_REGISTER
5325 ? regno_reg_rtx[first] : gen_rtx_REG (GET_MODE (src), first));
5327 /* We must use validate-change even for this, because this
5328 might be a special no-op instruction, suitable only to
5329 tag notes onto. */
5330 if (validate_change (insn, &SET_SRC (sets[i].rtl), new_src, 0))
5332 src = new_src;
5333 /* If we had a constant that is cheaper than what we are now
5334 setting SRC to, use that constant. We ignored it when we
5335 thought we could make this into a no-op. */
5336 if (src_const && COST (src_const) < COST (src)
5337 && validate_change (insn, &SET_SRC (sets[i].rtl),
5338 src_const, 0))
5339 src = src_const;
5344 /* If we made a change, recompute SRC values. */
5345 if (src != sets[i].src)
5347 do_not_record = 0;
5348 hash_arg_in_memory = 0;
5349 sets[i].src = src;
5350 sets[i].src_hash = HASH (src, mode);
5351 sets[i].src_volatile = do_not_record;
5352 sets[i].src_in_memory = hash_arg_in_memory;
5353 sets[i].src_elt = lookup (src, sets[i].src_hash, mode);
5356 /* If this is a single SET, we are setting a register, and we have an
5357 equivalent constant, we want to add a REG_EQUAL note if the constant
5358 is different from the source. We don't want to do it for a constant
5359 pseudo since verifying that this pseudo hasn't been eliminated is a
5360 pain; moreover such a note won't help anything.
5362 Avoid a REG_EQUAL note for (CONST (MINUS (LABEL_REF) (LABEL_REF)))
5363 which can be created for a reference to a compile time computable
5364 entry in a jump table. */
5365 if (n_sets == 1
5366 && REG_P (dest)
5367 && src_const
5368 && !REG_P (src_const)
5369 && !(GET_CODE (src_const) == SUBREG
5370 && REG_P (SUBREG_REG (src_const)))
5371 && !(GET_CODE (src_const) == CONST
5372 && GET_CODE (XEXP (src_const, 0)) == MINUS
5373 && GET_CODE (XEXP (XEXP (src_const, 0), 0)) == LABEL_REF
5374 && GET_CODE (XEXP (XEXP (src_const, 0), 1)) == LABEL_REF)
5375 && !rtx_equal_p (src, src_const))
5377 /* Make sure that the rtx is not shared. */
5378 src_const = copy_rtx (src_const);
5380 /* Record the actual constant value in a REG_EQUAL note,
5381 making a new one if one does not already exist. */
5382 set_unique_reg_note (insn, REG_EQUAL, src_const);
5383 df_notes_rescan (insn);
5386 /* Now deal with the destination. */
5387 do_not_record = 0;
5389 /* Look within any ZERO_EXTRACT to the MEM or REG within it. */
5390 while (GET_CODE (dest) == SUBREG
5391 || GET_CODE (dest) == ZERO_EXTRACT
5392 || GET_CODE (dest) == STRICT_LOW_PART)
5393 dest = XEXP (dest, 0);
5395 sets[i].inner_dest = dest;
5397 if (MEM_P (dest))
5399 #ifdef PUSH_ROUNDING
5400 /* Stack pushes invalidate the stack pointer. */
5401 rtx addr = XEXP (dest, 0);
5402 if (GET_RTX_CLASS (GET_CODE (addr)) == RTX_AUTOINC
5403 && XEXP (addr, 0) == stack_pointer_rtx)
5404 invalidate (stack_pointer_rtx, VOIDmode);
5405 #endif
5406 dest = fold_rtx (dest, insn);
5409 /* Compute the hash code of the destination now,
5410 before the effects of this instruction are recorded,
5411 since the register values used in the address computation
5412 are those before this instruction. */
5413 sets[i].dest_hash = HASH (dest, mode);
5415 /* Don't enter a bit-field in the hash table
5416 because the value in it after the store
5417 may not equal what was stored, due to truncation. */
5419 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
5421 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5423 if (src_const != 0 && CONST_INT_P (src_const)
5424 && CONST_INT_P (width)
5425 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
5426 && ! (INTVAL (src_const)
5427 & (HOST_WIDE_INT_M1U << INTVAL (width))))
5428 /* Exception: if the value is constant,
5429 and it won't be truncated, record it. */
5431 else
5433 /* This is chosen so that the destination will be invalidated
5434 but no new value will be recorded.
5435 We must invalidate because sometimes constant
5436 values can be recorded for bitfields. */
5437 sets[i].src_elt = 0;
5438 sets[i].src_volatile = 1;
5439 src_eqv = 0;
5440 src_eqv_elt = 0;
5444 /* If only one set in a JUMP_INSN and it is now a no-op, we can delete
5445 the insn. */
5446 else if (n_sets == 1 && dest == pc_rtx && src == pc_rtx)
5448 /* One less use of the label this insn used to jump to. */
5449 delete_insn_and_edges (insn);
5450 cse_jumps_altered = true;
5451 /* No more processing for this set. */
5452 sets[i].rtl = 0;
5455 /* If this SET is now setting PC to a label, we know it used to
5456 be a conditional or computed branch. */
5457 else if (dest == pc_rtx && GET_CODE (src) == LABEL_REF
5458 && !LABEL_REF_NONLOCAL_P (src))
5460 /* We reemit the jump in as many cases as possible just in
5461 case the form of an unconditional jump is significantly
5462 different than a computed jump or conditional jump.
5464 If this insn has multiple sets, then reemitting the
5465 jump is nontrivial. So instead we just force rerecognition
5466 and hope for the best. */
5467 if (n_sets == 1)
5469 rtx new_rtx, note;
5471 new_rtx = emit_jump_insn_before (gen_jump (XEXP (src, 0)), insn);
5472 JUMP_LABEL (new_rtx) = XEXP (src, 0);
5473 LABEL_NUSES (XEXP (src, 0))++;
5475 /* Make sure to copy over REG_NON_LOCAL_GOTO. */
5476 note = find_reg_note (insn, REG_NON_LOCAL_GOTO, 0);
5477 if (note)
5479 XEXP (note, 1) = NULL_RTX;
5480 REG_NOTES (new_rtx) = note;
5483 delete_insn_and_edges (insn);
5484 insn = new_rtx;
5486 else
5487 INSN_CODE (insn) = -1;
5489 /* Do not bother deleting any unreachable code, let jump do it. */
5490 cse_jumps_altered = true;
5491 sets[i].rtl = 0;
5494 /* If destination is volatile, invalidate it and then do no further
5495 processing for this assignment. */
5497 else if (do_not_record)
5499 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5500 invalidate (dest, VOIDmode);
5501 else if (MEM_P (dest))
5502 invalidate (dest, VOIDmode);
5503 else if (GET_CODE (dest) == STRICT_LOW_PART
5504 || GET_CODE (dest) == ZERO_EXTRACT)
5505 invalidate (XEXP (dest, 0), GET_MODE (dest));
5506 sets[i].rtl = 0;
5509 if (sets[i].rtl != 0 && dest != SET_DEST (sets[i].rtl))
5510 sets[i].dest_hash = HASH (SET_DEST (sets[i].rtl), mode);
5512 #ifdef HAVE_cc0
5513 /* If setting CC0, record what it was set to, or a constant, if it
5514 is equivalent to a constant. If it is being set to a floating-point
5515 value, make a COMPARE with the appropriate constant of 0. If we
5516 don't do this, later code can interpret this as a test against
5517 const0_rtx, which can cause problems if we try to put it into an
5518 insn as a floating-point operand. */
5519 if (dest == cc0_rtx)
5521 this_insn_cc0 = src_const && mode != VOIDmode ? src_const : src;
5522 this_insn_cc0_mode = mode;
5523 if (FLOAT_MODE_P (mode))
5524 this_insn_cc0 = gen_rtx_COMPARE (VOIDmode, this_insn_cc0,
5525 CONST0_RTX (mode));
5527 #endif
5530 /* Now enter all non-volatile source expressions in the hash table
5531 if they are not already present.
5532 Record their equivalence classes in src_elt.
5533 This way we can insert the corresponding destinations into
5534 the same classes even if the actual sources are no longer in them
5535 (having been invalidated). */
5537 if (src_eqv && src_eqv_elt == 0 && sets[0].rtl != 0 && ! src_eqv_volatile
5538 && ! rtx_equal_p (src_eqv, SET_DEST (sets[0].rtl)))
5540 struct table_elt *elt;
5541 struct table_elt *classp = sets[0].src_elt;
5542 rtx dest = SET_DEST (sets[0].rtl);
5543 enum machine_mode eqvmode = GET_MODE (dest);
5545 if (GET_CODE (dest) == STRICT_LOW_PART)
5547 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
5548 classp = 0;
5550 if (insert_regs (src_eqv, classp, 0))
5552 rehash_using_reg (src_eqv);
5553 src_eqv_hash = HASH (src_eqv, eqvmode);
5555 elt = insert (src_eqv, classp, src_eqv_hash, eqvmode);
5556 elt->in_memory = src_eqv_in_memory;
5557 src_eqv_elt = elt;
5559 /* Check to see if src_eqv_elt is the same as a set source which
5560 does not yet have an elt, and if so set the elt of the set source
5561 to src_eqv_elt. */
5562 for (i = 0; i < n_sets; i++)
5563 if (sets[i].rtl && sets[i].src_elt == 0
5564 && rtx_equal_p (SET_SRC (sets[i].rtl), src_eqv))
5565 sets[i].src_elt = src_eqv_elt;
5568 for (i = 0; i < n_sets; i++)
5569 if (sets[i].rtl && ! sets[i].src_volatile
5570 && ! rtx_equal_p (SET_SRC (sets[i].rtl), SET_DEST (sets[i].rtl)))
5572 if (GET_CODE (SET_DEST (sets[i].rtl)) == STRICT_LOW_PART)
5574 /* REG_EQUAL in setting a STRICT_LOW_PART
5575 gives an equivalent for the entire destination register,
5576 not just for the subreg being stored in now.
5577 This is a more interesting equivalence, so we arrange later
5578 to treat the entire reg as the destination. */
5579 sets[i].src_elt = src_eqv_elt;
5580 sets[i].src_hash = src_eqv_hash;
5582 else
5584 /* Insert source and constant equivalent into hash table, if not
5585 already present. */
5586 struct table_elt *classp = src_eqv_elt;
5587 rtx src = sets[i].src;
5588 rtx dest = SET_DEST (sets[i].rtl);
5589 enum machine_mode mode
5590 = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
5592 /* It's possible that we have a source value known to be
5593 constant but don't have a REG_EQUAL note on the insn.
5594 Lack of a note will mean src_eqv_elt will be NULL. This
5595 can happen where we've generated a SUBREG to access a
5596 CONST_INT that is already in a register in a wider mode.
5597 Ensure that the source expression is put in the proper
5598 constant class. */
5599 if (!classp)
5600 classp = sets[i].src_const_elt;
5602 if (sets[i].src_elt == 0)
5604 struct table_elt *elt;
5606 /* Note that these insert_regs calls cannot remove
5607 any of the src_elt's, because they would have failed to
5608 match if not still valid. */
5609 if (insert_regs (src, classp, 0))
5611 rehash_using_reg (src);
5612 sets[i].src_hash = HASH (src, mode);
5614 elt = insert (src, classp, sets[i].src_hash, mode);
5615 elt->in_memory = sets[i].src_in_memory;
5616 sets[i].src_elt = classp = elt;
5618 if (sets[i].src_const && sets[i].src_const_elt == 0
5619 && src != sets[i].src_const
5620 && ! rtx_equal_p (sets[i].src_const, src))
5621 sets[i].src_elt = insert (sets[i].src_const, classp,
5622 sets[i].src_const_hash, mode);
5625 else if (sets[i].src_elt == 0)
5626 /* If we did not insert the source into the hash table (e.g., it was
5627 volatile), note the equivalence class for the REG_EQUAL value, if any,
5628 so that the destination goes into that class. */
5629 sets[i].src_elt = src_eqv_elt;
5631 /* Record destination addresses in the hash table. This allows us to
5632 check if they are invalidated by other sets. */
5633 for (i = 0; i < n_sets; i++)
5635 if (sets[i].rtl)
5637 rtx x = sets[i].inner_dest;
5638 struct table_elt *elt;
5639 enum machine_mode mode;
5640 unsigned hash;
5642 if (MEM_P (x))
5644 x = XEXP (x, 0);
5645 mode = GET_MODE (x);
5646 hash = HASH (x, mode);
5647 elt = lookup (x, hash, mode);
5648 if (!elt)
5650 if (insert_regs (x, NULL, 0))
5652 rtx dest = SET_DEST (sets[i].rtl);
5654 rehash_using_reg (x);
5655 hash = HASH (x, mode);
5656 sets[i].dest_hash = HASH (dest, GET_MODE (dest));
5658 elt = insert (x, NULL, hash, mode);
5661 sets[i].dest_addr_elt = elt;
5663 else
5664 sets[i].dest_addr_elt = NULL;
5668 invalidate_from_clobbers (insn);
5670 /* Some registers are invalidated by subroutine calls. Memory is
5671 invalidated by non-constant calls. */
5673 if (CALL_P (insn))
5675 if (!(RTL_CONST_OR_PURE_CALL_P (insn)))
5676 invalidate_memory ();
5677 invalidate_for_call ();
5680 /* Now invalidate everything set by this instruction.
5681 If a SUBREG or other funny destination is being set,
5682 sets[i].rtl is still nonzero, so here we invalidate the reg
5683 a part of which is being set. */
5685 for (i = 0; i < n_sets; i++)
5686 if (sets[i].rtl)
5688 /* We can't use the inner dest, because the mode associated with
5689 a ZERO_EXTRACT is significant. */
5690 rtx dest = SET_DEST (sets[i].rtl);
5692 /* Needed for registers to remove the register from its
5693 previous quantity's chain.
5694 Needed for memory if this is a nonvarying address, unless
5695 we have just done an invalidate_memory that covers even those. */
5696 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5697 invalidate (dest, VOIDmode);
5698 else if (MEM_P (dest))
5699 invalidate (dest, VOIDmode);
5700 else if (GET_CODE (dest) == STRICT_LOW_PART
5701 || GET_CODE (dest) == ZERO_EXTRACT)
5702 invalidate (XEXP (dest, 0), GET_MODE (dest));
5705 /* Don't cse over a call to setjmp; on some machines (eg VAX)
5706 the regs restored by the longjmp come from a later time
5707 than the setjmp. */
5708 if (CALL_P (insn) && find_reg_note (insn, REG_SETJMP, NULL))
5710 flush_hash_table ();
5711 goto done;
5714 /* Make sure registers mentioned in destinations
5715 are safe for use in an expression to be inserted.
5716 This removes from the hash table
5717 any invalid entry that refers to one of these registers.
5719 We don't care about the return value from mention_regs because
5720 we are going to hash the SET_DEST values unconditionally. */
5722 for (i = 0; i < n_sets; i++)
5724 if (sets[i].rtl)
5726 rtx x = SET_DEST (sets[i].rtl);
5728 if (!REG_P (x))
5729 mention_regs (x);
5730 else
5732 /* We used to rely on all references to a register becoming
5733 inaccessible when a register changes to a new quantity,
5734 since that changes the hash code. However, that is not
5735 safe, since after HASH_SIZE new quantities we get a
5736 hash 'collision' of a register with its own invalid
5737 entries. And since SUBREGs have been changed not to
5738 change their hash code with the hash code of the register,
5739 it wouldn't work any longer at all. So we have to check
5740 for any invalid references lying around now.
5741 This code is similar to the REG case in mention_regs,
5742 but it knows that reg_tick has been incremented, and
5743 it leaves reg_in_table as -1 . */
5744 unsigned int regno = REGNO (x);
5745 unsigned int endregno = END_REGNO (x);
5746 unsigned int i;
5748 for (i = regno; i < endregno; i++)
5750 if (REG_IN_TABLE (i) >= 0)
5752 remove_invalid_refs (i);
5753 REG_IN_TABLE (i) = -1;
5760 /* We may have just removed some of the src_elt's from the hash table.
5761 So replace each one with the current head of the same class.
5762 Also check if destination addresses have been removed. */
5764 for (i = 0; i < n_sets; i++)
5765 if (sets[i].rtl)
5767 if (sets[i].dest_addr_elt
5768 && sets[i].dest_addr_elt->first_same_value == 0)
5770 /* The elt was removed, which means this destination is not
5771 valid after this instruction. */
5772 sets[i].rtl = NULL_RTX;
5774 else if (sets[i].src_elt && sets[i].src_elt->first_same_value == 0)
5775 /* If elt was removed, find current head of same class,
5776 or 0 if nothing remains of that class. */
5778 struct table_elt *elt = sets[i].src_elt;
5780 while (elt && elt->prev_same_value)
5781 elt = elt->prev_same_value;
5783 while (elt && elt->first_same_value == 0)
5784 elt = elt->next_same_value;
5785 sets[i].src_elt = elt ? elt->first_same_value : 0;
5789 /* Now insert the destinations into their equivalence classes. */
5791 for (i = 0; i < n_sets; i++)
5792 if (sets[i].rtl)
5794 rtx dest = SET_DEST (sets[i].rtl);
5795 struct table_elt *elt;
5797 /* Don't record value if we are not supposed to risk allocating
5798 floating-point values in registers that might be wider than
5799 memory. */
5800 if ((flag_float_store
5801 && MEM_P (dest)
5802 && FLOAT_MODE_P (GET_MODE (dest)))
5803 /* Don't record BLKmode values, because we don't know the
5804 size of it, and can't be sure that other BLKmode values
5805 have the same or smaller size. */
5806 || GET_MODE (dest) == BLKmode
5807 /* If we didn't put a REG_EQUAL value or a source into the hash
5808 table, there is no point is recording DEST. */
5809 || sets[i].src_elt == 0
5810 /* If DEST is a paradoxical SUBREG and SRC is a ZERO_EXTEND
5811 or SIGN_EXTEND, don't record DEST since it can cause
5812 some tracking to be wrong.
5814 ??? Think about this more later. */
5815 || (paradoxical_subreg_p (dest)
5816 && (GET_CODE (sets[i].src) == SIGN_EXTEND
5817 || GET_CODE (sets[i].src) == ZERO_EXTEND)))
5818 continue;
5820 /* STRICT_LOW_PART isn't part of the value BEING set,
5821 and neither is the SUBREG inside it.
5822 Note that in this case SETS[I].SRC_ELT is really SRC_EQV_ELT. */
5823 if (GET_CODE (dest) == STRICT_LOW_PART)
5824 dest = SUBREG_REG (XEXP (dest, 0));
5826 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5827 /* Registers must also be inserted into chains for quantities. */
5828 if (insert_regs (dest, sets[i].src_elt, 1))
5830 /* If `insert_regs' changes something, the hash code must be
5831 recalculated. */
5832 rehash_using_reg (dest);
5833 sets[i].dest_hash = HASH (dest, GET_MODE (dest));
5836 elt = insert (dest, sets[i].src_elt,
5837 sets[i].dest_hash, GET_MODE (dest));
5839 /* If this is a constant, insert the constant anchors with the
5840 equivalent register-offset expressions using register DEST. */
5841 if (targetm.const_anchor
5842 && REG_P (dest)
5843 && SCALAR_INT_MODE_P (GET_MODE (dest))
5844 && GET_CODE (sets[i].src_elt->exp) == CONST_INT)
5845 insert_const_anchors (dest, sets[i].src_elt->exp, GET_MODE (dest));
5847 elt->in_memory = (MEM_P (sets[i].inner_dest)
5848 && !MEM_READONLY_P (sets[i].inner_dest));
5850 /* If we have (set (subreg:m1 (reg:m2 foo) 0) (bar:m1)), M1 is no
5851 narrower than M2, and both M1 and M2 are the same number of words,
5852 we are also doing (set (reg:m2 foo) (subreg:m2 (bar:m1) 0)) so
5853 make that equivalence as well.
5855 However, BAR may have equivalences for which gen_lowpart
5856 will produce a simpler value than gen_lowpart applied to
5857 BAR (e.g., if BAR was ZERO_EXTENDed from M2), so we will scan all
5858 BAR's equivalences. If we don't get a simplified form, make
5859 the SUBREG. It will not be used in an equivalence, but will
5860 cause two similar assignments to be detected.
5862 Note the loop below will find SUBREG_REG (DEST) since we have
5863 already entered SRC and DEST of the SET in the table. */
5865 if (GET_CODE (dest) == SUBREG
5866 && (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))) - 1)
5867 / UNITS_PER_WORD)
5868 == (GET_MODE_SIZE (GET_MODE (dest)) - 1) / UNITS_PER_WORD)
5869 && (GET_MODE_SIZE (GET_MODE (dest))
5870 >= GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
5871 && sets[i].src_elt != 0)
5873 enum machine_mode new_mode = GET_MODE (SUBREG_REG (dest));
5874 struct table_elt *elt, *classp = 0;
5876 for (elt = sets[i].src_elt->first_same_value; elt;
5877 elt = elt->next_same_value)
5879 rtx new_src = 0;
5880 unsigned src_hash;
5881 struct table_elt *src_elt;
5882 int byte = 0;
5884 /* Ignore invalid entries. */
5885 if (!REG_P (elt->exp)
5886 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
5887 continue;
5889 /* We may have already been playing subreg games. If the
5890 mode is already correct for the destination, use it. */
5891 if (GET_MODE (elt->exp) == new_mode)
5892 new_src = elt->exp;
5893 else
5895 /* Calculate big endian correction for the SUBREG_BYTE.
5896 We have already checked that M1 (GET_MODE (dest))
5897 is not narrower than M2 (new_mode). */
5898 if (BYTES_BIG_ENDIAN)
5899 byte = (GET_MODE_SIZE (GET_MODE (dest))
5900 - GET_MODE_SIZE (new_mode));
5902 new_src = simplify_gen_subreg (new_mode, elt->exp,
5903 GET_MODE (dest), byte);
5906 /* The call to simplify_gen_subreg fails if the value
5907 is VOIDmode, yet we can't do any simplification, e.g.
5908 for EXPR_LISTs denoting function call results.
5909 It is invalid to construct a SUBREG with a VOIDmode
5910 SUBREG_REG, hence a zero new_src means we can't do
5911 this substitution. */
5912 if (! new_src)
5913 continue;
5915 src_hash = HASH (new_src, new_mode);
5916 src_elt = lookup (new_src, src_hash, new_mode);
5918 /* Put the new source in the hash table is if isn't
5919 already. */
5920 if (src_elt == 0)
5922 if (insert_regs (new_src, classp, 0))
5924 rehash_using_reg (new_src);
5925 src_hash = HASH (new_src, new_mode);
5927 src_elt = insert (new_src, classp, src_hash, new_mode);
5928 src_elt->in_memory = elt->in_memory;
5930 else if (classp && classp != src_elt->first_same_value)
5931 /* Show that two things that we've seen before are
5932 actually the same. */
5933 merge_equiv_classes (src_elt, classp);
5935 classp = src_elt->first_same_value;
5936 /* Ignore invalid entries. */
5937 while (classp
5938 && !REG_P (classp->exp)
5939 && ! exp_equiv_p (classp->exp, classp->exp, 1, false))
5940 classp = classp->next_same_value;
5945 /* Special handling for (set REG0 REG1) where REG0 is the
5946 "cheapest", cheaper than REG1. After cse, REG1 will probably not
5947 be used in the sequel, so (if easily done) change this insn to
5948 (set REG1 REG0) and replace REG1 with REG0 in the previous insn
5949 that computed their value. Then REG1 will become a dead store
5950 and won't cloud the situation for later optimizations.
5952 Do not make this change if REG1 is a hard register, because it will
5953 then be used in the sequel and we may be changing a two-operand insn
5954 into a three-operand insn.
5956 Also do not do this if we are operating on a copy of INSN. */
5958 if (n_sets == 1 && sets[0].rtl)
5959 try_back_substitute_reg (sets[0].rtl, insn);
5961 done:;
5964 /* Remove from the hash table all expressions that reference memory. */
5966 static void
5967 invalidate_memory (void)
5969 int i;
5970 struct table_elt *p, *next;
5972 for (i = 0; i < HASH_SIZE; i++)
5973 for (p = table[i]; p; p = next)
5975 next = p->next_same_hash;
5976 if (p->in_memory)
5977 remove_from_table (p, i);
5981 /* Perform invalidation on the basis of everything about INSN,
5982 except for invalidating the actual places that are SET in it.
5983 This includes the places CLOBBERed, and anything that might
5984 alias with something that is SET or CLOBBERed. */
5986 static void
5987 invalidate_from_clobbers (rtx insn)
5989 rtx x = PATTERN (insn);
5991 if (GET_CODE (x) == CLOBBER)
5993 rtx ref = XEXP (x, 0);
5994 if (ref)
5996 if (REG_P (ref) || GET_CODE (ref) == SUBREG
5997 || MEM_P (ref))
5998 invalidate (ref, VOIDmode);
5999 else if (GET_CODE (ref) == STRICT_LOW_PART
6000 || GET_CODE (ref) == ZERO_EXTRACT)
6001 invalidate (XEXP (ref, 0), GET_MODE (ref));
6004 else if (GET_CODE (x) == PARALLEL)
6006 int i;
6007 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
6009 rtx y = XVECEXP (x, 0, i);
6010 if (GET_CODE (y) == CLOBBER)
6012 rtx ref = XEXP (y, 0);
6013 if (REG_P (ref) || GET_CODE (ref) == SUBREG
6014 || MEM_P (ref))
6015 invalidate (ref, VOIDmode);
6016 else if (GET_CODE (ref) == STRICT_LOW_PART
6017 || GET_CODE (ref) == ZERO_EXTRACT)
6018 invalidate (XEXP (ref, 0), GET_MODE (ref));
6024 /* Perform invalidation on the basis of everything about INSN.
6025 This includes the places CLOBBERed, and anything that might
6026 alias with something that is SET or CLOBBERed. */
6028 static void
6029 invalidate_from_sets_and_clobbers (rtx insn)
6031 rtx tem;
6032 rtx x = PATTERN (insn);
6034 if (CALL_P (insn))
6036 for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
6037 if (GET_CODE (XEXP (tem, 0)) == CLOBBER)
6038 invalidate (SET_DEST (XEXP (tem, 0)), VOIDmode);
6041 /* Ensure we invalidate the destination register of a CALL insn.
6042 This is necessary for machines where this register is a fixed_reg,
6043 because no other code would invalidate it. */
6044 if (GET_CODE (x) == SET && GET_CODE (SET_SRC (x)) == CALL)
6045 invalidate (SET_DEST (x), VOIDmode);
6047 else if (GET_CODE (x) == PARALLEL)
6049 int i;
6051 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
6053 rtx y = XVECEXP (x, 0, i);
6054 if (GET_CODE (y) == CLOBBER)
6056 rtx clobbered = XEXP (y, 0);
6058 if (REG_P (clobbered)
6059 || GET_CODE (clobbered) == SUBREG)
6060 invalidate (clobbered, VOIDmode);
6061 else if (GET_CODE (clobbered) == STRICT_LOW_PART
6062 || GET_CODE (clobbered) == ZERO_EXTRACT)
6063 invalidate (XEXP (clobbered, 0), GET_MODE (clobbered));
6065 else if (GET_CODE (y) == SET && GET_CODE (SET_SRC (y)) == CALL)
6066 invalidate (SET_DEST (y), VOIDmode);
6071 /* Process X, part of the REG_NOTES of an insn. Look at any REG_EQUAL notes
6072 and replace any registers in them with either an equivalent constant
6073 or the canonical form of the register. If we are inside an address,
6074 only do this if the address remains valid.
6076 OBJECT is 0 except when within a MEM in which case it is the MEM.
6078 Return the replacement for X. */
6080 static rtx
6081 cse_process_notes_1 (rtx x, rtx object, bool *changed)
6083 enum rtx_code code = GET_CODE (x);
6084 const char *fmt = GET_RTX_FORMAT (code);
6085 int i;
6087 switch (code)
6089 case CONST:
6090 case SYMBOL_REF:
6091 case LABEL_REF:
6092 CASE_CONST_ANY:
6093 case PC:
6094 case CC0:
6095 case LO_SUM:
6096 return x;
6098 case MEM:
6099 validate_change (x, &XEXP (x, 0),
6100 cse_process_notes (XEXP (x, 0), x, changed), 0);
6101 return x;
6103 case EXPR_LIST:
6104 if (REG_NOTE_KIND (x) == REG_EQUAL)
6105 XEXP (x, 0) = cse_process_notes (XEXP (x, 0), NULL_RTX, changed);
6106 /* Fall through. */
6108 case INSN_LIST:
6109 case INT_LIST:
6110 if (XEXP (x, 1))
6111 XEXP (x, 1) = cse_process_notes (XEXP (x, 1), NULL_RTX, changed);
6112 return x;
6114 case SIGN_EXTEND:
6115 case ZERO_EXTEND:
6116 case SUBREG:
6118 rtx new_rtx = cse_process_notes (XEXP (x, 0), object, changed);
6119 /* We don't substitute VOIDmode constants into these rtx,
6120 since they would impede folding. */
6121 if (GET_MODE (new_rtx) != VOIDmode)
6122 validate_change (object, &XEXP (x, 0), new_rtx, 0);
6123 return x;
6126 case UNSIGNED_FLOAT:
6128 rtx new_rtx = cse_process_notes (XEXP (x, 0), object, changed);
6129 /* We don't substitute negative VOIDmode constants into these rtx,
6130 since they would impede folding. */
6131 if (GET_MODE (new_rtx) != VOIDmode
6132 || (CONST_INT_P (new_rtx) && INTVAL (new_rtx) >= 0)
6133 || (CONST_DOUBLE_P (new_rtx) && CONST_DOUBLE_HIGH (new_rtx) >= 0))
6134 validate_change (object, &XEXP (x, 0), new_rtx, 0);
6135 return x;
6138 case REG:
6139 i = REG_QTY (REGNO (x));
6141 /* Return a constant or a constant register. */
6142 if (REGNO_QTY_VALID_P (REGNO (x)))
6144 struct qty_table_elem *ent = &qty_table[i];
6146 if (ent->const_rtx != NULL_RTX
6147 && (CONSTANT_P (ent->const_rtx)
6148 || REG_P (ent->const_rtx)))
6150 rtx new_rtx = gen_lowpart (GET_MODE (x), ent->const_rtx);
6151 if (new_rtx)
6152 return copy_rtx (new_rtx);
6156 /* Otherwise, canonicalize this register. */
6157 return canon_reg (x, NULL_RTX);
6159 default:
6160 break;
6163 for (i = 0; i < GET_RTX_LENGTH (code); i++)
6164 if (fmt[i] == 'e')
6165 validate_change (object, &XEXP (x, i),
6166 cse_process_notes (XEXP (x, i), object, changed), 0);
6168 return x;
6171 static rtx
6172 cse_process_notes (rtx x, rtx object, bool *changed)
6174 rtx new_rtx = cse_process_notes_1 (x, object, changed);
6175 if (new_rtx != x)
6176 *changed = true;
6177 return new_rtx;
6181 /* Find a path in the CFG, starting with FIRST_BB to perform CSE on.
6183 DATA is a pointer to a struct cse_basic_block_data, that is used to
6184 describe the path.
6185 It is filled with a queue of basic blocks, starting with FIRST_BB
6186 and following a trace through the CFG.
6188 If all paths starting at FIRST_BB have been followed, or no new path
6189 starting at FIRST_BB can be constructed, this function returns FALSE.
6190 Otherwise, DATA->path is filled and the function returns TRUE indicating
6191 that a path to follow was found.
6193 If FOLLOW_JUMPS is false, the maximum path length is 1 and the only
6194 block in the path will be FIRST_BB. */
6196 static bool
6197 cse_find_path (basic_block first_bb, struct cse_basic_block_data *data,
6198 int follow_jumps)
6200 basic_block bb;
6201 edge e;
6202 int path_size;
6204 bitmap_set_bit (cse_visited_basic_blocks, first_bb->index);
6206 /* See if there is a previous path. */
6207 path_size = data->path_size;
6209 /* There is a previous path. Make sure it started with FIRST_BB. */
6210 if (path_size)
6211 gcc_assert (data->path[0].bb == first_bb);
6213 /* There was only one basic block in the last path. Clear the path and
6214 return, so that paths starting at another basic block can be tried. */
6215 if (path_size == 1)
6217 path_size = 0;
6218 goto done;
6221 /* If the path was empty from the beginning, construct a new path. */
6222 if (path_size == 0)
6223 data->path[path_size++].bb = first_bb;
6224 else
6226 /* Otherwise, path_size must be equal to or greater than 2, because
6227 a previous path exists that is at least two basic blocks long.
6229 Update the previous branch path, if any. If the last branch was
6230 previously along the branch edge, take the fallthrough edge now. */
6231 while (path_size >= 2)
6233 basic_block last_bb_in_path, previous_bb_in_path;
6234 edge e;
6236 --path_size;
6237 last_bb_in_path = data->path[path_size].bb;
6238 previous_bb_in_path = data->path[path_size - 1].bb;
6240 /* If we previously followed a path along the branch edge, try
6241 the fallthru edge now. */
6242 if (EDGE_COUNT (previous_bb_in_path->succs) == 2
6243 && any_condjump_p (BB_END (previous_bb_in_path))
6244 && (e = find_edge (previous_bb_in_path, last_bb_in_path))
6245 && e == BRANCH_EDGE (previous_bb_in_path))
6247 bb = FALLTHRU_EDGE (previous_bb_in_path)->dest;
6248 if (bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
6249 && single_pred_p (bb)
6250 /* We used to assert here that we would only see blocks
6251 that we have not visited yet. But we may end up
6252 visiting basic blocks twice if the CFG has changed
6253 in this run of cse_main, because when the CFG changes
6254 the topological sort of the CFG also changes. A basic
6255 blocks that previously had more than two predecessors
6256 may now have a single predecessor, and become part of
6257 a path that starts at another basic block.
6259 We still want to visit each basic block only once, so
6260 halt the path here if we have already visited BB. */
6261 && !bitmap_bit_p (cse_visited_basic_blocks, bb->index))
6263 bitmap_set_bit (cse_visited_basic_blocks, bb->index);
6264 data->path[path_size++].bb = bb;
6265 break;
6269 data->path[path_size].bb = NULL;
6272 /* If only one block remains in the path, bail. */
6273 if (path_size == 1)
6275 path_size = 0;
6276 goto done;
6280 /* Extend the path if possible. */
6281 if (follow_jumps)
6283 bb = data->path[path_size - 1].bb;
6284 while (bb && path_size < PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH))
6286 if (single_succ_p (bb))
6287 e = single_succ_edge (bb);
6288 else if (EDGE_COUNT (bb->succs) == 2
6289 && any_condjump_p (BB_END (bb)))
6291 /* First try to follow the branch. If that doesn't lead
6292 to a useful path, follow the fallthru edge. */
6293 e = BRANCH_EDGE (bb);
6294 if (!single_pred_p (e->dest))
6295 e = FALLTHRU_EDGE (bb);
6297 else
6298 e = NULL;
6300 if (e
6301 && !((e->flags & EDGE_ABNORMAL_CALL) && cfun->has_nonlocal_label)
6302 && e->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
6303 && single_pred_p (e->dest)
6304 /* Avoid visiting basic blocks twice. The large comment
6305 above explains why this can happen. */
6306 && !bitmap_bit_p (cse_visited_basic_blocks, e->dest->index))
6308 basic_block bb2 = e->dest;
6309 bitmap_set_bit (cse_visited_basic_blocks, bb2->index);
6310 data->path[path_size++].bb = bb2;
6311 bb = bb2;
6313 else
6314 bb = NULL;
6318 done:
6319 data->path_size = path_size;
6320 return path_size != 0;
6323 /* Dump the path in DATA to file F. NSETS is the number of sets
6324 in the path. */
6326 static void
6327 cse_dump_path (struct cse_basic_block_data *data, int nsets, FILE *f)
6329 int path_entry;
6331 fprintf (f, ";; Following path with %d sets: ", nsets);
6332 for (path_entry = 0; path_entry < data->path_size; path_entry++)
6333 fprintf (f, "%d ", (data->path[path_entry].bb)->index);
6334 fputc ('\n', dump_file);
6335 fflush (f);
6339 /* Return true if BB has exception handling successor edges. */
6341 static bool
6342 have_eh_succ_edges (basic_block bb)
6344 edge e;
6345 edge_iterator ei;
6347 FOR_EACH_EDGE (e, ei, bb->succs)
6348 if (e->flags & EDGE_EH)
6349 return true;
6351 return false;
6355 /* Scan to the end of the path described by DATA. Return an estimate of
6356 the total number of SETs of all insns in the path. */
6358 static void
6359 cse_prescan_path (struct cse_basic_block_data *data)
6361 int nsets = 0;
6362 int path_size = data->path_size;
6363 int path_entry;
6365 /* Scan to end of each basic block in the path. */
6366 for (path_entry = 0; path_entry < path_size; path_entry++)
6368 basic_block bb;
6369 rtx insn;
6371 bb = data->path[path_entry].bb;
6373 FOR_BB_INSNS (bb, insn)
6375 if (!INSN_P (insn))
6376 continue;
6378 /* A PARALLEL can have lots of SETs in it,
6379 especially if it is really an ASM_OPERANDS. */
6380 if (GET_CODE (PATTERN (insn)) == PARALLEL)
6381 nsets += XVECLEN (PATTERN (insn), 0);
6382 else
6383 nsets += 1;
6387 data->nsets = nsets;
6390 /* Process a single extended basic block described by EBB_DATA. */
6392 static void
6393 cse_extended_basic_block (struct cse_basic_block_data *ebb_data)
6395 int path_size = ebb_data->path_size;
6396 int path_entry;
6397 int num_insns = 0;
6399 /* Allocate the space needed by qty_table. */
6400 qty_table = XNEWVEC (struct qty_table_elem, max_qty);
6402 new_basic_block ();
6403 cse_ebb_live_in = df_get_live_in (ebb_data->path[0].bb);
6404 cse_ebb_live_out = df_get_live_out (ebb_data->path[path_size - 1].bb);
6405 for (path_entry = 0; path_entry < path_size; path_entry++)
6407 basic_block bb;
6408 rtx insn;
6410 bb = ebb_data->path[path_entry].bb;
6412 /* Invalidate recorded information for eh regs if there is an EH
6413 edge pointing to that bb. */
6414 if (bb_has_eh_pred (bb))
6416 df_ref def;
6418 FOR_EACH_ARTIFICIAL_DEF (def, bb->index)
6419 if (DF_REF_FLAGS (def) & DF_REF_AT_TOP)
6420 invalidate (DF_REF_REG (def), GET_MODE (DF_REF_REG (def)));
6423 optimize_this_for_speed_p = optimize_bb_for_speed_p (bb);
6424 FOR_BB_INSNS (bb, insn)
6426 /* If we have processed 1,000 insns, flush the hash table to
6427 avoid extreme quadratic behavior. We must not include NOTEs
6428 in the count since there may be more of them when generating
6429 debugging information. If we clear the table at different
6430 times, code generated with -g -O might be different than code
6431 generated with -O but not -g.
6433 FIXME: This is a real kludge and needs to be done some other
6434 way. */
6435 if (NONDEBUG_INSN_P (insn)
6436 && num_insns++ > PARAM_VALUE (PARAM_MAX_CSE_INSNS))
6438 flush_hash_table ();
6439 num_insns = 0;
6442 if (INSN_P (insn))
6444 /* Process notes first so we have all notes in canonical forms
6445 when looking for duplicate operations. */
6446 if (REG_NOTES (insn))
6448 bool changed = false;
6449 REG_NOTES (insn) = cse_process_notes (REG_NOTES (insn),
6450 NULL_RTX, &changed);
6451 if (changed)
6452 df_notes_rescan (insn);
6455 cse_insn (insn);
6457 /* If we haven't already found an insn where we added a LABEL_REF,
6458 check this one. */
6459 if (INSN_P (insn) && !recorded_label_ref
6460 && for_each_rtx (&PATTERN (insn), check_for_label_ref,
6461 (void *) insn))
6462 recorded_label_ref = true;
6464 #ifdef HAVE_cc0
6465 if (NONDEBUG_INSN_P (insn))
6467 /* If the previous insn sets CC0 and this insn no
6468 longer references CC0, delete the previous insn.
6469 Here we use fact that nothing expects CC0 to be
6470 valid over an insn, which is true until the final
6471 pass. */
6472 rtx prev_insn, tem;
6474 prev_insn = prev_nonnote_nondebug_insn (insn);
6475 if (prev_insn && NONJUMP_INSN_P (prev_insn)
6476 && (tem = single_set (prev_insn)) != NULL_RTX
6477 && SET_DEST (tem) == cc0_rtx
6478 && ! reg_mentioned_p (cc0_rtx, PATTERN (insn)))
6479 delete_insn (prev_insn);
6481 /* If this insn is not the last insn in the basic
6482 block, it will be PREV_INSN(insn) in the next
6483 iteration. If we recorded any CC0-related
6484 information for this insn, remember it. */
6485 if (insn != BB_END (bb))
6487 prev_insn_cc0 = this_insn_cc0;
6488 prev_insn_cc0_mode = this_insn_cc0_mode;
6491 #endif
6495 /* With non-call exceptions, we are not always able to update
6496 the CFG properly inside cse_insn. So clean up possibly
6497 redundant EH edges here. */
6498 if (cfun->can_throw_non_call_exceptions && have_eh_succ_edges (bb))
6499 cse_cfg_altered |= purge_dead_edges (bb);
6501 /* If we changed a conditional jump, we may have terminated
6502 the path we are following. Check that by verifying that
6503 the edge we would take still exists. If the edge does
6504 not exist anymore, purge the remainder of the path.
6505 Note that this will cause us to return to the caller. */
6506 if (path_entry < path_size - 1)
6508 basic_block next_bb = ebb_data->path[path_entry + 1].bb;
6509 if (!find_edge (bb, next_bb))
6513 path_size--;
6515 /* If we truncate the path, we must also reset the
6516 visited bit on the remaining blocks in the path,
6517 or we will never visit them at all. */
6518 bitmap_clear_bit (cse_visited_basic_blocks,
6519 ebb_data->path[path_size].bb->index);
6520 ebb_data->path[path_size].bb = NULL;
6522 while (path_size - 1 != path_entry);
6523 ebb_data->path_size = path_size;
6527 /* If this is a conditional jump insn, record any known
6528 equivalences due to the condition being tested. */
6529 insn = BB_END (bb);
6530 if (path_entry < path_size - 1
6531 && JUMP_P (insn)
6532 && single_set (insn)
6533 && any_condjump_p (insn))
6535 basic_block next_bb = ebb_data->path[path_entry + 1].bb;
6536 bool taken = (next_bb == BRANCH_EDGE (bb)->dest);
6537 record_jump_equiv (insn, taken);
6540 #ifdef HAVE_cc0
6541 /* Clear the CC0-tracking related insns, they can't provide
6542 useful information across basic block boundaries. */
6543 prev_insn_cc0 = 0;
6544 #endif
6547 gcc_assert (next_qty <= max_qty);
6549 free (qty_table);
6553 /* Perform cse on the instructions of a function.
6554 F is the first instruction.
6555 NREGS is one plus the highest pseudo-reg number used in the instruction.
6557 Return 2 if jump optimizations should be redone due to simplifications
6558 in conditional jump instructions.
6559 Return 1 if the CFG should be cleaned up because it has been modified.
6560 Return 0 otherwise. */
6562 static int
6563 cse_main (rtx f ATTRIBUTE_UNUSED, int nregs)
6565 struct cse_basic_block_data ebb_data;
6566 basic_block bb;
6567 int *rc_order = XNEWVEC (int, last_basic_block_for_fn (cfun));
6568 int i, n_blocks;
6570 df_set_flags (DF_LR_RUN_DCE);
6571 df_note_add_problem ();
6572 df_analyze ();
6573 df_set_flags (DF_DEFER_INSN_RESCAN);
6575 reg_scan (get_insns (), max_reg_num ());
6576 init_cse_reg_info (nregs);
6578 ebb_data.path = XNEWVEC (struct branch_path,
6579 PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH));
6581 cse_cfg_altered = false;
6582 cse_jumps_altered = false;
6583 recorded_label_ref = false;
6584 constant_pool_entries_cost = 0;
6585 constant_pool_entries_regcost = 0;
6586 ebb_data.path_size = 0;
6587 ebb_data.nsets = 0;
6588 rtl_hooks = cse_rtl_hooks;
6590 init_recog ();
6591 init_alias_analysis ();
6593 reg_eqv_table = XNEWVEC (struct reg_eqv_elem, nregs);
6595 /* Set up the table of already visited basic blocks. */
6596 cse_visited_basic_blocks = sbitmap_alloc (last_basic_block_for_fn (cfun));
6597 bitmap_clear (cse_visited_basic_blocks);
6599 /* Loop over basic blocks in reverse completion order (RPO),
6600 excluding the ENTRY and EXIT blocks. */
6601 n_blocks = pre_and_rev_post_order_compute (NULL, rc_order, false);
6602 i = 0;
6603 while (i < n_blocks)
6605 /* Find the first block in the RPO queue that we have not yet
6606 processed before. */
6609 bb = BASIC_BLOCK_FOR_FN (cfun, rc_order[i++]);
6611 while (bitmap_bit_p (cse_visited_basic_blocks, bb->index)
6612 && i < n_blocks);
6614 /* Find all paths starting with BB, and process them. */
6615 while (cse_find_path (bb, &ebb_data, flag_cse_follow_jumps))
6617 /* Pre-scan the path. */
6618 cse_prescan_path (&ebb_data);
6620 /* If this basic block has no sets, skip it. */
6621 if (ebb_data.nsets == 0)
6622 continue;
6624 /* Get a reasonable estimate for the maximum number of qty's
6625 needed for this path. For this, we take the number of sets
6626 and multiply that by MAX_RECOG_OPERANDS. */
6627 max_qty = ebb_data.nsets * MAX_RECOG_OPERANDS;
6629 /* Dump the path we're about to process. */
6630 if (dump_file)
6631 cse_dump_path (&ebb_data, ebb_data.nsets, dump_file);
6633 cse_extended_basic_block (&ebb_data);
6637 /* Clean up. */
6638 end_alias_analysis ();
6639 free (reg_eqv_table);
6640 free (ebb_data.path);
6641 sbitmap_free (cse_visited_basic_blocks);
6642 free (rc_order);
6643 rtl_hooks = general_rtl_hooks;
6645 if (cse_jumps_altered || recorded_label_ref)
6646 return 2;
6647 else if (cse_cfg_altered)
6648 return 1;
6649 else
6650 return 0;
6653 /* Called via for_each_rtx to see if an insn is using a LABEL_REF for
6654 which there isn't a REG_LABEL_OPERAND note.
6655 Return one if so. DATA is the insn. */
6657 static int
6658 check_for_label_ref (rtx *rtl, void *data)
6660 rtx insn = (rtx) data;
6662 /* If this insn uses a LABEL_REF and there isn't a REG_LABEL_OPERAND
6663 note for it, we must rerun jump since it needs to place the note. If
6664 this is a LABEL_REF for a CODE_LABEL that isn't in the insn chain,
6665 don't do this since no REG_LABEL_OPERAND will be added. */
6666 return (GET_CODE (*rtl) == LABEL_REF
6667 && ! LABEL_REF_NONLOCAL_P (*rtl)
6668 && (!JUMP_P (insn)
6669 || !label_is_jump_target_p (XEXP (*rtl, 0), insn))
6670 && LABEL_P (XEXP (*rtl, 0))
6671 && INSN_UID (XEXP (*rtl, 0)) != 0
6672 && ! find_reg_note (insn, REG_LABEL_OPERAND, XEXP (*rtl, 0)));
6675 /* Count the number of times registers are used (not set) in X.
6676 COUNTS is an array in which we accumulate the count, INCR is how much
6677 we count each register usage.
6679 Don't count a usage of DEST, which is the SET_DEST of a SET which
6680 contains X in its SET_SRC. This is because such a SET does not
6681 modify the liveness of DEST.
6682 DEST is set to pc_rtx for a trapping insn, or for an insn with side effects.
6683 We must then count uses of a SET_DEST regardless, because the insn can't be
6684 deleted here. */
6686 static void
6687 count_reg_usage (rtx x, int *counts, rtx dest, int incr)
6689 enum rtx_code code;
6690 rtx note;
6691 const char *fmt;
6692 int i, j;
6694 if (x == 0)
6695 return;
6697 switch (code = GET_CODE (x))
6699 case REG:
6700 if (x != dest)
6701 counts[REGNO (x)] += incr;
6702 return;
6704 case PC:
6705 case CC0:
6706 case CONST:
6707 CASE_CONST_ANY:
6708 case SYMBOL_REF:
6709 case LABEL_REF:
6710 return;
6712 case CLOBBER:
6713 /* If we are clobbering a MEM, mark any registers inside the address
6714 as being used. */
6715 if (MEM_P (XEXP (x, 0)))
6716 count_reg_usage (XEXP (XEXP (x, 0), 0), counts, NULL_RTX, incr);
6717 return;
6719 case SET:
6720 /* Unless we are setting a REG, count everything in SET_DEST. */
6721 if (!REG_P (SET_DEST (x)))
6722 count_reg_usage (SET_DEST (x), counts, NULL_RTX, incr);
6723 count_reg_usage (SET_SRC (x), counts,
6724 dest ? dest : SET_DEST (x),
6725 incr);
6726 return;
6728 case DEBUG_INSN:
6729 return;
6731 case CALL_INSN:
6732 case INSN:
6733 case JUMP_INSN:
6734 /* We expect dest to be NULL_RTX here. If the insn may throw,
6735 or if it cannot be deleted due to side-effects, mark this fact
6736 by setting DEST to pc_rtx. */
6737 if ((!cfun->can_delete_dead_exceptions && !insn_nothrow_p (x))
6738 || side_effects_p (PATTERN (x)))
6739 dest = pc_rtx;
6740 if (code == CALL_INSN)
6741 count_reg_usage (CALL_INSN_FUNCTION_USAGE (x), counts, dest, incr);
6742 count_reg_usage (PATTERN (x), counts, dest, incr);
6744 /* Things used in a REG_EQUAL note aren't dead since loop may try to
6745 use them. */
6747 note = find_reg_equal_equiv_note (x);
6748 if (note)
6750 rtx eqv = XEXP (note, 0);
6752 if (GET_CODE (eqv) == EXPR_LIST)
6753 /* This REG_EQUAL note describes the result of a function call.
6754 Process all the arguments. */
6757 count_reg_usage (XEXP (eqv, 0), counts, dest, incr);
6758 eqv = XEXP (eqv, 1);
6760 while (eqv && GET_CODE (eqv) == EXPR_LIST);
6761 else
6762 count_reg_usage (eqv, counts, dest, incr);
6764 return;
6766 case EXPR_LIST:
6767 if (REG_NOTE_KIND (x) == REG_EQUAL
6768 || (REG_NOTE_KIND (x) != REG_NONNEG && GET_CODE (XEXP (x,0)) == USE)
6769 /* FUNCTION_USAGE expression lists may include (CLOBBER (mem /u)),
6770 involving registers in the address. */
6771 || GET_CODE (XEXP (x, 0)) == CLOBBER)
6772 count_reg_usage (XEXP (x, 0), counts, NULL_RTX, incr);
6774 count_reg_usage (XEXP (x, 1), counts, NULL_RTX, incr);
6775 return;
6777 case ASM_OPERANDS:
6778 /* Iterate over just the inputs, not the constraints as well. */
6779 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
6780 count_reg_usage (ASM_OPERANDS_INPUT (x, i), counts, dest, incr);
6781 return;
6783 case INSN_LIST:
6784 case INT_LIST:
6785 gcc_unreachable ();
6787 default:
6788 break;
6791 fmt = GET_RTX_FORMAT (code);
6792 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
6794 if (fmt[i] == 'e')
6795 count_reg_usage (XEXP (x, i), counts, dest, incr);
6796 else if (fmt[i] == 'E')
6797 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
6798 count_reg_usage (XVECEXP (x, i, j), counts, dest, incr);
6802 /* Return true if X is a dead register. */
6804 static inline int
6805 is_dead_reg (rtx x, int *counts)
6807 return (REG_P (x)
6808 && REGNO (x) >= FIRST_PSEUDO_REGISTER
6809 && counts[REGNO (x)] == 0);
6812 /* Return true if set is live. */
6813 static bool
6814 set_live_p (rtx set, rtx insn ATTRIBUTE_UNUSED, /* Only used with HAVE_cc0. */
6815 int *counts)
6817 #ifdef HAVE_cc0
6818 rtx tem;
6819 #endif
6821 if (set_noop_p (set))
6824 #ifdef HAVE_cc0
6825 else if (GET_CODE (SET_DEST (set)) == CC0
6826 && !side_effects_p (SET_SRC (set))
6827 && ((tem = next_nonnote_nondebug_insn (insn)) == NULL_RTX
6828 || !INSN_P (tem)
6829 || !reg_referenced_p (cc0_rtx, PATTERN (tem))))
6830 return false;
6831 #endif
6832 else if (!is_dead_reg (SET_DEST (set), counts)
6833 || side_effects_p (SET_SRC (set)))
6834 return true;
6835 return false;
6838 /* Return true if insn is live. */
6840 static bool
6841 insn_live_p (rtx insn, int *counts)
6843 int i;
6844 if (!cfun->can_delete_dead_exceptions && !insn_nothrow_p (insn))
6845 return true;
6846 else if (GET_CODE (PATTERN (insn)) == SET)
6847 return set_live_p (PATTERN (insn), insn, counts);
6848 else if (GET_CODE (PATTERN (insn)) == PARALLEL)
6850 for (i = XVECLEN (PATTERN (insn), 0) - 1; i >= 0; i--)
6852 rtx elt = XVECEXP (PATTERN (insn), 0, i);
6854 if (GET_CODE (elt) == SET)
6856 if (set_live_p (elt, insn, counts))
6857 return true;
6859 else if (GET_CODE (elt) != CLOBBER && GET_CODE (elt) != USE)
6860 return true;
6862 return false;
6864 else if (DEBUG_INSN_P (insn))
6866 rtx next;
6868 for (next = NEXT_INSN (insn); next; next = NEXT_INSN (next))
6869 if (NOTE_P (next))
6870 continue;
6871 else if (!DEBUG_INSN_P (next))
6872 return true;
6873 else if (INSN_VAR_LOCATION_DECL (insn) == INSN_VAR_LOCATION_DECL (next))
6874 return false;
6876 return true;
6878 else
6879 return true;
6882 /* Count the number of stores into pseudo. Callback for note_stores. */
6884 static void
6885 count_stores (rtx x, const_rtx set ATTRIBUTE_UNUSED, void *data)
6887 int *counts = (int *) data;
6888 if (REG_P (x) && REGNO (x) >= FIRST_PSEUDO_REGISTER)
6889 counts[REGNO (x)]++;
6892 struct dead_debug_insn_data
6894 int *counts;
6895 rtx *replacements;
6896 bool seen_repl;
6899 /* Return if a DEBUG_INSN needs to be reset because some dead
6900 pseudo doesn't have a replacement. Callback for for_each_rtx. */
6902 static int
6903 is_dead_debug_insn (rtx *loc, void *data)
6905 rtx x = *loc;
6906 struct dead_debug_insn_data *ddid = (struct dead_debug_insn_data *) data;
6908 if (is_dead_reg (x, ddid->counts))
6910 if (ddid->replacements && ddid->replacements[REGNO (x)] != NULL_RTX)
6911 ddid->seen_repl = true;
6912 else
6913 return 1;
6915 return 0;
6918 /* Replace a dead pseudo in a DEBUG_INSN with replacement DEBUG_EXPR.
6919 Callback for simplify_replace_fn_rtx. */
6921 static rtx
6922 replace_dead_reg (rtx x, const_rtx old_rtx ATTRIBUTE_UNUSED, void *data)
6924 rtx *replacements = (rtx *) data;
6926 if (REG_P (x)
6927 && REGNO (x) >= FIRST_PSEUDO_REGISTER
6928 && replacements[REGNO (x)] != NULL_RTX)
6930 if (GET_MODE (x) == GET_MODE (replacements[REGNO (x)]))
6931 return replacements[REGNO (x)];
6932 return lowpart_subreg (GET_MODE (x), replacements[REGNO (x)],
6933 GET_MODE (replacements[REGNO (x)]));
6935 return NULL_RTX;
6938 /* Scan all the insns and delete any that are dead; i.e., they store a register
6939 that is never used or they copy a register to itself.
6941 This is used to remove insns made obviously dead by cse, loop or other
6942 optimizations. It improves the heuristics in loop since it won't try to
6943 move dead invariants out of loops or make givs for dead quantities. The
6944 remaining passes of the compilation are also sped up. */
6947 delete_trivially_dead_insns (rtx insns, int nreg)
6949 int *counts;
6950 rtx insn, prev;
6951 rtx *replacements = NULL;
6952 int ndead = 0;
6954 timevar_push (TV_DELETE_TRIVIALLY_DEAD);
6955 /* First count the number of times each register is used. */
6956 if (MAY_HAVE_DEBUG_INSNS)
6958 counts = XCNEWVEC (int, nreg * 3);
6959 for (insn = insns; insn; insn = NEXT_INSN (insn))
6960 if (DEBUG_INSN_P (insn))
6961 count_reg_usage (INSN_VAR_LOCATION_LOC (insn), counts + nreg,
6962 NULL_RTX, 1);
6963 else if (INSN_P (insn))
6965 count_reg_usage (insn, counts, NULL_RTX, 1);
6966 note_stores (PATTERN (insn), count_stores, counts + nreg * 2);
6968 /* If there can be debug insns, COUNTS are 3 consecutive arrays.
6969 First one counts how many times each pseudo is used outside
6970 of debug insns, second counts how many times each pseudo is
6971 used in debug insns and third counts how many times a pseudo
6972 is stored. */
6974 else
6976 counts = XCNEWVEC (int, nreg);
6977 for (insn = insns; insn; insn = NEXT_INSN (insn))
6978 if (INSN_P (insn))
6979 count_reg_usage (insn, counts, NULL_RTX, 1);
6980 /* If no debug insns can be present, COUNTS is just an array
6981 which counts how many times each pseudo is used. */
6983 /* Go from the last insn to the first and delete insns that only set unused
6984 registers or copy a register to itself. As we delete an insn, remove
6985 usage counts for registers it uses.
6987 The first jump optimization pass may leave a real insn as the last
6988 insn in the function. We must not skip that insn or we may end
6989 up deleting code that is not really dead.
6991 If some otherwise unused register is only used in DEBUG_INSNs,
6992 try to create a DEBUG_EXPR temporary and emit a DEBUG_INSN before
6993 the setter. Then go through DEBUG_INSNs and if a DEBUG_EXPR
6994 has been created for the unused register, replace it with
6995 the DEBUG_EXPR, otherwise reset the DEBUG_INSN. */
6996 for (insn = get_last_insn (); insn; insn = prev)
6998 int live_insn = 0;
7000 prev = PREV_INSN (insn);
7001 if (!INSN_P (insn))
7002 continue;
7004 live_insn = insn_live_p (insn, counts);
7006 /* If this is a dead insn, delete it and show registers in it aren't
7007 being used. */
7009 if (! live_insn && dbg_cnt (delete_trivial_dead))
7011 if (DEBUG_INSN_P (insn))
7012 count_reg_usage (INSN_VAR_LOCATION_LOC (insn), counts + nreg,
7013 NULL_RTX, -1);
7014 else
7016 rtx set;
7017 if (MAY_HAVE_DEBUG_INSNS
7018 && (set = single_set (insn)) != NULL_RTX
7019 && is_dead_reg (SET_DEST (set), counts)
7020 /* Used at least once in some DEBUG_INSN. */
7021 && counts[REGNO (SET_DEST (set)) + nreg] > 0
7022 /* And set exactly once. */
7023 && counts[REGNO (SET_DEST (set)) + nreg * 2] == 1
7024 && !side_effects_p (SET_SRC (set))
7025 && asm_noperands (PATTERN (insn)) < 0)
7027 rtx dval, bind;
7029 /* Create DEBUG_EXPR (and DEBUG_EXPR_DECL). */
7030 dval = make_debug_expr_from_rtl (SET_DEST (set));
7032 /* Emit a debug bind insn before the insn in which
7033 reg dies. */
7034 bind = gen_rtx_VAR_LOCATION (GET_MODE (SET_DEST (set)),
7035 DEBUG_EXPR_TREE_DECL (dval),
7036 SET_SRC (set),
7037 VAR_INIT_STATUS_INITIALIZED);
7038 count_reg_usage (bind, counts + nreg, NULL_RTX, 1);
7040 bind = emit_debug_insn_before (bind, insn);
7041 df_insn_rescan (bind);
7043 if (replacements == NULL)
7044 replacements = XCNEWVEC (rtx, nreg);
7045 replacements[REGNO (SET_DEST (set))] = dval;
7048 count_reg_usage (insn, counts, NULL_RTX, -1);
7049 ndead++;
7051 delete_insn_and_edges (insn);
7055 if (MAY_HAVE_DEBUG_INSNS)
7057 struct dead_debug_insn_data ddid;
7058 ddid.counts = counts;
7059 ddid.replacements = replacements;
7060 for (insn = get_last_insn (); insn; insn = PREV_INSN (insn))
7061 if (DEBUG_INSN_P (insn))
7063 /* If this debug insn references a dead register that wasn't replaced
7064 with an DEBUG_EXPR, reset the DEBUG_INSN. */
7065 ddid.seen_repl = false;
7066 if (for_each_rtx (&INSN_VAR_LOCATION_LOC (insn),
7067 is_dead_debug_insn, &ddid))
7069 INSN_VAR_LOCATION_LOC (insn) = gen_rtx_UNKNOWN_VAR_LOC ();
7070 df_insn_rescan (insn);
7072 else if (ddid.seen_repl)
7074 INSN_VAR_LOCATION_LOC (insn)
7075 = simplify_replace_fn_rtx (INSN_VAR_LOCATION_LOC (insn),
7076 NULL_RTX, replace_dead_reg,
7077 replacements);
7078 df_insn_rescan (insn);
7081 free (replacements);
7084 if (dump_file && ndead)
7085 fprintf (dump_file, "Deleted %i trivially dead insns\n",
7086 ndead);
7087 /* Clean up. */
7088 free (counts);
7089 timevar_pop (TV_DELETE_TRIVIALLY_DEAD);
7090 return ndead;
7093 /* This function is called via for_each_rtx. The argument, NEWREG, is
7094 a condition code register with the desired mode. If we are looking
7095 at the same register in a different mode, replace it with
7096 NEWREG. */
7098 static int
7099 cse_change_cc_mode (rtx *loc, void *data)
7101 struct change_cc_mode_args* args = (struct change_cc_mode_args*)data;
7103 if (*loc
7104 && REG_P (*loc)
7105 && REGNO (*loc) == REGNO (args->newreg)
7106 && GET_MODE (*loc) != GET_MODE (args->newreg))
7108 validate_change (args->insn, loc, args->newreg, 1);
7110 return -1;
7112 return 0;
7115 /* Change the mode of any reference to the register REGNO (NEWREG) to
7116 GET_MODE (NEWREG) in INSN. */
7118 static void
7119 cse_change_cc_mode_insn (rtx insn, rtx newreg)
7121 struct change_cc_mode_args args;
7122 int success;
7124 if (!INSN_P (insn))
7125 return;
7127 args.insn = insn;
7128 args.newreg = newreg;
7130 for_each_rtx (&PATTERN (insn), cse_change_cc_mode, &args);
7131 for_each_rtx (&REG_NOTES (insn), cse_change_cc_mode, &args);
7133 /* If the following assertion was triggered, there is most probably
7134 something wrong with the cc_modes_compatible back end function.
7135 CC modes only can be considered compatible if the insn - with the mode
7136 replaced by any of the compatible modes - can still be recognized. */
7137 success = apply_change_group ();
7138 gcc_assert (success);
7141 /* Change the mode of any reference to the register REGNO (NEWREG) to
7142 GET_MODE (NEWREG), starting at START. Stop before END. Stop at
7143 any instruction which modifies NEWREG. */
7145 static void
7146 cse_change_cc_mode_insns (rtx start, rtx end, rtx newreg)
7148 rtx insn;
7150 for (insn = start; insn != end; insn = NEXT_INSN (insn))
7152 if (! INSN_P (insn))
7153 continue;
7155 if (reg_set_p (newreg, insn))
7156 return;
7158 cse_change_cc_mode_insn (insn, newreg);
7162 /* BB is a basic block which finishes with CC_REG as a condition code
7163 register which is set to CC_SRC. Look through the successors of BB
7164 to find blocks which have a single predecessor (i.e., this one),
7165 and look through those blocks for an assignment to CC_REG which is
7166 equivalent to CC_SRC. CAN_CHANGE_MODE indicates whether we are
7167 permitted to change the mode of CC_SRC to a compatible mode. This
7168 returns VOIDmode if no equivalent assignments were found.
7169 Otherwise it returns the mode which CC_SRC should wind up with.
7170 ORIG_BB should be the same as BB in the outermost cse_cc_succs call,
7171 but is passed unmodified down to recursive calls in order to prevent
7172 endless recursion.
7174 The main complexity in this function is handling the mode issues.
7175 We may have more than one duplicate which we can eliminate, and we
7176 try to find a mode which will work for multiple duplicates. */
7178 static enum machine_mode
7179 cse_cc_succs (basic_block bb, basic_block orig_bb, rtx cc_reg, rtx cc_src,
7180 bool can_change_mode)
7182 bool found_equiv;
7183 enum machine_mode mode;
7184 unsigned int insn_count;
7185 edge e;
7186 rtx insns[2];
7187 enum machine_mode modes[2];
7188 rtx last_insns[2];
7189 unsigned int i;
7190 rtx newreg;
7191 edge_iterator ei;
7193 /* We expect to have two successors. Look at both before picking
7194 the final mode for the comparison. If we have more successors
7195 (i.e., some sort of table jump, although that seems unlikely),
7196 then we require all beyond the first two to use the same
7197 mode. */
7199 found_equiv = false;
7200 mode = GET_MODE (cc_src);
7201 insn_count = 0;
7202 FOR_EACH_EDGE (e, ei, bb->succs)
7204 rtx insn;
7205 rtx end;
7207 if (e->flags & EDGE_COMPLEX)
7208 continue;
7210 if (EDGE_COUNT (e->dest->preds) != 1
7211 || e->dest == EXIT_BLOCK_PTR_FOR_FN (cfun)
7212 /* Avoid endless recursion on unreachable blocks. */
7213 || e->dest == orig_bb)
7214 continue;
7216 end = NEXT_INSN (BB_END (e->dest));
7217 for (insn = BB_HEAD (e->dest); insn != end; insn = NEXT_INSN (insn))
7219 rtx set;
7221 if (! INSN_P (insn))
7222 continue;
7224 /* If CC_SRC is modified, we have to stop looking for
7225 something which uses it. */
7226 if (modified_in_p (cc_src, insn))
7227 break;
7229 /* Check whether INSN sets CC_REG to CC_SRC. */
7230 set = single_set (insn);
7231 if (set
7232 && REG_P (SET_DEST (set))
7233 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7235 bool found;
7236 enum machine_mode set_mode;
7237 enum machine_mode comp_mode;
7239 found = false;
7240 set_mode = GET_MODE (SET_SRC (set));
7241 comp_mode = set_mode;
7242 if (rtx_equal_p (cc_src, SET_SRC (set)))
7243 found = true;
7244 else if (GET_CODE (cc_src) == COMPARE
7245 && GET_CODE (SET_SRC (set)) == COMPARE
7246 && mode != set_mode
7247 && rtx_equal_p (XEXP (cc_src, 0),
7248 XEXP (SET_SRC (set), 0))
7249 && rtx_equal_p (XEXP (cc_src, 1),
7250 XEXP (SET_SRC (set), 1)))
7253 comp_mode = targetm.cc_modes_compatible (mode, set_mode);
7254 if (comp_mode != VOIDmode
7255 && (can_change_mode || comp_mode == mode))
7256 found = true;
7259 if (found)
7261 found_equiv = true;
7262 if (insn_count < ARRAY_SIZE (insns))
7264 insns[insn_count] = insn;
7265 modes[insn_count] = set_mode;
7266 last_insns[insn_count] = end;
7267 ++insn_count;
7269 if (mode != comp_mode)
7271 gcc_assert (can_change_mode);
7272 mode = comp_mode;
7274 /* The modified insn will be re-recognized later. */
7275 PUT_MODE (cc_src, mode);
7278 else
7280 if (set_mode != mode)
7282 /* We found a matching expression in the
7283 wrong mode, but we don't have room to
7284 store it in the array. Punt. This case
7285 should be rare. */
7286 break;
7288 /* INSN sets CC_REG to a value equal to CC_SRC
7289 with the right mode. We can simply delete
7290 it. */
7291 delete_insn (insn);
7294 /* We found an instruction to delete. Keep looking,
7295 in the hopes of finding a three-way jump. */
7296 continue;
7299 /* We found an instruction which sets the condition
7300 code, so don't look any farther. */
7301 break;
7304 /* If INSN sets CC_REG in some other way, don't look any
7305 farther. */
7306 if (reg_set_p (cc_reg, insn))
7307 break;
7310 /* If we fell off the bottom of the block, we can keep looking
7311 through successors. We pass CAN_CHANGE_MODE as false because
7312 we aren't prepared to handle compatibility between the
7313 further blocks and this block. */
7314 if (insn == end)
7316 enum machine_mode submode;
7318 submode = cse_cc_succs (e->dest, orig_bb, cc_reg, cc_src, false);
7319 if (submode != VOIDmode)
7321 gcc_assert (submode == mode);
7322 found_equiv = true;
7323 can_change_mode = false;
7328 if (! found_equiv)
7329 return VOIDmode;
7331 /* Now INSN_COUNT is the number of instructions we found which set
7332 CC_REG to a value equivalent to CC_SRC. The instructions are in
7333 INSNS. The modes used by those instructions are in MODES. */
7335 newreg = NULL_RTX;
7336 for (i = 0; i < insn_count; ++i)
7338 if (modes[i] != mode)
7340 /* We need to change the mode of CC_REG in INSNS[i] and
7341 subsequent instructions. */
7342 if (! newreg)
7344 if (GET_MODE (cc_reg) == mode)
7345 newreg = cc_reg;
7346 else
7347 newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7349 cse_change_cc_mode_insns (NEXT_INSN (insns[i]), last_insns[i],
7350 newreg);
7353 delete_insn_and_edges (insns[i]);
7356 return mode;
7359 /* If we have a fixed condition code register (or two), walk through
7360 the instructions and try to eliminate duplicate assignments. */
7362 static void
7363 cse_condition_code_reg (void)
7365 unsigned int cc_regno_1;
7366 unsigned int cc_regno_2;
7367 rtx cc_reg_1;
7368 rtx cc_reg_2;
7369 basic_block bb;
7371 if (! targetm.fixed_condition_code_regs (&cc_regno_1, &cc_regno_2))
7372 return;
7374 cc_reg_1 = gen_rtx_REG (CCmode, cc_regno_1);
7375 if (cc_regno_2 != INVALID_REGNUM)
7376 cc_reg_2 = gen_rtx_REG (CCmode, cc_regno_2);
7377 else
7378 cc_reg_2 = NULL_RTX;
7380 FOR_EACH_BB_FN (bb, cfun)
7382 rtx last_insn;
7383 rtx cc_reg;
7384 rtx insn;
7385 rtx cc_src_insn;
7386 rtx cc_src;
7387 enum machine_mode mode;
7388 enum machine_mode orig_mode;
7390 /* Look for blocks which end with a conditional jump based on a
7391 condition code register. Then look for the instruction which
7392 sets the condition code register. Then look through the
7393 successor blocks for instructions which set the condition
7394 code register to the same value. There are other possible
7395 uses of the condition code register, but these are by far the
7396 most common and the ones which we are most likely to be able
7397 to optimize. */
7399 last_insn = BB_END (bb);
7400 if (!JUMP_P (last_insn))
7401 continue;
7403 if (reg_referenced_p (cc_reg_1, PATTERN (last_insn)))
7404 cc_reg = cc_reg_1;
7405 else if (cc_reg_2 && reg_referenced_p (cc_reg_2, PATTERN (last_insn)))
7406 cc_reg = cc_reg_2;
7407 else
7408 continue;
7410 cc_src_insn = NULL_RTX;
7411 cc_src = NULL_RTX;
7412 for (insn = PREV_INSN (last_insn);
7413 insn && insn != PREV_INSN (BB_HEAD (bb));
7414 insn = PREV_INSN (insn))
7416 rtx set;
7418 if (! INSN_P (insn))
7419 continue;
7420 set = single_set (insn);
7421 if (set
7422 && REG_P (SET_DEST (set))
7423 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7425 cc_src_insn = insn;
7426 cc_src = SET_SRC (set);
7427 break;
7429 else if (reg_set_p (cc_reg, insn))
7430 break;
7433 if (! cc_src_insn)
7434 continue;
7436 if (modified_between_p (cc_src, cc_src_insn, NEXT_INSN (last_insn)))
7437 continue;
7439 /* Now CC_REG is a condition code register used for a
7440 conditional jump at the end of the block, and CC_SRC, in
7441 CC_SRC_INSN, is the value to which that condition code
7442 register is set, and CC_SRC is still meaningful at the end of
7443 the basic block. */
7445 orig_mode = GET_MODE (cc_src);
7446 mode = cse_cc_succs (bb, bb, cc_reg, cc_src, true);
7447 if (mode != VOIDmode)
7449 gcc_assert (mode == GET_MODE (cc_src));
7450 if (mode != orig_mode)
7452 rtx newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7454 cse_change_cc_mode_insn (cc_src_insn, newreg);
7456 /* Do the same in the following insns that use the
7457 current value of CC_REG within BB. */
7458 cse_change_cc_mode_insns (NEXT_INSN (cc_src_insn),
7459 NEXT_INSN (last_insn),
7460 newreg);
7467 /* Perform common subexpression elimination. Nonzero value from
7468 `cse_main' means that jumps were simplified and some code may now
7469 be unreachable, so do jump optimization again. */
7470 static unsigned int
7471 rest_of_handle_cse (void)
7473 int tem;
7475 if (dump_file)
7476 dump_flow_info (dump_file, dump_flags);
7478 tem = cse_main (get_insns (), max_reg_num ());
7480 /* If we are not running more CSE passes, then we are no longer
7481 expecting CSE to be run. But always rerun it in a cheap mode. */
7482 cse_not_expected = !flag_rerun_cse_after_loop && !flag_gcse;
7484 if (tem == 2)
7486 timevar_push (TV_JUMP);
7487 rebuild_jump_labels (get_insns ());
7488 cleanup_cfg (CLEANUP_CFG_CHANGED);
7489 timevar_pop (TV_JUMP);
7491 else if (tem == 1 || optimize > 1)
7492 cleanup_cfg (0);
7494 return 0;
7497 namespace {
7499 const pass_data pass_data_cse =
7501 RTL_PASS, /* type */
7502 "cse1", /* name */
7503 OPTGROUP_NONE, /* optinfo_flags */
7504 TV_CSE, /* tv_id */
7505 0, /* properties_required */
7506 0, /* properties_provided */
7507 0, /* properties_destroyed */
7508 0, /* todo_flags_start */
7509 TODO_df_finish, /* todo_flags_finish */
7512 class pass_cse : public rtl_opt_pass
7514 public:
7515 pass_cse (gcc::context *ctxt)
7516 : rtl_opt_pass (pass_data_cse, ctxt)
7519 /* opt_pass methods: */
7520 virtual bool gate (function *) { return optimize > 0; }
7521 virtual unsigned int execute (function *) { return rest_of_handle_cse (); }
7523 }; // class pass_cse
7525 } // anon namespace
7527 rtl_opt_pass *
7528 make_pass_cse (gcc::context *ctxt)
7530 return new pass_cse (ctxt);
7534 /* Run second CSE pass after loop optimizations. */
7535 static unsigned int
7536 rest_of_handle_cse2 (void)
7538 int tem;
7540 if (dump_file)
7541 dump_flow_info (dump_file, dump_flags);
7543 tem = cse_main (get_insns (), max_reg_num ());
7545 /* Run a pass to eliminate duplicated assignments to condition code
7546 registers. We have to run this after bypass_jumps, because it
7547 makes it harder for that pass to determine whether a jump can be
7548 bypassed safely. */
7549 cse_condition_code_reg ();
7551 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7553 if (tem == 2)
7555 timevar_push (TV_JUMP);
7556 rebuild_jump_labels (get_insns ());
7557 cleanup_cfg (CLEANUP_CFG_CHANGED);
7558 timevar_pop (TV_JUMP);
7560 else if (tem == 1)
7561 cleanup_cfg (0);
7563 cse_not_expected = 1;
7564 return 0;
7568 namespace {
7570 const pass_data pass_data_cse2 =
7572 RTL_PASS, /* type */
7573 "cse2", /* name */
7574 OPTGROUP_NONE, /* optinfo_flags */
7575 TV_CSE2, /* tv_id */
7576 0, /* properties_required */
7577 0, /* properties_provided */
7578 0, /* properties_destroyed */
7579 0, /* todo_flags_start */
7580 TODO_df_finish, /* todo_flags_finish */
7583 class pass_cse2 : public rtl_opt_pass
7585 public:
7586 pass_cse2 (gcc::context *ctxt)
7587 : rtl_opt_pass (pass_data_cse2, ctxt)
7590 /* opt_pass methods: */
7591 virtual bool gate (function *)
7593 return optimize > 0 && flag_rerun_cse_after_loop;
7596 virtual unsigned int execute (function *) { return rest_of_handle_cse2 (); }
7598 }; // class pass_cse2
7600 } // anon namespace
7602 rtl_opt_pass *
7603 make_pass_cse2 (gcc::context *ctxt)
7605 return new pass_cse2 (ctxt);
7608 /* Run second CSE pass after loop optimizations. */
7609 static unsigned int
7610 rest_of_handle_cse_after_global_opts (void)
7612 int save_cfj;
7613 int tem;
7615 /* We only want to do local CSE, so don't follow jumps. */
7616 save_cfj = flag_cse_follow_jumps;
7617 flag_cse_follow_jumps = 0;
7619 rebuild_jump_labels (get_insns ());
7620 tem = cse_main (get_insns (), max_reg_num ());
7621 purge_all_dead_edges ();
7622 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7624 cse_not_expected = !flag_rerun_cse_after_loop;
7626 /* If cse altered any jumps, rerun jump opts to clean things up. */
7627 if (tem == 2)
7629 timevar_push (TV_JUMP);
7630 rebuild_jump_labels (get_insns ());
7631 cleanup_cfg (CLEANUP_CFG_CHANGED);
7632 timevar_pop (TV_JUMP);
7634 else if (tem == 1)
7635 cleanup_cfg (0);
7637 flag_cse_follow_jumps = save_cfj;
7638 return 0;
7641 namespace {
7643 const pass_data pass_data_cse_after_global_opts =
7645 RTL_PASS, /* type */
7646 "cse_local", /* name */
7647 OPTGROUP_NONE, /* optinfo_flags */
7648 TV_CSE, /* tv_id */
7649 0, /* properties_required */
7650 0, /* properties_provided */
7651 0, /* properties_destroyed */
7652 0, /* todo_flags_start */
7653 TODO_df_finish, /* todo_flags_finish */
7656 class pass_cse_after_global_opts : public rtl_opt_pass
7658 public:
7659 pass_cse_after_global_opts (gcc::context *ctxt)
7660 : rtl_opt_pass (pass_data_cse_after_global_opts, ctxt)
7663 /* opt_pass methods: */
7664 virtual bool gate (function *)
7666 return optimize > 0 && flag_rerun_cse_after_global_opts;
7669 virtual unsigned int execute (function *)
7671 return rest_of_handle_cse_after_global_opts ();
7674 }; // class pass_cse_after_global_opts
7676 } // anon namespace
7678 rtl_opt_pass *
7679 make_pass_cse_after_global_opts (gcc::context *ctxt)
7681 return new pass_cse_after_global_opts (ctxt);