Merge from mainline
[official-gcc.git] / gcc / cse.c
blob7d6f46b0a3563ade022c2956357a5fbd54c56895
1 /* Common subexpression elimination for GNU compiler.
2 Copyright (C) 1987, 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998
3 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 2, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING. If not, write to the Free
19 Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
20 02110-1301, USA. */
22 #include "config.h"
23 /* stdio.h must precede rtl.h for FFS. */
24 #include "system.h"
25 #include "coretypes.h"
26 #include "tm.h"
27 #include "rtl.h"
28 #include "tm_p.h"
29 #include "hard-reg-set.h"
30 #include "regs.h"
31 #include "basic-block.h"
32 #include "flags.h"
33 #include "real.h"
34 #include "insn-config.h"
35 #include "recog.h"
36 #include "function.h"
37 #include "expr.h"
38 #include "toplev.h"
39 #include "output.h"
40 #include "ggc.h"
41 #include "timevar.h"
42 #include "except.h"
43 #include "target.h"
44 #include "params.h"
45 #include "rtlhooks-def.h"
46 #include "tree-pass.h"
48 /* The basic idea of common subexpression elimination is to go
49 through the code, keeping a record of expressions that would
50 have the same value at the current scan point, and replacing
51 expressions encountered with the cheapest equivalent expression.
53 It is too complicated to keep track of the different possibilities
54 when control paths merge in this code; so, at each label, we forget all
55 that is known and start fresh. This can be described as processing each
56 extended basic block separately. We have a separate pass to perform
57 global CSE.
59 Note CSE can turn a conditional or computed jump into a nop or
60 an unconditional jump. When this occurs we arrange to run the jump
61 optimizer after CSE to delete the unreachable code.
63 We use two data structures to record the equivalent expressions:
64 a hash table for most expressions, and a vector of "quantity
65 numbers" to record equivalent (pseudo) registers.
67 The use of the special data structure for registers is desirable
68 because it is faster. It is possible because registers references
69 contain a fairly small number, the register number, taken from
70 a contiguously allocated series, and two register references are
71 identical if they have the same number. General expressions
72 do not have any such thing, so the only way to retrieve the
73 information recorded on an expression other than a register
74 is to keep it in a hash table.
76 Registers and "quantity numbers":
78 At the start of each basic block, all of the (hardware and pseudo)
79 registers used in the function are given distinct quantity
80 numbers to indicate their contents. During scan, when the code
81 copies one register into another, we copy the quantity number.
82 When a register is loaded in any other way, we allocate a new
83 quantity number to describe the value generated by this operation.
84 `REG_QTY (N)' records what quantity register N is currently thought
85 of as containing.
87 All real quantity numbers are greater than or equal to zero.
88 If register N has not been assigned a quantity, `REG_QTY (N)' will
89 equal -N - 1, which is always negative.
91 Quantity numbers below zero do not exist and none of the `qty_table'
92 entries should be referenced with a negative index.
94 We also maintain a bidirectional chain of registers for each
95 quantity number. The `qty_table` members `first_reg' and `last_reg',
96 and `reg_eqv_table' members `next' and `prev' hold these chains.
98 The first register in a chain is the one whose lifespan is least local.
99 Among equals, it is the one that was seen first.
100 We replace any equivalent register with that one.
102 If two registers have the same quantity number, it must be true that
103 REG expressions with qty_table `mode' must be in the hash table for both
104 registers and must be in the same class.
106 The converse is not true. Since hard registers may be referenced in
107 any mode, two REG expressions might be equivalent in the hash table
108 but not have the same quantity number if the quantity number of one
109 of the registers is not the same mode as those expressions.
111 Constants and quantity numbers
113 When a quantity has a known constant value, that value is stored
114 in the appropriate qty_table `const_rtx'. This is in addition to
115 putting the constant in the hash table as is usual for non-regs.
117 Whether a reg or a constant is preferred is determined by the configuration
118 macro CONST_COSTS and will often depend on the constant value. In any
119 event, expressions containing constants can be simplified, by fold_rtx.
121 When a quantity has a known nearly constant value (such as an address
122 of a stack slot), that value is stored in the appropriate qty_table
123 `const_rtx'.
125 Integer constants don't have a machine mode. However, cse
126 determines the intended machine mode from the destination
127 of the instruction that moves the constant. The machine mode
128 is recorded in the hash table along with the actual RTL
129 constant expression so that different modes are kept separate.
131 Other expressions:
133 To record known equivalences among expressions in general
134 we use a hash table called `table'. It has a fixed number of buckets
135 that contain chains of `struct table_elt' elements for expressions.
136 These chains connect the elements whose expressions have the same
137 hash codes.
139 Other chains through the same elements connect the elements which
140 currently have equivalent values.
142 Register references in an expression are canonicalized before hashing
143 the expression. This is done using `reg_qty' and qty_table `first_reg'.
144 The hash code of a register reference is computed using the quantity
145 number, not the register number.
147 When the value of an expression changes, it is necessary to remove from the
148 hash table not just that expression but all expressions whose values
149 could be different as a result.
151 1. If the value changing is in memory, except in special cases
152 ANYTHING referring to memory could be changed. That is because
153 nobody knows where a pointer does not point.
154 The function `invalidate_memory' removes what is necessary.
156 The special cases are when the address is constant or is
157 a constant plus a fixed register such as the frame pointer
158 or a static chain pointer. When such addresses are stored in,
159 we can tell exactly which other such addresses must be invalidated
160 due to overlap. `invalidate' does this.
161 All expressions that refer to non-constant
162 memory addresses are also invalidated. `invalidate_memory' does this.
164 2. If the value changing is a register, all expressions
165 containing references to that register, and only those,
166 must be removed.
168 Because searching the entire hash table for expressions that contain
169 a register is very slow, we try to figure out when it isn't necessary.
170 Precisely, this is necessary only when expressions have been
171 entered in the hash table using this register, and then the value has
172 changed, and then another expression wants to be added to refer to
173 the register's new value. This sequence of circumstances is rare
174 within any one basic block.
176 `REG_TICK' and `REG_IN_TABLE', accessors for members of
177 cse_reg_info, are used to detect this case. REG_TICK (i) is
178 incremented whenever a value is stored in register i.
179 REG_IN_TABLE (i) holds -1 if no references to register i have been
180 entered in the table; otherwise, it contains the value REG_TICK (i)
181 had when the references were entered. If we want to enter a
182 reference and REG_IN_TABLE (i) != REG_TICK (i), we must scan and
183 remove old references. Until we want to enter a new entry, the
184 mere fact that the two vectors don't match makes the entries be
185 ignored if anyone tries to match them.
187 Registers themselves are entered in the hash table as well as in
188 the equivalent-register chains. However, `REG_TICK' and
189 `REG_IN_TABLE' do not apply to expressions which are simple
190 register references. These expressions are removed from the table
191 immediately when they become invalid, and this can be done even if
192 we do not immediately search for all the expressions that refer to
193 the register.
195 A CLOBBER rtx in an instruction invalidates its operand for further
196 reuse. A CLOBBER or SET rtx whose operand is a MEM:BLK
197 invalidates everything that resides in memory.
199 Related expressions:
201 Constant expressions that differ only by an additive integer
202 are called related. When a constant expression is put in
203 the table, the related expression with no constant term
204 is also entered. These are made to point at each other
205 so that it is possible to find out if there exists any
206 register equivalent to an expression related to a given expression. */
208 /* Length of qty_table vector. We know in advance we will not need
209 a quantity number this big. */
211 static int max_qty;
213 /* Next quantity number to be allocated.
214 This is 1 + the largest number needed so far. */
216 static int next_qty;
218 /* Per-qty information tracking.
220 `first_reg' and `last_reg' track the head and tail of the
221 chain of registers which currently contain this quantity.
223 `mode' contains the machine mode of this quantity.
225 `const_rtx' holds the rtx of the constant value of this
226 quantity, if known. A summations of the frame/arg pointer
227 and a constant can also be entered here. When this holds
228 a known value, `const_insn' is the insn which stored the
229 constant value.
231 `comparison_{code,const,qty}' are used to track when a
232 comparison between a quantity and some constant or register has
233 been passed. In such a case, we know the results of the comparison
234 in case we see it again. These members record a comparison that
235 is known to be true. `comparison_code' holds the rtx code of such
236 a comparison, else it is set to UNKNOWN and the other two
237 comparison members are undefined. `comparison_const' holds
238 the constant being compared against, or zero if the comparison
239 is not against a constant. `comparison_qty' holds the quantity
240 being compared against when the result is known. If the comparison
241 is not with a register, `comparison_qty' is -1. */
243 struct qty_table_elem
245 rtx const_rtx;
246 rtx const_insn;
247 rtx comparison_const;
248 int comparison_qty;
249 unsigned int first_reg, last_reg;
250 /* The sizes of these fields should match the sizes of the
251 code and mode fields of struct rtx_def (see rtl.h). */
252 ENUM_BITFIELD(rtx_code) comparison_code : 16;
253 ENUM_BITFIELD(machine_mode) mode : 8;
256 /* The table of all qtys, indexed by qty number. */
257 static struct qty_table_elem *qty_table;
259 /* Structure used to pass arguments via for_each_rtx to function
260 cse_change_cc_mode. */
261 struct change_cc_mode_args
263 rtx insn;
264 rtx newreg;
267 #ifdef HAVE_cc0
268 /* For machines that have a CC0, we do not record its value in the hash
269 table since its use is guaranteed to be the insn immediately following
270 its definition and any other insn is presumed to invalidate it.
272 Instead, we store below the value last assigned to CC0. If it should
273 happen to be a constant, it is stored in preference to the actual
274 assigned value. In case it is a constant, we store the mode in which
275 the constant should be interpreted. */
277 static rtx prev_insn_cc0;
278 static enum machine_mode prev_insn_cc0_mode;
280 /* Previous actual insn. 0 if at first insn of basic block. */
282 static rtx prev_insn;
283 #endif
285 /* Insn being scanned. */
287 static rtx this_insn;
289 /* Index by register number, gives the number of the next (or
290 previous) register in the chain of registers sharing the same
291 value.
293 Or -1 if this register is at the end of the chain.
295 If REG_QTY (N) == -N - 1, reg_eqv_table[N].next is undefined. */
297 /* Per-register equivalence chain. */
298 struct reg_eqv_elem
300 int next, prev;
303 /* The table of all register equivalence chains. */
304 static struct reg_eqv_elem *reg_eqv_table;
306 struct cse_reg_info
308 /* The timestamp at which this register is initialized. */
309 unsigned int timestamp;
311 /* The quantity number of the register's current contents. */
312 int reg_qty;
314 /* The number of times the register has been altered in the current
315 basic block. */
316 int reg_tick;
318 /* The REG_TICK value at which rtx's containing this register are
319 valid in the hash table. If this does not equal the current
320 reg_tick value, such expressions existing in the hash table are
321 invalid. */
322 int reg_in_table;
324 /* The SUBREG that was set when REG_TICK was last incremented. Set
325 to -1 if the last store was to the whole register, not a subreg. */
326 unsigned int subreg_ticked;
329 /* A table of cse_reg_info indexed by register numbers. */
330 static struct cse_reg_info *cse_reg_info_table;
332 /* The size of the above table. */
333 static unsigned int cse_reg_info_table_size;
335 /* The index of the first entry that has not been initialized. */
336 static unsigned int cse_reg_info_table_first_uninitialized;
338 /* The timestamp at the beginning of the current run of
339 cse_basic_block. We increment this variable at the beginning of
340 the current run of cse_basic_block. The timestamp field of a
341 cse_reg_info entry matches the value of this variable if and only
342 if the entry has been initialized during the current run of
343 cse_basic_block. */
344 static unsigned int cse_reg_info_timestamp;
346 /* A HARD_REG_SET containing all the hard registers for which there is
347 currently a REG expression in the hash table. Note the difference
348 from the above variables, which indicate if the REG is mentioned in some
349 expression in the table. */
351 static HARD_REG_SET hard_regs_in_table;
353 /* CUID of insn that starts the basic block currently being cse-processed. */
355 static int cse_basic_block_start;
357 /* CUID of insn that ends the basic block currently being cse-processed. */
359 static int cse_basic_block_end;
361 /* Vector mapping INSN_UIDs to cuids.
362 The cuids are like uids but increase monotonically always.
363 We use them to see whether a reg is used outside a given basic block. */
365 static int *uid_cuid;
367 /* Highest UID in UID_CUID. */
368 static int max_uid;
370 /* Get the cuid of an insn. */
372 #define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)])
374 /* Nonzero if this pass has made changes, and therefore it's
375 worthwhile to run the garbage collector. */
377 static int cse_altered;
379 /* Nonzero if cse has altered conditional jump insns
380 in such a way that jump optimization should be redone. */
382 static int cse_jumps_altered;
384 /* Nonzero if we put a LABEL_REF into the hash table for an INSN without a
385 REG_LABEL, we have to rerun jump after CSE to put in the note. */
386 static int recorded_label_ref;
388 /* canon_hash stores 1 in do_not_record
389 if it notices a reference to CC0, PC, or some other volatile
390 subexpression. */
392 static int do_not_record;
394 /* canon_hash stores 1 in hash_arg_in_memory
395 if it notices a reference to memory within the expression being hashed. */
397 static int hash_arg_in_memory;
399 /* The hash table contains buckets which are chains of `struct table_elt's,
400 each recording one expression's information.
401 That expression is in the `exp' field.
403 The canon_exp field contains a canonical (from the point of view of
404 alias analysis) version of the `exp' field.
406 Those elements with the same hash code are chained in both directions
407 through the `next_same_hash' and `prev_same_hash' fields.
409 Each set of expressions with equivalent values
410 are on a two-way chain through the `next_same_value'
411 and `prev_same_value' fields, and all point with
412 the `first_same_value' field at the first element in
413 that chain. The chain is in order of increasing cost.
414 Each element's cost value is in its `cost' field.
416 The `in_memory' field is nonzero for elements that
417 involve any reference to memory. These elements are removed
418 whenever a write is done to an unidentified location in memory.
419 To be safe, we assume that a memory address is unidentified unless
420 the address is either a symbol constant or a constant plus
421 the frame pointer or argument pointer.
423 The `related_value' field is used to connect related expressions
424 (that differ by adding an integer).
425 The related expressions are chained in a circular fashion.
426 `related_value' is zero for expressions for which this
427 chain is not useful.
429 The `cost' field stores the cost of this element's expression.
430 The `regcost' field stores the value returned by approx_reg_cost for
431 this element's expression.
433 The `is_const' flag is set if the element is a constant (including
434 a fixed address).
436 The `flag' field is used as a temporary during some search routines.
438 The `mode' field is usually the same as GET_MODE (`exp'), but
439 if `exp' is a CONST_INT and has no machine mode then the `mode'
440 field is the mode it was being used as. Each constant is
441 recorded separately for each mode it is used with. */
443 struct table_elt
445 rtx exp;
446 rtx canon_exp;
447 struct table_elt *next_same_hash;
448 struct table_elt *prev_same_hash;
449 struct table_elt *next_same_value;
450 struct table_elt *prev_same_value;
451 struct table_elt *first_same_value;
452 struct table_elt *related_value;
453 int cost;
454 int regcost;
455 /* The size of this field should match the size
456 of the mode field of struct rtx_def (see rtl.h). */
457 ENUM_BITFIELD(machine_mode) mode : 8;
458 char in_memory;
459 char is_const;
460 char flag;
463 /* We don't want a lot of buckets, because we rarely have very many
464 things stored in the hash table, and a lot of buckets slows
465 down a lot of loops that happen frequently. */
466 #define HASH_SHIFT 5
467 #define HASH_SIZE (1 << HASH_SHIFT)
468 #define HASH_MASK (HASH_SIZE - 1)
470 /* Compute hash code of X in mode M. Special-case case where X is a pseudo
471 register (hard registers may require `do_not_record' to be set). */
473 #define HASH(X, M) \
474 ((REG_P (X) && REGNO (X) >= FIRST_PSEUDO_REGISTER \
475 ? (((unsigned) REG << 7) + (unsigned) REG_QTY (REGNO (X))) \
476 : canon_hash (X, M)) & HASH_MASK)
478 /* Like HASH, but without side-effects. */
479 #define SAFE_HASH(X, M) \
480 ((REG_P (X) && REGNO (X) >= FIRST_PSEUDO_REGISTER \
481 ? (((unsigned) REG << 7) + (unsigned) REG_QTY (REGNO (X))) \
482 : safe_hash (X, M)) & HASH_MASK)
484 /* Determine whether register number N is considered a fixed register for the
485 purpose of approximating register costs.
486 It is desirable to replace other regs with fixed regs, to reduce need for
487 non-fixed hard regs.
488 A reg wins if it is either the frame pointer or designated as fixed. */
489 #define FIXED_REGNO_P(N) \
490 ((N) == FRAME_POINTER_REGNUM || (N) == HARD_FRAME_POINTER_REGNUM \
491 || fixed_regs[N] || global_regs[N])
493 /* Compute cost of X, as stored in the `cost' field of a table_elt. Fixed
494 hard registers and pointers into the frame are the cheapest with a cost
495 of 0. Next come pseudos with a cost of one and other hard registers with
496 a cost of 2. Aside from these special cases, call `rtx_cost'. */
498 #define CHEAP_REGNO(N) \
499 (REGNO_PTR_FRAME_P(N) \
500 || (HARD_REGISTER_NUM_P (N) \
501 && FIXED_REGNO_P (N) && REGNO_REG_CLASS (N) != NO_REGS))
503 #define COST(X) (REG_P (X) ? 0 : notreg_cost (X, SET))
504 #define COST_IN(X,OUTER) (REG_P (X) ? 0 : notreg_cost (X, OUTER))
506 /* Get the number of times this register has been updated in this
507 basic block. */
509 #define REG_TICK(N) (get_cse_reg_info (N)->reg_tick)
511 /* Get the point at which REG was recorded in the table. */
513 #define REG_IN_TABLE(N) (get_cse_reg_info (N)->reg_in_table)
515 /* Get the SUBREG set at the last increment to REG_TICK (-1 if not a
516 SUBREG). */
518 #define SUBREG_TICKED(N) (get_cse_reg_info (N)->subreg_ticked)
520 /* Get the quantity number for REG. */
522 #define REG_QTY(N) (get_cse_reg_info (N)->reg_qty)
524 /* Determine if the quantity number for register X represents a valid index
525 into the qty_table. */
527 #define REGNO_QTY_VALID_P(N) (REG_QTY (N) >= 0)
529 static struct table_elt *table[HASH_SIZE];
531 /* Chain of `struct table_elt's made so far for this function
532 but currently removed from the table. */
534 static struct table_elt *free_element_chain;
536 /* Set to the cost of a constant pool reference if one was found for a
537 symbolic constant. If this was found, it means we should try to
538 convert constants into constant pool entries if they don't fit in
539 the insn. */
541 static int constant_pool_entries_cost;
542 static int constant_pool_entries_regcost;
544 /* This data describes a block that will be processed by cse_basic_block. */
546 struct cse_basic_block_data
548 /* Lowest CUID value of insns in block. */
549 int low_cuid;
550 /* Highest CUID value of insns in block. */
551 int high_cuid;
552 /* Total number of SETs in block. */
553 int nsets;
554 /* Last insn in the block. */
555 rtx last;
556 /* Size of current branch path, if any. */
557 int path_size;
558 /* Current branch path, indicating which branches will be taken. */
559 struct branch_path
561 /* The branch insn. */
562 rtx branch;
563 /* Whether it should be taken or not. AROUND is the same as taken
564 except that it is used when the destination label is not preceded
565 by a BARRIER. */
566 enum taken {PATH_TAKEN, PATH_NOT_TAKEN, PATH_AROUND} status;
567 } *path;
570 static bool fixed_base_plus_p (rtx x);
571 static int notreg_cost (rtx, enum rtx_code);
572 static int approx_reg_cost_1 (rtx *, void *);
573 static int approx_reg_cost (rtx);
574 static int preferable (int, int, int, int);
575 static void new_basic_block (void);
576 static void make_new_qty (unsigned int, enum machine_mode);
577 static void make_regs_eqv (unsigned int, unsigned int);
578 static void delete_reg_equiv (unsigned int);
579 static int mention_regs (rtx);
580 static int insert_regs (rtx, struct table_elt *, int);
581 static void remove_from_table (struct table_elt *, unsigned);
582 static struct table_elt *lookup (rtx, unsigned, enum machine_mode);
583 static struct table_elt *lookup_for_remove (rtx, unsigned, enum machine_mode);
584 static rtx lookup_as_function (rtx, enum rtx_code);
585 static struct table_elt *insert (rtx, struct table_elt *, unsigned,
586 enum machine_mode);
587 static void merge_equiv_classes (struct table_elt *, struct table_elt *);
588 static void invalidate (rtx, enum machine_mode);
589 static int cse_rtx_varies_p (rtx, int);
590 static void remove_invalid_refs (unsigned int);
591 static void remove_invalid_subreg_refs (unsigned int, unsigned int,
592 enum machine_mode);
593 static void rehash_using_reg (rtx);
594 static void invalidate_memory (void);
595 static void invalidate_for_call (void);
596 static rtx use_related_value (rtx, struct table_elt *);
598 static inline unsigned canon_hash (rtx, enum machine_mode);
599 static inline unsigned safe_hash (rtx, enum machine_mode);
600 static unsigned hash_rtx_string (const char *);
602 static rtx canon_reg (rtx, rtx);
603 static void find_best_addr (rtx, rtx *, enum machine_mode);
604 static enum rtx_code find_comparison_args (enum rtx_code, rtx *, rtx *,
605 enum machine_mode *,
606 enum machine_mode *);
607 static rtx fold_rtx (rtx, rtx);
608 static rtx equiv_constant (rtx);
609 static void record_jump_equiv (rtx, int);
610 static void record_jump_cond (enum rtx_code, enum machine_mode, rtx, rtx,
611 int);
612 static void cse_insn (rtx, rtx);
613 static void cse_end_of_basic_block (rtx, struct cse_basic_block_data *,
614 int, int);
615 static int addr_affects_sp_p (rtx);
616 static void invalidate_from_clobbers (rtx);
617 static rtx cse_process_notes (rtx, rtx);
618 static void invalidate_skipped_set (rtx, rtx, void *);
619 static void invalidate_skipped_block (rtx);
620 static rtx cse_basic_block (rtx, rtx, struct branch_path *);
621 static void count_reg_usage (rtx, int *, rtx, int);
622 static int check_for_label_ref (rtx *, void *);
623 extern void dump_class (struct table_elt*);
624 static void get_cse_reg_info_1 (unsigned int regno);
625 static struct cse_reg_info * get_cse_reg_info (unsigned int regno);
626 static int check_dependence (rtx *, void *);
628 static void flush_hash_table (void);
629 static bool insn_live_p (rtx, int *);
630 static bool set_live_p (rtx, rtx, int *);
631 static bool dead_libcall_p (rtx, int *);
632 static int cse_change_cc_mode (rtx *, void *);
633 static void cse_change_cc_mode_insn (rtx, rtx);
634 static void cse_change_cc_mode_insns (rtx, rtx, rtx);
635 static enum machine_mode cse_cc_succs (basic_block, rtx, rtx, bool);
638 #undef RTL_HOOKS_GEN_LOWPART
639 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_if_possible
641 static const struct rtl_hooks cse_rtl_hooks = RTL_HOOKS_INITIALIZER;
643 /* Nonzero if X has the form (PLUS frame-pointer integer). We check for
644 virtual regs here because the simplify_*_operation routines are called
645 by integrate.c, which is called before virtual register instantiation. */
647 static bool
648 fixed_base_plus_p (rtx x)
650 switch (GET_CODE (x))
652 case REG:
653 if (x == frame_pointer_rtx || x == hard_frame_pointer_rtx)
654 return true;
655 if (x == arg_pointer_rtx && fixed_regs[ARG_POINTER_REGNUM])
656 return true;
657 if (REGNO (x) >= FIRST_VIRTUAL_REGISTER
658 && REGNO (x) <= LAST_VIRTUAL_REGISTER)
659 return true;
660 return false;
662 case PLUS:
663 if (GET_CODE (XEXP (x, 1)) != CONST_INT)
664 return false;
665 return fixed_base_plus_p (XEXP (x, 0));
667 default:
668 return false;
672 /* Dump the expressions in the equivalence class indicated by CLASSP.
673 This function is used only for debugging. */
674 void
675 dump_class (struct table_elt *classp)
677 struct table_elt *elt;
679 fprintf (stderr, "Equivalence chain for ");
680 print_rtl (stderr, classp->exp);
681 fprintf (stderr, ": \n");
683 for (elt = classp->first_same_value; elt; elt = elt->next_same_value)
685 print_rtl (stderr, elt->exp);
686 fprintf (stderr, "\n");
690 /* Subroutine of approx_reg_cost; called through for_each_rtx. */
692 static int
693 approx_reg_cost_1 (rtx *xp, void *data)
695 rtx x = *xp;
696 int *cost_p = data;
698 if (x && REG_P (x))
700 unsigned int regno = REGNO (x);
702 if (! CHEAP_REGNO (regno))
704 if (regno < FIRST_PSEUDO_REGISTER)
706 if (SMALL_REGISTER_CLASSES)
707 return 1;
708 *cost_p += 2;
710 else
711 *cost_p += 1;
715 return 0;
718 /* Return an estimate of the cost of the registers used in an rtx.
719 This is mostly the number of different REG expressions in the rtx;
720 however for some exceptions like fixed registers we use a cost of
721 0. If any other hard register reference occurs, return MAX_COST. */
723 static int
724 approx_reg_cost (rtx x)
726 int cost = 0;
728 if (for_each_rtx (&x, approx_reg_cost_1, (void *) &cost))
729 return MAX_COST;
731 return cost;
734 /* Returns a canonical version of X for the address, from the point of view,
735 that all multiplications are represented as MULT instead of the multiply
736 by a power of 2 being represented as ASHIFT. */
738 static rtx
739 canon_for_address (rtx x)
741 enum rtx_code code;
742 enum machine_mode mode;
743 rtx new = 0;
744 int i;
745 const char *fmt;
747 if (!x)
748 return x;
750 code = GET_CODE (x);
751 mode = GET_MODE (x);
753 switch (code)
755 case ASHIFT:
756 if (GET_CODE (XEXP (x, 1)) == CONST_INT
757 && INTVAL (XEXP (x, 1)) < GET_MODE_BITSIZE (mode)
758 && INTVAL (XEXP (x, 1)) >= 0)
760 new = canon_for_address (XEXP (x, 0));
761 new = gen_rtx_MULT (mode, new,
762 gen_int_mode ((HOST_WIDE_INT) 1
763 << INTVAL (XEXP (x, 1)),
764 mode));
766 break;
767 default:
768 break;
771 if (new)
772 return new;
774 /* Now recursively process each operand of this operation. */
775 fmt = GET_RTX_FORMAT (code);
776 for (i = 0; i < GET_RTX_LENGTH (code); i++)
777 if (fmt[i] == 'e')
779 new = canon_for_address (XEXP (x, i));
780 XEXP (x, i) = new;
782 return x;
785 /* Return a negative value if an rtx A, whose costs are given by COST_A
786 and REGCOST_A, is more desirable than an rtx B.
787 Return a positive value if A is less desirable, or 0 if the two are
788 equally good. */
789 static int
790 preferable (int cost_a, int regcost_a, int cost_b, int regcost_b)
792 /* First, get rid of cases involving expressions that are entirely
793 unwanted. */
794 if (cost_a != cost_b)
796 if (cost_a == MAX_COST)
797 return 1;
798 if (cost_b == MAX_COST)
799 return -1;
802 /* Avoid extending lifetimes of hardregs. */
803 if (regcost_a != regcost_b)
805 if (regcost_a == MAX_COST)
806 return 1;
807 if (regcost_b == MAX_COST)
808 return -1;
811 /* Normal operation costs take precedence. */
812 if (cost_a != cost_b)
813 return cost_a - cost_b;
814 /* Only if these are identical consider effects on register pressure. */
815 if (regcost_a != regcost_b)
816 return regcost_a - regcost_b;
817 return 0;
820 /* Internal function, to compute cost when X is not a register; called
821 from COST macro to keep it simple. */
823 static int
824 notreg_cost (rtx x, enum rtx_code outer)
826 return ((GET_CODE (x) == SUBREG
827 && REG_P (SUBREG_REG (x))
828 && GET_MODE_CLASS (GET_MODE (x)) == MODE_INT
829 && GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_INT
830 && (GET_MODE_SIZE (GET_MODE (x))
831 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
832 && subreg_lowpart_p (x)
833 && TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (GET_MODE (x)),
834 GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (x)))))
836 : rtx_cost (x, outer) * 2);
840 /* Initialize CSE_REG_INFO_TABLE. */
842 static void
843 init_cse_reg_info (unsigned int nregs)
845 /* Do we need to grow the table? */
846 if (nregs > cse_reg_info_table_size)
848 unsigned int new_size;
850 if (cse_reg_info_table_size < 2048)
852 /* Compute a new size that is a power of 2 and no smaller
853 than the large of NREGS and 64. */
854 new_size = (cse_reg_info_table_size
855 ? cse_reg_info_table_size : 64);
857 while (new_size < nregs)
858 new_size *= 2;
860 else
862 /* If we need a big table, allocate just enough to hold
863 NREGS registers. */
864 new_size = nregs;
867 /* Reallocate the table with NEW_SIZE entries. */
868 if (cse_reg_info_table)
869 free (cse_reg_info_table);
870 cse_reg_info_table = XNEWVEC (struct cse_reg_info, new_size);
871 cse_reg_info_table_size = new_size;
872 cse_reg_info_table_first_uninitialized = 0;
875 /* Do we have all of the first NREGS entries initialized? */
876 if (cse_reg_info_table_first_uninitialized < nregs)
878 unsigned int old_timestamp = cse_reg_info_timestamp - 1;
879 unsigned int i;
881 /* Put the old timestamp on newly allocated entries so that they
882 will all be considered out of date. We do not touch those
883 entries beyond the first NREGS entries to be nice to the
884 virtual memory. */
885 for (i = cse_reg_info_table_first_uninitialized; i < nregs; i++)
886 cse_reg_info_table[i].timestamp = old_timestamp;
888 cse_reg_info_table_first_uninitialized = nregs;
892 /* Given REGNO, initialize the cse_reg_info entry for REGNO. */
894 static void
895 get_cse_reg_info_1 (unsigned int regno)
897 /* Set TIMESTAMP field to CSE_REG_INFO_TIMESTAMP so that this
898 entry will be considered to have been initialized. */
899 cse_reg_info_table[regno].timestamp = cse_reg_info_timestamp;
901 /* Initialize the rest of the entry. */
902 cse_reg_info_table[regno].reg_tick = 1;
903 cse_reg_info_table[regno].reg_in_table = -1;
904 cse_reg_info_table[regno].subreg_ticked = -1;
905 cse_reg_info_table[regno].reg_qty = -regno - 1;
908 /* Find a cse_reg_info entry for REGNO. */
910 static inline struct cse_reg_info *
911 get_cse_reg_info (unsigned int regno)
913 struct cse_reg_info *p = &cse_reg_info_table[regno];
915 /* If this entry has not been initialized, go ahead and initialize
916 it. */
917 if (p->timestamp != cse_reg_info_timestamp)
918 get_cse_reg_info_1 (regno);
920 return p;
923 /* Clear the hash table and initialize each register with its own quantity,
924 for a new basic block. */
926 static void
927 new_basic_block (void)
929 int i;
931 next_qty = 0;
933 /* Invalidate cse_reg_info_table. */
934 cse_reg_info_timestamp++;
936 /* Clear out hash table state for this pass. */
937 CLEAR_HARD_REG_SET (hard_regs_in_table);
939 /* The per-quantity values used to be initialized here, but it is
940 much faster to initialize each as it is made in `make_new_qty'. */
942 for (i = 0; i < HASH_SIZE; i++)
944 struct table_elt *first;
946 first = table[i];
947 if (first != NULL)
949 struct table_elt *last = first;
951 table[i] = NULL;
953 while (last->next_same_hash != NULL)
954 last = last->next_same_hash;
956 /* Now relink this hash entire chain into
957 the free element list. */
959 last->next_same_hash = free_element_chain;
960 free_element_chain = first;
964 #ifdef HAVE_cc0
965 prev_insn = 0;
966 prev_insn_cc0 = 0;
967 #endif
970 /* Say that register REG contains a quantity in mode MODE not in any
971 register before and initialize that quantity. */
973 static void
974 make_new_qty (unsigned int reg, enum machine_mode mode)
976 int q;
977 struct qty_table_elem *ent;
978 struct reg_eqv_elem *eqv;
980 gcc_assert (next_qty < max_qty);
982 q = REG_QTY (reg) = next_qty++;
983 ent = &qty_table[q];
984 ent->first_reg = reg;
985 ent->last_reg = reg;
986 ent->mode = mode;
987 ent->const_rtx = ent->const_insn = NULL_RTX;
988 ent->comparison_code = UNKNOWN;
990 eqv = &reg_eqv_table[reg];
991 eqv->next = eqv->prev = -1;
994 /* Make reg NEW equivalent to reg OLD.
995 OLD is not changing; NEW is. */
997 static void
998 make_regs_eqv (unsigned int new, unsigned int old)
1000 unsigned int lastr, firstr;
1001 int q = REG_QTY (old);
1002 struct qty_table_elem *ent;
1004 ent = &qty_table[q];
1006 /* Nothing should become eqv until it has a "non-invalid" qty number. */
1007 gcc_assert (REGNO_QTY_VALID_P (old));
1009 REG_QTY (new) = q;
1010 firstr = ent->first_reg;
1011 lastr = ent->last_reg;
1013 /* Prefer fixed hard registers to anything. Prefer pseudo regs to other
1014 hard regs. Among pseudos, if NEW will live longer than any other reg
1015 of the same qty, and that is beyond the current basic block,
1016 make it the new canonical replacement for this qty. */
1017 if (! (firstr < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (firstr))
1018 /* Certain fixed registers might be of the class NO_REGS. This means
1019 that not only can they not be allocated by the compiler, but
1020 they cannot be used in substitutions or canonicalizations
1021 either. */
1022 && (new >= FIRST_PSEUDO_REGISTER || REGNO_REG_CLASS (new) != NO_REGS)
1023 && ((new < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (new))
1024 || (new >= FIRST_PSEUDO_REGISTER
1025 && (firstr < FIRST_PSEUDO_REGISTER
1026 || ((uid_cuid[REGNO_LAST_UID (new)] > cse_basic_block_end
1027 || (uid_cuid[REGNO_FIRST_UID (new)]
1028 < cse_basic_block_start))
1029 && (uid_cuid[REGNO_LAST_UID (new)]
1030 > uid_cuid[REGNO_LAST_UID (firstr)]))))))
1032 reg_eqv_table[firstr].prev = new;
1033 reg_eqv_table[new].next = firstr;
1034 reg_eqv_table[new].prev = -1;
1035 ent->first_reg = new;
1037 else
1039 /* If NEW is a hard reg (known to be non-fixed), insert at end.
1040 Otherwise, insert before any non-fixed hard regs that are at the
1041 end. Registers of class NO_REGS cannot be used as an
1042 equivalent for anything. */
1043 while (lastr < FIRST_PSEUDO_REGISTER && reg_eqv_table[lastr].prev >= 0
1044 && (REGNO_REG_CLASS (lastr) == NO_REGS || ! FIXED_REGNO_P (lastr))
1045 && new >= FIRST_PSEUDO_REGISTER)
1046 lastr = reg_eqv_table[lastr].prev;
1047 reg_eqv_table[new].next = reg_eqv_table[lastr].next;
1048 if (reg_eqv_table[lastr].next >= 0)
1049 reg_eqv_table[reg_eqv_table[lastr].next].prev = new;
1050 else
1051 qty_table[q].last_reg = new;
1052 reg_eqv_table[lastr].next = new;
1053 reg_eqv_table[new].prev = lastr;
1057 /* Remove REG from its equivalence class. */
1059 static void
1060 delete_reg_equiv (unsigned int reg)
1062 struct qty_table_elem *ent;
1063 int q = REG_QTY (reg);
1064 int p, n;
1066 /* If invalid, do nothing. */
1067 if (! REGNO_QTY_VALID_P (reg))
1068 return;
1070 ent = &qty_table[q];
1072 p = reg_eqv_table[reg].prev;
1073 n = reg_eqv_table[reg].next;
1075 if (n != -1)
1076 reg_eqv_table[n].prev = p;
1077 else
1078 ent->last_reg = p;
1079 if (p != -1)
1080 reg_eqv_table[p].next = n;
1081 else
1082 ent->first_reg = n;
1084 REG_QTY (reg) = -reg - 1;
1087 /* Remove any invalid expressions from the hash table
1088 that refer to any of the registers contained in expression X.
1090 Make sure that newly inserted references to those registers
1091 as subexpressions will be considered valid.
1093 mention_regs is not called when a register itself
1094 is being stored in the table.
1096 Return 1 if we have done something that may have changed the hash code
1097 of X. */
1099 static int
1100 mention_regs (rtx x)
1102 enum rtx_code code;
1103 int i, j;
1104 const char *fmt;
1105 int changed = 0;
1107 if (x == 0)
1108 return 0;
1110 code = GET_CODE (x);
1111 if (code == REG)
1113 unsigned int regno = REGNO (x);
1114 unsigned int endregno
1115 = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
1116 : hard_regno_nregs[regno][GET_MODE (x)]);
1117 unsigned int i;
1119 for (i = regno; i < endregno; i++)
1121 if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
1122 remove_invalid_refs (i);
1124 REG_IN_TABLE (i) = REG_TICK (i);
1125 SUBREG_TICKED (i) = -1;
1128 return 0;
1131 /* If this is a SUBREG, we don't want to discard other SUBREGs of the same
1132 pseudo if they don't use overlapping words. We handle only pseudos
1133 here for simplicity. */
1134 if (code == SUBREG && REG_P (SUBREG_REG (x))
1135 && REGNO (SUBREG_REG (x)) >= FIRST_PSEUDO_REGISTER)
1137 unsigned int i = REGNO (SUBREG_REG (x));
1139 if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
1141 /* If REG_IN_TABLE (i) differs from REG_TICK (i) by one, and
1142 the last store to this register really stored into this
1143 subreg, then remove the memory of this subreg.
1144 Otherwise, remove any memory of the entire register and
1145 all its subregs from the table. */
1146 if (REG_TICK (i) - REG_IN_TABLE (i) > 1
1147 || SUBREG_TICKED (i) != REGNO (SUBREG_REG (x)))
1148 remove_invalid_refs (i);
1149 else
1150 remove_invalid_subreg_refs (i, SUBREG_BYTE (x), GET_MODE (x));
1153 REG_IN_TABLE (i) = REG_TICK (i);
1154 SUBREG_TICKED (i) = REGNO (SUBREG_REG (x));
1155 return 0;
1158 /* If X is a comparison or a COMPARE and either operand is a register
1159 that does not have a quantity, give it one. This is so that a later
1160 call to record_jump_equiv won't cause X to be assigned a different
1161 hash code and not found in the table after that call.
1163 It is not necessary to do this here, since rehash_using_reg can
1164 fix up the table later, but doing this here eliminates the need to
1165 call that expensive function in the most common case where the only
1166 use of the register is in the comparison. */
1168 if (code == COMPARE || COMPARISON_P (x))
1170 if (REG_P (XEXP (x, 0))
1171 && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 0))))
1172 if (insert_regs (XEXP (x, 0), NULL, 0))
1174 rehash_using_reg (XEXP (x, 0));
1175 changed = 1;
1178 if (REG_P (XEXP (x, 1))
1179 && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 1))))
1180 if (insert_regs (XEXP (x, 1), NULL, 0))
1182 rehash_using_reg (XEXP (x, 1));
1183 changed = 1;
1187 fmt = GET_RTX_FORMAT (code);
1188 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
1189 if (fmt[i] == 'e')
1190 changed |= mention_regs (XEXP (x, i));
1191 else if (fmt[i] == 'E')
1192 for (j = 0; j < XVECLEN (x, i); j++)
1193 changed |= mention_regs (XVECEXP (x, i, j));
1195 return changed;
1198 /* Update the register quantities for inserting X into the hash table
1199 with a value equivalent to CLASSP.
1200 (If the class does not contain a REG, it is irrelevant.)
1201 If MODIFIED is nonzero, X is a destination; it is being modified.
1202 Note that delete_reg_equiv should be called on a register
1203 before insert_regs is done on that register with MODIFIED != 0.
1205 Nonzero value means that elements of reg_qty have changed
1206 so X's hash code may be different. */
1208 static int
1209 insert_regs (rtx x, struct table_elt *classp, int modified)
1211 if (REG_P (x))
1213 unsigned int regno = REGNO (x);
1214 int qty_valid;
1216 /* If REGNO is in the equivalence table already but is of the
1217 wrong mode for that equivalence, don't do anything here. */
1219 qty_valid = REGNO_QTY_VALID_P (regno);
1220 if (qty_valid)
1222 struct qty_table_elem *ent = &qty_table[REG_QTY (regno)];
1224 if (ent->mode != GET_MODE (x))
1225 return 0;
1228 if (modified || ! qty_valid)
1230 if (classp)
1231 for (classp = classp->first_same_value;
1232 classp != 0;
1233 classp = classp->next_same_value)
1234 if (REG_P (classp->exp)
1235 && GET_MODE (classp->exp) == GET_MODE (x))
1237 unsigned c_regno = REGNO (classp->exp);
1239 gcc_assert (REGNO_QTY_VALID_P (c_regno));
1241 /* Suppose that 5 is hard reg and 100 and 101 are
1242 pseudos. Consider
1244 (set (reg:si 100) (reg:si 5))
1245 (set (reg:si 5) (reg:si 100))
1246 (set (reg:di 101) (reg:di 5))
1248 We would now set REG_QTY (101) = REG_QTY (5), but the
1249 entry for 5 is in SImode. When we use this later in
1250 copy propagation, we get the register in wrong mode. */
1251 if (qty_table[REG_QTY (c_regno)].mode != GET_MODE (x))
1252 continue;
1254 make_regs_eqv (regno, c_regno);
1255 return 1;
1258 /* Mention_regs for a SUBREG checks if REG_TICK is exactly one larger
1259 than REG_IN_TABLE to find out if there was only a single preceding
1260 invalidation - for the SUBREG - or another one, which would be
1261 for the full register. However, if we find here that REG_TICK
1262 indicates that the register is invalid, it means that it has
1263 been invalidated in a separate operation. The SUBREG might be used
1264 now (then this is a recursive call), or we might use the full REG
1265 now and a SUBREG of it later. So bump up REG_TICK so that
1266 mention_regs will do the right thing. */
1267 if (! modified
1268 && REG_IN_TABLE (regno) >= 0
1269 && REG_TICK (regno) == REG_IN_TABLE (regno) + 1)
1270 REG_TICK (regno)++;
1271 make_new_qty (regno, GET_MODE (x));
1272 return 1;
1275 return 0;
1278 /* If X is a SUBREG, we will likely be inserting the inner register in the
1279 table. If that register doesn't have an assigned quantity number at
1280 this point but does later, the insertion that we will be doing now will
1281 not be accessible because its hash code will have changed. So assign
1282 a quantity number now. */
1284 else if (GET_CODE (x) == SUBREG && REG_P (SUBREG_REG (x))
1285 && ! REGNO_QTY_VALID_P (REGNO (SUBREG_REG (x))))
1287 insert_regs (SUBREG_REG (x), NULL, 0);
1288 mention_regs (x);
1289 return 1;
1291 else
1292 return mention_regs (x);
1295 /* Look in or update the hash table. */
1297 /* Remove table element ELT from use in the table.
1298 HASH is its hash code, made using the HASH macro.
1299 It's an argument because often that is known in advance
1300 and we save much time not recomputing it. */
1302 static void
1303 remove_from_table (struct table_elt *elt, unsigned int hash)
1305 if (elt == 0)
1306 return;
1308 /* Mark this element as removed. See cse_insn. */
1309 elt->first_same_value = 0;
1311 /* Remove the table element from its equivalence class. */
1314 struct table_elt *prev = elt->prev_same_value;
1315 struct table_elt *next = elt->next_same_value;
1317 if (next)
1318 next->prev_same_value = prev;
1320 if (prev)
1321 prev->next_same_value = next;
1322 else
1324 struct table_elt *newfirst = next;
1325 while (next)
1327 next->first_same_value = newfirst;
1328 next = next->next_same_value;
1333 /* Remove the table element from its hash bucket. */
1336 struct table_elt *prev = elt->prev_same_hash;
1337 struct table_elt *next = elt->next_same_hash;
1339 if (next)
1340 next->prev_same_hash = prev;
1342 if (prev)
1343 prev->next_same_hash = next;
1344 else if (table[hash] == elt)
1345 table[hash] = next;
1346 else
1348 /* This entry is not in the proper hash bucket. This can happen
1349 when two classes were merged by `merge_equiv_classes'. Search
1350 for the hash bucket that it heads. This happens only very
1351 rarely, so the cost is acceptable. */
1352 for (hash = 0; hash < HASH_SIZE; hash++)
1353 if (table[hash] == elt)
1354 table[hash] = next;
1358 /* Remove the table element from its related-value circular chain. */
1360 if (elt->related_value != 0 && elt->related_value != elt)
1362 struct table_elt *p = elt->related_value;
1364 while (p->related_value != elt)
1365 p = p->related_value;
1366 p->related_value = elt->related_value;
1367 if (p->related_value == p)
1368 p->related_value = 0;
1371 /* Now add it to the free element chain. */
1372 elt->next_same_hash = free_element_chain;
1373 free_element_chain = elt;
1376 /* Look up X in the hash table and return its table element,
1377 or 0 if X is not in the table.
1379 MODE is the machine-mode of X, or if X is an integer constant
1380 with VOIDmode then MODE is the mode with which X will be used.
1382 Here we are satisfied to find an expression whose tree structure
1383 looks like X. */
1385 static struct table_elt *
1386 lookup (rtx x, unsigned int hash, enum machine_mode mode)
1388 struct table_elt *p;
1390 for (p = table[hash]; p; p = p->next_same_hash)
1391 if (mode == p->mode && ((x == p->exp && REG_P (x))
1392 || exp_equiv_p (x, p->exp, !REG_P (x), false)))
1393 return p;
1395 return 0;
1398 /* Like `lookup' but don't care whether the table element uses invalid regs.
1399 Also ignore discrepancies in the machine mode of a register. */
1401 static struct table_elt *
1402 lookup_for_remove (rtx x, unsigned int hash, enum machine_mode mode)
1404 struct table_elt *p;
1406 if (REG_P (x))
1408 unsigned int regno = REGNO (x);
1410 /* Don't check the machine mode when comparing registers;
1411 invalidating (REG:SI 0) also invalidates (REG:DF 0). */
1412 for (p = table[hash]; p; p = p->next_same_hash)
1413 if (REG_P (p->exp)
1414 && REGNO (p->exp) == regno)
1415 return p;
1417 else
1419 for (p = table[hash]; p; p = p->next_same_hash)
1420 if (mode == p->mode
1421 && (x == p->exp || exp_equiv_p (x, p->exp, 0, false)))
1422 return p;
1425 return 0;
1428 /* Look for an expression equivalent to X and with code CODE.
1429 If one is found, return that expression. */
1431 static rtx
1432 lookup_as_function (rtx x, enum rtx_code code)
1434 struct table_elt *p
1435 = lookup (x, SAFE_HASH (x, VOIDmode), GET_MODE (x));
1437 /* If we are looking for a CONST_INT, the mode doesn't really matter, as
1438 long as we are narrowing. So if we looked in vain for a mode narrower
1439 than word_mode before, look for word_mode now. */
1440 if (p == 0 && code == CONST_INT
1441 && GET_MODE_SIZE (GET_MODE (x)) < GET_MODE_SIZE (word_mode))
1443 x = copy_rtx (x);
1444 PUT_MODE (x, word_mode);
1445 p = lookup (x, SAFE_HASH (x, VOIDmode), word_mode);
1448 if (p == 0)
1449 return 0;
1451 for (p = p->first_same_value; p; p = p->next_same_value)
1452 if (GET_CODE (p->exp) == code
1453 /* Make sure this is a valid entry in the table. */
1454 && exp_equiv_p (p->exp, p->exp, 1, false))
1455 return p->exp;
1457 return 0;
1460 /* Insert X in the hash table, assuming HASH is its hash code
1461 and CLASSP is an element of the class it should go in
1462 (or 0 if a new class should be made).
1463 It is inserted at the proper position to keep the class in
1464 the order cheapest first.
1466 MODE is the machine-mode of X, or if X is an integer constant
1467 with VOIDmode then MODE is the mode with which X will be used.
1469 For elements of equal cheapness, the most recent one
1470 goes in front, except that the first element in the list
1471 remains first unless a cheaper element is added. The order of
1472 pseudo-registers does not matter, as canon_reg will be called to
1473 find the cheapest when a register is retrieved from the table.
1475 The in_memory field in the hash table element is set to 0.
1476 The caller must set it nonzero if appropriate.
1478 You should call insert_regs (X, CLASSP, MODIFY) before calling here,
1479 and if insert_regs returns a nonzero value
1480 you must then recompute its hash code before calling here.
1482 If necessary, update table showing constant values of quantities. */
1484 #define CHEAPER(X, Y) \
1485 (preferable ((X)->cost, (X)->regcost, (Y)->cost, (Y)->regcost) < 0)
1487 static struct table_elt *
1488 insert (rtx x, struct table_elt *classp, unsigned int hash, enum machine_mode mode)
1490 struct table_elt *elt;
1492 /* If X is a register and we haven't made a quantity for it,
1493 something is wrong. */
1494 gcc_assert (!REG_P (x) || REGNO_QTY_VALID_P (REGNO (x)));
1496 /* If X is a hard register, show it is being put in the table. */
1497 if (REG_P (x) && REGNO (x) < FIRST_PSEUDO_REGISTER)
1499 unsigned int regno = REGNO (x);
1500 unsigned int endregno = regno + hard_regno_nregs[regno][GET_MODE (x)];
1501 unsigned int i;
1503 for (i = regno; i < endregno; i++)
1504 SET_HARD_REG_BIT (hard_regs_in_table, i);
1507 /* Put an element for X into the right hash bucket. */
1509 elt = free_element_chain;
1510 if (elt)
1511 free_element_chain = elt->next_same_hash;
1512 else
1513 elt = XNEW (struct table_elt);
1515 elt->exp = x;
1516 elt->canon_exp = NULL_RTX;
1517 elt->cost = COST (x);
1518 elt->regcost = approx_reg_cost (x);
1519 elt->next_same_value = 0;
1520 elt->prev_same_value = 0;
1521 elt->next_same_hash = table[hash];
1522 elt->prev_same_hash = 0;
1523 elt->related_value = 0;
1524 elt->in_memory = 0;
1525 elt->mode = mode;
1526 elt->is_const = (CONSTANT_P (x) || fixed_base_plus_p (x));
1528 if (table[hash])
1529 table[hash]->prev_same_hash = elt;
1530 table[hash] = elt;
1532 /* Put it into the proper value-class. */
1533 if (classp)
1535 classp = classp->first_same_value;
1536 if (CHEAPER (elt, classp))
1537 /* Insert at the head of the class. */
1539 struct table_elt *p;
1540 elt->next_same_value = classp;
1541 classp->prev_same_value = elt;
1542 elt->first_same_value = elt;
1544 for (p = classp; p; p = p->next_same_value)
1545 p->first_same_value = elt;
1547 else
1549 /* Insert not at head of the class. */
1550 /* Put it after the last element cheaper than X. */
1551 struct table_elt *p, *next;
1553 for (p = classp; (next = p->next_same_value) && CHEAPER (next, elt);
1554 p = next);
1556 /* Put it after P and before NEXT. */
1557 elt->next_same_value = next;
1558 if (next)
1559 next->prev_same_value = elt;
1561 elt->prev_same_value = p;
1562 p->next_same_value = elt;
1563 elt->first_same_value = classp;
1566 else
1567 elt->first_same_value = elt;
1569 /* If this is a constant being set equivalent to a register or a register
1570 being set equivalent to a constant, note the constant equivalence.
1572 If this is a constant, it cannot be equivalent to a different constant,
1573 and a constant is the only thing that can be cheaper than a register. So
1574 we know the register is the head of the class (before the constant was
1575 inserted).
1577 If this is a register that is not already known equivalent to a
1578 constant, we must check the entire class.
1580 If this is a register that is already known equivalent to an insn,
1581 update the qtys `const_insn' to show that `this_insn' is the latest
1582 insn making that quantity equivalent to the constant. */
1584 if (elt->is_const && classp && REG_P (classp->exp)
1585 && !REG_P (x))
1587 int exp_q = REG_QTY (REGNO (classp->exp));
1588 struct qty_table_elem *exp_ent = &qty_table[exp_q];
1590 exp_ent->const_rtx = gen_lowpart (exp_ent->mode, x);
1591 exp_ent->const_insn = this_insn;
1594 else if (REG_P (x)
1595 && classp
1596 && ! qty_table[REG_QTY (REGNO (x))].const_rtx
1597 && ! elt->is_const)
1599 struct table_elt *p;
1601 for (p = classp; p != 0; p = p->next_same_value)
1603 if (p->is_const && !REG_P (p->exp))
1605 int x_q = REG_QTY (REGNO (x));
1606 struct qty_table_elem *x_ent = &qty_table[x_q];
1608 x_ent->const_rtx
1609 = gen_lowpart (GET_MODE (x), p->exp);
1610 x_ent->const_insn = this_insn;
1611 break;
1616 else if (REG_P (x)
1617 && qty_table[REG_QTY (REGNO (x))].const_rtx
1618 && GET_MODE (x) == qty_table[REG_QTY (REGNO (x))].mode)
1619 qty_table[REG_QTY (REGNO (x))].const_insn = this_insn;
1621 /* If this is a constant with symbolic value,
1622 and it has a term with an explicit integer value,
1623 link it up with related expressions. */
1624 if (GET_CODE (x) == CONST)
1626 rtx subexp = get_related_value (x);
1627 unsigned subhash;
1628 struct table_elt *subelt, *subelt_prev;
1630 if (subexp != 0)
1632 /* Get the integer-free subexpression in the hash table. */
1633 subhash = SAFE_HASH (subexp, mode);
1634 subelt = lookup (subexp, subhash, mode);
1635 if (subelt == 0)
1636 subelt = insert (subexp, NULL, subhash, mode);
1637 /* Initialize SUBELT's circular chain if it has none. */
1638 if (subelt->related_value == 0)
1639 subelt->related_value = subelt;
1640 /* Find the element in the circular chain that precedes SUBELT. */
1641 subelt_prev = subelt;
1642 while (subelt_prev->related_value != subelt)
1643 subelt_prev = subelt_prev->related_value;
1644 /* Put new ELT into SUBELT's circular chain just before SUBELT.
1645 This way the element that follows SUBELT is the oldest one. */
1646 elt->related_value = subelt_prev->related_value;
1647 subelt_prev->related_value = elt;
1651 return elt;
1654 /* Given two equivalence classes, CLASS1 and CLASS2, put all the entries from
1655 CLASS2 into CLASS1. This is done when we have reached an insn which makes
1656 the two classes equivalent.
1658 CLASS1 will be the surviving class; CLASS2 should not be used after this
1659 call.
1661 Any invalid entries in CLASS2 will not be copied. */
1663 static void
1664 merge_equiv_classes (struct table_elt *class1, struct table_elt *class2)
1666 struct table_elt *elt, *next, *new;
1668 /* Ensure we start with the head of the classes. */
1669 class1 = class1->first_same_value;
1670 class2 = class2->first_same_value;
1672 /* If they were already equal, forget it. */
1673 if (class1 == class2)
1674 return;
1676 for (elt = class2; elt; elt = next)
1678 unsigned int hash;
1679 rtx exp = elt->exp;
1680 enum machine_mode mode = elt->mode;
1682 next = elt->next_same_value;
1684 /* Remove old entry, make a new one in CLASS1's class.
1685 Don't do this for invalid entries as we cannot find their
1686 hash code (it also isn't necessary). */
1687 if (REG_P (exp) || exp_equiv_p (exp, exp, 1, false))
1689 bool need_rehash = false;
1691 hash_arg_in_memory = 0;
1692 hash = HASH (exp, mode);
1694 if (REG_P (exp))
1696 need_rehash = REGNO_QTY_VALID_P (REGNO (exp));
1697 delete_reg_equiv (REGNO (exp));
1700 remove_from_table (elt, hash);
1702 if (insert_regs (exp, class1, 0) || need_rehash)
1704 rehash_using_reg (exp);
1705 hash = HASH (exp, mode);
1707 new = insert (exp, class1, hash, mode);
1708 new->in_memory = hash_arg_in_memory;
1713 /* Flush the entire hash table. */
1715 static void
1716 flush_hash_table (void)
1718 int i;
1719 struct table_elt *p;
1721 for (i = 0; i < HASH_SIZE; i++)
1722 for (p = table[i]; p; p = table[i])
1724 /* Note that invalidate can remove elements
1725 after P in the current hash chain. */
1726 if (REG_P (p->exp))
1727 invalidate (p->exp, p->mode);
1728 else
1729 remove_from_table (p, i);
1733 /* Function called for each rtx to check whether true dependence exist. */
1734 struct check_dependence_data
1736 enum machine_mode mode;
1737 rtx exp;
1738 rtx addr;
1741 static int
1742 check_dependence (rtx *x, void *data)
1744 struct check_dependence_data *d = (struct check_dependence_data *) data;
1745 if (*x && MEM_P (*x))
1746 return canon_true_dependence (d->exp, d->mode, d->addr, *x,
1747 cse_rtx_varies_p);
1748 else
1749 return 0;
1752 /* Remove from the hash table, or mark as invalid, all expressions whose
1753 values could be altered by storing in X. X is a register, a subreg, or
1754 a memory reference with nonvarying address (because, when a memory
1755 reference with a varying address is stored in, all memory references are
1756 removed by invalidate_memory so specific invalidation is superfluous).
1757 FULL_MODE, if not VOIDmode, indicates that this much should be
1758 invalidated instead of just the amount indicated by the mode of X. This
1759 is only used for bitfield stores into memory.
1761 A nonvarying address may be just a register or just a symbol reference,
1762 or it may be either of those plus a numeric offset. */
1764 static void
1765 invalidate (rtx x, enum machine_mode full_mode)
1767 int i;
1768 struct table_elt *p;
1769 rtx addr;
1771 switch (GET_CODE (x))
1773 case REG:
1775 /* If X is a register, dependencies on its contents are recorded
1776 through the qty number mechanism. Just change the qty number of
1777 the register, mark it as invalid for expressions that refer to it,
1778 and remove it itself. */
1779 unsigned int regno = REGNO (x);
1780 unsigned int hash = HASH (x, GET_MODE (x));
1782 /* Remove REGNO from any quantity list it might be on and indicate
1783 that its value might have changed. If it is a pseudo, remove its
1784 entry from the hash table.
1786 For a hard register, we do the first two actions above for any
1787 additional hard registers corresponding to X. Then, if any of these
1788 registers are in the table, we must remove any REG entries that
1789 overlap these registers. */
1791 delete_reg_equiv (regno);
1792 REG_TICK (regno)++;
1793 SUBREG_TICKED (regno) = -1;
1795 if (regno >= FIRST_PSEUDO_REGISTER)
1797 /* Because a register can be referenced in more than one mode,
1798 we might have to remove more than one table entry. */
1799 struct table_elt *elt;
1801 while ((elt = lookup_for_remove (x, hash, GET_MODE (x))))
1802 remove_from_table (elt, hash);
1804 else
1806 HOST_WIDE_INT in_table
1807 = TEST_HARD_REG_BIT (hard_regs_in_table, regno);
1808 unsigned int endregno
1809 = regno + hard_regno_nregs[regno][GET_MODE (x)];
1810 unsigned int tregno, tendregno, rn;
1811 struct table_elt *p, *next;
1813 CLEAR_HARD_REG_BIT (hard_regs_in_table, regno);
1815 for (rn = regno + 1; rn < endregno; rn++)
1817 in_table |= TEST_HARD_REG_BIT (hard_regs_in_table, rn);
1818 CLEAR_HARD_REG_BIT (hard_regs_in_table, rn);
1819 delete_reg_equiv (rn);
1820 REG_TICK (rn)++;
1821 SUBREG_TICKED (rn) = -1;
1824 if (in_table)
1825 for (hash = 0; hash < HASH_SIZE; hash++)
1826 for (p = table[hash]; p; p = next)
1828 next = p->next_same_hash;
1830 if (!REG_P (p->exp)
1831 || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
1832 continue;
1834 tregno = REGNO (p->exp);
1835 tendregno
1836 = tregno + hard_regno_nregs[tregno][GET_MODE (p->exp)];
1837 if (tendregno > regno && tregno < endregno)
1838 remove_from_table (p, hash);
1842 return;
1844 case SUBREG:
1845 invalidate (SUBREG_REG (x), VOIDmode);
1846 return;
1848 case PARALLEL:
1849 for (i = XVECLEN (x, 0) - 1; i >= 0; --i)
1850 invalidate (XVECEXP (x, 0, i), VOIDmode);
1851 return;
1853 case EXPR_LIST:
1854 /* This is part of a disjoint return value; extract the location in
1855 question ignoring the offset. */
1856 invalidate (XEXP (x, 0), VOIDmode);
1857 return;
1859 case MEM:
1860 addr = canon_rtx (get_addr (XEXP (x, 0)));
1861 /* Calculate the canonical version of X here so that
1862 true_dependence doesn't generate new RTL for X on each call. */
1863 x = canon_rtx (x);
1865 /* Remove all hash table elements that refer to overlapping pieces of
1866 memory. */
1867 if (full_mode == VOIDmode)
1868 full_mode = GET_MODE (x);
1870 for (i = 0; i < HASH_SIZE; i++)
1872 struct table_elt *next;
1874 for (p = table[i]; p; p = next)
1876 next = p->next_same_hash;
1877 if (p->in_memory)
1879 struct check_dependence_data d;
1881 /* Just canonicalize the expression once;
1882 otherwise each time we call invalidate
1883 true_dependence will canonicalize the
1884 expression again. */
1885 if (!p->canon_exp)
1886 p->canon_exp = canon_rtx (p->exp);
1887 d.exp = x;
1888 d.addr = addr;
1889 d.mode = full_mode;
1890 if (for_each_rtx (&p->canon_exp, check_dependence, &d))
1891 remove_from_table (p, i);
1895 return;
1897 default:
1898 gcc_unreachable ();
1902 /* Remove all expressions that refer to register REGNO,
1903 since they are already invalid, and we are about to
1904 mark that register valid again and don't want the old
1905 expressions to reappear as valid. */
1907 static void
1908 remove_invalid_refs (unsigned int regno)
1910 unsigned int i;
1911 struct table_elt *p, *next;
1913 for (i = 0; i < HASH_SIZE; i++)
1914 for (p = table[i]; p; p = next)
1916 next = p->next_same_hash;
1917 if (!REG_P (p->exp)
1918 && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
1919 remove_from_table (p, i);
1923 /* Likewise for a subreg with subreg_reg REGNO, subreg_byte OFFSET,
1924 and mode MODE. */
1925 static void
1926 remove_invalid_subreg_refs (unsigned int regno, unsigned int offset,
1927 enum machine_mode mode)
1929 unsigned int i;
1930 struct table_elt *p, *next;
1931 unsigned int end = offset + (GET_MODE_SIZE (mode) - 1);
1933 for (i = 0; i < HASH_SIZE; i++)
1934 for (p = table[i]; p; p = next)
1936 rtx exp = p->exp;
1937 next = p->next_same_hash;
1939 if (!REG_P (exp)
1940 && (GET_CODE (exp) != SUBREG
1941 || !REG_P (SUBREG_REG (exp))
1942 || REGNO (SUBREG_REG (exp)) != regno
1943 || (((SUBREG_BYTE (exp)
1944 + (GET_MODE_SIZE (GET_MODE (exp)) - 1)) >= offset)
1945 && SUBREG_BYTE (exp) <= end))
1946 && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
1947 remove_from_table (p, i);
1951 /* Recompute the hash codes of any valid entries in the hash table that
1952 reference X, if X is a register, or SUBREG_REG (X) if X is a SUBREG.
1954 This is called when we make a jump equivalence. */
1956 static void
1957 rehash_using_reg (rtx x)
1959 unsigned int i;
1960 struct table_elt *p, *next;
1961 unsigned hash;
1963 if (GET_CODE (x) == SUBREG)
1964 x = SUBREG_REG (x);
1966 /* If X is not a register or if the register is known not to be in any
1967 valid entries in the table, we have no work to do. */
1969 if (!REG_P (x)
1970 || REG_IN_TABLE (REGNO (x)) < 0
1971 || REG_IN_TABLE (REGNO (x)) != REG_TICK (REGNO (x)))
1972 return;
1974 /* Scan all hash chains looking for valid entries that mention X.
1975 If we find one and it is in the wrong hash chain, move it. */
1977 for (i = 0; i < HASH_SIZE; i++)
1978 for (p = table[i]; p; p = next)
1980 next = p->next_same_hash;
1981 if (reg_mentioned_p (x, p->exp)
1982 && exp_equiv_p (p->exp, p->exp, 1, false)
1983 && i != (hash = SAFE_HASH (p->exp, p->mode)))
1985 if (p->next_same_hash)
1986 p->next_same_hash->prev_same_hash = p->prev_same_hash;
1988 if (p->prev_same_hash)
1989 p->prev_same_hash->next_same_hash = p->next_same_hash;
1990 else
1991 table[i] = p->next_same_hash;
1993 p->next_same_hash = table[hash];
1994 p->prev_same_hash = 0;
1995 if (table[hash])
1996 table[hash]->prev_same_hash = p;
1997 table[hash] = p;
2002 /* Remove from the hash table any expression that is a call-clobbered
2003 register. Also update their TICK values. */
2005 static void
2006 invalidate_for_call (void)
2008 unsigned int regno, endregno;
2009 unsigned int i;
2010 unsigned hash;
2011 struct table_elt *p, *next;
2012 int in_table = 0;
2014 /* Go through all the hard registers. For each that is clobbered in
2015 a CALL_INSN, remove the register from quantity chains and update
2016 reg_tick if defined. Also see if any of these registers is currently
2017 in the table. */
2019 for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
2020 if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
2022 delete_reg_equiv (regno);
2023 if (REG_TICK (regno) >= 0)
2025 REG_TICK (regno)++;
2026 SUBREG_TICKED (regno) = -1;
2029 in_table |= (TEST_HARD_REG_BIT (hard_regs_in_table, regno) != 0);
2032 /* In the case where we have no call-clobbered hard registers in the
2033 table, we are done. Otherwise, scan the table and remove any
2034 entry that overlaps a call-clobbered register. */
2036 if (in_table)
2037 for (hash = 0; hash < HASH_SIZE; hash++)
2038 for (p = table[hash]; p; p = next)
2040 next = p->next_same_hash;
2042 if (!REG_P (p->exp)
2043 || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
2044 continue;
2046 regno = REGNO (p->exp);
2047 endregno = regno + hard_regno_nregs[regno][GET_MODE (p->exp)];
2049 for (i = regno; i < endregno; i++)
2050 if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
2052 remove_from_table (p, hash);
2053 break;
2058 /* Given an expression X of type CONST,
2059 and ELT which is its table entry (or 0 if it
2060 is not in the hash table),
2061 return an alternate expression for X as a register plus integer.
2062 If none can be found, return 0. */
2064 static rtx
2065 use_related_value (rtx x, struct table_elt *elt)
2067 struct table_elt *relt = 0;
2068 struct table_elt *p, *q;
2069 HOST_WIDE_INT offset;
2071 /* First, is there anything related known?
2072 If we have a table element, we can tell from that.
2073 Otherwise, must look it up. */
2075 if (elt != 0 && elt->related_value != 0)
2076 relt = elt;
2077 else if (elt == 0 && GET_CODE (x) == CONST)
2079 rtx subexp = get_related_value (x);
2080 if (subexp != 0)
2081 relt = lookup (subexp,
2082 SAFE_HASH (subexp, GET_MODE (subexp)),
2083 GET_MODE (subexp));
2086 if (relt == 0)
2087 return 0;
2089 /* Search all related table entries for one that has an
2090 equivalent register. */
2092 p = relt;
2093 while (1)
2095 /* This loop is strange in that it is executed in two different cases.
2096 The first is when X is already in the table. Then it is searching
2097 the RELATED_VALUE list of X's class (RELT). The second case is when
2098 X is not in the table. Then RELT points to a class for the related
2099 value.
2101 Ensure that, whatever case we are in, that we ignore classes that have
2102 the same value as X. */
2104 if (rtx_equal_p (x, p->exp))
2105 q = 0;
2106 else
2107 for (q = p->first_same_value; q; q = q->next_same_value)
2108 if (REG_P (q->exp))
2109 break;
2111 if (q)
2112 break;
2114 p = p->related_value;
2116 /* We went all the way around, so there is nothing to be found.
2117 Alternatively, perhaps RELT was in the table for some other reason
2118 and it has no related values recorded. */
2119 if (p == relt || p == 0)
2120 break;
2123 if (q == 0)
2124 return 0;
2126 offset = (get_integer_term (x) - get_integer_term (p->exp));
2127 /* Note: OFFSET may be 0 if P->xexp and X are related by commutativity. */
2128 return plus_constant (q->exp, offset);
2131 /* Hash a string. Just add its bytes up. */
2132 static inline unsigned
2133 hash_rtx_string (const char *ps)
2135 unsigned hash = 0;
2136 const unsigned char *p = (const unsigned char *) ps;
2138 if (p)
2139 while (*p)
2140 hash += *p++;
2142 return hash;
2145 /* Hash an rtx. We are careful to make sure the value is never negative.
2146 Equivalent registers hash identically.
2147 MODE is used in hashing for CONST_INTs only;
2148 otherwise the mode of X is used.
2150 Store 1 in DO_NOT_RECORD_P if any subexpression is volatile.
2152 If HASH_ARG_IN_MEMORY_P is not NULL, store 1 in it if X contains
2153 a MEM rtx which does not have the RTX_UNCHANGING_P bit set.
2155 Note that cse_insn knows that the hash code of a MEM expression
2156 is just (int) MEM plus the hash code of the address. */
2158 unsigned
2159 hash_rtx (rtx x, enum machine_mode mode, int *do_not_record_p,
2160 int *hash_arg_in_memory_p, bool have_reg_qty)
2162 int i, j;
2163 unsigned hash = 0;
2164 enum rtx_code code;
2165 const char *fmt;
2167 /* Used to turn recursion into iteration. We can't rely on GCC's
2168 tail-recursion elimination since we need to keep accumulating values
2169 in HASH. */
2170 repeat:
2171 if (x == 0)
2172 return hash;
2174 code = GET_CODE (x);
2175 switch (code)
2177 case REG:
2179 unsigned int regno = REGNO (x);
2181 if (!reload_completed)
2183 /* On some machines, we can't record any non-fixed hard register,
2184 because extending its life will cause reload problems. We
2185 consider ap, fp, sp, gp to be fixed for this purpose.
2187 We also consider CCmode registers to be fixed for this purpose;
2188 failure to do so leads to failure to simplify 0<100 type of
2189 conditionals.
2191 On all machines, we can't record any global registers.
2192 Nor should we record any register that is in a small
2193 class, as defined by CLASS_LIKELY_SPILLED_P. */
2194 bool record;
2196 if (regno >= FIRST_PSEUDO_REGISTER)
2197 record = true;
2198 else if (x == frame_pointer_rtx
2199 || x == hard_frame_pointer_rtx
2200 || x == arg_pointer_rtx
2201 || x == stack_pointer_rtx
2202 || x == pic_offset_table_rtx)
2203 record = true;
2204 else if (global_regs[regno])
2205 record = false;
2206 else if (fixed_regs[regno])
2207 record = true;
2208 else if (GET_MODE_CLASS (GET_MODE (x)) == MODE_CC)
2209 record = true;
2210 else if (SMALL_REGISTER_CLASSES)
2211 record = false;
2212 else if (CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (regno)))
2213 record = false;
2214 else
2215 record = true;
2217 if (!record)
2219 *do_not_record_p = 1;
2220 return 0;
2224 hash += ((unsigned int) REG << 7);
2225 hash += (have_reg_qty ? (unsigned) REG_QTY (regno) : regno);
2226 return hash;
2229 /* We handle SUBREG of a REG specially because the underlying
2230 reg changes its hash value with every value change; we don't
2231 want to have to forget unrelated subregs when one subreg changes. */
2232 case SUBREG:
2234 if (REG_P (SUBREG_REG (x)))
2236 hash += (((unsigned int) SUBREG << 7)
2237 + REGNO (SUBREG_REG (x))
2238 + (SUBREG_BYTE (x) / UNITS_PER_WORD));
2239 return hash;
2241 break;
2244 case CONST_INT:
2245 hash += (((unsigned int) CONST_INT << 7) + (unsigned int) mode
2246 + (unsigned int) INTVAL (x));
2247 return hash;
2249 case CONST_DOUBLE:
2250 /* This is like the general case, except that it only counts
2251 the integers representing the constant. */
2252 hash += (unsigned int) code + (unsigned int) GET_MODE (x);
2253 if (GET_MODE (x) != VOIDmode)
2254 hash += real_hash (CONST_DOUBLE_REAL_VALUE (x));
2255 else
2256 hash += ((unsigned int) CONST_DOUBLE_LOW (x)
2257 + (unsigned int) CONST_DOUBLE_HIGH (x));
2258 return hash;
2260 case CONST_VECTOR:
2262 int units;
2263 rtx elt;
2265 units = CONST_VECTOR_NUNITS (x);
2267 for (i = 0; i < units; ++i)
2269 elt = CONST_VECTOR_ELT (x, i);
2270 hash += hash_rtx (elt, GET_MODE (elt), do_not_record_p,
2271 hash_arg_in_memory_p, have_reg_qty);
2274 return hash;
2277 /* Assume there is only one rtx object for any given label. */
2278 case LABEL_REF:
2279 /* We don't hash on the address of the CODE_LABEL to avoid bootstrap
2280 differences and differences between each stage's debugging dumps. */
2281 hash += (((unsigned int) LABEL_REF << 7)
2282 + CODE_LABEL_NUMBER (XEXP (x, 0)));
2283 return hash;
2285 case SYMBOL_REF:
2287 /* Don't hash on the symbol's address to avoid bootstrap differences.
2288 Different hash values may cause expressions to be recorded in
2289 different orders and thus different registers to be used in the
2290 final assembler. This also avoids differences in the dump files
2291 between various stages. */
2292 unsigned int h = 0;
2293 const unsigned char *p = (const unsigned char *) XSTR (x, 0);
2295 while (*p)
2296 h += (h << 7) + *p++; /* ??? revisit */
2298 hash += ((unsigned int) SYMBOL_REF << 7) + h;
2299 return hash;
2302 case MEM:
2303 /* We don't record if marked volatile or if BLKmode since we don't
2304 know the size of the move. */
2305 if (MEM_VOLATILE_P (x) || GET_MODE (x) == BLKmode)
2307 *do_not_record_p = 1;
2308 return 0;
2310 if (hash_arg_in_memory_p && !MEM_READONLY_P (x))
2311 *hash_arg_in_memory_p = 1;
2313 /* Now that we have already found this special case,
2314 might as well speed it up as much as possible. */
2315 hash += (unsigned) MEM;
2316 x = XEXP (x, 0);
2317 goto repeat;
2319 case USE:
2320 /* A USE that mentions non-volatile memory needs special
2321 handling since the MEM may be BLKmode which normally
2322 prevents an entry from being made. Pure calls are
2323 marked by a USE which mentions BLKmode memory.
2324 See calls.c:emit_call_1. */
2325 if (MEM_P (XEXP (x, 0))
2326 && ! MEM_VOLATILE_P (XEXP (x, 0)))
2328 hash += (unsigned) USE;
2329 x = XEXP (x, 0);
2331 if (hash_arg_in_memory_p && !MEM_READONLY_P (x))
2332 *hash_arg_in_memory_p = 1;
2334 /* Now that we have already found this special case,
2335 might as well speed it up as much as possible. */
2336 hash += (unsigned) MEM;
2337 x = XEXP (x, 0);
2338 goto repeat;
2340 break;
2342 case PRE_DEC:
2343 case PRE_INC:
2344 case POST_DEC:
2345 case POST_INC:
2346 case PRE_MODIFY:
2347 case POST_MODIFY:
2348 case PC:
2349 case CC0:
2350 case CALL:
2351 case UNSPEC_VOLATILE:
2352 *do_not_record_p = 1;
2353 return 0;
2355 case ASM_OPERANDS:
2356 if (MEM_VOLATILE_P (x))
2358 *do_not_record_p = 1;
2359 return 0;
2361 else
2363 /* We don't want to take the filename and line into account. */
2364 hash += (unsigned) code + (unsigned) GET_MODE (x)
2365 + hash_rtx_string (ASM_OPERANDS_TEMPLATE (x))
2366 + hash_rtx_string (ASM_OPERANDS_OUTPUT_CONSTRAINT (x))
2367 + (unsigned) ASM_OPERANDS_OUTPUT_IDX (x);
2369 if (ASM_OPERANDS_INPUT_LENGTH (x))
2371 for (i = 1; i < ASM_OPERANDS_INPUT_LENGTH (x); i++)
2373 hash += (hash_rtx (ASM_OPERANDS_INPUT (x, i),
2374 GET_MODE (ASM_OPERANDS_INPUT (x, i)),
2375 do_not_record_p, hash_arg_in_memory_p,
2376 have_reg_qty)
2377 + hash_rtx_string
2378 (ASM_OPERANDS_INPUT_CONSTRAINT (x, i)));
2381 hash += hash_rtx_string (ASM_OPERANDS_INPUT_CONSTRAINT (x, 0));
2382 x = ASM_OPERANDS_INPUT (x, 0);
2383 mode = GET_MODE (x);
2384 goto repeat;
2387 return hash;
2389 break;
2391 default:
2392 break;
2395 i = GET_RTX_LENGTH (code) - 1;
2396 hash += (unsigned) code + (unsigned) GET_MODE (x);
2397 fmt = GET_RTX_FORMAT (code);
2398 for (; i >= 0; i--)
2400 switch (fmt[i])
2402 case 'e':
2403 /* If we are about to do the last recursive call
2404 needed at this level, change it into iteration.
2405 This function is called enough to be worth it. */
2406 if (i == 0)
2408 x = XEXP (x, i);
2409 goto repeat;
2412 hash += hash_rtx (XEXP (x, i), 0, do_not_record_p,
2413 hash_arg_in_memory_p, have_reg_qty);
2414 break;
2416 case 'E':
2417 for (j = 0; j < XVECLEN (x, i); j++)
2418 hash += hash_rtx (XVECEXP (x, i, j), 0, do_not_record_p,
2419 hash_arg_in_memory_p, have_reg_qty);
2420 break;
2422 case 's':
2423 hash += hash_rtx_string (XSTR (x, i));
2424 break;
2426 case 'i':
2427 hash += (unsigned int) XINT (x, i);
2428 break;
2430 case '0': case 't':
2431 /* Unused. */
2432 break;
2434 default:
2435 gcc_unreachable ();
2439 return hash;
2442 /* Hash an rtx X for cse via hash_rtx.
2443 Stores 1 in do_not_record if any subexpression is volatile.
2444 Stores 1 in hash_arg_in_memory if X contains a mem rtx which
2445 does not have the RTX_UNCHANGING_P bit set. */
2447 static inline unsigned
2448 canon_hash (rtx x, enum machine_mode mode)
2450 return hash_rtx (x, mode, &do_not_record, &hash_arg_in_memory, true);
2453 /* Like canon_hash but with no side effects, i.e. do_not_record
2454 and hash_arg_in_memory are not changed. */
2456 static inline unsigned
2457 safe_hash (rtx x, enum machine_mode mode)
2459 int dummy_do_not_record;
2460 return hash_rtx (x, mode, &dummy_do_not_record, NULL, true);
2463 /* Return 1 iff X and Y would canonicalize into the same thing,
2464 without actually constructing the canonicalization of either one.
2465 If VALIDATE is nonzero,
2466 we assume X is an expression being processed from the rtl
2467 and Y was found in the hash table. We check register refs
2468 in Y for being marked as valid.
2470 If FOR_GCSE is true, we compare X and Y for equivalence for GCSE. */
2473 exp_equiv_p (rtx x, rtx y, int validate, bool for_gcse)
2475 int i, j;
2476 enum rtx_code code;
2477 const char *fmt;
2479 /* Note: it is incorrect to assume an expression is equivalent to itself
2480 if VALIDATE is nonzero. */
2481 if (x == y && !validate)
2482 return 1;
2484 if (x == 0 || y == 0)
2485 return x == y;
2487 code = GET_CODE (x);
2488 if (code != GET_CODE (y))
2489 return 0;
2491 /* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent. */
2492 if (GET_MODE (x) != GET_MODE (y))
2493 return 0;
2495 switch (code)
2497 case PC:
2498 case CC0:
2499 case CONST_INT:
2500 case CONST_DOUBLE:
2501 return x == y;
2503 case LABEL_REF:
2504 return XEXP (x, 0) == XEXP (y, 0);
2506 case SYMBOL_REF:
2507 return XSTR (x, 0) == XSTR (y, 0);
2509 case REG:
2510 if (for_gcse)
2511 return REGNO (x) == REGNO (y);
2512 else
2514 unsigned int regno = REGNO (y);
2515 unsigned int i;
2516 unsigned int endregno
2517 = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
2518 : hard_regno_nregs[regno][GET_MODE (y)]);
2520 /* If the quantities are not the same, the expressions are not
2521 equivalent. If there are and we are not to validate, they
2522 are equivalent. Otherwise, ensure all regs are up-to-date. */
2524 if (REG_QTY (REGNO (x)) != REG_QTY (regno))
2525 return 0;
2527 if (! validate)
2528 return 1;
2530 for (i = regno; i < endregno; i++)
2531 if (REG_IN_TABLE (i) != REG_TICK (i))
2532 return 0;
2534 return 1;
2537 case MEM:
2538 if (for_gcse)
2540 /* A volatile mem should not be considered equivalent to any
2541 other. */
2542 if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
2543 return 0;
2545 /* Can't merge two expressions in different alias sets, since we
2546 can decide that the expression is transparent in a block when
2547 it isn't, due to it being set with the different alias set.
2549 Also, can't merge two expressions with different MEM_ATTRS.
2550 They could e.g. be two different entities allocated into the
2551 same space on the stack (see e.g. PR25130). In that case, the
2552 MEM addresses can be the same, even though the two MEMs are
2553 absolutely not equivalent.
2555 But because really all MEM attributes should be the same for
2556 equivalent MEMs, we just use the invariant that MEMs that have
2557 the same attributes share the same mem_attrs data structure. */
2558 if (MEM_ATTRS (x) != MEM_ATTRS (y))
2559 return 0;
2561 break;
2563 /* For commutative operations, check both orders. */
2564 case PLUS:
2565 case MULT:
2566 case AND:
2567 case IOR:
2568 case XOR:
2569 case NE:
2570 case EQ:
2571 return ((exp_equiv_p (XEXP (x, 0), XEXP (y, 0),
2572 validate, for_gcse)
2573 && exp_equiv_p (XEXP (x, 1), XEXP (y, 1),
2574 validate, for_gcse))
2575 || (exp_equiv_p (XEXP (x, 0), XEXP (y, 1),
2576 validate, for_gcse)
2577 && exp_equiv_p (XEXP (x, 1), XEXP (y, 0),
2578 validate, for_gcse)));
2580 case ASM_OPERANDS:
2581 /* We don't use the generic code below because we want to
2582 disregard filename and line numbers. */
2584 /* A volatile asm isn't equivalent to any other. */
2585 if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
2586 return 0;
2588 if (GET_MODE (x) != GET_MODE (y)
2589 || strcmp (ASM_OPERANDS_TEMPLATE (x), ASM_OPERANDS_TEMPLATE (y))
2590 || strcmp (ASM_OPERANDS_OUTPUT_CONSTRAINT (x),
2591 ASM_OPERANDS_OUTPUT_CONSTRAINT (y))
2592 || ASM_OPERANDS_OUTPUT_IDX (x) != ASM_OPERANDS_OUTPUT_IDX (y)
2593 || ASM_OPERANDS_INPUT_LENGTH (x) != ASM_OPERANDS_INPUT_LENGTH (y))
2594 return 0;
2596 if (ASM_OPERANDS_INPUT_LENGTH (x))
2598 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
2599 if (! exp_equiv_p (ASM_OPERANDS_INPUT (x, i),
2600 ASM_OPERANDS_INPUT (y, i),
2601 validate, for_gcse)
2602 || strcmp (ASM_OPERANDS_INPUT_CONSTRAINT (x, i),
2603 ASM_OPERANDS_INPUT_CONSTRAINT (y, i)))
2604 return 0;
2607 return 1;
2609 default:
2610 break;
2613 /* Compare the elements. If any pair of corresponding elements
2614 fail to match, return 0 for the whole thing. */
2616 fmt = GET_RTX_FORMAT (code);
2617 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2619 switch (fmt[i])
2621 case 'e':
2622 if (! exp_equiv_p (XEXP (x, i), XEXP (y, i),
2623 validate, for_gcse))
2624 return 0;
2625 break;
2627 case 'E':
2628 if (XVECLEN (x, i) != XVECLEN (y, i))
2629 return 0;
2630 for (j = 0; j < XVECLEN (x, i); j++)
2631 if (! exp_equiv_p (XVECEXP (x, i, j), XVECEXP (y, i, j),
2632 validate, for_gcse))
2633 return 0;
2634 break;
2636 case 's':
2637 if (strcmp (XSTR (x, i), XSTR (y, i)))
2638 return 0;
2639 break;
2641 case 'i':
2642 if (XINT (x, i) != XINT (y, i))
2643 return 0;
2644 break;
2646 case 'w':
2647 if (XWINT (x, i) != XWINT (y, i))
2648 return 0;
2649 break;
2651 case '0':
2652 case 't':
2653 break;
2655 default:
2656 gcc_unreachable ();
2660 return 1;
2663 /* Return 1 if X has a value that can vary even between two
2664 executions of the program. 0 means X can be compared reliably
2665 against certain constants or near-constants. */
2667 static int
2668 cse_rtx_varies_p (rtx x, int from_alias)
2670 /* We need not check for X and the equivalence class being of the same
2671 mode because if X is equivalent to a constant in some mode, it
2672 doesn't vary in any mode. */
2674 if (REG_P (x)
2675 && REGNO_QTY_VALID_P (REGNO (x)))
2677 int x_q = REG_QTY (REGNO (x));
2678 struct qty_table_elem *x_ent = &qty_table[x_q];
2680 if (GET_MODE (x) == x_ent->mode
2681 && x_ent->const_rtx != NULL_RTX)
2682 return 0;
2685 if (GET_CODE (x) == PLUS
2686 && GET_CODE (XEXP (x, 1)) == CONST_INT
2687 && REG_P (XEXP (x, 0))
2688 && REGNO_QTY_VALID_P (REGNO (XEXP (x, 0))))
2690 int x0_q = REG_QTY (REGNO (XEXP (x, 0)));
2691 struct qty_table_elem *x0_ent = &qty_table[x0_q];
2693 if ((GET_MODE (XEXP (x, 0)) == x0_ent->mode)
2694 && x0_ent->const_rtx != NULL_RTX)
2695 return 0;
2698 /* This can happen as the result of virtual register instantiation, if
2699 the initial constant is too large to be a valid address. This gives
2700 us a three instruction sequence, load large offset into a register,
2701 load fp minus a constant into a register, then a MEM which is the
2702 sum of the two `constant' registers. */
2703 if (GET_CODE (x) == PLUS
2704 && REG_P (XEXP (x, 0))
2705 && REG_P (XEXP (x, 1))
2706 && REGNO_QTY_VALID_P (REGNO (XEXP (x, 0)))
2707 && REGNO_QTY_VALID_P (REGNO (XEXP (x, 1))))
2709 int x0_q = REG_QTY (REGNO (XEXP (x, 0)));
2710 int x1_q = REG_QTY (REGNO (XEXP (x, 1)));
2711 struct qty_table_elem *x0_ent = &qty_table[x0_q];
2712 struct qty_table_elem *x1_ent = &qty_table[x1_q];
2714 if ((GET_MODE (XEXP (x, 0)) == x0_ent->mode)
2715 && x0_ent->const_rtx != NULL_RTX
2716 && (GET_MODE (XEXP (x, 1)) == x1_ent->mode)
2717 && x1_ent->const_rtx != NULL_RTX)
2718 return 0;
2721 return rtx_varies_p (x, from_alias);
2724 /* Subroutine of canon_reg. Pass *XLOC through canon_reg, and validate
2725 the result if necessary. INSN is as for canon_reg. */
2727 static void
2728 validate_canon_reg (rtx *xloc, rtx insn)
2730 rtx new = canon_reg (*xloc, insn);
2731 int insn_code;
2733 /* If replacing pseudo with hard reg or vice versa, ensure the
2734 insn remains valid. Likewise if the insn has MATCH_DUPs. */
2735 if (insn != 0 && new != 0
2736 && REG_P (new) && REG_P (*xloc)
2737 && (((REGNO (new) < FIRST_PSEUDO_REGISTER)
2738 != (REGNO (*xloc) < FIRST_PSEUDO_REGISTER))
2739 || GET_MODE (new) != GET_MODE (*xloc)
2740 || (insn_code = recog_memoized (insn)) < 0
2741 || insn_data[insn_code].n_dups > 0))
2742 validate_change (insn, xloc, new, 1);
2743 else
2744 *xloc = new;
2747 /* Canonicalize an expression:
2748 replace each register reference inside it
2749 with the "oldest" equivalent register.
2751 If INSN is nonzero and we are replacing a pseudo with a hard register
2752 or vice versa, validate_change is used to ensure that INSN remains valid
2753 after we make our substitution. The calls are made with IN_GROUP nonzero
2754 so apply_change_group must be called upon the outermost return from this
2755 function (unless INSN is zero). The result of apply_change_group can
2756 generally be discarded since the changes we are making are optional. */
2758 static rtx
2759 canon_reg (rtx x, rtx insn)
2761 int i;
2762 enum rtx_code code;
2763 const char *fmt;
2765 if (x == 0)
2766 return x;
2768 code = GET_CODE (x);
2769 switch (code)
2771 case PC:
2772 case CC0:
2773 case CONST:
2774 case CONST_INT:
2775 case CONST_DOUBLE:
2776 case CONST_VECTOR:
2777 case SYMBOL_REF:
2778 case LABEL_REF:
2779 case ADDR_VEC:
2780 case ADDR_DIFF_VEC:
2781 return x;
2783 case REG:
2785 int first;
2786 int q;
2787 struct qty_table_elem *ent;
2789 /* Never replace a hard reg, because hard regs can appear
2790 in more than one machine mode, and we must preserve the mode
2791 of each occurrence. Also, some hard regs appear in
2792 MEMs that are shared and mustn't be altered. Don't try to
2793 replace any reg that maps to a reg of class NO_REGS. */
2794 if (REGNO (x) < FIRST_PSEUDO_REGISTER
2795 || ! REGNO_QTY_VALID_P (REGNO (x)))
2796 return x;
2798 q = REG_QTY (REGNO (x));
2799 ent = &qty_table[q];
2800 first = ent->first_reg;
2801 return (first >= FIRST_PSEUDO_REGISTER ? regno_reg_rtx[first]
2802 : REGNO_REG_CLASS (first) == NO_REGS ? x
2803 : gen_rtx_REG (ent->mode, first));
2806 default:
2807 break;
2810 fmt = GET_RTX_FORMAT (code);
2811 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
2813 int j;
2815 if (fmt[i] == 'e')
2816 validate_canon_reg (&XEXP (x, i), insn);
2817 else if (fmt[i] == 'E')
2818 for (j = 0; j < XVECLEN (x, i); j++)
2819 validate_canon_reg (&XVECEXP (x, i, j), insn);
2822 return x;
2825 /* LOC is a location within INSN that is an operand address (the contents of
2826 a MEM). Find the best equivalent address to use that is valid for this
2827 insn.
2829 On most CISC machines, complicated address modes are costly, and rtx_cost
2830 is a good approximation for that cost. However, most RISC machines have
2831 only a few (usually only one) memory reference formats. If an address is
2832 valid at all, it is often just as cheap as any other address. Hence, for
2833 RISC machines, we use `address_cost' to compare the costs of various
2834 addresses. For two addresses of equal cost, choose the one with the
2835 highest `rtx_cost' value as that has the potential of eliminating the
2836 most insns. For equal costs, we choose the first in the equivalence
2837 class. Note that we ignore the fact that pseudo registers are cheaper than
2838 hard registers here because we would also prefer the pseudo registers. */
2840 static void
2841 find_best_addr (rtx insn, rtx *loc, enum machine_mode mode)
2843 struct table_elt *elt;
2844 rtx addr = *loc;
2845 struct table_elt *p;
2846 int found_better = 1;
2847 int save_do_not_record = do_not_record;
2848 int save_hash_arg_in_memory = hash_arg_in_memory;
2849 int addr_volatile;
2850 int regno;
2851 unsigned hash;
2853 /* Do not try to replace constant addresses or addresses of local and
2854 argument slots. These MEM expressions are made only once and inserted
2855 in many instructions, as well as being used to control symbol table
2856 output. It is not safe to clobber them.
2858 There are some uncommon cases where the address is already in a register
2859 for some reason, but we cannot take advantage of that because we have
2860 no easy way to unshare the MEM. In addition, looking up all stack
2861 addresses is costly. */
2862 if ((GET_CODE (addr) == PLUS
2863 && REG_P (XEXP (addr, 0))
2864 && GET_CODE (XEXP (addr, 1)) == CONST_INT
2865 && (regno = REGNO (XEXP (addr, 0)),
2866 regno == FRAME_POINTER_REGNUM || regno == HARD_FRAME_POINTER_REGNUM
2867 || regno == ARG_POINTER_REGNUM))
2868 || (REG_P (addr)
2869 && (regno = REGNO (addr), regno == FRAME_POINTER_REGNUM
2870 || regno == HARD_FRAME_POINTER_REGNUM
2871 || regno == ARG_POINTER_REGNUM))
2872 || CONSTANT_ADDRESS_P (addr))
2873 return;
2875 /* If this address is not simply a register, try to fold it. This will
2876 sometimes simplify the expression. Many simplifications
2877 will not be valid, but some, usually applying the associative rule, will
2878 be valid and produce better code. */
2879 if (!REG_P (addr))
2881 rtx folded = canon_for_address (fold_rtx (addr, NULL_RTX));
2883 if (folded != addr)
2885 int addr_folded_cost = address_cost (folded, mode);
2886 int addr_cost = address_cost (addr, mode);
2888 if ((addr_folded_cost < addr_cost
2889 || (addr_folded_cost == addr_cost
2890 /* ??? The rtx_cost comparison is left over from an older
2891 version of this code. It is probably no longer helpful.*/
2892 && (rtx_cost (folded, MEM) > rtx_cost (addr, MEM)
2893 || approx_reg_cost (folded) < approx_reg_cost (addr))))
2894 && validate_change (insn, loc, folded, 0))
2895 addr = folded;
2899 /* If this address is not in the hash table, we can't look for equivalences
2900 of the whole address. Also, ignore if volatile. */
2902 do_not_record = 0;
2903 hash = HASH (addr, Pmode);
2904 addr_volatile = do_not_record;
2905 do_not_record = save_do_not_record;
2906 hash_arg_in_memory = save_hash_arg_in_memory;
2908 if (addr_volatile)
2909 return;
2911 elt = lookup (addr, hash, Pmode);
2913 if (elt)
2915 /* We need to find the best (under the criteria documented above) entry
2916 in the class that is valid. We use the `flag' field to indicate
2917 choices that were invalid and iterate until we can't find a better
2918 one that hasn't already been tried. */
2920 for (p = elt->first_same_value; p; p = p->next_same_value)
2921 p->flag = 0;
2923 while (found_better)
2925 int best_addr_cost = address_cost (*loc, mode);
2926 int best_rtx_cost = (elt->cost + 1) >> 1;
2927 int exp_cost;
2928 struct table_elt *best_elt = elt;
2930 found_better = 0;
2931 for (p = elt->first_same_value; p; p = p->next_same_value)
2932 if (! p->flag)
2934 if ((REG_P (p->exp)
2935 || exp_equiv_p (p->exp, p->exp, 1, false))
2936 && ((exp_cost = address_cost (p->exp, mode)) < best_addr_cost
2937 || (exp_cost == best_addr_cost
2938 && ((p->cost + 1) >> 1) > best_rtx_cost)))
2940 found_better = 1;
2941 best_addr_cost = exp_cost;
2942 best_rtx_cost = (p->cost + 1) >> 1;
2943 best_elt = p;
2947 if (found_better)
2949 if (validate_change (insn, loc,
2950 canon_reg (copy_rtx (best_elt->exp),
2951 NULL_RTX), 0))
2952 return;
2953 else
2954 best_elt->flag = 1;
2959 /* If the address is a binary operation with the first operand a register
2960 and the second a constant, do the same as above, but looking for
2961 equivalences of the register. Then try to simplify before checking for
2962 the best address to use. This catches a few cases: First is when we
2963 have REG+const and the register is another REG+const. We can often merge
2964 the constants and eliminate one insn and one register. It may also be
2965 that a machine has a cheap REG+REG+const. Finally, this improves the
2966 code on the Alpha for unaligned byte stores. */
2968 if (flag_expensive_optimizations
2969 && ARITHMETIC_P (*loc)
2970 && REG_P (XEXP (*loc, 0)))
2972 rtx op1 = XEXP (*loc, 1);
2974 do_not_record = 0;
2975 hash = HASH (XEXP (*loc, 0), Pmode);
2976 do_not_record = save_do_not_record;
2977 hash_arg_in_memory = save_hash_arg_in_memory;
2979 elt = lookup (XEXP (*loc, 0), hash, Pmode);
2980 if (elt == 0)
2981 return;
2983 /* We need to find the best (under the criteria documented above) entry
2984 in the class that is valid. We use the `flag' field to indicate
2985 choices that were invalid and iterate until we can't find a better
2986 one that hasn't already been tried. */
2988 for (p = elt->first_same_value; p; p = p->next_same_value)
2989 p->flag = 0;
2991 while (found_better)
2993 int best_addr_cost = address_cost (*loc, mode);
2994 int best_rtx_cost = (COST (*loc) + 1) >> 1;
2995 struct table_elt *best_elt = elt;
2996 rtx best_rtx = *loc;
2997 int count;
2999 /* This is at worst case an O(n^2) algorithm, so limit our search
3000 to the first 32 elements on the list. This avoids trouble
3001 compiling code with very long basic blocks that can easily
3002 call simplify_gen_binary so many times that we run out of
3003 memory. */
3005 found_better = 0;
3006 for (p = elt->first_same_value, count = 0;
3007 p && count < 32;
3008 p = p->next_same_value, count++)
3009 if (! p->flag
3010 && (REG_P (p->exp)
3011 || (GET_CODE (p->exp) != EXPR_LIST
3012 && exp_equiv_p (p->exp, p->exp, 1, false))))
3015 rtx new = simplify_gen_binary (GET_CODE (*loc), Pmode,
3016 p->exp, op1);
3017 int new_cost;
3019 /* Get the canonical version of the address so we can accept
3020 more. */
3021 new = canon_for_address (new);
3023 new_cost = address_cost (new, mode);
3025 if (new_cost < best_addr_cost
3026 || (new_cost == best_addr_cost
3027 && (COST (new) + 1) >> 1 > best_rtx_cost))
3029 found_better = 1;
3030 best_addr_cost = new_cost;
3031 best_rtx_cost = (COST (new) + 1) >> 1;
3032 best_elt = p;
3033 best_rtx = new;
3037 if (found_better)
3039 if (validate_change (insn, loc,
3040 canon_reg (copy_rtx (best_rtx),
3041 NULL_RTX), 0))
3042 return;
3043 else
3044 best_elt->flag = 1;
3050 /* Given an operation (CODE, *PARG1, *PARG2), where code is a comparison
3051 operation (EQ, NE, GT, etc.), follow it back through the hash table and
3052 what values are being compared.
3054 *PARG1 and *PARG2 are updated to contain the rtx representing the values
3055 actually being compared. For example, if *PARG1 was (cc0) and *PARG2
3056 was (const_int 0), *PARG1 and *PARG2 will be set to the objects that were
3057 compared to produce cc0.
3059 The return value is the comparison operator and is either the code of
3060 A or the code corresponding to the inverse of the comparison. */
3062 static enum rtx_code
3063 find_comparison_args (enum rtx_code code, rtx *parg1, rtx *parg2,
3064 enum machine_mode *pmode1, enum machine_mode *pmode2)
3066 rtx arg1, arg2;
3068 arg1 = *parg1, arg2 = *parg2;
3070 /* If ARG2 is const0_rtx, see what ARG1 is equivalent to. */
3072 while (arg2 == CONST0_RTX (GET_MODE (arg1)))
3074 /* Set nonzero when we find something of interest. */
3075 rtx x = 0;
3076 int reverse_code = 0;
3077 struct table_elt *p = 0;
3079 /* If arg1 is a COMPARE, extract the comparison arguments from it.
3080 On machines with CC0, this is the only case that can occur, since
3081 fold_rtx will return the COMPARE or item being compared with zero
3082 when given CC0. */
3084 if (GET_CODE (arg1) == COMPARE && arg2 == const0_rtx)
3085 x = arg1;
3087 /* If ARG1 is a comparison operator and CODE is testing for
3088 STORE_FLAG_VALUE, get the inner arguments. */
3090 else if (COMPARISON_P (arg1))
3092 #ifdef FLOAT_STORE_FLAG_VALUE
3093 REAL_VALUE_TYPE fsfv;
3094 #endif
3096 if (code == NE
3097 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
3098 && code == LT && STORE_FLAG_VALUE == -1)
3099 #ifdef FLOAT_STORE_FLAG_VALUE
3100 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
3101 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3102 REAL_VALUE_NEGATIVE (fsfv)))
3103 #endif
3105 x = arg1;
3106 else if (code == EQ
3107 || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
3108 && code == GE && STORE_FLAG_VALUE == -1)
3109 #ifdef FLOAT_STORE_FLAG_VALUE
3110 || (SCALAR_FLOAT_MODE_P (GET_MODE (arg1))
3111 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3112 REAL_VALUE_NEGATIVE (fsfv)))
3113 #endif
3115 x = arg1, reverse_code = 1;
3118 /* ??? We could also check for
3120 (ne (and (eq (...) (const_int 1))) (const_int 0))
3122 and related forms, but let's wait until we see them occurring. */
3124 if (x == 0)
3125 /* Look up ARG1 in the hash table and see if it has an equivalence
3126 that lets us see what is being compared. */
3127 p = lookup (arg1, SAFE_HASH (arg1, GET_MODE (arg1)), GET_MODE (arg1));
3128 if (p)
3130 p = p->first_same_value;
3132 /* If what we compare is already known to be constant, that is as
3133 good as it gets.
3134 We need to break the loop in this case, because otherwise we
3135 can have an infinite loop when looking at a reg that is known
3136 to be a constant which is the same as a comparison of a reg
3137 against zero which appears later in the insn stream, which in
3138 turn is constant and the same as the comparison of the first reg
3139 against zero... */
3140 if (p->is_const)
3141 break;
3144 for (; p; p = p->next_same_value)
3146 enum machine_mode inner_mode = GET_MODE (p->exp);
3147 #ifdef FLOAT_STORE_FLAG_VALUE
3148 REAL_VALUE_TYPE fsfv;
3149 #endif
3151 /* If the entry isn't valid, skip it. */
3152 if (! exp_equiv_p (p->exp, p->exp, 1, false))
3153 continue;
3155 if (GET_CODE (p->exp) == COMPARE
3156 /* Another possibility is that this machine has a compare insn
3157 that includes the comparison code. In that case, ARG1 would
3158 be equivalent to a comparison operation that would set ARG1 to
3159 either STORE_FLAG_VALUE or zero. If this is an NE operation,
3160 ORIG_CODE is the actual comparison being done; if it is an EQ,
3161 we must reverse ORIG_CODE. On machine with a negative value
3162 for STORE_FLAG_VALUE, also look at LT and GE operations. */
3163 || ((code == NE
3164 || (code == LT
3165 && GET_MODE_CLASS (inner_mode) == MODE_INT
3166 && (GET_MODE_BITSIZE (inner_mode)
3167 <= HOST_BITS_PER_WIDE_INT)
3168 && (STORE_FLAG_VALUE
3169 & ((HOST_WIDE_INT) 1
3170 << (GET_MODE_BITSIZE (inner_mode) - 1))))
3171 #ifdef FLOAT_STORE_FLAG_VALUE
3172 || (code == LT
3173 && SCALAR_FLOAT_MODE_P (inner_mode)
3174 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3175 REAL_VALUE_NEGATIVE (fsfv)))
3176 #endif
3178 && COMPARISON_P (p->exp)))
3180 x = p->exp;
3181 break;
3183 else if ((code == EQ
3184 || (code == GE
3185 && GET_MODE_CLASS (inner_mode) == MODE_INT
3186 && (GET_MODE_BITSIZE (inner_mode)
3187 <= HOST_BITS_PER_WIDE_INT)
3188 && (STORE_FLAG_VALUE
3189 & ((HOST_WIDE_INT) 1
3190 << (GET_MODE_BITSIZE (inner_mode) - 1))))
3191 #ifdef FLOAT_STORE_FLAG_VALUE
3192 || (code == GE
3193 && SCALAR_FLOAT_MODE_P (inner_mode)
3194 && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
3195 REAL_VALUE_NEGATIVE (fsfv)))
3196 #endif
3198 && COMPARISON_P (p->exp))
3200 reverse_code = 1;
3201 x = p->exp;
3202 break;
3205 /* If this non-trapping address, e.g. fp + constant, the
3206 equivalent is a better operand since it may let us predict
3207 the value of the comparison. */
3208 else if (!rtx_addr_can_trap_p (p->exp))
3210 arg1 = p->exp;
3211 continue;
3215 /* If we didn't find a useful equivalence for ARG1, we are done.
3216 Otherwise, set up for the next iteration. */
3217 if (x == 0)
3218 break;
3220 /* If we need to reverse the comparison, make sure that that is
3221 possible -- we can't necessarily infer the value of GE from LT
3222 with floating-point operands. */
3223 if (reverse_code)
3225 enum rtx_code reversed = reversed_comparison_code (x, NULL_RTX);
3226 if (reversed == UNKNOWN)
3227 break;
3228 else
3229 code = reversed;
3231 else if (COMPARISON_P (x))
3232 code = GET_CODE (x);
3233 arg1 = XEXP (x, 0), arg2 = XEXP (x, 1);
3236 /* Return our results. Return the modes from before fold_rtx
3237 because fold_rtx might produce const_int, and then it's too late. */
3238 *pmode1 = GET_MODE (arg1), *pmode2 = GET_MODE (arg2);
3239 *parg1 = fold_rtx (arg1, 0), *parg2 = fold_rtx (arg2, 0);
3241 return code;
3244 /* Fold SUBREG. */
3246 static rtx
3247 fold_rtx_subreg (rtx x, rtx insn)
3249 enum machine_mode mode = GET_MODE (x);
3250 rtx folded_arg0;
3251 rtx const_arg0;
3252 rtx new;
3254 /* See if we previously assigned a constant value to this SUBREG. */
3255 if ((new = lookup_as_function (x, CONST_INT)) != 0
3256 || (new = lookup_as_function (x, CONST_DOUBLE)) != 0)
3257 return new;
3259 /* If this is a paradoxical SUBREG, we have no idea what value the
3260 extra bits would have. However, if the operand is equivalent to
3261 a SUBREG whose operand is the same as our mode, and all the modes
3262 are within a word, we can just use the inner operand because
3263 these SUBREGs just say how to treat the register.
3265 Similarly if we find an integer constant. */
3267 if (GET_MODE_SIZE (mode) > GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
3269 enum machine_mode imode = GET_MODE (SUBREG_REG (x));
3270 struct table_elt *elt;
3272 if (GET_MODE_SIZE (mode) <= UNITS_PER_WORD
3273 && GET_MODE_SIZE (imode) <= UNITS_PER_WORD
3274 && (elt = lookup (SUBREG_REG (x), HASH (SUBREG_REG (x), imode),
3275 imode)) != 0)
3276 for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
3278 if (CONSTANT_P (elt->exp)
3279 && GET_MODE (elt->exp) == VOIDmode)
3280 return elt->exp;
3282 if (GET_CODE (elt->exp) == SUBREG
3283 && GET_MODE (SUBREG_REG (elt->exp)) == mode
3284 && exp_equiv_p (elt->exp, elt->exp, 1, false))
3285 return copy_rtx (SUBREG_REG (elt->exp));
3288 return x;
3291 /* Fold SUBREG_REG. If it changed, see if we can simplify the
3292 SUBREG. We might be able to if the SUBREG is extracting a single
3293 word in an integral mode or extracting the low part. */
3295 folded_arg0 = fold_rtx (SUBREG_REG (x), insn);
3296 const_arg0 = equiv_constant (folded_arg0);
3297 if (const_arg0)
3298 folded_arg0 = const_arg0;
3300 if (folded_arg0 != SUBREG_REG (x))
3302 new = simplify_subreg (mode, folded_arg0,
3303 GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
3304 if (new)
3305 return new;
3308 if (REG_P (folded_arg0)
3309 && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (folded_arg0)))
3311 struct table_elt *elt;
3313 elt = lookup (folded_arg0,
3314 HASH (folded_arg0, GET_MODE (folded_arg0)),
3315 GET_MODE (folded_arg0));
3317 if (elt)
3318 elt = elt->first_same_value;
3320 if (subreg_lowpart_p (x))
3321 /* If this is a narrowing SUBREG and our operand is a REG, see
3322 if we can find an equivalence for REG that is an arithmetic
3323 operation in a wider mode where both operands are
3324 paradoxical SUBREGs from objects of our result mode. In
3325 that case, we couldn-t report an equivalent value for that
3326 operation, since we don't know what the extra bits will be.
3327 But we can find an equivalence for this SUBREG by folding
3328 that operation in the narrow mode. This allows us to fold
3329 arithmetic in narrow modes when the machine only supports
3330 word-sized arithmetic.
3332 Also look for a case where we have a SUBREG whose operand
3333 is the same as our result. If both modes are smaller than
3334 a word, we are simply interpreting a register in different
3335 modes and we can use the inner value. */
3337 for (; elt; elt = elt->next_same_value)
3339 enum rtx_code eltcode = GET_CODE (elt->exp);
3341 /* Just check for unary and binary operations. */
3342 if (UNARY_P (elt->exp)
3343 && eltcode != SIGN_EXTEND
3344 && eltcode != ZERO_EXTEND
3345 && GET_CODE (XEXP (elt->exp, 0)) == SUBREG
3346 && GET_MODE (SUBREG_REG (XEXP (elt->exp, 0))) == mode
3347 && (GET_MODE_CLASS (mode)
3348 == GET_MODE_CLASS (GET_MODE (XEXP (elt->exp, 0)))))
3350 rtx op0 = SUBREG_REG (XEXP (elt->exp, 0));
3352 if (!REG_P (op0) && ! CONSTANT_P (op0))
3353 op0 = fold_rtx (op0, NULL_RTX);
3355 op0 = equiv_constant (op0);
3356 if (op0)
3357 new = simplify_unary_operation (GET_CODE (elt->exp), mode,
3358 op0, mode);
3360 else if (ARITHMETIC_P (elt->exp)
3361 && eltcode != DIV && eltcode != MOD
3362 && eltcode != UDIV && eltcode != UMOD
3363 && eltcode != ASHIFTRT && eltcode != LSHIFTRT
3364 && eltcode != ROTATE && eltcode != ROTATERT
3365 && ((GET_CODE (XEXP (elt->exp, 0)) == SUBREG
3366 && (GET_MODE (SUBREG_REG (XEXP (elt->exp, 0)))
3367 == mode))
3368 || CONSTANT_P (XEXP (elt->exp, 0)))
3369 && ((GET_CODE (XEXP (elt->exp, 1)) == SUBREG
3370 && (GET_MODE (SUBREG_REG (XEXP (elt->exp, 1)))
3371 == mode))
3372 || CONSTANT_P (XEXP (elt->exp, 1))))
3374 rtx op0 = gen_lowpart_common (mode, XEXP (elt->exp, 0));
3375 rtx op1 = gen_lowpart_common (mode, XEXP (elt->exp, 1));
3377 if (op0 && !REG_P (op0) && ! CONSTANT_P (op0))
3378 op0 = fold_rtx (op0, NULL_RTX);
3380 if (op0)
3381 op0 = equiv_constant (op0);
3383 if (op1 && !REG_P (op1) && ! CONSTANT_P (op1))
3384 op1 = fold_rtx (op1, NULL_RTX);
3386 if (op1)
3387 op1 = equiv_constant (op1);
3389 /* If we are looking for the low SImode part of
3390 (ashift:DI c (const_int 32)), it doesn't work to
3391 compute that in SImode, because a 32-bit shift in
3392 SImode is unpredictable. We know the value is
3393 0. */
3394 if (op0 && op1
3395 && GET_CODE (elt->exp) == ASHIFT
3396 && GET_CODE (op1) == CONST_INT
3397 && INTVAL (op1) >= GET_MODE_BITSIZE (mode))
3399 if (INTVAL (op1)
3400 < GET_MODE_BITSIZE (GET_MODE (elt->exp)))
3401 /* If the count fits in the inner mode's width,
3402 but exceeds the outer mode's width, the value
3403 will get truncated to 0 by the subreg. */
3404 new = CONST0_RTX (mode);
3405 else
3406 /* If the count exceeds even the inner mode's width,
3407 don't fold this expression. */
3408 new = 0;
3410 else if (op0 && op1)
3411 new = simplify_binary_operation (GET_CODE (elt->exp),
3412 mode, op0, op1);
3415 else if (GET_CODE (elt->exp) == SUBREG
3416 && GET_MODE (SUBREG_REG (elt->exp)) == mode
3417 && (GET_MODE_SIZE (GET_MODE (folded_arg0))
3418 <= UNITS_PER_WORD)
3419 && exp_equiv_p (elt->exp, elt->exp, 1, false))
3420 new = copy_rtx (SUBREG_REG (elt->exp));
3422 if (new)
3423 return new;
3425 else
3426 /* A SUBREG resulting from a zero extension may fold to zero
3427 if it extracts higher bits than the ZERO_EXTEND's source
3428 bits. FIXME: if combine tried to, er, combine these
3429 instructions, this transformation may be moved to
3430 simplify_subreg. */
3431 for (; elt; elt = elt->next_same_value)
3433 if (GET_CODE (elt->exp) == ZERO_EXTEND
3434 && subreg_lsb (x)
3435 >= GET_MODE_BITSIZE (GET_MODE (XEXP (elt->exp, 0))))
3436 return CONST0_RTX (mode);
3440 return x;
3443 /* Fold MEM. */
3445 static rtx
3446 fold_rtx_mem (rtx x, rtx insn)
3448 enum machine_mode mode = GET_MODE (x);
3449 rtx new;
3451 /* If we are not actually processing an insn, don't try to find the
3452 best address. Not only don't we care, but we could modify the
3453 MEM in an invalid way since we have no insn to validate
3454 against. */
3455 if (insn != 0)
3456 find_best_addr (insn, &XEXP (x, 0), mode);
3459 /* Even if we don't fold in the insn itself, we can safely do so
3460 here, in hopes of getting a constant. */
3461 rtx addr = fold_rtx (XEXP (x, 0), NULL_RTX);
3462 rtx base = 0;
3463 HOST_WIDE_INT offset = 0;
3465 if (REG_P (addr)
3466 && REGNO_QTY_VALID_P (REGNO (addr)))
3468 int addr_q = REG_QTY (REGNO (addr));
3469 struct qty_table_elem *addr_ent = &qty_table[addr_q];
3471 if (GET_MODE (addr) == addr_ent->mode
3472 && addr_ent->const_rtx != NULL_RTX)
3473 addr = addr_ent->const_rtx;
3476 /* Call target hook to avoid the effects of -fpic etc.... */
3477 addr = targetm.delegitimize_address (addr);
3479 /* If address is constant, split it into a base and integer
3480 offset. */
3481 if (GET_CODE (addr) == SYMBOL_REF || GET_CODE (addr) == LABEL_REF)
3482 base = addr;
3483 else if (GET_CODE (addr) == CONST && GET_CODE (XEXP (addr, 0)) == PLUS
3484 && GET_CODE (XEXP (XEXP (addr, 0), 1)) == CONST_INT)
3486 base = XEXP (XEXP (addr, 0), 0);
3487 offset = INTVAL (XEXP (XEXP (addr, 0), 1));
3489 else if (GET_CODE (addr) == LO_SUM
3490 && GET_CODE (XEXP (addr, 1)) == SYMBOL_REF)
3491 base = XEXP (addr, 1);
3493 /* If this is a constant pool reference, we can fold it into its
3494 constant to allow better value tracking. */
3495 if (base && GET_CODE (base) == SYMBOL_REF
3496 && CONSTANT_POOL_ADDRESS_P (base))
3498 rtx constant = get_pool_constant (base);
3499 enum machine_mode const_mode = get_pool_mode (base);
3500 rtx new;
3502 if (CONSTANT_P (constant) && GET_CODE (constant) != CONST_INT)
3504 constant_pool_entries_cost = COST (constant);
3505 constant_pool_entries_regcost = approx_reg_cost (constant);
3508 /* If we are loading the full constant, we have an
3509 equivalence. */
3510 if (offset == 0 && mode == const_mode)
3511 return constant;
3513 /* If this actually isn't a constant (weird!), we can't do
3514 anything. Otherwise, handle the two most common cases:
3515 extracting a word from a multi-word constant, and
3516 extracting the low-order bits. Other cases don't seem
3517 common enough to worry about. */
3518 if (! CONSTANT_P (constant))
3519 return x;
3521 if (GET_MODE_CLASS (mode) == MODE_INT
3522 && GET_MODE_SIZE (mode) == UNITS_PER_WORD
3523 && offset % UNITS_PER_WORD == 0
3524 && (new = operand_subword (constant,
3525 offset / UNITS_PER_WORD,
3526 0, const_mode)) != 0)
3527 return new;
3529 if (((BYTES_BIG_ENDIAN
3530 && offset == GET_MODE_SIZE (GET_MODE (constant)) - 1)
3531 || (! BYTES_BIG_ENDIAN && offset == 0))
3532 && (new = gen_lowpart (mode, constant)) != 0)
3533 return new;
3536 /* If this is a reference to a label at a known position in a jump
3537 table, we also know its value. */
3538 if (base && GET_CODE (base) == LABEL_REF)
3540 rtx label = XEXP (base, 0);
3541 rtx table_insn = NEXT_INSN (label);
3543 if (table_insn && JUMP_P (table_insn)
3544 && GET_CODE (PATTERN (table_insn)) == ADDR_VEC)
3546 rtx table = PATTERN (table_insn);
3548 if (offset >= 0
3549 && (offset / GET_MODE_SIZE (GET_MODE (table))
3550 < XVECLEN (table, 0)))
3552 rtx label = XVECEXP
3553 (table, 0, offset / GET_MODE_SIZE (GET_MODE (table)));
3554 rtx set;
3556 /* If we have an insn that loads the label from the
3557 jumptable into a reg, we don't want to set the reg
3558 to the label, because this may cause a reference to
3559 the label to remain after the label is removed in
3560 some very obscure cases (PR middle-end/18628). */
3561 if (!insn)
3562 return label;
3564 set = single_set (insn);
3566 if (! set || SET_SRC (set) != x)
3567 return x;
3569 /* If it's a jump, it's safe to reference the label. */
3570 if (SET_DEST (set) == pc_rtx)
3571 return label;
3573 return x;
3576 if (table_insn && JUMP_P (table_insn)
3577 && GET_CODE (PATTERN (table_insn)) == ADDR_DIFF_VEC)
3579 rtx table = PATTERN (table_insn);
3581 if (offset >= 0
3582 && (offset / GET_MODE_SIZE (GET_MODE (table))
3583 < XVECLEN (table, 1)))
3585 offset /= GET_MODE_SIZE (GET_MODE (table));
3586 new = gen_rtx_MINUS (Pmode, XVECEXP (table, 1, offset),
3587 XEXP (table, 0));
3589 if (GET_MODE (table) != Pmode)
3590 new = gen_rtx_TRUNCATE (GET_MODE (table), new);
3592 /* Indicate this is a constant. This isn't a valid
3593 form of CONST, but it will only be used to fold the
3594 next insns and then discarded, so it should be
3595 safe.
3597 Note this expression must be explicitly discarded,
3598 by cse_insn, else it may end up in a REG_EQUAL note
3599 and "escape" to cause problems elsewhere. */
3600 return gen_rtx_CONST (GET_MODE (new), new);
3605 return x;
3609 /* If X is a nontrivial arithmetic operation on an argument
3610 for which a constant value can be determined, return
3611 the result of operating on that value, as a constant.
3612 Otherwise, return X, possibly with one or more operands
3613 modified by recursive calls to this function.
3615 If X is a register whose contents are known, we do NOT
3616 return those contents here. equiv_constant is called to
3617 perform that task.
3619 INSN is the insn that we may be modifying. If it is 0, make a copy
3620 of X before modifying it. */
3622 static rtx
3623 fold_rtx (rtx x, rtx insn)
3625 enum rtx_code code;
3626 enum machine_mode mode;
3627 const char *fmt;
3628 int i;
3629 rtx new = 0;
3630 int copied = 0;
3631 int must_swap = 0;
3633 /* Folded equivalents of first two operands of X. */
3634 rtx folded_arg0;
3635 rtx folded_arg1;
3637 /* Constant equivalents of first three operands of X;
3638 0 when no such equivalent is known. */
3639 rtx const_arg0;
3640 rtx const_arg1;
3641 rtx const_arg2;
3643 /* The mode of the first operand of X. We need this for sign and zero
3644 extends. */
3645 enum machine_mode mode_arg0;
3647 if (x == 0)
3648 return x;
3650 mode = GET_MODE (x);
3651 code = GET_CODE (x);
3652 switch (code)
3654 case CONST:
3655 case CONST_INT:
3656 case CONST_DOUBLE:
3657 case CONST_VECTOR:
3658 case SYMBOL_REF:
3659 case LABEL_REF:
3660 case REG:
3661 case PC:
3662 /* No use simplifying an EXPR_LIST
3663 since they are used only for lists of args
3664 in a function call's REG_EQUAL note. */
3665 case EXPR_LIST:
3666 return x;
3668 #ifdef HAVE_cc0
3669 case CC0:
3670 return prev_insn_cc0;
3671 #endif
3673 case SUBREG:
3674 return fold_rtx_subreg (x, insn);
3676 case NOT:
3677 case NEG:
3678 /* If we have (NOT Y), see if Y is known to be (NOT Z).
3679 If so, (NOT Y) simplifies to Z. Similarly for NEG. */
3680 new = lookup_as_function (XEXP (x, 0), code);
3681 if (new)
3682 return fold_rtx (copy_rtx (XEXP (new, 0)), insn);
3683 break;
3685 case MEM:
3686 return fold_rtx_mem (x, insn);
3688 #ifdef NO_FUNCTION_CSE
3689 case CALL:
3690 if (CONSTANT_P (XEXP (XEXP (x, 0), 0)))
3691 return x;
3692 break;
3693 #endif
3695 case ASM_OPERANDS:
3696 if (insn)
3698 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
3699 validate_change (insn, &ASM_OPERANDS_INPUT (x, i),
3700 fold_rtx (ASM_OPERANDS_INPUT (x, i), insn), 0);
3702 break;
3704 default:
3705 break;
3708 const_arg0 = 0;
3709 const_arg1 = 0;
3710 const_arg2 = 0;
3711 mode_arg0 = VOIDmode;
3713 /* Try folding our operands.
3714 Then see which ones have constant values known. */
3716 fmt = GET_RTX_FORMAT (code);
3717 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
3718 if (fmt[i] == 'e')
3720 rtx arg = XEXP (x, i);
3721 rtx folded_arg = arg, const_arg = 0;
3722 enum machine_mode mode_arg = GET_MODE (arg);
3723 rtx cheap_arg, expensive_arg;
3724 rtx replacements[2];
3725 int j;
3726 int old_cost = COST_IN (XEXP (x, i), code);
3728 /* Most arguments are cheap, so handle them specially. */
3729 switch (GET_CODE (arg))
3731 case REG:
3732 /* This is the same as calling equiv_constant; it is duplicated
3733 here for speed. */
3734 if (REGNO_QTY_VALID_P (REGNO (arg)))
3736 int arg_q = REG_QTY (REGNO (arg));
3737 struct qty_table_elem *arg_ent = &qty_table[arg_q];
3739 if (arg_ent->const_rtx != NULL_RTX
3740 && !REG_P (arg_ent->const_rtx)
3741 && GET_CODE (arg_ent->const_rtx) != PLUS)
3742 const_arg
3743 = gen_lowpart (GET_MODE (arg),
3744 arg_ent->const_rtx);
3746 break;
3748 case CONST:
3749 case CONST_INT:
3750 case SYMBOL_REF:
3751 case LABEL_REF:
3752 case CONST_DOUBLE:
3753 case CONST_VECTOR:
3754 const_arg = arg;
3755 break;
3757 #ifdef HAVE_cc0
3758 case CC0:
3759 folded_arg = prev_insn_cc0;
3760 mode_arg = prev_insn_cc0_mode;
3761 const_arg = equiv_constant (folded_arg);
3762 break;
3763 #endif
3765 default:
3766 folded_arg = fold_rtx (arg, insn);
3767 const_arg = equiv_constant (folded_arg);
3770 /* For the first three operands, see if the operand
3771 is constant or equivalent to a constant. */
3772 switch (i)
3774 case 0:
3775 folded_arg0 = folded_arg;
3776 const_arg0 = const_arg;
3777 mode_arg0 = mode_arg;
3778 break;
3779 case 1:
3780 folded_arg1 = folded_arg;
3781 const_arg1 = const_arg;
3782 break;
3783 case 2:
3784 const_arg2 = const_arg;
3785 break;
3788 /* Pick the least expensive of the folded argument and an
3789 equivalent constant argument. */
3790 if (const_arg == 0 || const_arg == folded_arg
3791 || COST_IN (const_arg, code) > COST_IN (folded_arg, code))
3792 cheap_arg = folded_arg, expensive_arg = const_arg;
3793 else
3794 cheap_arg = const_arg, expensive_arg = folded_arg;
3796 /* Try to replace the operand with the cheapest of the two
3797 possibilities. If it doesn't work and this is either of the first
3798 two operands of a commutative operation, try swapping them.
3799 If THAT fails, try the more expensive, provided it is cheaper
3800 than what is already there. */
3802 if (cheap_arg == XEXP (x, i))
3803 continue;
3805 if (insn == 0 && ! copied)
3807 x = copy_rtx (x);
3808 copied = 1;
3811 /* Order the replacements from cheapest to most expensive. */
3812 replacements[0] = cheap_arg;
3813 replacements[1] = expensive_arg;
3815 for (j = 0; j < 2 && replacements[j]; j++)
3817 int new_cost = COST_IN (replacements[j], code);
3819 /* Stop if what existed before was cheaper. Prefer constants
3820 in the case of a tie. */
3821 if (new_cost > old_cost
3822 || (new_cost == old_cost && CONSTANT_P (XEXP (x, i))))
3823 break;
3825 /* It's not safe to substitute the operand of a conversion
3826 operator with a constant, as the conversion's identity
3827 depends upon the mode of its operand. This optimization
3828 is handled by the call to simplify_unary_operation. */
3829 if (GET_RTX_CLASS (code) == RTX_UNARY
3830 && GET_MODE (replacements[j]) != mode_arg0
3831 && (code == ZERO_EXTEND
3832 || code == SIGN_EXTEND
3833 || code == TRUNCATE
3834 || code == FLOAT_TRUNCATE
3835 || code == FLOAT_EXTEND
3836 || code == FLOAT
3837 || code == FIX
3838 || code == UNSIGNED_FLOAT
3839 || code == UNSIGNED_FIX))
3840 continue;
3842 if (validate_change (insn, &XEXP (x, i), replacements[j], 0))
3843 break;
3845 if (GET_RTX_CLASS (code) == RTX_COMM_COMPARE
3846 || GET_RTX_CLASS (code) == RTX_COMM_ARITH)
3848 validate_change (insn, &XEXP (x, i), XEXP (x, 1 - i), 1);
3849 validate_change (insn, &XEXP (x, 1 - i), replacements[j], 1);
3851 if (apply_change_group ())
3853 /* Swap them back to be invalid so that this loop can
3854 continue and flag them to be swapped back later. */
3855 rtx tem;
3857 tem = XEXP (x, 0); XEXP (x, 0) = XEXP (x, 1);
3858 XEXP (x, 1) = tem;
3859 must_swap = 1;
3860 break;
3866 else
3868 if (fmt[i] == 'E')
3869 /* Don't try to fold inside of a vector of expressions.
3870 Doing nothing is harmless. */
3874 /* If a commutative operation, place a constant integer as the second
3875 operand unless the first operand is also a constant integer. Otherwise,
3876 place any constant second unless the first operand is also a constant. */
3878 if (COMMUTATIVE_P (x))
3880 if (must_swap
3881 || swap_commutative_operands_p (const_arg0 ? const_arg0
3882 : XEXP (x, 0),
3883 const_arg1 ? const_arg1
3884 : XEXP (x, 1)))
3886 rtx tem = XEXP (x, 0);
3888 if (insn == 0 && ! copied)
3890 x = copy_rtx (x);
3891 copied = 1;
3894 validate_change (insn, &XEXP (x, 0), XEXP (x, 1), 1);
3895 validate_change (insn, &XEXP (x, 1), tem, 1);
3896 if (apply_change_group ())
3898 tem = const_arg0, const_arg0 = const_arg1, const_arg1 = tem;
3899 tem = folded_arg0, folded_arg0 = folded_arg1, folded_arg1 = tem;
3904 /* If X is an arithmetic operation, see if we can simplify it. */
3906 switch (GET_RTX_CLASS (code))
3908 case RTX_UNARY:
3910 int is_const = 0;
3912 /* We can't simplify extension ops unless we know the
3913 original mode. */
3914 if ((code == ZERO_EXTEND || code == SIGN_EXTEND)
3915 && mode_arg0 == VOIDmode)
3916 break;
3918 /* If we had a CONST, strip it off and put it back later if we
3919 fold. */
3920 if (const_arg0 != 0 && GET_CODE (const_arg0) == CONST)
3921 is_const = 1, const_arg0 = XEXP (const_arg0, 0);
3923 new = simplify_unary_operation (code, mode,
3924 const_arg0 ? const_arg0 : folded_arg0,
3925 mode_arg0);
3926 /* NEG of PLUS could be converted into MINUS, but that causes
3927 expressions of the form
3928 (CONST (MINUS (CONST_INT) (SYMBOL_REF)))
3929 which many ports mistakenly treat as LEGITIMATE_CONSTANT_P.
3930 FIXME: those ports should be fixed. */
3931 if (new != 0 && is_const
3932 && GET_CODE (new) == PLUS
3933 && (GET_CODE (XEXP (new, 0)) == SYMBOL_REF
3934 || GET_CODE (XEXP (new, 0)) == LABEL_REF)
3935 && GET_CODE (XEXP (new, 1)) == CONST_INT)
3936 new = gen_rtx_CONST (mode, new);
3938 break;
3940 case RTX_COMPARE:
3941 case RTX_COMM_COMPARE:
3942 /* See what items are actually being compared and set FOLDED_ARG[01]
3943 to those values and CODE to the actual comparison code. If any are
3944 constant, set CONST_ARG0 and CONST_ARG1 appropriately. We needn't
3945 do anything if both operands are already known to be constant. */
3947 /* ??? Vector mode comparisons are not supported yet. */
3948 if (VECTOR_MODE_P (mode))
3949 break;
3951 if (const_arg0 == 0 || const_arg1 == 0)
3953 struct table_elt *p0, *p1;
3954 rtx true_rtx = const_true_rtx, false_rtx = const0_rtx;
3955 enum machine_mode mode_arg1;
3957 #ifdef FLOAT_STORE_FLAG_VALUE
3958 if (SCALAR_FLOAT_MODE_P (mode))
3960 true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
3961 (FLOAT_STORE_FLAG_VALUE (mode), mode));
3962 false_rtx = CONST0_RTX (mode);
3964 #endif
3966 code = find_comparison_args (code, &folded_arg0, &folded_arg1,
3967 &mode_arg0, &mode_arg1);
3969 /* If the mode is VOIDmode or a MODE_CC mode, we don't know
3970 what kinds of things are being compared, so we can't do
3971 anything with this comparison. */
3973 if (mode_arg0 == VOIDmode || GET_MODE_CLASS (mode_arg0) == MODE_CC)
3974 break;
3976 const_arg0 = equiv_constant (folded_arg0);
3977 const_arg1 = equiv_constant (folded_arg1);
3979 /* If we do not now have two constants being compared, see
3980 if we can nevertheless deduce some things about the
3981 comparison. */
3982 if (const_arg0 == 0 || const_arg1 == 0)
3984 if (const_arg1 != NULL)
3986 rtx cheapest_simplification;
3987 int cheapest_cost;
3988 rtx simp_result;
3989 struct table_elt *p;
3991 /* See if we can find an equivalent of folded_arg0
3992 that gets us a cheaper expression, possibly a
3993 constant through simplifications. */
3994 p = lookup (folded_arg0, SAFE_HASH (folded_arg0, mode_arg0),
3995 mode_arg0);
3997 if (p != NULL)
3999 cheapest_simplification = x;
4000 cheapest_cost = COST (x);
4002 for (p = p->first_same_value; p != NULL; p = p->next_same_value)
4004 int cost;
4006 /* If the entry isn't valid, skip it. */
4007 if (! exp_equiv_p (p->exp, p->exp, 1, false))
4008 continue;
4010 /* Try to simplify using this equivalence. */
4011 simp_result
4012 = simplify_relational_operation (code, mode,
4013 mode_arg0,
4014 p->exp,
4015 const_arg1);
4017 if (simp_result == NULL)
4018 continue;
4020 cost = COST (simp_result);
4021 if (cost < cheapest_cost)
4023 cheapest_cost = cost;
4024 cheapest_simplification = simp_result;
4028 /* If we have a cheaper expression now, use that
4029 and try folding it further, from the top. */
4030 if (cheapest_simplification != x)
4031 return fold_rtx (cheapest_simplification, insn);
4035 /* Some addresses are known to be nonzero. We don't know
4036 their sign, but equality comparisons are known. */
4037 if (const_arg1 == const0_rtx
4038 && nonzero_address_p (folded_arg0))
4040 if (code == EQ)
4041 return false_rtx;
4042 else if (code == NE)
4043 return true_rtx;
4046 /* See if the two operands are the same. */
4048 if (folded_arg0 == folded_arg1
4049 || (REG_P (folded_arg0)
4050 && REG_P (folded_arg1)
4051 && (REG_QTY (REGNO (folded_arg0))
4052 == REG_QTY (REGNO (folded_arg1))))
4053 || ((p0 = lookup (folded_arg0,
4054 SAFE_HASH (folded_arg0, mode_arg0),
4055 mode_arg0))
4056 && (p1 = lookup (folded_arg1,
4057 SAFE_HASH (folded_arg1, mode_arg0),
4058 mode_arg0))
4059 && p0->first_same_value == p1->first_same_value))
4061 /* Sadly two equal NaNs are not equivalent. */
4062 if (!HONOR_NANS (mode_arg0))
4063 return ((code == EQ || code == LE || code == GE
4064 || code == LEU || code == GEU || code == UNEQ
4065 || code == UNLE || code == UNGE
4066 || code == ORDERED)
4067 ? true_rtx : false_rtx);
4068 /* Take care for the FP compares we can resolve. */
4069 if (code == UNEQ || code == UNLE || code == UNGE)
4070 return true_rtx;
4071 if (code == LTGT || code == LT || code == GT)
4072 return false_rtx;
4075 /* If FOLDED_ARG0 is a register, see if the comparison we are
4076 doing now is either the same as we did before or the reverse
4077 (we only check the reverse if not floating-point). */
4078 else if (REG_P (folded_arg0))
4080 int qty = REG_QTY (REGNO (folded_arg0));
4082 if (REGNO_QTY_VALID_P (REGNO (folded_arg0)))
4084 struct qty_table_elem *ent = &qty_table[qty];
4086 if ((comparison_dominates_p (ent->comparison_code, code)
4087 || (! FLOAT_MODE_P (mode_arg0)
4088 && comparison_dominates_p (ent->comparison_code,
4089 reverse_condition (code))))
4090 && (rtx_equal_p (ent->comparison_const, folded_arg1)
4091 || (const_arg1
4092 && rtx_equal_p (ent->comparison_const,
4093 const_arg1))
4094 || (REG_P (folded_arg1)
4095 && (REG_QTY (REGNO (folded_arg1)) == ent->comparison_qty))))
4096 return (comparison_dominates_p (ent->comparison_code, code)
4097 ? true_rtx : false_rtx);
4103 /* If we are comparing against zero, see if the first operand is
4104 equivalent to an IOR with a constant. If so, we may be able to
4105 determine the result of this comparison. */
4107 if (const_arg1 == const0_rtx)
4109 rtx y = lookup_as_function (folded_arg0, IOR);
4110 rtx inner_const;
4112 if (y != 0
4113 && (inner_const = equiv_constant (XEXP (y, 1))) != 0
4114 && GET_CODE (inner_const) == CONST_INT
4115 && INTVAL (inner_const) != 0)
4117 int sign_bitnum = GET_MODE_BITSIZE (mode_arg0) - 1;
4118 int has_sign = (HOST_BITS_PER_WIDE_INT >= sign_bitnum
4119 && (INTVAL (inner_const)
4120 & ((HOST_WIDE_INT) 1 << sign_bitnum)));
4121 rtx true_rtx = const_true_rtx, false_rtx = const0_rtx;
4123 #ifdef FLOAT_STORE_FLAG_VALUE
4124 if (SCALAR_FLOAT_MODE_P (mode))
4126 true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
4127 (FLOAT_STORE_FLAG_VALUE (mode), mode));
4128 false_rtx = CONST0_RTX (mode);
4130 #endif
4132 switch (code)
4134 case EQ:
4135 return false_rtx;
4136 case NE:
4137 return true_rtx;
4138 case LT: case LE:
4139 if (has_sign)
4140 return true_rtx;
4141 break;
4142 case GT: case GE:
4143 if (has_sign)
4144 return false_rtx;
4145 break;
4146 default:
4147 break;
4153 rtx op0 = const_arg0 ? const_arg0 : folded_arg0;
4154 rtx op1 = const_arg1 ? const_arg1 : folded_arg1;
4155 new = simplify_relational_operation (code, mode, mode_arg0, op0, op1);
4157 break;
4159 case RTX_BIN_ARITH:
4160 case RTX_COMM_ARITH:
4161 switch (code)
4163 case PLUS:
4164 /* If the second operand is a LABEL_REF, see if the first is a MINUS
4165 with that LABEL_REF as its second operand. If so, the result is
4166 the first operand of that MINUS. This handles switches with an
4167 ADDR_DIFF_VEC table. */
4168 if (const_arg1 && GET_CODE (const_arg1) == LABEL_REF)
4170 rtx y
4171 = GET_CODE (folded_arg0) == MINUS ? folded_arg0
4172 : lookup_as_function (folded_arg0, MINUS);
4174 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
4175 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg1, 0))
4176 return XEXP (y, 0);
4178 /* Now try for a CONST of a MINUS like the above. */
4179 if ((y = (GET_CODE (folded_arg0) == CONST ? folded_arg0
4180 : lookup_as_function (folded_arg0, CONST))) != 0
4181 && GET_CODE (XEXP (y, 0)) == MINUS
4182 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
4183 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg1, 0))
4184 return XEXP (XEXP (y, 0), 0);
4187 /* Likewise if the operands are in the other order. */
4188 if (const_arg0 && GET_CODE (const_arg0) == LABEL_REF)
4190 rtx y
4191 = GET_CODE (folded_arg1) == MINUS ? folded_arg1
4192 : lookup_as_function (folded_arg1, MINUS);
4194 if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
4195 && XEXP (XEXP (y, 1), 0) == XEXP (const_arg0, 0))
4196 return XEXP (y, 0);
4198 /* Now try for a CONST of a MINUS like the above. */
4199 if ((y = (GET_CODE (folded_arg1) == CONST ? folded_arg1
4200 : lookup_as_function (folded_arg1, CONST))) != 0
4201 && GET_CODE (XEXP (y, 0)) == MINUS
4202 && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
4203 && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg0, 0))
4204 return XEXP (XEXP (y, 0), 0);
4207 /* If second operand is a register equivalent to a negative
4208 CONST_INT, see if we can find a register equivalent to the
4209 positive constant. Make a MINUS if so. Don't do this for
4210 a non-negative constant since we might then alternate between
4211 choosing positive and negative constants. Having the positive
4212 constant previously-used is the more common case. Be sure
4213 the resulting constant is non-negative; if const_arg1 were
4214 the smallest negative number this would overflow: depending
4215 on the mode, this would either just be the same value (and
4216 hence not save anything) or be incorrect. */
4217 if (const_arg1 != 0 && GET_CODE (const_arg1) == CONST_INT
4218 && INTVAL (const_arg1) < 0
4219 /* This used to test
4221 -INTVAL (const_arg1) >= 0
4223 But The Sun V5.0 compilers mis-compiled that test. So
4224 instead we test for the problematic value in a more direct
4225 manner and hope the Sun compilers get it correct. */
4226 && INTVAL (const_arg1) !=
4227 ((HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1))
4228 && REG_P (folded_arg1))
4230 rtx new_const = GEN_INT (-INTVAL (const_arg1));
4231 struct table_elt *p
4232 = lookup (new_const, SAFE_HASH (new_const, mode), mode);
4234 if (p)
4235 for (p = p->first_same_value; p; p = p->next_same_value)
4236 if (REG_P (p->exp))
4237 return simplify_gen_binary (MINUS, mode, folded_arg0,
4238 canon_reg (p->exp, NULL_RTX));
4240 goto from_plus;
4242 case MINUS:
4243 /* If we have (MINUS Y C), see if Y is known to be (PLUS Z C2).
4244 If so, produce (PLUS Z C2-C). */
4245 if (const_arg1 != 0 && GET_CODE (const_arg1) == CONST_INT)
4247 rtx y = lookup_as_function (XEXP (x, 0), PLUS);
4248 if (y && GET_CODE (XEXP (y, 1)) == CONST_INT)
4249 return fold_rtx (plus_constant (copy_rtx (y),
4250 -INTVAL (const_arg1)),
4251 NULL_RTX);
4254 /* Fall through. */
4256 from_plus:
4257 case SMIN: case SMAX: case UMIN: case UMAX:
4258 case IOR: case AND: case XOR:
4259 case MULT:
4260 case ASHIFT: case LSHIFTRT: case ASHIFTRT:
4261 /* If we have (<op> <reg> <const_int>) for an associative OP and REG
4262 is known to be of similar form, we may be able to replace the
4263 operation with a combined operation. This may eliminate the
4264 intermediate operation if every use is simplified in this way.
4265 Note that the similar optimization done by combine.c only works
4266 if the intermediate operation's result has only one reference. */
4268 if (REG_P (folded_arg0)
4269 && const_arg1 && GET_CODE (const_arg1) == CONST_INT)
4271 int is_shift
4272 = (code == ASHIFT || code == ASHIFTRT || code == LSHIFTRT);
4273 rtx y = lookup_as_function (folded_arg0, code);
4274 rtx inner_const;
4275 enum rtx_code associate_code;
4276 rtx new_const;
4278 if (y == 0
4279 || 0 == (inner_const
4280 = equiv_constant (fold_rtx (XEXP (y, 1), 0)))
4281 || GET_CODE (inner_const) != CONST_INT
4282 /* If we have compiled a statement like
4283 "if (x == (x & mask1))", and now are looking at
4284 "x & mask2", we will have a case where the first operand
4285 of Y is the same as our first operand. Unless we detect
4286 this case, an infinite loop will result. */
4287 || XEXP (y, 0) == folded_arg0)
4288 break;
4290 /* Don't associate these operations if they are a PLUS with the
4291 same constant and it is a power of two. These might be doable
4292 with a pre- or post-increment. Similarly for two subtracts of
4293 identical powers of two with post decrement. */
4295 if (code == PLUS && const_arg1 == inner_const
4296 && ((HAVE_PRE_INCREMENT
4297 && exact_log2 (INTVAL (const_arg1)) >= 0)
4298 || (HAVE_POST_INCREMENT
4299 && exact_log2 (INTVAL (const_arg1)) >= 0)
4300 || (HAVE_PRE_DECREMENT
4301 && exact_log2 (- INTVAL (const_arg1)) >= 0)
4302 || (HAVE_POST_DECREMENT
4303 && exact_log2 (- INTVAL (const_arg1)) >= 0)))
4304 break;
4306 /* Compute the code used to compose the constants. For example,
4307 A-C1-C2 is A-(C1 + C2), so if CODE == MINUS, we want PLUS. */
4309 associate_code = (is_shift || code == MINUS ? PLUS : code);
4311 new_const = simplify_binary_operation (associate_code, mode,
4312 const_arg1, inner_const);
4314 if (new_const == 0)
4315 break;
4317 /* If we are associating shift operations, don't let this
4318 produce a shift of the size of the object or larger.
4319 This could occur when we follow a sign-extend by a right
4320 shift on a machine that does a sign-extend as a pair
4321 of shifts. */
4323 if (is_shift && GET_CODE (new_const) == CONST_INT
4324 && INTVAL (new_const) >= GET_MODE_BITSIZE (mode))
4326 /* As an exception, we can turn an ASHIFTRT of this
4327 form into a shift of the number of bits - 1. */
4328 if (code == ASHIFTRT)
4329 new_const = GEN_INT (GET_MODE_BITSIZE (mode) - 1);
4330 else
4331 break;
4334 y = copy_rtx (XEXP (y, 0));
4336 /* If Y contains our first operand (the most common way this
4337 can happen is if Y is a MEM), we would do into an infinite
4338 loop if we tried to fold it. So don't in that case. */
4340 if (! reg_mentioned_p (folded_arg0, y))
4341 y = fold_rtx (y, insn);
4343 return simplify_gen_binary (code, mode, y, new_const);
4345 break;
4347 case DIV: case UDIV:
4348 /* ??? The associative optimization performed immediately above is
4349 also possible for DIV and UDIV using associate_code of MULT.
4350 However, we would need extra code to verify that the
4351 multiplication does not overflow, that is, there is no overflow
4352 in the calculation of new_const. */
4353 break;
4355 default:
4356 break;
4359 new = simplify_binary_operation (code, mode,
4360 const_arg0 ? const_arg0 : folded_arg0,
4361 const_arg1 ? const_arg1 : folded_arg1);
4362 break;
4364 case RTX_OBJ:
4365 /* (lo_sum (high X) X) is simply X. */
4366 if (code == LO_SUM && const_arg0 != 0
4367 && GET_CODE (const_arg0) == HIGH
4368 && rtx_equal_p (XEXP (const_arg0, 0), const_arg1))
4369 return const_arg1;
4370 break;
4372 case RTX_TERNARY:
4373 case RTX_BITFIELD_OPS:
4374 new = simplify_ternary_operation (code, mode, mode_arg0,
4375 const_arg0 ? const_arg0 : folded_arg0,
4376 const_arg1 ? const_arg1 : folded_arg1,
4377 const_arg2 ? const_arg2 : XEXP (x, 2));
4378 break;
4380 default:
4381 break;
4384 return new ? new : x;
4387 /* Return a constant value currently equivalent to X.
4388 Return 0 if we don't know one. */
4390 static rtx
4391 equiv_constant (rtx x)
4393 if (REG_P (x)
4394 && REGNO_QTY_VALID_P (REGNO (x)))
4396 int x_q = REG_QTY (REGNO (x));
4397 struct qty_table_elem *x_ent = &qty_table[x_q];
4399 if (x_ent->const_rtx)
4400 x = gen_lowpart (GET_MODE (x), x_ent->const_rtx);
4403 if (x == 0 || CONSTANT_P (x))
4404 return x;
4406 /* If X is a MEM, try to fold it outside the context of any insn to see if
4407 it might be equivalent to a constant. That handles the case where it
4408 is a constant-pool reference. Then try to look it up in the hash table
4409 in case it is something whose value we have seen before. */
4411 if (MEM_P (x))
4413 struct table_elt *elt;
4415 x = fold_rtx (x, NULL_RTX);
4416 if (CONSTANT_P (x))
4417 return x;
4419 elt = lookup (x, SAFE_HASH (x, GET_MODE (x)), GET_MODE (x));
4420 if (elt == 0)
4421 return 0;
4423 for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
4424 if (elt->is_const && CONSTANT_P (elt->exp))
4425 return elt->exp;
4428 return 0;
4431 /* Given INSN, a jump insn, PATH_TAKEN indicates if we are following the "taken"
4432 branch. It will be zero if not.
4434 In certain cases, this can cause us to add an equivalence. For example,
4435 if we are following the taken case of
4436 if (i == 2)
4437 we can add the fact that `i' and '2' are now equivalent.
4439 In any case, we can record that this comparison was passed. If the same
4440 comparison is seen later, we will know its value. */
4442 static void
4443 record_jump_equiv (rtx insn, int taken)
4445 int cond_known_true;
4446 rtx op0, op1;
4447 rtx set;
4448 enum machine_mode mode, mode0, mode1;
4449 int reversed_nonequality = 0;
4450 enum rtx_code code;
4452 /* Ensure this is the right kind of insn. */
4453 if (! any_condjump_p (insn))
4454 return;
4455 set = pc_set (insn);
4457 /* See if this jump condition is known true or false. */
4458 if (taken)
4459 cond_known_true = (XEXP (SET_SRC (set), 2) == pc_rtx);
4460 else
4461 cond_known_true = (XEXP (SET_SRC (set), 1) == pc_rtx);
4463 /* Get the type of comparison being done and the operands being compared.
4464 If we had to reverse a non-equality condition, record that fact so we
4465 know that it isn't valid for floating-point. */
4466 code = GET_CODE (XEXP (SET_SRC (set), 0));
4467 op0 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 0), insn);
4468 op1 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 1), insn);
4470 code = find_comparison_args (code, &op0, &op1, &mode0, &mode1);
4471 if (! cond_known_true)
4473 code = reversed_comparison_code_parts (code, op0, op1, insn);
4475 /* Don't remember if we can't find the inverse. */
4476 if (code == UNKNOWN)
4477 return;
4480 /* The mode is the mode of the non-constant. */
4481 mode = mode0;
4482 if (mode1 != VOIDmode)
4483 mode = mode1;
4485 record_jump_cond (code, mode, op0, op1, reversed_nonequality);
4488 /* Yet another form of subreg creation. In this case, we want something in
4489 MODE, and we should assume OP has MODE iff it is naturally modeless. */
4491 static rtx
4492 record_jump_cond_subreg (enum machine_mode mode, rtx op)
4494 enum machine_mode op_mode = GET_MODE (op);
4495 if (op_mode == mode || op_mode == VOIDmode)
4496 return op;
4497 return lowpart_subreg (mode, op, op_mode);
4500 /* We know that comparison CODE applied to OP0 and OP1 in MODE is true.
4501 REVERSED_NONEQUALITY is nonzero if CODE had to be swapped.
4502 Make any useful entries we can with that information. Called from
4503 above function and called recursively. */
4505 static void
4506 record_jump_cond (enum rtx_code code, enum machine_mode mode, rtx op0,
4507 rtx op1, int reversed_nonequality)
4509 unsigned op0_hash, op1_hash;
4510 int op0_in_memory, op1_in_memory;
4511 struct table_elt *op0_elt, *op1_elt;
4513 /* If OP0 and OP1 are known equal, and either is a paradoxical SUBREG,
4514 we know that they are also equal in the smaller mode (this is also
4515 true for all smaller modes whether or not there is a SUBREG, but
4516 is not worth testing for with no SUBREG). */
4518 /* Note that GET_MODE (op0) may not equal MODE. */
4519 if (code == EQ && GET_CODE (op0) == SUBREG
4520 && (GET_MODE_SIZE (GET_MODE (op0))
4521 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
4523 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
4524 rtx tem = record_jump_cond_subreg (inner_mode, op1);
4525 if (tem)
4526 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
4527 reversed_nonequality);
4530 if (code == EQ && GET_CODE (op1) == SUBREG
4531 && (GET_MODE_SIZE (GET_MODE (op1))
4532 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
4534 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
4535 rtx tem = record_jump_cond_subreg (inner_mode, op0);
4536 if (tem)
4537 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
4538 reversed_nonequality);
4541 /* Similarly, if this is an NE comparison, and either is a SUBREG
4542 making a smaller mode, we know the whole thing is also NE. */
4544 /* Note that GET_MODE (op0) may not equal MODE;
4545 if we test MODE instead, we can get an infinite recursion
4546 alternating between two modes each wider than MODE. */
4548 if (code == NE && GET_CODE (op0) == SUBREG
4549 && subreg_lowpart_p (op0)
4550 && (GET_MODE_SIZE (GET_MODE (op0))
4551 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
4553 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
4554 rtx tem = record_jump_cond_subreg (inner_mode, op1);
4555 if (tem)
4556 record_jump_cond (code, mode, SUBREG_REG (op0), tem,
4557 reversed_nonequality);
4560 if (code == NE && GET_CODE (op1) == SUBREG
4561 && subreg_lowpart_p (op1)
4562 && (GET_MODE_SIZE (GET_MODE (op1))
4563 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
4565 enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
4566 rtx tem = record_jump_cond_subreg (inner_mode, op0);
4567 if (tem)
4568 record_jump_cond (code, mode, SUBREG_REG (op1), tem,
4569 reversed_nonequality);
4572 /* Hash both operands. */
4574 do_not_record = 0;
4575 hash_arg_in_memory = 0;
4576 op0_hash = HASH (op0, mode);
4577 op0_in_memory = hash_arg_in_memory;
4579 if (do_not_record)
4580 return;
4582 do_not_record = 0;
4583 hash_arg_in_memory = 0;
4584 op1_hash = HASH (op1, mode);
4585 op1_in_memory = hash_arg_in_memory;
4587 if (do_not_record)
4588 return;
4590 /* Look up both operands. */
4591 op0_elt = lookup (op0, op0_hash, mode);
4592 op1_elt = lookup (op1, op1_hash, mode);
4594 /* If both operands are already equivalent or if they are not in the
4595 table but are identical, do nothing. */
4596 if ((op0_elt != 0 && op1_elt != 0
4597 && op0_elt->first_same_value == op1_elt->first_same_value)
4598 || op0 == op1 || rtx_equal_p (op0, op1))
4599 return;
4601 /* If we aren't setting two things equal all we can do is save this
4602 comparison. Similarly if this is floating-point. In the latter
4603 case, OP1 might be zero and both -0.0 and 0.0 are equal to it.
4604 If we record the equality, we might inadvertently delete code
4605 whose intent was to change -0 to +0. */
4607 if (code != EQ || FLOAT_MODE_P (GET_MODE (op0)))
4609 struct qty_table_elem *ent;
4610 int qty;
4612 /* If we reversed a floating-point comparison, if OP0 is not a
4613 register, or if OP1 is neither a register or constant, we can't
4614 do anything. */
4616 if (!REG_P (op1))
4617 op1 = equiv_constant (op1);
4619 if ((reversed_nonequality && FLOAT_MODE_P (mode))
4620 || !REG_P (op0) || op1 == 0)
4621 return;
4623 /* Put OP0 in the hash table if it isn't already. This gives it a
4624 new quantity number. */
4625 if (op0_elt == 0)
4627 if (insert_regs (op0, NULL, 0))
4629 rehash_using_reg (op0);
4630 op0_hash = HASH (op0, mode);
4632 /* If OP0 is contained in OP1, this changes its hash code
4633 as well. Faster to rehash than to check, except
4634 for the simple case of a constant. */
4635 if (! CONSTANT_P (op1))
4636 op1_hash = HASH (op1,mode);
4639 op0_elt = insert (op0, NULL, op0_hash, mode);
4640 op0_elt->in_memory = op0_in_memory;
4643 qty = REG_QTY (REGNO (op0));
4644 ent = &qty_table[qty];
4646 ent->comparison_code = code;
4647 if (REG_P (op1))
4649 /* Look it up again--in case op0 and op1 are the same. */
4650 op1_elt = lookup (op1, op1_hash, mode);
4652 /* Put OP1 in the hash table so it gets a new quantity number. */
4653 if (op1_elt == 0)
4655 if (insert_regs (op1, NULL, 0))
4657 rehash_using_reg (op1);
4658 op1_hash = HASH (op1, mode);
4661 op1_elt = insert (op1, NULL, op1_hash, mode);
4662 op1_elt->in_memory = op1_in_memory;
4665 ent->comparison_const = NULL_RTX;
4666 ent->comparison_qty = REG_QTY (REGNO (op1));
4668 else
4670 ent->comparison_const = op1;
4671 ent->comparison_qty = -1;
4674 return;
4677 /* If either side is still missing an equivalence, make it now,
4678 then merge the equivalences. */
4680 if (op0_elt == 0)
4682 if (insert_regs (op0, NULL, 0))
4684 rehash_using_reg (op0);
4685 op0_hash = HASH (op0, mode);
4688 op0_elt = insert (op0, NULL, op0_hash, mode);
4689 op0_elt->in_memory = op0_in_memory;
4692 if (op1_elt == 0)
4694 if (insert_regs (op1, NULL, 0))
4696 rehash_using_reg (op1);
4697 op1_hash = HASH (op1, mode);
4700 op1_elt = insert (op1, NULL, op1_hash, mode);
4701 op1_elt->in_memory = op1_in_memory;
4704 merge_equiv_classes (op0_elt, op1_elt);
4707 /* CSE processing for one instruction.
4708 First simplify sources and addresses of all assignments
4709 in the instruction, using previously-computed equivalents values.
4710 Then install the new sources and destinations in the table
4711 of available values.
4713 If LIBCALL_INSN is nonzero, don't record any equivalence made in
4714 the insn. It means that INSN is inside libcall block. In this
4715 case LIBCALL_INSN is the corresponding insn with REG_LIBCALL. */
4717 /* Data on one SET contained in the instruction. */
4719 struct set
4721 /* The SET rtx itself. */
4722 rtx rtl;
4723 /* The SET_SRC of the rtx (the original value, if it is changing). */
4724 rtx src;
4725 /* The hash-table element for the SET_SRC of the SET. */
4726 struct table_elt *src_elt;
4727 /* Hash value for the SET_SRC. */
4728 unsigned src_hash;
4729 /* Hash value for the SET_DEST. */
4730 unsigned dest_hash;
4731 /* The SET_DEST, with SUBREG, etc., stripped. */
4732 rtx inner_dest;
4733 /* Nonzero if the SET_SRC is in memory. */
4734 char src_in_memory;
4735 /* Nonzero if the SET_SRC contains something
4736 whose value cannot be predicted and understood. */
4737 char src_volatile;
4738 /* Original machine mode, in case it becomes a CONST_INT.
4739 The size of this field should match the size of the mode
4740 field of struct rtx_def (see rtl.h). */
4741 ENUM_BITFIELD(machine_mode) mode : 8;
4742 /* A constant equivalent for SET_SRC, if any. */
4743 rtx src_const;
4744 /* Original SET_SRC value used for libcall notes. */
4745 rtx orig_src;
4746 /* Hash value of constant equivalent for SET_SRC. */
4747 unsigned src_const_hash;
4748 /* Table entry for constant equivalent for SET_SRC, if any. */
4749 struct table_elt *src_const_elt;
4752 static void
4753 cse_insn (rtx insn, rtx libcall_insn)
4755 rtx x = PATTERN (insn);
4756 int i;
4757 rtx tem;
4758 int n_sets = 0;
4760 #ifdef HAVE_cc0
4761 /* Records what this insn does to set CC0. */
4762 rtx this_insn_cc0 = 0;
4763 enum machine_mode this_insn_cc0_mode = VOIDmode;
4764 #endif
4766 rtx src_eqv = 0;
4767 struct table_elt *src_eqv_elt = 0;
4768 int src_eqv_volatile = 0;
4769 int src_eqv_in_memory = 0;
4770 unsigned src_eqv_hash = 0;
4772 struct set *sets = (struct set *) 0;
4774 this_insn = insn;
4776 /* Find all the SETs and CLOBBERs in this instruction.
4777 Record all the SETs in the array `set' and count them.
4778 Also determine whether there is a CLOBBER that invalidates
4779 all memory references, or all references at varying addresses. */
4781 if (CALL_P (insn))
4783 for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
4785 if (GET_CODE (XEXP (tem, 0)) == CLOBBER)
4786 invalidate (SET_DEST (XEXP (tem, 0)), VOIDmode);
4787 XEXP (tem, 0) = canon_reg (XEXP (tem, 0), insn);
4791 if (GET_CODE (x) == SET)
4793 sets = alloca (sizeof (struct set));
4794 sets[0].rtl = x;
4796 /* Ignore SETs that are unconditional jumps.
4797 They never need cse processing, so this does not hurt.
4798 The reason is not efficiency but rather
4799 so that we can test at the end for instructions
4800 that have been simplified to unconditional jumps
4801 and not be misled by unchanged instructions
4802 that were unconditional jumps to begin with. */
4803 if (SET_DEST (x) == pc_rtx
4804 && GET_CODE (SET_SRC (x)) == LABEL_REF)
4807 /* Don't count call-insns, (set (reg 0) (call ...)), as a set.
4808 The hard function value register is used only once, to copy to
4809 someplace else, so it isn't worth cse'ing (and on 80386 is unsafe)!
4810 Ensure we invalidate the destination register. On the 80386 no
4811 other code would invalidate it since it is a fixed_reg.
4812 We need not check the return of apply_change_group; see canon_reg. */
4814 else if (GET_CODE (SET_SRC (x)) == CALL)
4816 canon_reg (SET_SRC (x), insn);
4817 apply_change_group ();
4818 fold_rtx (SET_SRC (x), insn);
4819 invalidate (SET_DEST (x), VOIDmode);
4821 else
4822 n_sets = 1;
4824 else if (GET_CODE (x) == PARALLEL)
4826 int lim = XVECLEN (x, 0);
4828 sets = alloca (lim * sizeof (struct set));
4830 /* Find all regs explicitly clobbered in this insn,
4831 and ensure they are not replaced with any other regs
4832 elsewhere in this insn.
4833 When a reg that is clobbered is also used for input,
4834 we should presume that that is for a reason,
4835 and we should not substitute some other register
4836 which is not supposed to be clobbered.
4837 Therefore, this loop cannot be merged into the one below
4838 because a CALL may precede a CLOBBER and refer to the
4839 value clobbered. We must not let a canonicalization do
4840 anything in that case. */
4841 for (i = 0; i < lim; i++)
4843 rtx y = XVECEXP (x, 0, i);
4844 if (GET_CODE (y) == CLOBBER)
4846 rtx clobbered = XEXP (y, 0);
4848 if (REG_P (clobbered)
4849 || GET_CODE (clobbered) == SUBREG)
4850 invalidate (clobbered, VOIDmode);
4851 else if (GET_CODE (clobbered) == STRICT_LOW_PART
4852 || GET_CODE (clobbered) == ZERO_EXTRACT)
4853 invalidate (XEXP (clobbered, 0), GET_MODE (clobbered));
4857 for (i = 0; i < lim; i++)
4859 rtx y = XVECEXP (x, 0, i);
4860 if (GET_CODE (y) == SET)
4862 /* As above, we ignore unconditional jumps and call-insns and
4863 ignore the result of apply_change_group. */
4864 if (GET_CODE (SET_SRC (y)) == CALL)
4866 canon_reg (SET_SRC (y), insn);
4867 apply_change_group ();
4868 fold_rtx (SET_SRC (y), insn);
4869 invalidate (SET_DEST (y), VOIDmode);
4871 else if (SET_DEST (y) == pc_rtx
4872 && GET_CODE (SET_SRC (y)) == LABEL_REF)
4874 else
4875 sets[n_sets++].rtl = y;
4877 else if (GET_CODE (y) == CLOBBER)
4879 /* If we clobber memory, canon the address.
4880 This does nothing when a register is clobbered
4881 because we have already invalidated the reg. */
4882 if (MEM_P (XEXP (y, 0)))
4883 canon_reg (XEXP (y, 0), NULL_RTX);
4885 else if (GET_CODE (y) == USE
4886 && ! (REG_P (XEXP (y, 0))
4887 && REGNO (XEXP (y, 0)) < FIRST_PSEUDO_REGISTER))
4888 canon_reg (y, NULL_RTX);
4889 else if (GET_CODE (y) == CALL)
4891 /* The result of apply_change_group can be ignored; see
4892 canon_reg. */
4893 canon_reg (y, insn);
4894 apply_change_group ();
4895 fold_rtx (y, insn);
4899 else if (GET_CODE (x) == CLOBBER)
4901 if (MEM_P (XEXP (x, 0)))
4902 canon_reg (XEXP (x, 0), NULL_RTX);
4905 /* Canonicalize a USE of a pseudo register or memory location. */
4906 else if (GET_CODE (x) == USE
4907 && ! (REG_P (XEXP (x, 0))
4908 && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER))
4909 canon_reg (XEXP (x, 0), NULL_RTX);
4910 else if (GET_CODE (x) == CALL)
4912 /* The result of apply_change_group can be ignored; see canon_reg. */
4913 canon_reg (x, insn);
4914 apply_change_group ();
4915 fold_rtx (x, insn);
4918 /* Store the equivalent value in SRC_EQV, if different, or if the DEST
4919 is a STRICT_LOW_PART. The latter condition is necessary because SRC_EQV
4920 is handled specially for this case, and if it isn't set, then there will
4921 be no equivalence for the destination. */
4922 if (n_sets == 1 && REG_NOTES (insn) != 0
4923 && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0
4924 && (! rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl))
4925 || GET_CODE (SET_DEST (sets[0].rtl)) == STRICT_LOW_PART))
4927 src_eqv = fold_rtx (canon_reg (XEXP (tem, 0), NULL_RTX), insn);
4928 XEXP (tem, 0) = src_eqv;
4931 /* Canonicalize sources and addresses of destinations.
4932 We do this in a separate pass to avoid problems when a MATCH_DUP is
4933 present in the insn pattern. In that case, we want to ensure that
4934 we don't break the duplicate nature of the pattern. So we will replace
4935 both operands at the same time. Otherwise, we would fail to find an
4936 equivalent substitution in the loop calling validate_change below.
4938 We used to suppress canonicalization of DEST if it appears in SRC,
4939 but we don't do this any more. */
4941 for (i = 0; i < n_sets; i++)
4943 rtx dest = SET_DEST (sets[i].rtl);
4944 rtx src = SET_SRC (sets[i].rtl);
4945 rtx new = canon_reg (src, insn);
4946 int insn_code;
4948 sets[i].orig_src = src;
4949 if ((REG_P (new) && REG_P (src)
4950 && ((REGNO (new) < FIRST_PSEUDO_REGISTER)
4951 != (REGNO (src) < FIRST_PSEUDO_REGISTER)))
4952 || (insn_code = recog_memoized (insn)) < 0
4953 || insn_data[insn_code].n_dups > 0)
4954 validate_change (insn, &SET_SRC (sets[i].rtl), new, 1);
4955 else
4956 SET_SRC (sets[i].rtl) = new;
4958 if (GET_CODE (dest) == ZERO_EXTRACT)
4960 validate_change (insn, &XEXP (dest, 1),
4961 canon_reg (XEXP (dest, 1), insn), 1);
4962 validate_change (insn, &XEXP (dest, 2),
4963 canon_reg (XEXP (dest, 2), insn), 1);
4966 while (GET_CODE (dest) == SUBREG
4967 || GET_CODE (dest) == ZERO_EXTRACT
4968 || GET_CODE (dest) == STRICT_LOW_PART)
4969 dest = XEXP (dest, 0);
4971 if (MEM_P (dest))
4972 canon_reg (dest, insn);
4975 /* Now that we have done all the replacements, we can apply the change
4976 group and see if they all work. Note that this will cause some
4977 canonicalizations that would have worked individually not to be applied
4978 because some other canonicalization didn't work, but this should not
4979 occur often.
4981 The result of apply_change_group can be ignored; see canon_reg. */
4983 apply_change_group ();
4985 /* Set sets[i].src_elt to the class each source belongs to.
4986 Detect assignments from or to volatile things
4987 and set set[i] to zero so they will be ignored
4988 in the rest of this function.
4990 Nothing in this loop changes the hash table or the register chains. */
4992 for (i = 0; i < n_sets; i++)
4994 rtx src, dest;
4995 rtx src_folded;
4996 struct table_elt *elt = 0, *p;
4997 enum machine_mode mode;
4998 rtx src_eqv_here;
4999 rtx src_const = 0;
5000 rtx src_related = 0;
5001 struct table_elt *src_const_elt = 0;
5002 int src_cost = MAX_COST;
5003 int src_eqv_cost = MAX_COST;
5004 int src_folded_cost = MAX_COST;
5005 int src_related_cost = MAX_COST;
5006 int src_elt_cost = MAX_COST;
5007 int src_regcost = MAX_COST;
5008 int src_eqv_regcost = MAX_COST;
5009 int src_folded_regcost = MAX_COST;
5010 int src_related_regcost = MAX_COST;
5011 int src_elt_regcost = MAX_COST;
5012 /* Set nonzero if we need to call force_const_mem on with the
5013 contents of src_folded before using it. */
5014 int src_folded_force_flag = 0;
5016 dest = SET_DEST (sets[i].rtl);
5017 src = SET_SRC (sets[i].rtl);
5019 /* If SRC is a constant that has no machine mode,
5020 hash it with the destination's machine mode.
5021 This way we can keep different modes separate. */
5023 mode = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
5024 sets[i].mode = mode;
5026 if (src_eqv)
5028 enum machine_mode eqvmode = mode;
5029 if (GET_CODE (dest) == STRICT_LOW_PART)
5030 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
5031 do_not_record = 0;
5032 hash_arg_in_memory = 0;
5033 src_eqv_hash = HASH (src_eqv, eqvmode);
5035 /* Find the equivalence class for the equivalent expression. */
5037 if (!do_not_record)
5038 src_eqv_elt = lookup (src_eqv, src_eqv_hash, eqvmode);
5040 src_eqv_volatile = do_not_record;
5041 src_eqv_in_memory = hash_arg_in_memory;
5044 /* If this is a STRICT_LOW_PART assignment, src_eqv corresponds to the
5045 value of the INNER register, not the destination. So it is not
5046 a valid substitution for the source. But save it for later. */
5047 if (GET_CODE (dest) == STRICT_LOW_PART)
5048 src_eqv_here = 0;
5049 else
5050 src_eqv_here = src_eqv;
5052 /* Simplify and foldable subexpressions in SRC. Then get the fully-
5053 simplified result, which may not necessarily be valid. */
5054 src_folded = fold_rtx (src, insn);
5056 #if 0
5057 /* ??? This caused bad code to be generated for the m68k port with -O2.
5058 Suppose src is (CONST_INT -1), and that after truncation src_folded
5059 is (CONST_INT 3). Suppose src_folded is then used for src_const.
5060 At the end we will add src and src_const to the same equivalence
5061 class. We now have 3 and -1 on the same equivalence class. This
5062 causes later instructions to be mis-optimized. */
5063 /* If storing a constant in a bitfield, pre-truncate the constant
5064 so we will be able to record it later. */
5065 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
5067 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5069 if (GET_CODE (src) == CONST_INT
5070 && GET_CODE (width) == CONST_INT
5071 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
5072 && (INTVAL (src) & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
5073 src_folded
5074 = GEN_INT (INTVAL (src) & (((HOST_WIDE_INT) 1
5075 << INTVAL (width)) - 1));
5077 #endif
5079 /* Compute SRC's hash code, and also notice if it
5080 should not be recorded at all. In that case,
5081 prevent any further processing of this assignment. */
5082 do_not_record = 0;
5083 hash_arg_in_memory = 0;
5085 sets[i].src = src;
5086 sets[i].src_hash = HASH (src, mode);
5087 sets[i].src_volatile = do_not_record;
5088 sets[i].src_in_memory = hash_arg_in_memory;
5090 /* If SRC is a MEM, there is a REG_EQUIV note for SRC, and DEST is
5091 a pseudo, do not record SRC. Using SRC as a replacement for
5092 anything else will be incorrect in that situation. Note that
5093 this usually occurs only for stack slots, in which case all the
5094 RTL would be referring to SRC, so we don't lose any optimization
5095 opportunities by not having SRC in the hash table. */
5097 if (MEM_P (src)
5098 && find_reg_note (insn, REG_EQUIV, NULL_RTX) != 0
5099 && REG_P (dest)
5100 && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
5101 sets[i].src_volatile = 1;
5103 #if 0
5104 /* It is no longer clear why we used to do this, but it doesn't
5105 appear to still be needed. So let's try without it since this
5106 code hurts cse'ing widened ops. */
5107 /* If source is a paradoxical subreg (such as QI treated as an SI),
5108 treat it as volatile. It may do the work of an SI in one context
5109 where the extra bits are not being used, but cannot replace an SI
5110 in general. */
5111 if (GET_CODE (src) == SUBREG
5112 && (GET_MODE_SIZE (GET_MODE (src))
5113 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))))
5114 sets[i].src_volatile = 1;
5115 #endif
5117 /* Locate all possible equivalent forms for SRC. Try to replace
5118 SRC in the insn with each cheaper equivalent.
5120 We have the following types of equivalents: SRC itself, a folded
5121 version, a value given in a REG_EQUAL note, or a value related
5122 to a constant.
5124 Each of these equivalents may be part of an additional class
5125 of equivalents (if more than one is in the table, they must be in
5126 the same class; we check for this).
5128 If the source is volatile, we don't do any table lookups.
5130 We note any constant equivalent for possible later use in a
5131 REG_NOTE. */
5133 if (!sets[i].src_volatile)
5134 elt = lookup (src, sets[i].src_hash, mode);
5136 sets[i].src_elt = elt;
5138 if (elt && src_eqv_here && src_eqv_elt)
5140 if (elt->first_same_value != src_eqv_elt->first_same_value)
5142 /* The REG_EQUAL is indicating that two formerly distinct
5143 classes are now equivalent. So merge them. */
5144 merge_equiv_classes (elt, src_eqv_elt);
5145 src_eqv_hash = HASH (src_eqv, elt->mode);
5146 src_eqv_elt = lookup (src_eqv, src_eqv_hash, elt->mode);
5149 src_eqv_here = 0;
5152 else if (src_eqv_elt)
5153 elt = src_eqv_elt;
5155 /* Try to find a constant somewhere and record it in `src_const'.
5156 Record its table element, if any, in `src_const_elt'. Look in
5157 any known equivalences first. (If the constant is not in the
5158 table, also set `sets[i].src_const_hash'). */
5159 if (elt)
5160 for (p = elt->first_same_value; p; p = p->next_same_value)
5161 if (p->is_const)
5163 src_const = p->exp;
5164 src_const_elt = elt;
5165 break;
5168 if (src_const == 0
5169 && (CONSTANT_P (src_folded)
5170 /* Consider (minus (label_ref L1) (label_ref L2)) as
5171 "constant" here so we will record it. This allows us
5172 to fold switch statements when an ADDR_DIFF_VEC is used. */
5173 || (GET_CODE (src_folded) == MINUS
5174 && GET_CODE (XEXP (src_folded, 0)) == LABEL_REF
5175 && GET_CODE (XEXP (src_folded, 1)) == LABEL_REF)))
5176 src_const = src_folded, src_const_elt = elt;
5177 else if (src_const == 0 && src_eqv_here && CONSTANT_P (src_eqv_here))
5178 src_const = src_eqv_here, src_const_elt = src_eqv_elt;
5180 /* If we don't know if the constant is in the table, get its
5181 hash code and look it up. */
5182 if (src_const && src_const_elt == 0)
5184 sets[i].src_const_hash = HASH (src_const, mode);
5185 src_const_elt = lookup (src_const, sets[i].src_const_hash, mode);
5188 sets[i].src_const = src_const;
5189 sets[i].src_const_elt = src_const_elt;
5191 /* If the constant and our source are both in the table, mark them as
5192 equivalent. Otherwise, if a constant is in the table but the source
5193 isn't, set ELT to it. */
5194 if (src_const_elt && elt
5195 && src_const_elt->first_same_value != elt->first_same_value)
5196 merge_equiv_classes (elt, src_const_elt);
5197 else if (src_const_elt && elt == 0)
5198 elt = src_const_elt;
5200 /* See if there is a register linearly related to a constant
5201 equivalent of SRC. */
5202 if (src_const
5203 && (GET_CODE (src_const) == CONST
5204 || (src_const_elt && src_const_elt->related_value != 0)))
5206 src_related = use_related_value (src_const, src_const_elt);
5207 if (src_related)
5209 struct table_elt *src_related_elt
5210 = lookup (src_related, HASH (src_related, mode), mode);
5211 if (src_related_elt && elt)
5213 if (elt->first_same_value
5214 != src_related_elt->first_same_value)
5215 /* This can occur when we previously saw a CONST
5216 involving a SYMBOL_REF and then see the SYMBOL_REF
5217 twice. Merge the involved classes. */
5218 merge_equiv_classes (elt, src_related_elt);
5220 src_related = 0;
5221 src_related_elt = 0;
5223 else if (src_related_elt && elt == 0)
5224 elt = src_related_elt;
5228 /* See if we have a CONST_INT that is already in a register in a
5229 wider mode. */
5231 if (src_const && src_related == 0 && GET_CODE (src_const) == CONST_INT
5232 && GET_MODE_CLASS (mode) == MODE_INT
5233 && GET_MODE_BITSIZE (mode) < BITS_PER_WORD)
5235 enum machine_mode wider_mode;
5237 for (wider_mode = GET_MODE_WIDER_MODE (mode);
5238 GET_MODE_BITSIZE (wider_mode) <= BITS_PER_WORD
5239 && src_related == 0;
5240 wider_mode = GET_MODE_WIDER_MODE (wider_mode))
5242 struct table_elt *const_elt
5243 = lookup (src_const, HASH (src_const, wider_mode), wider_mode);
5245 if (const_elt == 0)
5246 continue;
5248 for (const_elt = const_elt->first_same_value;
5249 const_elt; const_elt = const_elt->next_same_value)
5250 if (REG_P (const_elt->exp))
5252 src_related = gen_lowpart (mode,
5253 const_elt->exp);
5254 break;
5259 /* Another possibility is that we have an AND with a constant in
5260 a mode narrower than a word. If so, it might have been generated
5261 as part of an "if" which would narrow the AND. If we already
5262 have done the AND in a wider mode, we can use a SUBREG of that
5263 value. */
5265 if (flag_expensive_optimizations && ! src_related
5266 && GET_CODE (src) == AND && GET_CODE (XEXP (src, 1)) == CONST_INT
5267 && GET_MODE_SIZE (mode) < UNITS_PER_WORD)
5269 enum machine_mode tmode;
5270 rtx new_and = gen_rtx_AND (VOIDmode, NULL_RTX, XEXP (src, 1));
5272 for (tmode = GET_MODE_WIDER_MODE (mode);
5273 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
5274 tmode = GET_MODE_WIDER_MODE (tmode))
5276 rtx inner = gen_lowpart (tmode, XEXP (src, 0));
5277 struct table_elt *larger_elt;
5279 if (inner)
5281 PUT_MODE (new_and, tmode);
5282 XEXP (new_and, 0) = inner;
5283 larger_elt = lookup (new_and, HASH (new_and, tmode), tmode);
5284 if (larger_elt == 0)
5285 continue;
5287 for (larger_elt = larger_elt->first_same_value;
5288 larger_elt; larger_elt = larger_elt->next_same_value)
5289 if (REG_P (larger_elt->exp))
5291 src_related
5292 = gen_lowpart (mode, larger_elt->exp);
5293 break;
5296 if (src_related)
5297 break;
5302 #ifdef LOAD_EXTEND_OP
5303 /* See if a MEM has already been loaded with a widening operation;
5304 if it has, we can use a subreg of that. Many CISC machines
5305 also have such operations, but this is only likely to be
5306 beneficial on these machines. */
5308 if (flag_expensive_optimizations && src_related == 0
5309 && (GET_MODE_SIZE (mode) < UNITS_PER_WORD)
5310 && GET_MODE_CLASS (mode) == MODE_INT
5311 && MEM_P (src) && ! do_not_record
5312 && LOAD_EXTEND_OP (mode) != UNKNOWN)
5314 struct rtx_def memory_extend_buf;
5315 rtx memory_extend_rtx = &memory_extend_buf;
5316 enum machine_mode tmode;
5318 /* Set what we are trying to extend and the operation it might
5319 have been extended with. */
5320 memset (memory_extend_rtx, 0, sizeof(*memory_extend_rtx));
5321 PUT_CODE (memory_extend_rtx, LOAD_EXTEND_OP (mode));
5322 XEXP (memory_extend_rtx, 0) = src;
5324 for (tmode = GET_MODE_WIDER_MODE (mode);
5325 GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
5326 tmode = GET_MODE_WIDER_MODE (tmode))
5328 struct table_elt *larger_elt;
5330 PUT_MODE (memory_extend_rtx, tmode);
5331 larger_elt = lookup (memory_extend_rtx,
5332 HASH (memory_extend_rtx, tmode), tmode);
5333 if (larger_elt == 0)
5334 continue;
5336 for (larger_elt = larger_elt->first_same_value;
5337 larger_elt; larger_elt = larger_elt->next_same_value)
5338 if (REG_P (larger_elt->exp))
5340 src_related = gen_lowpart (mode,
5341 larger_elt->exp);
5342 break;
5345 if (src_related)
5346 break;
5349 #endif /* LOAD_EXTEND_OP */
5351 if (src == src_folded)
5352 src_folded = 0;
5354 /* At this point, ELT, if nonzero, points to a class of expressions
5355 equivalent to the source of this SET and SRC, SRC_EQV, SRC_FOLDED,
5356 and SRC_RELATED, if nonzero, each contain additional equivalent
5357 expressions. Prune these latter expressions by deleting expressions
5358 already in the equivalence class.
5360 Check for an equivalent identical to the destination. If found,
5361 this is the preferred equivalent since it will likely lead to
5362 elimination of the insn. Indicate this by placing it in
5363 `src_related'. */
5365 if (elt)
5366 elt = elt->first_same_value;
5367 for (p = elt; p; p = p->next_same_value)
5369 enum rtx_code code = GET_CODE (p->exp);
5371 /* If the expression is not valid, ignore it. Then we do not
5372 have to check for validity below. In most cases, we can use
5373 `rtx_equal_p', since canonicalization has already been done. */
5374 if (code != REG && ! exp_equiv_p (p->exp, p->exp, 1, false))
5375 continue;
5377 /* Also skip paradoxical subregs, unless that's what we're
5378 looking for. */
5379 if (code == SUBREG
5380 && (GET_MODE_SIZE (GET_MODE (p->exp))
5381 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))
5382 && ! (src != 0
5383 && GET_CODE (src) == SUBREG
5384 && GET_MODE (src) == GET_MODE (p->exp)
5385 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
5386 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))))
5387 continue;
5389 if (src && GET_CODE (src) == code && rtx_equal_p (src, p->exp))
5390 src = 0;
5391 else if (src_folded && GET_CODE (src_folded) == code
5392 && rtx_equal_p (src_folded, p->exp))
5393 src_folded = 0;
5394 else if (src_eqv_here && GET_CODE (src_eqv_here) == code
5395 && rtx_equal_p (src_eqv_here, p->exp))
5396 src_eqv_here = 0;
5397 else if (src_related && GET_CODE (src_related) == code
5398 && rtx_equal_p (src_related, p->exp))
5399 src_related = 0;
5401 /* This is the same as the destination of the insns, we want
5402 to prefer it. Copy it to src_related. The code below will
5403 then give it a negative cost. */
5404 if (GET_CODE (dest) == code && rtx_equal_p (p->exp, dest))
5405 src_related = dest;
5408 /* Find the cheapest valid equivalent, trying all the available
5409 possibilities. Prefer items not in the hash table to ones
5410 that are when they are equal cost. Note that we can never
5411 worsen an insn as the current contents will also succeed.
5412 If we find an equivalent identical to the destination, use it as best,
5413 since this insn will probably be eliminated in that case. */
5414 if (src)
5416 if (rtx_equal_p (src, dest))
5417 src_cost = src_regcost = -1;
5418 else
5420 src_cost = COST (src);
5421 src_regcost = approx_reg_cost (src);
5425 if (src_eqv_here)
5427 if (rtx_equal_p (src_eqv_here, dest))
5428 src_eqv_cost = src_eqv_regcost = -1;
5429 else
5431 src_eqv_cost = COST (src_eqv_here);
5432 src_eqv_regcost = approx_reg_cost (src_eqv_here);
5436 if (src_folded)
5438 if (rtx_equal_p (src_folded, dest))
5439 src_folded_cost = src_folded_regcost = -1;
5440 else
5442 src_folded_cost = COST (src_folded);
5443 src_folded_regcost = approx_reg_cost (src_folded);
5447 if (src_related)
5449 if (rtx_equal_p (src_related, dest))
5450 src_related_cost = src_related_regcost = -1;
5451 else
5453 src_related_cost = COST (src_related);
5454 src_related_regcost = approx_reg_cost (src_related);
5458 /* If this was an indirect jump insn, a known label will really be
5459 cheaper even though it looks more expensive. */
5460 if (dest == pc_rtx && src_const && GET_CODE (src_const) == LABEL_REF)
5461 src_folded = src_const, src_folded_cost = src_folded_regcost = -1;
5463 /* Terminate loop when replacement made. This must terminate since
5464 the current contents will be tested and will always be valid. */
5465 while (1)
5467 rtx trial;
5469 /* Skip invalid entries. */
5470 while (elt && !REG_P (elt->exp)
5471 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
5472 elt = elt->next_same_value;
5474 /* A paradoxical subreg would be bad here: it'll be the right
5475 size, but later may be adjusted so that the upper bits aren't
5476 what we want. So reject it. */
5477 if (elt != 0
5478 && GET_CODE (elt->exp) == SUBREG
5479 && (GET_MODE_SIZE (GET_MODE (elt->exp))
5480 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))
5481 /* It is okay, though, if the rtx we're trying to match
5482 will ignore any of the bits we can't predict. */
5483 && ! (src != 0
5484 && GET_CODE (src) == SUBREG
5485 && GET_MODE (src) == GET_MODE (elt->exp)
5486 && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
5487 < GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))))
5489 elt = elt->next_same_value;
5490 continue;
5493 if (elt)
5495 src_elt_cost = elt->cost;
5496 src_elt_regcost = elt->regcost;
5499 /* Find cheapest and skip it for the next time. For items
5500 of equal cost, use this order:
5501 src_folded, src, src_eqv, src_related and hash table entry. */
5502 if (src_folded
5503 && preferable (src_folded_cost, src_folded_regcost,
5504 src_cost, src_regcost) <= 0
5505 && preferable (src_folded_cost, src_folded_regcost,
5506 src_eqv_cost, src_eqv_regcost) <= 0
5507 && preferable (src_folded_cost, src_folded_regcost,
5508 src_related_cost, src_related_regcost) <= 0
5509 && preferable (src_folded_cost, src_folded_regcost,
5510 src_elt_cost, src_elt_regcost) <= 0)
5512 trial = src_folded, src_folded_cost = MAX_COST;
5513 if (src_folded_force_flag)
5515 rtx forced = force_const_mem (mode, trial);
5516 if (forced)
5517 trial = forced;
5520 else if (src
5521 && preferable (src_cost, src_regcost,
5522 src_eqv_cost, src_eqv_regcost) <= 0
5523 && preferable (src_cost, src_regcost,
5524 src_related_cost, src_related_regcost) <= 0
5525 && preferable (src_cost, src_regcost,
5526 src_elt_cost, src_elt_regcost) <= 0)
5527 trial = src, src_cost = MAX_COST;
5528 else if (src_eqv_here
5529 && preferable (src_eqv_cost, src_eqv_regcost,
5530 src_related_cost, src_related_regcost) <= 0
5531 && preferable (src_eqv_cost, src_eqv_regcost,
5532 src_elt_cost, src_elt_regcost) <= 0)
5533 trial = copy_rtx (src_eqv_here), src_eqv_cost = MAX_COST;
5534 else if (src_related
5535 && preferable (src_related_cost, src_related_regcost,
5536 src_elt_cost, src_elt_regcost) <= 0)
5537 trial = copy_rtx (src_related), src_related_cost = MAX_COST;
5538 else
5540 trial = copy_rtx (elt->exp);
5541 elt = elt->next_same_value;
5542 src_elt_cost = MAX_COST;
5545 /* We don't normally have an insn matching (set (pc) (pc)), so
5546 check for this separately here. We will delete such an
5547 insn below.
5549 For other cases such as a table jump or conditional jump
5550 where we know the ultimate target, go ahead and replace the
5551 operand. While that may not make a valid insn, we will
5552 reemit the jump below (and also insert any necessary
5553 barriers). */
5554 if (n_sets == 1 && dest == pc_rtx
5555 && (trial == pc_rtx
5556 || (GET_CODE (trial) == LABEL_REF
5557 && ! condjump_p (insn))))
5559 /* Don't substitute non-local labels, this confuses CFG. */
5560 if (GET_CODE (trial) == LABEL_REF
5561 && LABEL_REF_NONLOCAL_P (trial))
5562 continue;
5564 SET_SRC (sets[i].rtl) = trial;
5565 cse_jumps_altered = 1;
5566 break;
5569 /* Reject certain invalid forms of CONST that we create. */
5570 else if (CONSTANT_P (trial)
5571 && GET_CODE (trial) == CONST
5572 /* Reject cases that will cause decode_rtx_const to
5573 die. On the alpha when simplifying a switch, we
5574 get (const (truncate (minus (label_ref)
5575 (label_ref)))). */
5576 && (GET_CODE (XEXP (trial, 0)) == TRUNCATE
5577 /* Likewise on IA-64, except without the
5578 truncate. */
5579 || (GET_CODE (XEXP (trial, 0)) == MINUS
5580 && GET_CODE (XEXP (XEXP (trial, 0), 0)) == LABEL_REF
5581 && GET_CODE (XEXP (XEXP (trial, 0), 1)) == LABEL_REF)))
5582 /* Do nothing for this case. */
5585 /* Look for a substitution that makes a valid insn. */
5586 else if (validate_change (insn, &SET_SRC (sets[i].rtl), trial, 0))
5588 rtx new = canon_reg (SET_SRC (sets[i].rtl), insn);
5590 /* If we just made a substitution inside a libcall, then we
5591 need to make the same substitution in any notes attached
5592 to the RETVAL insn. */
5593 if (libcall_insn
5594 && (REG_P (sets[i].orig_src)
5595 || GET_CODE (sets[i].orig_src) == SUBREG
5596 || MEM_P (sets[i].orig_src)))
5598 rtx note = find_reg_equal_equiv_note (libcall_insn);
5599 if (note != 0)
5600 XEXP (note, 0) = simplify_replace_rtx (XEXP (note, 0),
5601 sets[i].orig_src,
5602 copy_rtx (new));
5605 /* The result of apply_change_group can be ignored; see
5606 canon_reg. */
5608 validate_change (insn, &SET_SRC (sets[i].rtl), new, 1);
5609 apply_change_group ();
5610 break;
5613 /* If we previously found constant pool entries for
5614 constants and this is a constant, try making a
5615 pool entry. Put it in src_folded unless we already have done
5616 this since that is where it likely came from. */
5618 else if (constant_pool_entries_cost
5619 && CONSTANT_P (trial)
5620 && (src_folded == 0
5621 || (!MEM_P (src_folded)
5622 && ! src_folded_force_flag))
5623 && GET_MODE_CLASS (mode) != MODE_CC
5624 && mode != VOIDmode)
5626 src_folded_force_flag = 1;
5627 src_folded = trial;
5628 src_folded_cost = constant_pool_entries_cost;
5629 src_folded_regcost = constant_pool_entries_regcost;
5633 src = SET_SRC (sets[i].rtl);
5635 /* In general, it is good to have a SET with SET_SRC == SET_DEST.
5636 However, there is an important exception: If both are registers
5637 that are not the head of their equivalence class, replace SET_SRC
5638 with the head of the class. If we do not do this, we will have
5639 both registers live over a portion of the basic block. This way,
5640 their lifetimes will likely abut instead of overlapping. */
5641 if (REG_P (dest)
5642 && REGNO_QTY_VALID_P (REGNO (dest)))
5644 int dest_q = REG_QTY (REGNO (dest));
5645 struct qty_table_elem *dest_ent = &qty_table[dest_q];
5647 if (dest_ent->mode == GET_MODE (dest)
5648 && dest_ent->first_reg != REGNO (dest)
5649 && REG_P (src) && REGNO (src) == REGNO (dest)
5650 /* Don't do this if the original insn had a hard reg as
5651 SET_SRC or SET_DEST. */
5652 && (!REG_P (sets[i].src)
5653 || REGNO (sets[i].src) >= FIRST_PSEUDO_REGISTER)
5654 && (!REG_P (dest) || REGNO (dest) >= FIRST_PSEUDO_REGISTER))
5655 /* We can't call canon_reg here because it won't do anything if
5656 SRC is a hard register. */
5658 int src_q = REG_QTY (REGNO (src));
5659 struct qty_table_elem *src_ent = &qty_table[src_q];
5660 int first = src_ent->first_reg;
5661 rtx new_src
5662 = (first >= FIRST_PSEUDO_REGISTER
5663 ? regno_reg_rtx[first] : gen_rtx_REG (GET_MODE (src), first));
5665 /* We must use validate-change even for this, because this
5666 might be a special no-op instruction, suitable only to
5667 tag notes onto. */
5668 if (validate_change (insn, &SET_SRC (sets[i].rtl), new_src, 0))
5670 src = new_src;
5671 /* If we had a constant that is cheaper than what we are now
5672 setting SRC to, use that constant. We ignored it when we
5673 thought we could make this into a no-op. */
5674 if (src_const && COST (src_const) < COST (src)
5675 && validate_change (insn, &SET_SRC (sets[i].rtl),
5676 src_const, 0))
5677 src = src_const;
5682 /* If we made a change, recompute SRC values. */
5683 if (src != sets[i].src)
5685 cse_altered = 1;
5686 do_not_record = 0;
5687 hash_arg_in_memory = 0;
5688 sets[i].src = src;
5689 sets[i].src_hash = HASH (src, mode);
5690 sets[i].src_volatile = do_not_record;
5691 sets[i].src_in_memory = hash_arg_in_memory;
5692 sets[i].src_elt = lookup (src, sets[i].src_hash, mode);
5695 /* If this is a single SET, we are setting a register, and we have an
5696 equivalent constant, we want to add a REG_NOTE. We don't want
5697 to write a REG_EQUAL note for a constant pseudo since verifying that
5698 that pseudo hasn't been eliminated is a pain. Such a note also
5699 won't help anything.
5701 Avoid a REG_EQUAL note for (CONST (MINUS (LABEL_REF) (LABEL_REF)))
5702 which can be created for a reference to a compile time computable
5703 entry in a jump table. */
5705 if (n_sets == 1 && src_const && REG_P (dest)
5706 && !REG_P (src_const)
5707 && ! (GET_CODE (src_const) == CONST
5708 && GET_CODE (XEXP (src_const, 0)) == MINUS
5709 && GET_CODE (XEXP (XEXP (src_const, 0), 0)) == LABEL_REF
5710 && GET_CODE (XEXP (XEXP (src_const, 0), 1)) == LABEL_REF))
5712 /* We only want a REG_EQUAL note if src_const != src. */
5713 if (! rtx_equal_p (src, src_const))
5715 /* Make sure that the rtx is not shared. */
5716 src_const = copy_rtx (src_const);
5718 /* Record the actual constant value in a REG_EQUAL note,
5719 making a new one if one does not already exist. */
5720 set_unique_reg_note (insn, REG_EQUAL, src_const);
5724 /* Now deal with the destination. */
5725 do_not_record = 0;
5727 /* Look within any ZERO_EXTRACT to the MEM or REG within it. */
5728 while (GET_CODE (dest) == SUBREG
5729 || GET_CODE (dest) == ZERO_EXTRACT
5730 || GET_CODE (dest) == STRICT_LOW_PART)
5731 dest = XEXP (dest, 0);
5733 sets[i].inner_dest = dest;
5735 if (MEM_P (dest))
5737 #ifdef PUSH_ROUNDING
5738 /* Stack pushes invalidate the stack pointer. */
5739 rtx addr = XEXP (dest, 0);
5740 if (GET_RTX_CLASS (GET_CODE (addr)) == RTX_AUTOINC
5741 && XEXP (addr, 0) == stack_pointer_rtx)
5742 invalidate (stack_pointer_rtx, Pmode);
5743 #endif
5744 dest = fold_rtx (dest, insn);
5747 /* Compute the hash code of the destination now,
5748 before the effects of this instruction are recorded,
5749 since the register values used in the address computation
5750 are those before this instruction. */
5751 sets[i].dest_hash = HASH (dest, mode);
5753 /* Don't enter a bit-field in the hash table
5754 because the value in it after the store
5755 may not equal what was stored, due to truncation. */
5757 if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT)
5759 rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
5761 if (src_const != 0 && GET_CODE (src_const) == CONST_INT
5762 && GET_CODE (width) == CONST_INT
5763 && INTVAL (width) < HOST_BITS_PER_WIDE_INT
5764 && ! (INTVAL (src_const)
5765 & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
5766 /* Exception: if the value is constant,
5767 and it won't be truncated, record it. */
5769 else
5771 /* This is chosen so that the destination will be invalidated
5772 but no new value will be recorded.
5773 We must invalidate because sometimes constant
5774 values can be recorded for bitfields. */
5775 sets[i].src_elt = 0;
5776 sets[i].src_volatile = 1;
5777 src_eqv = 0;
5778 src_eqv_elt = 0;
5782 /* If only one set in a JUMP_INSN and it is now a no-op, we can delete
5783 the insn. */
5784 else if (n_sets == 1 && dest == pc_rtx && src == pc_rtx)
5786 /* One less use of the label this insn used to jump to. */
5787 delete_insn (insn);
5788 cse_jumps_altered = 1;
5789 /* No more processing for this set. */
5790 sets[i].rtl = 0;
5793 /* If this SET is now setting PC to a label, we know it used to
5794 be a conditional or computed branch. */
5795 else if (dest == pc_rtx && GET_CODE (src) == LABEL_REF
5796 && !LABEL_REF_NONLOCAL_P (src))
5798 /* Now emit a BARRIER after the unconditional jump. */
5799 if (NEXT_INSN (insn) == 0
5800 || !BARRIER_P (NEXT_INSN (insn)))
5801 emit_barrier_after (insn);
5803 /* We reemit the jump in as many cases as possible just in
5804 case the form of an unconditional jump is significantly
5805 different than a computed jump or conditional jump.
5807 If this insn has multiple sets, then reemitting the
5808 jump is nontrivial. So instead we just force rerecognition
5809 and hope for the best. */
5810 if (n_sets == 1)
5812 rtx new, note;
5814 new = emit_jump_insn_after (gen_jump (XEXP (src, 0)), insn);
5815 JUMP_LABEL (new) = XEXP (src, 0);
5816 LABEL_NUSES (XEXP (src, 0))++;
5818 /* Make sure to copy over REG_NON_LOCAL_GOTO. */
5819 note = find_reg_note (insn, REG_NON_LOCAL_GOTO, 0);
5820 if (note)
5822 XEXP (note, 1) = NULL_RTX;
5823 REG_NOTES (new) = note;
5826 delete_insn (insn);
5827 insn = new;
5829 /* Now emit a BARRIER after the unconditional jump. */
5830 if (NEXT_INSN (insn) == 0
5831 || !BARRIER_P (NEXT_INSN (insn)))
5832 emit_barrier_after (insn);
5834 else
5835 INSN_CODE (insn) = -1;
5837 /* Do not bother deleting any unreachable code,
5838 let jump/flow do that. */
5840 cse_jumps_altered = 1;
5841 sets[i].rtl = 0;
5844 /* If destination is volatile, invalidate it and then do no further
5845 processing for this assignment. */
5847 else if (do_not_record)
5849 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
5850 invalidate (dest, VOIDmode);
5851 else if (MEM_P (dest))
5852 invalidate (dest, VOIDmode);
5853 else if (GET_CODE (dest) == STRICT_LOW_PART
5854 || GET_CODE (dest) == ZERO_EXTRACT)
5855 invalidate (XEXP (dest, 0), GET_MODE (dest));
5856 sets[i].rtl = 0;
5859 if (sets[i].rtl != 0 && dest != SET_DEST (sets[i].rtl))
5860 sets[i].dest_hash = HASH (SET_DEST (sets[i].rtl), mode);
5862 #ifdef HAVE_cc0
5863 /* If setting CC0, record what it was set to, or a constant, if it
5864 is equivalent to a constant. If it is being set to a floating-point
5865 value, make a COMPARE with the appropriate constant of 0. If we
5866 don't do this, later code can interpret this as a test against
5867 const0_rtx, which can cause problems if we try to put it into an
5868 insn as a floating-point operand. */
5869 if (dest == cc0_rtx)
5871 this_insn_cc0 = src_const && mode != VOIDmode ? src_const : src;
5872 this_insn_cc0_mode = mode;
5873 if (FLOAT_MODE_P (mode))
5874 this_insn_cc0 = gen_rtx_COMPARE (VOIDmode, this_insn_cc0,
5875 CONST0_RTX (mode));
5877 #endif
5880 /* Now enter all non-volatile source expressions in the hash table
5881 if they are not already present.
5882 Record their equivalence classes in src_elt.
5883 This way we can insert the corresponding destinations into
5884 the same classes even if the actual sources are no longer in them
5885 (having been invalidated). */
5887 if (src_eqv && src_eqv_elt == 0 && sets[0].rtl != 0 && ! src_eqv_volatile
5888 && ! rtx_equal_p (src_eqv, SET_DEST (sets[0].rtl)))
5890 struct table_elt *elt;
5891 struct table_elt *classp = sets[0].src_elt;
5892 rtx dest = SET_DEST (sets[0].rtl);
5893 enum machine_mode eqvmode = GET_MODE (dest);
5895 if (GET_CODE (dest) == STRICT_LOW_PART)
5897 eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
5898 classp = 0;
5900 if (insert_regs (src_eqv, classp, 0))
5902 rehash_using_reg (src_eqv);
5903 src_eqv_hash = HASH (src_eqv, eqvmode);
5905 elt = insert (src_eqv, classp, src_eqv_hash, eqvmode);
5906 elt->in_memory = src_eqv_in_memory;
5907 src_eqv_elt = elt;
5909 /* Check to see if src_eqv_elt is the same as a set source which
5910 does not yet have an elt, and if so set the elt of the set source
5911 to src_eqv_elt. */
5912 for (i = 0; i < n_sets; i++)
5913 if (sets[i].rtl && sets[i].src_elt == 0
5914 && rtx_equal_p (SET_SRC (sets[i].rtl), src_eqv))
5915 sets[i].src_elt = src_eqv_elt;
5918 for (i = 0; i < n_sets; i++)
5919 if (sets[i].rtl && ! sets[i].src_volatile
5920 && ! rtx_equal_p (SET_SRC (sets[i].rtl), SET_DEST (sets[i].rtl)))
5922 if (GET_CODE (SET_DEST (sets[i].rtl)) == STRICT_LOW_PART)
5924 /* REG_EQUAL in setting a STRICT_LOW_PART
5925 gives an equivalent for the entire destination register,
5926 not just for the subreg being stored in now.
5927 This is a more interesting equivalence, so we arrange later
5928 to treat the entire reg as the destination. */
5929 sets[i].src_elt = src_eqv_elt;
5930 sets[i].src_hash = src_eqv_hash;
5932 else
5934 /* Insert source and constant equivalent into hash table, if not
5935 already present. */
5936 struct table_elt *classp = src_eqv_elt;
5937 rtx src = sets[i].src;
5938 rtx dest = SET_DEST (sets[i].rtl);
5939 enum machine_mode mode
5940 = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
5942 /* It's possible that we have a source value known to be
5943 constant but don't have a REG_EQUAL note on the insn.
5944 Lack of a note will mean src_eqv_elt will be NULL. This
5945 can happen where we've generated a SUBREG to access a
5946 CONST_INT that is already in a register in a wider mode.
5947 Ensure that the source expression is put in the proper
5948 constant class. */
5949 if (!classp)
5950 classp = sets[i].src_const_elt;
5952 if (sets[i].src_elt == 0)
5954 /* Don't put a hard register source into the table if this is
5955 the last insn of a libcall. In this case, we only need
5956 to put src_eqv_elt in src_elt. */
5957 if (! find_reg_note (insn, REG_RETVAL, NULL_RTX))
5959 struct table_elt *elt;
5961 /* Note that these insert_regs calls cannot remove
5962 any of the src_elt's, because they would have failed to
5963 match if not still valid. */
5964 if (insert_regs (src, classp, 0))
5966 rehash_using_reg (src);
5967 sets[i].src_hash = HASH (src, mode);
5969 elt = insert (src, classp, sets[i].src_hash, mode);
5970 elt->in_memory = sets[i].src_in_memory;
5971 sets[i].src_elt = classp = elt;
5973 else
5974 sets[i].src_elt = classp;
5976 if (sets[i].src_const && sets[i].src_const_elt == 0
5977 && src != sets[i].src_const
5978 && ! rtx_equal_p (sets[i].src_const, src))
5979 sets[i].src_elt = insert (sets[i].src_const, classp,
5980 sets[i].src_const_hash, mode);
5983 else if (sets[i].src_elt == 0)
5984 /* If we did not insert the source into the hash table (e.g., it was
5985 volatile), note the equivalence class for the REG_EQUAL value, if any,
5986 so that the destination goes into that class. */
5987 sets[i].src_elt = src_eqv_elt;
5989 invalidate_from_clobbers (x);
5991 /* Some registers are invalidated by subroutine calls. Memory is
5992 invalidated by non-constant calls. */
5994 if (CALL_P (insn))
5996 if (! CONST_OR_PURE_CALL_P (insn))
5997 invalidate_memory ();
5998 invalidate_for_call ();
6001 /* Now invalidate everything set by this instruction.
6002 If a SUBREG or other funny destination is being set,
6003 sets[i].rtl is still nonzero, so here we invalidate the reg
6004 a part of which is being set. */
6006 for (i = 0; i < n_sets; i++)
6007 if (sets[i].rtl)
6009 /* We can't use the inner dest, because the mode associated with
6010 a ZERO_EXTRACT is significant. */
6011 rtx dest = SET_DEST (sets[i].rtl);
6013 /* Needed for registers to remove the register from its
6014 previous quantity's chain.
6015 Needed for memory if this is a nonvarying address, unless
6016 we have just done an invalidate_memory that covers even those. */
6017 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
6018 invalidate (dest, VOIDmode);
6019 else if (MEM_P (dest))
6020 invalidate (dest, VOIDmode);
6021 else if (GET_CODE (dest) == STRICT_LOW_PART
6022 || GET_CODE (dest) == ZERO_EXTRACT)
6023 invalidate (XEXP (dest, 0), GET_MODE (dest));
6026 /* A volatile ASM invalidates everything. */
6027 if (NONJUMP_INSN_P (insn)
6028 && GET_CODE (PATTERN (insn)) == ASM_OPERANDS
6029 && MEM_VOLATILE_P (PATTERN (insn)))
6030 flush_hash_table ();
6032 /* Make sure registers mentioned in destinations
6033 are safe for use in an expression to be inserted.
6034 This removes from the hash table
6035 any invalid entry that refers to one of these registers.
6037 We don't care about the return value from mention_regs because
6038 we are going to hash the SET_DEST values unconditionally. */
6040 for (i = 0; i < n_sets; i++)
6042 if (sets[i].rtl)
6044 rtx x = SET_DEST (sets[i].rtl);
6046 if (!REG_P (x))
6047 mention_regs (x);
6048 else
6050 /* We used to rely on all references to a register becoming
6051 inaccessible when a register changes to a new quantity,
6052 since that changes the hash code. However, that is not
6053 safe, since after HASH_SIZE new quantities we get a
6054 hash 'collision' of a register with its own invalid
6055 entries. And since SUBREGs have been changed not to
6056 change their hash code with the hash code of the register,
6057 it wouldn't work any longer at all. So we have to check
6058 for any invalid references lying around now.
6059 This code is similar to the REG case in mention_regs,
6060 but it knows that reg_tick has been incremented, and
6061 it leaves reg_in_table as -1 . */
6062 unsigned int regno = REGNO (x);
6063 unsigned int endregno
6064 = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
6065 : hard_regno_nregs[regno][GET_MODE (x)]);
6066 unsigned int i;
6068 for (i = regno; i < endregno; i++)
6070 if (REG_IN_TABLE (i) >= 0)
6072 remove_invalid_refs (i);
6073 REG_IN_TABLE (i) = -1;
6080 /* We may have just removed some of the src_elt's from the hash table.
6081 So replace each one with the current head of the same class. */
6083 for (i = 0; i < n_sets; i++)
6084 if (sets[i].rtl)
6086 if (sets[i].src_elt && sets[i].src_elt->first_same_value == 0)
6087 /* If elt was removed, find current head of same class,
6088 or 0 if nothing remains of that class. */
6090 struct table_elt *elt = sets[i].src_elt;
6092 while (elt && elt->prev_same_value)
6093 elt = elt->prev_same_value;
6095 while (elt && elt->first_same_value == 0)
6096 elt = elt->next_same_value;
6097 sets[i].src_elt = elt ? elt->first_same_value : 0;
6101 /* Now insert the destinations into their equivalence classes. */
6103 for (i = 0; i < n_sets; i++)
6104 if (sets[i].rtl)
6106 rtx dest = SET_DEST (sets[i].rtl);
6107 struct table_elt *elt;
6109 /* Don't record value if we are not supposed to risk allocating
6110 floating-point values in registers that might be wider than
6111 memory. */
6112 if ((flag_float_store
6113 && MEM_P (dest)
6114 && FLOAT_MODE_P (GET_MODE (dest)))
6115 /* Don't record BLKmode values, because we don't know the
6116 size of it, and can't be sure that other BLKmode values
6117 have the same or smaller size. */
6118 || GET_MODE (dest) == BLKmode
6119 /* Don't record values of destinations set inside a libcall block
6120 since we might delete the libcall. Things should have been set
6121 up so we won't want to reuse such a value, but we play it safe
6122 here. */
6123 || libcall_insn
6124 /* If we didn't put a REG_EQUAL value or a source into the hash
6125 table, there is no point is recording DEST. */
6126 || sets[i].src_elt == 0
6127 /* If DEST is a paradoxical SUBREG and SRC is a ZERO_EXTEND
6128 or SIGN_EXTEND, don't record DEST since it can cause
6129 some tracking to be wrong.
6131 ??? Think about this more later. */
6132 || (GET_CODE (dest) == SUBREG
6133 && (GET_MODE_SIZE (GET_MODE (dest))
6134 > GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
6135 && (GET_CODE (sets[i].src) == SIGN_EXTEND
6136 || GET_CODE (sets[i].src) == ZERO_EXTEND)))
6137 continue;
6139 /* STRICT_LOW_PART isn't part of the value BEING set,
6140 and neither is the SUBREG inside it.
6141 Note that in this case SETS[I].SRC_ELT is really SRC_EQV_ELT. */
6142 if (GET_CODE (dest) == STRICT_LOW_PART)
6143 dest = SUBREG_REG (XEXP (dest, 0));
6145 if (REG_P (dest) || GET_CODE (dest) == SUBREG)
6146 /* Registers must also be inserted into chains for quantities. */
6147 if (insert_regs (dest, sets[i].src_elt, 1))
6149 /* If `insert_regs' changes something, the hash code must be
6150 recalculated. */
6151 rehash_using_reg (dest);
6152 sets[i].dest_hash = HASH (dest, GET_MODE (dest));
6155 elt = insert (dest, sets[i].src_elt,
6156 sets[i].dest_hash, GET_MODE (dest));
6158 elt->in_memory = (MEM_P (sets[i].inner_dest)
6159 && !MEM_READONLY_P (sets[i].inner_dest));
6161 /* If we have (set (subreg:m1 (reg:m2 foo) 0) (bar:m1)), M1 is no
6162 narrower than M2, and both M1 and M2 are the same number of words,
6163 we are also doing (set (reg:m2 foo) (subreg:m2 (bar:m1) 0)) so
6164 make that equivalence as well.
6166 However, BAR may have equivalences for which gen_lowpart
6167 will produce a simpler value than gen_lowpart applied to
6168 BAR (e.g., if BAR was ZERO_EXTENDed from M2), so we will scan all
6169 BAR's equivalences. If we don't get a simplified form, make
6170 the SUBREG. It will not be used in an equivalence, but will
6171 cause two similar assignments to be detected.
6173 Note the loop below will find SUBREG_REG (DEST) since we have
6174 already entered SRC and DEST of the SET in the table. */
6176 if (GET_CODE (dest) == SUBREG
6177 && (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))) - 1)
6178 / UNITS_PER_WORD)
6179 == (GET_MODE_SIZE (GET_MODE (dest)) - 1) / UNITS_PER_WORD)
6180 && (GET_MODE_SIZE (GET_MODE (dest))
6181 >= GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
6182 && sets[i].src_elt != 0)
6184 enum machine_mode new_mode = GET_MODE (SUBREG_REG (dest));
6185 struct table_elt *elt, *classp = 0;
6187 for (elt = sets[i].src_elt->first_same_value; elt;
6188 elt = elt->next_same_value)
6190 rtx new_src = 0;
6191 unsigned src_hash;
6192 struct table_elt *src_elt;
6193 int byte = 0;
6195 /* Ignore invalid entries. */
6196 if (!REG_P (elt->exp)
6197 && ! exp_equiv_p (elt->exp, elt->exp, 1, false))
6198 continue;
6200 /* We may have already been playing subreg games. If the
6201 mode is already correct for the destination, use it. */
6202 if (GET_MODE (elt->exp) == new_mode)
6203 new_src = elt->exp;
6204 else
6206 /* Calculate big endian correction for the SUBREG_BYTE.
6207 We have already checked that M1 (GET_MODE (dest))
6208 is not narrower than M2 (new_mode). */
6209 if (BYTES_BIG_ENDIAN)
6210 byte = (GET_MODE_SIZE (GET_MODE (dest))
6211 - GET_MODE_SIZE (new_mode));
6213 new_src = simplify_gen_subreg (new_mode, elt->exp,
6214 GET_MODE (dest), byte);
6217 /* The call to simplify_gen_subreg fails if the value
6218 is VOIDmode, yet we can't do any simplification, e.g.
6219 for EXPR_LISTs denoting function call results.
6220 It is invalid to construct a SUBREG with a VOIDmode
6221 SUBREG_REG, hence a zero new_src means we can't do
6222 this substitution. */
6223 if (! new_src)
6224 continue;
6226 src_hash = HASH (new_src, new_mode);
6227 src_elt = lookup (new_src, src_hash, new_mode);
6229 /* Put the new source in the hash table is if isn't
6230 already. */
6231 if (src_elt == 0)
6233 if (insert_regs (new_src, classp, 0))
6235 rehash_using_reg (new_src);
6236 src_hash = HASH (new_src, new_mode);
6238 src_elt = insert (new_src, classp, src_hash, new_mode);
6239 src_elt->in_memory = elt->in_memory;
6241 else if (classp && classp != src_elt->first_same_value)
6242 /* Show that two things that we've seen before are
6243 actually the same. */
6244 merge_equiv_classes (src_elt, classp);
6246 classp = src_elt->first_same_value;
6247 /* Ignore invalid entries. */
6248 while (classp
6249 && !REG_P (classp->exp)
6250 && ! exp_equiv_p (classp->exp, classp->exp, 1, false))
6251 classp = classp->next_same_value;
6256 /* Special handling for (set REG0 REG1) where REG0 is the
6257 "cheapest", cheaper than REG1. After cse, REG1 will probably not
6258 be used in the sequel, so (if easily done) change this insn to
6259 (set REG1 REG0) and replace REG1 with REG0 in the previous insn
6260 that computed their value. Then REG1 will become a dead store
6261 and won't cloud the situation for later optimizations.
6263 Do not make this change if REG1 is a hard register, because it will
6264 then be used in the sequel and we may be changing a two-operand insn
6265 into a three-operand insn.
6267 Also do not do this if we are operating on a copy of INSN.
6269 Also don't do this if INSN ends a libcall; this would cause an unrelated
6270 register to be set in the middle of a libcall, and we then get bad code
6271 if the libcall is deleted. */
6273 if (n_sets == 1 && sets[0].rtl && REG_P (SET_DEST (sets[0].rtl))
6274 && NEXT_INSN (PREV_INSN (insn)) == insn
6275 && REG_P (SET_SRC (sets[0].rtl))
6276 && REGNO (SET_SRC (sets[0].rtl)) >= FIRST_PSEUDO_REGISTER
6277 && REGNO_QTY_VALID_P (REGNO (SET_SRC (sets[0].rtl))))
6279 int src_q = REG_QTY (REGNO (SET_SRC (sets[0].rtl)));
6280 struct qty_table_elem *src_ent = &qty_table[src_q];
6282 if ((src_ent->first_reg == REGNO (SET_DEST (sets[0].rtl)))
6283 && ! find_reg_note (insn, REG_RETVAL, NULL_RTX))
6285 rtx prev = insn;
6286 /* Scan for the previous nonnote insn, but stop at a basic
6287 block boundary. */
6290 prev = PREV_INSN (prev);
6292 while (prev && NOTE_P (prev)
6293 && NOTE_LINE_NUMBER (prev) != NOTE_INSN_BASIC_BLOCK);
6295 /* Do not swap the registers around if the previous instruction
6296 attaches a REG_EQUIV note to REG1.
6298 ??? It's not entirely clear whether we can transfer a REG_EQUIV
6299 from the pseudo that originally shadowed an incoming argument
6300 to another register. Some uses of REG_EQUIV might rely on it
6301 being attached to REG1 rather than REG2.
6303 This section previously turned the REG_EQUIV into a REG_EQUAL
6304 note. We cannot do that because REG_EQUIV may provide an
6305 uninitialized stack slot when REG_PARM_STACK_SPACE is used. */
6307 if (prev != 0 && NONJUMP_INSN_P (prev)
6308 && GET_CODE (PATTERN (prev)) == SET
6309 && SET_DEST (PATTERN (prev)) == SET_SRC (sets[0].rtl)
6310 && ! find_reg_note (prev, REG_EQUIV, NULL_RTX))
6312 rtx dest = SET_DEST (sets[0].rtl);
6313 rtx src = SET_SRC (sets[0].rtl);
6314 rtx note;
6316 validate_change (prev, &SET_DEST (PATTERN (prev)), dest, 1);
6317 validate_change (insn, &SET_DEST (sets[0].rtl), src, 1);
6318 validate_change (insn, &SET_SRC (sets[0].rtl), dest, 1);
6319 apply_change_group ();
6321 /* If INSN has a REG_EQUAL note, and this note mentions
6322 REG0, then we must delete it, because the value in
6323 REG0 has changed. If the note's value is REG1, we must
6324 also delete it because that is now this insn's dest. */
6325 note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
6326 if (note != 0
6327 && (reg_mentioned_p (dest, XEXP (note, 0))
6328 || rtx_equal_p (src, XEXP (note, 0))))
6329 remove_note (insn, note);
6334 /* If this is a conditional jump insn, record any known equivalences due to
6335 the condition being tested. */
6337 if (JUMP_P (insn)
6338 && n_sets == 1 && GET_CODE (x) == SET
6339 && GET_CODE (SET_SRC (x)) == IF_THEN_ELSE)
6340 record_jump_equiv (insn, 0);
6342 #ifdef HAVE_cc0
6343 /* If the previous insn set CC0 and this insn no longer references CC0,
6344 delete the previous insn. Here we use the fact that nothing expects CC0
6345 to be valid over an insn, which is true until the final pass. */
6346 if (prev_insn && NONJUMP_INSN_P (prev_insn)
6347 && (tem = single_set (prev_insn)) != 0
6348 && SET_DEST (tem) == cc0_rtx
6349 && ! reg_mentioned_p (cc0_rtx, x))
6350 delete_insn (prev_insn);
6352 prev_insn_cc0 = this_insn_cc0;
6353 prev_insn_cc0_mode = this_insn_cc0_mode;
6354 prev_insn = insn;
6355 #endif
6358 /* Remove from the hash table all expressions that reference memory. */
6360 static void
6361 invalidate_memory (void)
6363 int i;
6364 struct table_elt *p, *next;
6366 for (i = 0; i < HASH_SIZE; i++)
6367 for (p = table[i]; p; p = next)
6369 next = p->next_same_hash;
6370 if (p->in_memory)
6371 remove_from_table (p, i);
6375 /* If ADDR is an address that implicitly affects the stack pointer, return
6376 1 and update the register tables to show the effect. Else, return 0. */
6378 static int
6379 addr_affects_sp_p (rtx addr)
6381 if (GET_RTX_CLASS (GET_CODE (addr)) == RTX_AUTOINC
6382 && REG_P (XEXP (addr, 0))
6383 && REGNO (XEXP (addr, 0)) == STACK_POINTER_REGNUM)
6385 if (REG_TICK (STACK_POINTER_REGNUM) >= 0)
6387 REG_TICK (STACK_POINTER_REGNUM)++;
6388 /* Is it possible to use a subreg of SP? */
6389 SUBREG_TICKED (STACK_POINTER_REGNUM) = -1;
6392 /* This should be *very* rare. */
6393 if (TEST_HARD_REG_BIT (hard_regs_in_table, STACK_POINTER_REGNUM))
6394 invalidate (stack_pointer_rtx, VOIDmode);
6396 return 1;
6399 return 0;
6402 /* Perform invalidation on the basis of everything about an insn
6403 except for invalidating the actual places that are SET in it.
6404 This includes the places CLOBBERed, and anything that might
6405 alias with something that is SET or CLOBBERed.
6407 X is the pattern of the insn. */
6409 static void
6410 invalidate_from_clobbers (rtx x)
6412 if (GET_CODE (x) == CLOBBER)
6414 rtx ref = XEXP (x, 0);
6415 if (ref)
6417 if (REG_P (ref) || GET_CODE (ref) == SUBREG
6418 || MEM_P (ref))
6419 invalidate (ref, VOIDmode);
6420 else if (GET_CODE (ref) == STRICT_LOW_PART
6421 || GET_CODE (ref) == ZERO_EXTRACT)
6422 invalidate (XEXP (ref, 0), GET_MODE (ref));
6425 else if (GET_CODE (x) == PARALLEL)
6427 int i;
6428 for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
6430 rtx y = XVECEXP (x, 0, i);
6431 if (GET_CODE (y) == CLOBBER)
6433 rtx ref = XEXP (y, 0);
6434 if (REG_P (ref) || GET_CODE (ref) == SUBREG
6435 || MEM_P (ref))
6436 invalidate (ref, VOIDmode);
6437 else if (GET_CODE (ref) == STRICT_LOW_PART
6438 || GET_CODE (ref) == ZERO_EXTRACT)
6439 invalidate (XEXP (ref, 0), GET_MODE (ref));
6445 /* Process X, part of the REG_NOTES of an insn. Look at any REG_EQUAL notes
6446 and replace any registers in them with either an equivalent constant
6447 or the canonical form of the register. If we are inside an address,
6448 only do this if the address remains valid.
6450 OBJECT is 0 except when within a MEM in which case it is the MEM.
6452 Return the replacement for X. */
6454 static rtx
6455 cse_process_notes (rtx x, rtx object)
6457 enum rtx_code code = GET_CODE (x);
6458 const char *fmt = GET_RTX_FORMAT (code);
6459 int i;
6461 switch (code)
6463 case CONST_INT:
6464 case CONST:
6465 case SYMBOL_REF:
6466 case LABEL_REF:
6467 case CONST_DOUBLE:
6468 case CONST_VECTOR:
6469 case PC:
6470 case CC0:
6471 case LO_SUM:
6472 return x;
6474 case MEM:
6475 validate_change (x, &XEXP (x, 0),
6476 cse_process_notes (XEXP (x, 0), x), 0);
6477 return x;
6479 case EXPR_LIST:
6480 case INSN_LIST:
6481 if (REG_NOTE_KIND (x) == REG_EQUAL)
6482 XEXP (x, 0) = cse_process_notes (XEXP (x, 0), NULL_RTX);
6483 if (XEXP (x, 1))
6484 XEXP (x, 1) = cse_process_notes (XEXP (x, 1), NULL_RTX);
6485 return x;
6487 case SIGN_EXTEND:
6488 case ZERO_EXTEND:
6489 case SUBREG:
6491 rtx new = cse_process_notes (XEXP (x, 0), object);
6492 /* We don't substitute VOIDmode constants into these rtx,
6493 since they would impede folding. */
6494 if (GET_MODE (new) != VOIDmode)
6495 validate_change (object, &XEXP (x, 0), new, 0);
6496 return x;
6499 case REG:
6500 i = REG_QTY (REGNO (x));
6502 /* Return a constant or a constant register. */
6503 if (REGNO_QTY_VALID_P (REGNO (x)))
6505 struct qty_table_elem *ent = &qty_table[i];
6507 if (ent->const_rtx != NULL_RTX
6508 && (CONSTANT_P (ent->const_rtx)
6509 || REG_P (ent->const_rtx)))
6511 rtx new = gen_lowpart (GET_MODE (x), ent->const_rtx);
6512 if (new)
6513 return new;
6517 /* Otherwise, canonicalize this register. */
6518 return canon_reg (x, NULL_RTX);
6520 default:
6521 break;
6524 for (i = 0; i < GET_RTX_LENGTH (code); i++)
6525 if (fmt[i] == 'e')
6526 validate_change (object, &XEXP (x, i),
6527 cse_process_notes (XEXP (x, i), object), 0);
6529 return x;
6532 /* Process one SET of an insn that was skipped. We ignore CLOBBERs
6533 since they are done elsewhere. This function is called via note_stores. */
6535 static void
6536 invalidate_skipped_set (rtx dest, rtx set, void *data ATTRIBUTE_UNUSED)
6538 enum rtx_code code = GET_CODE (dest);
6540 if (code == MEM
6541 && ! addr_affects_sp_p (dest) /* If this is not a stack push ... */
6542 /* There are times when an address can appear varying and be a PLUS
6543 during this scan when it would be a fixed address were we to know
6544 the proper equivalences. So invalidate all memory if there is
6545 a BLKmode or nonscalar memory reference or a reference to a
6546 variable address. */
6547 && (MEM_IN_STRUCT_P (dest) || GET_MODE (dest) == BLKmode
6548 || cse_rtx_varies_p (XEXP (dest, 0), 0)))
6550 invalidate_memory ();
6551 return;
6554 if (GET_CODE (set) == CLOBBER
6555 || CC0_P (dest)
6556 || dest == pc_rtx)
6557 return;
6559 if (code == STRICT_LOW_PART || code == ZERO_EXTRACT)
6560 invalidate (XEXP (dest, 0), GET_MODE (dest));
6561 else if (code == REG || code == SUBREG || code == MEM)
6562 invalidate (dest, VOIDmode);
6565 /* Invalidate all insns from START up to the end of the function or the
6566 next label. This called when we wish to CSE around a block that is
6567 conditionally executed. */
6569 static void
6570 invalidate_skipped_block (rtx start)
6572 rtx insn;
6574 for (insn = start; insn && !LABEL_P (insn);
6575 insn = NEXT_INSN (insn))
6577 if (! INSN_P (insn))
6578 continue;
6580 if (CALL_P (insn))
6582 if (! CONST_OR_PURE_CALL_P (insn))
6583 invalidate_memory ();
6584 invalidate_for_call ();
6587 invalidate_from_clobbers (PATTERN (insn));
6588 note_stores (PATTERN (insn), invalidate_skipped_set, NULL);
6592 /* Find the end of INSN's basic block and return its range,
6593 the total number of SETs in all the insns of the block, the last insn of the
6594 block, and the branch path.
6596 The branch path indicates which branches should be followed. If a nonzero
6597 path size is specified, the block should be rescanned and a different set
6598 of branches will be taken. The branch path is only used if
6599 FLAG_CSE_FOLLOW_JUMPS or FLAG_CSE_SKIP_BLOCKS is nonzero.
6601 DATA is a pointer to a struct cse_basic_block_data, defined below, that is
6602 used to describe the block. It is filled in with the information about
6603 the current block. The incoming structure's branch path, if any, is used
6604 to construct the output branch path. */
6606 static void
6607 cse_end_of_basic_block (rtx insn, struct cse_basic_block_data *data,
6608 int follow_jumps, int skip_blocks)
6610 rtx p = insn, q;
6611 int nsets = 0;
6612 int low_cuid = INSN_CUID (insn), high_cuid = INSN_CUID (insn);
6613 rtx next = INSN_P (insn) ? insn : next_real_insn (insn);
6614 int path_size = data->path_size;
6615 int path_entry = 0;
6616 int i;
6618 /* Update the previous branch path, if any. If the last branch was
6619 previously PATH_TAKEN, mark it PATH_NOT_TAKEN.
6620 If it was previously PATH_NOT_TAKEN,
6621 shorten the path by one and look at the previous branch. We know that
6622 at least one branch must have been taken if PATH_SIZE is nonzero. */
6623 while (path_size > 0)
6625 if (data->path[path_size - 1].status != PATH_NOT_TAKEN)
6627 data->path[path_size - 1].status = PATH_NOT_TAKEN;
6628 break;
6630 else
6631 path_size--;
6634 /* If the first instruction is marked with QImode, that means we've
6635 already processed this block. Our caller will look at DATA->LAST
6636 to figure out where to go next. We want to return the next block
6637 in the instruction stream, not some branched-to block somewhere
6638 else. We accomplish this by pretending our called forbid us to
6639 follow jumps, or skip blocks. */
6640 if (GET_MODE (insn) == QImode)
6641 follow_jumps = skip_blocks = 0;
6643 /* Scan to end of this basic block. */
6644 while (p && !LABEL_P (p))
6646 /* Don't cse over a call to setjmp; on some machines (eg VAX)
6647 the regs restored by the longjmp come from
6648 a later time than the setjmp. */
6649 if (PREV_INSN (p) && CALL_P (PREV_INSN (p))
6650 && find_reg_note (PREV_INSN (p), REG_SETJMP, NULL))
6651 break;
6653 /* A PARALLEL can have lots of SETs in it,
6654 especially if it is really an ASM_OPERANDS. */
6655 if (INSN_P (p) && GET_CODE (PATTERN (p)) == PARALLEL)
6656 nsets += XVECLEN (PATTERN (p), 0);
6657 else if (!NOTE_P (p))
6658 nsets += 1;
6660 /* Ignore insns made by CSE; they cannot affect the boundaries of
6661 the basic block. */
6663 if (INSN_UID (p) <= max_uid && INSN_CUID (p) > high_cuid)
6664 high_cuid = INSN_CUID (p);
6665 if (INSN_UID (p) <= max_uid && INSN_CUID (p) < low_cuid)
6666 low_cuid = INSN_CUID (p);
6668 /* See if this insn is in our branch path. If it is and we are to
6669 take it, do so. */
6670 if (path_entry < path_size && data->path[path_entry].branch == p)
6672 if (data->path[path_entry].status != PATH_NOT_TAKEN)
6673 p = JUMP_LABEL (p);
6675 /* Point to next entry in path, if any. */
6676 path_entry++;
6679 /* If this is a conditional jump, we can follow it if -fcse-follow-jumps
6680 was specified, we haven't reached our maximum path length, there are
6681 insns following the target of the jump, this is the only use of the
6682 jump label, and the target label is preceded by a BARRIER.
6684 Alternatively, we can follow the jump if it branches around a
6685 block of code and there are no other branches into the block.
6686 In this case invalidate_skipped_block will be called to invalidate any
6687 registers set in the block when following the jump. */
6689 else if ((follow_jumps || skip_blocks) && path_size < PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH) - 1
6690 && JUMP_P (p)
6691 && GET_CODE (PATTERN (p)) == SET
6692 && GET_CODE (SET_SRC (PATTERN (p))) == IF_THEN_ELSE
6693 && JUMP_LABEL (p) != 0
6694 && LABEL_NUSES (JUMP_LABEL (p)) == 1
6695 && NEXT_INSN (JUMP_LABEL (p)) != 0)
6697 for (q = PREV_INSN (JUMP_LABEL (p)); q; q = PREV_INSN (q))
6698 if ((!NOTE_P (q)
6699 || NOTE_LINE_NUMBER (q) == NOTE_INSN_LOOP_END
6700 || (PREV_INSN (q) && CALL_P (PREV_INSN (q))
6701 && find_reg_note (PREV_INSN (q), REG_SETJMP, NULL)))
6702 && (!LABEL_P (q) || LABEL_NUSES (q) != 0))
6703 break;
6705 /* If we ran into a BARRIER, this code is an extension of the
6706 basic block when the branch is taken. */
6707 if (follow_jumps && q != 0 && BARRIER_P (q))
6709 /* Don't allow ourself to keep walking around an
6710 always-executed loop. */
6711 if (next_real_insn (q) == next)
6713 p = NEXT_INSN (p);
6714 continue;
6717 /* Similarly, don't put a branch in our path more than once. */
6718 for (i = 0; i < path_entry; i++)
6719 if (data->path[i].branch == p)
6720 break;
6722 if (i != path_entry)
6723 break;
6725 data->path[path_entry].branch = p;
6726 data->path[path_entry++].status = PATH_TAKEN;
6728 /* This branch now ends our path. It was possible that we
6729 didn't see this branch the last time around (when the
6730 insn in front of the target was a JUMP_INSN that was
6731 turned into a no-op). */
6732 path_size = path_entry;
6734 p = JUMP_LABEL (p);
6735 /* Mark block so we won't scan it again later. */
6736 PUT_MODE (NEXT_INSN (p), QImode);
6738 /* Detect a branch around a block of code. */
6739 else if (skip_blocks && q != 0 && !LABEL_P (q))
6741 rtx tmp;
6743 if (next_real_insn (q) == next)
6745 p = NEXT_INSN (p);
6746 continue;
6749 for (i = 0; i < path_entry; i++)
6750 if (data->path[i].branch == p)
6751 break;
6753 if (i != path_entry)
6754 break;
6756 /* This is no_labels_between_p (p, q) with an added check for
6757 reaching the end of a function (in case Q precedes P). */
6758 for (tmp = NEXT_INSN (p); tmp && tmp != q; tmp = NEXT_INSN (tmp))
6759 if (LABEL_P (tmp))
6760 break;
6762 if (tmp == q)
6764 data->path[path_entry].branch = p;
6765 data->path[path_entry++].status = PATH_AROUND;
6767 path_size = path_entry;
6769 p = JUMP_LABEL (p);
6770 /* Mark block so we won't scan it again later. */
6771 PUT_MODE (NEXT_INSN (p), QImode);
6775 p = NEXT_INSN (p);
6778 data->low_cuid = low_cuid;
6779 data->high_cuid = high_cuid;
6780 data->nsets = nsets;
6781 data->last = p;
6783 /* If all jumps in the path are not taken, set our path length to zero
6784 so a rescan won't be done. */
6785 for (i = path_size - 1; i >= 0; i--)
6786 if (data->path[i].status != PATH_NOT_TAKEN)
6787 break;
6789 if (i == -1)
6790 data->path_size = 0;
6791 else
6792 data->path_size = path_size;
6794 /* End the current branch path. */
6795 data->path[path_size].branch = 0;
6798 /* Perform cse on the instructions of a function.
6799 F is the first instruction.
6800 NREGS is one plus the highest pseudo-reg number used in the instruction.
6802 Returns 1 if jump_optimize should be redone due to simplifications
6803 in conditional jump instructions. */
6806 cse_main (rtx f, int nregs)
6808 struct cse_basic_block_data val;
6809 rtx insn = f;
6810 int i;
6812 init_cse_reg_info (nregs);
6814 val.path = XNEWVEC (struct branch_path, PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH));
6816 cse_jumps_altered = 0;
6817 recorded_label_ref = 0;
6818 constant_pool_entries_cost = 0;
6819 constant_pool_entries_regcost = 0;
6820 val.path_size = 0;
6821 rtl_hooks = cse_rtl_hooks;
6823 init_recog ();
6824 init_alias_analysis ();
6826 reg_eqv_table = XNEWVEC (struct reg_eqv_elem, nregs);
6828 /* Find the largest uid. */
6830 max_uid = get_max_uid ();
6831 uid_cuid = XCNEWVEC (int, max_uid + 1);
6833 /* Compute the mapping from uids to cuids.
6834 CUIDs are numbers assigned to insns, like uids,
6835 except that cuids increase monotonically through the code.
6836 Don't assign cuids to line-number NOTEs, so that the distance in cuids
6837 between two insns is not affected by -g. */
6839 for (insn = f, i = 0; insn; insn = NEXT_INSN (insn))
6841 if (!NOTE_P (insn)
6842 || NOTE_LINE_NUMBER (insn) < 0)
6843 INSN_CUID (insn) = ++i;
6844 else
6845 /* Give a line number note the same cuid as preceding insn. */
6846 INSN_CUID (insn) = i;
6849 /* Loop over basic blocks.
6850 Compute the maximum number of qty's needed for each basic block
6851 (which is 2 for each SET). */
6852 insn = f;
6853 while (insn)
6855 cse_altered = 0;
6856 cse_end_of_basic_block (insn, &val, flag_cse_follow_jumps,
6857 flag_cse_skip_blocks);
6859 /* If this basic block was already processed or has no sets, skip it. */
6860 if (val.nsets == 0 || GET_MODE (insn) == QImode)
6862 PUT_MODE (insn, VOIDmode);
6863 insn = (val.last ? NEXT_INSN (val.last) : 0);
6864 val.path_size = 0;
6865 continue;
6868 cse_basic_block_start = val.low_cuid;
6869 cse_basic_block_end = val.high_cuid;
6870 max_qty = val.nsets * 2;
6872 if (dump_file)
6873 fprintf (dump_file, ";; Processing block from %d to %d, %d sets.\n",
6874 INSN_UID (insn), val.last ? INSN_UID (val.last) : 0,
6875 val.nsets);
6877 /* Make MAX_QTY bigger to give us room to optimize
6878 past the end of this basic block, if that should prove useful. */
6879 if (max_qty < 500)
6880 max_qty = 500;
6882 /* If this basic block is being extended by following certain jumps,
6883 (see `cse_end_of_basic_block'), we reprocess the code from the start.
6884 Otherwise, we start after this basic block. */
6885 if (val.path_size > 0)
6886 cse_basic_block (insn, val.last, val.path);
6887 else
6889 int old_cse_jumps_altered = cse_jumps_altered;
6890 rtx temp;
6892 /* When cse changes a conditional jump to an unconditional
6893 jump, we want to reprocess the block, since it will give
6894 us a new branch path to investigate. */
6895 cse_jumps_altered = 0;
6896 temp = cse_basic_block (insn, val.last, val.path);
6897 if (cse_jumps_altered == 0
6898 || (flag_cse_follow_jumps == 0 && flag_cse_skip_blocks == 0))
6899 insn = temp;
6901 cse_jumps_altered |= old_cse_jumps_altered;
6904 if (cse_altered)
6905 ggc_collect ();
6907 #ifdef USE_C_ALLOCA
6908 alloca (0);
6909 #endif
6912 /* Clean up. */
6913 end_alias_analysis ();
6914 free (uid_cuid);
6915 free (reg_eqv_table);
6916 free (val.path);
6917 rtl_hooks = general_rtl_hooks;
6919 return cse_jumps_altered || recorded_label_ref;
6922 /* Process a single basic block. FROM and TO and the limits of the basic
6923 block. NEXT_BRANCH points to the branch path when following jumps or
6924 a null path when not following jumps. */
6926 static rtx
6927 cse_basic_block (rtx from, rtx to, struct branch_path *next_branch)
6929 rtx insn;
6930 int to_usage = 0;
6931 rtx libcall_insn = NULL_RTX;
6932 int num_insns = 0;
6933 int no_conflict = 0;
6935 /* Allocate the space needed by qty_table. */
6936 qty_table = XNEWVEC (struct qty_table_elem, max_qty);
6938 new_basic_block ();
6940 /* TO might be a label. If so, protect it from being deleted. */
6941 if (to != 0 && LABEL_P (to))
6942 ++LABEL_NUSES (to);
6944 for (insn = from; insn != to; insn = NEXT_INSN (insn))
6946 enum rtx_code code = GET_CODE (insn);
6948 /* If we have processed 1,000 insns, flush the hash table to
6949 avoid extreme quadratic behavior. We must not include NOTEs
6950 in the count since there may be more of them when generating
6951 debugging information. If we clear the table at different
6952 times, code generated with -g -O might be different than code
6953 generated with -O but not -g.
6955 ??? This is a real kludge and needs to be done some other way.
6956 Perhaps for 2.9. */
6957 if (code != NOTE && num_insns++ > PARAM_VALUE (PARAM_MAX_CSE_INSNS))
6959 flush_hash_table ();
6960 num_insns = 0;
6963 /* See if this is a branch that is part of the path. If so, and it is
6964 to be taken, do so. */
6965 if (next_branch->branch == insn)
6967 enum taken status = next_branch++->status;
6968 if (status != PATH_NOT_TAKEN)
6970 if (status == PATH_TAKEN)
6971 record_jump_equiv (insn, 1);
6972 else
6973 invalidate_skipped_block (NEXT_INSN (insn));
6975 /* Set the last insn as the jump insn; it doesn't affect cc0.
6976 Then follow this branch. */
6977 #ifdef HAVE_cc0
6978 prev_insn_cc0 = 0;
6979 prev_insn = insn;
6980 #endif
6981 insn = JUMP_LABEL (insn);
6982 continue;
6986 if (GET_MODE (insn) == QImode)
6987 PUT_MODE (insn, VOIDmode);
6989 if (GET_RTX_CLASS (code) == RTX_INSN)
6991 rtx p;
6993 /* Process notes first so we have all notes in canonical forms when
6994 looking for duplicate operations. */
6996 if (REG_NOTES (insn))
6997 REG_NOTES (insn) = cse_process_notes (REG_NOTES (insn), NULL_RTX);
6999 /* Track when we are inside in LIBCALL block. Inside such a block,
7000 we do not want to record destinations. The last insn of a
7001 LIBCALL block is not considered to be part of the block, since
7002 its destination is the result of the block and hence should be
7003 recorded. */
7005 if (REG_NOTES (insn) != 0)
7007 if ((p = find_reg_note (insn, REG_LIBCALL, NULL_RTX)))
7008 libcall_insn = XEXP (p, 0);
7009 else if (find_reg_note (insn, REG_RETVAL, NULL_RTX))
7011 /* Keep libcall_insn for the last SET insn of a no-conflict
7012 block to prevent changing the destination. */
7013 if (! no_conflict)
7014 libcall_insn = 0;
7015 else
7016 no_conflict = -1;
7018 else if (find_reg_note (insn, REG_NO_CONFLICT, NULL_RTX))
7019 no_conflict = 1;
7022 cse_insn (insn, libcall_insn);
7024 if (no_conflict == -1)
7026 libcall_insn = 0;
7027 no_conflict = 0;
7030 /* If we haven't already found an insn where we added a LABEL_REF,
7031 check this one. */
7032 if (NONJUMP_INSN_P (insn) && ! recorded_label_ref
7033 && for_each_rtx (&PATTERN (insn), check_for_label_ref,
7034 (void *) insn))
7035 recorded_label_ref = 1;
7038 /* If INSN is now an unconditional jump, skip to the end of our
7039 basic block by pretending that we just did the last insn in the
7040 basic block. If we are jumping to the end of our block, show
7041 that we can have one usage of TO. */
7043 if (any_uncondjump_p (insn))
7045 if (to == 0)
7047 free (qty_table);
7048 return 0;
7051 if (JUMP_LABEL (insn) == to)
7052 to_usage = 1;
7054 /* Maybe TO was deleted because the jump is unconditional.
7055 If so, there is nothing left in this basic block. */
7056 /* ??? Perhaps it would be smarter to set TO
7057 to whatever follows this insn,
7058 and pretend the basic block had always ended here. */
7059 if (INSN_DELETED_P (to))
7060 break;
7062 insn = PREV_INSN (to);
7065 /* See if it is ok to keep on going past the label
7066 which used to end our basic block. Remember that we incremented
7067 the count of that label, so we decrement it here. If we made
7068 a jump unconditional, TO_USAGE will be one; in that case, we don't
7069 want to count the use in that jump. */
7071 if (to != 0 && NEXT_INSN (insn) == to
7072 && LABEL_P (to) && --LABEL_NUSES (to) == to_usage)
7074 struct cse_basic_block_data val;
7075 rtx prev;
7077 insn = NEXT_INSN (to);
7079 /* If TO was the last insn in the function, we are done. */
7080 if (insn == 0)
7082 free (qty_table);
7083 return 0;
7086 /* If TO was preceded by a BARRIER we are done with this block
7087 because it has no continuation. */
7088 prev = prev_nonnote_insn (to);
7089 if (prev && BARRIER_P (prev))
7091 free (qty_table);
7092 return insn;
7095 /* Find the end of the following block. Note that we won't be
7096 following branches in this case. */
7097 to_usage = 0;
7098 val.path_size = 0;
7099 val.path = XNEWVEC (struct branch_path, PARAM_VALUE (PARAM_MAX_CSE_PATH_LENGTH));
7100 cse_end_of_basic_block (insn, &val, 0, 0);
7101 free (val.path);
7103 /* If the tables we allocated have enough space left
7104 to handle all the SETs in the next basic block,
7105 continue through it. Otherwise, return,
7106 and that block will be scanned individually. */
7107 if (val.nsets * 2 + next_qty > max_qty)
7108 break;
7110 cse_basic_block_start = val.low_cuid;
7111 cse_basic_block_end = val.high_cuid;
7112 to = val.last;
7114 /* Prevent TO from being deleted if it is a label. */
7115 if (to != 0 && LABEL_P (to))
7116 ++LABEL_NUSES (to);
7118 /* Back up so we process the first insn in the extension. */
7119 insn = PREV_INSN (insn);
7123 gcc_assert (next_qty <= max_qty);
7125 free (qty_table);
7127 return to ? NEXT_INSN (to) : 0;
7130 /* Called via for_each_rtx to see if an insn is using a LABEL_REF for which
7131 there isn't a REG_LABEL note. Return one if so. DATA is the insn. */
7133 static int
7134 check_for_label_ref (rtx *rtl, void *data)
7136 rtx insn = (rtx) data;
7138 /* If this insn uses a LABEL_REF and there isn't a REG_LABEL note for it,
7139 we must rerun jump since it needs to place the note. If this is a
7140 LABEL_REF for a CODE_LABEL that isn't in the insn chain, don't do this
7141 since no REG_LABEL will be added. */
7142 return (GET_CODE (*rtl) == LABEL_REF
7143 && ! LABEL_REF_NONLOCAL_P (*rtl)
7144 && LABEL_P (XEXP (*rtl, 0))
7145 && INSN_UID (XEXP (*rtl, 0)) != 0
7146 && ! find_reg_note (insn, REG_LABEL, XEXP (*rtl, 0)));
7149 /* Count the number of times registers are used (not set) in X.
7150 COUNTS is an array in which we accumulate the count, INCR is how much
7151 we count each register usage.
7153 Don't count a usage of DEST, which is the SET_DEST of a SET which
7154 contains X in its SET_SRC. This is because such a SET does not
7155 modify the liveness of DEST.
7156 DEST is set to pc_rtx for a trapping insn, which means that we must count
7157 uses of a SET_DEST regardless because the insn can't be deleted here. */
7159 static void
7160 count_reg_usage (rtx x, int *counts, rtx dest, int incr)
7162 enum rtx_code code;
7163 rtx note;
7164 const char *fmt;
7165 int i, j;
7167 if (x == 0)
7168 return;
7170 switch (code = GET_CODE (x))
7172 case REG:
7173 if (x != dest)
7174 counts[REGNO (x)] += incr;
7175 return;
7177 case PC:
7178 case CC0:
7179 case CONST:
7180 case CONST_INT:
7181 case CONST_DOUBLE:
7182 case CONST_VECTOR:
7183 case SYMBOL_REF:
7184 case LABEL_REF:
7185 return;
7187 case CLOBBER:
7188 /* If we are clobbering a MEM, mark any registers inside the address
7189 as being used. */
7190 if (MEM_P (XEXP (x, 0)))
7191 count_reg_usage (XEXP (XEXP (x, 0), 0), counts, NULL_RTX, incr);
7192 return;
7194 case SET:
7195 /* Unless we are setting a REG, count everything in SET_DEST. */
7196 if (!REG_P (SET_DEST (x)))
7197 count_reg_usage (SET_DEST (x), counts, NULL_RTX, incr);
7198 count_reg_usage (SET_SRC (x), counts,
7199 dest ? dest : SET_DEST (x),
7200 incr);
7201 return;
7203 case CALL_INSN:
7204 case INSN:
7205 case JUMP_INSN:
7206 /* We expect dest to be NULL_RTX here. If the insn may trap, mark
7207 this fact by setting DEST to pc_rtx. */
7208 if (flag_non_call_exceptions && may_trap_p (PATTERN (x)))
7209 dest = pc_rtx;
7210 if (code == CALL_INSN)
7211 count_reg_usage (CALL_INSN_FUNCTION_USAGE (x), counts, dest, incr);
7212 count_reg_usage (PATTERN (x), counts, dest, incr);
7214 /* Things used in a REG_EQUAL note aren't dead since loop may try to
7215 use them. */
7217 note = find_reg_equal_equiv_note (x);
7218 if (note)
7220 rtx eqv = XEXP (note, 0);
7222 if (GET_CODE (eqv) == EXPR_LIST)
7223 /* This REG_EQUAL note describes the result of a function call.
7224 Process all the arguments. */
7227 count_reg_usage (XEXP (eqv, 0), counts, dest, incr);
7228 eqv = XEXP (eqv, 1);
7230 while (eqv && GET_CODE (eqv) == EXPR_LIST);
7231 else
7232 count_reg_usage (eqv, counts, dest, incr);
7234 return;
7236 case EXPR_LIST:
7237 if (REG_NOTE_KIND (x) == REG_EQUAL
7238 || (REG_NOTE_KIND (x) != REG_NONNEG && GET_CODE (XEXP (x,0)) == USE)
7239 /* FUNCTION_USAGE expression lists may include (CLOBBER (mem /u)),
7240 involving registers in the address. */
7241 || GET_CODE (XEXP (x, 0)) == CLOBBER)
7242 count_reg_usage (XEXP (x, 0), counts, NULL_RTX, incr);
7244 count_reg_usage (XEXP (x, 1), counts, NULL_RTX, incr);
7245 return;
7247 case ASM_OPERANDS:
7248 /* If the asm is volatile, then this insn cannot be deleted,
7249 and so the inputs *must* be live. */
7250 if (MEM_VOLATILE_P (x))
7251 dest = NULL_RTX;
7252 /* Iterate over just the inputs, not the constraints as well. */
7253 for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
7254 count_reg_usage (ASM_OPERANDS_INPUT (x, i), counts, dest, incr);
7255 return;
7257 case INSN_LIST:
7258 gcc_unreachable ();
7260 default:
7261 break;
7264 fmt = GET_RTX_FORMAT (code);
7265 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
7267 if (fmt[i] == 'e')
7268 count_reg_usage (XEXP (x, i), counts, dest, incr);
7269 else if (fmt[i] == 'E')
7270 for (j = XVECLEN (x, i) - 1; j >= 0; j--)
7271 count_reg_usage (XVECEXP (x, i, j), counts, dest, incr);
7275 /* Return true if set is live. */
7276 static bool
7277 set_live_p (rtx set, rtx insn ATTRIBUTE_UNUSED, /* Only used with HAVE_cc0. */
7278 int *counts)
7280 #ifdef HAVE_cc0
7281 rtx tem;
7282 #endif
7284 if (set_noop_p (set))
7287 #ifdef HAVE_cc0
7288 else if (GET_CODE (SET_DEST (set)) == CC0
7289 && !side_effects_p (SET_SRC (set))
7290 && ((tem = next_nonnote_insn (insn)) == 0
7291 || !INSN_P (tem)
7292 || !reg_referenced_p (cc0_rtx, PATTERN (tem))))
7293 return false;
7294 #endif
7295 else if (!REG_P (SET_DEST (set))
7296 || REGNO (SET_DEST (set)) < FIRST_PSEUDO_REGISTER
7297 || counts[REGNO (SET_DEST (set))] != 0
7298 || side_effects_p (SET_SRC (set)))
7299 return true;
7300 return false;
7303 /* Return true if insn is live. */
7305 static bool
7306 insn_live_p (rtx insn, int *counts)
7308 int i;
7309 if (flag_non_call_exceptions && may_trap_p (PATTERN (insn)))
7310 return true;
7311 else if (GET_CODE (PATTERN (insn)) == SET)
7312 return set_live_p (PATTERN (insn), insn, counts);
7313 else if (GET_CODE (PATTERN (insn)) == PARALLEL)
7315 for (i = XVECLEN (PATTERN (insn), 0) - 1; i >= 0; i--)
7317 rtx elt = XVECEXP (PATTERN (insn), 0, i);
7319 if (GET_CODE (elt) == SET)
7321 if (set_live_p (elt, insn, counts))
7322 return true;
7324 else if (GET_CODE (elt) != CLOBBER && GET_CODE (elt) != USE)
7325 return true;
7327 return false;
7329 else
7330 return true;
7333 /* Return true if libcall is dead as a whole. */
7335 static bool
7336 dead_libcall_p (rtx insn, int *counts)
7338 rtx note, set, new;
7340 /* See if there's a REG_EQUAL note on this insn and try to
7341 replace the source with the REG_EQUAL expression.
7343 We assume that insns with REG_RETVALs can only be reg->reg
7344 copies at this point. */
7345 note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
7346 if (!note)
7347 return false;
7349 set = single_set (insn);
7350 if (!set)
7351 return false;
7353 new = simplify_rtx (XEXP (note, 0));
7354 if (!new)
7355 new = XEXP (note, 0);
7357 /* While changing insn, we must update the counts accordingly. */
7358 count_reg_usage (insn, counts, NULL_RTX, -1);
7360 if (validate_change (insn, &SET_SRC (set), new, 0))
7362 count_reg_usage (insn, counts, NULL_RTX, 1);
7363 remove_note (insn, find_reg_note (insn, REG_RETVAL, NULL_RTX));
7364 remove_note (insn, note);
7365 return true;
7368 if (CONSTANT_P (new))
7370 new = force_const_mem (GET_MODE (SET_DEST (set)), new);
7371 if (new && validate_change (insn, &SET_SRC (set), new, 0))
7373 count_reg_usage (insn, counts, NULL_RTX, 1);
7374 remove_note (insn, find_reg_note (insn, REG_RETVAL, NULL_RTX));
7375 remove_note (insn, note);
7376 return true;
7380 count_reg_usage (insn, counts, NULL_RTX, 1);
7381 return false;
7384 /* Scan all the insns and delete any that are dead; i.e., they store a register
7385 that is never used or they copy a register to itself.
7387 This is used to remove insns made obviously dead by cse, loop or other
7388 optimizations. It improves the heuristics in loop since it won't try to
7389 move dead invariants out of loops or make givs for dead quantities. The
7390 remaining passes of the compilation are also sped up. */
7393 delete_trivially_dead_insns (rtx insns, int nreg)
7395 int *counts;
7396 rtx insn, prev;
7397 int in_libcall = 0, dead_libcall = 0;
7398 int ndead = 0;
7400 timevar_push (TV_DELETE_TRIVIALLY_DEAD);
7401 /* First count the number of times each register is used. */
7402 counts = XCNEWVEC (int, nreg);
7403 for (insn = insns; insn; insn = NEXT_INSN (insn))
7404 if (INSN_P (insn))
7405 count_reg_usage (insn, counts, NULL_RTX, 1);
7407 /* Go from the last insn to the first and delete insns that only set unused
7408 registers or copy a register to itself. As we delete an insn, remove
7409 usage counts for registers it uses.
7411 The first jump optimization pass may leave a real insn as the last
7412 insn in the function. We must not skip that insn or we may end
7413 up deleting code that is not really dead. */
7414 for (insn = get_last_insn (); insn; insn = prev)
7416 int live_insn = 0;
7418 prev = PREV_INSN (insn);
7419 if (!INSN_P (insn))
7420 continue;
7422 /* Don't delete any insns that are part of a libcall block unless
7423 we can delete the whole libcall block.
7425 Flow or loop might get confused if we did that. Remember
7426 that we are scanning backwards. */
7427 if (find_reg_note (insn, REG_RETVAL, NULL_RTX))
7429 in_libcall = 1;
7430 live_insn = 1;
7431 dead_libcall = dead_libcall_p (insn, counts);
7433 else if (in_libcall)
7434 live_insn = ! dead_libcall;
7435 else
7436 live_insn = insn_live_p (insn, counts);
7438 /* If this is a dead insn, delete it and show registers in it aren't
7439 being used. */
7441 if (! live_insn)
7443 count_reg_usage (insn, counts, NULL_RTX, -1);
7444 delete_insn_and_edges (insn);
7445 ndead++;
7448 if (in_libcall && find_reg_note (insn, REG_LIBCALL, NULL_RTX))
7450 in_libcall = 0;
7451 dead_libcall = 0;
7455 if (dump_file && ndead)
7456 fprintf (dump_file, "Deleted %i trivially dead insns\n",
7457 ndead);
7458 /* Clean up. */
7459 free (counts);
7460 timevar_pop (TV_DELETE_TRIVIALLY_DEAD);
7461 return ndead;
7464 /* This function is called via for_each_rtx. The argument, NEWREG, is
7465 a condition code register with the desired mode. If we are looking
7466 at the same register in a different mode, replace it with
7467 NEWREG. */
7469 static int
7470 cse_change_cc_mode (rtx *loc, void *data)
7472 struct change_cc_mode_args* args = (struct change_cc_mode_args*)data;
7474 if (*loc
7475 && REG_P (*loc)
7476 && REGNO (*loc) == REGNO (args->newreg)
7477 && GET_MODE (*loc) != GET_MODE (args->newreg))
7479 validate_change (args->insn, loc, args->newreg, 1);
7481 return -1;
7483 return 0;
7486 /* Change the mode of any reference to the register REGNO (NEWREG) to
7487 GET_MODE (NEWREG) in INSN. */
7489 static void
7490 cse_change_cc_mode_insn (rtx insn, rtx newreg)
7492 struct change_cc_mode_args args;
7493 int success;
7495 if (!INSN_P (insn))
7496 return;
7498 args.insn = insn;
7499 args.newreg = newreg;
7501 for_each_rtx (&PATTERN (insn), cse_change_cc_mode, &args);
7502 for_each_rtx (&REG_NOTES (insn), cse_change_cc_mode, &args);
7504 /* If the following assertion was triggered, there is most probably
7505 something wrong with the cc_modes_compatible back end function.
7506 CC modes only can be considered compatible if the insn - with the mode
7507 replaced by any of the compatible modes - can still be recognized. */
7508 success = apply_change_group ();
7509 gcc_assert (success);
7512 /* Change the mode of any reference to the register REGNO (NEWREG) to
7513 GET_MODE (NEWREG), starting at START. Stop before END. Stop at
7514 any instruction which modifies NEWREG. */
7516 static void
7517 cse_change_cc_mode_insns (rtx start, rtx end, rtx newreg)
7519 rtx insn;
7521 for (insn = start; insn != end; insn = NEXT_INSN (insn))
7523 if (! INSN_P (insn))
7524 continue;
7526 if (reg_set_p (newreg, insn))
7527 return;
7529 cse_change_cc_mode_insn (insn, newreg);
7533 /* BB is a basic block which finishes with CC_REG as a condition code
7534 register which is set to CC_SRC. Look through the successors of BB
7535 to find blocks which have a single predecessor (i.e., this one),
7536 and look through those blocks for an assignment to CC_REG which is
7537 equivalent to CC_SRC. CAN_CHANGE_MODE indicates whether we are
7538 permitted to change the mode of CC_SRC to a compatible mode. This
7539 returns VOIDmode if no equivalent assignments were found.
7540 Otherwise it returns the mode which CC_SRC should wind up with.
7542 The main complexity in this function is handling the mode issues.
7543 We may have more than one duplicate which we can eliminate, and we
7544 try to find a mode which will work for multiple duplicates. */
7546 static enum machine_mode
7547 cse_cc_succs (basic_block bb, rtx cc_reg, rtx cc_src, bool can_change_mode)
7549 bool found_equiv;
7550 enum machine_mode mode;
7551 unsigned int insn_count;
7552 edge e;
7553 rtx insns[2];
7554 enum machine_mode modes[2];
7555 rtx last_insns[2];
7556 unsigned int i;
7557 rtx newreg;
7558 edge_iterator ei;
7560 /* We expect to have two successors. Look at both before picking
7561 the final mode for the comparison. If we have more successors
7562 (i.e., some sort of table jump, although that seems unlikely),
7563 then we require all beyond the first two to use the same
7564 mode. */
7566 found_equiv = false;
7567 mode = GET_MODE (cc_src);
7568 insn_count = 0;
7569 FOR_EACH_EDGE (e, ei, bb->succs)
7571 rtx insn;
7572 rtx end;
7574 if (e->flags & EDGE_COMPLEX)
7575 continue;
7577 if (EDGE_COUNT (e->dest->preds) != 1
7578 || e->dest == EXIT_BLOCK_PTR)
7579 continue;
7581 end = NEXT_INSN (BB_END (e->dest));
7582 for (insn = BB_HEAD (e->dest); insn != end; insn = NEXT_INSN (insn))
7584 rtx set;
7586 if (! INSN_P (insn))
7587 continue;
7589 /* If CC_SRC is modified, we have to stop looking for
7590 something which uses it. */
7591 if (modified_in_p (cc_src, insn))
7592 break;
7594 /* Check whether INSN sets CC_REG to CC_SRC. */
7595 set = single_set (insn);
7596 if (set
7597 && REG_P (SET_DEST (set))
7598 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7600 bool found;
7601 enum machine_mode set_mode;
7602 enum machine_mode comp_mode;
7604 found = false;
7605 set_mode = GET_MODE (SET_SRC (set));
7606 comp_mode = set_mode;
7607 if (rtx_equal_p (cc_src, SET_SRC (set)))
7608 found = true;
7609 else if (GET_CODE (cc_src) == COMPARE
7610 && GET_CODE (SET_SRC (set)) == COMPARE
7611 && mode != set_mode
7612 && rtx_equal_p (XEXP (cc_src, 0),
7613 XEXP (SET_SRC (set), 0))
7614 && rtx_equal_p (XEXP (cc_src, 1),
7615 XEXP (SET_SRC (set), 1)))
7618 comp_mode = targetm.cc_modes_compatible (mode, set_mode);
7619 if (comp_mode != VOIDmode
7620 && (can_change_mode || comp_mode == mode))
7621 found = true;
7624 if (found)
7626 found_equiv = true;
7627 if (insn_count < ARRAY_SIZE (insns))
7629 insns[insn_count] = insn;
7630 modes[insn_count] = set_mode;
7631 last_insns[insn_count] = end;
7632 ++insn_count;
7634 if (mode != comp_mode)
7636 gcc_assert (can_change_mode);
7637 mode = comp_mode;
7639 /* The modified insn will be re-recognized later. */
7640 PUT_MODE (cc_src, mode);
7643 else
7645 if (set_mode != mode)
7647 /* We found a matching expression in the
7648 wrong mode, but we don't have room to
7649 store it in the array. Punt. This case
7650 should be rare. */
7651 break;
7653 /* INSN sets CC_REG to a value equal to CC_SRC
7654 with the right mode. We can simply delete
7655 it. */
7656 delete_insn (insn);
7659 /* We found an instruction to delete. Keep looking,
7660 in the hopes of finding a three-way jump. */
7661 continue;
7664 /* We found an instruction which sets the condition
7665 code, so don't look any farther. */
7666 break;
7669 /* If INSN sets CC_REG in some other way, don't look any
7670 farther. */
7671 if (reg_set_p (cc_reg, insn))
7672 break;
7675 /* If we fell off the bottom of the block, we can keep looking
7676 through successors. We pass CAN_CHANGE_MODE as false because
7677 we aren't prepared to handle compatibility between the
7678 further blocks and this block. */
7679 if (insn == end)
7681 enum machine_mode submode;
7683 submode = cse_cc_succs (e->dest, cc_reg, cc_src, false);
7684 if (submode != VOIDmode)
7686 gcc_assert (submode == mode);
7687 found_equiv = true;
7688 can_change_mode = false;
7693 if (! found_equiv)
7694 return VOIDmode;
7696 /* Now INSN_COUNT is the number of instructions we found which set
7697 CC_REG to a value equivalent to CC_SRC. The instructions are in
7698 INSNS. The modes used by those instructions are in MODES. */
7700 newreg = NULL_RTX;
7701 for (i = 0; i < insn_count; ++i)
7703 if (modes[i] != mode)
7705 /* We need to change the mode of CC_REG in INSNS[i] and
7706 subsequent instructions. */
7707 if (! newreg)
7709 if (GET_MODE (cc_reg) == mode)
7710 newreg = cc_reg;
7711 else
7712 newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7714 cse_change_cc_mode_insns (NEXT_INSN (insns[i]), last_insns[i],
7715 newreg);
7718 delete_insn (insns[i]);
7721 return mode;
7724 /* If we have a fixed condition code register (or two), walk through
7725 the instructions and try to eliminate duplicate assignments. */
7727 static void
7728 cse_condition_code_reg (void)
7730 unsigned int cc_regno_1;
7731 unsigned int cc_regno_2;
7732 rtx cc_reg_1;
7733 rtx cc_reg_2;
7734 basic_block bb;
7736 if (! targetm.fixed_condition_code_regs (&cc_regno_1, &cc_regno_2))
7737 return;
7739 cc_reg_1 = gen_rtx_REG (CCmode, cc_regno_1);
7740 if (cc_regno_2 != INVALID_REGNUM)
7741 cc_reg_2 = gen_rtx_REG (CCmode, cc_regno_2);
7742 else
7743 cc_reg_2 = NULL_RTX;
7745 FOR_EACH_BB (bb)
7747 rtx last_insn;
7748 rtx cc_reg;
7749 rtx insn;
7750 rtx cc_src_insn;
7751 rtx cc_src;
7752 enum machine_mode mode;
7753 enum machine_mode orig_mode;
7755 /* Look for blocks which end with a conditional jump based on a
7756 condition code register. Then look for the instruction which
7757 sets the condition code register. Then look through the
7758 successor blocks for instructions which set the condition
7759 code register to the same value. There are other possible
7760 uses of the condition code register, but these are by far the
7761 most common and the ones which we are most likely to be able
7762 to optimize. */
7764 last_insn = BB_END (bb);
7765 if (!JUMP_P (last_insn))
7766 continue;
7768 if (reg_referenced_p (cc_reg_1, PATTERN (last_insn)))
7769 cc_reg = cc_reg_1;
7770 else if (cc_reg_2 && reg_referenced_p (cc_reg_2, PATTERN (last_insn)))
7771 cc_reg = cc_reg_2;
7772 else
7773 continue;
7775 cc_src_insn = NULL_RTX;
7776 cc_src = NULL_RTX;
7777 for (insn = PREV_INSN (last_insn);
7778 insn && insn != PREV_INSN (BB_HEAD (bb));
7779 insn = PREV_INSN (insn))
7781 rtx set;
7783 if (! INSN_P (insn))
7784 continue;
7785 set = single_set (insn);
7786 if (set
7787 && REG_P (SET_DEST (set))
7788 && REGNO (SET_DEST (set)) == REGNO (cc_reg))
7790 cc_src_insn = insn;
7791 cc_src = SET_SRC (set);
7792 break;
7794 else if (reg_set_p (cc_reg, insn))
7795 break;
7798 if (! cc_src_insn)
7799 continue;
7801 if (modified_between_p (cc_src, cc_src_insn, NEXT_INSN (last_insn)))
7802 continue;
7804 /* Now CC_REG is a condition code register used for a
7805 conditional jump at the end of the block, and CC_SRC, in
7806 CC_SRC_INSN, is the value to which that condition code
7807 register is set, and CC_SRC is still meaningful at the end of
7808 the basic block. */
7810 orig_mode = GET_MODE (cc_src);
7811 mode = cse_cc_succs (bb, cc_reg, cc_src, true);
7812 if (mode != VOIDmode)
7814 gcc_assert (mode == GET_MODE (cc_src));
7815 if (mode != orig_mode)
7817 rtx newreg = gen_rtx_REG (mode, REGNO (cc_reg));
7819 cse_change_cc_mode_insn (cc_src_insn, newreg);
7821 /* Do the same in the following insns that use the
7822 current value of CC_REG within BB. */
7823 cse_change_cc_mode_insns (NEXT_INSN (cc_src_insn),
7824 NEXT_INSN (last_insn),
7825 newreg);
7832 /* Perform common subexpression elimination. Nonzero value from
7833 `cse_main' means that jumps were simplified and some code may now
7834 be unreachable, so do jump optimization again. */
7835 static bool
7836 gate_handle_cse (void)
7838 return optimize > 0;
7841 static void
7842 rest_of_handle_cse (void)
7844 int tem;
7846 if (dump_file)
7847 dump_flow_info (dump_file, dump_flags);
7849 reg_scan (get_insns (), max_reg_num ());
7851 tem = cse_main (get_insns (), max_reg_num ());
7852 if (tem)
7853 rebuild_jump_labels (get_insns ());
7854 if (purge_all_dead_edges ())
7855 delete_unreachable_blocks ();
7857 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7859 /* If we are not running more CSE passes, then we are no longer
7860 expecting CSE to be run. But always rerun it in a cheap mode. */
7861 cse_not_expected = !flag_rerun_cse_after_loop && !flag_gcse;
7863 if (tem)
7864 delete_dead_jumptables ();
7866 if (tem || optimize > 1)
7867 cleanup_cfg (CLEANUP_EXPENSIVE | CLEANUP_PRE_LOOP);
7870 struct tree_opt_pass pass_cse =
7872 "cse1", /* name */
7873 gate_handle_cse, /* gate */
7874 rest_of_handle_cse, /* execute */
7875 NULL, /* sub */
7876 NULL, /* next */
7877 0, /* static_pass_number */
7878 TV_CSE, /* tv_id */
7879 0, /* properties_required */
7880 0, /* properties_provided */
7881 0, /* properties_destroyed */
7882 0, /* todo_flags_start */
7883 TODO_dump_func |
7884 TODO_ggc_collect, /* todo_flags_finish */
7885 's' /* letter */
7889 static bool
7890 gate_handle_cse2 (void)
7892 return optimize > 0 && flag_rerun_cse_after_loop;
7895 /* Run second CSE pass after loop optimizations. */
7896 static void
7897 rest_of_handle_cse2 (void)
7899 int tem;
7901 if (dump_file)
7902 dump_flow_info (dump_file, dump_flags);
7904 tem = cse_main (get_insns (), max_reg_num ());
7906 /* Run a pass to eliminate duplicated assignments to condition code
7907 registers. We have to run this after bypass_jumps, because it
7908 makes it harder for that pass to determine whether a jump can be
7909 bypassed safely. */
7910 cse_condition_code_reg ();
7912 purge_all_dead_edges ();
7913 delete_trivially_dead_insns (get_insns (), max_reg_num ());
7915 if (tem)
7917 timevar_push (TV_JUMP);
7918 rebuild_jump_labels (get_insns ());
7919 delete_dead_jumptables ();
7920 cleanup_cfg (CLEANUP_EXPENSIVE);
7921 timevar_pop (TV_JUMP);
7923 reg_scan (get_insns (), max_reg_num ());
7924 cse_not_expected = 1;
7928 struct tree_opt_pass pass_cse2 =
7930 "cse2", /* name */
7931 gate_handle_cse2, /* gate */
7932 rest_of_handle_cse2, /* execute */
7933 NULL, /* sub */
7934 NULL, /* next */
7935 0, /* static_pass_number */
7936 TV_CSE2, /* tv_id */
7937 0, /* properties_required */
7938 0, /* properties_provided */
7939 0, /* properties_destroyed */
7940 0, /* todo_flags_start */
7941 TODO_dump_func |
7942 TODO_ggc_collect, /* todo_flags_finish */
7943 't' /* letter */