gcc/testsuite/
[official-gcc.git] / gcc / cse.c
blob3ca8e179757555ae25defcd35c8464eb87beee07
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 "pointer-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 (x) != MEM_ATTRS (y))
2689 return 0;
2691 break;
2693 /* For commutative operations, check both orders. */
2694 case PLUS:
2695 case MULT:
2696 case AND:
2697 case IOR:
2698 case XOR:
2699 case NE:
2700 case EQ:
2701 return ((exp_equiv_p (XEXP (x, 0), XEXP (y, 0),
2702 validate, for_gcse)
2703 && exp_equiv_p (XEXP (x, 1), XEXP (y, 1),
2704 validate, for_gcse))
2705 || (exp_equiv_p (XEXP (x, 0), XEXP (y, 1),
2706 validate, for_gcse)
2707 && exp_equiv_p (XEXP (x, 1), XEXP (y, 0),
2708 validate, for_gcse)));
2710 case ASM_OPERANDS:
2711 /* We don't use the generic code below because we want to
2712 disregard filename and line numbers. */
2714 /* A volatile asm isn't equivalent to any other. */
2715 if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
2716 return 0;
2718 if (GET_MODE (x) != GET_MODE (y)
2719 || strcmp (ASM_OPERANDS_TEMPLATE (x), ASM_OPERANDS_TEMPLATE (y))
2720 || strcmp (ASM_OPERANDS_OUTPUT_CONSTRAINT (x),
2721 ASM_OPERANDS_OUTPUT_CONSTRAINT (y))
2722 || ASM_OPERANDS_OUTPUT_IDX (x) != ASM_OPERANDS_OUTPUT_IDX (y)
2723 || ASM_OPERANDS_INPUT_LENGTH (x) != ASM_OPERANDS_INPUT_LENGTH (y))
2724 return 0;
2726 if (ASM_OPERANDS_INPUT_LENGTH (x))
2728 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
2729 if (! exp_equiv_p (ASM_OPERANDS_INPUT (x, i),
2730 ASM_OPERANDS_INPUT (y, i),
2731 validate, for_gcse)
2732 || strcmp (ASM_OPERANDS_INPUT_CONSTRAINT (x, i),
2733 ASM_OPERANDS_INPUT_CONSTRAINT (y, i)))
2734 return 0;
2737 return 1;
2739 default:
2740 break;
2743 /* Compare the elements. If any pair of corresponding elements
2744 fail to match, return 0 for the whole thing. */
2746 fmt = GET_RTX_FORMAT (code);
2747 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2749 switch (fmt[i])
2751 case 'e':
2752 if (! exp_equiv_p (XEXP (x, i), XEXP (y, i),
2753 validate, for_gcse))
2754 return 0;
2755 break;
2757 case 'E':
2758 if (XVECLEN (x, i) != XVECLEN (y, i))
2759 return 0;
2760 for (j = 0; j < XVECLEN (x, i); j++)
2761 if (! exp_equiv_p (XVECEXP (x, i, j), XVECEXP (y, i, j),
2762 validate, for_gcse))
2763 return 0;
2764 break;
2766 case 's':
2767 if (strcmp (XSTR (x, i), XSTR (y, i)))
2768 return 0;
2769 break;
2771 case 'i':
2772 if (XINT (x, i) != XINT (y, i))
2773 return 0;
2774 break;
2776 case 'w':
2777 if (XWINT (x, i) != XWINT (y, i))
2778 return 0;
2779 break;
2781 case '0':
2782 case 't':
2783 break;
2785 default:
2786 gcc_unreachable ();
2790 return 1;
2793 /* Subroutine of canon_reg. Pass *XLOC through canon_reg, and validate
2794 the result if necessary. INSN is as for canon_reg. */
2796 static void
2797 validate_canon_reg (rtx *xloc, rtx insn)
2799 if (*xloc)
2801 rtx new_rtx = canon_reg (*xloc, insn);
2803 /* If replacing pseudo with hard reg or vice versa, ensure the
2804 insn remains valid. Likewise if the insn has MATCH_DUPs. */
2805 gcc_assert (insn && new_rtx);
2806 validate_change (insn, xloc, new_rtx, 1);
2810 /* Canonicalize an expression:
2811 replace each register reference inside it
2812 with the "oldest" equivalent register.
2814 If INSN is nonzero validate_change is used to ensure that INSN remains valid
2815 after we make our substitution. The calls are made with IN_GROUP nonzero
2816 so apply_change_group must be called upon the outermost return from this
2817 function (unless INSN is zero). The result of apply_change_group can
2818 generally be discarded since the changes we are making are optional. */
2820 static rtx
2821 canon_reg (rtx x, rtx insn)
2823 int i;
2824 enum rtx_code code;
2825 const char *fmt;
2827 if (x == 0)
2828 return x;
2830 code = GET_CODE (x);
2831 switch (code)
2833 case PC:
2834 case CC0:
2835 case CONST:
2836 CASE_CONST_ANY:
2837 case SYMBOL_REF:
2838 case LABEL_REF:
2839 case ADDR_VEC:
2840 case ADDR_DIFF_VEC:
2841 return x;
2843 case REG:
2845 int first;
2846 int q;
2847 struct qty_table_elem *ent;
2849 /* Never replace a hard reg, because hard regs can appear
2850 in more than one machine mode, and we must preserve the mode
2851 of each occurrence. Also, some hard regs appear in
2852 MEMs that are shared and mustn't be altered. Don't try to
2853 replace any reg that maps to a reg of class NO_REGS. */
2854 if (REGNO (x) < FIRST_PSEUDO_REGISTER
2855 || ! REGNO_QTY_VALID_P (REGNO (x)))
2856 return x;
2858 q = REG_QTY (REGNO (x));
2859 ent = &qty_table[q];
2860 first = ent->first_reg;
2861 return (first >= FIRST_PSEUDO_REGISTER ? regno_reg_rtx[first]
2862 : REGNO_REG_CLASS (first) == NO_REGS ? x
2863 : gen_rtx_REG (ent->mode, first));
2866 default:
2867 break;
2870 fmt = GET_RTX_FORMAT (code);
2871 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2873 int j;
2875 if (fmt[i] == 'e')
2876 validate_canon_reg (&XEXP (x, i), insn);
2877 else if (fmt[i] == 'E')
2878 for (j = 0; j < XVECLEN (x, i); j++)
2879 validate_canon_reg (&XVECEXP (x, i, j), insn);
2882 return x;
2885 /* Given an operation (CODE, *PARG1, *PARG2), where code is a comparison
2886 operation (EQ, NE, GT, etc.), follow it back through the hash table and
2887 what values are being compared.
2889 *PARG1 and *PARG2 are updated to contain the rtx representing the values
2890 actually being compared. For example, if *PARG1 was (cc0) and *PARG2
2891 was (const_int 0), *PARG1 and *PARG2 will be set to the objects that were
2892 compared to produce cc0.
2894 The return value is the comparison operator and is either the code of
2895 A or the code corresponding to the inverse of the comparison. */
2897 static enum rtx_code
2898 find_comparison_args (enum rtx_code code, rtx *parg1, rtx *parg2,
2899 enum machine_mode *pmode1, enum machine_mode *pmode2)
2901 rtx arg1, arg2;
2902 struct pointer_set_t *visited = NULL;
2903 /* Set nonzero when we find something of interest. */
2904 rtx x = NULL;
2906 arg1 = *parg1, arg2 = *parg2;
2908 /* If ARG2 is const0_rtx, see what ARG1 is equivalent to. */
2910 while (arg2 == CONST0_RTX (GET_MODE (arg1)))
2912 int reverse_code = 0;
2913 struct table_elt *p = 0;
2915 /* Remember state from previous iteration. */
2916 if (x)
2918 if (!visited)
2919 visited = pointer_set_create ();
2920 pointer_set_insert (visited, x);
2921 x = 0;
2924 /* If arg1 is a COMPARE, extract the comparison arguments from it.
2925 On machines with CC0, this is the only case that can occur, since
2926 fold_rtx will return the COMPARE or item being compared with zero
2927 when given CC0. */
2929 if (GET_CODE (arg1) == COMPARE && arg2 == const0_rtx)
2930 x = arg1;
2932 /* If ARG1 is a comparison operator and CODE is testing for
2933 STORE_FLAG_VALUE, get the inner arguments. */
2935 else if (COMPARISON_P (arg1))
2937 #ifdef FLOAT_STORE_FLAG_VALUE
2938 REAL_VALUE_TYPE fsfv;
2939 #endif
2941 if (code == NE
2942 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
2943 && code == LT && STORE_FLAG_VALUE == -1)
2944 #ifdef FLOAT_STORE_FLAG_VALUE
2945 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
2946 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
2947 REAL_VALUE_NEGATIVE (fsfv)))
2948 #endif
2950 x = arg1;
2951 else if (code == EQ
2952 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
2953 && code == GE && STORE_FLAG_VALUE == -1)
2954 #ifdef FLOAT_STORE_FLAG_VALUE
2955 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
2956 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
2957 REAL_VALUE_NEGATIVE (fsfv)))
2958 #endif
2960 x = arg1, reverse_code = 1;
2963 /* ??? We could also check for
2965 (ne (and (eq (...) (const_int 1))) (const_int 0))
2967 and related forms, but let's wait until we see them occurring. */
2969 if (x == 0)
2970 /* Look up ARG1 in the hash table and see if it has an equivalence
2971 that lets us see what is being compared. */
2972 p = lookup (arg1, SAFE_HASH (arg1, GET_MODE (arg1)), GET_MODE (arg1));
2973 if (p)
2975 p = p->first_same_value;
2977 /* If what we compare is already known to be constant, that is as
2978 good as it gets.
2979 We need to break the loop in this case, because otherwise we
2980 can have an infinite loop when looking at a reg that is known
2981 to be a constant which is the same as a comparison of a reg
2982 against zero which appears later in the insn stream, which in
2983 turn is constant and the same as the comparison of the first reg
2984 against zero... */
2985 if (p->is_const)
2986 break;
2989 for (; p; p = p->next_same_value)
2991 enum machine_mode inner_mode = GET_MODE (p->exp);
2992 #ifdef FLOAT_STORE_FLAG_VALUE
2993 REAL_VALUE_TYPE fsfv;
2994 #endif
2996 /* If the entry isn't valid, skip it. */
2997 if (! exp_equiv_p (p->exp, p->exp, 1, false))
2998 continue;
3000 /* If it's a comparison we've used before, skip it. */
3001 if (visited && pointer_set_contains (visited, p->exp))
3002 continue;
3004 if (GET_CODE (p->exp) == COMPARE
3005 /* Another possibility is that this machine has a compare insn
3006 that includes the comparison code. In that case, ARG1 would
3007 be equivalent to a comparison operation that would set ARG1 to
3008 either STORE_FLAG_VALUE or zero. If this is an NE operation,
3009 ORIG_CODE is the actual comparison being done; if it is an EQ,
3010 we must reverse ORIG_CODE. On machine with a negative value
3011 for STORE_FLAG_VALUE, also look at LT and GE operations. */
3012 || ((code == NE
3013 || (code == LT
3014 && val_signbit_known_set_p (inner_mode,
3015 STORE_FLAG_VALUE))
3016 #ifdef FLOAT_STORE_FLAG_VALUE
3017 || (code == LT
3018 && SCALAR_FLOAT_MODE_P (inner_mode)
3019 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3020 REAL_VALUE_NEGATIVE (fsfv)))
3021 #endif
3023 && COMPARISON_P (p->exp)))
3025 x = p->exp;
3026 break;
3028 else if ((code == EQ
3029 || (code == GE
3030 && val_signbit_known_set_p (inner_mode,
3031 STORE_FLAG_VALUE))
3032 #ifdef FLOAT_STORE_FLAG_VALUE
3033 || (code == GE
3034 && SCALAR_FLOAT_MODE_P (inner_mode)
3035 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3036 REAL_VALUE_NEGATIVE (fsfv)))
3037 #endif
3039 && COMPARISON_P (p->exp))
3041 reverse_code = 1;
3042 x = p->exp;
3043 break;
3046 /* If this non-trapping address, e.g. fp + constant, the
3047 equivalent is a better operand since it may let us predict
3048 the value of the comparison. */
3049 else if (!rtx_addr_can_trap_p (p->exp))
3051 arg1 = p->exp;
3052 continue;
3056 /* If we didn't find a useful equivalence for ARG1, we are done.
3057 Otherwise, set up for the next iteration. */
3058 if (x == 0)
3059 break;
3061 /* If we need to reverse the comparison, make sure that that is
3062 possible -- we can't necessarily infer the value of GE from LT
3063 with floating-point operands. */
3064 if (reverse_code)
3066 enum rtx_code reversed = reversed_comparison_code (x, NULL_RTX);
3067 if (reversed == UNKNOWN)
3068 break;
3069 else
3070 code = reversed;
3072 else if (COMPARISON_P (x))
3073 code = GET_CODE (x);
3074 arg1 = XEXP (x, 0), arg2 = XEXP (x, 1);
3077 /* Return our results. Return the modes from before fold_rtx
3078 because fold_rtx might produce const_int, and then it's too late. */
3079 *pmode1 = GET_MODE (arg1), *pmode2 = GET_MODE (arg2);
3080 *parg1 = fold_rtx (arg1, 0), *parg2 = fold_rtx (arg2, 0);
3082 if (visited)
3083 pointer_set_destroy (visited);
3084 return code;
3087 /* If X is a nontrivial arithmetic operation on an argument for which
3088 a constant value can be determined, return the result of operating
3089 on that value, as a constant. Otherwise, return X, possibly with
3090 one or more operands changed to a forward-propagated constant.
3092 If X is a register whose contents are known, we do NOT return
3093 those contents here; equiv_constant is called to perform that task.
3094 For SUBREGs and MEMs, we do that both here and in equiv_constant.
3096 INSN is the insn that we may be modifying. If it is 0, make a copy
3097 of X before modifying it. */
3099 static rtx
3100 fold_rtx (rtx x, rtx insn)
3102 enum rtx_code code;
3103 enum machine_mode mode;
3104 const char *fmt;
3105 int i;
3106 rtx new_rtx = 0;
3107 int changed = 0;
3109 /* Operands of X. */
3110 rtx folded_arg0;
3111 rtx folded_arg1;
3113 /* Constant equivalents of first three operands of X;
3114 0 when no such equivalent is known. */
3115 rtx const_arg0;
3116 rtx const_arg1;
3117 rtx const_arg2;
3119 /* The mode of the first operand of X. We need this for sign and zero
3120 extends. */
3121 enum machine_mode mode_arg0;
3123 if (x == 0)
3124 return x;
3126 /* Try to perform some initial simplifications on X. */
3127 code = GET_CODE (x);
3128 switch (code)
3130 case MEM:
3131 case SUBREG:
3132 if ((new_rtx = equiv_constant (x)) != NULL_RTX)
3133 return new_rtx;
3134 return x;
3136 case CONST:
3137 CASE_CONST_ANY:
3138 case SYMBOL_REF:
3139 case LABEL_REF:
3140 case REG:
3141 case PC:
3142 /* No use simplifying an EXPR_LIST
3143 since they are used only for lists of args
3144 in a function call's REG_EQUAL note. */
3145 case EXPR_LIST:
3146 return x;
3148 #ifdef HAVE_cc0
3149 case CC0:
3150 return prev_insn_cc0;
3151 #endif
3153 case ASM_OPERANDS:
3154 if (insn)
3156 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
3157 validate_change (insn, &ASM_OPERANDS_INPUT (x, i),
3158 fold_rtx (ASM_OPERANDS_INPUT (x, i), insn), 0);
3160 return x;
3162 #ifdef NO_FUNCTION_CSE
3163 case CALL:
3164 if (CONSTANT_P (XEXP (XEXP (x, 0), 0)))
3165 return x;
3166 break;
3167 #endif
3169 /* Anything else goes through the loop below. */
3170 default:
3171 break;
3174 mode = GET_MODE (x);
3175 const_arg0 = 0;
3176 const_arg1 = 0;
3177 const_arg2 = 0;
3178 mode_arg0 = VOIDmode;
3180 /* Try folding our operands.
3181 Then see which ones have constant values known. */
3183 fmt = GET_RTX_FORMAT (code);
3184 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
3185 if (fmt[i] == 'e')
3187 rtx folded_arg = XEXP (x, i), const_arg;
3188 enum machine_mode mode_arg = GET_MODE (folded_arg);
3190 switch (GET_CODE (folded_arg))
3192 case MEM:
3193 case REG:
3194 case SUBREG:
3195 const_arg = equiv_constant (folded_arg);
3196 break;
3198 case CONST:
3199 CASE_CONST_ANY:
3200 case SYMBOL_REF:
3201 case LABEL_REF:
3202 const_arg = folded_arg;
3203 break;
3205 #ifdef HAVE_cc0
3206 case CC0:
3207 /* The cc0-user and cc0-setter may be in different blocks if
3208 the cc0-setter potentially traps. In that case PREV_INSN_CC0
3209 will have been cleared as we exited the block with the
3210 setter.
3212 While we could potentially track cc0 in this case, it just
3213 doesn't seem to be worth it given that cc0 targets are not
3214 terribly common or important these days and trapping math
3215 is rarely used. The combination of those two conditions
3216 necessary to trip this situation is exceedingly rare in the
3217 real world. */
3218 if (!prev_insn_cc0)
3220 const_arg = NULL_RTX;
3222 else
3224 folded_arg = prev_insn_cc0;
3225 mode_arg = prev_insn_cc0_mode;
3226 const_arg = equiv_constant (folded_arg);
3228 break;
3229 #endif
3231 default:
3232 folded_arg = fold_rtx (folded_arg, insn);
3233 const_arg = equiv_constant (folded_arg);
3234 break;
3237 /* For the first three operands, see if the operand
3238 is constant or equivalent to a constant. */
3239 switch (i)
3241 case 0:
3242 folded_arg0 = folded_arg;
3243 const_arg0 = const_arg;
3244 mode_arg0 = mode_arg;
3245 break;
3246 case 1:
3247 folded_arg1 = folded_arg;
3248 const_arg1 = const_arg;
3249 break;
3250 case 2:
3251 const_arg2 = const_arg;
3252 break;
3255 /* Pick the least expensive of the argument and an equivalent constant
3256 argument. */
3257 if (const_arg != 0
3258 && const_arg != folded_arg
3259 && COST_IN (const_arg, code, i) <= COST_IN (folded_arg, code, i)
3261 /* It's not safe to substitute the operand of a conversion
3262 operator with a constant, as the conversion's identity
3263 depends upon the mode of its operand. This optimization
3264 is handled by the call to simplify_unary_operation. */
3265 && (GET_RTX_CLASS (code) != RTX_UNARY
3266 || GET_MODE (const_arg) == mode_arg0
3267 || (code != ZERO_EXTEND
3268 && code != SIGN_EXTEND
3269 && code != TRUNCATE
3270 && code != FLOAT_TRUNCATE
3271 && code != FLOAT_EXTEND
3272 && code != FLOAT
3273 && code != FIX
3274 && code != UNSIGNED_FLOAT
3275 && code != UNSIGNED_FIX)))
3276 folded_arg = const_arg;
3278 if (folded_arg == XEXP (x, i))
3279 continue;
3281 if (insn == NULL_RTX && !changed)
3282 x = copy_rtx (x);
3283 changed = 1;
3284 validate_unshare_change (insn, &XEXP (x, i), folded_arg, 1);
3287 if (changed)
3289 /* Canonicalize X if necessary, and keep const_argN and folded_argN
3290 consistent with the order in X. */
3291 if (canonicalize_change_group (insn, x))
3293 rtx tem;
3294 tem = const_arg0, const_arg0 = const_arg1, const_arg1 = tem;
3295 tem = folded_arg0, folded_arg0 = folded_arg1, folded_arg1 = tem;
3298 apply_change_group ();
3301 /* If X is an arithmetic operation, see if we can simplify it. */
3303 switch (GET_RTX_CLASS (code))
3305 case RTX_UNARY:
3307 /* We can't simplify extension ops unless we know the
3308 original mode. */
3309 if ((code == ZERO_EXTEND || code == SIGN_EXTEND)
3310 && mode_arg0 == VOIDmode)
3311 break;
3313 new_rtx = simplify_unary_operation (code, mode,
3314 const_arg0 ? const_arg0 : folded_arg0,
3315 mode_arg0);
3317 break;
3319 case RTX_COMPARE:
3320 case RTX_COMM_COMPARE:
3321 /* See what items are actually being compared and set FOLDED_ARG[01]
3322 to those values and CODE to the actual comparison code. If any are
3323 constant, set CONST_ARG0 and CONST_ARG1 appropriately. We needn't
3324 do anything if both operands are already known to be constant. */
3326 /* ??? Vector mode comparisons are not supported yet. */
3327 if (VECTOR_MODE_P (mode))
3328 break;
3330 if (const_arg0 == 0 || const_arg1 == 0)
3332 struct table_elt *p0, *p1;
3333 rtx true_rtx, false_rtx;
3334 enum machine_mode mode_arg1;
3336 if (SCALAR_FLOAT_MODE_P (mode))
3338 #ifdef FLOAT_STORE_FLAG_VALUE
3339 true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
3340 (FLOAT_STORE_FLAG_VALUE (mode), mode));
3341 #else
3342 true_rtx = NULL_RTX;
3343 #endif
3344 false_rtx = CONST0_RTX (mode);
3346 else
3348 true_rtx = const_true_rtx;
3349 false_rtx = const0_rtx;
3352 code = find_comparison_args (code, &folded_arg0, &folded_arg1,
3353 &mode_arg0, &mode_arg1);
3355 /* If the mode is VOIDmode or a MODE_CC mode, we don't know
3356 what kinds of things are being compared, so we can't do
3357 anything with this comparison. */
3359 if (mode_arg0 == VOIDmode || GET_MODE_CLASS (mode_arg0) == MODE_CC)
3360 break;
3362 const_arg0 = equiv_constant (folded_arg0);
3363 const_arg1 = equiv_constant (folded_arg1);
3365 /* If we do not now have two constants being compared, see
3366 if we can nevertheless deduce some things about the
3367 comparison. */
3368 if (const_arg0 == 0 || const_arg1 == 0)
3370 if (const_arg1 != NULL)
3372 rtx cheapest_simplification;
3373 int cheapest_cost;
3374 rtx simp_result;
3375 struct table_elt *p;
3377 /* See if we can find an equivalent of folded_arg0
3378 that gets us a cheaper expression, possibly a
3379 constant through simplifications. */
3380 p = lookup (folded_arg0, SAFE_HASH (folded_arg0, mode_arg0),
3381 mode_arg0);
3383 if (p != NULL)
3385 cheapest_simplification = x;
3386 cheapest_cost = COST (x);
3388 for (p = p->first_same_value; p != NULL; p = p->next_same_value)
3390 int cost;
3392 /* If the entry isn't valid, skip it. */
3393 if (! exp_equiv_p (p->exp, p->exp, 1, false))
3394 continue;
3396 /* Try to simplify using this equivalence. */
3397 simp_result
3398 = simplify_relational_operation (code, mode,
3399 mode_arg0,
3400 p->exp,
3401 const_arg1);
3403 if (simp_result == NULL)
3404 continue;
3406 cost = COST (simp_result);
3407 if (cost < cheapest_cost)
3409 cheapest_cost = cost;
3410 cheapest_simplification = simp_result;
3414 /* If we have a cheaper expression now, use that
3415 and try folding it further, from the top. */
3416 if (cheapest_simplification != x)
3417 return fold_rtx (copy_rtx (cheapest_simplification),
3418 insn);
3422 /* See if the two operands are the same. */
3424 if ((REG_P (folded_arg0)
3425 && REG_P (folded_arg1)
3426 && (REG_QTY (REGNO (folded_arg0))
3427 == REG_QTY (REGNO (folded_arg1))))
3428 || ((p0 = lookup (folded_arg0,
3429 SAFE_HASH (folded_arg0, mode_arg0),
3430 mode_arg0))
3431 && (p1 = lookup (folded_arg1,
3432 SAFE_HASH (folded_arg1, mode_arg0),
3433 mode_arg0))
3434 && p0->first_same_value == p1->first_same_value))
3435 folded_arg1 = folded_arg0;
3437 /* If FOLDED_ARG0 is a register, see if the comparison we are
3438 doing now is either the same as we did before or the reverse
3439 (we only check the reverse if not floating-point). */
3440 else if (REG_P (folded_arg0))
3442 int qty = REG_QTY (REGNO (folded_arg0));
3444 if (REGNO_QTY_VALID_P (REGNO (folded_arg0)))
3446 struct qty_table_elem *ent = &qty_table[qty];
3448 if ((comparison_dominates_p (ent->comparison_code, code)
3449 || (! FLOAT_MODE_P (mode_arg0)
3450 && comparison_dominates_p (ent->comparison_code,
3451 reverse_condition (code))))
3452 && (rtx_equal_p (ent->comparison_const, folded_arg1)
3453 || (const_arg1
3454 && rtx_equal_p (ent->comparison_const,
3455 const_arg1))
3456 || (REG_P (folded_arg1)
3457 && (REG_QTY (REGNO (folded_arg1)) == ent->comparison_qty))))
3459 if (comparison_dominates_p (ent->comparison_code, code))
3461 if (true_rtx)
3462 return true_rtx;
3463 else
3464 break;
3466 else
3467 return false_rtx;
3474 /* If we are comparing against zero, see if the first operand is
3475 equivalent to an IOR with a constant. If so, we may be able to
3476 determine the result of this comparison. */
3477 if (const_arg1 == const0_rtx && !const_arg0)
3479 rtx y = lookup_as_function (folded_arg0, IOR);
3480 rtx inner_const;
3482 if (y != 0
3483 && (inner_const = equiv_constant (XEXP (y, 1))) != 0
3484 && CONST_INT_P (inner_const)
3485 && INTVAL (inner_const) != 0)
3486 folded_arg0 = gen_rtx_IOR (mode_arg0, XEXP (y, 0), inner_const);
3490 rtx op0 = const_arg0 ? const_arg0 : copy_rtx (folded_arg0);
3491 rtx op1 = const_arg1 ? const_arg1 : copy_rtx (folded_arg1);
3492 new_rtx = simplify_relational_operation (code, mode, mode_arg0,
3493 op0, op1);
3495 break;
3497 case RTX_BIN_ARITH:
3498 case RTX_COMM_ARITH:
3499 switch (code)
3501 case PLUS:
3502 /* If the second operand is a LABEL_REF, see if the first is a MINUS
3503 with that LABEL_REF as its second operand. If so, the result is
3504 the first operand of that MINUS. This handles switches with an
3505 ADDR_DIFF_VEC table. */
3506 if (const_arg1 && GET_CODE (const_arg1) == LABEL_REF)
3508 rtx y
3509 = GET_CODE (folded_arg0) == MINUS ? folded_arg0
3510 : lookup_as_function (folded_arg0, MINUS);
3512 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
3513 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg1, 0))
3514 return XEXP (y, 0);
3516 /* Now try for a CONST of a MINUS like the above. */
3517 if ((y = (GET_CODE (folded_arg0) == CONST ? folded_arg0
3518 : lookup_as_function (folded_arg0, CONST))) != 0
3519 && GET_CODE (XEXP (y, 0)) == MINUS
3520 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
3521 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg1, 0))
3522 return XEXP (XEXP (y, 0), 0);
3525 /* Likewise if the operands are in the other order. */
3526 if (const_arg0 && GET_CODE (const_arg0) == LABEL_REF)
3528 rtx y
3529 = GET_CODE (folded_arg1) == MINUS ? folded_arg1
3530 : lookup_as_function (folded_arg1, MINUS);
3532 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
3533 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg0, 0))
3534 return XEXP (y, 0);
3536 /* Now try for a CONST of a MINUS like the above. */
3537 if ((y = (GET_CODE (folded_arg1) == CONST ? folded_arg1
3538 : lookup_as_function (folded_arg1, CONST))) != 0
3539 && GET_CODE (XEXP (y, 0)) == MINUS
3540 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
3541 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg0, 0))
3542 return XEXP (XEXP (y, 0), 0);
3545 /* If second operand is a register equivalent to a negative
3546 CONST_INT, see if we can find a register equivalent to the
3547 positive constant. Make a MINUS if so. Don't do this for
3548 a non-negative constant since we might then alternate between
3549 choosing positive and negative constants. Having the positive
3550 constant previously-used is the more common case. Be sure
3551 the resulting constant is non-negative; if const_arg1 were
3552 the smallest negative number this would overflow: depending
3553 on the mode, this would either just be the same value (and
3554 hence not save anything) or be incorrect. */
3555 if (const_arg1 != 0 && CONST_INT_P (const_arg1)
3556 && INTVAL (const_arg1) < 0
3557 /* This used to test
3559 -INTVAL (const_arg1) >= 0
3561 But The Sun V5.0 compilers mis-compiled that test. So
3562 instead we test for the problematic value in a more direct
3563 manner and hope the Sun compilers get it correct. */
3564 && INTVAL (const_arg1) !=
3565 ((HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1))
3566 && REG_P (folded_arg1))
3568 rtx new_const = GEN_INT (-INTVAL (const_arg1));
3569 struct table_elt *p
3570 = lookup (new_const, SAFE_HASH (new_const, mode), mode);
3572 if (p)
3573 for (p = p->first_same_value; p; p = p->next_same_value)
3574 if (REG_P (p->exp))
3575 return simplify_gen_binary (MINUS, mode, folded_arg0,
3576 canon_reg (p->exp, NULL_RTX));
3578 goto from_plus;
3580 case MINUS:
3581 /* If we have (MINUS Y C), see if Y is known to be (PLUS Z C2).
3582 If so, produce (PLUS Z C2-C). */
3583 if (const_arg1 != 0 && CONST_INT_P (const_arg1))
3585 rtx y = lookup_as_function (XEXP (x, 0), PLUS);
3586 if (y && CONST_INT_P (XEXP (y, 1)))
3587 return fold_rtx (plus_constant (mode, copy_rtx (y),
3588 -INTVAL (const_arg1)),
3589 NULL_RTX);
3592 /* Fall through. */
3594 from_plus:
3595 case SMIN: case SMAX: case UMIN: case UMAX:
3596 case IOR: case AND: case XOR:
3597 case MULT:
3598 case ASHIFT: case LSHIFTRT: case ASHIFTRT:
3599 /* If we have (<op> <reg> <const_int>) for an associative OP and REG
3600 is known to be of similar form, we may be able to replace the
3601 operation with a combined operation. This may eliminate the
3602 intermediate operation if every use is simplified in this way.
3603 Note that the similar optimization done by combine.c only works
3604 if the intermediate operation's result has only one reference. */
3606 if (REG_P (folded_arg0)
3607 && const_arg1 && CONST_INT_P (const_arg1))
3609 int is_shift
3610 = (code == ASHIFT || code == ASHIFTRT || code == LSHIFTRT);
3611 rtx y, inner_const, new_const;
3612 rtx canon_const_arg1 = const_arg1;
3613 enum rtx_code associate_code;
3615 if (is_shift
3616 && (INTVAL (const_arg1) >= GET_MODE_PRECISION (mode)
3617 || INTVAL (const_arg1) < 0))
3619 if (SHIFT_COUNT_TRUNCATED)
3620 canon_const_arg1 = GEN_INT (INTVAL (const_arg1)
3621 & (GET_MODE_BITSIZE (mode)
3622 - 1));
3623 else
3624 break;
3627 y = lookup_as_function (folded_arg0, code);
3628 if (y == 0)
3629 break;
3631 /* If we have compiled a statement like
3632 "if (x == (x & mask1))", and now are looking at
3633 "x & mask2", we will have a case where the first operand
3634 of Y is the same as our first operand. Unless we detect
3635 this case, an infinite loop will result. */
3636 if (XEXP (y, 0) == folded_arg0)
3637 break;
3639 inner_const = equiv_constant (fold_rtx (XEXP (y, 1), 0));
3640 if (!inner_const || !CONST_INT_P (inner_const))
3641 break;
3643 /* Don't associate these operations if they are a PLUS with the
3644 same constant and it is a power of two. These might be doable
3645 with a pre- or post-increment. Similarly for two subtracts of
3646 identical powers of two with post decrement. */
3648 if (code == PLUS && const_arg1 == inner_const
3649 && ((HAVE_PRE_INCREMENT
3650 && exact_log2 (INTVAL (const_arg1)) >= 0)
3651 || (HAVE_POST_INCREMENT
3652 && exact_log2 (INTVAL (const_arg1)) >= 0)
3653 || (HAVE_PRE_DECREMENT
3654 && exact_log2 (- INTVAL (const_arg1)) >= 0)
3655 || (HAVE_POST_DECREMENT
3656 && exact_log2 (- INTVAL (const_arg1)) >= 0)))
3657 break;
3659 /* ??? Vector mode shifts by scalar
3660 shift operand are not supported yet. */
3661 if (is_shift && VECTOR_MODE_P (mode))
3662 break;
3664 if (is_shift
3665 && (INTVAL (inner_const) >= GET_MODE_PRECISION (mode)
3666 || INTVAL (inner_const) < 0))
3668 if (SHIFT_COUNT_TRUNCATED)
3669 inner_const = GEN_INT (INTVAL (inner_const)
3670 & (GET_MODE_BITSIZE (mode) - 1));
3671 else
3672 break;
3675 /* Compute the code used to compose the constants. For example,
3676 A-C1-C2 is A-(C1 + C2), so if CODE == MINUS, we want PLUS. */
3678 associate_code = (is_shift || code == MINUS ? PLUS : code);
3680 new_const = simplify_binary_operation (associate_code, mode,
3681 canon_const_arg1,
3682 inner_const);
3684 if (new_const == 0)
3685 break;
3687 /* If we are associating shift operations, don't let this
3688 produce a shift of the size of the object or larger.
3689 This could occur when we follow a sign-extend by a right
3690 shift on a machine that does a sign-extend as a pair
3691 of shifts. */
3693 if (is_shift
3694 && CONST_INT_P (new_const)
3695 && INTVAL (new_const) >= GET_MODE_PRECISION (mode))
3697 /* As an exception, we can turn an ASHIFTRT of this
3698 form into a shift of the number of bits - 1. */
3699 if (code == ASHIFTRT)
3700 new_const = GEN_INT (GET_MODE_BITSIZE (mode) - 1);
3701 else if (!side_effects_p (XEXP (y, 0)))
3702 return CONST0_RTX (mode);
3703 else
3704 break;
3707 y = copy_rtx (XEXP (y, 0));
3709 /* If Y contains our first operand (the most common way this
3710 can happen is if Y is a MEM), we would do into an infinite
3711 loop if we tried to fold it. So don't in that case. */
3713 if (! reg_mentioned_p (folded_arg0, y))
3714 y = fold_rtx (y, insn);
3716 return simplify_gen_binary (code, mode, y, new_const);
3718 break;
3720 case DIV: case UDIV:
3721 /* ??? The associative optimization performed immediately above is
3722 also possible for DIV and UDIV using associate_code of MULT.
3723 However, we would need extra code to verify that the
3724 multiplication does not overflow, that is, there is no overflow
3725 in the calculation of new_const. */
3726 break;
3728 default:
3729 break;
3732 new_rtx = simplify_binary_operation (code, mode,
3733 const_arg0 ? const_arg0 : folded_arg0,
3734 const_arg1 ? const_arg1 : folded_arg1);
3735 break;
3737 case RTX_OBJ:
3738 /* (lo_sum (high X) X) is simply X. */
3739 if (code == LO_SUM && const_arg0 != 0
3740 && GET_CODE (const_arg0) == HIGH
3741 && rtx_equal_p (XEXP (const_arg0, 0), const_arg1))
3742 return const_arg1;
3743 break;
3745 case RTX_TERNARY:
3746 case RTX_BITFIELD_OPS:
3747 new_rtx = simplify_ternary_operation (code, mode, mode_arg0,
3748 const_arg0 ? const_arg0 : folded_arg0,
3749 const_arg1 ? const_arg1 : folded_arg1,
3750 const_arg2 ? const_arg2 : XEXP (x, 2));
3751 break;
3753 default:
3754 break;
3757 return new_rtx ? new_rtx : x;
3760 /* Return a constant value currently equivalent to X.
3761 Return 0 if we don't know one. */
3763 static rtx
3764 equiv_constant (rtx x)
3766 if (REG_P (x)
3767 && REGNO_QTY_VALID_P (REGNO (x)))
3769 int x_q = REG_QTY (REGNO (x));
3770 struct qty_table_elem *x_ent = &qty_table[x_q];
3772 if (x_ent->const_rtx)
3773 x = gen_lowpart (GET_MODE (x), x_ent->const_rtx);
3776 if (x == 0 || CONSTANT_P (x))
3777 return x;
3779 if (GET_CODE (x) == SUBREG)
3781 enum machine_mode mode = GET_MODE (x);
3782 enum machine_mode imode = GET_MODE (SUBREG_REG (x));
3783 rtx new_rtx;
3785 /* See if we previously assigned a constant value to this SUBREG. */
3786 if ((new_rtx = lookup_as_function (x, CONST_INT)) != 0
3787 || (new_rtx = lookup_as_function (x, CONST_WIDE_INT)) != 0
3788 || (new_rtx = lookup_as_function (x, CONST_DOUBLE)) != 0
3789 || (new_rtx = lookup_as_function (x, CONST_FIXED)) != 0)
3790 return new_rtx;
3792 /* If we didn't and if doing so makes sense, see if we previously
3793 assigned a constant value to the enclosing word mode SUBREG. */
3794 if (GET_MODE_SIZE (mode) < GET_MODE_SIZE (word_mode)
3795 && GET_MODE_SIZE (word_mode) < GET_MODE_SIZE (imode))
3797 int byte = SUBREG_BYTE (x) - subreg_lowpart_offset (mode, word_mode);
3798 if (byte >= 0 && (byte % UNITS_PER_WORD) == 0)
3800 rtx y = gen_rtx_SUBREG (word_mode, SUBREG_REG (x), byte);
3801 new_rtx = lookup_as_function (y, CONST_INT);
3802 if (new_rtx)
3803 return gen_lowpart (mode, new_rtx);
3807 /* Otherwise see if we already have a constant for the inner REG,
3808 and if that is enough to calculate an equivalent constant for
3809 the subreg. Note that the upper bits of paradoxical subregs
3810 are undefined, so they cannot be said to equal anything. */
3811 if (REG_P (SUBREG_REG (x))
3812 && GET_MODE_SIZE (mode) <= GET_MODE_SIZE (imode)
3813 && (new_rtx = equiv_constant (SUBREG_REG (x))) != 0)
3814 return simplify_subreg (mode, new_rtx, imode, SUBREG_BYTE (x));
3816 return 0;
3819 /* If X is a MEM, see if it is a constant-pool reference, or look it up in
3820 the hash table in case its value was seen before. */
3822 if (MEM_P (x))
3824 struct table_elt *elt;
3826 x = avoid_constant_pool_reference (x);
3827 if (CONSTANT_P (x))
3828 return x;
3830 elt = lookup (x, SAFE_HASH (x, GET_MODE (x)), GET_MODE (x));
3831 if (elt == 0)
3832 return 0;
3834 for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
3835 if (elt->is_const && CONSTANT_P (elt->exp))
3836 return elt->exp;
3839 return 0;
3842 /* Given INSN, a jump insn, TAKEN indicates if we are following the
3843 "taken" branch.
3845 In certain cases, this can cause us to add an equivalence. For example,
3846 if we are following the taken case of
3847 if (i == 2)
3848 we can add the fact that `i' and '2' are now equivalent.
3850 In any case, we can record that this comparison was passed. If the same
3851 comparison is seen later, we will know its value. */
3853 static void
3854 record_jump_equiv (rtx insn, bool taken)
3856 int cond_known_true;
3857 rtx op0, op1;
3858 rtx set;
3859 enum machine_mode mode, mode0, mode1;
3860 int reversed_nonequality = 0;
3861 enum rtx_code code;
3863 /* Ensure this is the right kind of insn. */
3864 gcc_assert (any_condjump_p (insn));
3866 set = pc_set (insn);
3868 /* See if this jump condition is known true or false. */
3869 if (taken)
3870 cond_known_true = (XEXP (SET_SRC (set), 2) == pc_rtx);
3871 else
3872 cond_known_true = (XEXP (SET_SRC (set), 1) == pc_rtx);
3874 /* Get the type of comparison being done and the operands being compared.
3875 If we had to reverse a non-equality condition, record that fact so we
3876 know that it isn't valid for floating-point. */
3877 code = GET_CODE (XEXP (SET_SRC (set), 0));
3878 op0 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 0), insn);
3879 op1 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 1), insn);
3881 code = find_comparison_args (code, &op0, &op1, &mode0, &mode1);
3882 if (! cond_known_true)
3884 code = reversed_comparison_code_parts (code, op0, op1, insn);
3886 /* Don't remember if we can't find the inverse. */
3887 if (code == UNKNOWN)
3888 return;
3891 /* The mode is the mode of the non-constant. */
3892 mode = mode0;
3893 if (mode1 != VOIDmode)
3894 mode = mode1;
3896 record_jump_cond (code, mode, op0, op1, reversed_nonequality);
3899 /* Yet another form of subreg creation. In this case, we want something in
3900 MODE, and we should assume OP has MODE iff it is naturally modeless. */
3902 static rtx
3903 record_jump_cond_subreg (enum machine_mode mode, rtx op)
3905 enum machine_mode op_mode = GET_MODE (op);
3906 if (op_mode == mode || op_mode == VOIDmode)
3907 return op;
3908 return lowpart_subreg (mode, op, op_mode);
3911 /* We know that comparison CODE applied to OP0 and OP1 in MODE is true.
3912 REVERSED_NONEQUALITY is nonzero if CODE had to be swapped.
3913 Make any useful entries we can with that information. Called from
3914 above function and called recursively. */
3916 static void
3917 record_jump_cond (enum rtx_code code, enum machine_mode mode, rtx op0,
3918 rtx op1, int reversed_nonequality)
3920 unsigned op0_hash, op1_hash;
3921 int op0_in_memory, op1_in_memory;
3922 struct table_elt *op0_elt, *op1_elt;
3924 /* If OP0 and OP1 are known equal, and either is a paradoxical SUBREG,
3925 we know that they are also equal in the smaller mode (this is also
3926 true for all smaller modes whether or not there is a SUBREG, but
3927 is not worth testing for with no SUBREG). */
3929 /* Note that GET_MODE (op0) may not equal MODE. */
3930 if (code == EQ && paradoxical_subreg_p (op0))
3932 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
3933 rtx tem = record_jump_cond_subreg (inner_mode, op1);
3934 if (tem)
3935 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
3936 reversed_nonequality);
3939 if (code == EQ && paradoxical_subreg_p (op1))
3941 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
3942 rtx tem = record_jump_cond_subreg (inner_mode, op0);
3943 if (tem)
3944 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
3945 reversed_nonequality);
3948 /* Similarly, if this is an NE comparison, and either is a SUBREG
3949 making a smaller mode, we know the whole thing is also NE. */
3951 /* Note that GET_MODE (op0) may not equal MODE;
3952 if we test MODE instead, we can get an infinite recursion
3953 alternating between two modes each wider than MODE. */
3955 if (code == NE && GET_CODE (op0) == SUBREG
3956 && subreg_lowpart_p (op0)
3957 && (GET_MODE_SIZE (GET_MODE (op0))
3958 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
3960 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
3961 rtx tem = record_jump_cond_subreg (inner_mode, op1);
3962 if (tem)
3963 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
3964 reversed_nonequality);
3967 if (code == NE && GET_CODE (op1) == SUBREG
3968 && subreg_lowpart_p (op1)
3969 && (GET_MODE_SIZE (GET_MODE (op1))
3970 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
3972 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
3973 rtx tem = record_jump_cond_subreg (inner_mode, op0);
3974 if (tem)
3975 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
3976 reversed_nonequality);
3979 /* Hash both operands. */
3981 do_not_record = 0;
3982 hash_arg_in_memory = 0;
3983 op0_hash = HASH (op0, mode);
3984 op0_in_memory = hash_arg_in_memory;
3986 if (do_not_record)
3987 return;
3989 do_not_record = 0;
3990 hash_arg_in_memory = 0;
3991 op1_hash = HASH (op1, mode);
3992 op1_in_memory = hash_arg_in_memory;
3994 if (do_not_record)
3995 return;
3997 /* Look up both operands. */
3998 op0_elt = lookup (op0, op0_hash, mode);
3999 op1_elt = lookup (op1, op1_hash, mode);
4001 /* If both operands are already equivalent or if they are not in the
4002 table but are identical, do nothing. */
4003 if ((op0_elt != 0 && op1_elt != 0
4004 && op0_elt->first_same_value == op1_elt->first_same_value)
4005 || op0 == op1 || rtx_equal_p (op0, op1))
4006 return;
4008 /* If we aren't setting two things equal all we can do is save this
4009 comparison. Similarly if this is floating-point. In the latter
4010 case, OP1 might be zero and both -0.0 and 0.0 are equal to it.
4011 If we record the equality, we might inadvertently delete code
4012 whose intent was to change -0 to +0. */
4014 if (code != EQ || FLOAT_MODE_P (GET_MODE (op0)))
4016 struct qty_table_elem *ent;
4017 int qty;
4019 /* If we reversed a floating-point comparison, if OP0 is not a
4020 register, or if OP1 is neither a register or constant, we can't
4021 do anything. */
4023 if (!REG_P (op1))
4024 op1 = equiv_constant (op1);
4026 if ((reversed_nonequality && FLOAT_MODE_P (mode))
4027 || !REG_P (op0) || op1 == 0)
4028 return;
4030 /* Put OP0 in the hash table if it isn't already. This gives it a
4031 new quantity number. */
4032 if (op0_elt == 0)
4034 if (insert_regs (op0, NULL, 0))
4036 rehash_using_reg (op0);
4037 op0_hash = HASH (op0, mode);
4039 /* If OP0 is contained in OP1, this changes its hash code
4040 as well. Faster to rehash than to check, except
4041 for the simple case of a constant. */
4042 if (! CONSTANT_P (op1))
4043 op1_hash = HASH (op1,mode);
4046 op0_elt = insert (op0, NULL, op0_hash, mode);
4047 op0_elt->in_memory = op0_in_memory;
4050 qty = REG_QTY (REGNO (op0));
4051 ent = &qty_table[qty];
4053 ent->comparison_code = code;
4054 if (REG_P (op1))
4056 /* Look it up again--in case op0 and op1 are the same. */
4057 op1_elt = lookup (op1, op1_hash, mode);
4059 /* Put OP1 in the hash table so it gets a new quantity number. */
4060 if (op1_elt == 0)
4062 if (insert_regs (op1, NULL, 0))
4064 rehash_using_reg (op1);
4065 op1_hash = HASH (op1, mode);
4068 op1_elt = insert (op1, NULL, op1_hash, mode);
4069 op1_elt->in_memory = op1_in_memory;
4072 ent->comparison_const = NULL_RTX;
4073 ent->comparison_qty = REG_QTY (REGNO (op1));
4075 else
4077 ent->comparison_const = op1;
4078 ent->comparison_qty = -1;
4081 return;
4084 /* If either side is still missing an equivalence, make it now,
4085 then merge the equivalences. */
4087 if (op0_elt == 0)
4089 if (insert_regs (op0, NULL, 0))
4091 rehash_using_reg (op0);
4092 op0_hash = HASH (op0, mode);
4095 op0_elt = insert (op0, NULL, op0_hash, mode);
4096 op0_elt->in_memory = op0_in_memory;
4099 if (op1_elt == 0)
4101 if (insert_regs (op1, NULL, 0))
4103 rehash_using_reg (op1);
4104 op1_hash = HASH (op1, mode);
4107 op1_elt = insert (op1, NULL, op1_hash, mode);
4108 op1_elt->in_memory = op1_in_memory;
4111 merge_equiv_classes (op0_elt, op1_elt);
4114 /* CSE processing for one instruction.
4116 Most "true" common subexpressions are mostly optimized away in GIMPLE,
4117 but the few that "leak through" are cleaned up by cse_insn, and complex
4118 addressing modes are often formed here.
4120 The main function is cse_insn, and between here and that function
4121 a couple of helper functions is defined to keep the size of cse_insn
4122 within reasonable proportions.
4124 Data is shared between the main and helper functions via STRUCT SET,
4125 that contains all data related for every set in the instruction that
4126 is being processed.
4128 Note that cse_main processes all sets in the instruction. Most
4129 passes in GCC only process simple SET insns or single_set insns, but
4130 CSE processes insns with multiple sets as well. */
4132 /* Data on one SET contained in the instruction. */
4134 struct set
4136 /* The SET rtx itself. */
4137 rtx rtl;
4138 /* The SET_SRC of the rtx (the original value, if it is changing). */
4139 rtx src;
4140 /* The hash-table element for the SET_SRC of the SET. */
4141 struct table_elt *src_elt;
4142 /* Hash value for the SET_SRC. */
4143 unsigned src_hash;
4144 /* Hash value for the SET_DEST. */
4145 unsigned dest_hash;
4146 /* The SET_DEST, with SUBREG, etc., stripped. */
4147 rtx inner_dest;
4148 /* Nonzero if the SET_SRC is in memory. */
4149 char src_in_memory;
4150 /* Nonzero if the SET_SRC contains something
4151 whose value cannot be predicted and understood. */
4152 char src_volatile;
4153 /* Original machine mode, in case it becomes a CONST_INT.
4154 The size of this field should match the size of the mode
4155 field of struct rtx_def (see rtl.h). */
4156 ENUM_BITFIELD(machine_mode) mode : 8;
4157 /* A constant equivalent for SET_SRC, if any. */
4158 rtx src_const;
4159 /* Hash value of constant equivalent for SET_SRC. */
4160 unsigned src_const_hash;
4161 /* Table entry for constant equivalent for SET_SRC, if any. */
4162 struct table_elt *src_const_elt;
4163 /* Table entry for the destination address. */
4164 struct table_elt *dest_addr_elt;
4167 /* Special handling for (set REG0 REG1) where REG0 is the
4168 "cheapest", cheaper than REG1. After cse, REG1 will probably not
4169 be used in the sequel, so (if easily done) change this insn to
4170 (set REG1 REG0) and replace REG1 with REG0 in the previous insn
4171 that computed their value. Then REG1 will become a dead store
4172 and won't cloud the situation for later optimizations.
4174 Do not make this change if REG1 is a hard register, because it will
4175 then be used in the sequel and we may be changing a two-operand insn
4176 into a three-operand insn.
4178 This is the last transformation that cse_insn will try to do. */
4180 static void
4181 try_back_substitute_reg (rtx set, rtx insn)
4183 rtx dest = SET_DEST (set);
4184 rtx src = SET_SRC (set);
4186 if (REG_P (dest)
4187 && REG_P (src) && ! HARD_REGISTER_P (src)
4188 && REGNO_QTY_VALID_P (REGNO (src)))
4190 int src_q = REG_QTY (REGNO (src));
4191 struct qty_table_elem *src_ent = &qty_table[src_q];
4193 if (src_ent->first_reg == REGNO (dest))
4195 /* Scan for the previous nonnote insn, but stop at a basic
4196 block boundary. */
4197 rtx prev = insn;
4198 rtx bb_head = BB_HEAD (BLOCK_FOR_INSN (insn));
4201 prev = PREV_INSN (prev);
4203 while (prev != bb_head && (NOTE_P (prev) || DEBUG_INSN_P (prev)));
4205 /* Do not swap the registers around if the previous instruction
4206 attaches a REG_EQUIV note to REG1.
4208 ??? It's not entirely clear whether we can transfer a REG_EQUIV
4209 from the pseudo that originally shadowed an incoming argument
4210 to another register. Some uses of REG_EQUIV might rely on it
4211 being attached to REG1 rather than REG2.
4213 This section previously turned the REG_EQUIV into a REG_EQUAL
4214 note. We cannot do that because REG_EQUIV may provide an
4215 uninitialized stack slot when REG_PARM_STACK_SPACE is used. */
4216 if (NONJUMP_INSN_P (prev)
4217 && GET_CODE (PATTERN (prev)) == SET
4218 && SET_DEST (PATTERN (prev)) == src
4219 && ! find_reg_note (prev, REG_EQUIV, NULL_RTX))
4221 rtx note;
4223 validate_change (prev, &SET_DEST (PATTERN (prev)), dest, 1);
4224 validate_change (insn, &SET_DEST (set), src, 1);
4225 validate_change (insn, &SET_SRC (set), dest, 1);
4226 apply_change_group ();
4228 /* If INSN has a REG_EQUAL note, and this note mentions
4229 REG0, then we must delete it, because the value in
4230 REG0 has changed. If the note's value is REG1, we must
4231 also delete it because that is now this insn's dest. */
4232 note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
4233 if (note != 0
4234 && (reg_mentioned_p (dest, XEXP (note, 0))
4235 || rtx_equal_p (src, XEXP (note, 0))))
4236 remove_note (insn, note);
4242 /* Record all the SETs in this instruction into SETS_PTR,
4243 and return the number of recorded sets. */
4244 static int
4245 find_sets_in_insn (rtx insn, struct set **psets)
4247 struct set *sets = *psets;
4248 int n_sets = 0;
4249 rtx x = PATTERN (insn);
4251 if (GET_CODE (x) == SET)
4253 /* Ignore SETs that are unconditional jumps.
4254 They never need cse processing, so this does not hurt.
4255 The reason is not efficiency but rather
4256 so that we can test at the end for instructions
4257 that have been simplified to unconditional jumps
4258 and not be misled by unchanged instructions
4259 that were unconditional jumps to begin with. */
4260 if (SET_DEST (x) == pc_rtx
4261 && GET_CODE (SET_SRC (x)) == LABEL_REF)
4263 /* Don't count call-insns, (set (reg 0) (call ...)), as a set.
4264 The hard function value register is used only once, to copy to
4265 someplace else, so it isn't worth cse'ing. */
4266 else if (GET_CODE (SET_SRC (x)) == CALL)
4268 else
4269 sets[n_sets++].rtl = x;
4271 else if (GET_CODE (x) == PARALLEL)
4273 int i, lim = XVECLEN (x, 0);
4275 /* Go over the epressions of the PARALLEL in forward order, to
4276 put them in the same order in the SETS array. */
4277 for (i = 0; i < lim; i++)
4279 rtx y = XVECEXP (x, 0, i);
4280 if (GET_CODE (y) == SET)
4282 /* As above, we ignore unconditional jumps and call-insns and
4283 ignore the result of apply_change_group. */
4284 if (SET_DEST (y) == pc_rtx
4285 && GET_CODE (SET_SRC (y)) == LABEL_REF)
4287 else if (GET_CODE (SET_SRC (y)) == CALL)
4289 else
4290 sets[n_sets++].rtl = y;
4295 return n_sets;
4298 /* Where possible, substitute every register reference in the N_SETS
4299 number of SETS in INSN with the the canonical register.
4301 Register canonicalization propagatest the earliest register (i.e.
4302 one that is set before INSN) with the same value. This is a very
4303 useful, simple form of CSE, to clean up warts from expanding GIMPLE
4304 to RTL. For instance, a CONST for an address is usually expanded
4305 multiple times to loads into different registers, thus creating many
4306 subexpressions of the form:
4308 (set (reg1) (some_const))
4309 (set (mem (... reg1 ...) (thing)))
4310 (set (reg2) (some_const))
4311 (set (mem (... reg2 ...) (thing)))
4313 After canonicalizing, the code takes the following form:
4315 (set (reg1) (some_const))
4316 (set (mem (... reg1 ...) (thing)))
4317 (set (reg2) (some_const))
4318 (set (mem (... reg1 ...) (thing)))
4320 The set to reg2 is now trivially dead, and the memory reference (or
4321 address, or whatever) may be a candidate for further CSEing.
4323 In this function, the result of apply_change_group can be ignored;
4324 see canon_reg. */
4326 static void
4327 canonicalize_insn (rtx insn, struct set **psets, int n_sets)
4329 struct set *sets = *psets;
4330 rtx tem;
4331 rtx x = PATTERN (insn);
4332 int i;
4334 if (CALL_P (insn))
4336 for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
4337 if (GET_CODE (XEXP (tem, 0)) != SET)
4338 XEXP (tem, 0) = canon_reg (XEXP (tem, 0), insn);
4341 if (GET_CODE (x) == SET && GET_CODE (SET_SRC (x)) == CALL)
4343 canon_reg (SET_SRC (x), insn);
4344 apply_change_group ();
4345 fold_rtx (SET_SRC (x), insn);
4347 else if (GET_CODE (x) == CLOBBER)
4349 /* If we clobber memory, canon the address.
4350 This does nothing when a register is clobbered
4351 because we have already invalidated the reg. */
4352 if (MEM_P (XEXP (x, 0)))
4353 canon_reg (XEXP (x, 0), insn);
4355 else if (GET_CODE (x) == USE
4356 && ! (REG_P (XEXP (x, 0))
4357 && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER))
4358 /* Canonicalize a USE of a pseudo register or memory location. */
4359 canon_reg (x, insn);
4360 else if (GET_CODE (x) == ASM_OPERANDS)
4362 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
4364 rtx input = ASM_OPERANDS_INPUT (x, i);
4365 if (!(REG_P (input) && REGNO (input) < FIRST_PSEUDO_REGISTER))
4367 input = canon_reg (input, insn);
4368 validate_change (insn, &ASM_OPERANDS_INPUT (x, i), input, 1);
4372 else if (GET_CODE (x) == CALL)
4374 canon_reg (x, insn);
4375 apply_change_group ();
4376 fold_rtx (x, insn);
4378 else if (DEBUG_INSN_P (insn))
4379 canon_reg (PATTERN (insn), insn);
4380 else if (GET_CODE (x) == PARALLEL)
4382 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
4384 rtx y = XVECEXP (x, 0, i);
4385 if (GET_CODE (y) == SET && GET_CODE (SET_SRC (y)) == CALL)
4387 canon_reg (SET_SRC (y), insn);
4388 apply_change_group ();
4389 fold_rtx (SET_SRC (y), insn);
4391 else if (GET_CODE (y) == CLOBBER)
4393 if (MEM_P (XEXP (y, 0)))
4394 canon_reg (XEXP (y, 0), insn);
4396 else if (GET_CODE (y) == USE
4397 && ! (REG_P (XEXP (y, 0))
4398 && REGNO (XEXP (y, 0)) < FIRST_PSEUDO_REGISTER))
4399 canon_reg (y, insn);
4400 else if (GET_CODE (y) == CALL)
4402 canon_reg (y, insn);
4403 apply_change_group ();
4404 fold_rtx (y, insn);
4409 if (n_sets == 1 && REG_NOTES (insn) != 0
4410 && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0)
4412 /* We potentially will process this insn many times. Therefore,
4413 drop the REG_EQUAL note if it is equal to the SET_SRC of the
4414 unique set in INSN.
4416 Do not do so if the REG_EQUAL note is for a STRICT_LOW_PART,
4417 because cse_insn handles those specially. */
4418 if (GET_CODE (SET_DEST (sets[0].rtl)) != STRICT_LOW_PART
4419 && rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl)))
4420 remove_note (insn, tem);
4421 else
4423 canon_reg (XEXP (tem, 0), insn);
4424 apply_change_group ();
4425 XEXP (tem, 0) = fold_rtx (XEXP (tem, 0), insn);
4426 df_notes_rescan (insn);
4430 /* Canonicalize sources and addresses of destinations.
4431 We do this in a separate pass to avoid problems when a MATCH_DUP is
4432 present in the insn pattern. In that case, we want to ensure that
4433 we don't break the duplicate nature of the pattern. So we will replace
4434 both operands at the same time. Otherwise, we would fail to find an
4435 equivalent substitution in the loop calling validate_change below.
4437 We used to suppress canonicalization of DEST if it appears in SRC,
4438 but we don't do this any more. */
4440 for (i = 0; i < n_sets; i++)
4442 rtx dest = SET_DEST (sets[i].rtl);
4443 rtx src = SET_SRC (sets[i].rtl);
4444 rtx new_rtx = canon_reg (src, insn);
4446 validate_change (insn, &SET_SRC (sets[i].rtl), new_rtx, 1);
4448 if (GET_CODE (dest) == ZERO_EXTRACT)
4450 validate_change (insn, &XEXP (dest, 1),
4451 canon_reg (XEXP (dest, 1), insn), 1);
4452 validate_change (insn, &XEXP (dest, 2),
4453 canon_reg (XEXP (dest, 2), insn), 1);
4456 while (GET_CODE (dest) == SUBREG
4457 || GET_CODE (dest) == ZERO_EXTRACT
4458 || GET_CODE (dest) == STRICT_LOW_PART)
4459 dest = XEXP (dest, 0);
4461 if (MEM_P (dest))
4462 canon_reg (dest, insn);
4465 /* Now that we have done all the replacements, we can apply the change
4466 group and see if they all work. Note that this will cause some
4467 canonicalizations that would have worked individually not to be applied
4468 because some other canonicalization didn't work, but this should not
4469 occur often.
4471 The result of apply_change_group can be ignored; see canon_reg. */
4473 apply_change_group ();
4476 /* Main function of CSE.
4477 First simplify sources and addresses of all assignments
4478 in the instruction, using previously-computed equivalents values.
4479 Then install the new sources and destinations in the table
4480 of available values. */
4482 static void
4483 cse_insn (rtx insn)
4485 rtx x = PATTERN (insn);
4486 int i;
4487 rtx tem;
4488 int n_sets = 0;
4490 rtx src_eqv = 0;
4491 struct table_elt *src_eqv_elt = 0;
4492 int src_eqv_volatile = 0;
4493 int src_eqv_in_memory = 0;
4494 unsigned src_eqv_hash = 0;
4496 struct set *sets = (struct set *) 0;
4498 if (GET_CODE (x) == SET)
4499 sets = XALLOCA (struct set);
4500 else if (GET_CODE (x) == PARALLEL)
4501 sets = XALLOCAVEC (struct set, XVECLEN (x, 0));
4503 this_insn = insn;
4504 #ifdef HAVE_cc0
4505 /* Records what this insn does to set CC0. */
4506 this_insn_cc0 = 0;
4507 this_insn_cc0_mode = VOIDmode;
4508 #endif
4510 /* Find all regs explicitly clobbered in this insn,
4511 to ensure they are not replaced with any other regs
4512 elsewhere in this insn. */
4513 invalidate_from_sets_and_clobbers (insn);
4515 /* Record all the SETs in this instruction. */
4516 n_sets = find_sets_in_insn (insn, &sets);
4518 /* Substitute the canonical register where possible. */
4519 canonicalize_insn (insn, &sets, n_sets);
4521 /* If this insn has a REG_EQUAL note, store the equivalent value in SRC_EQV,
4522 if different, or if the DEST is a STRICT_LOW_PART. The latter condition
4523 is necessary because SRC_EQV is handled specially for this case, and if
4524 it isn't set, then there will be no equivalence for the destination. */
4525 if (n_sets == 1 && REG_NOTES (insn) != 0
4526 && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0
4527 && (! rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl))
4528 || GET_CODE (SET_DEST (sets[0].rtl)) == STRICT_LOW_PART))
4529 src_eqv = copy_rtx (XEXP (tem, 0));
4531 /* Set sets[i].src_elt to the class each source belongs to.
4532 Detect assignments from or to volatile things
4533 and set set[i] to zero so they will be ignored
4534 in the rest of this function.
4536 Nothing in this loop changes the hash table or the register chains. */
4538 for (i = 0; i < n_sets; i++)
4540 bool repeat = false;
4541 rtx src, dest;
4542 rtx src_folded;
4543 struct table_elt *elt = 0, *p;
4544 enum machine_mode mode;
4545 rtx src_eqv_here;
4546 rtx src_const = 0;
4547 rtx src_related = 0;
4548 bool src_related_is_const_anchor = false;
4549 struct table_elt *src_const_elt = 0;
4550 int src_cost = MAX_COST;
4551 int src_eqv_cost = MAX_COST;
4552 int src_folded_cost = MAX_COST;
4553 int src_related_cost = MAX_COST;
4554 int src_elt_cost = MAX_COST;
4555 int src_regcost = MAX_COST;
4556 int src_eqv_regcost = MAX_COST;
4557 int src_folded_regcost = MAX_COST;
4558 int src_related_regcost = MAX_COST;
4559 int src_elt_regcost = MAX_COST;
4560 /* Set nonzero if we need to call force_const_mem on with the
4561 contents of src_folded before using it. */
4562 int src_folded_force_flag = 0;
4564 dest = SET_DEST (sets[i].rtl);
4565 src = SET_SRC (sets[i].rtl);
4567 /* If SRC is a constant that has no machine mode,
4568 hash it with the destination's machine mode.
4569 This way we can keep different modes separate. */
4571 mode = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
4572 sets[i].mode = mode;
4574 if (src_eqv)
4576 enum machine_mode eqvmode = mode;
4577 if (GET_CODE (dest) == STRICT_LOW_PART)
4578 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
4579 do_not_record = 0;
4580 hash_arg_in_memory = 0;
4581 src_eqv_hash = HASH (src_eqv, eqvmode);
4583 /* Find the equivalence class for the equivalent expression. */
4585 if (!do_not_record)
4586 src_eqv_elt = lookup (src_eqv, src_eqv_hash, eqvmode);
4588 src_eqv_volatile = do_not_record;
4589 src_eqv_in_memory = hash_arg_in_memory;
4592 /* If this is a STRICT_LOW_PART assignment, src_eqv corresponds to the
4593 value of the INNER register, not the destination. So it is not
4594 a valid substitution for the source. But save it for later. */
4595 if (GET_CODE (dest) == STRICT_LOW_PART)
4596 src_eqv_here = 0;
4597 else
4598 src_eqv_here = src_eqv;
4600 /* Simplify and foldable subexpressions in SRC. Then get the fully-
4601 simplified result, which may not necessarily be valid. */
4602 src_folded = fold_rtx (src, insn);
4604 #if 0
4605 /* ??? This caused bad code to be generated for the m68k port with -O2.
4606 Suppose src is (CONST_INT -1), and that after truncation src_folded
4607 is (CONST_INT 3). Suppose src_folded is then used for src_const.
4608 At the end we will add src and src_const to the same equivalence
4609 class. We now have 3 and -1 on the same equivalence class. This
4610 causes later instructions to be mis-optimized. */
4611 /* If storing a constant in a bitfield, pre-truncate the constant
4612 so we will be able to record it later. */
4613 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
4615 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
4617 if (CONST_INT_P (src)
4618 && CONST_INT_P (width)
4619 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
4620 && (INTVAL (src) & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
4621 src_folded
4622 = GEN_INT (INTVAL (src) & (((HOST_WIDE_INT) 1
4623 << INTVAL (width)) - 1));
4625 #endif
4627 /* Compute SRC's hash code, and also notice if it
4628 should not be recorded at all. In that case,
4629 prevent any further processing of this assignment. */
4630 do_not_record = 0;
4631 hash_arg_in_memory = 0;
4633 sets[i].src = src;
4634 sets[i].src_hash = HASH (src, mode);
4635 sets[i].src_volatile = do_not_record;
4636 sets[i].src_in_memory = hash_arg_in_memory;
4638 /* If SRC is a MEM, there is a REG_EQUIV note for SRC, and DEST is
4639 a pseudo, do not record SRC. Using SRC as a replacement for
4640 anything else will be incorrect in that situation. Note that
4641 this usually occurs only for stack slots, in which case all the
4642 RTL would be referring to SRC, so we don't lose any optimization
4643 opportunities by not having SRC in the hash table. */
4645 if (MEM_P (src)
4646 && find_reg_note (insn, REG_EQUIV, NULL_RTX) != 0
4647 && REG_P (dest)
4648 && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
4649 sets[i].src_volatile = 1;
4651 /* Also do not record result of a non-volatile inline asm with
4652 more than one result or with clobbers, we do not want CSE to
4653 break the inline asm apart. */
4654 else if (GET_CODE (src) == ASM_OPERANDS
4655 && GET_CODE (x) == PARALLEL)
4656 sets[i].src_volatile = 1;
4658 #if 0
4659 /* It is no longer clear why we used to do this, but it doesn't
4660 appear to still be needed. So let's try without it since this
4661 code hurts cse'ing widened ops. */
4662 /* If source is a paradoxical subreg (such as QI treated as an SI),
4663 treat it as volatile. It may do the work of an SI in one context
4664 where the extra bits are not being used, but cannot replace an SI
4665 in general. */
4666 if (paradoxical_subreg_p (src))
4667 sets[i].src_volatile = 1;
4668 #endif
4670 /* Locate all possible equivalent forms for SRC. Try to replace
4671 SRC in the insn with each cheaper equivalent.
4673 We have the following types of equivalents: SRC itself, a folded
4674 version, a value given in a REG_EQUAL note, or a value related
4675 to a constant.
4677 Each of these equivalents may be part of an additional class
4678 of equivalents (if more than one is in the table, they must be in
4679 the same class; we check for this).
4681 If the source is volatile, we don't do any table lookups.
4683 We note any constant equivalent for possible later use in a
4684 REG_NOTE. */
4686 if (!sets[i].src_volatile)
4687 elt = lookup (src, sets[i].src_hash, mode);
4689 sets[i].src_elt = elt;
4691 if (elt && src_eqv_here && src_eqv_elt)
4693 if (elt->first_same_value != src_eqv_elt->first_same_value)
4695 /* The REG_EQUAL is indicating that two formerly distinct
4696 classes are now equivalent. So merge them. */
4697 merge_equiv_classes (elt, src_eqv_elt);
4698 src_eqv_hash = HASH (src_eqv, elt->mode);
4699 src_eqv_elt = lookup (src_eqv, src_eqv_hash, elt->mode);
4702 src_eqv_here = 0;
4705 else if (src_eqv_elt)
4706 elt = src_eqv_elt;
4708 /* Try to find a constant somewhere and record it in `src_const'.
4709 Record its table element, if any, in `src_const_elt'. Look in
4710 any known equivalences first. (If the constant is not in the
4711 table, also set `sets[i].src_const_hash'). */
4712 if (elt)
4713 for (p = elt->first_same_value; p; p = p->next_same_value)
4714 if (p->is_const)
4716 src_const = p->exp;
4717 src_const_elt = elt;
4718 break;
4721 if (src_const == 0
4722 && (CONSTANT_P (src_folded)
4723 /* Consider (minus (label_ref L1) (label_ref L2)) as
4724 "constant" here so we will record it. This allows us
4725 to fold switch statements when an ADDR_DIFF_VEC is used. */
4726 || (GET_CODE (src_folded) == MINUS
4727 && GET_CODE (XEXP (src_folded, 0)) == LABEL_REF
4728 && GET_CODE (XEXP (src_folded, 1)) == LABEL_REF)))
4729 src_const = src_folded, src_const_elt = elt;
4730 else if (src_const == 0 && src_eqv_here && CONSTANT_P (src_eqv_here))
4731 src_const = src_eqv_here, src_const_elt = src_eqv_elt;
4733 /* If we don't know if the constant is in the table, get its
4734 hash code and look it up. */
4735 if (src_const && src_const_elt == 0)
4737 sets[i].src_const_hash = HASH (src_const, mode);
4738 src_const_elt = lookup (src_const, sets[i].src_const_hash, mode);
4741 sets[i].src_const = src_const;
4742 sets[i].src_const_elt = src_const_elt;
4744 /* If the constant and our source are both in the table, mark them as
4745 equivalent. Otherwise, if a constant is in the table but the source
4746 isn't, set ELT to it. */
4747 if (src_const_elt && elt
4748 && src_const_elt->first_same_value != elt->first_same_value)
4749 merge_equiv_classes (elt, src_const_elt);
4750 else if (src_const_elt && elt == 0)
4751 elt = src_const_elt;
4753 /* See if there is a register linearly related to a constant
4754 equivalent of SRC. */
4755 if (src_const
4756 && (GET_CODE (src_const) == CONST
4757 || (src_const_elt && src_const_elt->related_value != 0)))
4759 src_related = use_related_value (src_const, src_const_elt);
4760 if (src_related)
4762 struct table_elt *src_related_elt
4763 = lookup (src_related, HASH (src_related, mode), mode);
4764 if (src_related_elt && elt)
4766 if (elt->first_same_value
4767 != src_related_elt->first_same_value)
4768 /* This can occur when we previously saw a CONST
4769 involving a SYMBOL_REF and then see the SYMBOL_REF
4770 twice. Merge the involved classes. */
4771 merge_equiv_classes (elt, src_related_elt);
4773 src_related = 0;
4774 src_related_elt = 0;
4776 else if (src_related_elt && elt == 0)
4777 elt = src_related_elt;
4781 /* See if we have a CONST_INT that is already in a register in a
4782 wider mode. */
4784 if (src_const && src_related == 0 && CONST_INT_P (src_const)
4785 && GET_MODE_CLASS (mode) == MODE_INT
4786 && GET_MODE_PRECISION (mode) < BITS_PER_WORD)
4788 enum machine_mode wider_mode;
4790 for (wider_mode = GET_MODE_WIDER_MODE (mode);
4791 wider_mode != VOIDmode
4792 && GET_MODE_PRECISION (wider_mode) <= BITS_PER_WORD
4793 && src_related == 0;
4794 wider_mode = GET_MODE_WIDER_MODE (wider_mode))
4796 struct table_elt *const_elt
4797 = lookup (src_const, HASH (src_const, wider_mode), wider_mode);
4799 if (const_elt == 0)
4800 continue;
4802 for (const_elt = const_elt->first_same_value;
4803 const_elt; const_elt = const_elt->next_same_value)
4804 if (REG_P (const_elt->exp))
4806 src_related = gen_lowpart (mode, const_elt->exp);
4807 break;
4812 /* Another possibility is that we have an AND with a constant in
4813 a mode narrower than a word. If so, it might have been generated
4814 as part of an "if" which would narrow the AND. If we already
4815 have done the AND in a wider mode, we can use a SUBREG of that
4816 value. */
4818 if (flag_expensive_optimizations && ! src_related
4819 && GET_CODE (src) == AND && CONST_INT_P (XEXP (src, 1))
4820 && GET_MODE_SIZE (mode) < UNITS_PER_WORD)
4822 enum machine_mode tmode;
4823 rtx new_and = gen_rtx_AND (VOIDmode, NULL_RTX, XEXP (src, 1));
4825 for (tmode = GET_MODE_WIDER_MODE (mode);
4826 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
4827 tmode = GET_MODE_WIDER_MODE (tmode))
4829 rtx inner = gen_lowpart (tmode, XEXP (src, 0));
4830 struct table_elt *larger_elt;
4832 if (inner)
4834 PUT_MODE (new_and, tmode);
4835 XEXP (new_and, 0) = inner;
4836 larger_elt = lookup (new_and, HASH (new_and, tmode), tmode);
4837 if (larger_elt == 0)
4838 continue;
4840 for (larger_elt = larger_elt->first_same_value;
4841 larger_elt; larger_elt = larger_elt->next_same_value)
4842 if (REG_P (larger_elt->exp))
4844 src_related
4845 = gen_lowpart (mode, larger_elt->exp);
4846 break;
4849 if (src_related)
4850 break;
4855 #ifdef LOAD_EXTEND_OP
4856 /* See if a MEM has already been loaded with a widening operation;
4857 if it has, we can use a subreg of that. Many CISC machines
4858 also have such operations, but this is only likely to be
4859 beneficial on these machines. */
4861 if (flag_expensive_optimizations && src_related == 0
4862 && (GET_MODE_SIZE (mode) < UNITS_PER_WORD)
4863 && GET_MODE_CLASS (mode) == MODE_INT
4864 && MEM_P (src) && ! do_not_record
4865 && LOAD_EXTEND_OP (mode) != UNKNOWN)
4867 struct rtx_def memory_extend_buf;
4868 rtx memory_extend_rtx = &memory_extend_buf;
4869 enum machine_mode tmode;
4871 /* Set what we are trying to extend and the operation it might
4872 have been extended with. */
4873 memset (memory_extend_rtx, 0, sizeof (*memory_extend_rtx));
4874 PUT_CODE (memory_extend_rtx, LOAD_EXTEND_OP (mode));
4875 XEXP (memory_extend_rtx, 0) = src;
4877 for (tmode = GET_MODE_WIDER_MODE (mode);
4878 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
4879 tmode = GET_MODE_WIDER_MODE (tmode))
4881 struct table_elt *larger_elt;
4883 PUT_MODE (memory_extend_rtx, tmode);
4884 larger_elt = lookup (memory_extend_rtx,
4885 HASH (memory_extend_rtx, tmode), tmode);
4886 if (larger_elt == 0)
4887 continue;
4889 for (larger_elt = larger_elt->first_same_value;
4890 larger_elt; larger_elt = larger_elt->next_same_value)
4891 if (REG_P (larger_elt->exp))
4893 src_related = gen_lowpart (mode, larger_elt->exp);
4894 break;
4897 if (src_related)
4898 break;
4901 #endif /* LOAD_EXTEND_OP */
4903 /* Try to express the constant using a register+offset expression
4904 derived from a constant anchor. */
4906 if (targetm.const_anchor
4907 && !src_related
4908 && src_const
4909 && GET_CODE (src_const) == CONST_INT)
4911 src_related = try_const_anchors (src_const, mode);
4912 src_related_is_const_anchor = src_related != NULL_RTX;
4916 if (src == src_folded)
4917 src_folded = 0;
4919 /* At this point, ELT, if nonzero, points to a class of expressions
4920 equivalent to the source of this SET and SRC, SRC_EQV, SRC_FOLDED,
4921 and SRC_RELATED, if nonzero, each contain additional equivalent
4922 expressions. Prune these latter expressions by deleting expressions
4923 already in the equivalence class.
4925 Check for an equivalent identical to the destination. If found,
4926 this is the preferred equivalent since it will likely lead to
4927 elimination of the insn. Indicate this by placing it in
4928 `src_related'. */
4930 if (elt)
4931 elt = elt->first_same_value;
4932 for (p = elt; p; p = p->next_same_value)
4934 enum rtx_code code = GET_CODE (p->exp);
4936 /* If the expression is not valid, ignore it. Then we do not
4937 have to check for validity below. In most cases, we can use
4938 `rtx_equal_p', since canonicalization has already been done. */
4939 if (code != REG && ! exp_equiv_p (p->exp, p->exp, 1, false))
4940 continue;
4942 /* Also skip paradoxical subregs, unless that's what we're
4943 looking for. */
4944 if (paradoxical_subreg_p (p->exp)
4945 && ! (src != 0
4946 && GET_CODE (src) == SUBREG
4947 && GET_MODE (src) == GET_MODE (p->exp)
4948 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
4949 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))))
4950 continue;
4952 if (src && GET_CODE (src) == code && rtx_equal_p (src, p->exp))
4953 src = 0;
4954 else if (src_folded && GET_CODE (src_folded) == code
4955 && rtx_equal_p (src_folded, p->exp))
4956 src_folded = 0;
4957 else if (src_eqv_here && GET_CODE (src_eqv_here) == code
4958 && rtx_equal_p (src_eqv_here, p->exp))
4959 src_eqv_here = 0;
4960 else if (src_related && GET_CODE (src_related) == code
4961 && rtx_equal_p (src_related, p->exp))
4962 src_related = 0;
4964 /* This is the same as the destination of the insns, we want
4965 to prefer it. Copy it to src_related. The code below will
4966 then give it a negative cost. */
4967 if (GET_CODE (dest) == code && rtx_equal_p (p->exp, dest))
4968 src_related = dest;
4971 /* Find the cheapest valid equivalent, trying all the available
4972 possibilities. Prefer items not in the hash table to ones
4973 that are when they are equal cost. Note that we can never
4974 worsen an insn as the current contents will also succeed.
4975 If we find an equivalent identical to the destination, use it as best,
4976 since this insn will probably be eliminated in that case. */
4977 if (src)
4979 if (rtx_equal_p (src, dest))
4980 src_cost = src_regcost = -1;
4981 else
4983 src_cost = COST (src);
4984 src_regcost = approx_reg_cost (src);
4988 if (src_eqv_here)
4990 if (rtx_equal_p (src_eqv_here, dest))
4991 src_eqv_cost = src_eqv_regcost = -1;
4992 else
4994 src_eqv_cost = COST (src_eqv_here);
4995 src_eqv_regcost = approx_reg_cost (src_eqv_here);
4999 if (src_folded)
5001 if (rtx_equal_p (src_folded, dest))
5002 src_folded_cost = src_folded_regcost = -1;
5003 else
5005 src_folded_cost = COST (src_folded);
5006 src_folded_regcost = approx_reg_cost (src_folded);
5010 if (src_related)
5012 if (rtx_equal_p (src_related, dest))
5013 src_related_cost = src_related_regcost = -1;
5014 else
5016 src_related_cost = COST (src_related);
5017 src_related_regcost = approx_reg_cost (src_related);
5019 /* If a const-anchor is used to synthesize a constant that
5020 normally requires multiple instructions then slightly prefer
5021 it over the original sequence. These instructions are likely
5022 to become redundant now. We can't compare against the cost
5023 of src_eqv_here because, on MIPS for example, multi-insn
5024 constants have zero cost; they are assumed to be hoisted from
5025 loops. */
5026 if (src_related_is_const_anchor
5027 && src_related_cost == src_cost
5028 && src_eqv_here)
5029 src_related_cost--;
5033 /* If this was an indirect jump insn, a known label will really be
5034 cheaper even though it looks more expensive. */
5035 if (dest == pc_rtx && src_const && GET_CODE (src_const) == LABEL_REF)
5036 src_folded = src_const, src_folded_cost = src_folded_regcost = -1;
5038 /* Terminate loop when replacement made. This must terminate since
5039 the current contents will be tested and will always be valid. */
5040 while (1)
5042 rtx trial;
5044 /* Skip invalid entries. */
5045 while (elt && !REG_P (elt->exp)
5046 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
5047 elt = elt->next_same_value;
5049 /* A paradoxical subreg would be bad here: it'll be the right
5050 size, but later may be adjusted so that the upper bits aren't
5051 what we want. So reject it. */
5052 if (elt != 0
5053 && paradoxical_subreg_p (elt->exp)
5054 /* It is okay, though, if the rtx we're trying to match
5055 will ignore any of the bits we can't predict. */
5056 && ! (src != 0
5057 && GET_CODE (src) == SUBREG
5058 && GET_MODE (src) == GET_MODE (elt->exp)
5059 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
5060 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))))
5062 elt = elt->next_same_value;
5063 continue;
5066 if (elt)
5068 src_elt_cost = elt->cost;
5069 src_elt_regcost = elt->regcost;
5072 /* Find cheapest and skip it for the next time. For items
5073 of equal cost, use this order:
5074 src_folded, src, src_eqv, src_related and hash table entry. */
5075 if (src_folded
5076 && preferable (src_folded_cost, src_folded_regcost,
5077 src_cost, src_regcost) <= 0
5078 && preferable (src_folded_cost, src_folded_regcost,
5079 src_eqv_cost, src_eqv_regcost) <= 0
5080 && preferable (src_folded_cost, src_folded_regcost,
5081 src_related_cost, src_related_regcost) <= 0
5082 && preferable (src_folded_cost, src_folded_regcost,
5083 src_elt_cost, src_elt_regcost) <= 0)
5085 trial = src_folded, src_folded_cost = MAX_COST;
5086 if (src_folded_force_flag)
5088 rtx forced = force_const_mem (mode, trial);
5089 if (forced)
5090 trial = forced;
5093 else if (src
5094 && preferable (src_cost, src_regcost,
5095 src_eqv_cost, src_eqv_regcost) <= 0
5096 && preferable (src_cost, src_regcost,
5097 src_related_cost, src_related_regcost) <= 0
5098 && preferable (src_cost, src_regcost,
5099 src_elt_cost, src_elt_regcost) <= 0)
5100 trial = src, src_cost = MAX_COST;
5101 else if (src_eqv_here
5102 && preferable (src_eqv_cost, src_eqv_regcost,
5103 src_related_cost, src_related_regcost) <= 0
5104 && preferable (src_eqv_cost, src_eqv_regcost,
5105 src_elt_cost, src_elt_regcost) <= 0)
5106 trial = src_eqv_here, src_eqv_cost = MAX_COST;
5107 else if (src_related
5108 && preferable (src_related_cost, src_related_regcost,
5109 src_elt_cost, src_elt_regcost) <= 0)
5110 trial = src_related, src_related_cost = MAX_COST;
5111 else
5113 trial = elt->exp;
5114 elt = elt->next_same_value;
5115 src_elt_cost = MAX_COST;
5118 /* Avoid creation of overlapping memory moves. */
5119 if (MEM_P (trial) && MEM_P (SET_DEST (sets[i].rtl)))
5121 rtx src, dest;
5123 /* BLKmode moves are not handled by cse anyway. */
5124 if (GET_MODE (trial) == BLKmode)
5125 break;
5127 src = canon_rtx (trial);
5128 dest = canon_rtx (SET_DEST (sets[i].rtl));
5130 if (!MEM_P (src) || !MEM_P (dest)
5131 || !nonoverlapping_memrefs_p (src, dest, false))
5132 break;
5135 /* Try to optimize
5136 (set (reg:M N) (const_int A))
5137 (set (reg:M2 O) (const_int B))
5138 (set (zero_extract:M2 (reg:M N) (const_int C) (const_int D))
5139 (reg:M2 O)). */
5140 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT
5141 && CONST_INT_P (trial)
5142 && CONST_INT_P (XEXP (SET_DEST (sets[i].rtl), 1))
5143 && CONST_INT_P (XEXP (SET_DEST (sets[i].rtl), 2))
5144 && REG_P (XEXP (SET_DEST (sets[i].rtl), 0))
5145 && (GET_MODE_PRECISION (GET_MODE (SET_DEST (sets[i].rtl)))
5146 >= INTVAL (XEXP (SET_DEST (sets[i].rtl), 1)))
5147 && ((unsigned) INTVAL (XEXP (SET_DEST (sets[i].rtl), 1))
5148 + (unsigned) INTVAL (XEXP (SET_DEST (sets[i].rtl), 2))
5149 <= HOST_BITS_PER_WIDE_INT))
5151 rtx dest_reg = XEXP (SET_DEST (sets[i].rtl), 0);
5152 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5153 rtx pos = XEXP (SET_DEST (sets[i].rtl), 2);
5154 unsigned int dest_hash = HASH (dest_reg, GET_MODE (dest_reg));
5155 struct table_elt *dest_elt
5156 = lookup (dest_reg, dest_hash, GET_MODE (dest_reg));
5157 rtx dest_cst = NULL;
5159 if (dest_elt)
5160 for (p = dest_elt->first_same_value; p; p = p->next_same_value)
5161 if (p->is_const && CONST_INT_P (p->exp))
5163 dest_cst = p->exp;
5164 break;
5166 if (dest_cst)
5168 HOST_WIDE_INT val = INTVAL (dest_cst);
5169 HOST_WIDE_INT mask;
5170 unsigned int shift;
5171 if (BITS_BIG_ENDIAN)
5172 shift = GET_MODE_PRECISION (GET_MODE (dest_reg))
5173 - INTVAL (pos) - INTVAL (width);
5174 else
5175 shift = INTVAL (pos);
5176 if (INTVAL (width) == HOST_BITS_PER_WIDE_INT)
5177 mask = ~(HOST_WIDE_INT) 0;
5178 else
5179 mask = ((HOST_WIDE_INT) 1 << INTVAL (width)) - 1;
5180 val &= ~(mask << shift);
5181 val |= (INTVAL (trial) & mask) << shift;
5182 val = trunc_int_for_mode (val, GET_MODE (dest_reg));
5183 validate_unshare_change (insn, &SET_DEST (sets[i].rtl),
5184 dest_reg, 1);
5185 validate_unshare_change (insn, &SET_SRC (sets[i].rtl),
5186 GEN_INT (val), 1);
5187 if (apply_change_group ())
5189 rtx note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
5190 if (note)
5192 remove_note (insn, note);
5193 df_notes_rescan (insn);
5195 src_eqv = NULL_RTX;
5196 src_eqv_elt = NULL;
5197 src_eqv_volatile = 0;
5198 src_eqv_in_memory = 0;
5199 src_eqv_hash = 0;
5200 repeat = true;
5201 break;
5206 /* We don't normally have an insn matching (set (pc) (pc)), so
5207 check for this separately here. We will delete such an
5208 insn below.
5210 For other cases such as a table jump or conditional jump
5211 where we know the ultimate target, go ahead and replace the
5212 operand. While that may not make a valid insn, we will
5213 reemit the jump below (and also insert any necessary
5214 barriers). */
5215 if (n_sets == 1 && dest == pc_rtx
5216 && (trial == pc_rtx
5217 || (GET_CODE (trial) == LABEL_REF
5218 && ! condjump_p (insn))))
5220 /* Don't substitute non-local labels, this confuses CFG. */
5221 if (GET_CODE (trial) == LABEL_REF
5222 && LABEL_REF_NONLOCAL_P (trial))
5223 continue;
5225 SET_SRC (sets[i].rtl) = trial;
5226 cse_jumps_altered = true;
5227 break;
5230 /* Reject certain invalid forms of CONST that we create. */
5231 else if (CONSTANT_P (trial)
5232 && GET_CODE (trial) == CONST
5233 /* Reject cases that will cause decode_rtx_const to
5234 die. On the alpha when simplifying a switch, we
5235 get (const (truncate (minus (label_ref)
5236 (label_ref)))). */
5237 && (GET_CODE (XEXP (trial, 0)) == TRUNCATE
5238 /* Likewise on IA-64, except without the
5239 truncate. */
5240 || (GET_CODE (XEXP (trial, 0)) == MINUS
5241 && GET_CODE (XEXP (XEXP (trial, 0), 0)) == LABEL_REF
5242 && GET_CODE (XEXP (XEXP (trial, 0), 1)) == LABEL_REF)))
5243 /* Do nothing for this case. */
5246 /* Look for a substitution that makes a valid insn. */
5247 else if (validate_unshare_change
5248 (insn, &SET_SRC (sets[i].rtl), trial, 0))
5250 rtx new_rtx = canon_reg (SET_SRC (sets[i].rtl), insn);
5252 /* The result of apply_change_group can be ignored; see
5253 canon_reg. */
5255 validate_change (insn, &SET_SRC (sets[i].rtl), new_rtx, 1);
5256 apply_change_group ();
5258 break;
5261 /* If we previously found constant pool entries for
5262 constants and this is a constant, try making a
5263 pool entry. Put it in src_folded unless we already have done
5264 this since that is where it likely came from. */
5266 else if (constant_pool_entries_cost
5267 && CONSTANT_P (trial)
5268 && (src_folded == 0
5269 || (!MEM_P (src_folded)
5270 && ! src_folded_force_flag))
5271 && GET_MODE_CLASS (mode) != MODE_CC
5272 && mode != VOIDmode)
5274 src_folded_force_flag = 1;
5275 src_folded = trial;
5276 src_folded_cost = constant_pool_entries_cost;
5277 src_folded_regcost = constant_pool_entries_regcost;
5281 /* If we changed the insn too much, handle this set from scratch. */
5282 if (repeat)
5284 i--;
5285 continue;
5288 src = SET_SRC (sets[i].rtl);
5290 /* In general, it is good to have a SET with SET_SRC == SET_DEST.
5291 However, there is an important exception: If both are registers
5292 that are not the head of their equivalence class, replace SET_SRC
5293 with the head of the class. If we do not do this, we will have
5294 both registers live over a portion of the basic block. This way,
5295 their lifetimes will likely abut instead of overlapping. */
5296 if (REG_P (dest)
5297 && REGNO_QTY_VALID_P (REGNO (dest)))
5299 int dest_q = REG_QTY (REGNO (dest));
5300 struct qty_table_elem *dest_ent = &qty_table[dest_q];
5302 if (dest_ent->mode == GET_MODE (dest)
5303 && dest_ent->first_reg != REGNO (dest)
5304 && REG_P (src) && REGNO (src) == REGNO (dest)
5305 /* Don't do this if the original insn had a hard reg as
5306 SET_SRC or SET_DEST. */
5307 && (!REG_P (sets[i].src)
5308 || REGNO (sets[i].src) >= FIRST_PSEUDO_REGISTER)
5309 && (!REG_P (dest) || REGNO (dest) >= FIRST_PSEUDO_REGISTER))
5310 /* We can't call canon_reg here because it won't do anything if
5311 SRC is a hard register. */
5313 int src_q = REG_QTY (REGNO (src));
5314 struct qty_table_elem *src_ent = &qty_table[src_q];
5315 int first = src_ent->first_reg;
5316 rtx new_src
5317 = (first >= FIRST_PSEUDO_REGISTER
5318 ? regno_reg_rtx[first] : gen_rtx_REG (GET_MODE (src), first));
5320 /* We must use validate-change even for this, because this
5321 might be a special no-op instruction, suitable only to
5322 tag notes onto. */
5323 if (validate_change (insn, &SET_SRC (sets[i].rtl), new_src, 0))
5325 src = new_src;
5326 /* If we had a constant that is cheaper than what we are now
5327 setting SRC to, use that constant. We ignored it when we
5328 thought we could make this into a no-op. */
5329 if (src_const && COST (src_const) < COST (src)
5330 && validate_change (insn, &SET_SRC (sets[i].rtl),
5331 src_const, 0))
5332 src = src_const;
5337 /* If we made a change, recompute SRC values. */
5338 if (src != sets[i].src)
5340 do_not_record = 0;
5341 hash_arg_in_memory = 0;
5342 sets[i].src = src;
5343 sets[i].src_hash = HASH (src, mode);
5344 sets[i].src_volatile = do_not_record;
5345 sets[i].src_in_memory = hash_arg_in_memory;
5346 sets[i].src_elt = lookup (src, sets[i].src_hash, mode);
5349 /* If this is a single SET, we are setting a register, and we have an
5350 equivalent constant, we want to add a REG_EQUAL note if the constant
5351 is different from the source. We don't want to do it for a constant
5352 pseudo since verifying that this pseudo hasn't been eliminated is a
5353 pain; moreover such a note won't help anything.
5355 Avoid a REG_EQUAL note for (CONST (MINUS (LABEL_REF) (LABEL_REF)))
5356 which can be created for a reference to a compile time computable
5357 entry in a jump table. */
5358 if (n_sets == 1
5359 && REG_P (dest)
5360 && src_const
5361 && !REG_P (src_const)
5362 && !(GET_CODE (src_const) == SUBREG
5363 && REG_P (SUBREG_REG (src_const)))
5364 && !(GET_CODE (src_const) == CONST
5365 && GET_CODE (XEXP (src_const, 0)) == MINUS
5366 && GET_CODE (XEXP (XEXP (src_const, 0), 0)) == LABEL_REF
5367 && GET_CODE (XEXP (XEXP (src_const, 0), 1)) == LABEL_REF)
5368 && !rtx_equal_p (src, src_const))
5370 /* Make sure that the rtx is not shared. */
5371 src_const = copy_rtx (src_const);
5373 /* Record the actual constant value in a REG_EQUAL note,
5374 making a new one if one does not already exist. */
5375 set_unique_reg_note (insn, REG_EQUAL, src_const);
5376 df_notes_rescan (insn);
5379 /* Now deal with the destination. */
5380 do_not_record = 0;
5382 /* Look within any ZERO_EXTRACT to the MEM or REG within it. */
5383 while (GET_CODE (dest) == SUBREG
5384 || GET_CODE (dest) == ZERO_EXTRACT
5385 || GET_CODE (dest) == STRICT_LOW_PART)
5386 dest = XEXP (dest, 0);
5388 sets[i].inner_dest = dest;
5390 if (MEM_P (dest))
5392 #ifdef PUSH_ROUNDING
5393 /* Stack pushes invalidate the stack pointer. */
5394 rtx addr = XEXP (dest, 0);
5395 if (GET_RTX_CLASS (GET_CODE (addr)) == RTX_AUTOINC
5396 && XEXP (addr, 0) == stack_pointer_rtx)
5397 invalidate (stack_pointer_rtx, VOIDmode);
5398 #endif
5399 dest = fold_rtx (dest, insn);
5402 /* Compute the hash code of the destination now,
5403 before the effects of this instruction are recorded,
5404 since the register values used in the address computation
5405 are those before this instruction. */
5406 sets[i].dest_hash = HASH (dest, mode);
5408 /* Don't enter a bit-field in the hash table
5409 because the value in it after the store
5410 may not equal what was stored, due to truncation. */
5412 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
5414 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5416 if (src_const != 0 && CONST_INT_P (src_const)
5417 && CONST_INT_P (width)
5418 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
5419 && ! (INTVAL (src_const)
5420 & (HOST_WIDE_INT_M1U << INTVAL (width))))
5421 /* Exception: if the value is constant,
5422 and it won't be truncated, record it. */
5424 else
5426 /* This is chosen so that the destination will be invalidated
5427 but no new value will be recorded.
5428 We must invalidate because sometimes constant
5429 values can be recorded for bitfields. */
5430 sets[i].src_elt = 0;
5431 sets[i].src_volatile = 1;
5432 src_eqv = 0;
5433 src_eqv_elt = 0;
5437 /* If only one set in a JUMP_INSN and it is now a no-op, we can delete
5438 the insn. */
5439 else if (n_sets == 1 && dest == pc_rtx && src == pc_rtx)
5441 /* One less use of the label this insn used to jump to. */
5442 delete_insn_and_edges (insn);
5443 cse_jumps_altered = true;
5444 /* No more processing for this set. */
5445 sets[i].rtl = 0;
5448 /* If this SET is now setting PC to a label, we know it used to
5449 be a conditional or computed branch. */
5450 else if (dest == pc_rtx && GET_CODE (src) == LABEL_REF
5451 && !LABEL_REF_NONLOCAL_P (src))
5453 /* We reemit the jump in as many cases as possible just in
5454 case the form of an unconditional jump is significantly
5455 different than a computed jump or conditional jump.
5457 If this insn has multiple sets, then reemitting the
5458 jump is nontrivial. So instead we just force rerecognition
5459 and hope for the best. */
5460 if (n_sets == 1)
5462 rtx new_rtx, note;
5464 new_rtx = emit_jump_insn_before (gen_jump (XEXP (src, 0)), insn);
5465 JUMP_LABEL (new_rtx) = XEXP (src, 0);
5466 LABEL_NUSES (XEXP (src, 0))++;
5468 /* Make sure to copy over REG_NON_LOCAL_GOTO. */
5469 note = find_reg_note (insn, REG_NON_LOCAL_GOTO, 0);
5470 if (note)
5472 XEXP (note, 1) = NULL_RTX;
5473 REG_NOTES (new_rtx) = note;
5476 delete_insn_and_edges (insn);
5477 insn = new_rtx;
5479 else
5480 INSN_CODE (insn) = -1;
5482 /* Do not bother deleting any unreachable code, let jump do it. */
5483 cse_jumps_altered = true;
5484 sets[i].rtl = 0;
5487 /* If destination is volatile, invalidate it and then do no further
5488 processing for this assignment. */
5490 else if (do_not_record)
5492 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5493 invalidate (dest, VOIDmode);
5494 else if (MEM_P (dest))
5495 invalidate (dest, VOIDmode);
5496 else if (GET_CODE (dest) == STRICT_LOW_PART
5497 || GET_CODE (dest) == ZERO_EXTRACT)
5498 invalidate (XEXP (dest, 0), GET_MODE (dest));
5499 sets[i].rtl = 0;
5502 if (sets[i].rtl != 0 && dest != SET_DEST (sets[i].rtl))
5503 sets[i].dest_hash = HASH (SET_DEST (sets[i].rtl), mode);
5505 #ifdef HAVE_cc0
5506 /* If setting CC0, record what it was set to, or a constant, if it
5507 is equivalent to a constant. If it is being set to a floating-point
5508 value, make a COMPARE with the appropriate constant of 0. If we
5509 don't do this, later code can interpret this as a test against
5510 const0_rtx, which can cause problems if we try to put it into an
5511 insn as a floating-point operand. */
5512 if (dest == cc0_rtx)
5514 this_insn_cc0 = src_const && mode != VOIDmode ? src_const : src;
5515 this_insn_cc0_mode = mode;
5516 if (FLOAT_MODE_P (mode))
5517 this_insn_cc0 = gen_rtx_COMPARE (VOIDmode, this_insn_cc0,
5518 CONST0_RTX (mode));
5520 #endif
5523 /* Now enter all non-volatile source expressions in the hash table
5524 if they are not already present.
5525 Record their equivalence classes in src_elt.
5526 This way we can insert the corresponding destinations into
5527 the same classes even if the actual sources are no longer in them
5528 (having been invalidated). */
5530 if (src_eqv && src_eqv_elt == 0 && sets[0].rtl != 0 && ! src_eqv_volatile
5531 && ! rtx_equal_p (src_eqv, SET_DEST (sets[0].rtl)))
5533 struct table_elt *elt;
5534 struct table_elt *classp = sets[0].src_elt;
5535 rtx dest = SET_DEST (sets[0].rtl);
5536 enum machine_mode eqvmode = GET_MODE (dest);
5538 if (GET_CODE (dest) == STRICT_LOW_PART)
5540 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
5541 classp = 0;
5543 if (insert_regs (src_eqv, classp, 0))
5545 rehash_using_reg (src_eqv);
5546 src_eqv_hash = HASH (src_eqv, eqvmode);
5548 elt = insert (src_eqv, classp, src_eqv_hash, eqvmode);
5549 elt->in_memory = src_eqv_in_memory;
5550 src_eqv_elt = elt;
5552 /* Check to see if src_eqv_elt is the same as a set source which
5553 does not yet have an elt, and if so set the elt of the set source
5554 to src_eqv_elt. */
5555 for (i = 0; i < n_sets; i++)
5556 if (sets[i].rtl && sets[i].src_elt == 0
5557 && rtx_equal_p (SET_SRC (sets[i].rtl), src_eqv))
5558 sets[i].src_elt = src_eqv_elt;
5561 for (i = 0; i < n_sets; i++)
5562 if (sets[i].rtl && ! sets[i].src_volatile
5563 && ! rtx_equal_p (SET_SRC (sets[i].rtl), SET_DEST (sets[i].rtl)))
5565 if (GET_CODE (SET_DEST (sets[i].rtl)) == STRICT_LOW_PART)
5567 /* REG_EQUAL in setting a STRICT_LOW_PART
5568 gives an equivalent for the entire destination register,
5569 not just for the subreg being stored in now.
5570 This is a more interesting equivalence, so we arrange later
5571 to treat the entire reg as the destination. */
5572 sets[i].src_elt = src_eqv_elt;
5573 sets[i].src_hash = src_eqv_hash;
5575 else
5577 /* Insert source and constant equivalent into hash table, if not
5578 already present. */
5579 struct table_elt *classp = src_eqv_elt;
5580 rtx src = sets[i].src;
5581 rtx dest = SET_DEST (sets[i].rtl);
5582 enum machine_mode mode
5583 = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
5585 /* It's possible that we have a source value known to be
5586 constant but don't have a REG_EQUAL note on the insn.
5587 Lack of a note will mean src_eqv_elt will be NULL. This
5588 can happen where we've generated a SUBREG to access a
5589 CONST_INT that is already in a register in a wider mode.
5590 Ensure that the source expression is put in the proper
5591 constant class. */
5592 if (!classp)
5593 classp = sets[i].src_const_elt;
5595 if (sets[i].src_elt == 0)
5597 struct table_elt *elt;
5599 /* Note that these insert_regs calls cannot remove
5600 any of the src_elt's, because they would have failed to
5601 match if not still valid. */
5602 if (insert_regs (src, classp, 0))
5604 rehash_using_reg (src);
5605 sets[i].src_hash = HASH (src, mode);
5607 elt = insert (src, classp, sets[i].src_hash, mode);
5608 elt->in_memory = sets[i].src_in_memory;
5609 sets[i].src_elt = classp = elt;
5611 if (sets[i].src_const && sets[i].src_const_elt == 0
5612 && src != sets[i].src_const
5613 && ! rtx_equal_p (sets[i].src_const, src))
5614 sets[i].src_elt = insert (sets[i].src_const, classp,
5615 sets[i].src_const_hash, mode);
5618 else if (sets[i].src_elt == 0)
5619 /* If we did not insert the source into the hash table (e.g., it was
5620 volatile), note the equivalence class for the REG_EQUAL value, if any,
5621 so that the destination goes into that class. */
5622 sets[i].src_elt = src_eqv_elt;
5624 /* Record destination addresses in the hash table. This allows us to
5625 check if they are invalidated by other sets. */
5626 for (i = 0; i < n_sets; i++)
5628 if (sets[i].rtl)
5630 rtx x = sets[i].inner_dest;
5631 struct table_elt *elt;
5632 enum machine_mode mode;
5633 unsigned hash;
5635 if (MEM_P (x))
5637 x = XEXP (x, 0);
5638 mode = GET_MODE (x);
5639 hash = HASH (x, mode);
5640 elt = lookup (x, hash, mode);
5641 if (!elt)
5643 if (insert_regs (x, NULL, 0))
5645 rtx dest = SET_DEST (sets[i].rtl);
5647 rehash_using_reg (x);
5648 hash = HASH (x, mode);
5649 sets[i].dest_hash = HASH (dest, GET_MODE (dest));
5651 elt = insert (x, NULL, hash, mode);
5654 sets[i].dest_addr_elt = elt;
5656 else
5657 sets[i].dest_addr_elt = NULL;
5661 invalidate_from_clobbers (insn);
5663 /* Some registers are invalidated by subroutine calls. Memory is
5664 invalidated by non-constant calls. */
5666 if (CALL_P (insn))
5668 if (!(RTL_CONST_OR_PURE_CALL_P (insn)))
5669 invalidate_memory ();
5670 invalidate_for_call ();
5673 /* Now invalidate everything set by this instruction.
5674 If a SUBREG or other funny destination is being set,
5675 sets[i].rtl is still nonzero, so here we invalidate the reg
5676 a part of which is being set. */
5678 for (i = 0; i < n_sets; i++)
5679 if (sets[i].rtl)
5681 /* We can't use the inner dest, because the mode associated with
5682 a ZERO_EXTRACT is significant. */
5683 rtx dest = SET_DEST (sets[i].rtl);
5685 /* Needed for registers to remove the register from its
5686 previous quantity's chain.
5687 Needed for memory if this is a nonvarying address, unless
5688 we have just done an invalidate_memory that covers even those. */
5689 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5690 invalidate (dest, VOIDmode);
5691 else if (MEM_P (dest))
5692 invalidate (dest, VOIDmode);
5693 else if (GET_CODE (dest) == STRICT_LOW_PART
5694 || GET_CODE (dest) == ZERO_EXTRACT)
5695 invalidate (XEXP (dest, 0), GET_MODE (dest));
5698 /* Don't cse over a call to setjmp; on some machines (eg VAX)
5699 the regs restored by the longjmp come from a later time
5700 than the setjmp. */
5701 if (CALL_P (insn) && find_reg_note (insn, REG_SETJMP, NULL))
5703 flush_hash_table ();
5704 goto done;
5707 /* Make sure registers mentioned in destinations
5708 are safe for use in an expression to be inserted.
5709 This removes from the hash table
5710 any invalid entry that refers to one of these registers.
5712 We don't care about the return value from mention_regs because
5713 we are going to hash the SET_DEST values unconditionally. */
5715 for (i = 0; i < n_sets; i++)
5717 if (sets[i].rtl)
5719 rtx x = SET_DEST (sets[i].rtl);
5721 if (!REG_P (x))
5722 mention_regs (x);
5723 else
5725 /* We used to rely on all references to a register becoming
5726 inaccessible when a register changes to a new quantity,
5727 since that changes the hash code. However, that is not
5728 safe, since after HASH_SIZE new quantities we get a
5729 hash 'collision' of a register with its own invalid
5730 entries. And since SUBREGs have been changed not to
5731 change their hash code with the hash code of the register,
5732 it wouldn't work any longer at all. So we have to check
5733 for any invalid references lying around now.
5734 This code is similar to the REG case in mention_regs,
5735 but it knows that reg_tick has been incremented, and
5736 it leaves reg_in_table as -1 . */
5737 unsigned int regno = REGNO (x);
5738 unsigned int endregno = END_REGNO (x);
5739 unsigned int i;
5741 for (i = regno; i < endregno; i++)
5743 if (REG_IN_TABLE (i) >= 0)
5745 remove_invalid_refs (i);
5746 REG_IN_TABLE (i) = -1;
5753 /* We may have just removed some of the src_elt's from the hash table.
5754 So replace each one with the current head of the same class.
5755 Also check if destination addresses have been removed. */
5757 for (i = 0; i < n_sets; i++)
5758 if (sets[i].rtl)
5760 if (sets[i].dest_addr_elt
5761 && sets[i].dest_addr_elt->first_same_value == 0)
5763 /* The elt was removed, which means this destination is not
5764 valid after this instruction. */
5765 sets[i].rtl = NULL_RTX;
5767 else if (sets[i].src_elt && sets[i].src_elt->first_same_value == 0)
5768 /* If elt was removed, find current head of same class,
5769 or 0 if nothing remains of that class. */
5771 struct table_elt *elt = sets[i].src_elt;
5773 while (elt && elt->prev_same_value)
5774 elt = elt->prev_same_value;
5776 while (elt && elt->first_same_value == 0)
5777 elt = elt->next_same_value;
5778 sets[i].src_elt = elt ? elt->first_same_value : 0;
5782 /* Now insert the destinations into their equivalence classes. */
5784 for (i = 0; i < n_sets; i++)
5785 if (sets[i].rtl)
5787 rtx dest = SET_DEST (sets[i].rtl);
5788 struct table_elt *elt;
5790 /* Don't record value if we are not supposed to risk allocating
5791 floating-point values in registers that might be wider than
5792 memory. */
5793 if ((flag_float_store
5794 && MEM_P (dest)
5795 && FLOAT_MODE_P (GET_MODE (dest)))
5796 /* Don't record BLKmode values, because we don't know the
5797 size of it, and can't be sure that other BLKmode values
5798 have the same or smaller size. */
5799 || GET_MODE (dest) == BLKmode
5800 /* If we didn't put a REG_EQUAL value or a source into the hash
5801 table, there is no point is recording DEST. */
5802 || sets[i].src_elt == 0
5803 /* If DEST is a paradoxical SUBREG and SRC is a ZERO_EXTEND
5804 or SIGN_EXTEND, don't record DEST since it can cause
5805 some tracking to be wrong.
5807 ??? Think about this more later. */
5808 || (paradoxical_subreg_p (dest)
5809 && (GET_CODE (sets[i].src) == SIGN_EXTEND
5810 || GET_CODE (sets[i].src) == ZERO_EXTEND)))
5811 continue;
5813 /* STRICT_LOW_PART isn't part of the value BEING set,
5814 and neither is the SUBREG inside it.
5815 Note that in this case SETS[I].SRC_ELT is really SRC_EQV_ELT. */
5816 if (GET_CODE (dest) == STRICT_LOW_PART)
5817 dest = SUBREG_REG (XEXP (dest, 0));
5819 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5820 /* Registers must also be inserted into chains for quantities. */
5821 if (insert_regs (dest, sets[i].src_elt, 1))
5823 /* If `insert_regs' changes something, the hash code must be
5824 recalculated. */
5825 rehash_using_reg (dest);
5826 sets[i].dest_hash = HASH (dest, GET_MODE (dest));
5829 elt = insert (dest, sets[i].src_elt,
5830 sets[i].dest_hash, GET_MODE (dest));
5832 /* If this is a constant, insert the constant anchors with the
5833 equivalent register-offset expressions using register DEST. */
5834 if (targetm.const_anchor
5835 && REG_P (dest)
5836 && SCALAR_INT_MODE_P (GET_MODE (dest))
5837 && GET_CODE (sets[i].src_elt->exp) == CONST_INT)
5838 insert_const_anchors (dest, sets[i].src_elt->exp, GET_MODE (dest));
5840 elt->in_memory = (MEM_P (sets[i].inner_dest)
5841 && !MEM_READONLY_P (sets[i].inner_dest));
5843 /* If we have (set (subreg:m1 (reg:m2 foo) 0) (bar:m1)), M1 is no
5844 narrower than M2, and both M1 and M2 are the same number of words,
5845 we are also doing (set (reg:m2 foo) (subreg:m2 (bar:m1) 0)) so
5846 make that equivalence as well.
5848 However, BAR may have equivalences for which gen_lowpart
5849 will produce a simpler value than gen_lowpart applied to
5850 BAR (e.g., if BAR was ZERO_EXTENDed from M2), so we will scan all
5851 BAR's equivalences. If we don't get a simplified form, make
5852 the SUBREG. It will not be used in an equivalence, but will
5853 cause two similar assignments to be detected.
5855 Note the loop below will find SUBREG_REG (DEST) since we have
5856 already entered SRC and DEST of the SET in the table. */
5858 if (GET_CODE (dest) == SUBREG
5859 && (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))) - 1)
5860 / UNITS_PER_WORD)
5861 == (GET_MODE_SIZE (GET_MODE (dest)) - 1) / UNITS_PER_WORD)
5862 && (GET_MODE_SIZE (GET_MODE (dest))
5863 >= GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
5864 && sets[i].src_elt != 0)
5866 enum machine_mode new_mode = GET_MODE (SUBREG_REG (dest));
5867 struct table_elt *elt, *classp = 0;
5869 for (elt = sets[i].src_elt->first_same_value; elt;
5870 elt = elt->next_same_value)
5872 rtx new_src = 0;
5873 unsigned src_hash;
5874 struct table_elt *src_elt;
5875 int byte = 0;
5877 /* Ignore invalid entries. */
5878 if (!REG_P (elt->exp)
5879 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
5880 continue;
5882 /* We may have already been playing subreg games. If the
5883 mode is already correct for the destination, use it. */
5884 if (GET_MODE (elt->exp) == new_mode)
5885 new_src = elt->exp;
5886 else
5888 /* Calculate big endian correction for the SUBREG_BYTE.
5889 We have already checked that M1 (GET_MODE (dest))
5890 is not narrower than M2 (new_mode). */
5891 if (BYTES_BIG_ENDIAN)
5892 byte = (GET_MODE_SIZE (GET_MODE (dest))
5893 - GET_MODE_SIZE (new_mode));
5895 new_src = simplify_gen_subreg (new_mode, elt->exp,
5896 GET_MODE (dest), byte);
5899 /* The call to simplify_gen_subreg fails if the value
5900 is VOIDmode, yet we can't do any simplification, e.g.
5901 for EXPR_LISTs denoting function call results.
5902 It is invalid to construct a SUBREG with a VOIDmode
5903 SUBREG_REG, hence a zero new_src means we can't do
5904 this substitution. */
5905 if (! new_src)
5906 continue;
5908 src_hash = HASH (new_src, new_mode);
5909 src_elt = lookup (new_src, src_hash, new_mode);
5911 /* Put the new source in the hash table is if isn't
5912 already. */
5913 if (src_elt == 0)
5915 if (insert_regs (new_src, classp, 0))
5917 rehash_using_reg (new_src);
5918 src_hash = HASH (new_src, new_mode);
5920 src_elt = insert (new_src, classp, src_hash, new_mode);
5921 src_elt->in_memory = elt->in_memory;
5923 else if (classp && classp != src_elt->first_same_value)
5924 /* Show that two things that we've seen before are
5925 actually the same. */
5926 merge_equiv_classes (src_elt, classp);
5928 classp = src_elt->first_same_value;
5929 /* Ignore invalid entries. */
5930 while (classp
5931 && !REG_P (classp->exp)
5932 && ! exp_equiv_p (classp->exp, classp->exp, 1, false))
5933 classp = classp->next_same_value;
5938 /* Special handling for (set REG0 REG1) where REG0 is the
5939 "cheapest", cheaper than REG1. After cse, REG1 will probably not
5940 be used in the sequel, so (if easily done) change this insn to
5941 (set REG1 REG0) and replace REG1 with REG0 in the previous insn
5942 that computed their value. Then REG1 will become a dead store
5943 and won't cloud the situation for later optimizations.
5945 Do not make this change if REG1 is a hard register, because it will
5946 then be used in the sequel and we may be changing a two-operand insn
5947 into a three-operand insn.
5949 Also do not do this if we are operating on a copy of INSN. */
5951 if (n_sets == 1 && sets[0].rtl)
5952 try_back_substitute_reg (sets[0].rtl, insn);
5954 done:;
5957 /* Remove from the hash table all expressions that reference memory. */
5959 static void
5960 invalidate_memory (void)
5962 int i;
5963 struct table_elt *p, *next;
5965 for (i = 0; i < HASH_SIZE; i++)
5966 for (p = table[i]; p; p = next)
5968 next = p->next_same_hash;
5969 if (p->in_memory)
5970 remove_from_table (p, i);
5974 /* Perform invalidation on the basis of everything about INSN,
5975 except for invalidating the actual places that are SET in it.
5976 This includes the places CLOBBERed, and anything that might
5977 alias with something that is SET or CLOBBERed. */
5979 static void
5980 invalidate_from_clobbers (rtx insn)
5982 rtx x = PATTERN (insn);
5984 if (GET_CODE (x) == CLOBBER)
5986 rtx ref = XEXP (x, 0);
5987 if (ref)
5989 if (REG_P (ref) || GET_CODE (ref) == SUBREG
5990 || MEM_P (ref))
5991 invalidate (ref, VOIDmode);
5992 else if (GET_CODE (ref) == STRICT_LOW_PART
5993 || GET_CODE (ref) == ZERO_EXTRACT)
5994 invalidate (XEXP (ref, 0), GET_MODE (ref));
5997 else if (GET_CODE (x) == PARALLEL)
5999 int i;
6000 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
6002 rtx y = XVECEXP (x, 0, i);
6003 if (GET_CODE (y) == CLOBBER)
6005 rtx ref = XEXP (y, 0);
6006 if (REG_P (ref) || GET_CODE (ref) == SUBREG
6007 || MEM_P (ref))
6008 invalidate (ref, VOIDmode);
6009 else if (GET_CODE (ref) == STRICT_LOW_PART
6010 || GET_CODE (ref) == ZERO_EXTRACT)
6011 invalidate (XEXP (ref, 0), GET_MODE (ref));
6017 /* Perform invalidation on the basis of everything about INSN.
6018 This includes the places CLOBBERed, and anything that might
6019 alias with something that is SET or CLOBBERed. */
6021 static void
6022 invalidate_from_sets_and_clobbers (rtx insn)
6024 rtx tem;
6025 rtx x = PATTERN (insn);
6027 if (CALL_P (insn))
6029 for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
6030 if (GET_CODE (XEXP (tem, 0)) == CLOBBER)
6031 invalidate (SET_DEST (XEXP (tem, 0)), VOIDmode);
6034 /* Ensure we invalidate the destination register of a CALL insn.
6035 This is necessary for machines where this register is a fixed_reg,
6036 because no other code would invalidate it. */
6037 if (GET_CODE (x) == SET && GET_CODE (SET_SRC (x)) == CALL)
6038 invalidate (SET_DEST (x), VOIDmode);
6040 else if (GET_CODE (x) == PARALLEL)
6042 int i;
6044 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
6046 rtx y = XVECEXP (x, 0, i);
6047 if (GET_CODE (y) == CLOBBER)
6049 rtx clobbered = XEXP (y, 0);
6051 if (REG_P (clobbered)
6052 || GET_CODE (clobbered) == SUBREG)
6053 invalidate (clobbered, VOIDmode);
6054 else if (GET_CODE (clobbered) == STRICT_LOW_PART
6055 || GET_CODE (clobbered) == ZERO_EXTRACT)
6056 invalidate (XEXP (clobbered, 0), GET_MODE (clobbered));
6058 else if (GET_CODE (y) == SET && GET_CODE (SET_SRC (y)) == CALL)
6059 invalidate (SET_DEST (y), VOIDmode);
6064 /* Process X, part of the REG_NOTES of an insn. Look at any REG_EQUAL notes
6065 and replace any registers in them with either an equivalent constant
6066 or the canonical form of the register. If we are inside an address,
6067 only do this if the address remains valid.
6069 OBJECT is 0 except when within a MEM in which case it is the MEM.
6071 Return the replacement for X. */
6073 static rtx
6074 cse_process_notes_1 (rtx x, rtx object, bool *changed)
6076 enum rtx_code code = GET_CODE (x);
6077 const char *fmt = GET_RTX_FORMAT (code);
6078 int i;
6080 switch (code)
6082 case CONST:
6083 case SYMBOL_REF:
6084 case LABEL_REF:
6085 CASE_CONST_ANY:
6086 case PC:
6087 case CC0:
6088 case LO_SUM:
6089 return x;
6091 case MEM:
6092 validate_change (x, &XEXP (x, 0),
6093 cse_process_notes (XEXP (x, 0), x, changed), 0);
6094 return x;
6096 case EXPR_LIST:
6097 if (REG_NOTE_KIND (x) == REG_EQUAL)
6098 XEXP (x, 0) = cse_process_notes (XEXP (x, 0), NULL_RTX, changed);
6099 /* Fall through. */
6101 case INSN_LIST:
6102 case INT_LIST:
6103 if (XEXP (x, 1))
6104 XEXP (x, 1) = cse_process_notes (XEXP (x, 1), NULL_RTX, changed);
6105 return x;
6107 case SIGN_EXTEND:
6108 case ZERO_EXTEND:
6109 case SUBREG:
6111 rtx new_rtx = cse_process_notes (XEXP (x, 0), object, changed);
6112 /* We don't substitute VOIDmode constants into these rtx,
6113 since they would impede folding. */
6114 if (GET_MODE (new_rtx) != VOIDmode)
6115 validate_change (object, &XEXP (x, 0), new_rtx, 0);
6116 return x;
6119 case UNSIGNED_FLOAT:
6121 rtx new_rtx = cse_process_notes (XEXP (x, 0), object, changed);
6122 /* We don't substitute negative VOIDmode constants into these rtx,
6123 since they would impede folding. */
6124 if (GET_MODE (new_rtx) != VOIDmode
6125 || (CONST_INT_P (new_rtx) && INTVAL (new_rtx) >= 0)
6126 || (CONST_DOUBLE_P (new_rtx) && CONST_DOUBLE_HIGH (new_rtx) >= 0))
6127 validate_change (object, &XEXP (x, 0), new_rtx, 0);
6128 return x;
6131 case REG:
6132 i = REG_QTY (REGNO (x));
6134 /* Return a constant or a constant register. */
6135 if (REGNO_QTY_VALID_P (REGNO (x)))
6137 struct qty_table_elem *ent = &qty_table[i];
6139 if (ent->const_rtx != NULL_RTX
6140 && (CONSTANT_P (ent->const_rtx)
6141 || REG_P (ent->const_rtx)))
6143 rtx new_rtx = gen_lowpart (GET_MODE (x), ent->const_rtx);
6144 if (new_rtx)
6145 return copy_rtx (new_rtx);
6149 /* Otherwise, canonicalize this register. */
6150 return canon_reg (x, NULL_RTX);
6152 default:
6153 break;
6156 for (i = 0; i < GET_RTX_LENGTH (code); i++)
6157 if (fmt[i] == 'e')
6158 validate_change (object, &XEXP (x, i),
6159 cse_process_notes (XEXP (x, i), object, changed), 0);
6161 return x;
6164 static rtx
6165 cse_process_notes (rtx x, rtx object, bool *changed)
6167 rtx new_rtx = cse_process_notes_1 (x, object, changed);
6168 if (new_rtx != x)
6169 *changed = true;
6170 return new_rtx;
6174 /* Find a path in the CFG, starting with FIRST_BB to perform CSE on.
6176 DATA is a pointer to a struct cse_basic_block_data, that is used to
6177 describe the path.
6178 It is filled with a queue of basic blocks, starting with FIRST_BB
6179 and following a trace through the CFG.
6181 If all paths starting at FIRST_BB have been followed, or no new path
6182 starting at FIRST_BB can be constructed, this function returns FALSE.
6183 Otherwise, DATA->path is filled and the function returns TRUE indicating
6184 that a path to follow was found.
6186 If FOLLOW_JUMPS is false, the maximum path length is 1 and the only
6187 block in the path will be FIRST_BB. */
6189 static bool
6190 cse_find_path (basic_block first_bb, struct cse_basic_block_data *data,
6191 int follow_jumps)
6193 basic_block bb;
6194 edge e;
6195 int path_size;
6197 bitmap_set_bit (cse_visited_basic_blocks, first_bb->index);
6199 /* See if there is a previous path. */
6200 path_size = data->path_size;
6202 /* There is a previous path. Make sure it started with FIRST_BB. */
6203 if (path_size)
6204 gcc_assert (data->path[0].bb == first_bb);
6206 /* There was only one basic block in the last path. Clear the path and
6207 return, so that paths starting at another basic block can be tried. */
6208 if (path_size == 1)
6210 path_size = 0;
6211 goto done;
6214 /* If the path was empty from the beginning, construct a new path. */
6215 if (path_size == 0)
6216 data->path[path_size++].bb = first_bb;
6217 else
6219 /* Otherwise, path_size must be equal to or greater than 2, because
6220 a previous path exists that is at least two basic blocks long.
6222 Update the previous branch path, if any. If the last branch was
6223 previously along the branch edge, take the fallthrough edge now. */
6224 while (path_size >= 2)
6226 basic_block last_bb_in_path, previous_bb_in_path;
6227 edge e;
6229 --path_size;
6230 last_bb_in_path = data->path[path_size].bb;
6231 previous_bb_in_path = data->path[path_size - 1].bb;
6233 /* If we previously followed a path along the branch edge, try
6234 the fallthru edge now. */
6235 if (EDGE_COUNT (previous_bb_in_path->succs) == 2
6236 && any_condjump_p (BB_END (previous_bb_in_path))
6237 && (e = find_edge (previous_bb_in_path, last_bb_in_path))
6238 && e == BRANCH_EDGE (previous_bb_in_path))
6240 bb = FALLTHRU_EDGE (previous_bb_in_path)->dest;
6241 if (bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
6242 && single_pred_p (bb)
6243 /* We used to assert here that we would only see blocks
6244 that we have not visited yet. But we may end up
6245 visiting basic blocks twice if the CFG has changed
6246 in this run of cse_main, because when the CFG changes
6247 the topological sort of the CFG also changes. A basic
6248 blocks that previously had more than two predecessors
6249 may now have a single predecessor, and become part of
6250 a path that starts at another basic block.
6252 We still want to visit each basic block only once, so
6253 halt the path here if we have already visited BB. */
6254 && !bitmap_bit_p (cse_visited_basic_blocks, bb->index))
6256 bitmap_set_bit (cse_visited_basic_blocks, bb->index);
6257 data->path[path_size++].bb = bb;
6258 break;
6262 data->path[path_size].bb = NULL;
6265 /* If only one block remains in the path, bail. */
6266 if (path_size == 1)
6268 path_size = 0;
6269 goto done;
6273 /* Extend the path if possible. */
6274 if (follow_jumps)
6276 bb = data->path[path_size - 1].bb;
6277 while (bb && path_size < PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH))
6279 if (single_succ_p (bb))
6280 e = single_succ_edge (bb);
6281 else if (EDGE_COUNT (bb->succs) == 2
6282 && any_condjump_p (BB_END (bb)))
6284 /* First try to follow the branch. If that doesn't lead
6285 to a useful path, follow the fallthru edge. */
6286 e = BRANCH_EDGE (bb);
6287 if (!single_pred_p (e->dest))
6288 e = FALLTHRU_EDGE (bb);
6290 else
6291 e = NULL;
6293 if (e
6294 && !((e->flags & EDGE_ABNORMAL_CALL) && cfun->has_nonlocal_label)
6295 && e->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
6296 && single_pred_p (e->dest)
6297 /* Avoid visiting basic blocks twice. The large comment
6298 above explains why this can happen. */
6299 && !bitmap_bit_p (cse_visited_basic_blocks, e->dest->index))
6301 basic_block bb2 = e->dest;
6302 bitmap_set_bit (cse_visited_basic_blocks, bb2->index);
6303 data->path[path_size++].bb = bb2;
6304 bb = bb2;
6306 else
6307 bb = NULL;
6311 done:
6312 data->path_size = path_size;
6313 return path_size != 0;
6316 /* Dump the path in DATA to file F. NSETS is the number of sets
6317 in the path. */
6319 static void
6320 cse_dump_path (struct cse_basic_block_data *data, int nsets, FILE *f)
6322 int path_entry;
6324 fprintf (f, ";; Following path with %d sets: ", nsets);
6325 for (path_entry = 0; path_entry < data->path_size; path_entry++)
6326 fprintf (f, "%d ", (data->path[path_entry].bb)->index);
6327 fputc ('\n', dump_file);
6328 fflush (f);
6332 /* Return true if BB has exception handling successor edges. */
6334 static bool
6335 have_eh_succ_edges (basic_block bb)
6337 edge e;
6338 edge_iterator ei;
6340 FOR_EACH_EDGE (e, ei, bb->succs)
6341 if (e->flags & EDGE_EH)
6342 return true;
6344 return false;
6348 /* Scan to the end of the path described by DATA. Return an estimate of
6349 the total number of SETs of all insns in the path. */
6351 static void
6352 cse_prescan_path (struct cse_basic_block_data *data)
6354 int nsets = 0;
6355 int path_size = data->path_size;
6356 int path_entry;
6358 /* Scan to end of each basic block in the path. */
6359 for (path_entry = 0; path_entry < path_size; path_entry++)
6361 basic_block bb;
6362 rtx insn;
6364 bb = data->path[path_entry].bb;
6366 FOR_BB_INSNS (bb, insn)
6368 if (!INSN_P (insn))
6369 continue;
6371 /* A PARALLEL can have lots of SETs in it,
6372 especially if it is really an ASM_OPERANDS. */
6373 if (GET_CODE (PATTERN (insn)) == PARALLEL)
6374 nsets += XVECLEN (PATTERN (insn), 0);
6375 else
6376 nsets += 1;
6380 data->nsets = nsets;
6383 /* Process a single extended basic block described by EBB_DATA. */
6385 static void
6386 cse_extended_basic_block (struct cse_basic_block_data *ebb_data)
6388 int path_size = ebb_data->path_size;
6389 int path_entry;
6390 int num_insns = 0;
6392 /* Allocate the space needed by qty_table. */
6393 qty_table = XNEWVEC (struct qty_table_elem, max_qty);
6395 new_basic_block ();
6396 cse_ebb_live_in = df_get_live_in (ebb_data->path[0].bb);
6397 cse_ebb_live_out = df_get_live_out (ebb_data->path[path_size - 1].bb);
6398 for (path_entry = 0; path_entry < path_size; path_entry++)
6400 basic_block bb;
6401 rtx insn;
6403 bb = ebb_data->path[path_entry].bb;
6405 /* Invalidate recorded information for eh regs if there is an EH
6406 edge pointing to that bb. */
6407 if (bb_has_eh_pred (bb))
6409 df_ref *def_rec;
6411 for (def_rec = df_get_artificial_defs (bb->index); *def_rec; def_rec++)
6413 df_ref def = *def_rec;
6414 if (DF_REF_FLAGS (def) & DF_REF_AT_TOP)
6415 invalidate (DF_REF_REG (def), GET_MODE (DF_REF_REG (def)));
6419 optimize_this_for_speed_p = optimize_bb_for_speed_p (bb);
6420 FOR_BB_INSNS (bb, insn)
6422 /* If we have processed 1,000 insns, flush the hash table to
6423 avoid extreme quadratic behavior. We must not include NOTEs
6424 in the count since there may be more of them when generating
6425 debugging information. If we clear the table at different
6426 times, code generated with -g -O might be different than code
6427 generated with -O but not -g.
6429 FIXME: This is a real kludge and needs to be done some other
6430 way. */
6431 if (NONDEBUG_INSN_P (insn)
6432 && num_insns++ > PARAM_VALUE (PARAM_MAX_CSE_INSNS))
6434 flush_hash_table ();
6435 num_insns = 0;
6438 if (INSN_P (insn))
6440 /* Process notes first so we have all notes in canonical forms
6441 when looking for duplicate operations. */
6442 if (REG_NOTES (insn))
6444 bool changed = false;
6445 REG_NOTES (insn) = cse_process_notes (REG_NOTES (insn),
6446 NULL_RTX, &changed);
6447 if (changed)
6448 df_notes_rescan (insn);
6451 cse_insn (insn);
6453 /* If we haven't already found an insn where we added a LABEL_REF,
6454 check this one. */
6455 if (INSN_P (insn) && !recorded_label_ref
6456 && for_each_rtx (&PATTERN (insn), check_for_label_ref,
6457 (void *) insn))
6458 recorded_label_ref = true;
6460 #ifdef HAVE_cc0
6461 if (NONDEBUG_INSN_P (insn))
6463 /* If the previous insn sets CC0 and this insn no
6464 longer references CC0, delete the previous insn.
6465 Here we use fact that nothing expects CC0 to be
6466 valid over an insn, which is true until the final
6467 pass. */
6468 rtx prev_insn, tem;
6470 prev_insn = prev_nonnote_nondebug_insn (insn);
6471 if (prev_insn && NONJUMP_INSN_P (prev_insn)
6472 && (tem = single_set (prev_insn)) != NULL_RTX
6473 && SET_DEST (tem) == cc0_rtx
6474 && ! reg_mentioned_p (cc0_rtx, PATTERN (insn)))
6475 delete_insn (prev_insn);
6477 /* If this insn is not the last insn in the basic
6478 block, it will be PREV_INSN(insn) in the next
6479 iteration. If we recorded any CC0-related
6480 information for this insn, remember it. */
6481 if (insn != BB_END (bb))
6483 prev_insn_cc0 = this_insn_cc0;
6484 prev_insn_cc0_mode = this_insn_cc0_mode;
6487 #endif
6491 /* With non-call exceptions, we are not always able to update
6492 the CFG properly inside cse_insn. So clean up possibly
6493 redundant EH edges here. */
6494 if (cfun->can_throw_non_call_exceptions && have_eh_succ_edges (bb))
6495 cse_cfg_altered |= purge_dead_edges (bb);
6497 /* If we changed a conditional jump, we may have terminated
6498 the path we are following. Check that by verifying that
6499 the edge we would take still exists. If the edge does
6500 not exist anymore, purge the remainder of the path.
6501 Note that this will cause us to return to the caller. */
6502 if (path_entry < path_size - 1)
6504 basic_block next_bb = ebb_data->path[path_entry + 1].bb;
6505 if (!find_edge (bb, next_bb))
6509 path_size--;
6511 /* If we truncate the path, we must also reset the
6512 visited bit on the remaining blocks in the path,
6513 or we will never visit them at all. */
6514 bitmap_clear_bit (cse_visited_basic_blocks,
6515 ebb_data->path[path_size].bb->index);
6516 ebb_data->path[path_size].bb = NULL;
6518 while (path_size - 1 != path_entry);
6519 ebb_data->path_size = path_size;
6523 /* If this is a conditional jump insn, record any known
6524 equivalences due to the condition being tested. */
6525 insn = BB_END (bb);
6526 if (path_entry < path_size - 1
6527 && JUMP_P (insn)
6528 && single_set (insn)
6529 && any_condjump_p (insn))
6531 basic_block next_bb = ebb_data->path[path_entry + 1].bb;
6532 bool taken = (next_bb == BRANCH_EDGE (bb)->dest);
6533 record_jump_equiv (insn, taken);
6536 #ifdef HAVE_cc0
6537 /* Clear the CC0-tracking related insns, they can't provide
6538 useful information across basic block boundaries. */
6539 prev_insn_cc0 = 0;
6540 #endif
6543 gcc_assert (next_qty <= max_qty);
6545 free (qty_table);
6549 /* Perform cse on the instructions of a function.
6550 F is the first instruction.
6551 NREGS is one plus the highest pseudo-reg number used in the instruction.
6553 Return 2 if jump optimizations should be redone due to simplifications
6554 in conditional jump instructions.
6555 Return 1 if the CFG should be cleaned up because it has been modified.
6556 Return 0 otherwise. */
6558 static int
6559 cse_main (rtx f ATTRIBUTE_UNUSED, int nregs)
6561 struct cse_basic_block_data ebb_data;
6562 basic_block bb;
6563 int *rc_order = XNEWVEC (int, last_basic_block_for_fn (cfun));
6564 int i, n_blocks;
6566 df_set_flags (DF_LR_RUN_DCE);
6567 df_note_add_problem ();
6568 df_analyze ();
6569 df_set_flags (DF_DEFER_INSN_RESCAN);
6571 reg_scan (get_insns (), max_reg_num ());
6572 init_cse_reg_info (nregs);
6574 ebb_data.path = XNEWVEC (struct branch_path,
6575 PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH));
6577 cse_cfg_altered = false;
6578 cse_jumps_altered = false;
6579 recorded_label_ref = false;
6580 constant_pool_entries_cost = 0;
6581 constant_pool_entries_regcost = 0;
6582 ebb_data.path_size = 0;
6583 ebb_data.nsets = 0;
6584 rtl_hooks = cse_rtl_hooks;
6586 init_recog ();
6587 init_alias_analysis ();
6589 reg_eqv_table = XNEWVEC (struct reg_eqv_elem, nregs);
6591 /* Set up the table of already visited basic blocks. */
6592 cse_visited_basic_blocks = sbitmap_alloc (last_basic_block_for_fn (cfun));
6593 bitmap_clear (cse_visited_basic_blocks);
6595 /* Loop over basic blocks in reverse completion order (RPO),
6596 excluding the ENTRY and EXIT blocks. */
6597 n_blocks = pre_and_rev_post_order_compute (NULL, rc_order, false);
6598 i = 0;
6599 while (i < n_blocks)
6601 /* Find the first block in the RPO queue that we have not yet
6602 processed before. */
6605 bb = BASIC_BLOCK_FOR_FN (cfun, rc_order[i++]);
6607 while (bitmap_bit_p (cse_visited_basic_blocks, bb->index)
6608 && i < n_blocks);
6610 /* Find all paths starting with BB, and process them. */
6611 while (cse_find_path (bb, &ebb_data, flag_cse_follow_jumps))
6613 /* Pre-scan the path. */
6614 cse_prescan_path (&ebb_data);
6616 /* If this basic block has no sets, skip it. */
6617 if (ebb_data.nsets == 0)
6618 continue;
6620 /* Get a reasonable estimate for the maximum number of qty's
6621 needed for this path. For this, we take the number of sets
6622 and multiply that by MAX_RECOG_OPERANDS. */
6623 max_qty = ebb_data.nsets * MAX_RECOG_OPERANDS;
6625 /* Dump the path we're about to process. */
6626 if (dump_file)
6627 cse_dump_path (&ebb_data, ebb_data.nsets, dump_file);
6629 cse_extended_basic_block (&ebb_data);
6633 /* Clean up. */
6634 end_alias_analysis ();
6635 free (reg_eqv_table);
6636 free (ebb_data.path);
6637 sbitmap_free (cse_visited_basic_blocks);
6638 free (rc_order);
6639 rtl_hooks = general_rtl_hooks;
6641 if (cse_jumps_altered || recorded_label_ref)
6642 return 2;
6643 else if (cse_cfg_altered)
6644 return 1;
6645 else
6646 return 0;
6649 /* Called via for_each_rtx to see if an insn is using a LABEL_REF for
6650 which there isn't a REG_LABEL_OPERAND note.
6651 Return one if so. DATA is the insn. */
6653 static int
6654 check_for_label_ref (rtx *rtl, void *data)
6656 rtx insn = (rtx) data;
6658 /* If this insn uses a LABEL_REF and there isn't a REG_LABEL_OPERAND
6659 note for it, we must rerun jump since it needs to place the note. If
6660 this is a LABEL_REF for a CODE_LABEL that isn't in the insn chain,
6661 don't do this since no REG_LABEL_OPERAND will be added. */
6662 return (GET_CODE (*rtl) == LABEL_REF
6663 && ! LABEL_REF_NONLOCAL_P (*rtl)
6664 && (!JUMP_P (insn)
6665 || !label_is_jump_target_p (XEXP (*rtl, 0), insn))
6666 && LABEL_P (XEXP (*rtl, 0))
6667 && INSN_UID (XEXP (*rtl, 0)) != 0
6668 && ! find_reg_note (insn, REG_LABEL_OPERAND, XEXP (*rtl, 0)));
6671 /* Count the number of times registers are used (not set) in X.
6672 COUNTS is an array in which we accumulate the count, INCR is how much
6673 we count each register usage.
6675 Don't count a usage of DEST, which is the SET_DEST of a SET which
6676 contains X in its SET_SRC. This is because such a SET does not
6677 modify the liveness of DEST.
6678 DEST is set to pc_rtx for a trapping insn, or for an insn with side effects.
6679 We must then count uses of a SET_DEST regardless, because the insn can't be
6680 deleted here. */
6682 static void
6683 count_reg_usage (rtx x, int *counts, rtx dest, int incr)
6685 enum rtx_code code;
6686 rtx note;
6687 const char *fmt;
6688 int i, j;
6690 if (x == 0)
6691 return;
6693 switch (code = GET_CODE (x))
6695 case REG:
6696 if (x != dest)
6697 counts[REGNO (x)] += incr;
6698 return;
6700 case PC:
6701 case CC0:
6702 case CONST:
6703 CASE_CONST_ANY:
6704 case SYMBOL_REF:
6705 case LABEL_REF:
6706 return;
6708 case CLOBBER:
6709 /* If we are clobbering a MEM, mark any registers inside the address
6710 as being used. */
6711 if (MEM_P (XEXP (x, 0)))
6712 count_reg_usage (XEXP (XEXP (x, 0), 0), counts, NULL_RTX, incr);
6713 return;
6715 case SET:
6716 /* Unless we are setting a REG, count everything in SET_DEST. */
6717 if (!REG_P (SET_DEST (x)))
6718 count_reg_usage (SET_DEST (x), counts, NULL_RTX, incr);
6719 count_reg_usage (SET_SRC (x), counts,
6720 dest ? dest : SET_DEST (x),
6721 incr);
6722 return;
6724 case DEBUG_INSN:
6725 return;
6727 case CALL_INSN:
6728 case INSN:
6729 case JUMP_INSN:
6730 /* We expect dest to be NULL_RTX here. If the insn may throw,
6731 or if it cannot be deleted due to side-effects, mark this fact
6732 by setting DEST to pc_rtx. */
6733 if ((!cfun->can_delete_dead_exceptions && !insn_nothrow_p (x))
6734 || side_effects_p (PATTERN (x)))
6735 dest = pc_rtx;
6736 if (code == CALL_INSN)
6737 count_reg_usage (CALL_INSN_FUNCTION_USAGE (x), counts, dest, incr);
6738 count_reg_usage (PATTERN (x), counts, dest, incr);
6740 /* Things used in a REG_EQUAL note aren't dead since loop may try to
6741 use them. */
6743 note = find_reg_equal_equiv_note (x);
6744 if (note)
6746 rtx eqv = XEXP (note, 0);
6748 if (GET_CODE (eqv) == EXPR_LIST)
6749 /* This REG_EQUAL note describes the result of a function call.
6750 Process all the arguments. */
6753 count_reg_usage (XEXP (eqv, 0), counts, dest, incr);
6754 eqv = XEXP (eqv, 1);
6756 while (eqv && GET_CODE (eqv) == EXPR_LIST);
6757 else
6758 count_reg_usage (eqv, counts, dest, incr);
6760 return;
6762 case EXPR_LIST:
6763 if (REG_NOTE_KIND (x) == REG_EQUAL
6764 || (REG_NOTE_KIND (x) != REG_NONNEG && GET_CODE (XEXP (x,0)) == USE)
6765 /* FUNCTION_USAGE expression lists may include (CLOBBER (mem /u)),
6766 involving registers in the address. */
6767 || GET_CODE (XEXP (x, 0)) == CLOBBER)
6768 count_reg_usage (XEXP (x, 0), counts, NULL_RTX, incr);
6770 count_reg_usage (XEXP (x, 1), counts, NULL_RTX, incr);
6771 return;
6773 case ASM_OPERANDS:
6774 /* Iterate over just the inputs, not the constraints as well. */
6775 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
6776 count_reg_usage (ASM_OPERANDS_INPUT (x, i), counts, dest, incr);
6777 return;
6779 case INSN_LIST:
6780 case INT_LIST:
6781 gcc_unreachable ();
6783 default:
6784 break;
6787 fmt = GET_RTX_FORMAT (code);
6788 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
6790 if (fmt[i] == 'e')
6791 count_reg_usage (XEXP (x, i), counts, dest, incr);
6792 else if (fmt[i] == 'E')
6793 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
6794 count_reg_usage (XVECEXP (x, i, j), counts, dest, incr);
6798 /* Return true if X is a dead register. */
6800 static inline int
6801 is_dead_reg (rtx x, int *counts)
6803 return (REG_P (x)
6804 && REGNO (x) >= FIRST_PSEUDO_REGISTER
6805 && counts[REGNO (x)] == 0);
6808 /* Return true if set is live. */
6809 static bool
6810 set_live_p (rtx set, rtx insn ATTRIBUTE_UNUSED, /* Only used with HAVE_cc0. */
6811 int *counts)
6813 #ifdef HAVE_cc0
6814 rtx tem;
6815 #endif
6817 if (set_noop_p (set))
6820 #ifdef HAVE_cc0
6821 else if (GET_CODE (SET_DEST (set)) == CC0
6822 && !side_effects_p (SET_SRC (set))
6823 && ((tem = next_nonnote_nondebug_insn (insn)) == NULL_RTX
6824 || !INSN_P (tem)
6825 || !reg_referenced_p (cc0_rtx, PATTERN (tem))))
6826 return false;
6827 #endif
6828 else if (!is_dead_reg (SET_DEST (set), counts)
6829 || side_effects_p (SET_SRC (set)))
6830 return true;
6831 return false;
6834 /* Return true if insn is live. */
6836 static bool
6837 insn_live_p (rtx insn, int *counts)
6839 int i;
6840 if (!cfun->can_delete_dead_exceptions && !insn_nothrow_p (insn))
6841 return true;
6842 else if (GET_CODE (PATTERN (insn)) == SET)
6843 return set_live_p (PATTERN (insn), insn, counts);
6844 else if (GET_CODE (PATTERN (insn)) == PARALLEL)
6846 for (i = XVECLEN (PATTERN (insn), 0) - 1; i >= 0; i--)
6848 rtx elt = XVECEXP (PATTERN (insn), 0, i);
6850 if (GET_CODE (elt) == SET)
6852 if (set_live_p (elt, insn, counts))
6853 return true;
6855 else if (GET_CODE (elt) != CLOBBER && GET_CODE (elt) != USE)
6856 return true;
6858 return false;
6860 else if (DEBUG_INSN_P (insn))
6862 rtx next;
6864 for (next = NEXT_INSN (insn); next; next = NEXT_INSN (next))
6865 if (NOTE_P (next))
6866 continue;
6867 else if (!DEBUG_INSN_P (next))
6868 return true;
6869 else if (INSN_VAR_LOCATION_DECL (insn) == INSN_VAR_LOCATION_DECL (next))
6870 return false;
6872 return true;
6874 else
6875 return true;
6878 /* Count the number of stores into pseudo. Callback for note_stores. */
6880 static void
6881 count_stores (rtx x, const_rtx set ATTRIBUTE_UNUSED, void *data)
6883 int *counts = (int *) data;
6884 if (REG_P (x) && REGNO (x) >= FIRST_PSEUDO_REGISTER)
6885 counts[REGNO (x)]++;
6888 struct dead_debug_insn_data
6890 int *counts;
6891 rtx *replacements;
6892 bool seen_repl;
6895 /* Return if a DEBUG_INSN needs to be reset because some dead
6896 pseudo doesn't have a replacement. Callback for for_each_rtx. */
6898 static int
6899 is_dead_debug_insn (rtx *loc, void *data)
6901 rtx x = *loc;
6902 struct dead_debug_insn_data *ddid = (struct dead_debug_insn_data *) data;
6904 if (is_dead_reg (x, ddid->counts))
6906 if (ddid->replacements && ddid->replacements[REGNO (x)] != NULL_RTX)
6907 ddid->seen_repl = true;
6908 else
6909 return 1;
6911 return 0;
6914 /* Replace a dead pseudo in a DEBUG_INSN with replacement DEBUG_EXPR.
6915 Callback for simplify_replace_fn_rtx. */
6917 static rtx
6918 replace_dead_reg (rtx x, const_rtx old_rtx ATTRIBUTE_UNUSED, void *data)
6920 rtx *replacements = (rtx *) data;
6922 if (REG_P (x)
6923 && REGNO (x) >= FIRST_PSEUDO_REGISTER
6924 && replacements[REGNO (x)] != NULL_RTX)
6926 if (GET_MODE (x) == GET_MODE (replacements[REGNO (x)]))
6927 return replacements[REGNO (x)];
6928 return lowpart_subreg (GET_MODE (x), replacements[REGNO (x)],
6929 GET_MODE (replacements[REGNO (x)]));
6931 return NULL_RTX;
6934 /* Scan all the insns and delete any that are dead; i.e., they store a register
6935 that is never used or they copy a register to itself.
6937 This is used to remove insns made obviously dead by cse, loop or other
6938 optimizations. It improves the heuristics in loop since it won't try to
6939 move dead invariants out of loops or make givs for dead quantities. The
6940 remaining passes of the compilation are also sped up. */
6943 delete_trivially_dead_insns (rtx insns, int nreg)
6945 int *counts;
6946 rtx insn, prev;
6947 rtx *replacements = NULL;
6948 int ndead = 0;
6950 timevar_push (TV_DELETE_TRIVIALLY_DEAD);
6951 /* First count the number of times each register is used. */
6952 if (MAY_HAVE_DEBUG_INSNS)
6954 counts = XCNEWVEC (int, nreg * 3);
6955 for (insn = insns; insn; insn = NEXT_INSN (insn))
6956 if (DEBUG_INSN_P (insn))
6957 count_reg_usage (INSN_VAR_LOCATION_LOC (insn), counts + nreg,
6958 NULL_RTX, 1);
6959 else if (INSN_P (insn))
6961 count_reg_usage (insn, counts, NULL_RTX, 1);
6962 note_stores (PATTERN (insn), count_stores, counts + nreg * 2);
6964 /* If there can be debug insns, COUNTS are 3 consecutive arrays.
6965 First one counts how many times each pseudo is used outside
6966 of debug insns, second counts how many times each pseudo is
6967 used in debug insns and third counts how many times a pseudo
6968 is stored. */
6970 else
6972 counts = XCNEWVEC (int, nreg);
6973 for (insn = insns; insn; insn = NEXT_INSN (insn))
6974 if (INSN_P (insn))
6975 count_reg_usage (insn, counts, NULL_RTX, 1);
6976 /* If no debug insns can be present, COUNTS is just an array
6977 which counts how many times each pseudo is used. */
6979 /* Go from the last insn to the first and delete insns that only set unused
6980 registers or copy a register to itself. As we delete an insn, remove
6981 usage counts for registers it uses.
6983 The first jump optimization pass may leave a real insn as the last
6984 insn in the function. We must not skip that insn or we may end
6985 up deleting code that is not really dead.
6987 If some otherwise unused register is only used in DEBUG_INSNs,
6988 try to create a DEBUG_EXPR temporary and emit a DEBUG_INSN before
6989 the setter. Then go through DEBUG_INSNs and if a DEBUG_EXPR
6990 has been created for the unused register, replace it with
6991 the DEBUG_EXPR, otherwise reset the DEBUG_INSN. */
6992 for (insn = get_last_insn (); insn; insn = prev)
6994 int live_insn = 0;
6996 prev = PREV_INSN (insn);
6997 if (!INSN_P (insn))
6998 continue;
7000 live_insn = insn_live_p (insn, counts);
7002 /* If this is a dead insn, delete it and show registers in it aren't
7003 being used. */
7005 if (! live_insn && dbg_cnt (delete_trivial_dead))
7007 if (DEBUG_INSN_P (insn))
7008 count_reg_usage (INSN_VAR_LOCATION_LOC (insn), counts + nreg,
7009 NULL_RTX, -1);
7010 else
7012 rtx set;
7013 if (MAY_HAVE_DEBUG_INSNS
7014 && (set = single_set (insn)) != NULL_RTX
7015 && is_dead_reg (SET_DEST (set), counts)
7016 /* Used at least once in some DEBUG_INSN. */
7017 && counts[REGNO (SET_DEST (set)) + nreg] > 0
7018 /* And set exactly once. */
7019 && counts[REGNO (SET_DEST (set)) + nreg * 2] == 1
7020 && !side_effects_p (SET_SRC (set))
7021 && asm_noperands (PATTERN (insn)) < 0)
7023 rtx dval, bind;
7025 /* Create DEBUG_EXPR (and DEBUG_EXPR_DECL). */
7026 dval = make_debug_expr_from_rtl (SET_DEST (set));
7028 /* Emit a debug bind insn before the insn in which
7029 reg dies. */
7030 bind = gen_rtx_VAR_LOCATION (GET_MODE (SET_DEST (set)),
7031 DEBUG_EXPR_TREE_DECL (dval),
7032 SET_SRC (set),
7033 VAR_INIT_STATUS_INITIALIZED);
7034 count_reg_usage (bind, counts + nreg, NULL_RTX, 1);
7036 bind = emit_debug_insn_before (bind, insn);
7037 df_insn_rescan (bind);
7039 if (replacements == NULL)
7040 replacements = XCNEWVEC (rtx, nreg);
7041 replacements[REGNO (SET_DEST (set))] = dval;
7044 count_reg_usage (insn, counts, NULL_RTX, -1);
7045 ndead++;
7047 delete_insn_and_edges (insn);
7051 if (MAY_HAVE_DEBUG_INSNS)
7053 struct dead_debug_insn_data ddid;
7054 ddid.counts = counts;
7055 ddid.replacements = replacements;
7056 for (insn = get_last_insn (); insn; insn = PREV_INSN (insn))
7057 if (DEBUG_INSN_P (insn))
7059 /* If this debug insn references a dead register that wasn't replaced
7060 with an DEBUG_EXPR, reset the DEBUG_INSN. */
7061 ddid.seen_repl = false;
7062 if (for_each_rtx (&INSN_VAR_LOCATION_LOC (insn),
7063 is_dead_debug_insn, &ddid))
7065 INSN_VAR_LOCATION_LOC (insn) = gen_rtx_UNKNOWN_VAR_LOC ();
7066 df_insn_rescan (insn);
7068 else if (ddid.seen_repl)
7070 INSN_VAR_LOCATION_LOC (insn)
7071 = simplify_replace_fn_rtx (INSN_VAR_LOCATION_LOC (insn),
7072 NULL_RTX, replace_dead_reg,
7073 replacements);
7074 df_insn_rescan (insn);
7077 free (replacements);
7080 if (dump_file && ndead)
7081 fprintf (dump_file, "Deleted %i trivially dead insns\n",
7082 ndead);
7083 /* Clean up. */
7084 free (counts);
7085 timevar_pop (TV_DELETE_TRIVIALLY_DEAD);
7086 return ndead;
7089 /* This function is called via for_each_rtx. The argument, NEWREG, is
7090 a condition code register with the desired mode. If we are looking
7091 at the same register in a different mode, replace it with
7092 NEWREG. */
7094 static int
7095 cse_change_cc_mode (rtx *loc, void *data)
7097 struct change_cc_mode_args* args = (struct change_cc_mode_args*)data;
7099 if (*loc
7100 && REG_P (*loc)
7101 && REGNO (*loc) == REGNO (args->newreg)
7102 && GET_MODE (*loc) != GET_MODE (args->newreg))
7104 validate_change (args->insn, loc, args->newreg, 1);
7106 return -1;
7108 return 0;
7111 /* Change the mode of any reference to the register REGNO (NEWREG) to
7112 GET_MODE (NEWREG) in INSN. */
7114 static void
7115 cse_change_cc_mode_insn (rtx insn, rtx newreg)
7117 struct change_cc_mode_args args;
7118 int success;
7120 if (!INSN_P (insn))
7121 return;
7123 args.insn = insn;
7124 args.newreg = newreg;
7126 for_each_rtx (&PATTERN (insn), cse_change_cc_mode, &args);
7127 for_each_rtx (&REG_NOTES (insn), cse_change_cc_mode, &args);
7129 /* If the following assertion was triggered, there is most probably
7130 something wrong with the cc_modes_compatible back end function.
7131 CC modes only can be considered compatible if the insn - with the mode
7132 replaced by any of the compatible modes - can still be recognized. */
7133 success = apply_change_group ();
7134 gcc_assert (success);
7137 /* Change the mode of any reference to the register REGNO (NEWREG) to
7138 GET_MODE (NEWREG), starting at START. Stop before END. Stop at
7139 any instruction which modifies NEWREG. */
7141 static void
7142 cse_change_cc_mode_insns (rtx start, rtx end, rtx newreg)
7144 rtx insn;
7146 for (insn = start; insn != end; insn = NEXT_INSN (insn))
7148 if (! INSN_P (insn))
7149 continue;
7151 if (reg_set_p (newreg, insn))
7152 return;
7154 cse_change_cc_mode_insn (insn, newreg);
7158 /* BB is a basic block which finishes with CC_REG as a condition code
7159 register which is set to CC_SRC. Look through the successors of BB
7160 to find blocks which have a single predecessor (i.e., this one),
7161 and look through those blocks for an assignment to CC_REG which is
7162 equivalent to CC_SRC. CAN_CHANGE_MODE indicates whether we are
7163 permitted to change the mode of CC_SRC to a compatible mode. This
7164 returns VOIDmode if no equivalent assignments were found.
7165 Otherwise it returns the mode which CC_SRC should wind up with.
7166 ORIG_BB should be the same as BB in the outermost cse_cc_succs call,
7167 but is passed unmodified down to recursive calls in order to prevent
7168 endless recursion.
7170 The main complexity in this function is handling the mode issues.
7171 We may have more than one duplicate which we can eliminate, and we
7172 try to find a mode which will work for multiple duplicates. */
7174 static enum machine_mode
7175 cse_cc_succs (basic_block bb, basic_block orig_bb, rtx cc_reg, rtx cc_src,
7176 bool can_change_mode)
7178 bool found_equiv;
7179 enum machine_mode mode;
7180 unsigned int insn_count;
7181 edge e;
7182 rtx insns[2];
7183 enum machine_mode modes[2];
7184 rtx last_insns[2];
7185 unsigned int i;
7186 rtx newreg;
7187 edge_iterator ei;
7189 /* We expect to have two successors. Look at both before picking
7190 the final mode for the comparison. If we have more successors
7191 (i.e., some sort of table jump, although that seems unlikely),
7192 then we require all beyond the first two to use the same
7193 mode. */
7195 found_equiv = false;
7196 mode = GET_MODE (cc_src);
7197 insn_count = 0;
7198 FOR_EACH_EDGE (e, ei, bb->succs)
7200 rtx insn;
7201 rtx end;
7203 if (e->flags & EDGE_COMPLEX)
7204 continue;
7206 if (EDGE_COUNT (e->dest->preds) != 1
7207 || e->dest == EXIT_BLOCK_PTR_FOR_FN (cfun)
7208 /* Avoid endless recursion on unreachable blocks. */
7209 || e->dest == orig_bb)
7210 continue;
7212 end = NEXT_INSN (BB_END (e->dest));
7213 for (insn = BB_HEAD (e->dest); insn != end; insn = NEXT_INSN (insn))
7215 rtx set;
7217 if (! INSN_P (insn))
7218 continue;
7220 /* If CC_SRC is modified, we have to stop looking for
7221 something which uses it. */
7222 if (modified_in_p (cc_src, insn))
7223 break;
7225 /* Check whether INSN sets CC_REG to CC_SRC. */
7226 set = single_set (insn);
7227 if (set
7228 && REG_P (SET_DEST (set))
7229 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7231 bool found;
7232 enum machine_mode set_mode;
7233 enum machine_mode comp_mode;
7235 found = false;
7236 set_mode = GET_MODE (SET_SRC (set));
7237 comp_mode = set_mode;
7238 if (rtx_equal_p (cc_src, SET_SRC (set)))
7239 found = true;
7240 else if (GET_CODE (cc_src) == COMPARE
7241 && GET_CODE (SET_SRC (set)) == COMPARE
7242 && mode != set_mode
7243 && rtx_equal_p (XEXP (cc_src, 0),
7244 XEXP (SET_SRC (set), 0))
7245 && rtx_equal_p (XEXP (cc_src, 1),
7246 XEXP (SET_SRC (set), 1)))
7249 comp_mode = targetm.cc_modes_compatible (mode, set_mode);
7250 if (comp_mode != VOIDmode
7251 && (can_change_mode || comp_mode == mode))
7252 found = true;
7255 if (found)
7257 found_equiv = true;
7258 if (insn_count < ARRAY_SIZE (insns))
7260 insns[insn_count] = insn;
7261 modes[insn_count] = set_mode;
7262 last_insns[insn_count] = end;
7263 ++insn_count;
7265 if (mode != comp_mode)
7267 gcc_assert (can_change_mode);
7268 mode = comp_mode;
7270 /* The modified insn will be re-recognized later. */
7271 PUT_MODE (cc_src, mode);
7274 else
7276 if (set_mode != mode)
7278 /* We found a matching expression in the
7279 wrong mode, but we don't have room to
7280 store it in the array. Punt. This case
7281 should be rare. */
7282 break;
7284 /* INSN sets CC_REG to a value equal to CC_SRC
7285 with the right mode. We can simply delete
7286 it. */
7287 delete_insn (insn);
7290 /* We found an instruction to delete. Keep looking,
7291 in the hopes of finding a three-way jump. */
7292 continue;
7295 /* We found an instruction which sets the condition
7296 code, so don't look any farther. */
7297 break;
7300 /* If INSN sets CC_REG in some other way, don't look any
7301 farther. */
7302 if (reg_set_p (cc_reg, insn))
7303 break;
7306 /* If we fell off the bottom of the block, we can keep looking
7307 through successors. We pass CAN_CHANGE_MODE as false because
7308 we aren't prepared to handle compatibility between the
7309 further blocks and this block. */
7310 if (insn == end)
7312 enum machine_mode submode;
7314 submode = cse_cc_succs (e->dest, orig_bb, cc_reg, cc_src, false);
7315 if (submode != VOIDmode)
7317 gcc_assert (submode == mode);
7318 found_equiv = true;
7319 can_change_mode = false;
7324 if (! found_equiv)
7325 return VOIDmode;
7327 /* Now INSN_COUNT is the number of instructions we found which set
7328 CC_REG to a value equivalent to CC_SRC. The instructions are in
7329 INSNS. The modes used by those instructions are in MODES. */
7331 newreg = NULL_RTX;
7332 for (i = 0; i < insn_count; ++i)
7334 if (modes[i] != mode)
7336 /* We need to change the mode of CC_REG in INSNS[i] and
7337 subsequent instructions. */
7338 if (! newreg)
7340 if (GET_MODE (cc_reg) == mode)
7341 newreg = cc_reg;
7342 else
7343 newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7345 cse_change_cc_mode_insns (NEXT_INSN (insns[i]), last_insns[i],
7346 newreg);
7349 delete_insn_and_edges (insns[i]);
7352 return mode;
7355 /* If we have a fixed condition code register (or two), walk through
7356 the instructions and try to eliminate duplicate assignments. */
7358 static void
7359 cse_condition_code_reg (void)
7361 unsigned int cc_regno_1;
7362 unsigned int cc_regno_2;
7363 rtx cc_reg_1;
7364 rtx cc_reg_2;
7365 basic_block bb;
7367 if (! targetm.fixed_condition_code_regs (&cc_regno_1, &cc_regno_2))
7368 return;
7370 cc_reg_1 = gen_rtx_REG (CCmode, cc_regno_1);
7371 if (cc_regno_2 != INVALID_REGNUM)
7372 cc_reg_2 = gen_rtx_REG (CCmode, cc_regno_2);
7373 else
7374 cc_reg_2 = NULL_RTX;
7376 FOR_EACH_BB_FN (bb, cfun)
7378 rtx last_insn;
7379 rtx cc_reg;
7380 rtx insn;
7381 rtx cc_src_insn;
7382 rtx cc_src;
7383 enum machine_mode mode;
7384 enum machine_mode orig_mode;
7386 /* Look for blocks which end with a conditional jump based on a
7387 condition code register. Then look for the instruction which
7388 sets the condition code register. Then look through the
7389 successor blocks for instructions which set the condition
7390 code register to the same value. There are other possible
7391 uses of the condition code register, but these are by far the
7392 most common and the ones which we are most likely to be able
7393 to optimize. */
7395 last_insn = BB_END (bb);
7396 if (!JUMP_P (last_insn))
7397 continue;
7399 if (reg_referenced_p (cc_reg_1, PATTERN (last_insn)))
7400 cc_reg = cc_reg_1;
7401 else if (cc_reg_2 && reg_referenced_p (cc_reg_2, PATTERN (last_insn)))
7402 cc_reg = cc_reg_2;
7403 else
7404 continue;
7406 cc_src_insn = NULL_RTX;
7407 cc_src = NULL_RTX;
7408 for (insn = PREV_INSN (last_insn);
7409 insn && insn != PREV_INSN (BB_HEAD (bb));
7410 insn = PREV_INSN (insn))
7412 rtx set;
7414 if (! INSN_P (insn))
7415 continue;
7416 set = single_set (insn);
7417 if (set
7418 && REG_P (SET_DEST (set))
7419 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7421 cc_src_insn = insn;
7422 cc_src = SET_SRC (set);
7423 break;
7425 else if (reg_set_p (cc_reg, insn))
7426 break;
7429 if (! cc_src_insn)
7430 continue;
7432 if (modified_between_p (cc_src, cc_src_insn, NEXT_INSN (last_insn)))
7433 continue;
7435 /* Now CC_REG is a condition code register used for a
7436 conditional jump at the end of the block, and CC_SRC, in
7437 CC_SRC_INSN, is the value to which that condition code
7438 register is set, and CC_SRC is still meaningful at the end of
7439 the basic block. */
7441 orig_mode = GET_MODE (cc_src);
7442 mode = cse_cc_succs (bb, bb, cc_reg, cc_src, true);
7443 if (mode != VOIDmode)
7445 gcc_assert (mode == GET_MODE (cc_src));
7446 if (mode != orig_mode)
7448 rtx newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7450 cse_change_cc_mode_insn (cc_src_insn, newreg);
7452 /* Do the same in the following insns that use the
7453 current value of CC_REG within BB. */
7454 cse_change_cc_mode_insns (NEXT_INSN (cc_src_insn),
7455 NEXT_INSN (last_insn),
7456 newreg);
7463 /* Perform common subexpression elimination. Nonzero value from
7464 `cse_main' means that jumps were simplified and some code may now
7465 be unreachable, so do jump optimization again. */
7466 static unsigned int
7467 rest_of_handle_cse (void)
7469 int tem;
7471 if (dump_file)
7472 dump_flow_info (dump_file, dump_flags);
7474 tem = cse_main (get_insns (), max_reg_num ());
7476 /* If we are not running more CSE passes, then we are no longer
7477 expecting CSE to be run. But always rerun it in a cheap mode. */
7478 cse_not_expected = !flag_rerun_cse_after_loop && !flag_gcse;
7480 if (tem == 2)
7482 timevar_push (TV_JUMP);
7483 rebuild_jump_labels (get_insns ());
7484 cleanup_cfg (CLEANUP_CFG_CHANGED);
7485 timevar_pop (TV_JUMP);
7487 else if (tem == 1 || optimize > 1)
7488 cleanup_cfg (0);
7490 return 0;
7493 namespace {
7495 const pass_data pass_data_cse =
7497 RTL_PASS, /* type */
7498 "cse1", /* name */
7499 OPTGROUP_NONE, /* optinfo_flags */
7500 true, /* has_execute */
7501 TV_CSE, /* tv_id */
7502 0, /* properties_required */
7503 0, /* properties_provided */
7504 0, /* properties_destroyed */
7505 0, /* todo_flags_start */
7506 TODO_df_finish, /* todo_flags_finish */
7509 class pass_cse : public rtl_opt_pass
7511 public:
7512 pass_cse (gcc::context *ctxt)
7513 : rtl_opt_pass (pass_data_cse, ctxt)
7516 /* opt_pass methods: */
7517 virtual bool gate (function *) { return optimize > 0; }
7518 virtual unsigned int execute (function *) { return rest_of_handle_cse (); }
7520 }; // class pass_cse
7522 } // anon namespace
7524 rtl_opt_pass *
7525 make_pass_cse (gcc::context *ctxt)
7527 return new pass_cse (ctxt);
7531 /* Run second CSE pass after loop optimizations. */
7532 static unsigned int
7533 rest_of_handle_cse2 (void)
7535 int tem;
7537 if (dump_file)
7538 dump_flow_info (dump_file, dump_flags);
7540 tem = cse_main (get_insns (), max_reg_num ());
7542 /* Run a pass to eliminate duplicated assignments to condition code
7543 registers. We have to run this after bypass_jumps, because it
7544 makes it harder for that pass to determine whether a jump can be
7545 bypassed safely. */
7546 cse_condition_code_reg ();
7548 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7550 if (tem == 2)
7552 timevar_push (TV_JUMP);
7553 rebuild_jump_labels (get_insns ());
7554 cleanup_cfg (CLEANUP_CFG_CHANGED);
7555 timevar_pop (TV_JUMP);
7557 else if (tem == 1)
7558 cleanup_cfg (0);
7560 cse_not_expected = 1;
7561 return 0;
7565 namespace {
7567 const pass_data pass_data_cse2 =
7569 RTL_PASS, /* type */
7570 "cse2", /* name */
7571 OPTGROUP_NONE, /* optinfo_flags */
7572 true, /* has_execute */
7573 TV_CSE2, /* tv_id */
7574 0, /* properties_required */
7575 0, /* properties_provided */
7576 0, /* properties_destroyed */
7577 0, /* todo_flags_start */
7578 TODO_df_finish, /* todo_flags_finish */
7581 class pass_cse2 : public rtl_opt_pass
7583 public:
7584 pass_cse2 (gcc::context *ctxt)
7585 : rtl_opt_pass (pass_data_cse2, ctxt)
7588 /* opt_pass methods: */
7589 virtual bool gate (function *)
7591 return optimize > 0 && flag_rerun_cse_after_loop;
7594 virtual unsigned int execute (function *) { return rest_of_handle_cse2 (); }
7596 }; // class pass_cse2
7598 } // anon namespace
7600 rtl_opt_pass *
7601 make_pass_cse2 (gcc::context *ctxt)
7603 return new pass_cse2 (ctxt);
7606 /* Run second CSE pass after loop optimizations. */
7607 static unsigned int
7608 rest_of_handle_cse_after_global_opts (void)
7610 int save_cfj;
7611 int tem;
7613 /* We only want to do local CSE, so don't follow jumps. */
7614 save_cfj = flag_cse_follow_jumps;
7615 flag_cse_follow_jumps = 0;
7617 rebuild_jump_labels (get_insns ());
7618 tem = cse_main (get_insns (), max_reg_num ());
7619 purge_all_dead_edges ();
7620 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7622 cse_not_expected = !flag_rerun_cse_after_loop;
7624 /* If cse altered any jumps, rerun jump opts to clean things up. */
7625 if (tem == 2)
7627 timevar_push (TV_JUMP);
7628 rebuild_jump_labels (get_insns ());
7629 cleanup_cfg (CLEANUP_CFG_CHANGED);
7630 timevar_pop (TV_JUMP);
7632 else if (tem == 1)
7633 cleanup_cfg (0);
7635 flag_cse_follow_jumps = save_cfj;
7636 return 0;
7639 namespace {
7641 const pass_data pass_data_cse_after_global_opts =
7643 RTL_PASS, /* type */
7644 "cse_local", /* name */
7645 OPTGROUP_NONE, /* optinfo_flags */
7646 true, /* has_execute */
7647 TV_CSE, /* tv_id */
7648 0, /* properties_required */
7649 0, /* properties_provided */
7650 0, /* properties_destroyed */
7651 0, /* todo_flags_start */
7652 TODO_df_finish, /* todo_flags_finish */
7655 class pass_cse_after_global_opts : public rtl_opt_pass
7657 public:
7658 pass_cse_after_global_opts (gcc::context *ctxt)
7659 : rtl_opt_pass (pass_data_cse_after_global_opts, ctxt)
7662 /* opt_pass methods: */
7663 virtual bool gate (function *)
7665 return optimize > 0 && flag_rerun_cse_after_global_opts;
7668 virtual unsigned int execute (function *)
7670 return rest_of_handle_cse_after_global_opts ();
7673 }; // class pass_cse_after_global_opts
7675 } // anon namespace
7677 rtl_opt_pass *
7678 make_pass_cse_after_global_opts (gcc::context *ctxt)
7680 return new pass_cse_after_global_opts (ctxt);