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 Free Software Foundation, Inc.
|
---|
4 |
|
---|
5 | This file is part of GCC.
|
---|
6 |
|
---|
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.
|
---|
11 |
|
---|
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.
|
---|
16 |
|
---|
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, 59 Temple Place - Suite 330, Boston, MA
|
---|
20 | 02111-1307, USA. */
|
---|
21 |
|
---|
22 | #include "config.h"
|
---|
23 | /* stdio.h must precede rtl.h for FFS. */
|
---|
24 | #include "system.h"
|
---|
25 |
|
---|
26 | #include "rtl.h"
|
---|
27 | #include "tm_p.h"
|
---|
28 | #include "regs.h"
|
---|
29 | #include "hard-reg-set.h"
|
---|
30 | #include "basic-block.h"
|
---|
31 | #include "flags.h"
|
---|
32 | #include "real.h"
|
---|
33 | #include "insn-config.h"
|
---|
34 | #include "recog.h"
|
---|
35 | #include "function.h"
|
---|
36 | #include "expr.h"
|
---|
37 | #include "toplev.h"
|
---|
38 | #include "output.h"
|
---|
39 | #include "ggc.h"
|
---|
40 | #include "timevar.h"
|
---|
41 |
|
---|
42 | /* The basic idea of common subexpression elimination is to go
|
---|
43 | through the code, keeping a record of expressions that would
|
---|
44 | have the same value at the current scan point, and replacing
|
---|
45 | expressions encountered with the cheapest equivalent expression.
|
---|
46 |
|
---|
47 | It is too complicated to keep track of the different possibilities
|
---|
48 | when control paths merge in this code; so, at each label, we forget all
|
---|
49 | that is known and start fresh. This can be described as processing each
|
---|
50 | extended basic block separately. We have a separate pass to perform
|
---|
51 | global CSE.
|
---|
52 |
|
---|
53 | Note CSE can turn a conditional or computed jump into a nop or
|
---|
54 | an unconditional jump. When this occurs we arrange to run the jump
|
---|
55 | optimizer after CSE to delete the unreachable code.
|
---|
56 |
|
---|
57 | We use two data structures to record the equivalent expressions:
|
---|
58 | a hash table for most expressions, and a vector of "quantity
|
---|
59 | numbers" to record equivalent (pseudo) registers.
|
---|
60 |
|
---|
61 | The use of the special data structure for registers is desirable
|
---|
62 | because it is faster. It is possible because registers references
|
---|
63 | contain a fairly small number, the register number, taken from
|
---|
64 | a contiguously allocated series, and two register references are
|
---|
65 | identical if they have the same number. General expressions
|
---|
66 | do not have any such thing, so the only way to retrieve the
|
---|
67 | information recorded on an expression other than a register
|
---|
68 | is to keep it in a hash table.
|
---|
69 |
|
---|
70 | Registers and "quantity numbers":
|
---|
71 |
|
---|
72 | At the start of each basic block, all of the (hardware and pseudo)
|
---|
73 | registers used in the function are given distinct quantity
|
---|
74 | numbers to indicate their contents. During scan, when the code
|
---|
75 | copies one register into another, we copy the quantity number.
|
---|
76 | When a register is loaded in any other way, we allocate a new
|
---|
77 | quantity number to describe the value generated by this operation.
|
---|
78 | `reg_qty' records what quantity a register is currently thought
|
---|
79 | of as containing.
|
---|
80 |
|
---|
81 | All real quantity numbers are greater than or equal to `max_reg'.
|
---|
82 | If register N has not been assigned a quantity, reg_qty[N] will equal N.
|
---|
83 |
|
---|
84 | Quantity numbers below `max_reg' do not exist and none of the `qty_table'
|
---|
85 | entries should be referenced with an index below `max_reg'.
|
---|
86 |
|
---|
87 | We also maintain a bidirectional chain of registers for each
|
---|
88 | quantity number. The `qty_table` members `first_reg' and `last_reg',
|
---|
89 | and `reg_eqv_table' members `next' and `prev' hold these chains.
|
---|
90 |
|
---|
91 | The first register in a chain is the one whose lifespan is least local.
|
---|
92 | Among equals, it is the one that was seen first.
|
---|
93 | We replace any equivalent register with that one.
|
---|
94 |
|
---|
95 | If two registers have the same quantity number, it must be true that
|
---|
96 | REG expressions with qty_table `mode' must be in the hash table for both
|
---|
97 | registers and must be in the same class.
|
---|
98 |
|
---|
99 | The converse is not true. Since hard registers may be referenced in
|
---|
100 | any mode, two REG expressions might be equivalent in the hash table
|
---|
101 | but not have the same quantity number if the quantity number of one
|
---|
102 | of the registers is not the same mode as those expressions.
|
---|
103 |
|
---|
104 | Constants and quantity numbers
|
---|
105 |
|
---|
106 | When a quantity has a known constant value, that value is stored
|
---|
107 | in the appropriate qty_table `const_rtx'. This is in addition to
|
---|
108 | putting the constant in the hash table as is usual for non-regs.
|
---|
109 |
|
---|
110 | Whether a reg or a constant is preferred is determined by the configuration
|
---|
111 | macro CONST_COSTS and will often depend on the constant value. In any
|
---|
112 | event, expressions containing constants can be simplified, by fold_rtx.
|
---|
113 |
|
---|
114 | When a quantity has a known nearly constant value (such as an address
|
---|
115 | of a stack slot), that value is stored in the appropriate qty_table
|
---|
116 | `const_rtx'.
|
---|
117 |
|
---|
118 | Integer constants don't have a machine mode. However, cse
|
---|
119 | determines the intended machine mode from the destination
|
---|
120 | of the instruction that moves the constant. The machine mode
|
---|
121 | is recorded in the hash table along with the actual RTL
|
---|
122 | constant expression so that different modes are kept separate.
|
---|
123 |
|
---|
124 | Other expressions:
|
---|
125 |
|
---|
126 | To record known equivalences among expressions in general
|
---|
127 | we use a hash table called `table'. It has a fixed number of buckets
|
---|
128 | that contain chains of `struct table_elt' elements for expressions.
|
---|
129 | These chains connect the elements whose expressions have the same
|
---|
130 | hash codes.
|
---|
131 |
|
---|
132 | Other chains through the same elements connect the elements which
|
---|
133 | currently have equivalent values.
|
---|
134 |
|
---|
135 | Register references in an expression are canonicalized before hashing
|
---|
136 | the expression. This is done using `reg_qty' and qty_table `first_reg'.
|
---|
137 | The hash code of a register reference is computed using the quantity
|
---|
138 | number, not the register number.
|
---|
139 |
|
---|
140 | When the value of an expression changes, it is necessary to remove from the
|
---|
141 | hash table not just that expression but all expressions whose values
|
---|
142 | could be different as a result.
|
---|
143 |
|
---|
144 | 1. If the value changing is in memory, except in special cases
|
---|
145 | ANYTHING referring to memory could be changed. That is because
|
---|
146 | nobody knows where a pointer does not point.
|
---|
147 | The function `invalidate_memory' removes what is necessary.
|
---|
148 |
|
---|
149 | The special cases are when the address is constant or is
|
---|
150 | a constant plus a fixed register such as the frame pointer
|
---|
151 | or a static chain pointer. When such addresses are stored in,
|
---|
152 | we can tell exactly which other such addresses must be invalidated
|
---|
153 | due to overlap. `invalidate' does this.
|
---|
154 | All expressions that refer to non-constant
|
---|
155 | memory addresses are also invalidated. `invalidate_memory' does this.
|
---|
156 |
|
---|
157 | 2. If the value changing is a register, all expressions
|
---|
158 | containing references to that register, and only those,
|
---|
159 | must be removed.
|
---|
160 |
|
---|
161 | Because searching the entire hash table for expressions that contain
|
---|
162 | a register is very slow, we try to figure out when it isn't necessary.
|
---|
163 | Precisely, this is necessary only when expressions have been
|
---|
164 | entered in the hash table using this register, and then the value has
|
---|
165 | changed, and then another expression wants to be added to refer to
|
---|
166 | the register's new value. This sequence of circumstances is rare
|
---|
167 | within any one basic block.
|
---|
168 |
|
---|
169 | The vectors `reg_tick' and `reg_in_table' are used to detect this case.
|
---|
170 | reg_tick[i] is incremented whenever a value is stored in register i.
|
---|
171 | reg_in_table[i] holds -1 if no references to register i have been
|
---|
172 | entered in the table; otherwise, it contains the value reg_tick[i] had
|
---|
173 | when the references were entered. If we want to enter a reference
|
---|
174 | and reg_in_table[i] != reg_tick[i], we must scan and remove old references.
|
---|
175 | Until we want to enter a new entry, the mere fact that the two vectors
|
---|
176 | don't match makes the entries be ignored if anyone tries to match them.
|
---|
177 |
|
---|
178 | Registers themselves are entered in the hash table as well as in
|
---|
179 | the equivalent-register chains. However, the vectors `reg_tick'
|
---|
180 | and `reg_in_table' do not apply to expressions which are simple
|
---|
181 | register references. These expressions are removed from the table
|
---|
182 | immediately when they become invalid, and this can be done even if
|
---|
183 | we do not immediately search for all the expressions that refer to
|
---|
184 | the register.
|
---|
185 |
|
---|
186 | A CLOBBER rtx in an instruction invalidates its operand for further
|
---|
187 | reuse. A CLOBBER or SET rtx whose operand is a MEM:BLK
|
---|
188 | invalidates everything that resides in memory.
|
---|
189 |
|
---|
190 | Related expressions:
|
---|
191 |
|
---|
192 | Constant expressions that differ only by an additive integer
|
---|
193 | are called related. When a constant expression is put in
|
---|
194 | the table, the related expression with no constant term
|
---|
195 | is also entered. These are made to point at each other
|
---|
196 | so that it is possible to find out if there exists any
|
---|
197 | register equivalent to an expression related to a given expression. */
|
---|
198 |
|
---|
199 | /* One plus largest register number used in this function. */
|
---|
200 |
|
---|
201 | static int max_reg;
|
---|
202 |
|
---|
203 | /* One plus largest instruction UID used in this function at time of
|
---|
204 | cse_main call. */
|
---|
205 |
|
---|
206 | static int max_insn_uid;
|
---|
207 |
|
---|
208 | /* Length of qty_table vector. We know in advance we will not need
|
---|
209 | a quantity number this big. */
|
---|
210 |
|
---|
211 | static int max_qty;
|
---|
212 |
|
---|
213 | /* Next quantity number to be allocated.
|
---|
214 | This is 1 + the largest number needed so far. */
|
---|
215 |
|
---|
216 | static int next_qty;
|
---|
217 |
|
---|
218 | /* Per-qty information tracking.
|
---|
219 |
|
---|
220 | `first_reg' and `last_reg' track the head and tail of the
|
---|
221 | chain of registers which currently contain this quantity.
|
---|
222 |
|
---|
223 | `mode' contains the machine mode of this quantity.
|
---|
224 |
|
---|
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.
|
---|
230 |
|
---|
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. */
|
---|
242 |
|
---|
243 | struct qty_table_elem
|
---|
244 | {
|
---|
245 | rtx const_rtx;
|
---|
246 | rtx const_insn;
|
---|
247 | rtx comparison_const;
|
---|
248 | int comparison_qty;
|
---|
249 | unsigned int first_reg, last_reg;
|
---|
250 | enum machine_mode mode;
|
---|
251 | enum rtx_code comparison_code;
|
---|
252 | };
|
---|
253 |
|
---|
254 | /* The table of all qtys, indexed by qty number. */
|
---|
255 | static struct qty_table_elem *qty_table;
|
---|
256 |
|
---|
257 | #ifdef HAVE_cc0
|
---|
258 | /* For machines that have a CC0, we do not record its value in the hash
|
---|
259 | table since its use is guaranteed to be the insn immediately following
|
---|
260 | its definition and any other insn is presumed to invalidate it.
|
---|
261 |
|
---|
262 | Instead, we store below the value last assigned to CC0. If it should
|
---|
263 | happen to be a constant, it is stored in preference to the actual
|
---|
264 | assigned value. In case it is a constant, we store the mode in which
|
---|
265 | the constant should be interpreted. */
|
---|
266 |
|
---|
267 | static rtx prev_insn_cc0;
|
---|
268 | static enum machine_mode prev_insn_cc0_mode;
|
---|
269 | #endif
|
---|
270 |
|
---|
271 | /* Previous actual insn. 0 if at first insn of basic block. */
|
---|
272 |
|
---|
273 | static rtx prev_insn;
|
---|
274 |
|
---|
275 | /* Insn being scanned. */
|
---|
276 |
|
---|
277 | static rtx this_insn;
|
---|
278 |
|
---|
279 | /* Index by register number, gives the number of the next (or
|
---|
280 | previous) register in the chain of registers sharing the same
|
---|
281 | value.
|
---|
282 |
|
---|
283 | Or -1 if this register is at the end of the chain.
|
---|
284 |
|
---|
285 | If reg_qty[N] == N, reg_eqv_table[N].next is undefined. */
|
---|
286 |
|
---|
287 | /* Per-register equivalence chain. */
|
---|
288 | struct reg_eqv_elem
|
---|
289 | {
|
---|
290 | int next, prev;
|
---|
291 | };
|
---|
292 |
|
---|
293 | /* The table of all register equivalence chains. */
|
---|
294 | static struct reg_eqv_elem *reg_eqv_table;
|
---|
295 |
|
---|
296 | struct cse_reg_info
|
---|
297 | {
|
---|
298 | /* Next in hash chain. */
|
---|
299 | struct cse_reg_info *hash_next;
|
---|
300 |
|
---|
301 | /* The next cse_reg_info structure in the free or used list. */
|
---|
302 | struct cse_reg_info *next;
|
---|
303 |
|
---|
304 | /* Search key */
|
---|
305 | unsigned int regno;
|
---|
306 |
|
---|
307 | /* The quantity number of the register's current contents. */
|
---|
308 | int reg_qty;
|
---|
309 |
|
---|
310 | /* The number of times the register has been altered in the current
|
---|
311 | basic block. */
|
---|
312 | int reg_tick;
|
---|
313 |
|
---|
314 | /* The REG_TICK value at which rtx's containing this register are
|
---|
315 | valid in the hash table. If this does not equal the current
|
---|
316 | reg_tick value, such expressions existing in the hash table are
|
---|
317 | invalid. */
|
---|
318 | int reg_in_table;
|
---|
319 |
|
---|
320 | /* The SUBREG that was set when REG_TICK was last incremented. Set
|
---|
321 | to -1 if the last store was to the whole register, not a subreg. */
|
---|
322 | unsigned int subreg_ticked;
|
---|
323 | };
|
---|
324 |
|
---|
325 | /* A free list of cse_reg_info entries. */
|
---|
326 | static struct cse_reg_info *cse_reg_info_free_list;
|
---|
327 |
|
---|
328 | /* A used list of cse_reg_info entries. */
|
---|
329 | static struct cse_reg_info *cse_reg_info_used_list;
|
---|
330 | static struct cse_reg_info *cse_reg_info_used_list_end;
|
---|
331 |
|
---|
332 | /* A mapping from registers to cse_reg_info data structures. */
|
---|
333 | #define REGHASH_SHIFT 7
|
---|
334 | #define REGHASH_SIZE (1 << REGHASH_SHIFT)
|
---|
335 | #define REGHASH_MASK (REGHASH_SIZE - 1)
|
---|
336 | static struct cse_reg_info *reg_hash[REGHASH_SIZE];
|
---|
337 |
|
---|
338 | #define REGHASH_FN(REGNO) \
|
---|
339 | (((REGNO) ^ ((REGNO) >> REGHASH_SHIFT)) & REGHASH_MASK)
|
---|
340 |
|
---|
341 | /* The last lookup we did into the cse_reg_info_tree. This allows us
|
---|
342 | to cache repeated lookups. */
|
---|
343 | static unsigned int cached_regno;
|
---|
344 | static struct cse_reg_info *cached_cse_reg_info;
|
---|
345 |
|
---|
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. */
|
---|
350 |
|
---|
351 | static HARD_REG_SET hard_regs_in_table;
|
---|
352 |
|
---|
353 | /* CUID of insn that starts the basic block currently being cse-processed. */
|
---|
354 |
|
---|
355 | static int cse_basic_block_start;
|
---|
356 |
|
---|
357 | /* CUID of insn that ends the basic block currently being cse-processed. */
|
---|
358 |
|
---|
359 | static int cse_basic_block_end;
|
---|
360 |
|
---|
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. */
|
---|
364 |
|
---|
365 | static int *uid_cuid;
|
---|
366 |
|
---|
367 | /* Highest UID in UID_CUID. */
|
---|
368 | static int max_uid;
|
---|
369 |
|
---|
370 | /* Get the cuid of an insn. */
|
---|
371 |
|
---|
372 | #define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)])
|
---|
373 |
|
---|
374 | /* Nonzero if this pass has made changes, and therefore it's
|
---|
375 | worthwhile to run the garbage collector. */
|
---|
376 |
|
---|
377 | static int cse_altered;
|
---|
378 |
|
---|
379 | /* Nonzero if cse has altered conditional jump insns
|
---|
380 | in such a way that jump optimization should be redone. */
|
---|
381 |
|
---|
382 | static int cse_jumps_altered;
|
---|
383 |
|
---|
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;
|
---|
387 |
|
---|
388 | /* canon_hash stores 1 in do_not_record
|
---|
389 | if it notices a reference to CC0, PC, or some other volatile
|
---|
390 | subexpression. */
|
---|
391 |
|
---|
392 | static int do_not_record;
|
---|
393 |
|
---|
394 | #ifdef LOAD_EXTEND_OP
|
---|
395 |
|
---|
396 | /* Scratch rtl used when looking for load-extended copy of a MEM. */
|
---|
397 | static rtx memory_extend_rtx;
|
---|
398 | #endif
|
---|
399 |
|
---|
400 | /* canon_hash stores 1 in hash_arg_in_memory
|
---|
401 | if it notices a reference to memory within the expression being hashed. */
|
---|
402 |
|
---|
403 | static int hash_arg_in_memory;
|
---|
404 |
|
---|
405 | /* The hash table contains buckets which are chains of `struct table_elt's,
|
---|
406 | each recording one expression's information.
|
---|
407 | That expression is in the `exp' field.
|
---|
408 |
|
---|
409 | The canon_exp field contains a canonical (from the point of view of
|
---|
410 | alias analysis) version of the `exp' field.
|
---|
411 |
|
---|
412 | Those elements with the same hash code are chained in both directions
|
---|
413 | through the `next_same_hash' and `prev_same_hash' fields.
|
---|
414 |
|
---|
415 | Each set of expressions with equivalent values
|
---|
416 | are on a two-way chain through the `next_same_value'
|
---|
417 | and `prev_same_value' fields, and all point with
|
---|
418 | the `first_same_value' field at the first element in
|
---|
419 | that chain. The chain is in order of increasing cost.
|
---|
420 | Each element's cost value is in its `cost' field.
|
---|
421 |
|
---|
422 | The `in_memory' field is nonzero for elements that
|
---|
423 | involve any reference to memory. These elements are removed
|
---|
424 | whenever a write is done to an unidentified location in memory.
|
---|
425 | To be safe, we assume that a memory address is unidentified unless
|
---|
426 | the address is either a symbol constant or a constant plus
|
---|
427 | the frame pointer or argument pointer.
|
---|
428 |
|
---|
429 | The `related_value' field is used to connect related expressions
|
---|
430 | (that differ by adding an integer).
|
---|
431 | The related expressions are chained in a circular fashion.
|
---|
432 | `related_value' is zero for expressions for which this
|
---|
433 | chain is not useful.
|
---|
434 |
|
---|
435 | The `cost' field stores the cost of this element's expression.
|
---|
436 | The `regcost' field stores the value returned by approx_reg_cost for
|
---|
437 | this element's expression.
|
---|
438 |
|
---|
439 | The `is_const' flag is set if the element is a constant (including
|
---|
440 | a fixed address).
|
---|
441 |
|
---|
442 | The `flag' field is used as a temporary during some search routines.
|
---|
443 |
|
---|
444 | The `mode' field is usually the same as GET_MODE (`exp'), but
|
---|
445 | if `exp' is a CONST_INT and has no machine mode then the `mode'
|
---|
446 | field is the mode it was being used as. Each constant is
|
---|
447 | recorded separately for each mode it is used with. */
|
---|
448 |
|
---|
449 | struct table_elt
|
---|
450 | {
|
---|
451 | rtx exp;
|
---|
452 | rtx canon_exp;
|
---|
453 | struct table_elt *next_same_hash;
|
---|
454 | struct table_elt *prev_same_hash;
|
---|
455 | struct table_elt *next_same_value;
|
---|
456 | struct table_elt *prev_same_value;
|
---|
457 | struct table_elt *first_same_value;
|
---|
458 | struct table_elt *related_value;
|
---|
459 | int cost;
|
---|
460 | int regcost;
|
---|
461 | enum machine_mode mode;
|
---|
462 | char in_memory;
|
---|
463 | char is_const;
|
---|
464 | char flag;
|
---|
465 | };
|
---|
466 |
|
---|
467 | /* We don't want a lot of buckets, because we rarely have very many
|
---|
468 | things stored in the hash table, and a lot of buckets slows
|
---|
469 | down a lot of loops that happen frequently. */
|
---|
470 | #define HASH_SHIFT 5
|
---|
471 | #define HASH_SIZE (1 << HASH_SHIFT)
|
---|
472 | #define HASH_MASK (HASH_SIZE - 1)
|
---|
473 |
|
---|
474 | /* Compute hash code of X in mode M. Special-case case where X is a pseudo
|
---|
475 | register (hard registers may require `do_not_record' to be set). */
|
---|
476 |
|
---|
477 | #define HASH(X, M) \
|
---|
478 | ((GET_CODE (X) == REG && REGNO (X) >= FIRST_PSEUDO_REGISTER \
|
---|
479 | ? (((unsigned) REG << 7) + (unsigned) REG_QTY (REGNO (X))) \
|
---|
480 | : canon_hash (X, M)) & HASH_MASK)
|
---|
481 |
|
---|
482 | /* Determine whether register number N is considered a fixed register for the
|
---|
483 | purpose of approximating register costs.
|
---|
484 | It is desirable to replace other regs with fixed regs, to reduce need for
|
---|
485 | non-fixed hard regs.
|
---|
486 | A reg wins if it is either the frame pointer or designated as fixed. */
|
---|
487 | #define FIXED_REGNO_P(N) \
|
---|
488 | ((N) == FRAME_POINTER_REGNUM || (N) == HARD_FRAME_POINTER_REGNUM \
|
---|
489 | || fixed_regs[N] || global_regs[N])
|
---|
490 |
|
---|
491 | /* Compute cost of X, as stored in the `cost' field of a table_elt. Fixed
|
---|
492 | hard registers and pointers into the frame are the cheapest with a cost
|
---|
493 | of 0. Next come pseudos with a cost of one and other hard registers with
|
---|
494 | a cost of 2. Aside from these special cases, call `rtx_cost'. */
|
---|
495 |
|
---|
496 | #define CHEAP_REGNO(N) \
|
---|
497 | ((N) == FRAME_POINTER_REGNUM || (N) == HARD_FRAME_POINTER_REGNUM \
|
---|
498 | || (N) == STACK_POINTER_REGNUM || (N) == ARG_POINTER_REGNUM \
|
---|
499 | || ((N) >= FIRST_VIRTUAL_REGISTER && (N) <= LAST_VIRTUAL_REGISTER) \
|
---|
500 | || ((N) < FIRST_PSEUDO_REGISTER \
|
---|
501 | && FIXED_REGNO_P (N) && REGNO_REG_CLASS (N) != NO_REGS))
|
---|
502 |
|
---|
503 | #define COST(X) (GET_CODE (X) == REG ? 0 : notreg_cost (X, SET))
|
---|
504 | #define COST_IN(X,OUTER) (GET_CODE (X) == REG ? 0 : notreg_cost (X, OUTER))
|
---|
505 |
|
---|
506 | /* Get the info associated with register N. */
|
---|
507 |
|
---|
508 | #define GET_CSE_REG_INFO(N) \
|
---|
509 | (((N) == cached_regno && cached_cse_reg_info) \
|
---|
510 | ? cached_cse_reg_info : get_cse_reg_info ((N)))
|
---|
511 |
|
---|
512 | /* Get the number of times this register has been updated in this
|
---|
513 | basic block. */
|
---|
514 |
|
---|
515 | #define REG_TICK(N) ((GET_CSE_REG_INFO (N))->reg_tick)
|
---|
516 |
|
---|
517 | /* Get the point at which REG was recorded in the table. */
|
---|
518 |
|
---|
519 | #define REG_IN_TABLE(N) ((GET_CSE_REG_INFO (N))->reg_in_table)
|
---|
520 |
|
---|
521 | /* Get the SUBREG set at the last increment to REG_TICK (-1 if not a
|
---|
522 | SUBREG). */
|
---|
523 |
|
---|
524 | #define SUBREG_TICKED(N) ((GET_CSE_REG_INFO (N))->subreg_ticked)
|
---|
525 |
|
---|
526 | /* Get the quantity number for REG. */
|
---|
527 |
|
---|
528 | #define REG_QTY(N) ((GET_CSE_REG_INFO (N))->reg_qty)
|
---|
529 |
|
---|
530 | /* Determine if the quantity number for register X represents a valid index
|
---|
531 | into the qty_table. */
|
---|
532 |
|
---|
533 | #define REGNO_QTY_VALID_P(N) (REG_QTY (N) != (int) (N))
|
---|
534 |
|
---|
535 | static struct table_elt *table[HASH_SIZE];
|
---|
536 |
|
---|
537 | /* Chain of `struct table_elt's made so far for this function
|
---|
538 | but currently removed from the table. */
|
---|
539 |
|
---|
540 | static struct table_elt *free_element_chain;
|
---|
541 |
|
---|
542 | /* Number of `struct table_elt' structures made so far for this function. */
|
---|
543 |
|
---|
544 | static int n_elements_made;
|
---|
545 |
|
---|
546 | /* Maximum value `n_elements_made' has had so far in this compilation
|
---|
547 | for functions previously processed. */
|
---|
548 |
|
---|
549 | static int max_elements_made;
|
---|
550 |
|
---|
551 | /* Surviving equivalence class when two equivalence classes are merged
|
---|
552 | by recording the effects of a jump in the last insn. Zero if the
|
---|
553 | last insn was not a conditional jump. */
|
---|
554 |
|
---|
555 | static struct table_elt *last_jump_equiv_class;
|
---|
556 |
|
---|
557 | /* Set to the cost of a constant pool reference if one was found for a
|
---|
558 | symbolic constant. If this was found, it means we should try to
|
---|
559 | convert constants into constant pool entries if they don't fit in
|
---|
560 | the insn. */
|
---|
561 |
|
---|
562 | static int constant_pool_entries_cost;
|
---|
563 |
|
---|
564 | /* Define maximum length of a branch path. */
|
---|
565 |
|
---|
566 | #define PATHLENGTH 10
|
---|
567 |
|
---|
568 | /* This data describes a block that will be processed by cse_basic_block. */
|
---|
569 |
|
---|
570 | struct cse_basic_block_data
|
---|
571 | {
|
---|
572 | /* Lowest CUID value of insns in block. */
|
---|
573 | int low_cuid;
|
---|
574 | /* Highest CUID value of insns in block. */
|
---|
575 | int high_cuid;
|
---|
576 | /* Total number of SETs in block. */
|
---|
577 | int nsets;
|
---|
578 | /* Last insn in the block. */
|
---|
579 | rtx last;
|
---|
580 | /* Size of current branch path, if any. */
|
---|
581 | int path_size;
|
---|
582 | /* Current branch path, indicating which branches will be taken. */
|
---|
583 | struct branch_path
|
---|
584 | {
|
---|
585 | /* The branch insn. */
|
---|
586 | rtx branch;
|
---|
587 | /* Whether it should be taken or not. AROUND is the same as taken
|
---|
588 | except that it is used when the destination label is not preceded
|
---|
589 | by a BARRIER. */
|
---|
590 | enum taken {TAKEN, NOT_TAKEN, AROUND} status;
|
---|
591 | } path[PATHLENGTH];
|
---|
592 | };
|
---|
593 |
|
---|
594 | /* Nonzero if X has the form (PLUS frame-pointer integer). We check for
|
---|
595 | virtual regs here because the simplify_*_operation routines are called
|
---|
596 | by integrate.c, which is called before virtual register instantiation.
|
---|
597 |
|
---|
598 | ?!? FIXED_BASE_PLUS_P and NONZERO_BASE_PLUS_P need to move into
|
---|
599 | a header file so that their definitions can be shared with the
|
---|
600 | simplification routines in simplify-rtx.c. Until then, do not
|
---|
601 | change these macros without also changing the copy in simplify-rtx.c. */
|
---|
602 |
|
---|
603 | #define FIXED_BASE_PLUS_P(X) \
|
---|
604 | ((X) == frame_pointer_rtx || (X) == hard_frame_pointer_rtx \
|
---|
605 | || ((X) == arg_pointer_rtx && fixed_regs[ARG_POINTER_REGNUM])\
|
---|
606 | || (X) == virtual_stack_vars_rtx \
|
---|
607 | || (X) == virtual_incoming_args_rtx \
|
---|
608 | || (GET_CODE (X) == PLUS && GET_CODE (XEXP (X, 1)) == CONST_INT \
|
---|
609 | && (XEXP (X, 0) == frame_pointer_rtx \
|
---|
610 | || XEXP (X, 0) == hard_frame_pointer_rtx \
|
---|
611 | || ((X) == arg_pointer_rtx \
|
---|
612 | && fixed_regs[ARG_POINTER_REGNUM]) \
|
---|
613 | || XEXP (X, 0) == virtual_stack_vars_rtx \
|
---|
614 | || XEXP (X, 0) == virtual_incoming_args_rtx)) \
|
---|
615 | || GET_CODE (X) == ADDRESSOF)
|
---|
616 |
|
---|
617 | /* Similar, but also allows reference to the stack pointer.
|
---|
618 |
|
---|
619 | This used to include FIXED_BASE_PLUS_P, however, we can't assume that
|
---|
620 | arg_pointer_rtx by itself is nonzero, because on at least one machine,
|
---|
621 | the i960, the arg pointer is zero when it is unused. */
|
---|
622 |
|
---|
623 | #define NONZERO_BASE_PLUS_P(X) \
|
---|
624 | ((X) == frame_pointer_rtx || (X) == hard_frame_pointer_rtx \
|
---|
625 | || (X) == virtual_stack_vars_rtx \
|
---|
626 | || (X) == virtual_incoming_args_rtx \
|
---|
627 | || (GET_CODE (X) == PLUS && GET_CODE (XEXP (X, 1)) == CONST_INT \
|
---|
628 | && (XEXP (X, 0) == frame_pointer_rtx \
|
---|
629 | || XEXP (X, 0) == hard_frame_pointer_rtx \
|
---|
630 | || ((X) == arg_pointer_rtx \
|
---|
631 | && fixed_regs[ARG_POINTER_REGNUM]) \
|
---|
632 | || XEXP (X, 0) == virtual_stack_vars_rtx \
|
---|
633 | || XEXP (X, 0) == virtual_incoming_args_rtx)) \
|
---|
634 | || (X) == stack_pointer_rtx \
|
---|
635 | || (X) == virtual_stack_dynamic_rtx \
|
---|
636 | || (X) == virtual_outgoing_args_rtx \
|
---|
637 | || (GET_CODE (X) == PLUS && GET_CODE (XEXP (X, 1)) == CONST_INT \
|
---|
638 | && (XEXP (X, 0) == stack_pointer_rtx \
|
---|
639 | || XEXP (X, 0) == virtual_stack_dynamic_rtx \
|
---|
640 | || XEXP (X, 0) == virtual_outgoing_args_rtx)) \
|
---|
641 | || GET_CODE (X) == ADDRESSOF)
|
---|
642 |
|
---|
643 | static int notreg_cost PARAMS ((rtx, enum rtx_code));
|
---|
644 | static int approx_reg_cost_1 PARAMS ((rtx *, void *));
|
---|
645 | static int approx_reg_cost PARAMS ((rtx));
|
---|
646 | static int preferrable PARAMS ((int, int, int, int));
|
---|
647 | static void new_basic_block PARAMS ((void));
|
---|
648 | static void make_new_qty PARAMS ((unsigned int, enum machine_mode));
|
---|
649 | static void make_regs_eqv PARAMS ((unsigned int, unsigned int));
|
---|
650 | static void delete_reg_equiv PARAMS ((unsigned int));
|
---|
651 | static int mention_regs PARAMS ((rtx));
|
---|
652 | static int insert_regs PARAMS ((rtx, struct table_elt *, int));
|
---|
653 | static void remove_from_table PARAMS ((struct table_elt *, unsigned));
|
---|
654 | static struct table_elt *lookup PARAMS ((rtx, unsigned, enum machine_mode)),
|
---|
655 | *lookup_for_remove PARAMS ((rtx, unsigned, enum machine_mode));
|
---|
656 | static rtx lookup_as_function PARAMS ((rtx, enum rtx_code));
|
---|
657 | static struct table_elt *insert PARAMS ((rtx, struct table_elt *, unsigned,
|
---|
658 | enum machine_mode));
|
---|
659 | static void merge_equiv_classes PARAMS ((struct table_elt *,
|
---|
660 | struct table_elt *));
|
---|
661 | static void invalidate PARAMS ((rtx, enum machine_mode));
|
---|
662 | static int cse_rtx_varies_p PARAMS ((rtx, int));
|
---|
663 | static void remove_invalid_refs PARAMS ((unsigned int));
|
---|
664 | static void remove_invalid_subreg_refs PARAMS ((unsigned int, unsigned int,
|
---|
665 | enum machine_mode));
|
---|
666 | static void rehash_using_reg PARAMS ((rtx));
|
---|
667 | static void invalidate_memory PARAMS ((void));
|
---|
668 | static void invalidate_for_call PARAMS ((void));
|
---|
669 | static rtx use_related_value PARAMS ((rtx, struct table_elt *));
|
---|
670 | static unsigned canon_hash PARAMS ((rtx, enum machine_mode));
|
---|
671 | static unsigned canon_hash_string PARAMS ((const char *));
|
---|
672 | static unsigned safe_hash PARAMS ((rtx, enum machine_mode));
|
---|
673 | static int exp_equiv_p PARAMS ((rtx, rtx, int, int));
|
---|
674 | static rtx canon_reg PARAMS ((rtx, rtx));
|
---|
675 | static void find_best_addr PARAMS ((rtx, rtx *, enum machine_mode));
|
---|
676 | static enum rtx_code find_comparison_args PARAMS ((enum rtx_code, rtx *, rtx *,
|
---|
677 | enum machine_mode *,
|
---|
678 | enum machine_mode *));
|
---|
679 | static rtx fold_rtx PARAMS ((rtx, rtx));
|
---|
680 | static rtx equiv_constant PARAMS ((rtx));
|
---|
681 | static void record_jump_equiv PARAMS ((rtx, int));
|
---|
682 | static void record_jump_cond PARAMS ((enum rtx_code, enum machine_mode,
|
---|
683 | rtx, rtx, int));
|
---|
684 | static void cse_insn PARAMS ((rtx, rtx));
|
---|
685 | static int addr_affects_sp_p PARAMS ((rtx));
|
---|
686 | static void invalidate_from_clobbers PARAMS ((rtx));
|
---|
687 | static rtx cse_process_notes PARAMS ((rtx, rtx));
|
---|
688 | static void cse_around_loop PARAMS ((rtx));
|
---|
689 | static void invalidate_skipped_set PARAMS ((rtx, rtx, void *));
|
---|
690 | static void invalidate_skipped_block PARAMS ((rtx));
|
---|
691 | static void cse_check_loop_start PARAMS ((rtx, rtx, void *));
|
---|
692 | static void cse_set_around_loop PARAMS ((rtx, rtx, rtx));
|
---|
693 | static rtx cse_basic_block PARAMS ((rtx, rtx, struct branch_path *, int));
|
---|
694 | static void count_reg_usage PARAMS ((rtx, int *, rtx, int));
|
---|
695 | static int check_for_label_ref PARAMS ((rtx *, void *));
|
---|
696 | extern void dump_class PARAMS ((struct table_elt*));
|
---|
697 | static struct cse_reg_info * get_cse_reg_info PARAMS ((unsigned int));
|
---|
698 | static int check_dependence PARAMS ((rtx *, void *));
|
---|
699 |
|
---|
700 | static void flush_hash_table PARAMS ((void));
|
---|
701 | static bool insn_live_p PARAMS ((rtx, int *));
|
---|
702 | static bool set_live_p PARAMS ((rtx, rtx, int *));
|
---|
703 | static bool dead_libcall_p PARAMS ((rtx, int *));
|
---|
704 | |
---|
705 |
|
---|
706 | /* Dump the expressions in the equivalence class indicated by CLASSP.
|
---|
707 | This function is used only for debugging. */
|
---|
708 | void
|
---|
709 | dump_class (classp)
|
---|
710 | struct table_elt *classp;
|
---|
711 | {
|
---|
712 | struct table_elt *elt;
|
---|
713 |
|
---|
714 | fprintf (stderr, "Equivalence chain for ");
|
---|
715 | print_rtl (stderr, classp->exp);
|
---|
716 | fprintf (stderr, ": \n");
|
---|
717 |
|
---|
718 | for (elt = classp->first_same_value; elt; elt = elt->next_same_value)
|
---|
719 | {
|
---|
720 | print_rtl (stderr, elt->exp);
|
---|
721 | fprintf (stderr, "\n");
|
---|
722 | }
|
---|
723 | }
|
---|
724 |
|
---|
725 | /* Subroutine of approx_reg_cost; called through for_each_rtx. */
|
---|
726 |
|
---|
727 | static int
|
---|
728 | approx_reg_cost_1 (xp, data)
|
---|
729 | rtx *xp;
|
---|
730 | void *data;
|
---|
731 | {
|
---|
732 | rtx x = *xp;
|
---|
733 | int *cost_p = data;
|
---|
734 |
|
---|
735 | if (x && GET_CODE (x) == REG)
|
---|
736 | {
|
---|
737 | unsigned int regno = REGNO (x);
|
---|
738 |
|
---|
739 | if (! CHEAP_REGNO (regno))
|
---|
740 | {
|
---|
741 | if (regno < FIRST_PSEUDO_REGISTER)
|
---|
742 | {
|
---|
743 | if (SMALL_REGISTER_CLASSES)
|
---|
744 | return 1;
|
---|
745 | *cost_p += 2;
|
---|
746 | }
|
---|
747 | else
|
---|
748 | *cost_p += 1;
|
---|
749 | }
|
---|
750 | }
|
---|
751 |
|
---|
752 | return 0;
|
---|
753 | }
|
---|
754 |
|
---|
755 | /* Return an estimate of the cost of the registers used in an rtx.
|
---|
756 | This is mostly the number of different REG expressions in the rtx;
|
---|
757 | however for some exceptions like fixed registers we use a cost of
|
---|
758 | 0. If any other hard register reference occurs, return MAX_COST. */
|
---|
759 |
|
---|
760 | static int
|
---|
761 | approx_reg_cost (x)
|
---|
762 | rtx x;
|
---|
763 | {
|
---|
764 | int cost = 0;
|
---|
765 |
|
---|
766 | if (for_each_rtx (&x, approx_reg_cost_1, (void *) &cost))
|
---|
767 | return MAX_COST;
|
---|
768 |
|
---|
769 | return cost;
|
---|
770 | }
|
---|
771 |
|
---|
772 | /* Return a negative value if an rtx A, whose costs are given by COST_A
|
---|
773 | and REGCOST_A, is more desirable than an rtx B.
|
---|
774 | Return a positive value if A is less desirable, or 0 if the two are
|
---|
775 | equally good. */
|
---|
776 | static int
|
---|
777 | preferrable (cost_a, regcost_a, cost_b, regcost_b)
|
---|
778 | int cost_a, regcost_a, cost_b, regcost_b;
|
---|
779 | {
|
---|
780 | /* First, get rid of cases involving expressions that are entirely
|
---|
781 | unwanted. */
|
---|
782 | if (cost_a != cost_b)
|
---|
783 | {
|
---|
784 | if (cost_a == MAX_COST)
|
---|
785 | return 1;
|
---|
786 | if (cost_b == MAX_COST)
|
---|
787 | return -1;
|
---|
788 | }
|
---|
789 |
|
---|
790 | /* Avoid extending lifetimes of hardregs. */
|
---|
791 | if (regcost_a != regcost_b)
|
---|
792 | {
|
---|
793 | if (regcost_a == MAX_COST)
|
---|
794 | return 1;
|
---|
795 | if (regcost_b == MAX_COST)
|
---|
796 | return -1;
|
---|
797 | }
|
---|
798 |
|
---|
799 | /* Normal operation costs take precedence. */
|
---|
800 | if (cost_a != cost_b)
|
---|
801 | return cost_a - cost_b;
|
---|
802 | /* Only if these are identical consider effects on register pressure. */
|
---|
803 | if (regcost_a != regcost_b)
|
---|
804 | return regcost_a - regcost_b;
|
---|
805 | return 0;
|
---|
806 | }
|
---|
807 |
|
---|
808 | /* Internal function, to compute cost when X is not a register; called
|
---|
809 | from COST macro to keep it simple. */
|
---|
810 |
|
---|
811 | static int
|
---|
812 | notreg_cost (x, outer)
|
---|
813 | rtx x;
|
---|
814 | enum rtx_code outer;
|
---|
815 | {
|
---|
816 | return ((GET_CODE (x) == SUBREG
|
---|
817 | && GET_CODE (SUBREG_REG (x)) == REG
|
---|
818 | && GET_MODE_CLASS (GET_MODE (x)) == MODE_INT
|
---|
819 | && GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_INT
|
---|
820 | && (GET_MODE_SIZE (GET_MODE (x))
|
---|
821 | < GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
|
---|
822 | && subreg_lowpart_p (x)
|
---|
823 | && TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (GET_MODE (x)),
|
---|
824 | GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (x)))))
|
---|
825 | ? 0
|
---|
826 | : rtx_cost (x, outer) * 2);
|
---|
827 | }
|
---|
828 |
|
---|
829 | /* Return an estimate of the cost of computing rtx X.
|
---|
830 | One use is in cse, to decide which expression to keep in the hash table.
|
---|
831 | Another is in rtl generation, to pick the cheapest way to multiply.
|
---|
832 | Other uses like the latter are expected in the future. */
|
---|
833 |
|
---|
834 | int
|
---|
835 | rtx_cost (x, outer_code)
|
---|
836 | rtx x;
|
---|
837 | enum rtx_code outer_code ATTRIBUTE_UNUSED;
|
---|
838 | {
|
---|
839 | int i, j;
|
---|
840 | enum rtx_code code;
|
---|
841 | const char *fmt;
|
---|
842 | int total;
|
---|
843 |
|
---|
844 | if (x == 0)
|
---|
845 | return 0;
|
---|
846 |
|
---|
847 | /* Compute the default costs of certain things.
|
---|
848 | Note that RTX_COSTS can override the defaults. */
|
---|
849 |
|
---|
850 | code = GET_CODE (x);
|
---|
851 | switch (code)
|
---|
852 | {
|
---|
853 | case MULT:
|
---|
854 | total = COSTS_N_INSNS (5);
|
---|
855 | break;
|
---|
856 | case DIV:
|
---|
857 | case UDIV:
|
---|
858 | case MOD:
|
---|
859 | case UMOD:
|
---|
860 | total = COSTS_N_INSNS (7);
|
---|
861 | break;
|
---|
862 | case USE:
|
---|
863 | /* Used in loop.c and combine.c as a marker. */
|
---|
864 | total = 0;
|
---|
865 | break;
|
---|
866 | default:
|
---|
867 | total = COSTS_N_INSNS (1);
|
---|
868 | }
|
---|
869 |
|
---|
870 | switch (code)
|
---|
871 | {
|
---|
872 | case REG:
|
---|
873 | return 0;
|
---|
874 |
|
---|
875 | case SUBREG:
|
---|
876 | /* If we can't tie these modes, make this expensive. The larger
|
---|
877 | the mode, the more expensive it is. */
|
---|
878 | if (! MODES_TIEABLE_P (GET_MODE (x), GET_MODE (SUBREG_REG (x))))
|
---|
879 | return COSTS_N_INSNS (2
|
---|
880 | + GET_MODE_SIZE (GET_MODE (x)) / UNITS_PER_WORD);
|
---|
881 | break;
|
---|
882 |
|
---|
883 | #ifdef RTX_COSTS
|
---|
884 | RTX_COSTS (x, code, outer_code);
|
---|
885 | #endif
|
---|
886 | #ifdef CONST_COSTS
|
---|
887 | CONST_COSTS (x, code, outer_code);
|
---|
888 | #endif
|
---|
889 |
|
---|
890 | default:
|
---|
891 | #ifdef DEFAULT_RTX_COSTS
|
---|
892 | DEFAULT_RTX_COSTS (x, code, outer_code);
|
---|
893 | #endif
|
---|
894 | break;
|
---|
895 | }
|
---|
896 |
|
---|
897 | /* Sum the costs of the sub-rtx's, plus cost of this operation,
|
---|
898 | which is already in total. */
|
---|
899 |
|
---|
900 | fmt = GET_RTX_FORMAT (code);
|
---|
901 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
902 | if (fmt[i] == 'e')
|
---|
903 | total += rtx_cost (XEXP (x, i), code);
|
---|
904 | else if (fmt[i] == 'E')
|
---|
905 | for (j = 0; j < XVECLEN (x, i); j++)
|
---|
906 | total += rtx_cost (XVECEXP (x, i, j), code);
|
---|
907 |
|
---|
908 | return total;
|
---|
909 | }
|
---|
910 | |
---|
911 |
|
---|
912 | /* Return cost of address expression X.
|
---|
913 | Expect that X is properly formed address reference. */
|
---|
914 |
|
---|
915 | int
|
---|
916 | address_cost (x, mode)
|
---|
917 | rtx x;
|
---|
918 | enum machine_mode mode;
|
---|
919 | {
|
---|
920 | /* The ADDRESS_COST macro does not deal with ADDRESSOF nodes. But,
|
---|
921 | during CSE, such nodes are present. Using an ADDRESSOF node which
|
---|
922 | refers to the address of a REG is a good thing because we can then
|
---|
923 | turn (MEM (ADDRESSSOF (REG))) into just plain REG. */
|
---|
924 |
|
---|
925 | if (GET_CODE (x) == ADDRESSOF && REG_P (XEXP ((x), 0)))
|
---|
926 | return -1;
|
---|
927 |
|
---|
928 | /* We may be asked for cost of various unusual addresses, such as operands
|
---|
929 | of push instruction. It is not worthwhile to complicate writing
|
---|
930 | of ADDRESS_COST macro by such cases. */
|
---|
931 |
|
---|
932 | if (!memory_address_p (mode, x))
|
---|
933 | return 1000;
|
---|
934 | #ifdef ADDRESS_COST
|
---|
935 | return ADDRESS_COST (x);
|
---|
936 | #else
|
---|
937 | return rtx_cost (x, MEM);
|
---|
938 | #endif
|
---|
939 | }
|
---|
940 |
|
---|
941 | |
---|
942 |
|
---|
943 | static struct cse_reg_info *
|
---|
944 | get_cse_reg_info (regno)
|
---|
945 | unsigned int regno;
|
---|
946 | {
|
---|
947 | struct cse_reg_info **hash_head = ®_hash[REGHASH_FN (regno)];
|
---|
948 | struct cse_reg_info *p;
|
---|
949 |
|
---|
950 | for (p = *hash_head; p != NULL; p = p->hash_next)
|
---|
951 | if (p->regno == regno)
|
---|
952 | break;
|
---|
953 |
|
---|
954 | if (p == NULL)
|
---|
955 | {
|
---|
956 | /* Get a new cse_reg_info structure. */
|
---|
957 | if (cse_reg_info_free_list)
|
---|
958 | {
|
---|
959 | p = cse_reg_info_free_list;
|
---|
960 | cse_reg_info_free_list = p->next;
|
---|
961 | }
|
---|
962 | else
|
---|
963 | p = (struct cse_reg_info *) xmalloc (sizeof (struct cse_reg_info));
|
---|
964 |
|
---|
965 | /* Insert into hash table. */
|
---|
966 | p->hash_next = *hash_head;
|
---|
967 | *hash_head = p;
|
---|
968 |
|
---|
969 | /* Initialize it. */
|
---|
970 | p->reg_tick = 1;
|
---|
971 | p->reg_in_table = -1;
|
---|
972 | p->subreg_ticked = -1;
|
---|
973 | p->reg_qty = regno;
|
---|
974 | p->regno = regno;
|
---|
975 | p->next = cse_reg_info_used_list;
|
---|
976 | cse_reg_info_used_list = p;
|
---|
977 | if (!cse_reg_info_used_list_end)
|
---|
978 | cse_reg_info_used_list_end = p;
|
---|
979 | }
|
---|
980 |
|
---|
981 | /* Cache this lookup; we tend to be looking up information about the
|
---|
982 | same register several times in a row. */
|
---|
983 | cached_regno = regno;
|
---|
984 | cached_cse_reg_info = p;
|
---|
985 |
|
---|
986 | return p;
|
---|
987 | }
|
---|
988 |
|
---|
989 | /* Clear the hash table and initialize each register with its own quantity,
|
---|
990 | for a new basic block. */
|
---|
991 |
|
---|
992 | static void
|
---|
993 | new_basic_block ()
|
---|
994 | {
|
---|
995 | int i;
|
---|
996 |
|
---|
997 | next_qty = max_reg;
|
---|
998 |
|
---|
999 | /* Clear out hash table state for this pass. */
|
---|
1000 |
|
---|
1001 | memset ((char *) reg_hash, 0, sizeof reg_hash);
|
---|
1002 |
|
---|
1003 | if (cse_reg_info_used_list)
|
---|
1004 | {
|
---|
1005 | cse_reg_info_used_list_end->next = cse_reg_info_free_list;
|
---|
1006 | cse_reg_info_free_list = cse_reg_info_used_list;
|
---|
1007 | cse_reg_info_used_list = cse_reg_info_used_list_end = 0;
|
---|
1008 | }
|
---|
1009 | cached_cse_reg_info = 0;
|
---|
1010 |
|
---|
1011 | CLEAR_HARD_REG_SET (hard_regs_in_table);
|
---|
1012 |
|
---|
1013 | /* The per-quantity values used to be initialized here, but it is
|
---|
1014 | much faster to initialize each as it is made in `make_new_qty'. */
|
---|
1015 |
|
---|
1016 | for (i = 0; i < HASH_SIZE; i++)
|
---|
1017 | {
|
---|
1018 | struct table_elt *first;
|
---|
1019 |
|
---|
1020 | first = table[i];
|
---|
1021 | if (first != NULL)
|
---|
1022 | {
|
---|
1023 | struct table_elt *last = first;
|
---|
1024 |
|
---|
1025 | table[i] = NULL;
|
---|
1026 |
|
---|
1027 | while (last->next_same_hash != NULL)
|
---|
1028 | last = last->next_same_hash;
|
---|
1029 |
|
---|
1030 | /* Now relink this hash entire chain into
|
---|
1031 | the free element list. */
|
---|
1032 |
|
---|
1033 | last->next_same_hash = free_element_chain;
|
---|
1034 | free_element_chain = first;
|
---|
1035 | }
|
---|
1036 | }
|
---|
1037 |
|
---|
1038 | prev_insn = 0;
|
---|
1039 |
|
---|
1040 | #ifdef HAVE_cc0
|
---|
1041 | prev_insn_cc0 = 0;
|
---|
1042 | #endif
|
---|
1043 | }
|
---|
1044 |
|
---|
1045 | /* Say that register REG contains a quantity in mode MODE not in any
|
---|
1046 | register before and initialize that quantity. */
|
---|
1047 |
|
---|
1048 | static void
|
---|
1049 | make_new_qty (reg, mode)
|
---|
1050 | unsigned int reg;
|
---|
1051 | enum machine_mode mode;
|
---|
1052 | {
|
---|
1053 | int q;
|
---|
1054 | struct qty_table_elem *ent;
|
---|
1055 | struct reg_eqv_elem *eqv;
|
---|
1056 |
|
---|
1057 | if (next_qty >= max_qty)
|
---|
1058 | abort ();
|
---|
1059 |
|
---|
1060 | q = REG_QTY (reg) = next_qty++;
|
---|
1061 | ent = &qty_table[q];
|
---|
1062 | ent->first_reg = reg;
|
---|
1063 | ent->last_reg = reg;
|
---|
1064 | ent->mode = mode;
|
---|
1065 | ent->const_rtx = ent->const_insn = NULL_RTX;
|
---|
1066 | ent->comparison_code = UNKNOWN;
|
---|
1067 |
|
---|
1068 | eqv = ®_eqv_table[reg];
|
---|
1069 | eqv->next = eqv->prev = -1;
|
---|
1070 | }
|
---|
1071 |
|
---|
1072 | /* Make reg NEW equivalent to reg OLD.
|
---|
1073 | OLD is not changing; NEW is. */
|
---|
1074 |
|
---|
1075 | static void
|
---|
1076 | make_regs_eqv (new, old)
|
---|
1077 | unsigned int new, old;
|
---|
1078 | {
|
---|
1079 | unsigned int lastr, firstr;
|
---|
1080 | int q = REG_QTY (old);
|
---|
1081 | struct qty_table_elem *ent;
|
---|
1082 |
|
---|
1083 | ent = &qty_table[q];
|
---|
1084 |
|
---|
1085 | /* Nothing should become eqv until it has a "non-invalid" qty number. */
|
---|
1086 | if (! REGNO_QTY_VALID_P (old))
|
---|
1087 | abort ();
|
---|
1088 |
|
---|
1089 | REG_QTY (new) = q;
|
---|
1090 | firstr = ent->first_reg;
|
---|
1091 | lastr = ent->last_reg;
|
---|
1092 |
|
---|
1093 | /* Prefer fixed hard registers to anything. Prefer pseudo regs to other
|
---|
1094 | hard regs. Among pseudos, if NEW will live longer than any other reg
|
---|
1095 | of the same qty, and that is beyond the current basic block,
|
---|
1096 | make it the new canonical replacement for this qty. */
|
---|
1097 | if (! (firstr < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (firstr))
|
---|
1098 | /* Certain fixed registers might be of the class NO_REGS. This means
|
---|
1099 | that not only can they not be allocated by the compiler, but
|
---|
1100 | they cannot be used in substitutions or canonicalizations
|
---|
1101 | either. */
|
---|
1102 | && (new >= FIRST_PSEUDO_REGISTER || REGNO_REG_CLASS (new) != NO_REGS)
|
---|
1103 | && ((new < FIRST_PSEUDO_REGISTER && FIXED_REGNO_P (new))
|
---|
1104 | || (new >= FIRST_PSEUDO_REGISTER
|
---|
1105 | && (firstr < FIRST_PSEUDO_REGISTER
|
---|
1106 | || ((uid_cuid[REGNO_LAST_UID (new)] > cse_basic_block_end
|
---|
1107 | || (uid_cuid[REGNO_FIRST_UID (new)]
|
---|
1108 | < cse_basic_block_start))
|
---|
1109 | && (uid_cuid[REGNO_LAST_UID (new)]
|
---|
1110 | > uid_cuid[REGNO_LAST_UID (firstr)]))))))
|
---|
1111 | {
|
---|
1112 | reg_eqv_table[firstr].prev = new;
|
---|
1113 | reg_eqv_table[new].next = firstr;
|
---|
1114 | reg_eqv_table[new].prev = -1;
|
---|
1115 | ent->first_reg = new;
|
---|
1116 | }
|
---|
1117 | else
|
---|
1118 | {
|
---|
1119 | /* If NEW is a hard reg (known to be non-fixed), insert at end.
|
---|
1120 | Otherwise, insert before any non-fixed hard regs that are at the
|
---|
1121 | end. Registers of class NO_REGS cannot be used as an
|
---|
1122 | equivalent for anything. */
|
---|
1123 | while (lastr < FIRST_PSEUDO_REGISTER && reg_eqv_table[lastr].prev >= 0
|
---|
1124 | && (REGNO_REG_CLASS (lastr) == NO_REGS || ! FIXED_REGNO_P (lastr))
|
---|
1125 | && new >= FIRST_PSEUDO_REGISTER)
|
---|
1126 | lastr = reg_eqv_table[lastr].prev;
|
---|
1127 | reg_eqv_table[new].next = reg_eqv_table[lastr].next;
|
---|
1128 | if (reg_eqv_table[lastr].next >= 0)
|
---|
1129 | reg_eqv_table[reg_eqv_table[lastr].next].prev = new;
|
---|
1130 | else
|
---|
1131 | qty_table[q].last_reg = new;
|
---|
1132 | reg_eqv_table[lastr].next = new;
|
---|
1133 | reg_eqv_table[new].prev = lastr;
|
---|
1134 | }
|
---|
1135 | }
|
---|
1136 |
|
---|
1137 | /* Remove REG from its equivalence class. */
|
---|
1138 |
|
---|
1139 | static void
|
---|
1140 | delete_reg_equiv (reg)
|
---|
1141 | unsigned int reg;
|
---|
1142 | {
|
---|
1143 | struct qty_table_elem *ent;
|
---|
1144 | int q = REG_QTY (reg);
|
---|
1145 | int p, n;
|
---|
1146 |
|
---|
1147 | /* If invalid, do nothing. */
|
---|
1148 | if (q == (int) reg)
|
---|
1149 | return;
|
---|
1150 |
|
---|
1151 | ent = &qty_table[q];
|
---|
1152 |
|
---|
1153 | p = reg_eqv_table[reg].prev;
|
---|
1154 | n = reg_eqv_table[reg].next;
|
---|
1155 |
|
---|
1156 | if (n != -1)
|
---|
1157 | reg_eqv_table[n].prev = p;
|
---|
1158 | else
|
---|
1159 | ent->last_reg = p;
|
---|
1160 | if (p != -1)
|
---|
1161 | reg_eqv_table[p].next = n;
|
---|
1162 | else
|
---|
1163 | ent->first_reg = n;
|
---|
1164 |
|
---|
1165 | REG_QTY (reg) = reg;
|
---|
1166 | }
|
---|
1167 |
|
---|
1168 | /* Remove any invalid expressions from the hash table
|
---|
1169 | that refer to any of the registers contained in expression X.
|
---|
1170 |
|
---|
1171 | Make sure that newly inserted references to those registers
|
---|
1172 | as subexpressions will be considered valid.
|
---|
1173 |
|
---|
1174 | mention_regs is not called when a register itself
|
---|
1175 | is being stored in the table.
|
---|
1176 |
|
---|
1177 | Return 1 if we have done something that may have changed the hash code
|
---|
1178 | of X. */
|
---|
1179 |
|
---|
1180 | static int
|
---|
1181 | mention_regs (x)
|
---|
1182 | rtx x;
|
---|
1183 | {
|
---|
1184 | enum rtx_code code;
|
---|
1185 | int i, j;
|
---|
1186 | const char *fmt;
|
---|
1187 | int changed = 0;
|
---|
1188 |
|
---|
1189 | if (x == 0)
|
---|
1190 | return 0;
|
---|
1191 |
|
---|
1192 | code = GET_CODE (x);
|
---|
1193 | if (code == REG)
|
---|
1194 | {
|
---|
1195 | unsigned int regno = REGNO (x);
|
---|
1196 | unsigned int endregno
|
---|
1197 | = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
|
---|
1198 | : HARD_REGNO_NREGS (regno, GET_MODE (x)));
|
---|
1199 | unsigned int i;
|
---|
1200 |
|
---|
1201 | for (i = regno; i < endregno; i++)
|
---|
1202 | {
|
---|
1203 | if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
|
---|
1204 | remove_invalid_refs (i);
|
---|
1205 |
|
---|
1206 | REG_IN_TABLE (i) = REG_TICK (i);
|
---|
1207 | SUBREG_TICKED (i) = -1;
|
---|
1208 | }
|
---|
1209 |
|
---|
1210 | return 0;
|
---|
1211 | }
|
---|
1212 |
|
---|
1213 | /* If this is a SUBREG, we don't want to discard other SUBREGs of the same
|
---|
1214 | pseudo if they don't use overlapping words. We handle only pseudos
|
---|
1215 | here for simplicity. */
|
---|
1216 | if (code == SUBREG && GET_CODE (SUBREG_REG (x)) == REG
|
---|
1217 | && REGNO (SUBREG_REG (x)) >= FIRST_PSEUDO_REGISTER)
|
---|
1218 | {
|
---|
1219 | unsigned int i = REGNO (SUBREG_REG (x));
|
---|
1220 |
|
---|
1221 | if (REG_IN_TABLE (i) >= 0 && REG_IN_TABLE (i) != REG_TICK (i))
|
---|
1222 | {
|
---|
1223 | /* If REG_IN_TABLE (i) differs from REG_TICK (i) by one, and
|
---|
1224 | the last store to this register really stored into this
|
---|
1225 | subreg, then remove the memory of this subreg.
|
---|
1226 | Otherwise, remove any memory of the entire register and
|
---|
1227 | all its subregs from the table. */
|
---|
1228 | if (REG_TICK (i) - REG_IN_TABLE (i) > 1
|
---|
1229 | || SUBREG_TICKED (i) != REGNO (SUBREG_REG (x)))
|
---|
1230 | remove_invalid_refs (i);
|
---|
1231 | else
|
---|
1232 | remove_invalid_subreg_refs (i, SUBREG_BYTE (x), GET_MODE (x));
|
---|
1233 | }
|
---|
1234 |
|
---|
1235 | REG_IN_TABLE (i) = REG_TICK (i);
|
---|
1236 | SUBREG_TICKED (i) = REGNO (SUBREG_REG (x));
|
---|
1237 | return 0;
|
---|
1238 | }
|
---|
1239 |
|
---|
1240 | /* If X is a comparison or a COMPARE and either operand is a register
|
---|
1241 | that does not have a quantity, give it one. This is so that a later
|
---|
1242 | call to record_jump_equiv won't cause X to be assigned a different
|
---|
1243 | hash code and not found in the table after that call.
|
---|
1244 |
|
---|
1245 | It is not necessary to do this here, since rehash_using_reg can
|
---|
1246 | fix up the table later, but doing this here eliminates the need to
|
---|
1247 | call that expensive function in the most common case where the only
|
---|
1248 | use of the register is in the comparison. */
|
---|
1249 |
|
---|
1250 | if (code == COMPARE || GET_RTX_CLASS (code) == '<')
|
---|
1251 | {
|
---|
1252 | if (GET_CODE (XEXP (x, 0)) == REG
|
---|
1253 | && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 0))))
|
---|
1254 | if (insert_regs (XEXP (x, 0), NULL, 0))
|
---|
1255 | {
|
---|
1256 | rehash_using_reg (XEXP (x, 0));
|
---|
1257 | changed = 1;
|
---|
1258 | }
|
---|
1259 |
|
---|
1260 | if (GET_CODE (XEXP (x, 1)) == REG
|
---|
1261 | && ! REGNO_QTY_VALID_P (REGNO (XEXP (x, 1))))
|
---|
1262 | if (insert_regs (XEXP (x, 1), NULL, 0))
|
---|
1263 | {
|
---|
1264 | rehash_using_reg (XEXP (x, 1));
|
---|
1265 | changed = 1;
|
---|
1266 | }
|
---|
1267 | }
|
---|
1268 |
|
---|
1269 | fmt = GET_RTX_FORMAT (code);
|
---|
1270 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
1271 | if (fmt[i] == 'e')
|
---|
1272 | changed |= mention_regs (XEXP (x, i));
|
---|
1273 | else if (fmt[i] == 'E')
|
---|
1274 | for (j = 0; j < XVECLEN (x, i); j++)
|
---|
1275 | changed |= mention_regs (XVECEXP (x, i, j));
|
---|
1276 |
|
---|
1277 | return changed;
|
---|
1278 | }
|
---|
1279 |
|
---|
1280 | /* Update the register quantities for inserting X into the hash table
|
---|
1281 | with a value equivalent to CLASSP.
|
---|
1282 | (If the class does not contain a REG, it is irrelevant.)
|
---|
1283 | If MODIFIED is nonzero, X is a destination; it is being modified.
|
---|
1284 | Note that delete_reg_equiv should be called on a register
|
---|
1285 | before insert_regs is done on that register with MODIFIED != 0.
|
---|
1286 |
|
---|
1287 | Nonzero value means that elements of reg_qty have changed
|
---|
1288 | so X's hash code may be different. */
|
---|
1289 |
|
---|
1290 | static int
|
---|
1291 | insert_regs (x, classp, modified)
|
---|
1292 | rtx x;
|
---|
1293 | struct table_elt *classp;
|
---|
1294 | int modified;
|
---|
1295 | {
|
---|
1296 | if (GET_CODE (x) == REG)
|
---|
1297 | {
|
---|
1298 | unsigned int regno = REGNO (x);
|
---|
1299 | int qty_valid;
|
---|
1300 |
|
---|
1301 | /* If REGNO is in the equivalence table already but is of the
|
---|
1302 | wrong mode for that equivalence, don't do anything here. */
|
---|
1303 |
|
---|
1304 | qty_valid = REGNO_QTY_VALID_P (regno);
|
---|
1305 | if (qty_valid)
|
---|
1306 | {
|
---|
1307 | struct qty_table_elem *ent = &qty_table[REG_QTY (regno)];
|
---|
1308 |
|
---|
1309 | if (ent->mode != GET_MODE (x))
|
---|
1310 | return 0;
|
---|
1311 | }
|
---|
1312 |
|
---|
1313 | if (modified || ! qty_valid)
|
---|
1314 | {
|
---|
1315 | if (classp)
|
---|
1316 | for (classp = classp->first_same_value;
|
---|
1317 | classp != 0;
|
---|
1318 | classp = classp->next_same_value)
|
---|
1319 | if (GET_CODE (classp->exp) == REG
|
---|
1320 | && GET_MODE (classp->exp) == GET_MODE (x))
|
---|
1321 | {
|
---|
1322 | make_regs_eqv (regno, REGNO (classp->exp));
|
---|
1323 | return 1;
|
---|
1324 | }
|
---|
1325 |
|
---|
1326 | /* Mention_regs for a SUBREG checks if REG_TICK is exactly one larger
|
---|
1327 | than REG_IN_TABLE to find out if there was only a single preceding
|
---|
1328 | invalidation - for the SUBREG - or another one, which would be
|
---|
1329 | for the full register. However, if we find here that REG_TICK
|
---|
1330 | indicates that the register is invalid, it means that it has
|
---|
1331 | been invalidated in a separate operation. The SUBREG might be used
|
---|
1332 | now (then this is a recursive call), or we might use the full REG
|
---|
1333 | now and a SUBREG of it later. So bump up REG_TICK so that
|
---|
1334 | mention_regs will do the right thing. */
|
---|
1335 | if (! modified
|
---|
1336 | && REG_IN_TABLE (regno) >= 0
|
---|
1337 | && REG_TICK (regno) == REG_IN_TABLE (regno) + 1)
|
---|
1338 | REG_TICK (regno)++;
|
---|
1339 | make_new_qty (regno, GET_MODE (x));
|
---|
1340 | return 1;
|
---|
1341 | }
|
---|
1342 |
|
---|
1343 | return 0;
|
---|
1344 | }
|
---|
1345 |
|
---|
1346 | /* If X is a SUBREG, we will likely be inserting the inner register in the
|
---|
1347 | table. If that register doesn't have an assigned quantity number at
|
---|
1348 | this point but does later, the insertion that we will be doing now will
|
---|
1349 | not be accessible because its hash code will have changed. So assign
|
---|
1350 | a quantity number now. */
|
---|
1351 |
|
---|
1352 | else if (GET_CODE (x) == SUBREG && GET_CODE (SUBREG_REG (x)) == REG
|
---|
1353 | && ! REGNO_QTY_VALID_P (REGNO (SUBREG_REG (x))))
|
---|
1354 | {
|
---|
1355 | insert_regs (SUBREG_REG (x), NULL, 0);
|
---|
1356 | mention_regs (x);
|
---|
1357 | return 1;
|
---|
1358 | }
|
---|
1359 | else
|
---|
1360 | return mention_regs (x);
|
---|
1361 | }
|
---|
1362 | |
---|
1363 |
|
---|
1364 | /* Look in or update the hash table. */
|
---|
1365 |
|
---|
1366 | /* Remove table element ELT from use in the table.
|
---|
1367 | HASH is its hash code, made using the HASH macro.
|
---|
1368 | It's an argument because often that is known in advance
|
---|
1369 | and we save much time not recomputing it. */
|
---|
1370 |
|
---|
1371 | static void
|
---|
1372 | remove_from_table (elt, hash)
|
---|
1373 | struct table_elt *elt;
|
---|
1374 | unsigned hash;
|
---|
1375 | {
|
---|
1376 | if (elt == 0)
|
---|
1377 | return;
|
---|
1378 |
|
---|
1379 | /* Mark this element as removed. See cse_insn. */
|
---|
1380 | elt->first_same_value = 0;
|
---|
1381 |
|
---|
1382 | /* Remove the table element from its equivalence class. */
|
---|
1383 |
|
---|
1384 | {
|
---|
1385 | struct table_elt *prev = elt->prev_same_value;
|
---|
1386 | struct table_elt *next = elt->next_same_value;
|
---|
1387 |
|
---|
1388 | if (next)
|
---|
1389 | next->prev_same_value = prev;
|
---|
1390 |
|
---|
1391 | if (prev)
|
---|
1392 | prev->next_same_value = next;
|
---|
1393 | else
|
---|
1394 | {
|
---|
1395 | struct table_elt *newfirst = next;
|
---|
1396 | while (next)
|
---|
1397 | {
|
---|
1398 | next->first_same_value = newfirst;
|
---|
1399 | next = next->next_same_value;
|
---|
1400 | }
|
---|
1401 | }
|
---|
1402 | }
|
---|
1403 |
|
---|
1404 | /* Remove the table element from its hash bucket. */
|
---|
1405 |
|
---|
1406 | {
|
---|
1407 | struct table_elt *prev = elt->prev_same_hash;
|
---|
1408 | struct table_elt *next = elt->next_same_hash;
|
---|
1409 |
|
---|
1410 | if (next)
|
---|
1411 | next->prev_same_hash = prev;
|
---|
1412 |
|
---|
1413 | if (prev)
|
---|
1414 | prev->next_same_hash = next;
|
---|
1415 | else if (table[hash] == elt)
|
---|
1416 | table[hash] = next;
|
---|
1417 | else
|
---|
1418 | {
|
---|
1419 | /* This entry is not in the proper hash bucket. This can happen
|
---|
1420 | when two classes were merged by `merge_equiv_classes'. Search
|
---|
1421 | for the hash bucket that it heads. This happens only very
|
---|
1422 | rarely, so the cost is acceptable. */
|
---|
1423 | for (hash = 0; hash < HASH_SIZE; hash++)
|
---|
1424 | if (table[hash] == elt)
|
---|
1425 | table[hash] = next;
|
---|
1426 | }
|
---|
1427 | }
|
---|
1428 |
|
---|
1429 | /* Remove the table element from its related-value circular chain. */
|
---|
1430 |
|
---|
1431 | if (elt->related_value != 0 && elt->related_value != elt)
|
---|
1432 | {
|
---|
1433 | struct table_elt *p = elt->related_value;
|
---|
1434 |
|
---|
1435 | while (p->related_value != elt)
|
---|
1436 | p = p->related_value;
|
---|
1437 | p->related_value = elt->related_value;
|
---|
1438 | if (p->related_value == p)
|
---|
1439 | p->related_value = 0;
|
---|
1440 | }
|
---|
1441 |
|
---|
1442 | /* Now add it to the free element chain. */
|
---|
1443 | elt->next_same_hash = free_element_chain;
|
---|
1444 | free_element_chain = elt;
|
---|
1445 | }
|
---|
1446 |
|
---|
1447 | /* Look up X in the hash table and return its table element,
|
---|
1448 | or 0 if X is not in the table.
|
---|
1449 |
|
---|
1450 | MODE is the machine-mode of X, or if X is an integer constant
|
---|
1451 | with VOIDmode then MODE is the mode with which X will be used.
|
---|
1452 |
|
---|
1453 | Here we are satisfied to find an expression whose tree structure
|
---|
1454 | looks like X. */
|
---|
1455 |
|
---|
1456 | static struct table_elt *
|
---|
1457 | lookup (x, hash, mode)
|
---|
1458 | rtx x;
|
---|
1459 | unsigned hash;
|
---|
1460 | enum machine_mode mode;
|
---|
1461 | {
|
---|
1462 | struct table_elt *p;
|
---|
1463 |
|
---|
1464 | for (p = table[hash]; p; p = p->next_same_hash)
|
---|
1465 | if (mode == p->mode && ((x == p->exp && GET_CODE (x) == REG)
|
---|
1466 | || exp_equiv_p (x, p->exp, GET_CODE (x) != REG, 0)))
|
---|
1467 | return p;
|
---|
1468 |
|
---|
1469 | return 0;
|
---|
1470 | }
|
---|
1471 |
|
---|
1472 | /* Like `lookup' but don't care whether the table element uses invalid regs.
|
---|
1473 | Also ignore discrepancies in the machine mode of a register. */
|
---|
1474 |
|
---|
1475 | static struct table_elt *
|
---|
1476 | lookup_for_remove (x, hash, mode)
|
---|
1477 | rtx x;
|
---|
1478 | unsigned hash;
|
---|
1479 | enum machine_mode mode;
|
---|
1480 | {
|
---|
1481 | struct table_elt *p;
|
---|
1482 |
|
---|
1483 | if (GET_CODE (x) == REG)
|
---|
1484 | {
|
---|
1485 | unsigned int regno = REGNO (x);
|
---|
1486 |
|
---|
1487 | /* Don't check the machine mode when comparing registers;
|
---|
1488 | invalidating (REG:SI 0) also invalidates (REG:DF 0). */
|
---|
1489 | for (p = table[hash]; p; p = p->next_same_hash)
|
---|
1490 | if (GET_CODE (p->exp) == REG
|
---|
1491 | && REGNO (p->exp) == regno)
|
---|
1492 | return p;
|
---|
1493 | }
|
---|
1494 | else
|
---|
1495 | {
|
---|
1496 | for (p = table[hash]; p; p = p->next_same_hash)
|
---|
1497 | if (mode == p->mode && (x == p->exp || exp_equiv_p (x, p->exp, 0, 0)))
|
---|
1498 | return p;
|
---|
1499 | }
|
---|
1500 |
|
---|
1501 | return 0;
|
---|
1502 | }
|
---|
1503 |
|
---|
1504 | /* Look for an expression equivalent to X and with code CODE.
|
---|
1505 | If one is found, return that expression. */
|
---|
1506 |
|
---|
1507 | static rtx
|
---|
1508 | lookup_as_function (x, code)
|
---|
1509 | rtx x;
|
---|
1510 | enum rtx_code code;
|
---|
1511 | {
|
---|
1512 | struct table_elt *p
|
---|
1513 | = lookup (x, safe_hash (x, VOIDmode) & HASH_MASK, GET_MODE (x));
|
---|
1514 |
|
---|
1515 | /* If we are looking for a CONST_INT, the mode doesn't really matter, as
|
---|
1516 | long as we are narrowing. So if we looked in vain for a mode narrower
|
---|
1517 | than word_mode before, look for word_mode now. */
|
---|
1518 | if (p == 0 && code == CONST_INT
|
---|
1519 | && GET_MODE_SIZE (GET_MODE (x)) < GET_MODE_SIZE (word_mode))
|
---|
1520 | {
|
---|
1521 | x = copy_rtx (x);
|
---|
1522 | PUT_MODE (x, word_mode);
|
---|
1523 | p = lookup (x, safe_hash (x, VOIDmode) & HASH_MASK, word_mode);
|
---|
1524 | }
|
---|
1525 |
|
---|
1526 | if (p == 0)
|
---|
1527 | return 0;
|
---|
1528 |
|
---|
1529 | for (p = p->first_same_value; p; p = p->next_same_value)
|
---|
1530 | if (GET_CODE (p->exp) == code
|
---|
1531 | /* Make sure this is a valid entry in the table. */
|
---|
1532 | && exp_equiv_p (p->exp, p->exp, 1, 0))
|
---|
1533 | return p->exp;
|
---|
1534 |
|
---|
1535 | return 0;
|
---|
1536 | }
|
---|
1537 |
|
---|
1538 | /* Insert X in the hash table, assuming HASH is its hash code
|
---|
1539 | and CLASSP is an element of the class it should go in
|
---|
1540 | (or 0 if a new class should be made).
|
---|
1541 | It is inserted at the proper position to keep the class in
|
---|
1542 | the order cheapest first.
|
---|
1543 |
|
---|
1544 | MODE is the machine-mode of X, or if X is an integer constant
|
---|
1545 | with VOIDmode then MODE is the mode with which X will be used.
|
---|
1546 |
|
---|
1547 | For elements of equal cheapness, the most recent one
|
---|
1548 | goes in front, except that the first element in the list
|
---|
1549 | remains first unless a cheaper element is added. The order of
|
---|
1550 | pseudo-registers does not matter, as canon_reg will be called to
|
---|
1551 | find the cheapest when a register is retrieved from the table.
|
---|
1552 |
|
---|
1553 | The in_memory field in the hash table element is set to 0.
|
---|
1554 | The caller must set it nonzero if appropriate.
|
---|
1555 |
|
---|
1556 | You should call insert_regs (X, CLASSP, MODIFY) before calling here,
|
---|
1557 | and if insert_regs returns a nonzero value
|
---|
1558 | you must then recompute its hash code before calling here.
|
---|
1559 |
|
---|
1560 | If necessary, update table showing constant values of quantities. */
|
---|
1561 |
|
---|
1562 | #define CHEAPER(X, Y) \
|
---|
1563 | (preferrable ((X)->cost, (X)->regcost, (Y)->cost, (Y)->regcost) < 0)
|
---|
1564 |
|
---|
1565 | static struct table_elt *
|
---|
1566 | insert (x, classp, hash, mode)
|
---|
1567 | rtx x;
|
---|
1568 | struct table_elt *classp;
|
---|
1569 | unsigned hash;
|
---|
1570 | enum machine_mode mode;
|
---|
1571 | {
|
---|
1572 | struct table_elt *elt;
|
---|
1573 |
|
---|
1574 | /* If X is a register and we haven't made a quantity for it,
|
---|
1575 | something is wrong. */
|
---|
1576 | if (GET_CODE (x) == REG && ! REGNO_QTY_VALID_P (REGNO (x)))
|
---|
1577 | abort ();
|
---|
1578 |
|
---|
1579 | /* If X is a hard register, show it is being put in the table. */
|
---|
1580 | if (GET_CODE (x) == REG && REGNO (x) < FIRST_PSEUDO_REGISTER)
|
---|
1581 | {
|
---|
1582 | unsigned int regno = REGNO (x);
|
---|
1583 | unsigned int endregno = regno + HARD_REGNO_NREGS (regno, GET_MODE (x));
|
---|
1584 | unsigned int i;
|
---|
1585 |
|
---|
1586 | for (i = regno; i < endregno; i++)
|
---|
1587 | SET_HARD_REG_BIT (hard_regs_in_table, i);
|
---|
1588 | }
|
---|
1589 |
|
---|
1590 | /* Put an element for X into the right hash bucket. */
|
---|
1591 |
|
---|
1592 | elt = free_element_chain;
|
---|
1593 | if (elt)
|
---|
1594 | free_element_chain = elt->next_same_hash;
|
---|
1595 | else
|
---|
1596 | {
|
---|
1597 | n_elements_made++;
|
---|
1598 | elt = (struct table_elt *) xmalloc (sizeof (struct table_elt));
|
---|
1599 | }
|
---|
1600 |
|
---|
1601 | elt->exp = x;
|
---|
1602 | elt->canon_exp = NULL_RTX;
|
---|
1603 | elt->cost = COST (x);
|
---|
1604 | elt->regcost = approx_reg_cost (x);
|
---|
1605 | elt->next_same_value = 0;
|
---|
1606 | elt->prev_same_value = 0;
|
---|
1607 | elt->next_same_hash = table[hash];
|
---|
1608 | elt->prev_same_hash = 0;
|
---|
1609 | elt->related_value = 0;
|
---|
1610 | elt->in_memory = 0;
|
---|
1611 | elt->mode = mode;
|
---|
1612 | elt->is_const = (CONSTANT_P (x)
|
---|
1613 | /* GNU C++ takes advantage of this for `this'
|
---|
1614 | (and other const values). */
|
---|
1615 | || (GET_CODE (x) == REG
|
---|
1616 | && RTX_UNCHANGING_P (x)
|
---|
1617 | && REGNO (x) >= FIRST_PSEUDO_REGISTER)
|
---|
1618 | || FIXED_BASE_PLUS_P (x));
|
---|
1619 |
|
---|
1620 | if (table[hash])
|
---|
1621 | table[hash]->prev_same_hash = elt;
|
---|
1622 | table[hash] = elt;
|
---|
1623 |
|
---|
1624 | /* Put it into the proper value-class. */
|
---|
1625 | if (classp)
|
---|
1626 | {
|
---|
1627 | classp = classp->first_same_value;
|
---|
1628 | if (CHEAPER (elt, classp))
|
---|
1629 | /* Insert at the head of the class */
|
---|
1630 | {
|
---|
1631 | struct table_elt *p;
|
---|
1632 | elt->next_same_value = classp;
|
---|
1633 | classp->prev_same_value = elt;
|
---|
1634 | elt->first_same_value = elt;
|
---|
1635 |
|
---|
1636 | for (p = classp; p; p = p->next_same_value)
|
---|
1637 | p->first_same_value = elt;
|
---|
1638 | }
|
---|
1639 | else
|
---|
1640 | {
|
---|
1641 | /* Insert not at head of the class. */
|
---|
1642 | /* Put it after the last element cheaper than X. */
|
---|
1643 | struct table_elt *p, *next;
|
---|
1644 |
|
---|
1645 | for (p = classp; (next = p->next_same_value) && CHEAPER (next, elt);
|
---|
1646 | p = next);
|
---|
1647 |
|
---|
1648 | /* Put it after P and before NEXT. */
|
---|
1649 | elt->next_same_value = next;
|
---|
1650 | if (next)
|
---|
1651 | next->prev_same_value = elt;
|
---|
1652 |
|
---|
1653 | elt->prev_same_value = p;
|
---|
1654 | p->next_same_value = elt;
|
---|
1655 | elt->first_same_value = classp;
|
---|
1656 | }
|
---|
1657 | }
|
---|
1658 | else
|
---|
1659 | elt->first_same_value = elt;
|
---|
1660 |
|
---|
1661 | /* If this is a constant being set equivalent to a register or a register
|
---|
1662 | being set equivalent to a constant, note the constant equivalence.
|
---|
1663 |
|
---|
1664 | If this is a constant, it cannot be equivalent to a different constant,
|
---|
1665 | and a constant is the only thing that can be cheaper than a register. So
|
---|
1666 | we know the register is the head of the class (before the constant was
|
---|
1667 | inserted).
|
---|
1668 |
|
---|
1669 | If this is a register that is not already known equivalent to a
|
---|
1670 | constant, we must check the entire class.
|
---|
1671 |
|
---|
1672 | If this is a register that is already known equivalent to an insn,
|
---|
1673 | update the qtys `const_insn' to show that `this_insn' is the latest
|
---|
1674 | insn making that quantity equivalent to the constant. */
|
---|
1675 |
|
---|
1676 | if (elt->is_const && classp && GET_CODE (classp->exp) == REG
|
---|
1677 | && GET_CODE (x) != REG)
|
---|
1678 | {
|
---|
1679 | int exp_q = REG_QTY (REGNO (classp->exp));
|
---|
1680 | struct qty_table_elem *exp_ent = &qty_table[exp_q];
|
---|
1681 |
|
---|
1682 | exp_ent->const_rtx = gen_lowpart_if_possible (exp_ent->mode, x);
|
---|
1683 | exp_ent->const_insn = this_insn;
|
---|
1684 | }
|
---|
1685 |
|
---|
1686 | else if (GET_CODE (x) == REG
|
---|
1687 | && classp
|
---|
1688 | && ! qty_table[REG_QTY (REGNO (x))].const_rtx
|
---|
1689 | && ! elt->is_const)
|
---|
1690 | {
|
---|
1691 | struct table_elt *p;
|
---|
1692 |
|
---|
1693 | for (p = classp; p != 0; p = p->next_same_value)
|
---|
1694 | {
|
---|
1695 | if (p->is_const && GET_CODE (p->exp) != REG)
|
---|
1696 | {
|
---|
1697 | int x_q = REG_QTY (REGNO (x));
|
---|
1698 | struct qty_table_elem *x_ent = &qty_table[x_q];
|
---|
1699 |
|
---|
1700 | x_ent->const_rtx
|
---|
1701 | = gen_lowpart_if_possible (GET_MODE (x), p->exp);
|
---|
1702 | x_ent->const_insn = this_insn;
|
---|
1703 | break;
|
---|
1704 | }
|
---|
1705 | }
|
---|
1706 | }
|
---|
1707 |
|
---|
1708 | else if (GET_CODE (x) == REG
|
---|
1709 | && qty_table[REG_QTY (REGNO (x))].const_rtx
|
---|
1710 | && GET_MODE (x) == qty_table[REG_QTY (REGNO (x))].mode)
|
---|
1711 | qty_table[REG_QTY (REGNO (x))].const_insn = this_insn;
|
---|
1712 |
|
---|
1713 | /* If this is a constant with symbolic value,
|
---|
1714 | and it has a term with an explicit integer value,
|
---|
1715 | link it up with related expressions. */
|
---|
1716 | if (GET_CODE (x) == CONST)
|
---|
1717 | {
|
---|
1718 | rtx subexp = get_related_value (x);
|
---|
1719 | unsigned subhash;
|
---|
1720 | struct table_elt *subelt, *subelt_prev;
|
---|
1721 |
|
---|
1722 | if (subexp != 0)
|
---|
1723 | {
|
---|
1724 | /* Get the integer-free subexpression in the hash table. */
|
---|
1725 | subhash = safe_hash (subexp, mode) & HASH_MASK;
|
---|
1726 | subelt = lookup (subexp, subhash, mode);
|
---|
1727 | if (subelt == 0)
|
---|
1728 | subelt = insert (subexp, NULL, subhash, mode);
|
---|
1729 | /* Initialize SUBELT's circular chain if it has none. */
|
---|
1730 | if (subelt->related_value == 0)
|
---|
1731 | subelt->related_value = subelt;
|
---|
1732 | /* Find the element in the circular chain that precedes SUBELT. */
|
---|
1733 | subelt_prev = subelt;
|
---|
1734 | while (subelt_prev->related_value != subelt)
|
---|
1735 | subelt_prev = subelt_prev->related_value;
|
---|
1736 | /* Put new ELT into SUBELT's circular chain just before SUBELT.
|
---|
1737 | This way the element that follows SUBELT is the oldest one. */
|
---|
1738 | elt->related_value = subelt_prev->related_value;
|
---|
1739 | subelt_prev->related_value = elt;
|
---|
1740 | }
|
---|
1741 | }
|
---|
1742 |
|
---|
1743 | return elt;
|
---|
1744 | }
|
---|
1745 | |
---|
1746 |
|
---|
1747 | /* Given two equivalence classes, CLASS1 and CLASS2, put all the entries from
|
---|
1748 | CLASS2 into CLASS1. This is done when we have reached an insn which makes
|
---|
1749 | the two classes equivalent.
|
---|
1750 |
|
---|
1751 | CLASS1 will be the surviving class; CLASS2 should not be used after this
|
---|
1752 | call.
|
---|
1753 |
|
---|
1754 | Any invalid entries in CLASS2 will not be copied. */
|
---|
1755 |
|
---|
1756 | static void
|
---|
1757 | merge_equiv_classes (class1, class2)
|
---|
1758 | struct table_elt *class1, *class2;
|
---|
1759 | {
|
---|
1760 | struct table_elt *elt, *next, *new;
|
---|
1761 |
|
---|
1762 | /* Ensure we start with the head of the classes. */
|
---|
1763 | class1 = class1->first_same_value;
|
---|
1764 | class2 = class2->first_same_value;
|
---|
1765 |
|
---|
1766 | /* If they were already equal, forget it. */
|
---|
1767 | if (class1 == class2)
|
---|
1768 | return;
|
---|
1769 |
|
---|
1770 | for (elt = class2; elt; elt = next)
|
---|
1771 | {
|
---|
1772 | unsigned int hash;
|
---|
1773 | rtx exp = elt->exp;
|
---|
1774 | enum machine_mode mode = elt->mode;
|
---|
1775 |
|
---|
1776 | next = elt->next_same_value;
|
---|
1777 |
|
---|
1778 | /* Remove old entry, make a new one in CLASS1's class.
|
---|
1779 | Don't do this for invalid entries as we cannot find their
|
---|
1780 | hash code (it also isn't necessary). */
|
---|
1781 | if (GET_CODE (exp) == REG || exp_equiv_p (exp, exp, 1, 0))
|
---|
1782 | {
|
---|
1783 | hash_arg_in_memory = 0;
|
---|
1784 | hash = HASH (exp, mode);
|
---|
1785 |
|
---|
1786 | if (GET_CODE (exp) == REG)
|
---|
1787 | delete_reg_equiv (REGNO (exp));
|
---|
1788 |
|
---|
1789 | remove_from_table (elt, hash);
|
---|
1790 |
|
---|
1791 | if (insert_regs (exp, class1, 0))
|
---|
1792 | {
|
---|
1793 | rehash_using_reg (exp);
|
---|
1794 | hash = HASH (exp, mode);
|
---|
1795 | }
|
---|
1796 | new = insert (exp, class1, hash, mode);
|
---|
1797 | new->in_memory = hash_arg_in_memory;
|
---|
1798 | }
|
---|
1799 | }
|
---|
1800 | }
|
---|
1801 | |
---|
1802 |
|
---|
1803 | /* Flush the entire hash table. */
|
---|
1804 |
|
---|
1805 | static void
|
---|
1806 | flush_hash_table ()
|
---|
1807 | {
|
---|
1808 | int i;
|
---|
1809 | struct table_elt *p;
|
---|
1810 |
|
---|
1811 | for (i = 0; i < HASH_SIZE; i++)
|
---|
1812 | for (p = table[i]; p; p = table[i])
|
---|
1813 | {
|
---|
1814 | /* Note that invalidate can remove elements
|
---|
1815 | after P in the current hash chain. */
|
---|
1816 | if (GET_CODE (p->exp) == REG)
|
---|
1817 | invalidate (p->exp, p->mode);
|
---|
1818 | else
|
---|
1819 | remove_from_table (p, i);
|
---|
1820 | }
|
---|
1821 | }
|
---|
1822 | |
---|
1823 |
|
---|
1824 | /* Function called for each rtx to check whether true dependence exist. */
|
---|
1825 | struct check_dependence_data
|
---|
1826 | {
|
---|
1827 | enum machine_mode mode;
|
---|
1828 | rtx exp;
|
---|
1829 | };
|
---|
1830 |
|
---|
1831 | static int
|
---|
1832 | check_dependence (x, data)
|
---|
1833 | rtx *x;
|
---|
1834 | void *data;
|
---|
1835 | {
|
---|
1836 | struct check_dependence_data *d = (struct check_dependence_data *) data;
|
---|
1837 | if (*x && GET_CODE (*x) == MEM)
|
---|
1838 | return true_dependence (d->exp, d->mode, *x, cse_rtx_varies_p);
|
---|
1839 | else
|
---|
1840 | return 0;
|
---|
1841 | }
|
---|
1842 | |
---|
1843 |
|
---|
1844 | /* Remove from the hash table, or mark as invalid, all expressions whose
|
---|
1845 | values could be altered by storing in X. X is a register, a subreg, or
|
---|
1846 | a memory reference with nonvarying address (because, when a memory
|
---|
1847 | reference with a varying address is stored in, all memory references are
|
---|
1848 | removed by invalidate_memory so specific invalidation is superfluous).
|
---|
1849 | FULL_MODE, if not VOIDmode, indicates that this much should be
|
---|
1850 | invalidated instead of just the amount indicated by the mode of X. This
|
---|
1851 | is only used for bitfield stores into memory.
|
---|
1852 |
|
---|
1853 | A nonvarying address may be just a register or just a symbol reference,
|
---|
1854 | or it may be either of those plus a numeric offset. */
|
---|
1855 |
|
---|
1856 | static void
|
---|
1857 | invalidate (x, full_mode)
|
---|
1858 | rtx x;
|
---|
1859 | enum machine_mode full_mode;
|
---|
1860 | {
|
---|
1861 | int i;
|
---|
1862 | struct table_elt *p;
|
---|
1863 |
|
---|
1864 | switch (GET_CODE (x))
|
---|
1865 | {
|
---|
1866 | case REG:
|
---|
1867 | {
|
---|
1868 | /* If X is a register, dependencies on its contents are recorded
|
---|
1869 | through the qty number mechanism. Just change the qty number of
|
---|
1870 | the register, mark it as invalid for expressions that refer to it,
|
---|
1871 | and remove it itself. */
|
---|
1872 | unsigned int regno = REGNO (x);
|
---|
1873 | unsigned int hash = HASH (x, GET_MODE (x));
|
---|
1874 |
|
---|
1875 | /* Remove REGNO from any quantity list it might be on and indicate
|
---|
1876 | that its value might have changed. If it is a pseudo, remove its
|
---|
1877 | entry from the hash table.
|
---|
1878 |
|
---|
1879 | For a hard register, we do the first two actions above for any
|
---|
1880 | additional hard registers corresponding to X. Then, if any of these
|
---|
1881 | registers are in the table, we must remove any REG entries that
|
---|
1882 | overlap these registers. */
|
---|
1883 |
|
---|
1884 | delete_reg_equiv (regno);
|
---|
1885 | REG_TICK (regno)++;
|
---|
1886 | SUBREG_TICKED (regno) = -1;
|
---|
1887 |
|
---|
1888 | if (regno >= FIRST_PSEUDO_REGISTER)
|
---|
1889 | {
|
---|
1890 | /* Because a register can be referenced in more than one mode,
|
---|
1891 | we might have to remove more than one table entry. */
|
---|
1892 | struct table_elt *elt;
|
---|
1893 |
|
---|
1894 | while ((elt = lookup_for_remove (x, hash, GET_MODE (x))))
|
---|
1895 | remove_from_table (elt, hash);
|
---|
1896 | }
|
---|
1897 | else
|
---|
1898 | {
|
---|
1899 | HOST_WIDE_INT in_table
|
---|
1900 | = TEST_HARD_REG_BIT (hard_regs_in_table, regno);
|
---|
1901 | unsigned int endregno
|
---|
1902 | = regno + HARD_REGNO_NREGS (regno, GET_MODE (x));
|
---|
1903 | unsigned int tregno, tendregno, rn;
|
---|
1904 | struct table_elt *p, *next;
|
---|
1905 |
|
---|
1906 | CLEAR_HARD_REG_BIT (hard_regs_in_table, regno);
|
---|
1907 |
|
---|
1908 | for (rn = regno + 1; rn < endregno; rn++)
|
---|
1909 | {
|
---|
1910 | in_table |= TEST_HARD_REG_BIT (hard_regs_in_table, rn);
|
---|
1911 | CLEAR_HARD_REG_BIT (hard_regs_in_table, rn);
|
---|
1912 | delete_reg_equiv (rn);
|
---|
1913 | REG_TICK (rn)++;
|
---|
1914 | SUBREG_TICKED (rn) = -1;
|
---|
1915 | }
|
---|
1916 |
|
---|
1917 | if (in_table)
|
---|
1918 | for (hash = 0; hash < HASH_SIZE; hash++)
|
---|
1919 | for (p = table[hash]; p; p = next)
|
---|
1920 | {
|
---|
1921 | next = p->next_same_hash;
|
---|
1922 |
|
---|
1923 | if (GET_CODE (p->exp) != REG
|
---|
1924 | || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
|
---|
1925 | continue;
|
---|
1926 |
|
---|
1927 | tregno = REGNO (p->exp);
|
---|
1928 | tendregno
|
---|
1929 | = tregno + HARD_REGNO_NREGS (tregno, GET_MODE (p->exp));
|
---|
1930 | if (tendregno > regno && tregno < endregno)
|
---|
1931 | remove_from_table (p, hash);
|
---|
1932 | }
|
---|
1933 | }
|
---|
1934 | }
|
---|
1935 | return;
|
---|
1936 |
|
---|
1937 | case SUBREG:
|
---|
1938 | invalidate (SUBREG_REG (x), VOIDmode);
|
---|
1939 | return;
|
---|
1940 |
|
---|
1941 | case PARALLEL:
|
---|
1942 | for (i = XVECLEN (x, 0) - 1; i >= 0; --i)
|
---|
1943 | invalidate (XVECEXP (x, 0, i), VOIDmode);
|
---|
1944 | return;
|
---|
1945 |
|
---|
1946 | case EXPR_LIST:
|
---|
1947 | /* This is part of a disjoint return value; extract the location in
|
---|
1948 | question ignoring the offset. */
|
---|
1949 | invalidate (XEXP (x, 0), VOIDmode);
|
---|
1950 | return;
|
---|
1951 |
|
---|
1952 | case MEM:
|
---|
1953 | /* Calculate the canonical version of X here so that
|
---|
1954 | true_dependence doesn't generate new RTL for X on each call. */
|
---|
1955 | x = canon_rtx (x);
|
---|
1956 |
|
---|
1957 | /* Remove all hash table elements that refer to overlapping pieces of
|
---|
1958 | memory. */
|
---|
1959 | if (full_mode == VOIDmode)
|
---|
1960 | full_mode = GET_MODE (x);
|
---|
1961 |
|
---|
1962 | for (i = 0; i < HASH_SIZE; i++)
|
---|
1963 | {
|
---|
1964 | struct table_elt *next;
|
---|
1965 |
|
---|
1966 | for (p = table[i]; p; p = next)
|
---|
1967 | {
|
---|
1968 | next = p->next_same_hash;
|
---|
1969 | if (p->in_memory)
|
---|
1970 | {
|
---|
1971 | struct check_dependence_data d;
|
---|
1972 |
|
---|
1973 | /* Just canonicalize the expression once;
|
---|
1974 | otherwise each time we call invalidate
|
---|
1975 | true_dependence will canonicalize the
|
---|
1976 | expression again. */
|
---|
1977 | if (!p->canon_exp)
|
---|
1978 | p->canon_exp = canon_rtx (p->exp);
|
---|
1979 | d.exp = x;
|
---|
1980 | d.mode = full_mode;
|
---|
1981 | if (for_each_rtx (&p->canon_exp, check_dependence, &d))
|
---|
1982 | remove_from_table (p, i);
|
---|
1983 | }
|
---|
1984 | }
|
---|
1985 | }
|
---|
1986 | return;
|
---|
1987 |
|
---|
1988 | default:
|
---|
1989 | abort ();
|
---|
1990 | }
|
---|
1991 | }
|
---|
1992 | |
---|
1993 |
|
---|
1994 | /* Remove all expressions that refer to register REGNO,
|
---|
1995 | since they are already invalid, and we are about to
|
---|
1996 | mark that register valid again and don't want the old
|
---|
1997 | expressions to reappear as valid. */
|
---|
1998 |
|
---|
1999 | static void
|
---|
2000 | remove_invalid_refs (regno)
|
---|
2001 | unsigned int regno;
|
---|
2002 | {
|
---|
2003 | unsigned int i;
|
---|
2004 | struct table_elt *p, *next;
|
---|
2005 |
|
---|
2006 | for (i = 0; i < HASH_SIZE; i++)
|
---|
2007 | for (p = table[i]; p; p = next)
|
---|
2008 | {
|
---|
2009 | next = p->next_same_hash;
|
---|
2010 | if (GET_CODE (p->exp) != REG
|
---|
2011 | && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
|
---|
2012 | remove_from_table (p, i);
|
---|
2013 | }
|
---|
2014 | }
|
---|
2015 |
|
---|
2016 | /* Likewise for a subreg with subreg_reg REGNO, subreg_byte OFFSET,
|
---|
2017 | and mode MODE. */
|
---|
2018 | static void
|
---|
2019 | remove_invalid_subreg_refs (regno, offset, mode)
|
---|
2020 | unsigned int regno;
|
---|
2021 | unsigned int offset;
|
---|
2022 | enum machine_mode mode;
|
---|
2023 | {
|
---|
2024 | unsigned int i;
|
---|
2025 | struct table_elt *p, *next;
|
---|
2026 | unsigned int end = offset + (GET_MODE_SIZE (mode) - 1);
|
---|
2027 |
|
---|
2028 | for (i = 0; i < HASH_SIZE; i++)
|
---|
2029 | for (p = table[i]; p; p = next)
|
---|
2030 | {
|
---|
2031 | rtx exp = p->exp;
|
---|
2032 | next = p->next_same_hash;
|
---|
2033 |
|
---|
2034 | if (GET_CODE (exp) != REG
|
---|
2035 | && (GET_CODE (exp) != SUBREG
|
---|
2036 | || GET_CODE (SUBREG_REG (exp)) != REG
|
---|
2037 | || REGNO (SUBREG_REG (exp)) != regno
|
---|
2038 | || (((SUBREG_BYTE (exp)
|
---|
2039 | + (GET_MODE_SIZE (GET_MODE (exp)) - 1)) >= offset)
|
---|
2040 | && SUBREG_BYTE (exp) <= end))
|
---|
2041 | && refers_to_regno_p (regno, regno + 1, p->exp, (rtx *) 0))
|
---|
2042 | remove_from_table (p, i);
|
---|
2043 | }
|
---|
2044 | }
|
---|
2045 | |
---|
2046 |
|
---|
2047 | /* Recompute the hash codes of any valid entries in the hash table that
|
---|
2048 | reference X, if X is a register, or SUBREG_REG (X) if X is a SUBREG.
|
---|
2049 |
|
---|
2050 | This is called when we make a jump equivalence. */
|
---|
2051 |
|
---|
2052 | static void
|
---|
2053 | rehash_using_reg (x)
|
---|
2054 | rtx x;
|
---|
2055 | {
|
---|
2056 | unsigned int i;
|
---|
2057 | struct table_elt *p, *next;
|
---|
2058 | unsigned hash;
|
---|
2059 |
|
---|
2060 | if (GET_CODE (x) == SUBREG)
|
---|
2061 | x = SUBREG_REG (x);
|
---|
2062 |
|
---|
2063 | /* If X is not a register or if the register is known not to be in any
|
---|
2064 | valid entries in the table, we have no work to do. */
|
---|
2065 |
|
---|
2066 | if (GET_CODE (x) != REG
|
---|
2067 | || REG_IN_TABLE (REGNO (x)) < 0
|
---|
2068 | || REG_IN_TABLE (REGNO (x)) != REG_TICK (REGNO (x)))
|
---|
2069 | return;
|
---|
2070 |
|
---|
2071 | /* Scan all hash chains looking for valid entries that mention X.
|
---|
2072 | If we find one and it is in the wrong hash chain, move it. We can skip
|
---|
2073 | objects that are registers, since they are handled specially. */
|
---|
2074 |
|
---|
2075 | for (i = 0; i < HASH_SIZE; i++)
|
---|
2076 | for (p = table[i]; p; p = next)
|
---|
2077 | {
|
---|
2078 | next = p->next_same_hash;
|
---|
2079 | if (GET_CODE (p->exp) != REG && reg_mentioned_p (x, p->exp)
|
---|
2080 | && exp_equiv_p (p->exp, p->exp, 1, 0)
|
---|
2081 | && i != (hash = safe_hash (p->exp, p->mode) & HASH_MASK))
|
---|
2082 | {
|
---|
2083 | if (p->next_same_hash)
|
---|
2084 | p->next_same_hash->prev_same_hash = p->prev_same_hash;
|
---|
2085 |
|
---|
2086 | if (p->prev_same_hash)
|
---|
2087 | p->prev_same_hash->next_same_hash = p->next_same_hash;
|
---|
2088 | else
|
---|
2089 | table[i] = p->next_same_hash;
|
---|
2090 |
|
---|
2091 | p->next_same_hash = table[hash];
|
---|
2092 | p->prev_same_hash = 0;
|
---|
2093 | if (table[hash])
|
---|
2094 | table[hash]->prev_same_hash = p;
|
---|
2095 | table[hash] = p;
|
---|
2096 | }
|
---|
2097 | }
|
---|
2098 | }
|
---|
2099 | |
---|
2100 |
|
---|
2101 | /* Remove from the hash table any expression that is a call-clobbered
|
---|
2102 | register. Also update their TICK values. */
|
---|
2103 |
|
---|
2104 | static void
|
---|
2105 | invalidate_for_call ()
|
---|
2106 | {
|
---|
2107 | unsigned int regno, endregno;
|
---|
2108 | unsigned int i;
|
---|
2109 | unsigned hash;
|
---|
2110 | struct table_elt *p, *next;
|
---|
2111 | int in_table = 0;
|
---|
2112 |
|
---|
2113 | /* Go through all the hard registers. For each that is clobbered in
|
---|
2114 | a CALL_INSN, remove the register from quantity chains and update
|
---|
2115 | reg_tick if defined. Also see if any of these registers is currently
|
---|
2116 | in the table. */
|
---|
2117 |
|
---|
2118 | for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
---|
2119 | if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
---|
2120 | {
|
---|
2121 | delete_reg_equiv (regno);
|
---|
2122 | if (REG_TICK (regno) >= 0)
|
---|
2123 | {
|
---|
2124 | REG_TICK (regno)++;
|
---|
2125 | SUBREG_TICKED (regno) = -1;
|
---|
2126 | }
|
---|
2127 |
|
---|
2128 | in_table |= (TEST_HARD_REG_BIT (hard_regs_in_table, regno) != 0);
|
---|
2129 | }
|
---|
2130 |
|
---|
2131 | /* In the case where we have no call-clobbered hard registers in the
|
---|
2132 | table, we are done. Otherwise, scan the table and remove any
|
---|
2133 | entry that overlaps a call-clobbered register. */
|
---|
2134 |
|
---|
2135 | if (in_table)
|
---|
2136 | for (hash = 0; hash < HASH_SIZE; hash++)
|
---|
2137 | for (p = table[hash]; p; p = next)
|
---|
2138 | {
|
---|
2139 | next = p->next_same_hash;
|
---|
2140 |
|
---|
2141 | if (GET_CODE (p->exp) != REG
|
---|
2142 | || REGNO (p->exp) >= FIRST_PSEUDO_REGISTER)
|
---|
2143 | continue;
|
---|
2144 |
|
---|
2145 | regno = REGNO (p->exp);
|
---|
2146 | endregno = regno + HARD_REGNO_NREGS (regno, GET_MODE (p->exp));
|
---|
2147 |
|
---|
2148 | for (i = regno; i < endregno; i++)
|
---|
2149 | if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
|
---|
2150 | {
|
---|
2151 | remove_from_table (p, hash);
|
---|
2152 | break;
|
---|
2153 | }
|
---|
2154 | }
|
---|
2155 | }
|
---|
2156 | |
---|
2157 |
|
---|
2158 | /* Given an expression X of type CONST,
|
---|
2159 | and ELT which is its table entry (or 0 if it
|
---|
2160 | is not in the hash table),
|
---|
2161 | return an alternate expression for X as a register plus integer.
|
---|
2162 | If none can be found, return 0. */
|
---|
2163 |
|
---|
2164 | static rtx
|
---|
2165 | use_related_value (x, elt)
|
---|
2166 | rtx x;
|
---|
2167 | struct table_elt *elt;
|
---|
2168 | {
|
---|
2169 | struct table_elt *relt = 0;
|
---|
2170 | struct table_elt *p, *q;
|
---|
2171 | HOST_WIDE_INT offset;
|
---|
2172 |
|
---|
2173 | /* First, is there anything related known?
|
---|
2174 | If we have a table element, we can tell from that.
|
---|
2175 | Otherwise, must look it up. */
|
---|
2176 |
|
---|
2177 | if (elt != 0 && elt->related_value != 0)
|
---|
2178 | relt = elt;
|
---|
2179 | else if (elt == 0 && GET_CODE (x) == CONST)
|
---|
2180 | {
|
---|
2181 | rtx subexp = get_related_value (x);
|
---|
2182 | if (subexp != 0)
|
---|
2183 | relt = lookup (subexp,
|
---|
2184 | safe_hash (subexp, GET_MODE (subexp)) & HASH_MASK,
|
---|
2185 | GET_MODE (subexp));
|
---|
2186 | }
|
---|
2187 |
|
---|
2188 | if (relt == 0)
|
---|
2189 | return 0;
|
---|
2190 |
|
---|
2191 | /* Search all related table entries for one that has an
|
---|
2192 | equivalent register. */
|
---|
2193 |
|
---|
2194 | p = relt;
|
---|
2195 | while (1)
|
---|
2196 | {
|
---|
2197 | /* This loop is strange in that it is executed in two different cases.
|
---|
2198 | The first is when X is already in the table. Then it is searching
|
---|
2199 | the RELATED_VALUE list of X's class (RELT). The second case is when
|
---|
2200 | X is not in the table. Then RELT points to a class for the related
|
---|
2201 | value.
|
---|
2202 |
|
---|
2203 | Ensure that, whatever case we are in, that we ignore classes that have
|
---|
2204 | the same value as X. */
|
---|
2205 |
|
---|
2206 | if (rtx_equal_p (x, p->exp))
|
---|
2207 | q = 0;
|
---|
2208 | else
|
---|
2209 | for (q = p->first_same_value; q; q = q->next_same_value)
|
---|
2210 | if (GET_CODE (q->exp) == REG)
|
---|
2211 | break;
|
---|
2212 |
|
---|
2213 | if (q)
|
---|
2214 | break;
|
---|
2215 |
|
---|
2216 | p = p->related_value;
|
---|
2217 |
|
---|
2218 | /* We went all the way around, so there is nothing to be found.
|
---|
2219 | Alternatively, perhaps RELT was in the table for some other reason
|
---|
2220 | and it has no related values recorded. */
|
---|
2221 | if (p == relt || p == 0)
|
---|
2222 | break;
|
---|
2223 | }
|
---|
2224 |
|
---|
2225 | if (q == 0)
|
---|
2226 | return 0;
|
---|
2227 |
|
---|
2228 | offset = (get_integer_term (x) - get_integer_term (p->exp));
|
---|
2229 | /* Note: OFFSET may be 0 if P->xexp and X are related by commutativity. */
|
---|
2230 | return plus_constant (q->exp, offset);
|
---|
2231 | }
|
---|
2232 | |
---|
2233 |
|
---|
2234 | /* Hash a string. Just add its bytes up. */
|
---|
2235 | static inline unsigned
|
---|
2236 | canon_hash_string (ps)
|
---|
2237 | const char *ps;
|
---|
2238 | {
|
---|
2239 | unsigned hash = 0;
|
---|
2240 | const unsigned char *p = (const unsigned char *) ps;
|
---|
2241 |
|
---|
2242 | if (p)
|
---|
2243 | while (*p)
|
---|
2244 | hash += *p++;
|
---|
2245 |
|
---|
2246 | return hash;
|
---|
2247 | }
|
---|
2248 |
|
---|
2249 | /* Hash an rtx. We are careful to make sure the value is never negative.
|
---|
2250 | Equivalent registers hash identically.
|
---|
2251 | MODE is used in hashing for CONST_INTs only;
|
---|
2252 | otherwise the mode of X is used.
|
---|
2253 |
|
---|
2254 | Store 1 in do_not_record if any subexpression is volatile.
|
---|
2255 |
|
---|
2256 | Store 1 in hash_arg_in_memory if X contains a MEM rtx
|
---|
2257 | which does not have the RTX_UNCHANGING_P bit set.
|
---|
2258 |
|
---|
2259 | Note that cse_insn knows that the hash code of a MEM expression
|
---|
2260 | is just (int) MEM plus the hash code of the address. */
|
---|
2261 |
|
---|
2262 | static unsigned
|
---|
2263 | canon_hash (x, mode)
|
---|
2264 | rtx x;
|
---|
2265 | enum machine_mode mode;
|
---|
2266 | {
|
---|
2267 | int i, j;
|
---|
2268 | unsigned hash = 0;
|
---|
2269 | enum rtx_code code;
|
---|
2270 | const char *fmt;
|
---|
2271 |
|
---|
2272 | /* repeat is used to turn tail-recursion into iteration. */
|
---|
2273 | repeat:
|
---|
2274 | if (x == 0)
|
---|
2275 | return hash;
|
---|
2276 |
|
---|
2277 | code = GET_CODE (x);
|
---|
2278 | switch (code)
|
---|
2279 | {
|
---|
2280 | case REG:
|
---|
2281 | {
|
---|
2282 | unsigned int regno = REGNO (x);
|
---|
2283 | bool record;
|
---|
2284 |
|
---|
2285 | /* On some machines, we can't record any non-fixed hard register,
|
---|
2286 | because extending its life will cause reload problems. We
|
---|
2287 | consider ap, fp, sp, gp to be fixed for this purpose.
|
---|
2288 |
|
---|
2289 | We also consider CCmode registers to be fixed for this purpose;
|
---|
2290 | failure to do so leads to failure to simplify 0<100 type of
|
---|
2291 | conditionals.
|
---|
2292 |
|
---|
2293 | On all machines, we can't record any global registers.
|
---|
2294 | Nor should we record any register that is in a small
|
---|
2295 | class, as defined by CLASS_LIKELY_SPILLED_P. */
|
---|
2296 |
|
---|
2297 | if (regno >= FIRST_PSEUDO_REGISTER)
|
---|
2298 | record = true;
|
---|
2299 | else if (x == frame_pointer_rtx
|
---|
2300 | || x == hard_frame_pointer_rtx
|
---|
2301 | || x == arg_pointer_rtx
|
---|
2302 | || x == stack_pointer_rtx
|
---|
2303 | || x == pic_offset_table_rtx)
|
---|
2304 | record = true;
|
---|
2305 | else if (global_regs[regno])
|
---|
2306 | record = false;
|
---|
2307 | else if (fixed_regs[regno])
|
---|
2308 | record = true;
|
---|
2309 | else if (GET_MODE_CLASS (GET_MODE (x)) == MODE_CC)
|
---|
2310 | record = true;
|
---|
2311 | else if (SMALL_REGISTER_CLASSES)
|
---|
2312 | record = false;
|
---|
2313 | else if (CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (regno)))
|
---|
2314 | record = false;
|
---|
2315 | else
|
---|
2316 | record = true;
|
---|
2317 |
|
---|
2318 | if (!record)
|
---|
2319 | {
|
---|
2320 | do_not_record = 1;
|
---|
2321 | return 0;
|
---|
2322 | }
|
---|
2323 |
|
---|
2324 | hash += ((unsigned) REG << 7) + (unsigned) REG_QTY (regno);
|
---|
2325 | return hash;
|
---|
2326 | }
|
---|
2327 |
|
---|
2328 | /* We handle SUBREG of a REG specially because the underlying
|
---|
2329 | reg changes its hash value with every value change; we don't
|
---|
2330 | want to have to forget unrelated subregs when one subreg changes. */
|
---|
2331 | case SUBREG:
|
---|
2332 | {
|
---|
2333 | if (GET_CODE (SUBREG_REG (x)) == REG)
|
---|
2334 | {
|
---|
2335 | hash += (((unsigned) SUBREG << 7)
|
---|
2336 | + REGNO (SUBREG_REG (x))
|
---|
2337 | + (SUBREG_BYTE (x) / UNITS_PER_WORD));
|
---|
2338 | return hash;
|
---|
2339 | }
|
---|
2340 | break;
|
---|
2341 | }
|
---|
2342 |
|
---|
2343 | case CONST_INT:
|
---|
2344 | {
|
---|
2345 | unsigned HOST_WIDE_INT tem = INTVAL (x);
|
---|
2346 | hash += ((unsigned) CONST_INT << 7) + (unsigned) mode + tem;
|
---|
2347 | return hash;
|
---|
2348 | }
|
---|
2349 |
|
---|
2350 | case CONST_DOUBLE:
|
---|
2351 | /* This is like the general case, except that it only counts
|
---|
2352 | the integers representing the constant. */
|
---|
2353 | hash += (unsigned) code + (unsigned) GET_MODE (x);
|
---|
2354 | if (GET_MODE (x) != VOIDmode)
|
---|
2355 | hash += real_hash (CONST_DOUBLE_REAL_VALUE (x));
|
---|
2356 | else
|
---|
2357 | hash += ((unsigned) CONST_DOUBLE_LOW (x)
|
---|
2358 | + (unsigned) CONST_DOUBLE_HIGH (x));
|
---|
2359 | return hash;
|
---|
2360 |
|
---|
2361 | case CONST_VECTOR:
|
---|
2362 | {
|
---|
2363 | int units;
|
---|
2364 | rtx elt;
|
---|
2365 |
|
---|
2366 | units = CONST_VECTOR_NUNITS (x);
|
---|
2367 |
|
---|
2368 | for (i = 0; i < units; ++i)
|
---|
2369 | {
|
---|
2370 | elt = CONST_VECTOR_ELT (x, i);
|
---|
2371 | hash += canon_hash (elt, GET_MODE (elt));
|
---|
2372 | }
|
---|
2373 |
|
---|
2374 | return hash;
|
---|
2375 | }
|
---|
2376 |
|
---|
2377 | /* Assume there is only one rtx object for any given label. */
|
---|
2378 | case LABEL_REF:
|
---|
2379 | hash += ((unsigned) LABEL_REF << 7) + (unsigned long) XEXP (x, 0);
|
---|
2380 | return hash;
|
---|
2381 |
|
---|
2382 | case SYMBOL_REF:
|
---|
2383 | hash += ((unsigned) SYMBOL_REF << 7) + (unsigned long) XSTR (x, 0);
|
---|
2384 | return hash;
|
---|
2385 |
|
---|
2386 | case MEM:
|
---|
2387 | /* We don't record if marked volatile or if BLKmode since we don't
|
---|
2388 | know the size of the move. */
|
---|
2389 | if (MEM_VOLATILE_P (x) || GET_MODE (x) == BLKmode)
|
---|
2390 | {
|
---|
2391 | do_not_record = 1;
|
---|
2392 | return 0;
|
---|
2393 | }
|
---|
2394 | if (! RTX_UNCHANGING_P (x) || FIXED_BASE_PLUS_P (XEXP (x, 0)))
|
---|
2395 | {
|
---|
2396 | hash_arg_in_memory = 1;
|
---|
2397 | }
|
---|
2398 | /* Now that we have already found this special case,
|
---|
2399 | might as well speed it up as much as possible. */
|
---|
2400 | hash += (unsigned) MEM;
|
---|
2401 | x = XEXP (x, 0);
|
---|
2402 | goto repeat;
|
---|
2403 |
|
---|
2404 | case USE:
|
---|
2405 | /* A USE that mentions non-volatile memory needs special
|
---|
2406 | handling since the MEM may be BLKmode which normally
|
---|
2407 | prevents an entry from being made. Pure calls are
|
---|
2408 | marked by a USE which mentions BLKmode memory. */
|
---|
2409 | if (GET_CODE (XEXP (x, 0)) == MEM
|
---|
2410 | && ! MEM_VOLATILE_P (XEXP (x, 0)))
|
---|
2411 | {
|
---|
2412 | hash += (unsigned) USE;
|
---|
2413 | x = XEXP (x, 0);
|
---|
2414 |
|
---|
2415 | if (! RTX_UNCHANGING_P (x) || FIXED_BASE_PLUS_P (XEXP (x, 0)))
|
---|
2416 | hash_arg_in_memory = 1;
|
---|
2417 |
|
---|
2418 | /* Now that we have already found this special case,
|
---|
2419 | might as well speed it up as much as possible. */
|
---|
2420 | hash += (unsigned) MEM;
|
---|
2421 | x = XEXP (x, 0);
|
---|
2422 | goto repeat;
|
---|
2423 | }
|
---|
2424 | break;
|
---|
2425 |
|
---|
2426 | case PRE_DEC:
|
---|
2427 | case PRE_INC:
|
---|
2428 | case POST_DEC:
|
---|
2429 | case POST_INC:
|
---|
2430 | case PRE_MODIFY:
|
---|
2431 | case POST_MODIFY:
|
---|
2432 | case PC:
|
---|
2433 | case CC0:
|
---|
2434 | case CALL:
|
---|
2435 | case UNSPEC_VOLATILE:
|
---|
2436 | do_not_record = 1;
|
---|
2437 | return 0;
|
---|
2438 |
|
---|
2439 | case ASM_OPERANDS:
|
---|
2440 | if (MEM_VOLATILE_P (x))
|
---|
2441 | {
|
---|
2442 | do_not_record = 1;
|
---|
2443 | return 0;
|
---|
2444 | }
|
---|
2445 | else
|
---|
2446 | {
|
---|
2447 | /* We don't want to take the filename and line into account. */
|
---|
2448 | hash += (unsigned) code + (unsigned) GET_MODE (x)
|
---|
2449 | + canon_hash_string (ASM_OPERANDS_TEMPLATE (x))
|
---|
2450 | + canon_hash_string (ASM_OPERANDS_OUTPUT_CONSTRAINT (x))
|
---|
2451 | + (unsigned) ASM_OPERANDS_OUTPUT_IDX (x);
|
---|
2452 |
|
---|
2453 | if (ASM_OPERANDS_INPUT_LENGTH (x))
|
---|
2454 | {
|
---|
2455 | for (i = 1; i < ASM_OPERANDS_INPUT_LENGTH (x); i++)
|
---|
2456 | {
|
---|
2457 | hash += (canon_hash (ASM_OPERANDS_INPUT (x, i),
|
---|
2458 | GET_MODE (ASM_OPERANDS_INPUT (x, i)))
|
---|
2459 | + canon_hash_string (ASM_OPERANDS_INPUT_CONSTRAINT
|
---|
2460 | (x, i)));
|
---|
2461 | }
|
---|
2462 |
|
---|
2463 | hash += canon_hash_string (ASM_OPERANDS_INPUT_CONSTRAINT (x, 0));
|
---|
2464 | x = ASM_OPERANDS_INPUT (x, 0);
|
---|
2465 | mode = GET_MODE (x);
|
---|
2466 | goto repeat;
|
---|
2467 | }
|
---|
2468 |
|
---|
2469 | return hash;
|
---|
2470 | }
|
---|
2471 | break;
|
---|
2472 |
|
---|
2473 | default:
|
---|
2474 | break;
|
---|
2475 | }
|
---|
2476 |
|
---|
2477 | i = GET_RTX_LENGTH (code) - 1;
|
---|
2478 | hash += (unsigned) code + (unsigned) GET_MODE (x);
|
---|
2479 | fmt = GET_RTX_FORMAT (code);
|
---|
2480 | for (; i >= 0; i--)
|
---|
2481 | {
|
---|
2482 | if (fmt[i] == 'e')
|
---|
2483 | {
|
---|
2484 | rtx tem = XEXP (x, i);
|
---|
2485 |
|
---|
2486 | /* If we are about to do the last recursive call
|
---|
2487 | needed at this level, change it into iteration.
|
---|
2488 | This function is called enough to be worth it. */
|
---|
2489 | if (i == 0)
|
---|
2490 | {
|
---|
2491 | x = tem;
|
---|
2492 | goto repeat;
|
---|
2493 | }
|
---|
2494 | hash += canon_hash (tem, 0);
|
---|
2495 | }
|
---|
2496 | else if (fmt[i] == 'E')
|
---|
2497 | for (j = 0; j < XVECLEN (x, i); j++)
|
---|
2498 | hash += canon_hash (XVECEXP (x, i, j), 0);
|
---|
2499 | else if (fmt[i] == 's')
|
---|
2500 | hash += canon_hash_string (XSTR (x, i));
|
---|
2501 | else if (fmt[i] == 'i')
|
---|
2502 | {
|
---|
2503 | unsigned tem = XINT (x, i);
|
---|
2504 | hash += tem;
|
---|
2505 | }
|
---|
2506 | else if (fmt[i] == '0' || fmt[i] == 't')
|
---|
2507 | /* Unused. */
|
---|
2508 | ;
|
---|
2509 | else
|
---|
2510 | abort ();
|
---|
2511 | }
|
---|
2512 | return hash;
|
---|
2513 | }
|
---|
2514 |
|
---|
2515 | /* Like canon_hash but with no side effects. */
|
---|
2516 |
|
---|
2517 | static unsigned
|
---|
2518 | safe_hash (x, mode)
|
---|
2519 | rtx x;
|
---|
2520 | enum machine_mode mode;
|
---|
2521 | {
|
---|
2522 | int save_do_not_record = do_not_record;
|
---|
2523 | int save_hash_arg_in_memory = hash_arg_in_memory;
|
---|
2524 | unsigned hash = canon_hash (x, mode);
|
---|
2525 | hash_arg_in_memory = save_hash_arg_in_memory;
|
---|
2526 | do_not_record = save_do_not_record;
|
---|
2527 | return hash;
|
---|
2528 | }
|
---|
2529 | |
---|
2530 |
|
---|
2531 | /* Return 1 iff X and Y would canonicalize into the same thing,
|
---|
2532 | without actually constructing the canonicalization of either one.
|
---|
2533 | If VALIDATE is nonzero,
|
---|
2534 | we assume X is an expression being processed from the rtl
|
---|
2535 | and Y was found in the hash table. We check register refs
|
---|
2536 | in Y for being marked as valid.
|
---|
2537 |
|
---|
2538 | If EQUAL_VALUES is nonzero, we allow a register to match a constant value
|
---|
2539 | that is known to be in the register. Ordinarily, we don't allow them
|
---|
2540 | to match, because letting them match would cause unpredictable results
|
---|
2541 | in all the places that search a hash table chain for an equivalent
|
---|
2542 | for a given value. A possible equivalent that has different structure
|
---|
2543 | has its hash code computed from different data. Whether the hash code
|
---|
2544 | is the same as that of the given value is pure luck. */
|
---|
2545 |
|
---|
2546 | static int
|
---|
2547 | exp_equiv_p (x, y, validate, equal_values)
|
---|
2548 | rtx x, y;
|
---|
2549 | int validate;
|
---|
2550 | int equal_values;
|
---|
2551 | {
|
---|
2552 | int i, j;
|
---|
2553 | enum rtx_code code;
|
---|
2554 | const char *fmt;
|
---|
2555 |
|
---|
2556 | /* Note: it is incorrect to assume an expression is equivalent to itself
|
---|
2557 | if VALIDATE is nonzero. */
|
---|
2558 | if (x == y && !validate)
|
---|
2559 | return 1;
|
---|
2560 | if (x == 0 || y == 0)
|
---|
2561 | return x == y;
|
---|
2562 |
|
---|
2563 | code = GET_CODE (x);
|
---|
2564 | if (code != GET_CODE (y))
|
---|
2565 | {
|
---|
2566 | if (!equal_values)
|
---|
2567 | return 0;
|
---|
2568 |
|
---|
2569 | /* If X is a constant and Y is a register or vice versa, they may be
|
---|
2570 | equivalent. We only have to validate if Y is a register. */
|
---|
2571 | if (CONSTANT_P (x) && GET_CODE (y) == REG
|
---|
2572 | && REGNO_QTY_VALID_P (REGNO (y)))
|
---|
2573 | {
|
---|
2574 | int y_q = REG_QTY (REGNO (y));
|
---|
2575 | struct qty_table_elem *y_ent = &qty_table[y_q];
|
---|
2576 |
|
---|
2577 | if (GET_MODE (y) == y_ent->mode
|
---|
2578 | && rtx_equal_p (x, y_ent->const_rtx)
|
---|
2579 | && (! validate || REG_IN_TABLE (REGNO (y)) == REG_TICK (REGNO (y))))
|
---|
2580 | return 1;
|
---|
2581 | }
|
---|
2582 |
|
---|
2583 | if (CONSTANT_P (y) && code == REG
|
---|
2584 | && REGNO_QTY_VALID_P (REGNO (x)))
|
---|
2585 | {
|
---|
2586 | int x_q = REG_QTY (REGNO (x));
|
---|
2587 | struct qty_table_elem *x_ent = &qty_table[x_q];
|
---|
2588 |
|
---|
2589 | if (GET_MODE (x) == x_ent->mode
|
---|
2590 | && rtx_equal_p (y, x_ent->const_rtx))
|
---|
2591 | return 1;
|
---|
2592 | }
|
---|
2593 |
|
---|
2594 | return 0;
|
---|
2595 | }
|
---|
2596 |
|
---|
2597 | /* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent. */
|
---|
2598 | if (GET_MODE (x) != GET_MODE (y))
|
---|
2599 | return 0;
|
---|
2600 |
|
---|
2601 | switch (code)
|
---|
2602 | {
|
---|
2603 | case PC:
|
---|
2604 | case CC0:
|
---|
2605 | case CONST_INT:
|
---|
2606 | return x == y;
|
---|
2607 |
|
---|
2608 | case LABEL_REF:
|
---|
2609 | return XEXP (x, 0) == XEXP (y, 0);
|
---|
2610 |
|
---|
2611 | case SYMBOL_REF:
|
---|
2612 | return XSTR (x, 0) == XSTR (y, 0);
|
---|
2613 |
|
---|
2614 | case REG:
|
---|
2615 | {
|
---|
2616 | unsigned int regno = REGNO (y);
|
---|
2617 | unsigned int endregno
|
---|
2618 | = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
|
---|
2619 | : HARD_REGNO_NREGS (regno, GET_MODE (y)));
|
---|
2620 | unsigned int i;
|
---|
2621 |
|
---|
2622 | /* If the quantities are not the same, the expressions are not
|
---|
2623 | equivalent. If there are and we are not to validate, they
|
---|
2624 | are equivalent. Otherwise, ensure all regs are up-to-date. */
|
---|
2625 |
|
---|
2626 | if (REG_QTY (REGNO (x)) != REG_QTY (regno))
|
---|
2627 | return 0;
|
---|
2628 |
|
---|
2629 | if (! validate)
|
---|
2630 | return 1;
|
---|
2631 |
|
---|
2632 | for (i = regno; i < endregno; i++)
|
---|
2633 | if (REG_IN_TABLE (i) != REG_TICK (i))
|
---|
2634 | return 0;
|
---|
2635 |
|
---|
2636 | return 1;
|
---|
2637 | }
|
---|
2638 |
|
---|
2639 | /* For commutative operations, check both orders. */
|
---|
2640 | case PLUS:
|
---|
2641 | case MULT:
|
---|
2642 | case AND:
|
---|
2643 | case IOR:
|
---|
2644 | case XOR:
|
---|
2645 | case NE:
|
---|
2646 | case EQ:
|
---|
2647 | return ((exp_equiv_p (XEXP (x, 0), XEXP (y, 0), validate, equal_values)
|
---|
2648 | && exp_equiv_p (XEXP (x, 1), XEXP (y, 1),
|
---|
2649 | validate, equal_values))
|
---|
2650 | || (exp_equiv_p (XEXP (x, 0), XEXP (y, 1),
|
---|
2651 | validate, equal_values)
|
---|
2652 | && exp_equiv_p (XEXP (x, 1), XEXP (y, 0),
|
---|
2653 | validate, equal_values)));
|
---|
2654 |
|
---|
2655 | case ASM_OPERANDS:
|
---|
2656 | /* We don't use the generic code below because we want to
|
---|
2657 | disregard filename and line numbers. */
|
---|
2658 |
|
---|
2659 | /* A volatile asm isn't equivalent to any other. */
|
---|
2660 | if (MEM_VOLATILE_P (x) || MEM_VOLATILE_P (y))
|
---|
2661 | return 0;
|
---|
2662 |
|
---|
2663 | if (GET_MODE (x) != GET_MODE (y)
|
---|
2664 | || strcmp (ASM_OPERANDS_TEMPLATE (x), ASM_OPERANDS_TEMPLATE (y))
|
---|
2665 | || strcmp (ASM_OPERANDS_OUTPUT_CONSTRAINT (x),
|
---|
2666 | ASM_OPERANDS_OUTPUT_CONSTRAINT (y))
|
---|
2667 | || ASM_OPERANDS_OUTPUT_IDX (x) != ASM_OPERANDS_OUTPUT_IDX (y)
|
---|
2668 | || ASM_OPERANDS_INPUT_LENGTH (x) != ASM_OPERANDS_INPUT_LENGTH (y))
|
---|
2669 | return 0;
|
---|
2670 |
|
---|
2671 | if (ASM_OPERANDS_INPUT_LENGTH (x))
|
---|
2672 | {
|
---|
2673 | for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
|
---|
2674 | if (! exp_equiv_p (ASM_OPERANDS_INPUT (x, i),
|
---|
2675 | ASM_OPERANDS_INPUT (y, i),
|
---|
2676 | validate, equal_values)
|
---|
2677 | || strcmp (ASM_OPERANDS_INPUT_CONSTRAINT (x, i),
|
---|
2678 | ASM_OPERANDS_INPUT_CONSTRAINT (y, i)))
|
---|
2679 | return 0;
|
---|
2680 | }
|
---|
2681 |
|
---|
2682 | return 1;
|
---|
2683 |
|
---|
2684 | default:
|
---|
2685 | break;
|
---|
2686 | }
|
---|
2687 |
|
---|
2688 | /* Compare the elements. If any pair of corresponding elements
|
---|
2689 | fail to match, return 0 for the whole things. */
|
---|
2690 |
|
---|
2691 | fmt = GET_RTX_FORMAT (code);
|
---|
2692 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
2693 | {
|
---|
2694 | switch (fmt[i])
|
---|
2695 | {
|
---|
2696 | case 'e':
|
---|
2697 | if (! exp_equiv_p (XEXP (x, i), XEXP (y, i), validate, equal_values))
|
---|
2698 | return 0;
|
---|
2699 | break;
|
---|
2700 |
|
---|
2701 | case 'E':
|
---|
2702 | if (XVECLEN (x, i) != XVECLEN (y, i))
|
---|
2703 | return 0;
|
---|
2704 | for (j = 0; j < XVECLEN (x, i); j++)
|
---|
2705 | if (! exp_equiv_p (XVECEXP (x, i, j), XVECEXP (y, i, j),
|
---|
2706 | validate, equal_values))
|
---|
2707 | return 0;
|
---|
2708 | break;
|
---|
2709 |
|
---|
2710 | case 's':
|
---|
2711 | if (strcmp (XSTR (x, i), XSTR (y, i)))
|
---|
2712 | return 0;
|
---|
2713 | break;
|
---|
2714 |
|
---|
2715 | case 'i':
|
---|
2716 | if (XINT (x, i) != XINT (y, i))
|
---|
2717 | return 0;
|
---|
2718 | break;
|
---|
2719 |
|
---|
2720 | case 'w':
|
---|
2721 | if (XWINT (x, i) != XWINT (y, i))
|
---|
2722 | return 0;
|
---|
2723 | break;
|
---|
2724 |
|
---|
2725 | case '0':
|
---|
2726 | case 't':
|
---|
2727 | break;
|
---|
2728 |
|
---|
2729 | default:
|
---|
2730 | abort ();
|
---|
2731 | }
|
---|
2732 | }
|
---|
2733 |
|
---|
2734 | return 1;
|
---|
2735 | }
|
---|
2736 | |
---|
2737 |
|
---|
2738 | /* Return 1 if X has a value that can vary even between two
|
---|
2739 | executions of the program. 0 means X can be compared reliably
|
---|
2740 | against certain constants or near-constants. */
|
---|
2741 |
|
---|
2742 | static int
|
---|
2743 | cse_rtx_varies_p (x, from_alias)
|
---|
2744 | rtx x;
|
---|
2745 | int from_alias;
|
---|
2746 | {
|
---|
2747 | /* We need not check for X and the equivalence class being of the same
|
---|
2748 | mode because if X is equivalent to a constant in some mode, it
|
---|
2749 | doesn't vary in any mode. */
|
---|
2750 |
|
---|
2751 | if (GET_CODE (x) == REG
|
---|
2752 | && REGNO_QTY_VALID_P (REGNO (x)))
|
---|
2753 | {
|
---|
2754 | int x_q = REG_QTY (REGNO (x));
|
---|
2755 | struct qty_table_elem *x_ent = &qty_table[x_q];
|
---|
2756 |
|
---|
2757 | if (GET_MODE (x) == x_ent->mode
|
---|
2758 | && x_ent->const_rtx != NULL_RTX)
|
---|
2759 | return 0;
|
---|
2760 | }
|
---|
2761 |
|
---|
2762 | if (GET_CODE (x) == PLUS
|
---|
2763 | && GET_CODE (XEXP (x, 1)) == CONST_INT
|
---|
2764 | && GET_CODE (XEXP (x, 0)) == REG
|
---|
2765 | && REGNO_QTY_VALID_P (REGNO (XEXP (x, 0))))
|
---|
2766 | {
|
---|
2767 | int x0_q = REG_QTY (REGNO (XEXP (x, 0)));
|
---|
2768 | struct qty_table_elem *x0_ent = &qty_table[x0_q];
|
---|
2769 |
|
---|
2770 | if ((GET_MODE (XEXP (x, 0)) == x0_ent->mode)
|
---|
2771 | && x0_ent->const_rtx != NULL_RTX)
|
---|
2772 | return 0;
|
---|
2773 | }
|
---|
2774 |
|
---|
2775 | /* This can happen as the result of virtual register instantiation, if
|
---|
2776 | the initial constant is too large to be a valid address. This gives
|
---|
2777 | us a three instruction sequence, load large offset into a register,
|
---|
2778 | load fp minus a constant into a register, then a MEM which is the
|
---|
2779 | sum of the two `constant' registers. */
|
---|
2780 | if (GET_CODE (x) == PLUS
|
---|
2781 | && GET_CODE (XEXP (x, 0)) == REG
|
---|
2782 | && GET_CODE (XEXP (x, 1)) == REG
|
---|
2783 | && REGNO_QTY_VALID_P (REGNO (XEXP (x, 0)))
|
---|
2784 | && REGNO_QTY_VALID_P (REGNO (XEXP (x, 1))))
|
---|
2785 | {
|
---|
2786 | int x0_q = REG_QTY (REGNO (XEXP (x, 0)));
|
---|
2787 | int x1_q = REG_QTY (REGNO (XEXP (x, 1)));
|
---|
2788 | struct qty_table_elem *x0_ent = &qty_table[x0_q];
|
---|
2789 | struct qty_table_elem *x1_ent = &qty_table[x1_q];
|
---|
2790 |
|
---|
2791 | if ((GET_MODE (XEXP (x, 0)) == x0_ent->mode)
|
---|
2792 | && x0_ent->const_rtx != NULL_RTX
|
---|
2793 | && (GET_MODE (XEXP (x, 1)) == x1_ent->mode)
|
---|
2794 | && x1_ent->const_rtx != NULL_RTX)
|
---|
2795 | return 0;
|
---|
2796 | }
|
---|
2797 |
|
---|
2798 | return rtx_varies_p (x, from_alias);
|
---|
2799 | }
|
---|
2800 | |
---|
2801 |
|
---|
2802 | /* Canonicalize an expression:
|
---|
2803 | replace each register reference inside it
|
---|
2804 | with the "oldest" equivalent register.
|
---|
2805 |
|
---|
2806 | If INSN is nonzero and we are replacing a pseudo with a hard register
|
---|
2807 | or vice versa, validate_change is used to ensure that INSN remains valid
|
---|
2808 | after we make our substitution. The calls are made with IN_GROUP nonzero
|
---|
2809 | so apply_change_group must be called upon the outermost return from this
|
---|
2810 | function (unless INSN is zero). The result of apply_change_group can
|
---|
2811 | generally be discarded since the changes we are making are optional. */
|
---|
2812 |
|
---|
2813 | static rtx
|
---|
2814 | canon_reg (x, insn)
|
---|
2815 | rtx x;
|
---|
2816 | rtx insn;
|
---|
2817 | {
|
---|
2818 | int i;
|
---|
2819 | enum rtx_code code;
|
---|
2820 | const char *fmt;
|
---|
2821 |
|
---|
2822 | if (x == 0)
|
---|
2823 | return x;
|
---|
2824 |
|
---|
2825 | code = GET_CODE (x);
|
---|
2826 | switch (code)
|
---|
2827 | {
|
---|
2828 | case PC:
|
---|
2829 | case CC0:
|
---|
2830 | case CONST:
|
---|
2831 | case CONST_INT:
|
---|
2832 | case CONST_DOUBLE:
|
---|
2833 | case CONST_VECTOR:
|
---|
2834 | case SYMBOL_REF:
|
---|
2835 | case LABEL_REF:
|
---|
2836 | case ADDR_VEC:
|
---|
2837 | case ADDR_DIFF_VEC:
|
---|
2838 | return x;
|
---|
2839 |
|
---|
2840 | case REG:
|
---|
2841 | {
|
---|
2842 | int first;
|
---|
2843 | int q;
|
---|
2844 | struct qty_table_elem *ent;
|
---|
2845 |
|
---|
2846 | /* Never replace a hard reg, because hard regs can appear
|
---|
2847 | in more than one machine mode, and we must preserve the mode
|
---|
2848 | of each occurrence. Also, some hard regs appear in
|
---|
2849 | MEMs that are shared and mustn't be altered. Don't try to
|
---|
2850 | replace any reg that maps to a reg of class NO_REGS. */
|
---|
2851 | if (REGNO (x) < FIRST_PSEUDO_REGISTER
|
---|
2852 | || ! REGNO_QTY_VALID_P (REGNO (x)))
|
---|
2853 | return x;
|
---|
2854 |
|
---|
2855 | q = REG_QTY (REGNO (x));
|
---|
2856 | ent = &qty_table[q];
|
---|
2857 | first = ent->first_reg;
|
---|
2858 | return (first >= FIRST_PSEUDO_REGISTER ? regno_reg_rtx[first]
|
---|
2859 | : REGNO_REG_CLASS (first) == NO_REGS ? x
|
---|
2860 | : gen_rtx_REG (ent->mode, first));
|
---|
2861 | }
|
---|
2862 |
|
---|
2863 | default:
|
---|
2864 | break;
|
---|
2865 | }
|
---|
2866 |
|
---|
2867 | fmt = GET_RTX_FORMAT (code);
|
---|
2868 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
2869 | {
|
---|
2870 | int j;
|
---|
2871 |
|
---|
2872 | if (fmt[i] == 'e')
|
---|
2873 | {
|
---|
2874 | rtx new = canon_reg (XEXP (x, i), insn);
|
---|
2875 | int insn_code;
|
---|
2876 |
|
---|
2877 | /* If replacing pseudo with hard reg or vice versa, ensure the
|
---|
2878 | insn remains valid. Likewise if the insn has MATCH_DUPs. */
|
---|
2879 | if (insn != 0 && new != 0
|
---|
2880 | && GET_CODE (new) == REG && GET_CODE (XEXP (x, i)) == REG
|
---|
2881 | && (((REGNO (new) < FIRST_PSEUDO_REGISTER)
|
---|
2882 | != (REGNO (XEXP (x, i)) < FIRST_PSEUDO_REGISTER))
|
---|
2883 | || (insn_code = recog_memoized (insn)) < 0
|
---|
2884 | || insn_data[insn_code].n_dups > 0))
|
---|
2885 | validate_change (insn, &XEXP (x, i), new, 1);
|
---|
2886 | else
|
---|
2887 | XEXP (x, i) = new;
|
---|
2888 | }
|
---|
2889 | else if (fmt[i] == 'E')
|
---|
2890 | for (j = 0; j < XVECLEN (x, i); j++)
|
---|
2891 | XVECEXP (x, i, j) = canon_reg (XVECEXP (x, i, j), insn);
|
---|
2892 | }
|
---|
2893 |
|
---|
2894 | return x;
|
---|
2895 | }
|
---|
2896 | |
---|
2897 |
|
---|
2898 | /* LOC is a location within INSN that is an operand address (the contents of
|
---|
2899 | a MEM). Find the best equivalent address to use that is valid for this
|
---|
2900 | insn.
|
---|
2901 |
|
---|
2902 | On most CISC machines, complicated address modes are costly, and rtx_cost
|
---|
2903 | is a good approximation for that cost. However, most RISC machines have
|
---|
2904 | only a few (usually only one) memory reference formats. If an address is
|
---|
2905 | valid at all, it is often just as cheap as any other address. Hence, for
|
---|
2906 | RISC machines, we use the configuration macro `ADDRESS_COST' to compare the
|
---|
2907 | costs of various addresses. For two addresses of equal cost, choose the one
|
---|
2908 | with the highest `rtx_cost' value as that has the potential of eliminating
|
---|
2909 | the most insns. For equal costs, we choose the first in the equivalence
|
---|
2910 | class. Note that we ignore the fact that pseudo registers are cheaper
|
---|
2911 | than hard registers here because we would also prefer the pseudo registers.
|
---|
2912 | */
|
---|
2913 |
|
---|
2914 | static void
|
---|
2915 | find_best_addr (insn, loc, mode)
|
---|
2916 | rtx insn;
|
---|
2917 | rtx *loc;
|
---|
2918 | enum machine_mode mode;
|
---|
2919 | {
|
---|
2920 | struct table_elt *elt;
|
---|
2921 | rtx addr = *loc;
|
---|
2922 | #ifdef ADDRESS_COST
|
---|
2923 | struct table_elt *p;
|
---|
2924 | int found_better = 1;
|
---|
2925 | #endif
|
---|
2926 | int save_do_not_record = do_not_record;
|
---|
2927 | int save_hash_arg_in_memory = hash_arg_in_memory;
|
---|
2928 | int addr_volatile;
|
---|
2929 | int regno;
|
---|
2930 | unsigned hash;
|
---|
2931 |
|
---|
2932 | /* Do not try to replace constant addresses or addresses of local and
|
---|
2933 | argument slots. These MEM expressions are made only once and inserted
|
---|
2934 | in many instructions, as well as being used to control symbol table
|
---|
2935 | output. It is not safe to clobber them.
|
---|
2936 |
|
---|
2937 | There are some uncommon cases where the address is already in a register
|
---|
2938 | for some reason, but we cannot take advantage of that because we have
|
---|
2939 | no easy way to unshare the MEM. In addition, looking up all stack
|
---|
2940 | addresses is costly. */
|
---|
2941 | if ((GET_CODE (addr) == PLUS
|
---|
2942 | && GET_CODE (XEXP (addr, 0)) == REG
|
---|
2943 | && GET_CODE (XEXP (addr, 1)) == CONST_INT
|
---|
2944 | && (regno = REGNO (XEXP (addr, 0)),
|
---|
2945 | regno == FRAME_POINTER_REGNUM || regno == HARD_FRAME_POINTER_REGNUM
|
---|
2946 | || regno == ARG_POINTER_REGNUM))
|
---|
2947 | || (GET_CODE (addr) == REG
|
---|
2948 | && (regno = REGNO (addr), regno == FRAME_POINTER_REGNUM
|
---|
2949 | || regno == HARD_FRAME_POINTER_REGNUM
|
---|
2950 | || regno == ARG_POINTER_REGNUM))
|
---|
2951 | || GET_CODE (addr) == ADDRESSOF
|
---|
2952 | || CONSTANT_ADDRESS_P (addr))
|
---|
2953 | return;
|
---|
2954 |
|
---|
2955 | /* If this address is not simply a register, try to fold it. This will
|
---|
2956 | sometimes simplify the expression. Many simplifications
|
---|
2957 | will not be valid, but some, usually applying the associative rule, will
|
---|
2958 | be valid and produce better code. */
|
---|
2959 | if (GET_CODE (addr) != REG)
|
---|
2960 | {
|
---|
2961 | rtx folded = fold_rtx (copy_rtx (addr), NULL_RTX);
|
---|
2962 | int addr_folded_cost = address_cost (folded, mode);
|
---|
2963 | int addr_cost = address_cost (addr, mode);
|
---|
2964 |
|
---|
2965 | if ((addr_folded_cost < addr_cost
|
---|
2966 | || (addr_folded_cost == addr_cost
|
---|
2967 | /* ??? The rtx_cost comparison is left over from an older
|
---|
2968 | version of this code. It is probably no longer helpful. */
|
---|
2969 | && (rtx_cost (folded, MEM) > rtx_cost (addr, MEM)
|
---|
2970 | || approx_reg_cost (folded) < approx_reg_cost (addr))))
|
---|
2971 | && validate_change (insn, loc, folded, 0))
|
---|
2972 | addr = folded;
|
---|
2973 | }
|
---|
2974 |
|
---|
2975 | /* If this address is not in the hash table, we can't look for equivalences
|
---|
2976 | of the whole address. Also, ignore if volatile. */
|
---|
2977 |
|
---|
2978 | do_not_record = 0;
|
---|
2979 | hash = HASH (addr, Pmode);
|
---|
2980 | addr_volatile = do_not_record;
|
---|
2981 | do_not_record = save_do_not_record;
|
---|
2982 | hash_arg_in_memory = save_hash_arg_in_memory;
|
---|
2983 |
|
---|
2984 | if (addr_volatile)
|
---|
2985 | return;
|
---|
2986 |
|
---|
2987 | elt = lookup (addr, hash, Pmode);
|
---|
2988 |
|
---|
2989 | #ifndef ADDRESS_COST
|
---|
2990 | if (elt)
|
---|
2991 | {
|
---|
2992 | int our_cost = elt->cost;
|
---|
2993 |
|
---|
2994 | /* Find the lowest cost below ours that works. */
|
---|
2995 | for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
|
---|
2996 | if (elt->cost < our_cost
|
---|
2997 | && (GET_CODE (elt->exp) == REG
|
---|
2998 | || exp_equiv_p (elt->exp, elt->exp, 1, 0))
|
---|
2999 | && validate_change (insn, loc,
|
---|
3000 | canon_reg (copy_rtx (elt->exp), NULL_RTX), 0))
|
---|
3001 | return;
|
---|
3002 | }
|
---|
3003 | #else
|
---|
3004 |
|
---|
3005 | if (elt)
|
---|
3006 | {
|
---|
3007 | /* We need to find the best (under the criteria documented above) entry
|
---|
3008 | in the class that is valid. We use the `flag' field to indicate
|
---|
3009 | choices that were invalid and iterate until we can't find a better
|
---|
3010 | one that hasn't already been tried. */
|
---|
3011 |
|
---|
3012 | for (p = elt->first_same_value; p; p = p->next_same_value)
|
---|
3013 | p->flag = 0;
|
---|
3014 |
|
---|
3015 | while (found_better)
|
---|
3016 | {
|
---|
3017 | int best_addr_cost = address_cost (*loc, mode);
|
---|
3018 | int best_rtx_cost = (elt->cost + 1) >> 1;
|
---|
3019 | int exp_cost;
|
---|
3020 | struct table_elt *best_elt = elt;
|
---|
3021 |
|
---|
3022 | found_better = 0;
|
---|
3023 | for (p = elt->first_same_value; p; p = p->next_same_value)
|
---|
3024 | if (! p->flag)
|
---|
3025 | {
|
---|
3026 | if ((GET_CODE (p->exp) == REG
|
---|
3027 | || exp_equiv_p (p->exp, p->exp, 1, 0))
|
---|
3028 | && ((exp_cost = address_cost (p->exp, mode)) < best_addr_cost
|
---|
3029 | || (exp_cost == best_addr_cost
|
---|
3030 | && ((p->cost + 1) >> 1) > best_rtx_cost)))
|
---|
3031 | {
|
---|
3032 | found_better = 1;
|
---|
3033 | best_addr_cost = exp_cost;
|
---|
3034 | best_rtx_cost = (p->cost + 1) >> 1;
|
---|
3035 | best_elt = p;
|
---|
3036 | }
|
---|
3037 | }
|
---|
3038 |
|
---|
3039 | if (found_better)
|
---|
3040 | {
|
---|
3041 | if (validate_change (insn, loc,
|
---|
3042 | canon_reg (copy_rtx (best_elt->exp),
|
---|
3043 | NULL_RTX), 0))
|
---|
3044 | return;
|
---|
3045 | else
|
---|
3046 | best_elt->flag = 1;
|
---|
3047 | }
|
---|
3048 | }
|
---|
3049 | }
|
---|
3050 |
|
---|
3051 | /* If the address is a binary operation with the first operand a register
|
---|
3052 | and the second a constant, do the same as above, but looking for
|
---|
3053 | equivalences of the register. Then try to simplify before checking for
|
---|
3054 | the best address to use. This catches a few cases: First is when we
|
---|
3055 | have REG+const and the register is another REG+const. We can often merge
|
---|
3056 | the constants and eliminate one insn and one register. It may also be
|
---|
3057 | that a machine has a cheap REG+REG+const. Finally, this improves the
|
---|
3058 | code on the Alpha for unaligned byte stores. */
|
---|
3059 |
|
---|
3060 | if (flag_expensive_optimizations
|
---|
3061 | && (GET_RTX_CLASS (GET_CODE (*loc)) == '2'
|
---|
3062 | || GET_RTX_CLASS (GET_CODE (*loc)) == 'c')
|
---|
3063 | && GET_CODE (XEXP (*loc, 0)) == REG
|
---|
3064 | && GET_CODE (XEXP (*loc, 1)) == CONST_INT)
|
---|
3065 | {
|
---|
3066 | rtx c = XEXP (*loc, 1);
|
---|
3067 |
|
---|
3068 | do_not_record = 0;
|
---|
3069 | hash = HASH (XEXP (*loc, 0), Pmode);
|
---|
3070 | do_not_record = save_do_not_record;
|
---|
3071 | hash_arg_in_memory = save_hash_arg_in_memory;
|
---|
3072 |
|
---|
3073 | elt = lookup (XEXP (*loc, 0), hash, Pmode);
|
---|
3074 | if (elt == 0)
|
---|
3075 | return;
|
---|
3076 |
|
---|
3077 | /* We need to find the best (under the criteria documented above) entry
|
---|
3078 | in the class that is valid. We use the `flag' field to indicate
|
---|
3079 | choices that were invalid and iterate until we can't find a better
|
---|
3080 | one that hasn't already been tried. */
|
---|
3081 |
|
---|
3082 | for (p = elt->first_same_value; p; p = p->next_same_value)
|
---|
3083 | p->flag = 0;
|
---|
3084 |
|
---|
3085 | while (found_better)
|
---|
3086 | {
|
---|
3087 | int best_addr_cost = address_cost (*loc, mode);
|
---|
3088 | int best_rtx_cost = (COST (*loc) + 1) >> 1;
|
---|
3089 | struct table_elt *best_elt = elt;
|
---|
3090 | rtx best_rtx = *loc;
|
---|
3091 | int count;
|
---|
3092 |
|
---|
3093 | /* This is at worst case an O(n^2) algorithm, so limit our search
|
---|
3094 | to the first 32 elements on the list. This avoids trouble
|
---|
3095 | compiling code with very long basic blocks that can easily
|
---|
3096 | call simplify_gen_binary so many times that we run out of
|
---|
3097 | memory. */
|
---|
3098 |
|
---|
3099 | found_better = 0;
|
---|
3100 | for (p = elt->first_same_value, count = 0;
|
---|
3101 | p && count < 32;
|
---|
3102 | p = p->next_same_value, count++)
|
---|
3103 | if (! p->flag
|
---|
3104 | && (GET_CODE (p->exp) == REG
|
---|
3105 | || exp_equiv_p (p->exp, p->exp, 1, 0)))
|
---|
3106 | {
|
---|
3107 | rtx new = simplify_gen_binary (GET_CODE (*loc), Pmode,
|
---|
3108 | p->exp, c);
|
---|
3109 | int new_cost;
|
---|
3110 | new_cost = address_cost (new, mode);
|
---|
3111 |
|
---|
3112 | if (new_cost < best_addr_cost
|
---|
3113 | || (new_cost == best_addr_cost
|
---|
3114 | && (COST (new) + 1) >> 1 > best_rtx_cost))
|
---|
3115 | {
|
---|
3116 | found_better = 1;
|
---|
3117 | best_addr_cost = new_cost;
|
---|
3118 | best_rtx_cost = (COST (new) + 1) >> 1;
|
---|
3119 | best_elt = p;
|
---|
3120 | best_rtx = new;
|
---|
3121 | }
|
---|
3122 | }
|
---|
3123 |
|
---|
3124 | if (found_better)
|
---|
3125 | {
|
---|
3126 | if (validate_change (insn, loc,
|
---|
3127 | canon_reg (copy_rtx (best_rtx),
|
---|
3128 | NULL_RTX), 0))
|
---|
3129 | return;
|
---|
3130 | else
|
---|
3131 | best_elt->flag = 1;
|
---|
3132 | }
|
---|
3133 | }
|
---|
3134 | }
|
---|
3135 | #endif
|
---|
3136 | }
|
---|
3137 | |
---|
3138 |
|
---|
3139 | /* Given an operation (CODE, *PARG1, *PARG2), where code is a comparison
|
---|
3140 | operation (EQ, NE, GT, etc.), follow it back through the hash table and
|
---|
3141 | what values are being compared.
|
---|
3142 |
|
---|
3143 | *PARG1 and *PARG2 are updated to contain the rtx representing the values
|
---|
3144 | actually being compared. For example, if *PARG1 was (cc0) and *PARG2
|
---|
3145 | was (const_int 0), *PARG1 and *PARG2 will be set to the objects that were
|
---|
3146 | compared to produce cc0.
|
---|
3147 |
|
---|
3148 | The return value is the comparison operator and is either the code of
|
---|
3149 | A or the code corresponding to the inverse of the comparison. */
|
---|
3150 |
|
---|
3151 | static enum rtx_code
|
---|
3152 | find_comparison_args (code, parg1, parg2, pmode1, pmode2)
|
---|
3153 | enum rtx_code code;
|
---|
3154 | rtx *parg1, *parg2;
|
---|
3155 | enum machine_mode *pmode1, *pmode2;
|
---|
3156 | {
|
---|
3157 | rtx arg1, arg2;
|
---|
3158 |
|
---|
3159 | arg1 = *parg1, arg2 = *parg2;
|
---|
3160 |
|
---|
3161 | /* If ARG2 is const0_rtx, see what ARG1 is equivalent to. */
|
---|
3162 |
|
---|
3163 | while (arg2 == CONST0_RTX (GET_MODE (arg1)))
|
---|
3164 | {
|
---|
3165 | /* Set nonzero when we find something of interest. */
|
---|
3166 | rtx x = 0;
|
---|
3167 | int reverse_code = 0;
|
---|
3168 | struct table_elt *p = 0;
|
---|
3169 |
|
---|
3170 | /* If arg1 is a COMPARE, extract the comparison arguments from it.
|
---|
3171 | On machines with CC0, this is the only case that can occur, since
|
---|
3172 | fold_rtx will return the COMPARE or item being compared with zero
|
---|
3173 | when given CC0. */
|
---|
3174 |
|
---|
3175 | if (GET_CODE (arg1) == COMPARE && arg2 == const0_rtx)
|
---|
3176 | x = arg1;
|
---|
3177 |
|
---|
3178 | /* If ARG1 is a comparison operator and CODE is testing for
|
---|
3179 | STORE_FLAG_VALUE, get the inner arguments. */
|
---|
3180 |
|
---|
3181 | else if (GET_RTX_CLASS (GET_CODE (arg1)) == '<')
|
---|
3182 | {
|
---|
3183 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3184 | REAL_VALUE_TYPE fsfv;
|
---|
3185 | #endif
|
---|
3186 |
|
---|
3187 | if (code == NE
|
---|
3188 | || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
|
---|
3189 | && code == LT && STORE_FLAG_VALUE == -1)
|
---|
3190 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3191 | || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_FLOAT
|
---|
3192 | && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
|
---|
3193 | REAL_VALUE_NEGATIVE (fsfv)))
|
---|
3194 | #endif
|
---|
3195 | )
|
---|
3196 | x = arg1;
|
---|
3197 | else if (code == EQ
|
---|
3198 | || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_INT
|
---|
3199 | && code == GE && STORE_FLAG_VALUE == -1)
|
---|
3200 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3201 | || (GET_MODE_CLASS (GET_MODE (arg1)) == MODE_FLOAT
|
---|
3202 | && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
|
---|
3203 | REAL_VALUE_NEGATIVE (fsfv)))
|
---|
3204 | #endif
|
---|
3205 | )
|
---|
3206 | x = arg1, reverse_code = 1;
|
---|
3207 | }
|
---|
3208 |
|
---|
3209 | /* ??? We could also check for
|
---|
3210 |
|
---|
3211 | (ne (and (eq (...) (const_int 1))) (const_int 0))
|
---|
3212 |
|
---|
3213 | and related forms, but let's wait until we see them occurring. */
|
---|
3214 |
|
---|
3215 | if (x == 0)
|
---|
3216 | /* Look up ARG1 in the hash table and see if it has an equivalence
|
---|
3217 | that lets us see what is being compared. */
|
---|
3218 | p = lookup (arg1, safe_hash (arg1, GET_MODE (arg1)) & HASH_MASK,
|
---|
3219 | GET_MODE (arg1));
|
---|
3220 | if (p)
|
---|
3221 | {
|
---|
3222 | p = p->first_same_value;
|
---|
3223 |
|
---|
3224 | /* If what we compare is already known to be constant, that is as
|
---|
3225 | good as it gets.
|
---|
3226 | We need to break the loop in this case, because otherwise we
|
---|
3227 | can have an infinite loop when looking at a reg that is known
|
---|
3228 | to be a constant which is the same as a comparison of a reg
|
---|
3229 | against zero which appears later in the insn stream, which in
|
---|
3230 | turn is constant and the same as the comparison of the first reg
|
---|
3231 | against zero... */
|
---|
3232 | if (p->is_const)
|
---|
3233 | break;
|
---|
3234 | }
|
---|
3235 |
|
---|
3236 | for (; p; p = p->next_same_value)
|
---|
3237 | {
|
---|
3238 | enum machine_mode inner_mode = GET_MODE (p->exp);
|
---|
3239 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3240 | REAL_VALUE_TYPE fsfv;
|
---|
3241 | #endif
|
---|
3242 |
|
---|
3243 | /* If the entry isn't valid, skip it. */
|
---|
3244 | if (! exp_equiv_p (p->exp, p->exp, 1, 0))
|
---|
3245 | continue;
|
---|
3246 |
|
---|
3247 | if (GET_CODE (p->exp) == COMPARE
|
---|
3248 | /* Another possibility is that this machine has a compare insn
|
---|
3249 | that includes the comparison code. In that case, ARG1 would
|
---|
3250 | be equivalent to a comparison operation that would set ARG1 to
|
---|
3251 | either STORE_FLAG_VALUE or zero. If this is an NE operation,
|
---|
3252 | ORIG_CODE is the actual comparison being done; if it is an EQ,
|
---|
3253 | we must reverse ORIG_CODE. On machine with a negative value
|
---|
3254 | for STORE_FLAG_VALUE, also look at LT and GE operations. */
|
---|
3255 | || ((code == NE
|
---|
3256 | || (code == LT
|
---|
3257 | && GET_MODE_CLASS (inner_mode) == MODE_INT
|
---|
3258 | && (GET_MODE_BITSIZE (inner_mode)
|
---|
3259 | <= HOST_BITS_PER_WIDE_INT)
|
---|
3260 | && (STORE_FLAG_VALUE
|
---|
3261 | & ((HOST_WIDE_INT) 1
|
---|
3262 | << (GET_MODE_BITSIZE (inner_mode) - 1))))
|
---|
3263 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3264 | || (code == LT
|
---|
3265 | && GET_MODE_CLASS (inner_mode) == MODE_FLOAT
|
---|
3266 | && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
|
---|
3267 | REAL_VALUE_NEGATIVE (fsfv)))
|
---|
3268 | #endif
|
---|
3269 | )
|
---|
3270 | && GET_RTX_CLASS (GET_CODE (p->exp)) == '<'))
|
---|
3271 | {
|
---|
3272 | x = p->exp;
|
---|
3273 | break;
|
---|
3274 | }
|
---|
3275 | else if ((code == EQ
|
---|
3276 | || (code == GE
|
---|
3277 | && GET_MODE_CLASS (inner_mode) == MODE_INT
|
---|
3278 | && (GET_MODE_BITSIZE (inner_mode)
|
---|
3279 | <= HOST_BITS_PER_WIDE_INT)
|
---|
3280 | && (STORE_FLAG_VALUE
|
---|
3281 | & ((HOST_WIDE_INT) 1
|
---|
3282 | << (GET_MODE_BITSIZE (inner_mode) - 1))))
|
---|
3283 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3284 | || (code == GE
|
---|
3285 | && GET_MODE_CLASS (inner_mode) == MODE_FLOAT
|
---|
3286 | && (fsfv = FLOAT_STORE_FLAG_VALUE (GET_MODE (arg1)),
|
---|
3287 | REAL_VALUE_NEGATIVE (fsfv)))
|
---|
3288 | #endif
|
---|
3289 | )
|
---|
3290 | && GET_RTX_CLASS (GET_CODE (p->exp)) == '<')
|
---|
3291 | {
|
---|
3292 | reverse_code = 1;
|
---|
3293 | x = p->exp;
|
---|
3294 | break;
|
---|
3295 | }
|
---|
3296 |
|
---|
3297 | /* If this is fp + constant, the equivalent is a better operand since
|
---|
3298 | it may let us predict the value of the comparison. */
|
---|
3299 | else if (NONZERO_BASE_PLUS_P (p->exp))
|
---|
3300 | {
|
---|
3301 | arg1 = p->exp;
|
---|
3302 | continue;
|
---|
3303 | }
|
---|
3304 | }
|
---|
3305 |
|
---|
3306 | /* If we didn't find a useful equivalence for ARG1, we are done.
|
---|
3307 | Otherwise, set up for the next iteration. */
|
---|
3308 | if (x == 0)
|
---|
3309 | break;
|
---|
3310 |
|
---|
3311 | /* If we need to reverse the comparison, make sure that that is
|
---|
3312 | possible -- we can't necessarily infer the value of GE from LT
|
---|
3313 | with floating-point operands. */
|
---|
3314 | if (reverse_code)
|
---|
3315 | {
|
---|
3316 | enum rtx_code reversed = reversed_comparison_code (x, NULL_RTX);
|
---|
3317 | if (reversed == UNKNOWN)
|
---|
3318 | break;
|
---|
3319 | else
|
---|
3320 | code = reversed;
|
---|
3321 | }
|
---|
3322 | else if (GET_RTX_CLASS (GET_CODE (x)) == '<')
|
---|
3323 | code = GET_CODE (x);
|
---|
3324 | arg1 = XEXP (x, 0), arg2 = XEXP (x, 1);
|
---|
3325 | }
|
---|
3326 |
|
---|
3327 | /* Return our results. Return the modes from before fold_rtx
|
---|
3328 | because fold_rtx might produce const_int, and then it's too late. */
|
---|
3329 | *pmode1 = GET_MODE (arg1), *pmode2 = GET_MODE (arg2);
|
---|
3330 | *parg1 = fold_rtx (arg1, 0), *parg2 = fold_rtx (arg2, 0);
|
---|
3331 |
|
---|
3332 | return code;
|
---|
3333 | }
|
---|
3334 | |
---|
3335 |
|
---|
3336 | /* If X is a nontrivial arithmetic operation on an argument
|
---|
3337 | for which a constant value can be determined, return
|
---|
3338 | the result of operating on that value, as a constant.
|
---|
3339 | Otherwise, return X, possibly with one or more operands
|
---|
3340 | modified by recursive calls to this function.
|
---|
3341 |
|
---|
3342 | If X is a register whose contents are known, we do NOT
|
---|
3343 | return those contents here. equiv_constant is called to
|
---|
3344 | perform that task.
|
---|
3345 |
|
---|
3346 | INSN is the insn that we may be modifying. If it is 0, make a copy
|
---|
3347 | of X before modifying it. */
|
---|
3348 |
|
---|
3349 | static rtx
|
---|
3350 | fold_rtx (x, insn)
|
---|
3351 | rtx x;
|
---|
3352 | rtx insn;
|
---|
3353 | {
|
---|
3354 | enum rtx_code code;
|
---|
3355 | enum machine_mode mode;
|
---|
3356 | const char *fmt;
|
---|
3357 | int i;
|
---|
3358 | rtx new = 0;
|
---|
3359 | int copied = 0;
|
---|
3360 | int must_swap = 0;
|
---|
3361 |
|
---|
3362 | /* Folded equivalents of first two operands of X. */
|
---|
3363 | rtx folded_arg0;
|
---|
3364 | rtx folded_arg1;
|
---|
3365 |
|
---|
3366 | /* Constant equivalents of first three operands of X;
|
---|
3367 | 0 when no such equivalent is known. */
|
---|
3368 | rtx const_arg0;
|
---|
3369 | rtx const_arg1;
|
---|
3370 | rtx const_arg2;
|
---|
3371 |
|
---|
3372 | /* The mode of the first operand of X. We need this for sign and zero
|
---|
3373 | extends. */
|
---|
3374 | enum machine_mode mode_arg0;
|
---|
3375 |
|
---|
3376 | if (x == 0)
|
---|
3377 | return x;
|
---|
3378 |
|
---|
3379 | mode = GET_MODE (x);
|
---|
3380 | code = GET_CODE (x);
|
---|
3381 | switch (code)
|
---|
3382 | {
|
---|
3383 | case CONST:
|
---|
3384 | case CONST_INT:
|
---|
3385 | case CONST_DOUBLE:
|
---|
3386 | case CONST_VECTOR:
|
---|
3387 | case SYMBOL_REF:
|
---|
3388 | case LABEL_REF:
|
---|
3389 | case REG:
|
---|
3390 | /* No use simplifying an EXPR_LIST
|
---|
3391 | since they are used only for lists of args
|
---|
3392 | in a function call's REG_EQUAL note. */
|
---|
3393 | case EXPR_LIST:
|
---|
3394 | /* Changing anything inside an ADDRESSOF is incorrect; we don't
|
---|
3395 | want to (e.g.,) make (addressof (const_int 0)) just because
|
---|
3396 | the location is known to be zero. */
|
---|
3397 | case ADDRESSOF:
|
---|
3398 | return x;
|
---|
3399 |
|
---|
3400 | #ifdef HAVE_cc0
|
---|
3401 | case CC0:
|
---|
3402 | return prev_insn_cc0;
|
---|
3403 | #endif
|
---|
3404 |
|
---|
3405 | case PC:
|
---|
3406 | /* If the next insn is a CODE_LABEL followed by a jump table,
|
---|
3407 | PC's value is a LABEL_REF pointing to that label. That
|
---|
3408 | lets us fold switch statements on the VAX. */
|
---|
3409 | if (insn && GET_CODE (insn) == JUMP_INSN)
|
---|
3410 | {
|
---|
3411 | rtx next = next_nonnote_insn (insn);
|
---|
3412 |
|
---|
3413 | if (next && GET_CODE (next) == CODE_LABEL
|
---|
3414 | && NEXT_INSN (next) != 0
|
---|
3415 | && GET_CODE (NEXT_INSN (next)) == JUMP_INSN
|
---|
3416 | && (GET_CODE (PATTERN (NEXT_INSN (next))) == ADDR_VEC
|
---|
3417 | || GET_CODE (PATTERN (NEXT_INSN (next))) == ADDR_DIFF_VEC))
|
---|
3418 | return gen_rtx_LABEL_REF (Pmode, next);
|
---|
3419 | }
|
---|
3420 | break;
|
---|
3421 |
|
---|
3422 | case SUBREG:
|
---|
3423 | /* See if we previously assigned a constant value to this SUBREG. */
|
---|
3424 | if ((new = lookup_as_function (x, CONST_INT)) != 0
|
---|
3425 | || (new = lookup_as_function (x, CONST_DOUBLE)) != 0)
|
---|
3426 | return new;
|
---|
3427 |
|
---|
3428 | /* If this is a paradoxical SUBREG, we have no idea what value the
|
---|
3429 | extra bits would have. However, if the operand is equivalent
|
---|
3430 | to a SUBREG whose operand is the same as our mode, and all the
|
---|
3431 | modes are within a word, we can just use the inner operand
|
---|
3432 | because these SUBREGs just say how to treat the register.
|
---|
3433 |
|
---|
3434 | Similarly if we find an integer constant. */
|
---|
3435 |
|
---|
3436 | if (GET_MODE_SIZE (mode) > GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
|
---|
3437 | {
|
---|
3438 | enum machine_mode imode = GET_MODE (SUBREG_REG (x));
|
---|
3439 | struct table_elt *elt;
|
---|
3440 |
|
---|
3441 | if (GET_MODE_SIZE (mode) <= UNITS_PER_WORD
|
---|
3442 | && GET_MODE_SIZE (imode) <= UNITS_PER_WORD
|
---|
3443 | && (elt = lookup (SUBREG_REG (x), HASH (SUBREG_REG (x), imode),
|
---|
3444 | imode)) != 0)
|
---|
3445 | for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
|
---|
3446 | {
|
---|
3447 | if (CONSTANT_P (elt->exp)
|
---|
3448 | && GET_MODE (elt->exp) == VOIDmode)
|
---|
3449 | return elt->exp;
|
---|
3450 |
|
---|
3451 | if (GET_CODE (elt->exp) == SUBREG
|
---|
3452 | && GET_MODE (SUBREG_REG (elt->exp)) == mode
|
---|
3453 | && exp_equiv_p (elt->exp, elt->exp, 1, 0))
|
---|
3454 | return copy_rtx (SUBREG_REG (elt->exp));
|
---|
3455 | }
|
---|
3456 |
|
---|
3457 | return x;
|
---|
3458 | }
|
---|
3459 |
|
---|
3460 | /* Fold SUBREG_REG. If it changed, see if we can simplify the SUBREG.
|
---|
3461 | We might be able to if the SUBREG is extracting a single word in an
|
---|
3462 | integral mode or extracting the low part. */
|
---|
3463 |
|
---|
3464 | folded_arg0 = fold_rtx (SUBREG_REG (x), insn);
|
---|
3465 | const_arg0 = equiv_constant (folded_arg0);
|
---|
3466 | if (const_arg0)
|
---|
3467 | folded_arg0 = const_arg0;
|
---|
3468 |
|
---|
3469 | if (folded_arg0 != SUBREG_REG (x))
|
---|
3470 | {
|
---|
3471 | new = simplify_subreg (mode, folded_arg0,
|
---|
3472 | GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
|
---|
3473 | if (new)
|
---|
3474 | return new;
|
---|
3475 | }
|
---|
3476 |
|
---|
3477 | /* If this is a narrowing SUBREG and our operand is a REG, see if
|
---|
3478 | we can find an equivalence for REG that is an arithmetic operation
|
---|
3479 | in a wider mode where both operands are paradoxical SUBREGs
|
---|
3480 | from objects of our result mode. In that case, we couldn't report
|
---|
3481 | an equivalent value for that operation, since we don't know what the
|
---|
3482 | extra bits will be. But we can find an equivalence for this SUBREG
|
---|
3483 | by folding that operation is the narrow mode. This allows us to
|
---|
3484 | fold arithmetic in narrow modes when the machine only supports
|
---|
3485 | word-sized arithmetic.
|
---|
3486 |
|
---|
3487 | Also look for a case where we have a SUBREG whose operand is the
|
---|
3488 | same as our result. If both modes are smaller than a word, we
|
---|
3489 | are simply interpreting a register in different modes and we
|
---|
3490 | can use the inner value. */
|
---|
3491 |
|
---|
3492 | if (GET_CODE (folded_arg0) == REG
|
---|
3493 | && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (folded_arg0))
|
---|
3494 | && subreg_lowpart_p (x))
|
---|
3495 | {
|
---|
3496 | struct table_elt *elt;
|
---|
3497 |
|
---|
3498 | /* We can use HASH here since we know that canon_hash won't be
|
---|
3499 | called. */
|
---|
3500 | elt = lookup (folded_arg0,
|
---|
3501 | HASH (folded_arg0, GET_MODE (folded_arg0)),
|
---|
3502 | GET_MODE (folded_arg0));
|
---|
3503 |
|
---|
3504 | if (elt)
|
---|
3505 | elt = elt->first_same_value;
|
---|
3506 |
|
---|
3507 | for (; elt; elt = elt->next_same_value)
|
---|
3508 | {
|
---|
3509 | enum rtx_code eltcode = GET_CODE (elt->exp);
|
---|
3510 |
|
---|
3511 | /* Just check for unary and binary operations. */
|
---|
3512 | if (GET_RTX_CLASS (GET_CODE (elt->exp)) == '1'
|
---|
3513 | && GET_CODE (elt->exp) != SIGN_EXTEND
|
---|
3514 | && GET_CODE (elt->exp) != ZERO_EXTEND
|
---|
3515 | && GET_CODE (XEXP (elt->exp, 0)) == SUBREG
|
---|
3516 | && GET_MODE (SUBREG_REG (XEXP (elt->exp, 0))) == mode
|
---|
3517 | && (GET_MODE_CLASS (mode)
|
---|
3518 | == GET_MODE_CLASS (GET_MODE (XEXP (elt->exp, 0)))))
|
---|
3519 | {
|
---|
3520 | rtx op0 = SUBREG_REG (XEXP (elt->exp, 0));
|
---|
3521 |
|
---|
3522 | if (GET_CODE (op0) != REG && ! CONSTANT_P (op0))
|
---|
3523 | op0 = fold_rtx (op0, NULL_RTX);
|
---|
3524 |
|
---|
3525 | op0 = equiv_constant (op0);
|
---|
3526 | if (op0)
|
---|
3527 | new = simplify_unary_operation (GET_CODE (elt->exp), mode,
|
---|
3528 | op0, mode);
|
---|
3529 | }
|
---|
3530 | else if ((GET_RTX_CLASS (GET_CODE (elt->exp)) == '2'
|
---|
3531 | || GET_RTX_CLASS (GET_CODE (elt->exp)) == 'c')
|
---|
3532 | && eltcode != DIV && eltcode != MOD
|
---|
3533 | && eltcode != UDIV && eltcode != UMOD
|
---|
3534 | && eltcode != ASHIFTRT && eltcode != LSHIFTRT
|
---|
3535 | && eltcode != ROTATE && eltcode != ROTATERT
|
---|
3536 | && ((GET_CODE (XEXP (elt->exp, 0)) == SUBREG
|
---|
3537 | && (GET_MODE (SUBREG_REG (XEXP (elt->exp, 0)))
|
---|
3538 | == mode))
|
---|
3539 | || CONSTANT_P (XEXP (elt->exp, 0)))
|
---|
3540 | && ((GET_CODE (XEXP (elt->exp, 1)) == SUBREG
|
---|
3541 | && (GET_MODE (SUBREG_REG (XEXP (elt->exp, 1)))
|
---|
3542 | == mode))
|
---|
3543 | || CONSTANT_P (XEXP (elt->exp, 1))))
|
---|
3544 | {
|
---|
3545 | rtx op0 = gen_lowpart_common (mode, XEXP (elt->exp, 0));
|
---|
3546 | rtx op1 = gen_lowpart_common (mode, XEXP (elt->exp, 1));
|
---|
3547 |
|
---|
3548 | if (op0 && GET_CODE (op0) != REG && ! CONSTANT_P (op0))
|
---|
3549 | op0 = fold_rtx (op0, NULL_RTX);
|
---|
3550 |
|
---|
3551 | if (op0)
|
---|
3552 | op0 = equiv_constant (op0);
|
---|
3553 |
|
---|
3554 | if (op1 && GET_CODE (op1) != REG && ! CONSTANT_P (op1))
|
---|
3555 | op1 = fold_rtx (op1, NULL_RTX);
|
---|
3556 |
|
---|
3557 | if (op1)
|
---|
3558 | op1 = equiv_constant (op1);
|
---|
3559 |
|
---|
3560 | /* If we are looking for the low SImode part of
|
---|
3561 | (ashift:DI c (const_int 32)), it doesn't work
|
---|
3562 | to compute that in SImode, because a 32-bit shift
|
---|
3563 | in SImode is unpredictable. We know the value is 0. */
|
---|
3564 | if (op0 && op1
|
---|
3565 | && GET_CODE (elt->exp) == ASHIFT
|
---|
3566 | && GET_CODE (op1) == CONST_INT
|
---|
3567 | && INTVAL (op1) >= GET_MODE_BITSIZE (mode))
|
---|
3568 | {
|
---|
3569 | if (INTVAL (op1) < GET_MODE_BITSIZE (GET_MODE (elt->exp)))
|
---|
3570 |
|
---|
3571 | /* If the count fits in the inner mode's width,
|
---|
3572 | but exceeds the outer mode's width,
|
---|
3573 | the value will get truncated to 0
|
---|
3574 | by the subreg. */
|
---|
3575 | new = const0_rtx;
|
---|
3576 | else
|
---|
3577 | /* If the count exceeds even the inner mode's width,
|
---|
3578 | don't fold this expression. */
|
---|
3579 | new = 0;
|
---|
3580 | }
|
---|
3581 | else if (op0 && op1)
|
---|
3582 | new = simplify_binary_operation (GET_CODE (elt->exp), mode,
|
---|
3583 | op0, op1);
|
---|
3584 | }
|
---|
3585 |
|
---|
3586 | else if (GET_CODE (elt->exp) == SUBREG
|
---|
3587 | && GET_MODE (SUBREG_REG (elt->exp)) == mode
|
---|
3588 | && (GET_MODE_SIZE (GET_MODE (folded_arg0))
|
---|
3589 | <= UNITS_PER_WORD)
|
---|
3590 | && exp_equiv_p (elt->exp, elt->exp, 1, 0))
|
---|
3591 | new = copy_rtx (SUBREG_REG (elt->exp));
|
---|
3592 |
|
---|
3593 | if (new)
|
---|
3594 | return new;
|
---|
3595 | }
|
---|
3596 | }
|
---|
3597 |
|
---|
3598 | return x;
|
---|
3599 |
|
---|
3600 | case NOT:
|
---|
3601 | case NEG:
|
---|
3602 | /* If we have (NOT Y), see if Y is known to be (NOT Z).
|
---|
3603 | If so, (NOT Y) simplifies to Z. Similarly for NEG. */
|
---|
3604 | new = lookup_as_function (XEXP (x, 0), code);
|
---|
3605 | if (new)
|
---|
3606 | return fold_rtx (copy_rtx (XEXP (new, 0)), insn);
|
---|
3607 | break;
|
---|
3608 |
|
---|
3609 | case MEM:
|
---|
3610 | /* If we are not actually processing an insn, don't try to find the
|
---|
3611 | best address. Not only don't we care, but we could modify the
|
---|
3612 | MEM in an invalid way since we have no insn to validate against. */
|
---|
3613 | if (insn != 0)
|
---|
3614 | find_best_addr (insn, &XEXP (x, 0), GET_MODE (x));
|
---|
3615 |
|
---|
3616 | {
|
---|
3617 | /* Even if we don't fold in the insn itself,
|
---|
3618 | we can safely do so here, in hopes of getting a constant. */
|
---|
3619 | rtx addr = fold_rtx (XEXP (x, 0), NULL_RTX);
|
---|
3620 | rtx base = 0;
|
---|
3621 | HOST_WIDE_INT offset = 0;
|
---|
3622 |
|
---|
3623 | if (GET_CODE (addr) == REG
|
---|
3624 | && REGNO_QTY_VALID_P (REGNO (addr)))
|
---|
3625 | {
|
---|
3626 | int addr_q = REG_QTY (REGNO (addr));
|
---|
3627 | struct qty_table_elem *addr_ent = &qty_table[addr_q];
|
---|
3628 |
|
---|
3629 | if (GET_MODE (addr) == addr_ent->mode
|
---|
3630 | && addr_ent->const_rtx != NULL_RTX)
|
---|
3631 | addr = addr_ent->const_rtx;
|
---|
3632 | }
|
---|
3633 |
|
---|
3634 | /* If address is constant, split it into a base and integer offset. */
|
---|
3635 | if (GET_CODE (addr) == SYMBOL_REF || GET_CODE (addr) == LABEL_REF)
|
---|
3636 | base = addr;
|
---|
3637 | else if (GET_CODE (addr) == CONST && GET_CODE (XEXP (addr, 0)) == PLUS
|
---|
3638 | && GET_CODE (XEXP (XEXP (addr, 0), 1)) == CONST_INT)
|
---|
3639 | {
|
---|
3640 | base = XEXP (XEXP (addr, 0), 0);
|
---|
3641 | offset = INTVAL (XEXP (XEXP (addr, 0), 1));
|
---|
3642 | }
|
---|
3643 | else if (GET_CODE (addr) == LO_SUM
|
---|
3644 | && GET_CODE (XEXP (addr, 1)) == SYMBOL_REF)
|
---|
3645 | base = XEXP (addr, 1);
|
---|
3646 | else if (GET_CODE (addr) == ADDRESSOF)
|
---|
3647 | return change_address (x, VOIDmode, addr);
|
---|
3648 |
|
---|
3649 | /* If this is a constant pool reference, we can fold it into its
|
---|
3650 | constant to allow better value tracking. */
|
---|
3651 | if (base && GET_CODE (base) == SYMBOL_REF
|
---|
3652 | && CONSTANT_POOL_ADDRESS_P (base))
|
---|
3653 | {
|
---|
3654 | rtx constant = get_pool_constant (base);
|
---|
3655 | enum machine_mode const_mode = get_pool_mode (base);
|
---|
3656 | rtx new;
|
---|
3657 |
|
---|
3658 | if (CONSTANT_P (constant) && GET_CODE (constant) != CONST_INT)
|
---|
3659 | constant_pool_entries_cost = COST (constant);
|
---|
3660 |
|
---|
3661 | /* If we are loading the full constant, we have an equivalence. */
|
---|
3662 | if (offset == 0 && mode == const_mode)
|
---|
3663 | return constant;
|
---|
3664 |
|
---|
3665 | /* If this actually isn't a constant (weird!), we can't do
|
---|
3666 | anything. Otherwise, handle the two most common cases:
|
---|
3667 | extracting a word from a multi-word constant, and extracting
|
---|
3668 | the low-order bits. Other cases don't seem common enough to
|
---|
3669 | worry about. */
|
---|
3670 | if (! CONSTANT_P (constant))
|
---|
3671 | return x;
|
---|
3672 |
|
---|
3673 | if (GET_MODE_CLASS (mode) == MODE_INT
|
---|
3674 | && GET_MODE_SIZE (mode) == UNITS_PER_WORD
|
---|
3675 | && offset % UNITS_PER_WORD == 0
|
---|
3676 | && (new = operand_subword (constant,
|
---|
3677 | offset / UNITS_PER_WORD,
|
---|
3678 | 0, const_mode)) != 0)
|
---|
3679 | return new;
|
---|
3680 |
|
---|
3681 | if (((BYTES_BIG_ENDIAN
|
---|
3682 | && offset == GET_MODE_SIZE (GET_MODE (constant)) - 1)
|
---|
3683 | || (! BYTES_BIG_ENDIAN && offset == 0))
|
---|
3684 | && (new = gen_lowpart_if_possible (mode, constant)) != 0)
|
---|
3685 | return new;
|
---|
3686 | }
|
---|
3687 |
|
---|
3688 | /* If this is a reference to a label at a known position in a jump
|
---|
3689 | table, we also know its value. */
|
---|
3690 | if (base && GET_CODE (base) == LABEL_REF)
|
---|
3691 | {
|
---|
3692 | rtx label = XEXP (base, 0);
|
---|
3693 | rtx table_insn = NEXT_INSN (label);
|
---|
3694 |
|
---|
3695 | if (table_insn && GET_CODE (table_insn) == JUMP_INSN
|
---|
3696 | && GET_CODE (PATTERN (table_insn)) == ADDR_VEC)
|
---|
3697 | {
|
---|
3698 | rtx table = PATTERN (table_insn);
|
---|
3699 |
|
---|
3700 | if (offset >= 0
|
---|
3701 | && (offset / GET_MODE_SIZE (GET_MODE (table))
|
---|
3702 | < XVECLEN (table, 0)))
|
---|
3703 | return XVECEXP (table, 0,
|
---|
3704 | offset / GET_MODE_SIZE (GET_MODE (table)));
|
---|
3705 | }
|
---|
3706 | if (table_insn && GET_CODE (table_insn) == JUMP_INSN
|
---|
3707 | && GET_CODE (PATTERN (table_insn)) == ADDR_DIFF_VEC)
|
---|
3708 | {
|
---|
3709 | rtx table = PATTERN (table_insn);
|
---|
3710 |
|
---|
3711 | if (offset >= 0
|
---|
3712 | && (offset / GET_MODE_SIZE (GET_MODE (table))
|
---|
3713 | < XVECLEN (table, 1)))
|
---|
3714 | {
|
---|
3715 | offset /= GET_MODE_SIZE (GET_MODE (table));
|
---|
3716 | new = gen_rtx_MINUS (Pmode, XVECEXP (table, 1, offset),
|
---|
3717 | XEXP (table, 0));
|
---|
3718 |
|
---|
3719 | if (GET_MODE (table) != Pmode)
|
---|
3720 | new = gen_rtx_TRUNCATE (GET_MODE (table), new);
|
---|
3721 |
|
---|
3722 | /* Indicate this is a constant. This isn't a
|
---|
3723 | valid form of CONST, but it will only be used
|
---|
3724 | to fold the next insns and then discarded, so
|
---|
3725 | it should be safe.
|
---|
3726 |
|
---|
3727 | Note this expression must be explicitly discarded,
|
---|
3728 | by cse_insn, else it may end up in a REG_EQUAL note
|
---|
3729 | and "escape" to cause problems elsewhere. */
|
---|
3730 | return gen_rtx_CONST (GET_MODE (new), new);
|
---|
3731 | }
|
---|
3732 | }
|
---|
3733 | }
|
---|
3734 |
|
---|
3735 | return x;
|
---|
3736 | }
|
---|
3737 |
|
---|
3738 | #ifdef NO_FUNCTION_CSE
|
---|
3739 | case CALL:
|
---|
3740 | if (CONSTANT_P (XEXP (XEXP (x, 0), 0)))
|
---|
3741 | return x;
|
---|
3742 | break;
|
---|
3743 | #endif
|
---|
3744 |
|
---|
3745 | case ASM_OPERANDS:
|
---|
3746 | for (i = ASM_OPERANDS_INPUT_LENGTH (x) - 1; i >= 0; i--)
|
---|
3747 | validate_change (insn, &ASM_OPERANDS_INPUT (x, i),
|
---|
3748 | fold_rtx (ASM_OPERANDS_INPUT (x, i), insn), 0);
|
---|
3749 | break;
|
---|
3750 |
|
---|
3751 | default:
|
---|
3752 | break;
|
---|
3753 | }
|
---|
3754 |
|
---|
3755 | const_arg0 = 0;
|
---|
3756 | const_arg1 = 0;
|
---|
3757 | const_arg2 = 0;
|
---|
3758 | mode_arg0 = VOIDmode;
|
---|
3759 |
|
---|
3760 | /* Try folding our operands.
|
---|
3761 | Then see which ones have constant values known. */
|
---|
3762 |
|
---|
3763 | fmt = GET_RTX_FORMAT (code);
|
---|
3764 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
3765 | if (fmt[i] == 'e')
|
---|
3766 | {
|
---|
3767 | rtx arg = XEXP (x, i);
|
---|
3768 | rtx folded_arg = arg, const_arg = 0;
|
---|
3769 | enum machine_mode mode_arg = GET_MODE (arg);
|
---|
3770 | rtx cheap_arg, expensive_arg;
|
---|
3771 | rtx replacements[2];
|
---|
3772 | int j;
|
---|
3773 | int old_cost = COST_IN (XEXP (x, i), code);
|
---|
3774 |
|
---|
3775 | /* Most arguments are cheap, so handle them specially. */
|
---|
3776 | switch (GET_CODE (arg))
|
---|
3777 | {
|
---|
3778 | case REG:
|
---|
3779 | /* This is the same as calling equiv_constant; it is duplicated
|
---|
3780 | here for speed. */
|
---|
3781 | if (REGNO_QTY_VALID_P (REGNO (arg)))
|
---|
3782 | {
|
---|
3783 | int arg_q = REG_QTY (REGNO (arg));
|
---|
3784 | struct qty_table_elem *arg_ent = &qty_table[arg_q];
|
---|
3785 |
|
---|
3786 | if (arg_ent->const_rtx != NULL_RTX
|
---|
3787 | && GET_CODE (arg_ent->const_rtx) != REG
|
---|
3788 | && GET_CODE (arg_ent->const_rtx) != PLUS)
|
---|
3789 | const_arg
|
---|
3790 | = gen_lowpart_if_possible (GET_MODE (arg),
|
---|
3791 | arg_ent->const_rtx);
|
---|
3792 | }
|
---|
3793 | break;
|
---|
3794 |
|
---|
3795 | case CONST:
|
---|
3796 | case CONST_INT:
|
---|
3797 | case SYMBOL_REF:
|
---|
3798 | case LABEL_REF:
|
---|
3799 | case CONST_DOUBLE:
|
---|
3800 | case CONST_VECTOR:
|
---|
3801 | const_arg = arg;
|
---|
3802 | break;
|
---|
3803 |
|
---|
3804 | #ifdef HAVE_cc0
|
---|
3805 | case CC0:
|
---|
3806 | folded_arg = prev_insn_cc0;
|
---|
3807 | mode_arg = prev_insn_cc0_mode;
|
---|
3808 | const_arg = equiv_constant (folded_arg);
|
---|
3809 | break;
|
---|
3810 | #endif
|
---|
3811 |
|
---|
3812 | default:
|
---|
3813 | folded_arg = fold_rtx (arg, insn);
|
---|
3814 | const_arg = equiv_constant (folded_arg);
|
---|
3815 | }
|
---|
3816 |
|
---|
3817 | /* For the first three operands, see if the operand
|
---|
3818 | is constant or equivalent to a constant. */
|
---|
3819 | switch (i)
|
---|
3820 | {
|
---|
3821 | case 0:
|
---|
3822 | folded_arg0 = folded_arg;
|
---|
3823 | const_arg0 = const_arg;
|
---|
3824 | mode_arg0 = mode_arg;
|
---|
3825 | break;
|
---|
3826 | case 1:
|
---|
3827 | folded_arg1 = folded_arg;
|
---|
3828 | const_arg1 = const_arg;
|
---|
3829 | break;
|
---|
3830 | case 2:
|
---|
3831 | const_arg2 = const_arg;
|
---|
3832 | break;
|
---|
3833 | }
|
---|
3834 |
|
---|
3835 | /* Pick the least expensive of the folded argument and an
|
---|
3836 | equivalent constant argument. */
|
---|
3837 | if (const_arg == 0 || const_arg == folded_arg
|
---|
3838 | || COST_IN (const_arg, code) > COST_IN (folded_arg, code))
|
---|
3839 | cheap_arg = folded_arg, expensive_arg = const_arg;
|
---|
3840 | else
|
---|
3841 | cheap_arg = const_arg, expensive_arg = folded_arg;
|
---|
3842 |
|
---|
3843 | /* Try to replace the operand with the cheapest of the two
|
---|
3844 | possibilities. If it doesn't work and this is either of the first
|
---|
3845 | two operands of a commutative operation, try swapping them.
|
---|
3846 | If THAT fails, try the more expensive, provided it is cheaper
|
---|
3847 | than what is already there. */
|
---|
3848 |
|
---|
3849 | if (cheap_arg == XEXP (x, i))
|
---|
3850 | continue;
|
---|
3851 |
|
---|
3852 | if (insn == 0 && ! copied)
|
---|
3853 | {
|
---|
3854 | x = copy_rtx (x);
|
---|
3855 | copied = 1;
|
---|
3856 | }
|
---|
3857 |
|
---|
3858 | /* Order the replacements from cheapest to most expensive. */
|
---|
3859 | replacements[0] = cheap_arg;
|
---|
3860 | replacements[1] = expensive_arg;
|
---|
3861 |
|
---|
3862 | for (j = 0; j < 2 && replacements[j]; j++)
|
---|
3863 | {
|
---|
3864 | int new_cost = COST_IN (replacements[j], code);
|
---|
3865 |
|
---|
3866 | /* Stop if what existed before was cheaper. Prefer constants
|
---|
3867 | in the case of a tie. */
|
---|
3868 | if (new_cost > old_cost
|
---|
3869 | || (new_cost == old_cost && CONSTANT_P (XEXP (x, i))))
|
---|
3870 | break;
|
---|
3871 |
|
---|
3872 | if (validate_change (insn, &XEXP (x, i), replacements[j], 0))
|
---|
3873 | break;
|
---|
3874 |
|
---|
3875 | if (code == NE || code == EQ || GET_RTX_CLASS (code) == 'c'
|
---|
3876 | || code == LTGT || code == UNEQ || code == ORDERED
|
---|
3877 | || code == UNORDERED)
|
---|
3878 | {
|
---|
3879 | validate_change (insn, &XEXP (x, i), XEXP (x, 1 - i), 1);
|
---|
3880 | validate_change (insn, &XEXP (x, 1 - i), replacements[j], 1);
|
---|
3881 |
|
---|
3882 | if (apply_change_group ())
|
---|
3883 | {
|
---|
3884 | /* Swap them back to be invalid so that this loop can
|
---|
3885 | continue and flag them to be swapped back later. */
|
---|
3886 | rtx tem;
|
---|
3887 |
|
---|
3888 | tem = XEXP (x, 0); XEXP (x, 0) = XEXP (x, 1);
|
---|
3889 | XEXP (x, 1) = tem;
|
---|
3890 | must_swap = 1;
|
---|
3891 | break;
|
---|
3892 | }
|
---|
3893 | }
|
---|
3894 | }
|
---|
3895 | }
|
---|
3896 |
|
---|
3897 | else
|
---|
3898 | {
|
---|
3899 | if (fmt[i] == 'E')
|
---|
3900 | /* Don't try to fold inside of a vector of expressions.
|
---|
3901 | Doing nothing is harmless. */
|
---|
3902 | {;}
|
---|
3903 | }
|
---|
3904 |
|
---|
3905 | /* If a commutative operation, place a constant integer as the second
|
---|
3906 | operand unless the first operand is also a constant integer. Otherwise,
|
---|
3907 | place any constant second unless the first operand is also a constant. */
|
---|
3908 |
|
---|
3909 | if (code == EQ || code == NE || GET_RTX_CLASS (code) == 'c'
|
---|
3910 | || code == LTGT || code == UNEQ || code == ORDERED
|
---|
3911 | || code == UNORDERED)
|
---|
3912 | {
|
---|
3913 | if (must_swap || (const_arg0
|
---|
3914 | && (const_arg1 == 0
|
---|
3915 | || (GET_CODE (const_arg0) == CONST_INT
|
---|
3916 | && GET_CODE (const_arg1) != CONST_INT))))
|
---|
3917 | {
|
---|
3918 | rtx tem = XEXP (x, 0);
|
---|
3919 |
|
---|
3920 | if (insn == 0 && ! copied)
|
---|
3921 | {
|
---|
3922 | x = copy_rtx (x);
|
---|
3923 | copied = 1;
|
---|
3924 | }
|
---|
3925 |
|
---|
3926 | validate_change (insn, &XEXP (x, 0), XEXP (x, 1), 1);
|
---|
3927 | validate_change (insn, &XEXP (x, 1), tem, 1);
|
---|
3928 | if (apply_change_group ())
|
---|
3929 | {
|
---|
3930 | tem = const_arg0, const_arg0 = const_arg1, const_arg1 = tem;
|
---|
3931 | tem = folded_arg0, folded_arg0 = folded_arg1, folded_arg1 = tem;
|
---|
3932 | }
|
---|
3933 | }
|
---|
3934 | }
|
---|
3935 |
|
---|
3936 | /* If X is an arithmetic operation, see if we can simplify it. */
|
---|
3937 |
|
---|
3938 | switch (GET_RTX_CLASS (code))
|
---|
3939 | {
|
---|
3940 | case '1':
|
---|
3941 | {
|
---|
3942 | int is_const = 0;
|
---|
3943 |
|
---|
3944 | /* We can't simplify extension ops unless we know the
|
---|
3945 | original mode. */
|
---|
3946 | if ((code == ZERO_EXTEND || code == SIGN_EXTEND)
|
---|
3947 | && mode_arg0 == VOIDmode)
|
---|
3948 | break;
|
---|
3949 |
|
---|
3950 | /* If we had a CONST, strip it off and put it back later if we
|
---|
3951 | fold. */
|
---|
3952 | if (const_arg0 != 0 && GET_CODE (const_arg0) == CONST)
|
---|
3953 | is_const = 1, const_arg0 = XEXP (const_arg0, 0);
|
---|
3954 |
|
---|
3955 | new = simplify_unary_operation (code, mode,
|
---|
3956 | const_arg0 ? const_arg0 : folded_arg0,
|
---|
3957 | mode_arg0);
|
---|
3958 | if (new != 0 && is_const)
|
---|
3959 | new = gen_rtx_CONST (mode, new);
|
---|
3960 | }
|
---|
3961 | break;
|
---|
3962 |
|
---|
3963 | case '<':
|
---|
3964 | /* See what items are actually being compared and set FOLDED_ARG[01]
|
---|
3965 | to those values and CODE to the actual comparison code. If any are
|
---|
3966 | constant, set CONST_ARG0 and CONST_ARG1 appropriately. We needn't
|
---|
3967 | do anything if both operands are already known to be constant. */
|
---|
3968 |
|
---|
3969 | if (const_arg0 == 0 || const_arg1 == 0)
|
---|
3970 | {
|
---|
3971 | struct table_elt *p0, *p1;
|
---|
3972 | rtx true_rtx = const_true_rtx, false_rtx = const0_rtx;
|
---|
3973 | enum machine_mode mode_arg1;
|
---|
3974 |
|
---|
3975 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
3976 | if (GET_MODE_CLASS (mode) == MODE_FLOAT)
|
---|
3977 | {
|
---|
3978 | true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
|
---|
3979 | (FLOAT_STORE_FLAG_VALUE (mode), mode));
|
---|
3980 | false_rtx = CONST0_RTX (mode);
|
---|
3981 | }
|
---|
3982 | #endif
|
---|
3983 |
|
---|
3984 | code = find_comparison_args (code, &folded_arg0, &folded_arg1,
|
---|
3985 | &mode_arg0, &mode_arg1);
|
---|
3986 | const_arg0 = equiv_constant (folded_arg0);
|
---|
3987 | const_arg1 = equiv_constant (folded_arg1);
|
---|
3988 |
|
---|
3989 | /* If the mode is VOIDmode or a MODE_CC mode, we don't know
|
---|
3990 | what kinds of things are being compared, so we can't do
|
---|
3991 | anything with this comparison. */
|
---|
3992 |
|
---|
3993 | if (mode_arg0 == VOIDmode || GET_MODE_CLASS (mode_arg0) == MODE_CC)
|
---|
3994 | break;
|
---|
3995 |
|
---|
3996 | /* If we do not now have two constants being compared, see
|
---|
3997 | if we can nevertheless deduce some things about the
|
---|
3998 | comparison. */
|
---|
3999 | if (const_arg0 == 0 || const_arg1 == 0)
|
---|
4000 | {
|
---|
4001 | /* Is FOLDED_ARG0 frame-pointer plus a constant? Or
|
---|
4002 | non-explicit constant? These aren't zero, but we
|
---|
4003 | don't know their sign. */
|
---|
4004 | if (const_arg1 == const0_rtx
|
---|
4005 | && (NONZERO_BASE_PLUS_P (folded_arg0)
|
---|
4006 | #if 0 /* Sad to say, on sysvr4, #pragma weak can make a symbol address
|
---|
4007 | come out as 0. */
|
---|
4008 | || GET_CODE (folded_arg0) == SYMBOL_REF
|
---|
4009 | #endif
|
---|
4010 | || GET_CODE (folded_arg0) == LABEL_REF
|
---|
4011 | || GET_CODE (folded_arg0) == CONST))
|
---|
4012 | {
|
---|
4013 | if (code == EQ)
|
---|
4014 | return false_rtx;
|
---|
4015 | else if (code == NE)
|
---|
4016 | return true_rtx;
|
---|
4017 | }
|
---|
4018 |
|
---|
4019 | /* See if the two operands are the same. */
|
---|
4020 |
|
---|
4021 | if (folded_arg0 == folded_arg1
|
---|
4022 | || (GET_CODE (folded_arg0) == REG
|
---|
4023 | && GET_CODE (folded_arg1) == REG
|
---|
4024 | && (REG_QTY (REGNO (folded_arg0))
|
---|
4025 | == REG_QTY (REGNO (folded_arg1))))
|
---|
4026 | || ((p0 = lookup (folded_arg0,
|
---|
4027 | (safe_hash (folded_arg0, mode_arg0)
|
---|
4028 | & HASH_MASK), mode_arg0))
|
---|
4029 | && (p1 = lookup (folded_arg1,
|
---|
4030 | (safe_hash (folded_arg1, mode_arg0)
|
---|
4031 | & HASH_MASK), mode_arg0))
|
---|
4032 | && p0->first_same_value == p1->first_same_value))
|
---|
4033 | {
|
---|
4034 | /* Sadly two equal NaNs are not equivalent. */
|
---|
4035 | if (!HONOR_NANS (mode_arg0))
|
---|
4036 | return ((code == EQ || code == LE || code == GE
|
---|
4037 | || code == LEU || code == GEU || code == UNEQ
|
---|
4038 | || code == UNLE || code == UNGE
|
---|
4039 | || code == ORDERED)
|
---|
4040 | ? true_rtx : false_rtx);
|
---|
4041 | /* Take care for the FP compares we can resolve. */
|
---|
4042 | if (code == UNEQ || code == UNLE || code == UNGE)
|
---|
4043 | return true_rtx;
|
---|
4044 | if (code == LTGT || code == LT || code == GT)
|
---|
4045 | return false_rtx;
|
---|
4046 | }
|
---|
4047 |
|
---|
4048 | /* If FOLDED_ARG0 is a register, see if the comparison we are
|
---|
4049 | doing now is either the same as we did before or the reverse
|
---|
4050 | (we only check the reverse if not floating-point). */
|
---|
4051 | else if (GET_CODE (folded_arg0) == REG)
|
---|
4052 | {
|
---|
4053 | int qty = REG_QTY (REGNO (folded_arg0));
|
---|
4054 |
|
---|
4055 | if (REGNO_QTY_VALID_P (REGNO (folded_arg0)))
|
---|
4056 | {
|
---|
4057 | struct qty_table_elem *ent = &qty_table[qty];
|
---|
4058 |
|
---|
4059 | if ((comparison_dominates_p (ent->comparison_code, code)
|
---|
4060 | || (! FLOAT_MODE_P (mode_arg0)
|
---|
4061 | && comparison_dominates_p (ent->comparison_code,
|
---|
4062 | reverse_condition (code))))
|
---|
4063 | && (rtx_equal_p (ent->comparison_const, folded_arg1)
|
---|
4064 | || (const_arg1
|
---|
4065 | && rtx_equal_p (ent->comparison_const,
|
---|
4066 | const_arg1))
|
---|
4067 | || (GET_CODE (folded_arg1) == REG
|
---|
4068 | && (REG_QTY (REGNO (folded_arg1)) == ent->comparison_qty))))
|
---|
4069 | return (comparison_dominates_p (ent->comparison_code, code)
|
---|
4070 | ? true_rtx : false_rtx);
|
---|
4071 | }
|
---|
4072 | }
|
---|
4073 | }
|
---|
4074 | }
|
---|
4075 |
|
---|
4076 | /* If we are comparing against zero, see if the first operand is
|
---|
4077 | equivalent to an IOR with a constant. If so, we may be able to
|
---|
4078 | determine the result of this comparison. */
|
---|
4079 |
|
---|
4080 | if (const_arg1 == const0_rtx)
|
---|
4081 | {
|
---|
4082 | rtx y = lookup_as_function (folded_arg0, IOR);
|
---|
4083 | rtx inner_const;
|
---|
4084 |
|
---|
4085 | if (y != 0
|
---|
4086 | && (inner_const = equiv_constant (XEXP (y, 1))) != 0
|
---|
4087 | && GET_CODE (inner_const) == CONST_INT
|
---|
4088 | && INTVAL (inner_const) != 0)
|
---|
4089 | {
|
---|
4090 | int sign_bitnum = GET_MODE_BITSIZE (mode_arg0) - 1;
|
---|
4091 | int has_sign = (HOST_BITS_PER_WIDE_INT >= sign_bitnum
|
---|
4092 | && (INTVAL (inner_const)
|
---|
4093 | & ((HOST_WIDE_INT) 1 << sign_bitnum)));
|
---|
4094 | rtx true_rtx = const_true_rtx, false_rtx = const0_rtx;
|
---|
4095 |
|
---|
4096 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
4097 | if (GET_MODE_CLASS (mode) == MODE_FLOAT)
|
---|
4098 | {
|
---|
4099 | true_rtx = (CONST_DOUBLE_FROM_REAL_VALUE
|
---|
4100 | (FLOAT_STORE_FLAG_VALUE (mode), mode));
|
---|
4101 | false_rtx = CONST0_RTX (mode);
|
---|
4102 | }
|
---|
4103 | #endif
|
---|
4104 |
|
---|
4105 | switch (code)
|
---|
4106 | {
|
---|
4107 | case EQ:
|
---|
4108 | return false_rtx;
|
---|
4109 | case NE:
|
---|
4110 | return true_rtx;
|
---|
4111 | case LT: case LE:
|
---|
4112 | if (has_sign)
|
---|
4113 | return true_rtx;
|
---|
4114 | break;
|
---|
4115 | case GT: case GE:
|
---|
4116 | if (has_sign)
|
---|
4117 | return false_rtx;
|
---|
4118 | break;
|
---|
4119 | default:
|
---|
4120 | break;
|
---|
4121 | }
|
---|
4122 | }
|
---|
4123 | }
|
---|
4124 |
|
---|
4125 | new = simplify_relational_operation (code,
|
---|
4126 | (mode_arg0 != VOIDmode
|
---|
4127 | ? mode_arg0
|
---|
4128 | : (GET_MODE (const_arg0
|
---|
4129 | ? const_arg0
|
---|
4130 | : folded_arg0)
|
---|
4131 | != VOIDmode)
|
---|
4132 | ? GET_MODE (const_arg0
|
---|
4133 | ? const_arg0
|
---|
4134 | : folded_arg0)
|
---|
4135 | : GET_MODE (const_arg1
|
---|
4136 | ? const_arg1
|
---|
4137 | : folded_arg1)),
|
---|
4138 | const_arg0 ? const_arg0 : folded_arg0,
|
---|
4139 | const_arg1 ? const_arg1 : folded_arg1);
|
---|
4140 | #ifdef FLOAT_STORE_FLAG_VALUE
|
---|
4141 | if (new != 0 && GET_MODE_CLASS (mode) == MODE_FLOAT)
|
---|
4142 | {
|
---|
4143 | if (new == const0_rtx)
|
---|
4144 | new = CONST0_RTX (mode);
|
---|
4145 | else
|
---|
4146 | new = (CONST_DOUBLE_FROM_REAL_VALUE
|
---|
4147 | (FLOAT_STORE_FLAG_VALUE (mode), mode));
|
---|
4148 | }
|
---|
4149 | #endif
|
---|
4150 | break;
|
---|
4151 |
|
---|
4152 | case '2':
|
---|
4153 | case 'c':
|
---|
4154 | switch (code)
|
---|
4155 | {
|
---|
4156 | case PLUS:
|
---|
4157 | /* If the second operand is a LABEL_REF, see if the first is a MINUS
|
---|
4158 | with that LABEL_REF as its second operand. If so, the result is
|
---|
4159 | the first operand of that MINUS. This handles switches with an
|
---|
4160 | ADDR_DIFF_VEC table. */
|
---|
4161 | if (const_arg1 && GET_CODE (const_arg1) == LABEL_REF)
|
---|
4162 | {
|
---|
4163 | rtx y
|
---|
4164 | = GET_CODE (folded_arg0) == MINUS ? folded_arg0
|
---|
4165 | : lookup_as_function (folded_arg0, MINUS);
|
---|
4166 |
|
---|
4167 | if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
|
---|
4168 | && XEXP (XEXP (y, 1), 0) == XEXP (const_arg1, 0))
|
---|
4169 | return XEXP (y, 0);
|
---|
4170 |
|
---|
4171 | /* Now try for a CONST of a MINUS like the above. */
|
---|
4172 | if ((y = (GET_CODE (folded_arg0) == CONST ? folded_arg0
|
---|
4173 | : lookup_as_function (folded_arg0, CONST))) != 0
|
---|
4174 | && GET_CODE (XEXP (y, 0)) == MINUS
|
---|
4175 | && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
|
---|
4176 | && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg1, 0))
|
---|
4177 | return XEXP (XEXP (y, 0), 0);
|
---|
4178 | }
|
---|
4179 |
|
---|
4180 | /* Likewise if the operands are in the other order. */
|
---|
4181 | if (const_arg0 && GET_CODE (const_arg0) == LABEL_REF)
|
---|
4182 | {
|
---|
4183 | rtx y
|
---|
4184 | = GET_CODE (folded_arg1) == MINUS ? folded_arg1
|
---|
4185 | : lookup_as_function (folded_arg1, MINUS);
|
---|
4186 |
|
---|
4187 | if (y != 0 && GET_CODE (XEXP (y, 1)) == LABEL_REF
|
---|
4188 | && XEXP (XEXP (y, 1), 0) == XEXP (const_arg0, 0))
|
---|
4189 | return XEXP (y, 0);
|
---|
4190 |
|
---|
4191 | /* Now try for a CONST of a MINUS like the above. */
|
---|
4192 | if ((y = (GET_CODE (folded_arg1) == CONST ? folded_arg1
|
---|
4193 | : lookup_as_function (folded_arg1, CONST))) != 0
|
---|
4194 | && GET_CODE (XEXP (y, 0)) == MINUS
|
---|
4195 | && GET_CODE (XEXP (XEXP (y, 0), 1)) == LABEL_REF
|
---|
4196 | && XEXP (XEXP (XEXP (y, 0), 1), 0) == XEXP (const_arg0, 0))
|
---|
4197 | return XEXP (XEXP (y, 0), 0);
|
---|
4198 | }
|
---|
4199 |
|
---|
4200 | /* If second operand is a register equivalent to a negative
|
---|
4201 | CONST_INT, see if we can find a register equivalent to the
|
---|
4202 | positive constant. Make a MINUS if so. Don't do this for
|
---|
4203 | a non-negative constant since we might then alternate between
|
---|
4204 | choosing positive and negative constants. Having the positive
|
---|
4205 | constant previously-used is the more common case. Be sure
|
---|
4206 | the resulting constant is non-negative; if const_arg1 were
|
---|
4207 | the smallest negative number this would overflow: depending
|
---|
4208 | on the mode, this would either just be the same value (and
|
---|
4209 | hence not save anything) or be incorrect. */
|
---|
4210 | if (const_arg1 != 0 && GET_CODE (const_arg1) == CONST_INT
|
---|
4211 | && INTVAL (const_arg1) < 0
|
---|
4212 | /* This used to test
|
---|
4213 |
|
---|
4214 | -INTVAL (const_arg1) >= 0
|
---|
4215 |
|
---|
4216 | But The Sun V5.0 compilers mis-compiled that test. So
|
---|
4217 | instead we test for the problematic value in a more direct
|
---|
4218 | manner and hope the Sun compilers get it correct. */
|
---|
4219 | && INTVAL (const_arg1) !=
|
---|
4220 | ((HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1))
|
---|
4221 | && GET_CODE (folded_arg1) == REG)
|
---|
4222 | {
|
---|
4223 | rtx new_const = GEN_INT (-INTVAL (const_arg1));
|
---|
4224 | struct table_elt *p
|
---|
4225 | = lookup (new_const, safe_hash (new_const, mode) & HASH_MASK,
|
---|
4226 | mode);
|
---|
4227 |
|
---|
4228 | if (p)
|
---|
4229 | for (p = p->first_same_value; p; p = p->next_same_value)
|
---|
4230 | if (GET_CODE (p->exp) == REG)
|
---|
4231 | return simplify_gen_binary (MINUS, mode, folded_arg0,
|
---|
4232 | canon_reg (p->exp, NULL_RTX));
|
---|
4233 | }
|
---|
4234 | goto from_plus;
|
---|
4235 |
|
---|
4236 | case MINUS:
|
---|
4237 | /* If we have (MINUS Y C), see if Y is known to be (PLUS Z C2).
|
---|
4238 | If so, produce (PLUS Z C2-C). */
|
---|
4239 | if (const_arg1 != 0 && GET_CODE (const_arg1) == CONST_INT)
|
---|
4240 | {
|
---|
4241 | rtx y = lookup_as_function (XEXP (x, 0), PLUS);
|
---|
4242 | if (y && GET_CODE (XEXP (y, 1)) == CONST_INT)
|
---|
4243 | return fold_rtx (plus_constant (copy_rtx (y),
|
---|
4244 | -INTVAL (const_arg1)),
|
---|
4245 | NULL_RTX);
|
---|
4246 | }
|
---|
4247 |
|
---|
4248 | /* Fall through. */
|
---|
4249 |
|
---|
4250 | from_plus:
|
---|
4251 | case SMIN: case SMAX: case UMIN: case UMAX:
|
---|
4252 | case IOR: case AND: case XOR:
|
---|
4253 | case MULT:
|
---|
4254 | case ASHIFT: case LSHIFTRT: case ASHIFTRT:
|
---|
4255 | /* If we have (<op> <reg> <const_int>) for an associative OP and REG
|
---|
4256 | is known to be of similar form, we may be able to replace the
|
---|
4257 | operation with a combined operation. This may eliminate the
|
---|
4258 | intermediate operation if every use is simplified in this way.
|
---|
4259 | Note that the similar optimization done by combine.c only works
|
---|
4260 | if the intermediate operation's result has only one reference. */
|
---|
4261 |
|
---|
4262 | if (GET_CODE (folded_arg0) == REG
|
---|
4263 | && const_arg1 && GET_CODE (const_arg1) == CONST_INT)
|
---|
4264 | {
|
---|
4265 | int is_shift
|
---|
4266 | = (code == ASHIFT || code == ASHIFTRT || code == LSHIFTRT);
|
---|
4267 | rtx y = lookup_as_function (folded_arg0, code);
|
---|
4268 | rtx inner_const;
|
---|
4269 | enum rtx_code associate_code;
|
---|
4270 | rtx new_const;
|
---|
4271 |
|
---|
4272 | if (y == 0
|
---|
4273 | || 0 == (inner_const
|
---|
4274 | = equiv_constant (fold_rtx (XEXP (y, 1), 0)))
|
---|
4275 | || GET_CODE (inner_const) != CONST_INT
|
---|
4276 | /* If we have compiled a statement like
|
---|
4277 | "if (x == (x & mask1))", and now are looking at
|
---|
4278 | "x & mask2", we will have a case where the first operand
|
---|
4279 | of Y is the same as our first operand. Unless we detect
|
---|
4280 | this case, an infinite loop will result. */
|
---|
4281 | || XEXP (y, 0) == folded_arg0)
|
---|
4282 | break;
|
---|
4283 |
|
---|
4284 | /* Don't associate these operations if they are a PLUS with the
|
---|
4285 | same constant and it is a power of two. These might be doable
|
---|
4286 | with a pre- or post-increment. Similarly for two subtracts of
|
---|
4287 | identical powers of two with post decrement. */
|
---|
4288 |
|
---|
4289 | if (code == PLUS && INTVAL (const_arg1) == INTVAL (inner_const)
|
---|
4290 | && ((HAVE_PRE_INCREMENT
|
---|
4291 | && exact_log2 (INTVAL (const_arg1)) >= 0)
|
---|
4292 | || (HAVE_POST_INCREMENT
|
---|
4293 | && exact_log2 (INTVAL (const_arg1)) >= 0)
|
---|
4294 | || (HAVE_PRE_DECREMENT
|
---|
4295 | && exact_log2 (- INTVAL (const_arg1)) >= 0)
|
---|
4296 | || (HAVE_POST_DECREMENT
|
---|
4297 | && exact_log2 (- INTVAL (const_arg1)) >= 0)))
|
---|
4298 | break;
|
---|
4299 |
|
---|
4300 | /* Compute the code used to compose the constants. For example,
|
---|
4301 | A-C1-C2 is A-(C1 + C2), so if CODE == MINUS, we want PLUS. */
|
---|
4302 |
|
---|
4303 | associate_code = (is_shift || code == MINUS ? PLUS : code);
|
---|
4304 |
|
---|
4305 | new_const = simplify_binary_operation (associate_code, mode,
|
---|
4306 | const_arg1, inner_const);
|
---|
4307 |
|
---|
4308 | if (new_const == 0)
|
---|
4309 | break;
|
---|
4310 |
|
---|
4311 | /* If we are associating shift operations, don't let this
|
---|
4312 | produce a shift of the size of the object or larger.
|
---|
4313 | This could occur when we follow a sign-extend by a right
|
---|
4314 | shift on a machine that does a sign-extend as a pair
|
---|
4315 | of shifts. */
|
---|
4316 |
|
---|
4317 | if (is_shift && GET_CODE (new_const) == CONST_INT
|
---|
4318 | && INTVAL (new_const) >= GET_MODE_BITSIZE (mode))
|
---|
4319 | {
|
---|
4320 | /* As an exception, we can turn an ASHIFTRT of this
|
---|
4321 | form into a shift of the number of bits - 1. */
|
---|
4322 | if (code == ASHIFTRT)
|
---|
4323 | new_const = GEN_INT (GET_MODE_BITSIZE (mode) - 1);
|
---|
4324 | else
|
---|
4325 | break;
|
---|
4326 | }
|
---|
4327 |
|
---|
4328 | y = copy_rtx (XEXP (y, 0));
|
---|
4329 |
|
---|
4330 | /* If Y contains our first operand (the most common way this
|
---|
4331 | can happen is if Y is a MEM), we would do into an infinite
|
---|
4332 | loop if we tried to fold it. So don't in that case. */
|
---|
4333 |
|
---|
4334 | if (! reg_mentioned_p (folded_arg0, y))
|
---|
4335 | y = fold_rtx (y, insn);
|
---|
4336 |
|
---|
4337 | return simplify_gen_binary (code, mode, y, new_const);
|
---|
4338 | }
|
---|
4339 | break;
|
---|
4340 |
|
---|
4341 | case DIV: case UDIV:
|
---|
4342 | /* ??? The associative optimization performed immediately above is
|
---|
4343 | also possible for DIV and UDIV using associate_code of MULT.
|
---|
4344 | However, we would need extra code to verify that the
|
---|
4345 | multiplication does not overflow, that is, there is no overflow
|
---|
4346 | in the calculation of new_const. */
|
---|
4347 | break;
|
---|
4348 |
|
---|
4349 | default:
|
---|
4350 | break;
|
---|
4351 | }
|
---|
4352 |
|
---|
4353 | new = simplify_binary_operation (code, mode,
|
---|
4354 | const_arg0 ? const_arg0 : folded_arg0,
|
---|
4355 | const_arg1 ? const_arg1 : folded_arg1);
|
---|
4356 | break;
|
---|
4357 |
|
---|
4358 | case 'o':
|
---|
4359 | /* (lo_sum (high X) X) is simply X. */
|
---|
4360 | if (code == LO_SUM && const_arg0 != 0
|
---|
4361 | && GET_CODE (const_arg0) == HIGH
|
---|
4362 | && rtx_equal_p (XEXP (const_arg0, 0), const_arg1))
|
---|
4363 | return const_arg1;
|
---|
4364 | break;
|
---|
4365 |
|
---|
4366 | case '3':
|
---|
4367 | case 'b':
|
---|
4368 | new = simplify_ternary_operation (code, mode, mode_arg0,
|
---|
4369 | const_arg0 ? const_arg0 : folded_arg0,
|
---|
4370 | const_arg1 ? const_arg1 : folded_arg1,
|
---|
4371 | const_arg2 ? const_arg2 : XEXP (x, 2));
|
---|
4372 | break;
|
---|
4373 |
|
---|
4374 | case 'x':
|
---|
4375 | /* Always eliminate CONSTANT_P_RTX at this stage. */
|
---|
4376 | if (code == CONSTANT_P_RTX)
|
---|
4377 | return (const_arg0 ? const1_rtx : const0_rtx);
|
---|
4378 | break;
|
---|
4379 | }
|
---|
4380 |
|
---|
4381 | return new ? new : x;
|
---|
4382 | }
|
---|
4383 | |
---|
4384 |
|
---|
4385 | /* Return a constant value currently equivalent to X.
|
---|
4386 | Return 0 if we don't know one. */
|
---|
4387 |
|
---|
4388 | static rtx
|
---|
4389 | equiv_constant (x)
|
---|
4390 | rtx x;
|
---|
4391 | {
|
---|
4392 | if (GET_CODE (x) == REG
|
---|
4393 | && REGNO_QTY_VALID_P (REGNO (x)))
|
---|
4394 | {
|
---|
4395 | int x_q = REG_QTY (REGNO (x));
|
---|
4396 | struct qty_table_elem *x_ent = &qty_table[x_q];
|
---|
4397 |
|
---|
4398 | if (x_ent->const_rtx)
|
---|
4399 | x = gen_lowpart_if_possible (GET_MODE (x), x_ent->const_rtx);
|
---|
4400 | }
|
---|
4401 |
|
---|
4402 | if (x == 0 || CONSTANT_P (x))
|
---|
4403 | return x;
|
---|
4404 |
|
---|
4405 | /* If X is a MEM, try to fold it outside the context of any insn to see if
|
---|
4406 | it might be equivalent to a constant. That handles the case where it
|
---|
4407 | is a constant-pool reference. Then try to look it up in the hash table
|
---|
4408 | in case it is something whose value we have seen before. */
|
---|
4409 |
|
---|
4410 | if (GET_CODE (x) == MEM)
|
---|
4411 | {
|
---|
4412 | struct table_elt *elt;
|
---|
4413 |
|
---|
4414 | x = fold_rtx (x, NULL_RTX);
|
---|
4415 | if (CONSTANT_P (x))
|
---|
4416 | return x;
|
---|
4417 |
|
---|
4418 | elt = lookup (x, safe_hash (x, GET_MODE (x)) & HASH_MASK, GET_MODE (x));
|
---|
4419 | if (elt == 0)
|
---|
4420 | return 0;
|
---|
4421 |
|
---|
4422 | for (elt = elt->first_same_value; elt; elt = elt->next_same_value)
|
---|
4423 | if (elt->is_const && CONSTANT_P (elt->exp))
|
---|
4424 | return elt->exp;
|
---|
4425 | }
|
---|
4426 |
|
---|
4427 | return 0;
|
---|
4428 | }
|
---|
4429 | |
---|
4430 |
|
---|
4431 | /* Assuming that X is an rtx (e.g., MEM, REG or SUBREG) for a fixed-point
|
---|
4432 | number, return an rtx (MEM, SUBREG, or CONST_INT) that refers to the
|
---|
4433 | least-significant part of X.
|
---|
4434 | MODE specifies how big a part of X to return.
|
---|
4435 |
|
---|
4436 | If the requested operation cannot be done, 0 is returned.
|
---|
4437 |
|
---|
4438 | This is similar to gen_lowpart in emit-rtl.c. */
|
---|
4439 |
|
---|
4440 | rtx
|
---|
4441 | gen_lowpart_if_possible (mode, x)
|
---|
4442 | enum machine_mode mode;
|
---|
4443 | rtx x;
|
---|
4444 | {
|
---|
4445 | rtx result = gen_lowpart_common (mode, x);
|
---|
4446 |
|
---|
4447 | if (result)
|
---|
4448 | return result;
|
---|
4449 | else if (GET_CODE (x) == MEM)
|
---|
4450 | {
|
---|
4451 | /* This is the only other case we handle. */
|
---|
4452 | int offset = 0;
|
---|
4453 | rtx new;
|
---|
4454 |
|
---|
4455 | if (WORDS_BIG_ENDIAN)
|
---|
4456 | offset = (MAX (GET_MODE_SIZE (GET_MODE (x)), UNITS_PER_WORD)
|
---|
4457 | - MAX (GET_MODE_SIZE (mode), UNITS_PER_WORD));
|
---|
4458 | if (BYTES_BIG_ENDIAN)
|
---|
4459 | /* Adjust the address so that the address-after-the-data is
|
---|
4460 | unchanged. */
|
---|
4461 | offset -= (MIN (UNITS_PER_WORD, GET_MODE_SIZE (mode))
|
---|
4462 | - MIN (UNITS_PER_WORD, GET_MODE_SIZE (GET_MODE (x))));
|
---|
4463 |
|
---|
4464 | new = adjust_address_nv (x, mode, offset);
|
---|
4465 | if (! memory_address_p (mode, XEXP (new, 0)))
|
---|
4466 | return 0;
|
---|
4467 |
|
---|
4468 | return new;
|
---|
4469 | }
|
---|
4470 | else
|
---|
4471 | return 0;
|
---|
4472 | }
|
---|
4473 | |
---|
4474 |
|
---|
4475 | /* Given INSN, a jump insn, TAKEN indicates if we are following the "taken"
|
---|
4476 | branch. It will be zero if not.
|
---|
4477 |
|
---|
4478 | In certain cases, this can cause us to add an equivalence. For example,
|
---|
4479 | if we are following the taken case of
|
---|
4480 | if (i == 2)
|
---|
4481 | we can add the fact that `i' and '2' are now equivalent.
|
---|
4482 |
|
---|
4483 | In any case, we can record that this comparison was passed. If the same
|
---|
4484 | comparison is seen later, we will know its value. */
|
---|
4485 |
|
---|
4486 | static void
|
---|
4487 | record_jump_equiv (insn, taken)
|
---|
4488 | rtx insn;
|
---|
4489 | int taken;
|
---|
4490 | {
|
---|
4491 | int cond_known_true;
|
---|
4492 | rtx op0, op1;
|
---|
4493 | rtx set;
|
---|
4494 | enum machine_mode mode, mode0, mode1;
|
---|
4495 | int reversed_nonequality = 0;
|
---|
4496 | enum rtx_code code;
|
---|
4497 |
|
---|
4498 | /* Ensure this is the right kind of insn. */
|
---|
4499 | if (! any_condjump_p (insn))
|
---|
4500 | return;
|
---|
4501 | set = pc_set (insn);
|
---|
4502 |
|
---|
4503 | /* See if this jump condition is known true or false. */
|
---|
4504 | if (taken)
|
---|
4505 | cond_known_true = (XEXP (SET_SRC (set), 2) == pc_rtx);
|
---|
4506 | else
|
---|
4507 | cond_known_true = (XEXP (SET_SRC (set), 1) == pc_rtx);
|
---|
4508 |
|
---|
4509 | /* Get the type of comparison being done and the operands being compared.
|
---|
4510 | If we had to reverse a non-equality condition, record that fact so we
|
---|
4511 | know that it isn't valid for floating-point. */
|
---|
4512 | code = GET_CODE (XEXP (SET_SRC (set), 0));
|
---|
4513 | op0 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 0), insn);
|
---|
4514 | op1 = fold_rtx (XEXP (XEXP (SET_SRC (set), 0), 1), insn);
|
---|
4515 |
|
---|
4516 | code = find_comparison_args (code, &op0, &op1, &mode0, &mode1);
|
---|
4517 | if (! cond_known_true)
|
---|
4518 | {
|
---|
4519 | code = reversed_comparison_code_parts (code, op0, op1, insn);
|
---|
4520 |
|
---|
4521 | /* Don't remember if we can't find the inverse. */
|
---|
4522 | if (code == UNKNOWN)
|
---|
4523 | return;
|
---|
4524 | }
|
---|
4525 |
|
---|
4526 | /* The mode is the mode of the non-constant. */
|
---|
4527 | mode = mode0;
|
---|
4528 | if (mode1 != VOIDmode)
|
---|
4529 | mode = mode1;
|
---|
4530 |
|
---|
4531 | record_jump_cond (code, mode, op0, op1, reversed_nonequality);
|
---|
4532 | }
|
---|
4533 |
|
---|
4534 | /* We know that comparison CODE applied to OP0 and OP1 in MODE is true.
|
---|
4535 | REVERSED_NONEQUALITY is nonzero if CODE had to be swapped.
|
---|
4536 | Make any useful entries we can with that information. Called from
|
---|
4537 | above function and called recursively. */
|
---|
4538 |
|
---|
4539 | static void
|
---|
4540 | record_jump_cond (code, mode, op0, op1, reversed_nonequality)
|
---|
4541 | enum rtx_code code;
|
---|
4542 | enum machine_mode mode;
|
---|
4543 | rtx op0, op1;
|
---|
4544 | int reversed_nonequality;
|
---|
4545 | {
|
---|
4546 | unsigned op0_hash, op1_hash;
|
---|
4547 | int op0_in_memory, op1_in_memory;
|
---|
4548 | struct table_elt *op0_elt, *op1_elt;
|
---|
4549 |
|
---|
4550 | /* If OP0 and OP1 are known equal, and either is a paradoxical SUBREG,
|
---|
4551 | we know that they are also equal in the smaller mode (this is also
|
---|
4552 | true for all smaller modes whether or not there is a SUBREG, but
|
---|
4553 | is not worth testing for with no SUBREG). */
|
---|
4554 |
|
---|
4555 | /* Note that GET_MODE (op0) may not equal MODE. */
|
---|
4556 | if (code == EQ && GET_CODE (op0) == SUBREG
|
---|
4557 | && (GET_MODE_SIZE (GET_MODE (op0))
|
---|
4558 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
|
---|
4559 | {
|
---|
4560 | enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
|
---|
4561 | rtx tem = gen_lowpart_if_possible (inner_mode, op1);
|
---|
4562 |
|
---|
4563 | record_jump_cond (code, mode, SUBREG_REG (op0),
|
---|
4564 | tem ? tem : gen_rtx_SUBREG (inner_mode, op1, 0),
|
---|
4565 | reversed_nonequality);
|
---|
4566 | }
|
---|
4567 |
|
---|
4568 | if (code == EQ && GET_CODE (op1) == SUBREG
|
---|
4569 | && (GET_MODE_SIZE (GET_MODE (op1))
|
---|
4570 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
|
---|
4571 | {
|
---|
4572 | enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
|
---|
4573 | rtx tem = gen_lowpart_if_possible (inner_mode, op0);
|
---|
4574 |
|
---|
4575 | record_jump_cond (code, mode, SUBREG_REG (op1),
|
---|
4576 | tem ? tem : gen_rtx_SUBREG (inner_mode, op0, 0),
|
---|
4577 | reversed_nonequality);
|
---|
4578 | }
|
---|
4579 |
|
---|
4580 | /* Similarly, if this is an NE comparison, and either is a SUBREG
|
---|
4581 | making a smaller mode, we know the whole thing is also NE. */
|
---|
4582 |
|
---|
4583 | /* Note that GET_MODE (op0) may not equal MODE;
|
---|
4584 | if we test MODE instead, we can get an infinite recursion
|
---|
4585 | alternating between two modes each wider than MODE. */
|
---|
4586 |
|
---|
4587 | if (code == NE && GET_CODE (op0) == SUBREG
|
---|
4588 | && subreg_lowpart_p (op0)
|
---|
4589 | && (GET_MODE_SIZE (GET_MODE (op0))
|
---|
4590 | < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
|
---|
4591 | {
|
---|
4592 | enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
|
---|
4593 | rtx tem = gen_lowpart_if_possible (inner_mode, op1);
|
---|
4594 |
|
---|
4595 | record_jump_cond (code, mode, SUBREG_REG (op0),
|
---|
4596 | tem ? tem : gen_rtx_SUBREG (inner_mode, op1, 0),
|
---|
4597 | reversed_nonequality);
|
---|
4598 | }
|
---|
4599 |
|
---|
4600 | if (code == NE && GET_CODE (op1) == SUBREG
|
---|
4601 | && subreg_lowpart_p (op1)
|
---|
4602 | && (GET_MODE_SIZE (GET_MODE (op1))
|
---|
4603 | < GET_MODE_SIZE (GET_MODE (SUBREG_REG (op1)))))
|
---|
4604 | {
|
---|
4605 | enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op1));
|
---|
4606 | rtx tem = gen_lowpart_if_possible (inner_mode, op0);
|
---|
4607 |
|
---|
4608 | record_jump_cond (code, mode, SUBREG_REG (op1),
|
---|
4609 | tem ? tem : gen_rtx_SUBREG (inner_mode, op0, 0),
|
---|
4610 | reversed_nonequality);
|
---|
4611 | }
|
---|
4612 |
|
---|
4613 | /* Hash both operands. */
|
---|
4614 |
|
---|
4615 | do_not_record = 0;
|
---|
4616 | hash_arg_in_memory = 0;
|
---|
4617 | op0_hash = HASH (op0, mode);
|
---|
4618 | op0_in_memory = hash_arg_in_memory;
|
---|
4619 |
|
---|
4620 | if (do_not_record)
|
---|
4621 | return;
|
---|
4622 |
|
---|
4623 | do_not_record = 0;
|
---|
4624 | hash_arg_in_memory = 0;
|
---|
4625 | op1_hash = HASH (op1, mode);
|
---|
4626 | op1_in_memory = hash_arg_in_memory;
|
---|
4627 |
|
---|
4628 | if (do_not_record)
|
---|
4629 | return;
|
---|
4630 |
|
---|
4631 | /* Look up both operands. */
|
---|
4632 | op0_elt = lookup (op0, op0_hash, mode);
|
---|
4633 | op1_elt = lookup (op1, op1_hash, mode);
|
---|
4634 |
|
---|
4635 | /* If both operands are already equivalent or if they are not in the
|
---|
4636 | table but are identical, do nothing. */
|
---|
4637 | if ((op0_elt != 0 && op1_elt != 0
|
---|
4638 | && op0_elt->first_same_value == op1_elt->first_same_value)
|
---|
4639 | || op0 == op1 || rtx_equal_p (op0, op1))
|
---|
4640 | return;
|
---|
4641 |
|
---|
4642 | /* If we aren't setting two things equal all we can do is save this
|
---|
4643 | comparison. Similarly if this is floating-point. In the latter
|
---|
4644 | case, OP1 might be zero and both -0.0 and 0.0 are equal to it.
|
---|
4645 | If we record the equality, we might inadvertently delete code
|
---|
4646 | whose intent was to change -0 to +0. */
|
---|
4647 |
|
---|
4648 | if (code != EQ || FLOAT_MODE_P (GET_MODE (op0)))
|
---|
4649 | {
|
---|
4650 | struct qty_table_elem *ent;
|
---|
4651 | int qty;
|
---|
4652 |
|
---|
4653 | /* If we reversed a floating-point comparison, if OP0 is not a
|
---|
4654 | register, or if OP1 is neither a register or constant, we can't
|
---|
4655 | do anything. */
|
---|
4656 |
|
---|
4657 | if (GET_CODE (op1) != REG)
|
---|
4658 | op1 = equiv_constant (op1);
|
---|
4659 |
|
---|
4660 | if ((reversed_nonequality && FLOAT_MODE_P (mode))
|
---|
4661 | || GET_CODE (op0) != REG || op1 == 0)
|
---|
4662 | return;
|
---|
4663 |
|
---|
4664 | /* Put OP0 in the hash table if it isn't already. This gives it a
|
---|
4665 | new quantity number. */
|
---|
4666 | if (op0_elt == 0)
|
---|
4667 | {
|
---|
4668 | if (insert_regs (op0, NULL, 0))
|
---|
4669 | {
|
---|
4670 | rehash_using_reg (op0);
|
---|
4671 | op0_hash = HASH (op0, mode);
|
---|
4672 |
|
---|
4673 | /* If OP0 is contained in OP1, this changes its hash code
|
---|
4674 | as well. Faster to rehash than to check, except
|
---|
4675 | for the simple case of a constant. */
|
---|
4676 | if (! CONSTANT_P (op1))
|
---|
4677 | op1_hash = HASH (op1,mode);
|
---|
4678 | }
|
---|
4679 |
|
---|
4680 | op0_elt = insert (op0, NULL, op0_hash, mode);
|
---|
4681 | op0_elt->in_memory = op0_in_memory;
|
---|
4682 | }
|
---|
4683 |
|
---|
4684 | qty = REG_QTY (REGNO (op0));
|
---|
4685 | ent = &qty_table[qty];
|
---|
4686 |
|
---|
4687 | ent->comparison_code = code;
|
---|
4688 | if (GET_CODE (op1) == REG)
|
---|
4689 | {
|
---|
4690 | /* Look it up again--in case op0 and op1 are the same. */
|
---|
4691 | op1_elt = lookup (op1, op1_hash, mode);
|
---|
4692 |
|
---|
4693 | /* Put OP1 in the hash table so it gets a new quantity number. */
|
---|
4694 | if (op1_elt == 0)
|
---|
4695 | {
|
---|
4696 | if (insert_regs (op1, NULL, 0))
|
---|
4697 | {
|
---|
4698 | rehash_using_reg (op1);
|
---|
4699 | op1_hash = HASH (op1, mode);
|
---|
4700 | }
|
---|
4701 |
|
---|
4702 | op1_elt = insert (op1, NULL, op1_hash, mode);
|
---|
4703 | op1_elt->in_memory = op1_in_memory;
|
---|
4704 | }
|
---|
4705 |
|
---|
4706 | ent->comparison_const = NULL_RTX;
|
---|
4707 | ent->comparison_qty = REG_QTY (REGNO (op1));
|
---|
4708 | }
|
---|
4709 | else
|
---|
4710 | {
|
---|
4711 | ent->comparison_const = op1;
|
---|
4712 | ent->comparison_qty = -1;
|
---|
4713 | }
|
---|
4714 |
|
---|
4715 | return;
|
---|
4716 | }
|
---|
4717 |
|
---|
4718 | /* If either side is still missing an equivalence, make it now,
|
---|
4719 | then merge the equivalences. */
|
---|
4720 |
|
---|
4721 | if (op0_elt == 0)
|
---|
4722 | {
|
---|
4723 | if (insert_regs (op0, NULL, 0))
|
---|
4724 | {
|
---|
4725 | rehash_using_reg (op0);
|
---|
4726 | op0_hash = HASH (op0, mode);
|
---|
4727 | }
|
---|
4728 |
|
---|
4729 | op0_elt = insert (op0, NULL, op0_hash, mode);
|
---|
4730 | op0_elt->in_memory = op0_in_memory;
|
---|
4731 | }
|
---|
4732 |
|
---|
4733 | if (op1_elt == 0)
|
---|
4734 | {
|
---|
4735 | if (insert_regs (op1, NULL, 0))
|
---|
4736 | {
|
---|
4737 | rehash_using_reg (op1);
|
---|
4738 | op1_hash = HASH (op1, mode);
|
---|
4739 | }
|
---|
4740 |
|
---|
4741 | op1_elt = insert (op1, NULL, op1_hash, mode);
|
---|
4742 | op1_elt->in_memory = op1_in_memory;
|
---|
4743 | }
|
---|
4744 |
|
---|
4745 | merge_equiv_classes (op0_elt, op1_elt);
|
---|
4746 | last_jump_equiv_class = op0_elt;
|
---|
4747 | }
|
---|
4748 | |
---|
4749 |
|
---|
4750 | /* CSE processing for one instruction.
|
---|
4751 | First simplify sources and addresses of all assignments
|
---|
4752 | in the instruction, using previously-computed equivalents values.
|
---|
4753 | Then install the new sources and destinations in the table
|
---|
4754 | of available values.
|
---|
4755 |
|
---|
4756 | If LIBCALL_INSN is nonzero, don't record any equivalence made in
|
---|
4757 | the insn. It means that INSN is inside libcall block. In this
|
---|
4758 | case LIBCALL_INSN is the corresponding insn with REG_LIBCALL. */
|
---|
4759 |
|
---|
4760 | /* Data on one SET contained in the instruction. */
|
---|
4761 |
|
---|
4762 | struct set
|
---|
4763 | {
|
---|
4764 | /* The SET rtx itself. */
|
---|
4765 | rtx rtl;
|
---|
4766 | /* The SET_SRC of the rtx (the original value, if it is changing). */
|
---|
4767 | rtx src;
|
---|
4768 | /* The hash-table element for the SET_SRC of the SET. */
|
---|
4769 | struct table_elt *src_elt;
|
---|
4770 | /* Hash value for the SET_SRC. */
|
---|
4771 | unsigned src_hash;
|
---|
4772 | /* Hash value for the SET_DEST. */
|
---|
4773 | unsigned dest_hash;
|
---|
4774 | /* The SET_DEST, with SUBREG, etc., stripped. */
|
---|
4775 | rtx inner_dest;
|
---|
4776 | /* Nonzero if the SET_SRC is in memory. */
|
---|
4777 | char src_in_memory;
|
---|
4778 | /* Nonzero if the SET_SRC contains something
|
---|
4779 | whose value cannot be predicted and understood. */
|
---|
4780 | char src_volatile;
|
---|
4781 | /* Original machine mode, in case it becomes a CONST_INT. */
|
---|
4782 | enum machine_mode mode;
|
---|
4783 | /* A constant equivalent for SET_SRC, if any. */
|
---|
4784 | rtx src_const;
|
---|
4785 | /* Original SET_SRC value used for libcall notes. */
|
---|
4786 | rtx orig_src;
|
---|
4787 | /* Hash value of constant equivalent for SET_SRC. */
|
---|
4788 | unsigned src_const_hash;
|
---|
4789 | /* Table entry for constant equivalent for SET_SRC, if any. */
|
---|
4790 | struct table_elt *src_const_elt;
|
---|
4791 | };
|
---|
4792 |
|
---|
4793 | static void
|
---|
4794 | cse_insn (insn, libcall_insn)
|
---|
4795 | rtx insn;
|
---|
4796 | rtx libcall_insn;
|
---|
4797 | {
|
---|
4798 | rtx x = PATTERN (insn);
|
---|
4799 | int i;
|
---|
4800 | rtx tem;
|
---|
4801 | int n_sets = 0;
|
---|
4802 |
|
---|
4803 | #ifdef HAVE_cc0
|
---|
4804 | /* Records what this insn does to set CC0. */
|
---|
4805 | rtx this_insn_cc0 = 0;
|
---|
4806 | enum machine_mode this_insn_cc0_mode = VOIDmode;
|
---|
4807 | #endif
|
---|
4808 |
|
---|
4809 | rtx src_eqv = 0;
|
---|
4810 | struct table_elt *src_eqv_elt = 0;
|
---|
4811 | int src_eqv_volatile = 0;
|
---|
4812 | int src_eqv_in_memory = 0;
|
---|
4813 | unsigned src_eqv_hash = 0;
|
---|
4814 |
|
---|
4815 | struct set *sets = (struct set *) 0;
|
---|
4816 |
|
---|
4817 | this_insn = insn;
|
---|
4818 |
|
---|
4819 | /* Find all the SETs and CLOBBERs in this instruction.
|
---|
4820 | Record all the SETs in the array `set' and count them.
|
---|
4821 | Also determine whether there is a CLOBBER that invalidates
|
---|
4822 | all memory references, or all references at varying addresses. */
|
---|
4823 |
|
---|
4824 | if (GET_CODE (insn) == CALL_INSN)
|
---|
4825 | {
|
---|
4826 | for (tem = CALL_INSN_FUNCTION_USAGE (insn); tem; tem = XEXP (tem, 1))
|
---|
4827 | {
|
---|
4828 | if (GET_CODE (XEXP (tem, 0)) == CLOBBER)
|
---|
4829 | invalidate (SET_DEST (XEXP (tem, 0)), VOIDmode);
|
---|
4830 | XEXP (tem, 0) = canon_reg (XEXP (tem, 0), insn);
|
---|
4831 | }
|
---|
4832 | }
|
---|
4833 |
|
---|
4834 | if (GET_CODE (x) == SET)
|
---|
4835 | {
|
---|
4836 | sets = (struct set *) alloca (sizeof (struct set));
|
---|
4837 | sets[0].rtl = x;
|
---|
4838 |
|
---|
4839 | /* Ignore SETs that are unconditional jumps.
|
---|
4840 | They never need cse processing, so this does not hurt.
|
---|
4841 | The reason is not efficiency but rather
|
---|
4842 | so that we can test at the end for instructions
|
---|
4843 | that have been simplified to unconditional jumps
|
---|
4844 | and not be misled by unchanged instructions
|
---|
4845 | that were unconditional jumps to begin with. */
|
---|
4846 | if (SET_DEST (x) == pc_rtx
|
---|
4847 | && GET_CODE (SET_SRC (x)) == LABEL_REF)
|
---|
4848 | ;
|
---|
4849 |
|
---|
4850 | /* Don't count call-insns, (set (reg 0) (call ...)), as a set.
|
---|
4851 | The hard function value register is used only once, to copy to
|
---|
4852 | someplace else, so it isn't worth cse'ing (and on 80386 is unsafe)!
|
---|
4853 | Ensure we invalidate the destination register. On the 80386 no
|
---|
4854 | other code would invalidate it since it is a fixed_reg.
|
---|
4855 | We need not check the return of apply_change_group; see canon_reg. */
|
---|
4856 |
|
---|
4857 | else if (GET_CODE (SET_SRC (x)) == CALL)
|
---|
4858 | {
|
---|
4859 | canon_reg (SET_SRC (x), insn);
|
---|
4860 | apply_change_group ();
|
---|
4861 | fold_rtx (SET_SRC (x), insn);
|
---|
4862 | invalidate (SET_DEST (x), VOIDmode);
|
---|
4863 | }
|
---|
4864 | else
|
---|
4865 | n_sets = 1;
|
---|
4866 | }
|
---|
4867 | else if (GET_CODE (x) == PARALLEL)
|
---|
4868 | {
|
---|
4869 | int lim = XVECLEN (x, 0);
|
---|
4870 |
|
---|
4871 | sets = (struct set *) alloca (lim * sizeof (struct set));
|
---|
4872 |
|
---|
4873 | /* Find all regs explicitly clobbered in this insn,
|
---|
4874 | and ensure they are not replaced with any other regs
|
---|
4875 | elsewhere in this insn.
|
---|
4876 | When a reg that is clobbered is also used for input,
|
---|
4877 | we should presume that that is for a reason,
|
---|
4878 | and we should not substitute some other register
|
---|
4879 | which is not supposed to be clobbered.
|
---|
4880 | Therefore, this loop cannot be merged into the one below
|
---|
4881 | because a CALL may precede a CLOBBER and refer to the
|
---|
4882 | value clobbered. We must not let a canonicalization do
|
---|
4883 | anything in that case. */
|
---|
4884 | for (i = 0; i < lim; i++)
|
---|
4885 | {
|
---|
4886 | rtx y = XVECEXP (x, 0, i);
|
---|
4887 | if (GET_CODE (y) == CLOBBER)
|
---|
4888 | {
|
---|
4889 | rtx clobbered = XEXP (y, 0);
|
---|
4890 |
|
---|
4891 | if (GET_CODE (clobbered) == REG
|
---|
4892 | || GET_CODE (clobbered) == SUBREG)
|
---|
4893 | invalidate (clobbered, VOIDmode);
|
---|
4894 | else if (GET_CODE (clobbered) == STRICT_LOW_PART
|
---|
4895 | || GET_CODE (clobbered) == ZERO_EXTRACT)
|
---|
4896 | invalidate (XEXP (clobbered, 0), GET_MODE (clobbered));
|
---|
4897 | }
|
---|
4898 | }
|
---|
4899 |
|
---|
4900 | for (i = 0; i < lim; i++)
|
---|
4901 | {
|
---|
4902 | rtx y = XVECEXP (x, 0, i);
|
---|
4903 | if (GET_CODE (y) == SET)
|
---|
4904 | {
|
---|
4905 | /* As above, we ignore unconditional jumps and call-insns and
|
---|
4906 | ignore the result of apply_change_group. */
|
---|
4907 | if (GET_CODE (SET_SRC (y)) == CALL)
|
---|
4908 | {
|
---|
4909 | canon_reg (SET_SRC (y), insn);
|
---|
4910 | apply_change_group ();
|
---|
4911 | fold_rtx (SET_SRC (y), insn);
|
---|
4912 | invalidate (SET_DEST (y), VOIDmode);
|
---|
4913 | }
|
---|
4914 | else if (SET_DEST (y) == pc_rtx
|
---|
4915 | && GET_CODE (SET_SRC (y)) == LABEL_REF)
|
---|
4916 | ;
|
---|
4917 | else
|
---|
4918 | sets[n_sets++].rtl = y;
|
---|
4919 | }
|
---|
4920 | else if (GET_CODE (y) == CLOBBER)
|
---|
4921 | {
|
---|
4922 | /* If we clobber memory, canon the address.
|
---|
4923 | This does nothing when a register is clobbered
|
---|
4924 | because we have already invalidated the reg. */
|
---|
4925 | if (GET_CODE (XEXP (y, 0)) == MEM)
|
---|
4926 | canon_reg (XEXP (y, 0), NULL_RTX);
|
---|
4927 | }
|
---|
4928 | else if (GET_CODE (y) == USE
|
---|
4929 | && ! (GET_CODE (XEXP (y, 0)) == REG
|
---|
4930 | && REGNO (XEXP (y, 0)) < FIRST_PSEUDO_REGISTER))
|
---|
4931 | canon_reg (y, NULL_RTX);
|
---|
4932 | else if (GET_CODE (y) == CALL)
|
---|
4933 | {
|
---|
4934 | /* The result of apply_change_group can be ignored; see
|
---|
4935 | canon_reg. */
|
---|
4936 | canon_reg (y, insn);
|
---|
4937 | apply_change_group ();
|
---|
4938 | fold_rtx (y, insn);
|
---|
4939 | }
|
---|
4940 | }
|
---|
4941 | }
|
---|
4942 | else if (GET_CODE (x) == CLOBBER)
|
---|
4943 | {
|
---|
4944 | if (GET_CODE (XEXP (x, 0)) == MEM)
|
---|
4945 | canon_reg (XEXP (x, 0), NULL_RTX);
|
---|
4946 | }
|
---|
4947 |
|
---|
4948 | /* Canonicalize a USE of a pseudo register or memory location. */
|
---|
4949 | else if (GET_CODE (x) == USE
|
---|
4950 | && ! (GET_CODE (XEXP (x, 0)) == REG
|
---|
4951 | && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER))
|
---|
4952 | canon_reg (XEXP (x, 0), NULL_RTX);
|
---|
4953 | else if (GET_CODE (x) == CALL)
|
---|
4954 | {
|
---|
4955 | /* The result of apply_change_group can be ignored; see canon_reg. */
|
---|
4956 | canon_reg (x, insn);
|
---|
4957 | apply_change_group ();
|
---|
4958 | fold_rtx (x, insn);
|
---|
4959 | }
|
---|
4960 |
|
---|
4961 | /* Store the equivalent value in SRC_EQV, if different, or if the DEST
|
---|
4962 | is a STRICT_LOW_PART. The latter condition is necessary because SRC_EQV
|
---|
4963 | is handled specially for this case, and if it isn't set, then there will
|
---|
4964 | be no equivalence for the destination. */
|
---|
4965 | if (n_sets == 1 && REG_NOTES (insn) != 0
|
---|
4966 | && (tem = find_reg_note (insn, REG_EQUAL, NULL_RTX)) != 0
|
---|
4967 | && (! rtx_equal_p (XEXP (tem, 0), SET_SRC (sets[0].rtl))
|
---|
4968 | || GET_CODE (SET_DEST (sets[0].rtl)) == STRICT_LOW_PART))
|
---|
4969 | {
|
---|
4970 | src_eqv = fold_rtx (canon_reg (XEXP (tem, 0), NULL_RTX), insn);
|
---|
4971 | XEXP (tem, 0) = src_eqv;
|
---|
4972 | }
|
---|
4973 |
|
---|
4974 | /* Canonicalize sources and addresses of destinations.
|
---|
4975 | We do this in a separate pass to avoid problems when a MATCH_DUP is
|
---|
4976 | present in the insn pattern. In that case, we want to ensure that
|
---|
4977 | we don't break the duplicate nature of the pattern. So we will replace
|
---|
4978 | both operands at the same time. Otherwise, we would fail to find an
|
---|
4979 | equivalent substitution in the loop calling validate_change below.
|
---|
4980 |
|
---|
4981 | We used to suppress canonicalization of DEST if it appears in SRC,
|
---|
4982 | but we don't do this any more. */
|
---|
4983 |
|
---|
4984 | for (i = 0; i < n_sets; i++)
|
---|
4985 | {
|
---|
4986 | rtx dest = SET_DEST (sets[i].rtl);
|
---|
4987 | rtx src = SET_SRC (sets[i].rtl);
|
---|
4988 | rtx new = canon_reg (src, insn);
|
---|
4989 | int insn_code;
|
---|
4990 |
|
---|
4991 | sets[i].orig_src = src;
|
---|
4992 | if ((GET_CODE (new) == REG && GET_CODE (src) == REG
|
---|
4993 | && ((REGNO (new) < FIRST_PSEUDO_REGISTER)
|
---|
4994 | != (REGNO (src) < FIRST_PSEUDO_REGISTER)))
|
---|
4995 | || (insn_code = recog_memoized (insn)) < 0
|
---|
4996 | || insn_data[insn_code].n_dups > 0)
|
---|
4997 | validate_change (insn, &SET_SRC (sets[i].rtl), new, 1);
|
---|
4998 | else
|
---|
4999 | SET_SRC (sets[i].rtl) = new;
|
---|
5000 |
|
---|
5001 | if (GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == SIGN_EXTRACT)
|
---|
5002 | {
|
---|
5003 | validate_change (insn, &XEXP (dest, 1),
|
---|
5004 | canon_reg (XEXP (dest, 1), insn), 1);
|
---|
5005 | validate_change (insn, &XEXP (dest, 2),
|
---|
5006 | canon_reg (XEXP (dest, 2), insn), 1);
|
---|
5007 | }
|
---|
5008 |
|
---|
5009 | while (GET_CODE (dest) == SUBREG || GET_CODE (dest) == STRICT_LOW_PART
|
---|
5010 | || GET_CODE (dest) == ZERO_EXTRACT
|
---|
5011 | || GET_CODE (dest) == SIGN_EXTRACT)
|
---|
5012 | dest = XEXP (dest, 0);
|
---|
5013 |
|
---|
5014 | if (GET_CODE (dest) == MEM)
|
---|
5015 | canon_reg (dest, insn);
|
---|
5016 | }
|
---|
5017 |
|
---|
5018 | /* Now that we have done all the replacements, we can apply the change
|
---|
5019 | group and see if they all work. Note that this will cause some
|
---|
5020 | canonicalizations that would have worked individually not to be applied
|
---|
5021 | because some other canonicalization didn't work, but this should not
|
---|
5022 | occur often.
|
---|
5023 |
|
---|
5024 | The result of apply_change_group can be ignored; see canon_reg. */
|
---|
5025 |
|
---|
5026 | apply_change_group ();
|
---|
5027 |
|
---|
5028 | /* Set sets[i].src_elt to the class each source belongs to.
|
---|
5029 | Detect assignments from or to volatile things
|
---|
5030 | and set set[i] to zero so they will be ignored
|
---|
5031 | in the rest of this function.
|
---|
5032 |
|
---|
5033 | Nothing in this loop changes the hash table or the register chains. */
|
---|
5034 |
|
---|
5035 | for (i = 0; i < n_sets; i++)
|
---|
5036 | {
|
---|
5037 | rtx src, dest;
|
---|
5038 | rtx src_folded;
|
---|
5039 | struct table_elt *elt = 0, *p;
|
---|
5040 | enum machine_mode mode;
|
---|
5041 | rtx src_eqv_here;
|
---|
5042 | rtx src_const = 0;
|
---|
5043 | rtx src_related = 0;
|
---|
5044 | struct table_elt *src_const_elt = 0;
|
---|
5045 | int src_cost = MAX_COST;
|
---|
5046 | int src_eqv_cost = MAX_COST;
|
---|
5047 | int src_folded_cost = MAX_COST;
|
---|
5048 | int src_related_cost = MAX_COST;
|
---|
5049 | int src_elt_cost = MAX_COST;
|
---|
5050 | int src_regcost = MAX_COST;
|
---|
5051 | int src_eqv_regcost = MAX_COST;
|
---|
5052 | int src_folded_regcost = MAX_COST;
|
---|
5053 | int src_related_regcost = MAX_COST;
|
---|
5054 | int src_elt_regcost = MAX_COST;
|
---|
5055 | /* Set nonzero if we need to call force_const_mem on with the
|
---|
5056 | contents of src_folded before using it. */
|
---|
5057 | int src_folded_force_flag = 0;
|
---|
5058 |
|
---|
5059 | dest = SET_DEST (sets[i].rtl);
|
---|
5060 | src = SET_SRC (sets[i].rtl);
|
---|
5061 |
|
---|
5062 | /* If SRC is a constant that has no machine mode,
|
---|
5063 | hash it with the destination's machine mode.
|
---|
5064 | This way we can keep different modes separate. */
|
---|
5065 |
|
---|
5066 | mode = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
|
---|
5067 | sets[i].mode = mode;
|
---|
5068 |
|
---|
5069 | if (src_eqv)
|
---|
5070 | {
|
---|
5071 | enum machine_mode eqvmode = mode;
|
---|
5072 | if (GET_CODE (dest) == STRICT_LOW_PART)
|
---|
5073 | eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
|
---|
5074 | do_not_record = 0;
|
---|
5075 | hash_arg_in_memory = 0;
|
---|
5076 | src_eqv_hash = HASH (src_eqv, eqvmode);
|
---|
5077 |
|
---|
5078 | /* Find the equivalence class for the equivalent expression. */
|
---|
5079 |
|
---|
5080 | if (!do_not_record)
|
---|
5081 | src_eqv_elt = lookup (src_eqv, src_eqv_hash, eqvmode);
|
---|
5082 |
|
---|
5083 | src_eqv_volatile = do_not_record;
|
---|
5084 | src_eqv_in_memory = hash_arg_in_memory;
|
---|
5085 | }
|
---|
5086 |
|
---|
5087 | /* If this is a STRICT_LOW_PART assignment, src_eqv corresponds to the
|
---|
5088 | value of the INNER register, not the destination. So it is not
|
---|
5089 | a valid substitution for the source. But save it for later. */
|
---|
5090 | if (GET_CODE (dest) == STRICT_LOW_PART)
|
---|
5091 | src_eqv_here = 0;
|
---|
5092 | else
|
---|
5093 | src_eqv_here = src_eqv;
|
---|
5094 |
|
---|
5095 | /* Simplify and foldable subexpressions in SRC. Then get the fully-
|
---|
5096 | simplified result, which may not necessarily be valid. */
|
---|
5097 | src_folded = fold_rtx (src, insn);
|
---|
5098 |
|
---|
5099 | #if 0
|
---|
5100 | /* ??? This caused bad code to be generated for the m68k port with -O2.
|
---|
5101 | Suppose src is (CONST_INT -1), and that after truncation src_folded
|
---|
5102 | is (CONST_INT 3). Suppose src_folded is then used for src_const.
|
---|
5103 | At the end we will add src and src_const to the same equivalence
|
---|
5104 | class. We now have 3 and -1 on the same equivalence class. This
|
---|
5105 | causes later instructions to be mis-optimized. */
|
---|
5106 | /* If storing a constant in a bitfield, pre-truncate the constant
|
---|
5107 | so we will be able to record it later. */
|
---|
5108 | if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT
|
---|
5109 | || GET_CODE (SET_DEST (sets[i].rtl)) == SIGN_EXTRACT)
|
---|
5110 | {
|
---|
5111 | rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
|
---|
5112 |
|
---|
5113 | if (GET_CODE (src) == CONST_INT
|
---|
5114 | && GET_CODE (width) == CONST_INT
|
---|
5115 | && INTVAL (width) < HOST_BITS_PER_WIDE_INT
|
---|
5116 | && (INTVAL (src) & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
|
---|
5117 | src_folded
|
---|
5118 | = GEN_INT (INTVAL (src) & (((HOST_WIDE_INT) 1
|
---|
5119 | << INTVAL (width)) - 1));
|
---|
5120 | }
|
---|
5121 | #endif
|
---|
5122 |
|
---|
5123 | /* Compute SRC's hash code, and also notice if it
|
---|
5124 | should not be recorded at all. In that case,
|
---|
5125 | prevent any further processing of this assignment. */
|
---|
5126 | do_not_record = 0;
|
---|
5127 | hash_arg_in_memory = 0;
|
---|
5128 |
|
---|
5129 | sets[i].src = src;
|
---|
5130 | sets[i].src_hash = HASH (src, mode);
|
---|
5131 | sets[i].src_volatile = do_not_record;
|
---|
5132 | sets[i].src_in_memory = hash_arg_in_memory;
|
---|
5133 |
|
---|
5134 | /* If SRC is a MEM, there is a REG_EQUIV note for SRC, and DEST is
|
---|
5135 | a pseudo, do not record SRC. Using SRC as a replacement for
|
---|
5136 | anything else will be incorrect in that situation. Note that
|
---|
5137 | this usually occurs only for stack slots, in which case all the
|
---|
5138 | RTL would be referring to SRC, so we don't lose any optimization
|
---|
5139 | opportunities by not having SRC in the hash table. */
|
---|
5140 |
|
---|
5141 | if (GET_CODE (src) == MEM
|
---|
5142 | && find_reg_note (insn, REG_EQUIV, NULL_RTX) != 0
|
---|
5143 | && GET_CODE (dest) == REG
|
---|
5144 | && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
|
---|
5145 | sets[i].src_volatile = 1;
|
---|
5146 |
|
---|
5147 | #if 0
|
---|
5148 | /* It is no longer clear why we used to do this, but it doesn't
|
---|
5149 | appear to still be needed. So let's try without it since this
|
---|
5150 | code hurts cse'ing widened ops. */
|
---|
5151 | /* If source is a perverse subreg (such as QI treated as an SI),
|
---|
5152 | treat it as volatile. It may do the work of an SI in one context
|
---|
5153 | where the extra bits are not being used, but cannot replace an SI
|
---|
5154 | in general. */
|
---|
5155 | if (GET_CODE (src) == SUBREG
|
---|
5156 | && (GET_MODE_SIZE (GET_MODE (src))
|
---|
5157 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))))
|
---|
5158 | sets[i].src_volatile = 1;
|
---|
5159 | #endif
|
---|
5160 |
|
---|
5161 | /* Locate all possible equivalent forms for SRC. Try to replace
|
---|
5162 | SRC in the insn with each cheaper equivalent.
|
---|
5163 |
|
---|
5164 | We have the following types of equivalents: SRC itself, a folded
|
---|
5165 | version, a value given in a REG_EQUAL note, or a value related
|
---|
5166 | to a constant.
|
---|
5167 |
|
---|
5168 | Each of these equivalents may be part of an additional class
|
---|
5169 | of equivalents (if more than one is in the table, they must be in
|
---|
5170 | the same class; we check for this).
|
---|
5171 |
|
---|
5172 | If the source is volatile, we don't do any table lookups.
|
---|
5173 |
|
---|
5174 | We note any constant equivalent for possible later use in a
|
---|
5175 | REG_NOTE. */
|
---|
5176 |
|
---|
5177 | if (!sets[i].src_volatile)
|
---|
5178 | elt = lookup (src, sets[i].src_hash, mode);
|
---|
5179 |
|
---|
5180 | sets[i].src_elt = elt;
|
---|
5181 |
|
---|
5182 | if (elt && src_eqv_here && src_eqv_elt)
|
---|
5183 | {
|
---|
5184 | if (elt->first_same_value != src_eqv_elt->first_same_value)
|
---|
5185 | {
|
---|
5186 | /* The REG_EQUAL is indicating that two formerly distinct
|
---|
5187 | classes are now equivalent. So merge them. */
|
---|
5188 | merge_equiv_classes (elt, src_eqv_elt);
|
---|
5189 | src_eqv_hash = HASH (src_eqv, elt->mode);
|
---|
5190 | src_eqv_elt = lookup (src_eqv, src_eqv_hash, elt->mode);
|
---|
5191 | }
|
---|
5192 |
|
---|
5193 | src_eqv_here = 0;
|
---|
5194 | }
|
---|
5195 |
|
---|
5196 | else if (src_eqv_elt)
|
---|
5197 | elt = src_eqv_elt;
|
---|
5198 |
|
---|
5199 | /* Try to find a constant somewhere and record it in `src_const'.
|
---|
5200 | Record its table element, if any, in `src_const_elt'. Look in
|
---|
5201 | any known equivalences first. (If the constant is not in the
|
---|
5202 | table, also set `sets[i].src_const_hash'). */
|
---|
5203 | if (elt)
|
---|
5204 | for (p = elt->first_same_value; p; p = p->next_same_value)
|
---|
5205 | if (p->is_const)
|
---|
5206 | {
|
---|
5207 | src_const = p->exp;
|
---|
5208 | src_const_elt = elt;
|
---|
5209 | break;
|
---|
5210 | }
|
---|
5211 |
|
---|
5212 | if (src_const == 0
|
---|
5213 | && (CONSTANT_P (src_folded)
|
---|
5214 | /* Consider (minus (label_ref L1) (label_ref L2)) as
|
---|
5215 | "constant" here so we will record it. This allows us
|
---|
5216 | to fold switch statements when an ADDR_DIFF_VEC is used. */
|
---|
5217 | || (GET_CODE (src_folded) == MINUS
|
---|
5218 | && GET_CODE (XEXP (src_folded, 0)) == LABEL_REF
|
---|
5219 | && GET_CODE (XEXP (src_folded, 1)) == LABEL_REF)))
|
---|
5220 | src_const = src_folded, src_const_elt = elt;
|
---|
5221 | else if (src_const == 0 && src_eqv_here && CONSTANT_P (src_eqv_here))
|
---|
5222 | src_const = src_eqv_here, src_const_elt = src_eqv_elt;
|
---|
5223 |
|
---|
5224 | /* If we don't know if the constant is in the table, get its
|
---|
5225 | hash code and look it up. */
|
---|
5226 | if (src_const && src_const_elt == 0)
|
---|
5227 | {
|
---|
5228 | sets[i].src_const_hash = HASH (src_const, mode);
|
---|
5229 | src_const_elt = lookup (src_const, sets[i].src_const_hash, mode);
|
---|
5230 | }
|
---|
5231 |
|
---|
5232 | sets[i].src_const = src_const;
|
---|
5233 | sets[i].src_const_elt = src_const_elt;
|
---|
5234 |
|
---|
5235 | /* If the constant and our source are both in the table, mark them as
|
---|
5236 | equivalent. Otherwise, if a constant is in the table but the source
|
---|
5237 | isn't, set ELT to it. */
|
---|
5238 | if (src_const_elt && elt
|
---|
5239 | && src_const_elt->first_same_value != elt->first_same_value)
|
---|
5240 | merge_equiv_classes (elt, src_const_elt);
|
---|
5241 | else if (src_const_elt && elt == 0)
|
---|
5242 | elt = src_const_elt;
|
---|
5243 |
|
---|
5244 | /* See if there is a register linearly related to a constant
|
---|
5245 | equivalent of SRC. */
|
---|
5246 | if (src_const
|
---|
5247 | && (GET_CODE (src_const) == CONST
|
---|
5248 | || (src_const_elt && src_const_elt->related_value != 0)))
|
---|
5249 | {
|
---|
5250 | src_related = use_related_value (src_const, src_const_elt);
|
---|
5251 | if (src_related)
|
---|
5252 | {
|
---|
5253 | struct table_elt *src_related_elt
|
---|
5254 | = lookup (src_related, HASH (src_related, mode), mode);
|
---|
5255 | if (src_related_elt && elt)
|
---|
5256 | {
|
---|
5257 | if (elt->first_same_value
|
---|
5258 | != src_related_elt->first_same_value)
|
---|
5259 | /* This can occur when we previously saw a CONST
|
---|
5260 | involving a SYMBOL_REF and then see the SYMBOL_REF
|
---|
5261 | twice. Merge the involved classes. */
|
---|
5262 | merge_equiv_classes (elt, src_related_elt);
|
---|
5263 |
|
---|
5264 | src_related = 0;
|
---|
5265 | src_related_elt = 0;
|
---|
5266 | }
|
---|
5267 | else if (src_related_elt && elt == 0)
|
---|
5268 | elt = src_related_elt;
|
---|
5269 | }
|
---|
5270 | }
|
---|
5271 |
|
---|
5272 | /* See if we have a CONST_INT that is already in a register in a
|
---|
5273 | wider mode. */
|
---|
5274 |
|
---|
5275 | if (src_const && src_related == 0 && GET_CODE (src_const) == CONST_INT
|
---|
5276 | && GET_MODE_CLASS (mode) == MODE_INT
|
---|
5277 | && GET_MODE_BITSIZE (mode) < BITS_PER_WORD)
|
---|
5278 | {
|
---|
5279 | enum machine_mode wider_mode;
|
---|
5280 |
|
---|
5281 | for (wider_mode = GET_MODE_WIDER_MODE (mode);
|
---|
5282 | GET_MODE_BITSIZE (wider_mode) <= BITS_PER_WORD
|
---|
5283 | && src_related == 0;
|
---|
5284 | wider_mode = GET_MODE_WIDER_MODE (wider_mode))
|
---|
5285 | {
|
---|
5286 | struct table_elt *const_elt
|
---|
5287 | = lookup (src_const, HASH (src_const, wider_mode), wider_mode);
|
---|
5288 |
|
---|
5289 | if (const_elt == 0)
|
---|
5290 | continue;
|
---|
5291 |
|
---|
5292 | for (const_elt = const_elt->first_same_value;
|
---|
5293 | const_elt; const_elt = const_elt->next_same_value)
|
---|
5294 | if (GET_CODE (const_elt->exp) == REG)
|
---|
5295 | {
|
---|
5296 | src_related = gen_lowpart_if_possible (mode,
|
---|
5297 | const_elt->exp);
|
---|
5298 | break;
|
---|
5299 | }
|
---|
5300 | }
|
---|
5301 | }
|
---|
5302 |
|
---|
5303 | /* Another possibility is that we have an AND with a constant in
|
---|
5304 | a mode narrower than a word. If so, it might have been generated
|
---|
5305 | as part of an "if" which would narrow the AND. If we already
|
---|
5306 | have done the AND in a wider mode, we can use a SUBREG of that
|
---|
5307 | value. */
|
---|
5308 |
|
---|
5309 | if (flag_expensive_optimizations && ! src_related
|
---|
5310 | && GET_CODE (src) == AND && GET_CODE (XEXP (src, 1)) == CONST_INT
|
---|
5311 | && GET_MODE_SIZE (mode) < UNITS_PER_WORD)
|
---|
5312 | {
|
---|
5313 | enum machine_mode tmode;
|
---|
5314 | rtx new_and = gen_rtx_AND (VOIDmode, NULL_RTX, XEXP (src, 1));
|
---|
5315 |
|
---|
5316 | for (tmode = GET_MODE_WIDER_MODE (mode);
|
---|
5317 | GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
|
---|
5318 | tmode = GET_MODE_WIDER_MODE (tmode))
|
---|
5319 | {
|
---|
5320 | rtx inner = gen_lowpart_if_possible (tmode, XEXP (src, 0));
|
---|
5321 | struct table_elt *larger_elt;
|
---|
5322 |
|
---|
5323 | if (inner)
|
---|
5324 | {
|
---|
5325 | PUT_MODE (new_and, tmode);
|
---|
5326 | XEXP (new_and, 0) = inner;
|
---|
5327 | larger_elt = lookup (new_and, HASH (new_and, tmode), tmode);
|
---|
5328 | if (larger_elt == 0)
|
---|
5329 | continue;
|
---|
5330 |
|
---|
5331 | for (larger_elt = larger_elt->first_same_value;
|
---|
5332 | larger_elt; larger_elt = larger_elt->next_same_value)
|
---|
5333 | if (GET_CODE (larger_elt->exp) == REG)
|
---|
5334 | {
|
---|
5335 | src_related
|
---|
5336 | = gen_lowpart_if_possible (mode, larger_elt->exp);
|
---|
5337 | break;
|
---|
5338 | }
|
---|
5339 |
|
---|
5340 | if (src_related)
|
---|
5341 | break;
|
---|
5342 | }
|
---|
5343 | }
|
---|
5344 | }
|
---|
5345 |
|
---|
5346 | #ifdef LOAD_EXTEND_OP
|
---|
5347 | /* See if a MEM has already been loaded with a widening operation;
|
---|
5348 | if it has, we can use a subreg of that. Many CISC machines
|
---|
5349 | also have such operations, but this is only likely to be
|
---|
5350 | beneficial these machines. */
|
---|
5351 |
|
---|
5352 | if (flag_expensive_optimizations && src_related == 0
|
---|
5353 | && (GET_MODE_SIZE (mode) < UNITS_PER_WORD)
|
---|
5354 | && GET_MODE_CLASS (mode) == MODE_INT
|
---|
5355 | && GET_CODE (src) == MEM && ! do_not_record
|
---|
5356 | && LOAD_EXTEND_OP (mode) != NIL)
|
---|
5357 | {
|
---|
5358 | enum machine_mode tmode;
|
---|
5359 |
|
---|
5360 | /* Set what we are trying to extend and the operation it might
|
---|
5361 | have been extended with. */
|
---|
5362 | PUT_CODE (memory_extend_rtx, LOAD_EXTEND_OP (mode));
|
---|
5363 | XEXP (memory_extend_rtx, 0) = src;
|
---|
5364 |
|
---|
5365 | for (tmode = GET_MODE_WIDER_MODE (mode);
|
---|
5366 | GET_MODE_SIZE (tmode) <= UNITS_PER_WORD;
|
---|
5367 | tmode = GET_MODE_WIDER_MODE (tmode))
|
---|
5368 | {
|
---|
5369 | struct table_elt *larger_elt;
|
---|
5370 |
|
---|
5371 | PUT_MODE (memory_extend_rtx, tmode);
|
---|
5372 | larger_elt = lookup (memory_extend_rtx,
|
---|
5373 | HASH (memory_extend_rtx, tmode), tmode);
|
---|
5374 | if (larger_elt == 0)
|
---|
5375 | continue;
|
---|
5376 |
|
---|
5377 | for (larger_elt = larger_elt->first_same_value;
|
---|
5378 | larger_elt; larger_elt = larger_elt->next_same_value)
|
---|
5379 | if (GET_CODE (larger_elt->exp) == REG)
|
---|
5380 | {
|
---|
5381 | src_related = gen_lowpart_if_possible (mode,
|
---|
5382 | larger_elt->exp);
|
---|
5383 | break;
|
---|
5384 | }
|
---|
5385 |
|
---|
5386 | if (src_related)
|
---|
5387 | break;
|
---|
5388 | }
|
---|
5389 | }
|
---|
5390 | #endif /* LOAD_EXTEND_OP */
|
---|
5391 |
|
---|
5392 | if (src == src_folded)
|
---|
5393 | src_folded = 0;
|
---|
5394 |
|
---|
5395 | /* At this point, ELT, if nonzero, points to a class of expressions
|
---|
5396 | equivalent to the source of this SET and SRC, SRC_EQV, SRC_FOLDED,
|
---|
5397 | and SRC_RELATED, if nonzero, each contain additional equivalent
|
---|
5398 | expressions. Prune these latter expressions by deleting expressions
|
---|
5399 | already in the equivalence class.
|
---|
5400 |
|
---|
5401 | Check for an equivalent identical to the destination. If found,
|
---|
5402 | this is the preferred equivalent since it will likely lead to
|
---|
5403 | elimination of the insn. Indicate this by placing it in
|
---|
5404 | `src_related'. */
|
---|
5405 |
|
---|
5406 | if (elt)
|
---|
5407 | elt = elt->first_same_value;
|
---|
5408 | for (p = elt; p; p = p->next_same_value)
|
---|
5409 | {
|
---|
5410 | enum rtx_code code = GET_CODE (p->exp);
|
---|
5411 |
|
---|
5412 | /* If the expression is not valid, ignore it. Then we do not
|
---|
5413 | have to check for validity below. In most cases, we can use
|
---|
5414 | `rtx_equal_p', since canonicalization has already been done. */
|
---|
5415 | if (code != REG && ! exp_equiv_p (p->exp, p->exp, 1, 0))
|
---|
5416 | continue;
|
---|
5417 |
|
---|
5418 | /* Also skip paradoxical subregs, unless that's what we're
|
---|
5419 | looking for. */
|
---|
5420 | if (code == SUBREG
|
---|
5421 | && (GET_MODE_SIZE (GET_MODE (p->exp))
|
---|
5422 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))
|
---|
5423 | && ! (src != 0
|
---|
5424 | && GET_CODE (src) == SUBREG
|
---|
5425 | && GET_MODE (src) == GET_MODE (p->exp)
|
---|
5426 | && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
|
---|
5427 | < GET_MODE_SIZE (GET_MODE (SUBREG_REG (p->exp))))))
|
---|
5428 | continue;
|
---|
5429 |
|
---|
5430 | if (src && GET_CODE (src) == code && rtx_equal_p (src, p->exp))
|
---|
5431 | src = 0;
|
---|
5432 | else if (src_folded && GET_CODE (src_folded) == code
|
---|
5433 | && rtx_equal_p (src_folded, p->exp))
|
---|
5434 | src_folded = 0;
|
---|
5435 | else if (src_eqv_here && GET_CODE (src_eqv_here) == code
|
---|
5436 | && rtx_equal_p (src_eqv_here, p->exp))
|
---|
5437 | src_eqv_here = 0;
|
---|
5438 | else if (src_related && GET_CODE (src_related) == code
|
---|
5439 | && rtx_equal_p (src_related, p->exp))
|
---|
5440 | src_related = 0;
|
---|
5441 |
|
---|
5442 | /* This is the same as the destination of the insns, we want
|
---|
5443 | to prefer it. Copy it to src_related. The code below will
|
---|
5444 | then give it a negative cost. */
|
---|
5445 | if (GET_CODE (dest) == code && rtx_equal_p (p->exp, dest))
|
---|
5446 | src_related = dest;
|
---|
5447 | }
|
---|
5448 |
|
---|
5449 | /* Find the cheapest valid equivalent, trying all the available
|
---|
5450 | possibilities. Prefer items not in the hash table to ones
|
---|
5451 | that are when they are equal cost. Note that we can never
|
---|
5452 | worsen an insn as the current contents will also succeed.
|
---|
5453 | If we find an equivalent identical to the destination, use it as best,
|
---|
5454 | since this insn will probably be eliminated in that case. */
|
---|
5455 | if (src)
|
---|
5456 | {
|
---|
5457 | if (rtx_equal_p (src, dest))
|
---|
5458 | src_cost = src_regcost = -1;
|
---|
5459 | else
|
---|
5460 | {
|
---|
5461 | src_cost = COST (src);
|
---|
5462 | src_regcost = approx_reg_cost (src);
|
---|
5463 | }
|
---|
5464 | }
|
---|
5465 |
|
---|
5466 | if (src_eqv_here)
|
---|
5467 | {
|
---|
5468 | if (rtx_equal_p (src_eqv_here, dest))
|
---|
5469 | src_eqv_cost = src_eqv_regcost = -1;
|
---|
5470 | else
|
---|
5471 | {
|
---|
5472 | src_eqv_cost = COST (src_eqv_here);
|
---|
5473 | src_eqv_regcost = approx_reg_cost (src_eqv_here);
|
---|
5474 | }
|
---|
5475 | }
|
---|
5476 |
|
---|
5477 | if (src_folded)
|
---|
5478 | {
|
---|
5479 | if (rtx_equal_p (src_folded, dest))
|
---|
5480 | src_folded_cost = src_folded_regcost = -1;
|
---|
5481 | else
|
---|
5482 | {
|
---|
5483 | src_folded_cost = COST (src_folded);
|
---|
5484 | src_folded_regcost = approx_reg_cost (src_folded);
|
---|
5485 | }
|
---|
5486 | }
|
---|
5487 |
|
---|
5488 | if (src_related)
|
---|
5489 | {
|
---|
5490 | if (rtx_equal_p (src_related, dest))
|
---|
5491 | src_related_cost = src_related_regcost = -1;
|
---|
5492 | else
|
---|
5493 | {
|
---|
5494 | src_related_cost = COST (src_related);
|
---|
5495 | src_related_regcost = approx_reg_cost (src_related);
|
---|
5496 | }
|
---|
5497 | }
|
---|
5498 |
|
---|
5499 | /* If this was an indirect jump insn, a known label will really be
|
---|
5500 | cheaper even though it looks more expensive. */
|
---|
5501 | if (dest == pc_rtx && src_const && GET_CODE (src_const) == LABEL_REF)
|
---|
5502 | src_folded = src_const, src_folded_cost = src_folded_regcost = -1;
|
---|
5503 |
|
---|
5504 | /* Terminate loop when replacement made. This must terminate since
|
---|
5505 | the current contents will be tested and will always be valid. */
|
---|
5506 | while (1)
|
---|
5507 | {
|
---|
5508 | rtx trial;
|
---|
5509 |
|
---|
5510 | /* Skip invalid entries. */
|
---|
5511 | while (elt && GET_CODE (elt->exp) != REG
|
---|
5512 | && ! exp_equiv_p (elt->exp, elt->exp, 1, 0))
|
---|
5513 | elt = elt->next_same_value;
|
---|
5514 |
|
---|
5515 | /* A paradoxical subreg would be bad here: it'll be the right
|
---|
5516 | size, but later may be adjusted so that the upper bits aren't
|
---|
5517 | what we want. So reject it. */
|
---|
5518 | if (elt != 0
|
---|
5519 | && GET_CODE (elt->exp) == SUBREG
|
---|
5520 | && (GET_MODE_SIZE (GET_MODE (elt->exp))
|
---|
5521 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))
|
---|
5522 | /* It is okay, though, if the rtx we're trying to match
|
---|
5523 | will ignore any of the bits we can't predict. */
|
---|
5524 | && ! (src != 0
|
---|
5525 | && GET_CODE (src) == SUBREG
|
---|
5526 | && GET_MODE (src) == GET_MODE (elt->exp)
|
---|
5527 | && (GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
|
---|
5528 | < GET_MODE_SIZE (GET_MODE (SUBREG_REG (elt->exp))))))
|
---|
5529 | {
|
---|
5530 | elt = elt->next_same_value;
|
---|
5531 | continue;
|
---|
5532 | }
|
---|
5533 |
|
---|
5534 | if (elt)
|
---|
5535 | {
|
---|
5536 | src_elt_cost = elt->cost;
|
---|
5537 | src_elt_regcost = elt->regcost;
|
---|
5538 | }
|
---|
5539 |
|
---|
5540 | /* Find cheapest and skip it for the next time. For items
|
---|
5541 | of equal cost, use this order:
|
---|
5542 | src_folded, src, src_eqv, src_related and hash table entry. */
|
---|
5543 | if (src_folded
|
---|
5544 | && preferrable (src_folded_cost, src_folded_regcost,
|
---|
5545 | src_cost, src_regcost) <= 0
|
---|
5546 | && preferrable (src_folded_cost, src_folded_regcost,
|
---|
5547 | src_eqv_cost, src_eqv_regcost) <= 0
|
---|
5548 | && preferrable (src_folded_cost, src_folded_regcost,
|
---|
5549 | src_related_cost, src_related_regcost) <= 0
|
---|
5550 | && preferrable (src_folded_cost, src_folded_regcost,
|
---|
5551 | src_elt_cost, src_elt_regcost) <= 0)
|
---|
5552 | {
|
---|
5553 | trial = src_folded, src_folded_cost = MAX_COST;
|
---|
5554 | if (src_folded_force_flag)
|
---|
5555 | trial = force_const_mem (mode, trial);
|
---|
5556 | }
|
---|
5557 | else if (src
|
---|
5558 | && preferrable (src_cost, src_regcost,
|
---|
5559 | src_eqv_cost, src_eqv_regcost) <= 0
|
---|
5560 | && preferrable (src_cost, src_regcost,
|
---|
5561 | src_related_cost, src_related_regcost) <= 0
|
---|
5562 | && preferrable (src_cost, src_regcost,
|
---|
5563 | src_elt_cost, src_elt_regcost) <= 0)
|
---|
5564 | trial = src, src_cost = MAX_COST;
|
---|
5565 | else if (src_eqv_here
|
---|
5566 | && preferrable (src_eqv_cost, src_eqv_regcost,
|
---|
5567 | src_related_cost, src_related_regcost) <= 0
|
---|
5568 | && preferrable (src_eqv_cost, src_eqv_regcost,
|
---|
5569 | src_elt_cost, src_elt_regcost) <= 0)
|
---|
5570 | trial = copy_rtx (src_eqv_here), src_eqv_cost = MAX_COST;
|
---|
5571 | else if (src_related
|
---|
5572 | && preferrable (src_related_cost, src_related_regcost,
|
---|
5573 | src_elt_cost, src_elt_regcost) <= 0)
|
---|
5574 | trial = copy_rtx (src_related), src_related_cost = MAX_COST;
|
---|
5575 | else
|
---|
5576 | {
|
---|
5577 | trial = copy_rtx (elt->exp);
|
---|
5578 | elt = elt->next_same_value;
|
---|
5579 | src_elt_cost = MAX_COST;
|
---|
5580 | }
|
---|
5581 |
|
---|
5582 | /* We don't normally have an insn matching (set (pc) (pc)), so
|
---|
5583 | check for this separately here. We will delete such an
|
---|
5584 | insn below.
|
---|
5585 |
|
---|
5586 | For other cases such as a table jump or conditional jump
|
---|
5587 | where we know the ultimate target, go ahead and replace the
|
---|
5588 | operand. While that may not make a valid insn, we will
|
---|
5589 | reemit the jump below (and also insert any necessary
|
---|
5590 | barriers). */
|
---|
5591 | if (n_sets == 1 && dest == pc_rtx
|
---|
5592 | && (trial == pc_rtx
|
---|
5593 | || (GET_CODE (trial) == LABEL_REF
|
---|
5594 | && ! condjump_p (insn))))
|
---|
5595 | {
|
---|
5596 | SET_SRC (sets[i].rtl) = trial;
|
---|
5597 | cse_jumps_altered = 1;
|
---|
5598 | break;
|
---|
5599 | }
|
---|
5600 |
|
---|
5601 | /* Look for a substitution that makes a valid insn. */
|
---|
5602 | else if (validate_change (insn, &SET_SRC (sets[i].rtl), trial, 0))
|
---|
5603 | {
|
---|
5604 | /* If we just made a substitution inside a libcall, then we
|
---|
5605 | need to make the same substitution in any notes attached
|
---|
5606 | to the RETVAL insn. */
|
---|
5607 | if (libcall_insn
|
---|
5608 | && (GET_CODE (sets[i].orig_src) == REG
|
---|
5609 | || GET_CODE (sets[i].orig_src) == SUBREG
|
---|
5610 | || GET_CODE (sets[i].orig_src) == MEM))
|
---|
5611 | replace_rtx (REG_NOTES (libcall_insn), sets[i].orig_src,
|
---|
5612 | canon_reg (SET_SRC (sets[i].rtl), insn));
|
---|
5613 |
|
---|
5614 | /* The result of apply_change_group can be ignored; see
|
---|
5615 | canon_reg. */
|
---|
5616 |
|
---|
5617 | validate_change (insn, &SET_SRC (sets[i].rtl),
|
---|
5618 | canon_reg (SET_SRC (sets[i].rtl), insn),
|
---|
5619 | 1);
|
---|
5620 | apply_change_group ();
|
---|
5621 | break;
|
---|
5622 | }
|
---|
5623 |
|
---|
5624 | /* If we previously found constant pool entries for
|
---|
5625 | constants and this is a constant, try making a
|
---|
5626 | pool entry. Put it in src_folded unless we already have done
|
---|
5627 | this since that is where it likely came from. */
|
---|
5628 |
|
---|
5629 | else if (constant_pool_entries_cost
|
---|
5630 | && CONSTANT_P (trial)
|
---|
5631 | /* Reject cases that will abort in decode_rtx_const.
|
---|
5632 | On the alpha when simplifying a switch, we get
|
---|
5633 | (const (truncate (minus (label_ref) (label_ref)))). */
|
---|
5634 | && ! (GET_CODE (trial) == CONST
|
---|
5635 | && GET_CODE (XEXP (trial, 0)) == TRUNCATE)
|
---|
5636 | /* Likewise on IA-64, except without the truncate. */
|
---|
5637 | && ! (GET_CODE (trial) == CONST
|
---|
5638 | && GET_CODE (XEXP (trial, 0)) == MINUS
|
---|
5639 | && GET_CODE (XEXP (XEXP (trial, 0), 0)) == LABEL_REF
|
---|
5640 | && GET_CODE (XEXP (XEXP (trial, 0), 1)) == LABEL_REF)
|
---|
5641 | && (src_folded == 0
|
---|
5642 | || (GET_CODE (src_folded) != MEM
|
---|
5643 | && ! src_folded_force_flag))
|
---|
5644 | && GET_MODE_CLASS (mode) != MODE_CC
|
---|
5645 | && mode != VOIDmode)
|
---|
5646 | {
|
---|
5647 | src_folded_force_flag = 1;
|
---|
5648 | src_folded = trial;
|
---|
5649 | src_folded_cost = constant_pool_entries_cost;
|
---|
5650 | }
|
---|
5651 | }
|
---|
5652 |
|
---|
5653 | src = SET_SRC (sets[i].rtl);
|
---|
5654 |
|
---|
5655 | /* In general, it is good to have a SET with SET_SRC == SET_DEST.
|
---|
5656 | However, there is an important exception: If both are registers
|
---|
5657 | that are not the head of their equivalence class, replace SET_SRC
|
---|
5658 | with the head of the class. If we do not do this, we will have
|
---|
5659 | both registers live over a portion of the basic block. This way,
|
---|
5660 | their lifetimes will likely abut instead of overlapping. */
|
---|
5661 | if (GET_CODE (dest) == REG
|
---|
5662 | && REGNO_QTY_VALID_P (REGNO (dest)))
|
---|
5663 | {
|
---|
5664 | int dest_q = REG_QTY (REGNO (dest));
|
---|
5665 | struct qty_table_elem *dest_ent = &qty_table[dest_q];
|
---|
5666 |
|
---|
5667 | if (dest_ent->mode == GET_MODE (dest)
|
---|
5668 | && dest_ent->first_reg != REGNO (dest)
|
---|
5669 | && GET_CODE (src) == REG && REGNO (src) == REGNO (dest)
|
---|
5670 | /* Don't do this if the original insn had a hard reg as
|
---|
5671 | SET_SRC or SET_DEST. */
|
---|
5672 | && (GET_CODE (sets[i].src) != REG
|
---|
5673 | || REGNO (sets[i].src) >= FIRST_PSEUDO_REGISTER)
|
---|
5674 | && (GET_CODE (dest) != REG || REGNO (dest) >= FIRST_PSEUDO_REGISTER))
|
---|
5675 | /* We can't call canon_reg here because it won't do anything if
|
---|
5676 | SRC is a hard register. */
|
---|
5677 | {
|
---|
5678 | int src_q = REG_QTY (REGNO (src));
|
---|
5679 | struct qty_table_elem *src_ent = &qty_table[src_q];
|
---|
5680 | int first = src_ent->first_reg;
|
---|
5681 | rtx new_src
|
---|
5682 | = (first >= FIRST_PSEUDO_REGISTER
|
---|
5683 | ? regno_reg_rtx[first] : gen_rtx_REG (GET_MODE (src), first));
|
---|
5684 |
|
---|
5685 | /* We must use validate-change even for this, because this
|
---|
5686 | might be a special no-op instruction, suitable only to
|
---|
5687 | tag notes onto. */
|
---|
5688 | if (validate_change (insn, &SET_SRC (sets[i].rtl), new_src, 0))
|
---|
5689 | {
|
---|
5690 | src = new_src;
|
---|
5691 | /* If we had a constant that is cheaper than what we are now
|
---|
5692 | setting SRC to, use that constant. We ignored it when we
|
---|
5693 | thought we could make this into a no-op. */
|
---|
5694 | if (src_const && COST (src_const) < COST (src)
|
---|
5695 | && validate_change (insn, &SET_SRC (sets[i].rtl),
|
---|
5696 | src_const, 0))
|
---|
5697 | src = src_const;
|
---|
5698 | }
|
---|
5699 | }
|
---|
5700 | }
|
---|
5701 |
|
---|
5702 | /* If we made a change, recompute SRC values. */
|
---|
5703 | if (src != sets[i].src)
|
---|
5704 | {
|
---|
5705 | cse_altered = 1;
|
---|
5706 | do_not_record = 0;
|
---|
5707 | hash_arg_in_memory = 0;
|
---|
5708 | sets[i].src = src;
|
---|
5709 | sets[i].src_hash = HASH (src, mode);
|
---|
5710 | sets[i].src_volatile = do_not_record;
|
---|
5711 | sets[i].src_in_memory = hash_arg_in_memory;
|
---|
5712 | sets[i].src_elt = lookup (src, sets[i].src_hash, mode);
|
---|
5713 | }
|
---|
5714 |
|
---|
5715 | /* If this is a single SET, we are setting a register, and we have an
|
---|
5716 | equivalent constant, we want to add a REG_NOTE. We don't want
|
---|
5717 | to write a REG_EQUAL note for a constant pseudo since verifying that
|
---|
5718 | that pseudo hasn't been eliminated is a pain. Such a note also
|
---|
5719 | won't help anything.
|
---|
5720 |
|
---|
5721 | Avoid a REG_EQUAL note for (CONST (MINUS (LABEL_REF) (LABEL_REF)))
|
---|
5722 | which can be created for a reference to a compile time computable
|
---|
5723 | entry in a jump table. */
|
---|
5724 |
|
---|
5725 | if (n_sets == 1 && src_const && GET_CODE (dest) == REG
|
---|
5726 | && GET_CODE (src_const) != REG
|
---|
5727 | && ! (GET_CODE (src_const) == CONST
|
---|
5728 | && GET_CODE (XEXP (src_const, 0)) == MINUS
|
---|
5729 | && GET_CODE (XEXP (XEXP (src_const, 0), 0)) == LABEL_REF
|
---|
5730 | && GET_CODE (XEXP (XEXP (src_const, 0), 1)) == LABEL_REF))
|
---|
5731 | {
|
---|
5732 | /* Make sure that the rtx is not shared with any other insn. */
|
---|
5733 | src_const = copy_rtx (src_const);
|
---|
5734 |
|
---|
5735 | /* Record the actual constant value in a REG_EQUAL note, making
|
---|
5736 | a new one if one does not already exist. */
|
---|
5737 | set_unique_reg_note (insn, REG_EQUAL, src_const);
|
---|
5738 |
|
---|
5739 | /* If storing a constant value in a register that
|
---|
5740 | previously held the constant value 0,
|
---|
5741 | record this fact with a REG_WAS_0 note on this insn.
|
---|
5742 |
|
---|
5743 | Note that the *register* is required to have previously held 0,
|
---|
5744 | not just any register in the quantity and we must point to the
|
---|
5745 | insn that set that register to zero.
|
---|
5746 |
|
---|
5747 | Rather than track each register individually, we just see if
|
---|
5748 | the last set for this quantity was for this register. */
|
---|
5749 |
|
---|
5750 | if (REGNO_QTY_VALID_P (REGNO (dest)))
|
---|
5751 | {
|
---|
5752 | int dest_q = REG_QTY (REGNO (dest));
|
---|
5753 | struct qty_table_elem *dest_ent = &qty_table[dest_q];
|
---|
5754 |
|
---|
5755 | if (dest_ent->const_rtx == const0_rtx)
|
---|
5756 | {
|
---|
5757 | /* See if we previously had a REG_WAS_0 note. */
|
---|
5758 | rtx note = find_reg_note (insn, REG_WAS_0, NULL_RTX);
|
---|
5759 | rtx const_insn = dest_ent->const_insn;
|
---|
5760 |
|
---|
5761 | if ((tem = single_set (const_insn)) != 0
|
---|
5762 | && rtx_equal_p (SET_DEST (tem), dest))
|
---|
5763 | {
|
---|
5764 | if (note)
|
---|
5765 | XEXP (note, 0) = const_insn;
|
---|
5766 | else
|
---|
5767 | REG_NOTES (insn)
|
---|
5768 | = gen_rtx_INSN_LIST (REG_WAS_0, const_insn,
|
---|
5769 | REG_NOTES (insn));
|
---|
5770 | }
|
---|
5771 | }
|
---|
5772 | }
|
---|
5773 | }
|
---|
5774 |
|
---|
5775 | /* Now deal with the destination. */
|
---|
5776 | do_not_record = 0;
|
---|
5777 |
|
---|
5778 | /* Look within any SIGN_EXTRACT or ZERO_EXTRACT
|
---|
5779 | to the MEM or REG within it. */
|
---|
5780 | while (GET_CODE (dest) == SIGN_EXTRACT
|
---|
5781 | || GET_CODE (dest) == ZERO_EXTRACT
|
---|
5782 | || GET_CODE (dest) == SUBREG
|
---|
5783 | || GET_CODE (dest) == STRICT_LOW_PART)
|
---|
5784 | dest = XEXP (dest, 0);
|
---|
5785 |
|
---|
5786 | sets[i].inner_dest = dest;
|
---|
5787 |
|
---|
5788 | if (GET_CODE (dest) == MEM)
|
---|
5789 | {
|
---|
5790 | #ifdef PUSH_ROUNDING
|
---|
5791 | /* Stack pushes invalidate the stack pointer. */
|
---|
5792 | rtx addr = XEXP (dest, 0);
|
---|
5793 | if (GET_RTX_CLASS (GET_CODE (addr)) == 'a'
|
---|
5794 | && XEXP (addr, 0) == stack_pointer_rtx)
|
---|
5795 | invalidate (stack_pointer_rtx, Pmode);
|
---|
5796 | #endif
|
---|
5797 | dest = fold_rtx (dest, insn);
|
---|
5798 | }
|
---|
5799 |
|
---|
5800 | /* Compute the hash code of the destination now,
|
---|
5801 | before the effects of this instruction are recorded,
|
---|
5802 | since the register values used in the address computation
|
---|
5803 | are those before this instruction. */
|
---|
5804 | sets[i].dest_hash = HASH (dest, mode);
|
---|
5805 |
|
---|
5806 | /* Don't enter a bit-field in the hash table
|
---|
5807 | because the value in it after the store
|
---|
5808 | may not equal what was stored, due to truncation. */
|
---|
5809 |
|
---|
5810 | if (GET_CODE (SET_DEST (sets[i].rtl)) == ZERO_EXTRACT
|
---|
5811 | || GET_CODE (SET_DEST (sets[i].rtl)) == SIGN_EXTRACT)
|
---|
5812 | {
|
---|
5813 | rtx width = XEXP (SET_DEST (sets[i].rtl), 1);
|
---|
5814 |
|
---|
5815 | if (src_const != 0 && GET_CODE (src_const) == CONST_INT
|
---|
5816 | && GET_CODE (width) == CONST_INT
|
---|
5817 | && INTVAL (width) < HOST_BITS_PER_WIDE_INT
|
---|
5818 | && ! (INTVAL (src_const)
|
---|
5819 | & ((HOST_WIDE_INT) (-1) << INTVAL (width))))
|
---|
5820 | /* Exception: if the value is constant,
|
---|
5821 | and it won't be truncated, record it. */
|
---|
5822 | ;
|
---|
5823 | else
|
---|
5824 | {
|
---|
5825 | /* This is chosen so that the destination will be invalidated
|
---|
5826 | but no new value will be recorded.
|
---|
5827 | We must invalidate because sometimes constant
|
---|
5828 | values can be recorded for bitfields. */
|
---|
5829 | sets[i].src_elt = 0;
|
---|
5830 | sets[i].src_volatile = 1;
|
---|
5831 | src_eqv = 0;
|
---|
5832 | src_eqv_elt = 0;
|
---|
5833 | }
|
---|
5834 | }
|
---|
5835 |
|
---|
5836 | /* If only one set in a JUMP_INSN and it is now a no-op, we can delete
|
---|
5837 | the insn. */
|
---|
5838 | else if (n_sets == 1 && dest == pc_rtx && src == pc_rtx)
|
---|
5839 | {
|
---|
5840 | /* One less use of the label this insn used to jump to. */
|
---|
5841 | delete_insn (insn);
|
---|
5842 | cse_jumps_altered = 1;
|
---|
5843 | /* No more processing for this set. */
|
---|
5844 | sets[i].rtl = 0;
|
---|
5845 | }
|
---|
5846 |
|
---|
5847 | /* If this SET is now setting PC to a label, we know it used to
|
---|
5848 | be a conditional or computed branch. */
|
---|
5849 | else if (dest == pc_rtx && GET_CODE (src) == LABEL_REF)
|
---|
5850 | {
|
---|
5851 | /* Now emit a BARRIER after the unconditional jump. */
|
---|
5852 | if (NEXT_INSN (insn) == 0
|
---|
5853 | || GET_CODE (NEXT_INSN (insn)) != BARRIER)
|
---|
5854 | emit_barrier_after (insn);
|
---|
5855 |
|
---|
5856 | /* We reemit the jump in as many cases as possible just in
|
---|
5857 | case the form of an unconditional jump is significantly
|
---|
5858 | different than a computed jump or conditional jump.
|
---|
5859 |
|
---|
5860 | If this insn has multiple sets, then reemitting the
|
---|
5861 | jump is nontrivial. So instead we just force rerecognition
|
---|
5862 | and hope for the best. */
|
---|
5863 | if (n_sets == 1)
|
---|
5864 | {
|
---|
5865 | rtx new = emit_jump_insn_after (gen_jump (XEXP (src, 0)), insn);
|
---|
5866 |
|
---|
5867 | JUMP_LABEL (new) = XEXP (src, 0);
|
---|
5868 | LABEL_NUSES (XEXP (src, 0))++;
|
---|
5869 | delete_insn (insn);
|
---|
5870 | insn = new;
|
---|
5871 |
|
---|
5872 | /* Now emit a BARRIER after the unconditional jump. */
|
---|
5873 | if (NEXT_INSN (insn) == 0
|
---|
5874 | || GET_CODE (NEXT_INSN (insn)) != BARRIER)
|
---|
5875 | emit_barrier_after (insn);
|
---|
5876 | }
|
---|
5877 | else
|
---|
5878 | INSN_CODE (insn) = -1;
|
---|
5879 |
|
---|
5880 | never_reached_warning (insn, NULL);
|
---|
5881 |
|
---|
5882 | /* Do not bother deleting any unreachable code,
|
---|
5883 | let jump/flow do that. */
|
---|
5884 |
|
---|
5885 | cse_jumps_altered = 1;
|
---|
5886 | sets[i].rtl = 0;
|
---|
5887 | }
|
---|
5888 |
|
---|
5889 | /* If destination is volatile, invalidate it and then do no further
|
---|
5890 | processing for this assignment. */
|
---|
5891 |
|
---|
5892 | else if (do_not_record)
|
---|
5893 | {
|
---|
5894 | if (GET_CODE (dest) == REG || GET_CODE (dest) == SUBREG)
|
---|
5895 | invalidate (dest, VOIDmode);
|
---|
5896 | else if (GET_CODE (dest) == MEM)
|
---|
5897 | {
|
---|
5898 | /* Outgoing arguments for a libcall don't
|
---|
5899 | affect any recorded expressions. */
|
---|
5900 | if (! libcall_insn || insn == libcall_insn)
|
---|
5901 | invalidate (dest, VOIDmode);
|
---|
5902 | }
|
---|
5903 | else if (GET_CODE (dest) == STRICT_LOW_PART
|
---|
5904 | || GET_CODE (dest) == ZERO_EXTRACT)
|
---|
5905 | invalidate (XEXP (dest, 0), GET_MODE (dest));
|
---|
5906 | sets[i].rtl = 0;
|
---|
5907 | }
|
---|
5908 |
|
---|
5909 | if (sets[i].rtl != 0 && dest != SET_DEST (sets[i].rtl))
|
---|
5910 | sets[i].dest_hash = HASH (SET_DEST (sets[i].rtl), mode);
|
---|
5911 |
|
---|
5912 | #ifdef HAVE_cc0
|
---|
5913 | /* If setting CC0, record what it was set to, or a constant, if it
|
---|
5914 | is equivalent to a constant. If it is being set to a floating-point
|
---|
5915 | value, make a COMPARE with the appropriate constant of 0. If we
|
---|
5916 | don't do this, later code can interpret this as a test against
|
---|
5917 | const0_rtx, which can cause problems if we try to put it into an
|
---|
5918 | insn as a floating-point operand. */
|
---|
5919 | if (dest == cc0_rtx)
|
---|
5920 | {
|
---|
5921 | this_insn_cc0 = src_const && mode != VOIDmode ? src_const : src;
|
---|
5922 | this_insn_cc0_mode = mode;
|
---|
5923 | if (FLOAT_MODE_P (mode))
|
---|
5924 | this_insn_cc0 = gen_rtx_COMPARE (VOIDmode, this_insn_cc0,
|
---|
5925 | CONST0_RTX (mode));
|
---|
5926 | }
|
---|
5927 | #endif
|
---|
5928 | }
|
---|
5929 |
|
---|
5930 | /* Now enter all non-volatile source expressions in the hash table
|
---|
5931 | if they are not already present.
|
---|
5932 | Record their equivalence classes in src_elt.
|
---|
5933 | This way we can insert the corresponding destinations into
|
---|
5934 | the same classes even if the actual sources are no longer in them
|
---|
5935 | (having been invalidated). */
|
---|
5936 |
|
---|
5937 | if (src_eqv && src_eqv_elt == 0 && sets[0].rtl != 0 && ! src_eqv_volatile
|
---|
5938 | && ! rtx_equal_p (src_eqv, SET_DEST (sets[0].rtl)))
|
---|
5939 | {
|
---|
5940 | struct table_elt *elt;
|
---|
5941 | struct table_elt *classp = sets[0].src_elt;
|
---|
5942 | rtx dest = SET_DEST (sets[0].rtl);
|
---|
5943 | enum machine_mode eqvmode = GET_MODE (dest);
|
---|
5944 |
|
---|
5945 | if (GET_CODE (dest) == STRICT_LOW_PART)
|
---|
5946 | {
|
---|
5947 | eqvmode = GET_MODE (SUBREG_REG (XEXP (dest, 0)));
|
---|
5948 | classp = 0;
|
---|
5949 | }
|
---|
5950 | if (insert_regs (src_eqv, classp, 0))
|
---|
5951 | {
|
---|
5952 | rehash_using_reg (src_eqv);
|
---|
5953 | src_eqv_hash = HASH (src_eqv, eqvmode);
|
---|
5954 | }
|
---|
5955 | elt = insert (src_eqv, classp, src_eqv_hash, eqvmode);
|
---|
5956 | elt->in_memory = src_eqv_in_memory;
|
---|
5957 | src_eqv_elt = elt;
|
---|
5958 |
|
---|
5959 | /* Check to see if src_eqv_elt is the same as a set source which
|
---|
5960 | does not yet have an elt, and if so set the elt of the set source
|
---|
5961 | to src_eqv_elt. */
|
---|
5962 | for (i = 0; i < n_sets; i++)
|
---|
5963 | if (sets[i].rtl && sets[i].src_elt == 0
|
---|
5964 | && rtx_equal_p (SET_SRC (sets[i].rtl), src_eqv))
|
---|
5965 | sets[i].src_elt = src_eqv_elt;
|
---|
5966 | }
|
---|
5967 |
|
---|
5968 | for (i = 0; i < n_sets; i++)
|
---|
5969 | if (sets[i].rtl && ! sets[i].src_volatile
|
---|
5970 | && ! rtx_equal_p (SET_SRC (sets[i].rtl), SET_DEST (sets[i].rtl)))
|
---|
5971 | {
|
---|
5972 | if (GET_CODE (SET_DEST (sets[i].rtl)) == STRICT_LOW_PART)
|
---|
5973 | {
|
---|
5974 | /* REG_EQUAL in setting a STRICT_LOW_PART
|
---|
5975 | gives an equivalent for the entire destination register,
|
---|
5976 | not just for the subreg being stored in now.
|
---|
5977 | This is a more interesting equivalence, so we arrange later
|
---|
5978 | to treat the entire reg as the destination. */
|
---|
5979 | sets[i].src_elt = src_eqv_elt;
|
---|
5980 | sets[i].src_hash = src_eqv_hash;
|
---|
5981 | }
|
---|
5982 | else
|
---|
5983 | {
|
---|
5984 | /* Insert source and constant equivalent into hash table, if not
|
---|
5985 | already present. */
|
---|
5986 | struct table_elt *classp = src_eqv_elt;
|
---|
5987 | rtx src = sets[i].src;
|
---|
5988 | rtx dest = SET_DEST (sets[i].rtl);
|
---|
5989 | enum machine_mode mode
|
---|
5990 | = GET_MODE (src) == VOIDmode ? GET_MODE (dest) : GET_MODE (src);
|
---|
5991 |
|
---|
5992 | if (sets[i].src_elt == 0)
|
---|
5993 | {
|
---|
5994 | /* Don't put a hard register source into the table if this is
|
---|
5995 | the last insn of a libcall. In this case, we only need
|
---|
5996 | to put src_eqv_elt in src_elt. */
|
---|
5997 | if (! find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
---|
5998 | {
|
---|
5999 | struct table_elt *elt;
|
---|
6000 |
|
---|
6001 | /* Note that these insert_regs calls cannot remove
|
---|
6002 | any of the src_elt's, because they would have failed to
|
---|
6003 | match if not still valid. */
|
---|
6004 | if (insert_regs (src, classp, 0))
|
---|
6005 | {
|
---|
6006 | rehash_using_reg (src);
|
---|
6007 | sets[i].src_hash = HASH (src, mode);
|
---|
6008 | }
|
---|
6009 | elt = insert (src, classp, sets[i].src_hash, mode);
|
---|
6010 | elt->in_memory = sets[i].src_in_memory;
|
---|
6011 | sets[i].src_elt = classp = elt;
|
---|
6012 | }
|
---|
6013 | else
|
---|
6014 | sets[i].src_elt = classp;
|
---|
6015 | }
|
---|
6016 | if (sets[i].src_const && sets[i].src_const_elt == 0
|
---|
6017 | && src != sets[i].src_const
|
---|
6018 | && ! rtx_equal_p (sets[i].src_const, src))
|
---|
6019 | sets[i].src_elt = insert (sets[i].src_const, classp,
|
---|
6020 | sets[i].src_const_hash, mode);
|
---|
6021 | }
|
---|
6022 | }
|
---|
6023 | else if (sets[i].src_elt == 0)
|
---|
6024 | /* If we did not insert the source into the hash table (e.g., it was
|
---|
6025 | volatile), note the equivalence class for the REG_EQUAL value, if any,
|
---|
6026 | so that the destination goes into that class. */
|
---|
6027 | sets[i].src_elt = src_eqv_elt;
|
---|
6028 |
|
---|
6029 | invalidate_from_clobbers (x);
|
---|
6030 |
|
---|
6031 | /* Some registers are invalidated by subroutine calls. Memory is
|
---|
6032 | invalidated by non-constant calls. */
|
---|
6033 |
|
---|
6034 | if (GET_CODE (insn) == CALL_INSN)
|
---|
6035 | {
|
---|
6036 | if (! CONST_OR_PURE_CALL_P (insn))
|
---|
6037 | invalidate_memory ();
|
---|
6038 | invalidate_for_call ();
|
---|
6039 | }
|
---|
6040 |
|
---|
6041 | /* Now invalidate everything set by this instruction.
|
---|
6042 | If a SUBREG or other funny destination is being set,
|
---|
6043 | sets[i].rtl is still nonzero, so here we invalidate the reg
|
---|
6044 | a part of which is being set. */
|
---|
6045 |
|
---|
6046 | for (i = 0; i < n_sets; i++)
|
---|
6047 | if (sets[i].rtl)
|
---|
6048 | {
|
---|
6049 | /* We can't use the inner dest, because the mode associated with
|
---|
6050 | a ZERO_EXTRACT is significant. */
|
---|
6051 | rtx dest = SET_DEST (sets[i].rtl);
|
---|
6052 |
|
---|
6053 | /* Needed for registers to remove the register from its
|
---|
6054 | previous quantity's chain.
|
---|
6055 | Needed for memory if this is a nonvarying address, unless
|
---|
6056 | we have just done an invalidate_memory that covers even those. */
|
---|
6057 | if (GET_CODE (dest) == REG || GET_CODE (dest) == SUBREG)
|
---|
6058 | invalidate (dest, VOIDmode);
|
---|
6059 | else if (GET_CODE (dest) == MEM)
|
---|
6060 | {
|
---|
6061 | /* Outgoing arguments for a libcall don't
|
---|
6062 | affect any recorded expressions. */
|
---|
6063 | if (! libcall_insn || insn == libcall_insn)
|
---|
6064 | invalidate (dest, VOIDmode);
|
---|
6065 | }
|
---|
6066 | else if (GET_CODE (dest) == STRICT_LOW_PART
|
---|
6067 | || GET_CODE (dest) == ZERO_EXTRACT)
|
---|
6068 | invalidate (XEXP (dest, 0), GET_MODE (dest));
|
---|
6069 | }
|
---|
6070 |
|
---|
6071 | /* A volatile ASM invalidates everything. */
|
---|
6072 | if (GET_CODE (insn) == INSN
|
---|
6073 | && GET_CODE (PATTERN (insn)) == ASM_OPERANDS
|
---|
6074 | && MEM_VOLATILE_P (PATTERN (insn)))
|
---|
6075 | flush_hash_table ();
|
---|
6076 |
|
---|
6077 | /* Make sure registers mentioned in destinations
|
---|
6078 | are safe for use in an expression to be inserted.
|
---|
6079 | This removes from the hash table
|
---|
6080 | any invalid entry that refers to one of these registers.
|
---|
6081 |
|
---|
6082 | We don't care about the return value from mention_regs because
|
---|
6083 | we are going to hash the SET_DEST values unconditionally. */
|
---|
6084 |
|
---|
6085 | for (i = 0; i < n_sets; i++)
|
---|
6086 | {
|
---|
6087 | if (sets[i].rtl)
|
---|
6088 | {
|
---|
6089 | rtx x = SET_DEST (sets[i].rtl);
|
---|
6090 |
|
---|
6091 | if (GET_CODE (x) != REG)
|
---|
6092 | mention_regs (x);
|
---|
6093 | else
|
---|
6094 | {
|
---|
6095 | /* We used to rely on all references to a register becoming
|
---|
6096 | inaccessible when a register changes to a new quantity,
|
---|
6097 | since that changes the hash code. However, that is not
|
---|
6098 | safe, since after HASH_SIZE new quantities we get a
|
---|
6099 | hash 'collision' of a register with its own invalid
|
---|
6100 | entries. And since SUBREGs have been changed not to
|
---|
6101 | change their hash code with the hash code of the register,
|
---|
6102 | it wouldn't work any longer at all. So we have to check
|
---|
6103 | for any invalid references lying around now.
|
---|
6104 | This code is similar to the REG case in mention_regs,
|
---|
6105 | but it knows that reg_tick has been incremented, and
|
---|
6106 | it leaves reg_in_table as -1 . */
|
---|
6107 | unsigned int regno = REGNO (x);
|
---|
6108 | unsigned int endregno
|
---|
6109 | = regno + (regno >= FIRST_PSEUDO_REGISTER ? 1
|
---|
6110 | : HARD_REGNO_NREGS (regno, GET_MODE (x)));
|
---|
6111 | unsigned int i;
|
---|
6112 |
|
---|
6113 | for (i = regno; i < endregno; i++)
|
---|
6114 | {
|
---|
6115 | if (REG_IN_TABLE (i) >= 0)
|
---|
6116 | {
|
---|
6117 | remove_invalid_refs (i);
|
---|
6118 | REG_IN_TABLE (i) = -1;
|
---|
6119 | }
|
---|
6120 | }
|
---|
6121 | }
|
---|
6122 | }
|
---|
6123 | }
|
---|
6124 |
|
---|
6125 | /* We may have just removed some of the src_elt's from the hash table.
|
---|
6126 | So replace each one with the current head of the same class. */
|
---|
6127 |
|
---|
6128 | for (i = 0; i < n_sets; i++)
|
---|
6129 | if (sets[i].rtl)
|
---|
6130 | {
|
---|
6131 | if (sets[i].src_elt && sets[i].src_elt->first_same_value == 0)
|
---|
6132 | /* If elt was removed, find current head of same class,
|
---|
6133 | or 0 if nothing remains of that class. */
|
---|
6134 | {
|
---|
6135 | struct table_elt *elt = sets[i].src_elt;
|
---|
6136 |
|
---|
6137 | while (elt && elt->prev_same_value)
|
---|
6138 | elt = elt->prev_same_value;
|
---|
6139 |
|
---|
6140 | while (elt && elt->first_same_value == 0)
|
---|
6141 | elt = elt->next_same_value;
|
---|
6142 | sets[i].src_elt = elt ? elt->first_same_value : 0;
|
---|
6143 | }
|
---|
6144 | }
|
---|
6145 |
|
---|
6146 | /* Now insert the destinations into their equivalence classes. */
|
---|
6147 |
|
---|
6148 | for (i = 0; i < n_sets; i++)
|
---|
6149 | if (sets[i].rtl)
|
---|
6150 | {
|
---|
6151 | rtx dest = SET_DEST (sets[i].rtl);
|
---|
6152 | rtx inner_dest = sets[i].inner_dest;
|
---|
6153 | struct table_elt *elt;
|
---|
6154 |
|
---|
6155 | /* Don't record value if we are not supposed to risk allocating
|
---|
6156 | floating-point values in registers that might be wider than
|
---|
6157 | memory. */
|
---|
6158 | if ((flag_float_store
|
---|
6159 | && GET_CODE (dest) == MEM
|
---|
6160 | && FLOAT_MODE_P (GET_MODE (dest)))
|
---|
6161 | /* Don't record BLKmode values, because we don't know the
|
---|
6162 | size of it, and can't be sure that other BLKmode values
|
---|
6163 | have the same or smaller size. */
|
---|
6164 | || GET_MODE (dest) == BLKmode
|
---|
6165 | /* Don't record values of destinations set inside a libcall block
|
---|
6166 | since we might delete the libcall. Things should have been set
|
---|
6167 | up so we won't want to reuse such a value, but we play it safe
|
---|
6168 | here. */
|
---|
6169 | || libcall_insn
|
---|
6170 | /* If we didn't put a REG_EQUAL value or a source into the hash
|
---|
6171 | table, there is no point is recording DEST. */
|
---|
6172 | || sets[i].src_elt == 0
|
---|
6173 | /* If DEST is a paradoxical SUBREG and SRC is a ZERO_EXTEND
|
---|
6174 | or SIGN_EXTEND, don't record DEST since it can cause
|
---|
6175 | some tracking to be wrong.
|
---|
6176 |
|
---|
6177 | ??? Think about this more later. */
|
---|
6178 | || (GET_CODE (dest) == SUBREG
|
---|
6179 | && (GET_MODE_SIZE (GET_MODE (dest))
|
---|
6180 | > GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
|
---|
6181 | && (GET_CODE (sets[i].src) == SIGN_EXTEND
|
---|
6182 | || GET_CODE (sets[i].src) == ZERO_EXTEND)))
|
---|
6183 | continue;
|
---|
6184 |
|
---|
6185 | /* STRICT_LOW_PART isn't part of the value BEING set,
|
---|
6186 | and neither is the SUBREG inside it.
|
---|
6187 | Note that in this case SETS[I].SRC_ELT is really SRC_EQV_ELT. */
|
---|
6188 | if (GET_CODE (dest) == STRICT_LOW_PART)
|
---|
6189 | dest = SUBREG_REG (XEXP (dest, 0));
|
---|
6190 |
|
---|
6191 | if (GET_CODE (dest) == REG || GET_CODE (dest) == SUBREG)
|
---|
6192 | /* Registers must also be inserted into chains for quantities. */
|
---|
6193 | if (insert_regs (dest, sets[i].src_elt, 1))
|
---|
6194 | {
|
---|
6195 | /* If `insert_regs' changes something, the hash code must be
|
---|
6196 | recalculated. */
|
---|
6197 | rehash_using_reg (dest);
|
---|
6198 | sets[i].dest_hash = HASH (dest, GET_MODE (dest));
|
---|
6199 | }
|
---|
6200 |
|
---|
6201 | if (GET_CODE (inner_dest) == MEM
|
---|
6202 | && GET_CODE (XEXP (inner_dest, 0)) == ADDRESSOF)
|
---|
6203 | /* Given (SET (MEM (ADDRESSOF (X))) Y) we don't want to say
|
---|
6204 | that (MEM (ADDRESSOF (X))) is equivalent to Y.
|
---|
6205 | Consider the case in which the address of the MEM is
|
---|
6206 | passed to a function, which alters the MEM. Then, if we
|
---|
6207 | later use Y instead of the MEM we'll miss the update. */
|
---|
6208 | elt = insert (dest, 0, sets[i].dest_hash, GET_MODE (dest));
|
---|
6209 | else
|
---|
6210 | elt = insert (dest, sets[i].src_elt,
|
---|
6211 | sets[i].dest_hash, GET_MODE (dest));
|
---|
6212 |
|
---|
6213 | elt->in_memory = (GET_CODE (sets[i].inner_dest) == MEM
|
---|
6214 | && (! RTX_UNCHANGING_P (sets[i].inner_dest)
|
---|
6215 | || FIXED_BASE_PLUS_P (XEXP (sets[i].inner_dest,
|
---|
6216 | 0))));
|
---|
6217 |
|
---|
6218 | /* If we have (set (subreg:m1 (reg:m2 foo) 0) (bar:m1)), M1 is no
|
---|
6219 | narrower than M2, and both M1 and M2 are the same number of words,
|
---|
6220 | we are also doing (set (reg:m2 foo) (subreg:m2 (bar:m1) 0)) so
|
---|
6221 | make that equivalence as well.
|
---|
6222 |
|
---|
6223 | However, BAR may have equivalences for which gen_lowpart_if_possible
|
---|
6224 | will produce a simpler value than gen_lowpart_if_possible applied to
|
---|
6225 | BAR (e.g., if BAR was ZERO_EXTENDed from M2), so we will scan all
|
---|
6226 | BAR's equivalences. If we don't get a simplified form, make
|
---|
6227 | the SUBREG. It will not be used in an equivalence, but will
|
---|
6228 | cause two similar assignments to be detected.
|
---|
6229 |
|
---|
6230 | Note the loop below will find SUBREG_REG (DEST) since we have
|
---|
6231 | already entered SRC and DEST of the SET in the table. */
|
---|
6232 |
|
---|
6233 | if (GET_CODE (dest) == SUBREG
|
---|
6234 | && (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))) - 1)
|
---|
6235 | / UNITS_PER_WORD)
|
---|
6236 | == (GET_MODE_SIZE (GET_MODE (dest)) - 1) / UNITS_PER_WORD)
|
---|
6237 | && (GET_MODE_SIZE (GET_MODE (dest))
|
---|
6238 | >= GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest))))
|
---|
6239 | && sets[i].src_elt != 0)
|
---|
6240 | {
|
---|
6241 | enum machine_mode new_mode = GET_MODE (SUBREG_REG (dest));
|
---|
6242 | struct table_elt *elt, *classp = 0;
|
---|
6243 |
|
---|
6244 | for (elt = sets[i].src_elt->first_same_value; elt;
|
---|
6245 | elt = elt->next_same_value)
|
---|
6246 | {
|
---|
6247 | rtx new_src = 0;
|
---|
6248 | unsigned src_hash;
|
---|
6249 | struct table_elt *src_elt;
|
---|
6250 | int byte = 0;
|
---|
6251 |
|
---|
6252 | /* Ignore invalid entries. */
|
---|
6253 | if (GET_CODE (elt->exp) != REG
|
---|
6254 | && ! exp_equiv_p (elt->exp, elt->exp, 1, 0))
|
---|
6255 | continue;
|
---|
6256 |
|
---|
6257 | /* We may have already been playing subreg games. If the
|
---|
6258 | mode is already correct for the destination, use it. */
|
---|
6259 | if (GET_MODE (elt->exp) == new_mode)
|
---|
6260 | new_src = elt->exp;
|
---|
6261 | else
|
---|
6262 | {
|
---|
6263 | /* Calculate big endian correction for the SUBREG_BYTE.
|
---|
6264 | We have already checked that M1 (GET_MODE (dest))
|
---|
6265 | is not narrower than M2 (new_mode). */
|
---|
6266 | if (BYTES_BIG_ENDIAN)
|
---|
6267 | byte = (GET_MODE_SIZE (GET_MODE (dest))
|
---|
6268 | - GET_MODE_SIZE (new_mode));
|
---|
6269 |
|
---|
6270 | new_src = simplify_gen_subreg (new_mode, elt->exp,
|
---|
6271 | GET_MODE (dest), byte);
|
---|
6272 | }
|
---|
6273 |
|
---|
6274 | /* The call to simplify_gen_subreg fails if the value
|
---|
6275 | is VOIDmode, yet we can't do any simplification, e.g.
|
---|
6276 | for EXPR_LISTs denoting function call results.
|
---|
6277 | It is invalid to construct a SUBREG with a VOIDmode
|
---|
6278 | SUBREG_REG, hence a zero new_src means we can't do
|
---|
6279 | this substitution. */
|
---|
6280 | if (! new_src)
|
---|
6281 | continue;
|
---|
6282 |
|
---|
6283 | src_hash = HASH (new_src, new_mode);
|
---|
6284 | src_elt = lookup (new_src, src_hash, new_mode);
|
---|
6285 |
|
---|
6286 | /* Put the new source in the hash table is if isn't
|
---|
6287 | already. */
|
---|
6288 | if (src_elt == 0)
|
---|
6289 | {
|
---|
6290 | if (insert_regs (new_src, classp, 0))
|
---|
6291 | {
|
---|
6292 | rehash_using_reg (new_src);
|
---|
6293 | src_hash = HASH (new_src, new_mode);
|
---|
6294 | }
|
---|
6295 | src_elt = insert (new_src, classp, src_hash, new_mode);
|
---|
6296 | src_elt->in_memory = elt->in_memory;
|
---|
6297 | }
|
---|
6298 | else if (classp && classp != src_elt->first_same_value)
|
---|
6299 | /* Show that two things that we've seen before are
|
---|
6300 | actually the same. */
|
---|
6301 | merge_equiv_classes (src_elt, classp);
|
---|
6302 |
|
---|
6303 | classp = src_elt->first_same_value;
|
---|
6304 | /* Ignore invalid entries. */
|
---|
6305 | while (classp
|
---|
6306 | && GET_CODE (classp->exp) != REG
|
---|
6307 | && ! exp_equiv_p (classp->exp, classp->exp, 1, 0))
|
---|
6308 | classp = classp->next_same_value;
|
---|
6309 | }
|
---|
6310 | }
|
---|
6311 | }
|
---|
6312 |
|
---|
6313 | /* Special handling for (set REG0 REG1) where REG0 is the
|
---|
6314 | "cheapest", cheaper than REG1. After cse, REG1 will probably not
|
---|
6315 | be used in the sequel, so (if easily done) change this insn to
|
---|
6316 | (set REG1 REG0) and replace REG1 with REG0 in the previous insn
|
---|
6317 | that computed their value. Then REG1 will become a dead store
|
---|
6318 | and won't cloud the situation for later optimizations.
|
---|
6319 |
|
---|
6320 | Do not make this change if REG1 is a hard register, because it will
|
---|
6321 | then be used in the sequel and we may be changing a two-operand insn
|
---|
6322 | into a three-operand insn.
|
---|
6323 |
|
---|
6324 | Also do not do this if we are operating on a copy of INSN.
|
---|
6325 |
|
---|
6326 | Also don't do this if INSN ends a libcall; this would cause an unrelated
|
---|
6327 | register to be set in the middle of a libcall, and we then get bad code
|
---|
6328 | if the libcall is deleted. */
|
---|
6329 |
|
---|
6330 | if (n_sets == 1 && sets[0].rtl && GET_CODE (SET_DEST (sets[0].rtl)) == REG
|
---|
6331 | && NEXT_INSN (PREV_INSN (insn)) == insn
|
---|
6332 | && GET_CODE (SET_SRC (sets[0].rtl)) == REG
|
---|
6333 | && REGNO (SET_SRC (sets[0].rtl)) >= FIRST_PSEUDO_REGISTER
|
---|
6334 | && REGNO_QTY_VALID_P (REGNO (SET_SRC (sets[0].rtl))))
|
---|
6335 | {
|
---|
6336 | int src_q = REG_QTY (REGNO (SET_SRC (sets[0].rtl)));
|
---|
6337 | struct qty_table_elem *src_ent = &qty_table[src_q];
|
---|
6338 |
|
---|
6339 | if ((src_ent->first_reg == REGNO (SET_DEST (sets[0].rtl)))
|
---|
6340 | && ! find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
---|
6341 | {
|
---|
6342 | rtx prev = insn;
|
---|
6343 | /* Scan for the previous nonnote insn, but stop at a basic
|
---|
6344 | block boundary. */
|
---|
6345 | do
|
---|
6346 | {
|
---|
6347 | prev = PREV_INSN (prev);
|
---|
6348 | }
|
---|
6349 | while (prev && GET_CODE (prev) == NOTE
|
---|
6350 | && NOTE_LINE_NUMBER (prev) != NOTE_INSN_BASIC_BLOCK);
|
---|
6351 |
|
---|
6352 | /* Do not swap the registers around if the previous instruction
|
---|
6353 | attaches a REG_EQUIV note to REG1.
|
---|
6354 |
|
---|
6355 | ??? It's not entirely clear whether we can transfer a REG_EQUIV
|
---|
6356 | from the pseudo that originally shadowed an incoming argument
|
---|
6357 | to another register. Some uses of REG_EQUIV might rely on it
|
---|
6358 | being attached to REG1 rather than REG2.
|
---|
6359 |
|
---|
6360 | This section previously turned the REG_EQUIV into a REG_EQUAL
|
---|
6361 | note. We cannot do that because REG_EQUIV may provide an
|
---|
6362 | uninitialized stack slot when REG_PARM_STACK_SPACE is used. */
|
---|
6363 |
|
---|
6364 | if (prev != 0 && GET_CODE (prev) == INSN
|
---|
6365 | && GET_CODE (PATTERN (prev)) == SET
|
---|
6366 | && SET_DEST (PATTERN (prev)) == SET_SRC (sets[0].rtl)
|
---|
6367 | && ! find_reg_note (prev, REG_EQUIV, NULL_RTX))
|
---|
6368 | {
|
---|
6369 | rtx dest = SET_DEST (sets[0].rtl);
|
---|
6370 | rtx src = SET_SRC (sets[0].rtl);
|
---|
6371 | rtx note;
|
---|
6372 |
|
---|
6373 | validate_change (prev, &SET_DEST (PATTERN (prev)), dest, 1);
|
---|
6374 | validate_change (insn, &SET_DEST (sets[0].rtl), src, 1);
|
---|
6375 | validate_change (insn, &SET_SRC (sets[0].rtl), dest, 1);
|
---|
6376 | apply_change_group ();
|
---|
6377 |
|
---|
6378 | /* If there was a REG_WAS_0 note on PREV, remove it. Move
|
---|
6379 | any REG_WAS_0 note on INSN to PREV. */
|
---|
6380 | note = find_reg_note (prev, REG_WAS_0, NULL_RTX);
|
---|
6381 | if (note)
|
---|
6382 | remove_note (prev, note);
|
---|
6383 |
|
---|
6384 | note = find_reg_note (insn, REG_WAS_0, NULL_RTX);
|
---|
6385 | if (note)
|
---|
6386 | {
|
---|
6387 | remove_note (insn, note);
|
---|
6388 | XEXP (note, 1) = REG_NOTES (prev);
|
---|
6389 | REG_NOTES (prev) = note;
|
---|
6390 | }
|
---|
6391 |
|
---|
6392 | /* If INSN has a REG_EQUAL note, and this note mentions
|
---|
6393 | REG0, then we must delete it, because the value in
|
---|
6394 | REG0 has changed. If the note's value is REG1, we must
|
---|
6395 | also delete it because that is now this insn's dest. */
|
---|
6396 | note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
|
---|
6397 | if (note != 0
|
---|
6398 | && (reg_mentioned_p (dest, XEXP (note, 0))
|
---|
6399 | || rtx_equal_p (src, XEXP (note, 0))))
|
---|
6400 | remove_note (insn, note);
|
---|
6401 | }
|
---|
6402 | }
|
---|
6403 | }
|
---|
6404 |
|
---|
6405 | /* If this is a conditional jump insn, record any known equivalences due to
|
---|
6406 | the condition being tested. */
|
---|
6407 |
|
---|
6408 | last_jump_equiv_class = 0;
|
---|
6409 | if (GET_CODE (insn) == JUMP_INSN
|
---|
6410 | && n_sets == 1 && GET_CODE (x) == SET
|
---|
6411 | && GET_CODE (SET_SRC (x)) == IF_THEN_ELSE)
|
---|
6412 | record_jump_equiv (insn, 0);
|
---|
6413 |
|
---|
6414 | #ifdef HAVE_cc0
|
---|
6415 | /* If the previous insn set CC0 and this insn no longer references CC0,
|
---|
6416 | delete the previous insn. Here we use the fact that nothing expects CC0
|
---|
6417 | to be valid over an insn, which is true until the final pass. */
|
---|
6418 | if (prev_insn && GET_CODE (prev_insn) == INSN
|
---|
6419 | && (tem = single_set (prev_insn)) != 0
|
---|
6420 | && SET_DEST (tem) == cc0_rtx
|
---|
6421 | && ! reg_mentioned_p (cc0_rtx, x))
|
---|
6422 | delete_insn (prev_insn);
|
---|
6423 |
|
---|
6424 | prev_insn_cc0 = this_insn_cc0;
|
---|
6425 | prev_insn_cc0_mode = this_insn_cc0_mode;
|
---|
6426 | #endif
|
---|
6427 |
|
---|
6428 | prev_insn = insn;
|
---|
6429 | }
|
---|
6430 | |
---|
6431 |
|
---|
6432 | /* Remove from the hash table all expressions that reference memory. */
|
---|
6433 |
|
---|
6434 | static void
|
---|
6435 | invalidate_memory ()
|
---|
6436 | {
|
---|
6437 | int i;
|
---|
6438 | struct table_elt *p, *next;
|
---|
6439 |
|
---|
6440 | for (i = 0; i < HASH_SIZE; i++)
|
---|
6441 | for (p = table[i]; p; p = next)
|
---|
6442 | {
|
---|
6443 | next = p->next_same_hash;
|
---|
6444 | if (p->in_memory)
|
---|
6445 | remove_from_table (p, i);
|
---|
6446 | }
|
---|
6447 | }
|
---|
6448 |
|
---|
6449 | /* If ADDR is an address that implicitly affects the stack pointer, return
|
---|
6450 | 1 and update the register tables to show the effect. Else, return 0. */
|
---|
6451 |
|
---|
6452 | static int
|
---|
6453 | addr_affects_sp_p (addr)
|
---|
6454 | rtx addr;
|
---|
6455 | {
|
---|
6456 | if (GET_RTX_CLASS (GET_CODE (addr)) == 'a'
|
---|
6457 | && GET_CODE (XEXP (addr, 0)) == REG
|
---|
6458 | && REGNO (XEXP (addr, 0)) == STACK_POINTER_REGNUM)
|
---|
6459 | {
|
---|
6460 | if (REG_TICK (STACK_POINTER_REGNUM) >= 0)
|
---|
6461 | {
|
---|
6462 | REG_TICK (STACK_POINTER_REGNUM)++;
|
---|
6463 | /* Is it possible to use a subreg of SP? */
|
---|
6464 | SUBREG_TICKED (STACK_POINTER_REGNUM) = -1;
|
---|
6465 | }
|
---|
6466 |
|
---|
6467 | /* This should be *very* rare. */
|
---|
6468 | if (TEST_HARD_REG_BIT (hard_regs_in_table, STACK_POINTER_REGNUM))
|
---|
6469 | invalidate (stack_pointer_rtx, VOIDmode);
|
---|
6470 |
|
---|
6471 | return 1;
|
---|
6472 | }
|
---|
6473 |
|
---|
6474 | return 0;
|
---|
6475 | }
|
---|
6476 |
|
---|
6477 | /* Perform invalidation on the basis of everything about an insn
|
---|
6478 | except for invalidating the actual places that are SET in it.
|
---|
6479 | This includes the places CLOBBERed, and anything that might
|
---|
6480 | alias with something that is SET or CLOBBERed.
|
---|
6481 |
|
---|
6482 | X is the pattern of the insn. */
|
---|
6483 |
|
---|
6484 | static void
|
---|
6485 | invalidate_from_clobbers (x)
|
---|
6486 | rtx x;
|
---|
6487 | {
|
---|
6488 | if (GET_CODE (x) == CLOBBER)
|
---|
6489 | {
|
---|
6490 | rtx ref = XEXP (x, 0);
|
---|
6491 | if (ref)
|
---|
6492 | {
|
---|
6493 | if (GET_CODE (ref) == REG || GET_CODE (ref) == SUBREG
|
---|
6494 | || GET_CODE (ref) == MEM)
|
---|
6495 | invalidate (ref, VOIDmode);
|
---|
6496 | else if (GET_CODE (ref) == STRICT_LOW_PART
|
---|
6497 | || GET_CODE (ref) == ZERO_EXTRACT)
|
---|
6498 | invalidate (XEXP (ref, 0), GET_MODE (ref));
|
---|
6499 | }
|
---|
6500 | }
|
---|
6501 | else if (GET_CODE (x) == PARALLEL)
|
---|
6502 | {
|
---|
6503 | int i;
|
---|
6504 | for (i = XVECLEN (x, 0) - 1; i >= 0; i--)
|
---|
6505 | {
|
---|
6506 | rtx y = XVECEXP (x, 0, i);
|
---|
6507 | if (GET_CODE (y) == CLOBBER)
|
---|
6508 | {
|
---|
6509 | rtx ref = XEXP (y, 0);
|
---|
6510 | if (GET_CODE (ref) == REG || GET_CODE (ref) == SUBREG
|
---|
6511 | || GET_CODE (ref) == MEM)
|
---|
6512 | invalidate (ref, VOIDmode);
|
---|
6513 | else if (GET_CODE (ref) == STRICT_LOW_PART
|
---|
6514 | || GET_CODE (ref) == ZERO_EXTRACT)
|
---|
6515 | invalidate (XEXP (ref, 0), GET_MODE (ref));
|
---|
6516 | }
|
---|
6517 | }
|
---|
6518 | }
|
---|
6519 | }
|
---|
6520 | |
---|
6521 |
|
---|
6522 | /* Process X, part of the REG_NOTES of an insn. Look at any REG_EQUAL notes
|
---|
6523 | and replace any registers in them with either an equivalent constant
|
---|
6524 | or the canonical form of the register. If we are inside an address,
|
---|
6525 | only do this if the address remains valid.
|
---|
6526 |
|
---|
6527 | OBJECT is 0 except when within a MEM in which case it is the MEM.
|
---|
6528 |
|
---|
6529 | Return the replacement for X. */
|
---|
6530 |
|
---|
6531 | static rtx
|
---|
6532 | cse_process_notes (x, object)
|
---|
6533 | rtx x;
|
---|
6534 | rtx object;
|
---|
6535 | {
|
---|
6536 | enum rtx_code code = GET_CODE (x);
|
---|
6537 | const char *fmt = GET_RTX_FORMAT (code);
|
---|
6538 | int i;
|
---|
6539 |
|
---|
6540 | switch (code)
|
---|
6541 | {
|
---|
6542 | case CONST_INT:
|
---|
6543 | case CONST:
|
---|
6544 | case SYMBOL_REF:
|
---|
6545 | case LABEL_REF:
|
---|
6546 | case CONST_DOUBLE:
|
---|
6547 | case CONST_VECTOR:
|
---|
6548 | case PC:
|
---|
6549 | case CC0:
|
---|
6550 | case LO_SUM:
|
---|
6551 | return x;
|
---|
6552 |
|
---|
6553 | case MEM:
|
---|
6554 | validate_change (x, &XEXP (x, 0),
|
---|
6555 | cse_process_notes (XEXP (x, 0), x), 0);
|
---|
6556 | return x;
|
---|
6557 |
|
---|
6558 | case EXPR_LIST:
|
---|
6559 | case INSN_LIST:
|
---|
6560 | if (REG_NOTE_KIND (x) == REG_EQUAL)
|
---|
6561 | XEXP (x, 0) = cse_process_notes (XEXP (x, 0), NULL_RTX);
|
---|
6562 | if (XEXP (x, 1))
|
---|
6563 | XEXP (x, 1) = cse_process_notes (XEXP (x, 1), NULL_RTX);
|
---|
6564 | return x;
|
---|
6565 |
|
---|
6566 | case SIGN_EXTEND:
|
---|
6567 | case ZERO_EXTEND:
|
---|
6568 | case SUBREG:
|
---|
6569 | {
|
---|
6570 | rtx new = cse_process_notes (XEXP (x, 0), object);
|
---|
6571 | /* We don't substitute VOIDmode constants into these rtx,
|
---|
6572 | since they would impede folding. */
|
---|
6573 | if (GET_MODE (new) != VOIDmode)
|
---|
6574 | validate_change (object, &XEXP (x, 0), new, 0);
|
---|
6575 | return x;
|
---|
6576 | }
|
---|
6577 |
|
---|
6578 | case REG:
|
---|
6579 | i = REG_QTY (REGNO (x));
|
---|
6580 |
|
---|
6581 | /* Return a constant or a constant register. */
|
---|
6582 | if (REGNO_QTY_VALID_P (REGNO (x)))
|
---|
6583 | {
|
---|
6584 | struct qty_table_elem *ent = &qty_table[i];
|
---|
6585 |
|
---|
6586 | if (ent->const_rtx != NULL_RTX
|
---|
6587 | && (CONSTANT_P (ent->const_rtx)
|
---|
6588 | || GET_CODE (ent->const_rtx) == REG))
|
---|
6589 | {
|
---|
6590 | rtx new = gen_lowpart_if_possible (GET_MODE (x), ent->const_rtx);
|
---|
6591 | if (new)
|
---|
6592 | return new;
|
---|
6593 | }
|
---|
6594 | }
|
---|
6595 |
|
---|
6596 | /* Otherwise, canonicalize this register. */
|
---|
6597 | return canon_reg (x, NULL_RTX);
|
---|
6598 |
|
---|
6599 | default:
|
---|
6600 | break;
|
---|
6601 | }
|
---|
6602 |
|
---|
6603 | for (i = 0; i < GET_RTX_LENGTH (code); i++)
|
---|
6604 | if (fmt[i] == 'e')
|
---|
6605 | validate_change (object, &XEXP (x, i),
|
---|
6606 | cse_process_notes (XEXP (x, i), object), 0);
|
---|
6607 |
|
---|
6608 | return x;
|
---|
6609 | }
|
---|
6610 | |
---|
6611 |
|
---|
6612 | /* Find common subexpressions between the end test of a loop and the beginning
|
---|
6613 | of the loop. LOOP_START is the CODE_LABEL at the start of a loop.
|
---|
6614 |
|
---|
6615 | Often we have a loop where an expression in the exit test is used
|
---|
6616 | in the body of the loop. For example "while (*p) *q++ = *p++;".
|
---|
6617 | Because of the way we duplicate the loop exit test in front of the loop,
|
---|
6618 | however, we don't detect that common subexpression. This will be caught
|
---|
6619 | when global cse is implemented, but this is a quite common case.
|
---|
6620 |
|
---|
6621 | This function handles the most common cases of these common expressions.
|
---|
6622 | It is called after we have processed the basic block ending with the
|
---|
6623 | NOTE_INSN_LOOP_END note that ends a loop and the previous JUMP_INSN
|
---|
6624 | jumps to a label used only once. */
|
---|
6625 |
|
---|
6626 | static void
|
---|
6627 | cse_around_loop (loop_start)
|
---|
6628 | rtx loop_start;
|
---|
6629 | {
|
---|
6630 | rtx insn;
|
---|
6631 | int i;
|
---|
6632 | struct table_elt *p;
|
---|
6633 |
|
---|
6634 | /* If the jump at the end of the loop doesn't go to the start, we don't
|
---|
6635 | do anything. */
|
---|
6636 | for (insn = PREV_INSN (loop_start);
|
---|
6637 | insn && (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) >= 0);
|
---|
6638 | insn = PREV_INSN (insn))
|
---|
6639 | ;
|
---|
6640 |
|
---|
6641 | if (insn == 0
|
---|
6642 | || GET_CODE (insn) != NOTE
|
---|
6643 | || NOTE_LINE_NUMBER (insn) != NOTE_INSN_LOOP_BEG)
|
---|
6644 | return;
|
---|
6645 |
|
---|
6646 | /* If the last insn of the loop (the end test) was an NE comparison,
|
---|
6647 | we will interpret it as an EQ comparison, since we fell through
|
---|
6648 | the loop. Any equivalences resulting from that comparison are
|
---|
6649 | therefore not valid and must be invalidated. */
|
---|
6650 | if (last_jump_equiv_class)
|
---|
6651 | for (p = last_jump_equiv_class->first_same_value; p;
|
---|
6652 | p = p->next_same_value)
|
---|
6653 | {
|
---|
6654 | if (GET_CODE (p->exp) == MEM || GET_CODE (p->exp) == REG
|
---|
6655 | || (GET_CODE (p->exp) == SUBREG
|
---|
6656 | && GET_CODE (SUBREG_REG (p->exp)) == REG))
|
---|
6657 | invalidate (p->exp, VOIDmode);
|
---|
6658 | else if (GET_CODE (p->exp) == STRICT_LOW_PART
|
---|
6659 | || GET_CODE (p->exp) == ZERO_EXTRACT)
|
---|
6660 | invalidate (XEXP (p->exp, 0), GET_MODE (p->exp));
|
---|
6661 | }
|
---|
6662 |
|
---|
6663 | /* Process insns starting after LOOP_START until we hit a CALL_INSN or
|
---|
6664 | a CODE_LABEL (we could handle a CALL_INSN, but it isn't worth it).
|
---|
6665 |
|
---|
6666 | The only thing we do with SET_DEST is invalidate entries, so we
|
---|
6667 | can safely process each SET in order. It is slightly less efficient
|
---|
6668 | to do so, but we only want to handle the most common cases.
|
---|
6669 |
|
---|
6670 | The gen_move_insn call in cse_set_around_loop may create new pseudos.
|
---|
6671 | These pseudos won't have valid entries in any of the tables indexed
|
---|
6672 | by register number, such as reg_qty. We avoid out-of-range array
|
---|
6673 | accesses by not processing any instructions created after cse started. */
|
---|
6674 |
|
---|
6675 | for (insn = NEXT_INSN (loop_start);
|
---|
6676 | GET_CODE (insn) != CALL_INSN && GET_CODE (insn) != CODE_LABEL
|
---|
6677 | && INSN_UID (insn) < max_insn_uid
|
---|
6678 | && ! (GET_CODE (insn) == NOTE
|
---|
6679 | && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_END);
|
---|
6680 | insn = NEXT_INSN (insn))
|
---|
6681 | {
|
---|
6682 | if (INSN_P (insn)
|
---|
6683 | && (GET_CODE (PATTERN (insn)) == SET
|
---|
6684 | || GET_CODE (PATTERN (insn)) == CLOBBER))
|
---|
6685 | cse_set_around_loop (PATTERN (insn), insn, loop_start);
|
---|
6686 | else if (INSN_P (insn) && GET_CODE (PATTERN (insn)) == PARALLEL)
|
---|
6687 | for (i = XVECLEN (PATTERN (insn), 0) - 1; i >= 0; i--)
|
---|
6688 | if (GET_CODE (XVECEXP (PATTERN (insn), 0, i)) == SET
|
---|
6689 | || GET_CODE (XVECEXP (PATTERN (insn), 0, i)) == CLOBBER)
|
---|
6690 | cse_set_around_loop (XVECEXP (PATTERN (insn), 0, i), insn,
|
---|
6691 | loop_start);
|
---|
6692 | }
|
---|
6693 | }
|
---|
6694 | |
---|
6695 |
|
---|
6696 | /* Process one SET of an insn that was skipped. We ignore CLOBBERs
|
---|
6697 | since they are done elsewhere. This function is called via note_stores. */
|
---|
6698 |
|
---|
6699 | static void
|
---|
6700 | invalidate_skipped_set (dest, set, data)
|
---|
6701 | rtx set;
|
---|
6702 | rtx dest;
|
---|
6703 | void *data ATTRIBUTE_UNUSED;
|
---|
6704 | {
|
---|
6705 | enum rtx_code code = GET_CODE (dest);
|
---|
6706 |
|
---|
6707 | if (code == MEM
|
---|
6708 | && ! addr_affects_sp_p (dest) /* If this is not a stack push ... */
|
---|
6709 | /* There are times when an address can appear varying and be a PLUS
|
---|
6710 | during this scan when it would be a fixed address were we to know
|
---|
6711 | the proper equivalences. So invalidate all memory if there is
|
---|
6712 | a BLKmode or nonscalar memory reference or a reference to a
|
---|
6713 | variable address. */
|
---|
6714 | && (MEM_IN_STRUCT_P (dest) || GET_MODE (dest) == BLKmode
|
---|
6715 | || cse_rtx_varies_p (XEXP (dest, 0), 0)))
|
---|
6716 | {
|
---|
6717 | invalidate_memory ();
|
---|
6718 | return;
|
---|
6719 | }
|
---|
6720 |
|
---|
6721 | if (GET_CODE (set) == CLOBBER
|
---|
6722 | #ifdef HAVE_cc0
|
---|
6723 | || dest == cc0_rtx
|
---|
6724 | #endif
|
---|
6725 | || dest == pc_rtx)
|
---|
6726 | return;
|
---|
6727 |
|
---|
6728 | if (code == STRICT_LOW_PART || code == ZERO_EXTRACT)
|
---|
6729 | invalidate (XEXP (dest, 0), GET_MODE (dest));
|
---|
6730 | else if (code == REG || code == SUBREG || code == MEM)
|
---|
6731 | invalidate (dest, VOIDmode);
|
---|
6732 | }
|
---|
6733 |
|
---|
6734 | /* Invalidate all insns from START up to the end of the function or the
|
---|
6735 | next label. This called when we wish to CSE around a block that is
|
---|
6736 | conditionally executed. */
|
---|
6737 |
|
---|
6738 | static void
|
---|
6739 | invalidate_skipped_block (start)
|
---|
6740 | rtx start;
|
---|
6741 | {
|
---|
6742 | rtx insn;
|
---|
6743 |
|
---|
6744 | for (insn = start; insn && GET_CODE (insn) != CODE_LABEL;
|
---|
6745 | insn = NEXT_INSN (insn))
|
---|
6746 | {
|
---|
6747 | if (! INSN_P (insn))
|
---|
6748 | continue;
|
---|
6749 |
|
---|
6750 | if (GET_CODE (insn) == CALL_INSN)
|
---|
6751 | {
|
---|
6752 | if (! CONST_OR_PURE_CALL_P (insn))
|
---|
6753 | invalidate_memory ();
|
---|
6754 | invalidate_for_call ();
|
---|
6755 | }
|
---|
6756 |
|
---|
6757 | invalidate_from_clobbers (PATTERN (insn));
|
---|
6758 | note_stores (PATTERN (insn), invalidate_skipped_set, NULL);
|
---|
6759 | }
|
---|
6760 | }
|
---|
6761 | |
---|
6762 |
|
---|
6763 | /* If modifying X will modify the value in *DATA (which is really an
|
---|
6764 | `rtx *'), indicate that fact by setting the pointed to value to
|
---|
6765 | NULL_RTX. */
|
---|
6766 |
|
---|
6767 | static void
|
---|
6768 | cse_check_loop_start (x, set, data)
|
---|
6769 | rtx x;
|
---|
6770 | rtx set ATTRIBUTE_UNUSED;
|
---|
6771 | void *data;
|
---|
6772 | {
|
---|
6773 | rtx *cse_check_loop_start_value = (rtx *) data;
|
---|
6774 |
|
---|
6775 | if (*cse_check_loop_start_value == NULL_RTX
|
---|
6776 | || GET_CODE (x) == CC0 || GET_CODE (x) == PC)
|
---|
6777 | return;
|
---|
6778 |
|
---|
6779 | if ((GET_CODE (x) == MEM && GET_CODE (*cse_check_loop_start_value) == MEM)
|
---|
6780 | || reg_overlap_mentioned_p (x, *cse_check_loop_start_value))
|
---|
6781 | *cse_check_loop_start_value = NULL_RTX;
|
---|
6782 | }
|
---|
6783 |
|
---|
6784 | /* X is a SET or CLOBBER contained in INSN that was found near the start of
|
---|
6785 | a loop that starts with the label at LOOP_START.
|
---|
6786 |
|
---|
6787 | If X is a SET, we see if its SET_SRC is currently in our hash table.
|
---|
6788 | If so, we see if it has a value equal to some register used only in the
|
---|
6789 | loop exit code (as marked by jump.c).
|
---|
6790 |
|
---|
6791 | If those two conditions are true, we search backwards from the start of
|
---|
6792 | the loop to see if that same value was loaded into a register that still
|
---|
6793 | retains its value at the start of the loop.
|
---|
6794 |
|
---|
6795 | If so, we insert an insn after the load to copy the destination of that
|
---|
6796 | load into the equivalent register and (try to) replace our SET_SRC with that
|
---|
6797 | register.
|
---|
6798 |
|
---|
6799 | In any event, we invalidate whatever this SET or CLOBBER modifies. */
|
---|
6800 |
|
---|
6801 | static void
|
---|
6802 | cse_set_around_loop (x, insn, loop_start)
|
---|
6803 | rtx x;
|
---|
6804 | rtx insn;
|
---|
6805 | rtx loop_start;
|
---|
6806 | {
|
---|
6807 | struct table_elt *src_elt;
|
---|
6808 |
|
---|
6809 | /* If this is a SET, see if we can replace SET_SRC, but ignore SETs that
|
---|
6810 | are setting PC or CC0 or whose SET_SRC is already a register. */
|
---|
6811 | if (GET_CODE (x) == SET
|
---|
6812 | && GET_CODE (SET_DEST (x)) != PC && GET_CODE (SET_DEST (x)) != CC0
|
---|
6813 | && GET_CODE (SET_SRC (x)) != REG)
|
---|
6814 | {
|
---|
6815 | src_elt = lookup (SET_SRC (x),
|
---|
6816 | HASH (SET_SRC (x), GET_MODE (SET_DEST (x))),
|
---|
6817 | GET_MODE (SET_DEST (x)));
|
---|
6818 |
|
---|
6819 | if (src_elt)
|
---|
6820 | for (src_elt = src_elt->first_same_value; src_elt;
|
---|
6821 | src_elt = src_elt->next_same_value)
|
---|
6822 | if (GET_CODE (src_elt->exp) == REG && REG_LOOP_TEST_P (src_elt->exp)
|
---|
6823 | && COST (src_elt->exp) < COST (SET_SRC (x)))
|
---|
6824 | {
|
---|
6825 | rtx p, set;
|
---|
6826 |
|
---|
6827 | /* Look for an insn in front of LOOP_START that sets
|
---|
6828 | something in the desired mode to SET_SRC (x) before we hit
|
---|
6829 | a label or CALL_INSN. */
|
---|
6830 |
|
---|
6831 | for (p = prev_nonnote_insn (loop_start);
|
---|
6832 | p && GET_CODE (p) != CALL_INSN
|
---|
6833 | && GET_CODE (p) != CODE_LABEL;
|
---|
6834 | p = prev_nonnote_insn (p))
|
---|
6835 | if ((set = single_set (p)) != 0
|
---|
6836 | && GET_CODE (SET_DEST (set)) == REG
|
---|
6837 | && GET_MODE (SET_DEST (set)) == src_elt->mode
|
---|
6838 | && rtx_equal_p (SET_SRC (set), SET_SRC (x)))
|
---|
6839 | {
|
---|
6840 | /* We now have to ensure that nothing between P
|
---|
6841 | and LOOP_START modified anything referenced in
|
---|
6842 | SET_SRC (x). We know that nothing within the loop
|
---|
6843 | can modify it, or we would have invalidated it in
|
---|
6844 | the hash table. */
|
---|
6845 | rtx q;
|
---|
6846 | rtx cse_check_loop_start_value = SET_SRC (x);
|
---|
6847 | for (q = p; q != loop_start; q = NEXT_INSN (q))
|
---|
6848 | if (INSN_P (q))
|
---|
6849 | note_stores (PATTERN (q),
|
---|
6850 | cse_check_loop_start,
|
---|
6851 | &cse_check_loop_start_value);
|
---|
6852 |
|
---|
6853 | /* If nothing was changed and we can replace our
|
---|
6854 | SET_SRC, add an insn after P to copy its destination
|
---|
6855 | to what we will be replacing SET_SRC with. */
|
---|
6856 | if (cse_check_loop_start_value
|
---|
6857 | && validate_change (insn, &SET_SRC (x),
|
---|
6858 | src_elt->exp, 0))
|
---|
6859 | {
|
---|
6860 | /* If this creates new pseudos, this is unsafe,
|
---|
6861 | because the regno of new pseudo is unsuitable
|
---|
6862 | to index into reg_qty when cse_insn processes
|
---|
6863 | the new insn. Therefore, if a new pseudo was
|
---|
6864 | created, discard this optimization. */
|
---|
6865 | int nregs = max_reg_num ();
|
---|
6866 | rtx move
|
---|
6867 | = gen_move_insn (src_elt->exp, SET_DEST (set));
|
---|
6868 | if (nregs != max_reg_num ())
|
---|
6869 | {
|
---|
6870 | if (! validate_change (insn, &SET_SRC (x),
|
---|
6871 | SET_SRC (set), 0))
|
---|
6872 | abort ();
|
---|
6873 | }
|
---|
6874 | else
|
---|
6875 | {
|
---|
6876 | if (control_flow_insn_p (p))
|
---|
6877 | /* p can cause a control flow transfer so it
|
---|
6878 | is the last insn of a basic block. We can't
|
---|
6879 | therefore use emit_insn_after. */
|
---|
6880 | emit_insn_before (move, next_nonnote_insn (p));
|
---|
6881 | else
|
---|
6882 | emit_insn_after (move, p);
|
---|
6883 | }
|
---|
6884 | }
|
---|
6885 | break;
|
---|
6886 | }
|
---|
6887 | }
|
---|
6888 | }
|
---|
6889 |
|
---|
6890 | /* Deal with the destination of X affecting the stack pointer. */
|
---|
6891 | addr_affects_sp_p (SET_DEST (x));
|
---|
6892 |
|
---|
6893 | /* See comment on similar code in cse_insn for explanation of these
|
---|
6894 | tests. */
|
---|
6895 | if (GET_CODE (SET_DEST (x)) == REG || GET_CODE (SET_DEST (x)) == SUBREG
|
---|
6896 | || GET_CODE (SET_DEST (x)) == MEM)
|
---|
6897 | invalidate (SET_DEST (x), VOIDmode);
|
---|
6898 | else if (GET_CODE (SET_DEST (x)) == STRICT_LOW_PART
|
---|
6899 | || GET_CODE (SET_DEST (x)) == ZERO_EXTRACT)
|
---|
6900 | invalidate (XEXP (SET_DEST (x), 0), GET_MODE (SET_DEST (x)));
|
---|
6901 | }
|
---|
6902 | |
---|
6903 |
|
---|
6904 | /* Find the end of INSN's basic block and return its range,
|
---|
6905 | the total number of SETs in all the insns of the block, the last insn of the
|
---|
6906 | block, and the branch path.
|
---|
6907 |
|
---|
6908 | The branch path indicates which branches should be followed. If a nonzero
|
---|
6909 | path size is specified, the block should be rescanned and a different set
|
---|
6910 | of branches will be taken. The branch path is only used if
|
---|
6911 | FLAG_CSE_FOLLOW_JUMPS or FLAG_CSE_SKIP_BLOCKS is nonzero.
|
---|
6912 |
|
---|
6913 | DATA is a pointer to a struct cse_basic_block_data, defined below, that is
|
---|
6914 | used to describe the block. It is filled in with the information about
|
---|
6915 | the current block. The incoming structure's branch path, if any, is used
|
---|
6916 | to construct the output branch path. */
|
---|
6917 |
|
---|
6918 | void
|
---|
6919 | cse_end_of_basic_block (insn, data, follow_jumps, after_loop, skip_blocks)
|
---|
6920 | rtx insn;
|
---|
6921 | struct cse_basic_block_data *data;
|
---|
6922 | int follow_jumps;
|
---|
6923 | int after_loop;
|
---|
6924 | int skip_blocks;
|
---|
6925 | {
|
---|
6926 | rtx p = insn, q;
|
---|
6927 | int nsets = 0;
|
---|
6928 | int low_cuid = INSN_CUID (insn), high_cuid = INSN_CUID (insn);
|
---|
6929 | rtx next = INSN_P (insn) ? insn : next_real_insn (insn);
|
---|
6930 | int path_size = data->path_size;
|
---|
6931 | int path_entry = 0;
|
---|
6932 | int i;
|
---|
6933 |
|
---|
6934 | /* Update the previous branch path, if any. If the last branch was
|
---|
6935 | previously TAKEN, mark it NOT_TAKEN. If it was previously NOT_TAKEN,
|
---|
6936 | shorten the path by one and look at the previous branch. We know that
|
---|
6937 | at least one branch must have been taken if PATH_SIZE is nonzero. */
|
---|
6938 | while (path_size > 0)
|
---|
6939 | {
|
---|
6940 | if (data->path[path_size - 1].status != NOT_TAKEN)
|
---|
6941 | {
|
---|
6942 | data->path[path_size - 1].status = NOT_TAKEN;
|
---|
6943 | break;
|
---|
6944 | }
|
---|
6945 | else
|
---|
6946 | path_size--;
|
---|
6947 | }
|
---|
6948 |
|
---|
6949 | /* If the first instruction is marked with QImode, that means we've
|
---|
6950 | already processed this block. Our caller will look at DATA->LAST
|
---|
6951 | to figure out where to go next. We want to return the next block
|
---|
6952 | in the instruction stream, not some branched-to block somewhere
|
---|
6953 | else. We accomplish this by pretending our called forbid us to
|
---|
6954 | follow jumps, or skip blocks. */
|
---|
6955 | if (GET_MODE (insn) == QImode)
|
---|
6956 | follow_jumps = skip_blocks = 0;
|
---|
6957 |
|
---|
6958 | /* Scan to end of this basic block. */
|
---|
6959 | while (p && GET_CODE (p) != CODE_LABEL)
|
---|
6960 | {
|
---|
6961 | /* Don't cse out the end of a loop. This makes a difference
|
---|
6962 | only for the unusual loops that always execute at least once;
|
---|
6963 | all other loops have labels there so we will stop in any case.
|
---|
6964 | Cse'ing out the end of the loop is dangerous because it
|
---|
6965 | might cause an invariant expression inside the loop
|
---|
6966 | to be reused after the end of the loop. This would make it
|
---|
6967 | hard to move the expression out of the loop in loop.c,
|
---|
6968 | especially if it is one of several equivalent expressions
|
---|
6969 | and loop.c would like to eliminate it.
|
---|
6970 |
|
---|
6971 | If we are running after loop.c has finished, we can ignore
|
---|
6972 | the NOTE_INSN_LOOP_END. */
|
---|
6973 |
|
---|
6974 | if (! after_loop && GET_CODE (p) == NOTE
|
---|
6975 | && NOTE_LINE_NUMBER (p) == NOTE_INSN_LOOP_END)
|
---|
6976 | break;
|
---|
6977 |
|
---|
6978 | /* Don't cse over a call to setjmp; on some machines (eg VAX)
|
---|
6979 | the regs restored by the longjmp come from
|
---|
6980 | a later time than the setjmp. */
|
---|
6981 | if (PREV_INSN (p) && GET_CODE (PREV_INSN (p)) == CALL_INSN
|
---|
6982 | && find_reg_note (PREV_INSN (p), REG_SETJMP, NULL))
|
---|
6983 | break;
|
---|
6984 |
|
---|
6985 | /* A PARALLEL can have lots of SETs in it,
|
---|
6986 | especially if it is really an ASM_OPERANDS. */
|
---|
6987 | if (INSN_P (p) && GET_CODE (PATTERN (p)) == PARALLEL)
|
---|
6988 | nsets += XVECLEN (PATTERN (p), 0);
|
---|
6989 | else if (GET_CODE (p) != NOTE)
|
---|
6990 | nsets += 1;
|
---|
6991 |
|
---|
6992 | /* Ignore insns made by CSE; they cannot affect the boundaries of
|
---|
6993 | the basic block. */
|
---|
6994 |
|
---|
6995 | if (INSN_UID (p) <= max_uid && INSN_CUID (p) > high_cuid)
|
---|
6996 | high_cuid = INSN_CUID (p);
|
---|
6997 | if (INSN_UID (p) <= max_uid && INSN_CUID (p) < low_cuid)
|
---|
6998 | low_cuid = INSN_CUID (p);
|
---|
6999 |
|
---|
7000 | /* See if this insn is in our branch path. If it is and we are to
|
---|
7001 | take it, do so. */
|
---|
7002 | if (path_entry < path_size && data->path[path_entry].branch == p)
|
---|
7003 | {
|
---|
7004 | if (data->path[path_entry].status != NOT_TAKEN)
|
---|
7005 | p = JUMP_LABEL (p);
|
---|
7006 |
|
---|
7007 | /* Point to next entry in path, if any. */
|
---|
7008 | path_entry++;
|
---|
7009 | }
|
---|
7010 |
|
---|
7011 | /* If this is a conditional jump, we can follow it if -fcse-follow-jumps
|
---|
7012 | was specified, we haven't reached our maximum path length, there are
|
---|
7013 | insns following the target of the jump, this is the only use of the
|
---|
7014 | jump label, and the target label is preceded by a BARRIER.
|
---|
7015 |
|
---|
7016 | Alternatively, we can follow the jump if it branches around a
|
---|
7017 | block of code and there are no other branches into the block.
|
---|
7018 | In this case invalidate_skipped_block will be called to invalidate any
|
---|
7019 | registers set in the block when following the jump. */
|
---|
7020 |
|
---|
7021 | else if ((follow_jumps || skip_blocks) && path_size < PATHLENGTH - 1
|
---|
7022 | && GET_CODE (p) == JUMP_INSN
|
---|
7023 | && GET_CODE (PATTERN (p)) == SET
|
---|
7024 | && GET_CODE (SET_SRC (PATTERN (p))) == IF_THEN_ELSE
|
---|
7025 | && JUMP_LABEL (p) != 0
|
---|
7026 | && LABEL_NUSES (JUMP_LABEL (p)) == 1
|
---|
7027 | && NEXT_INSN (JUMP_LABEL (p)) != 0)
|
---|
7028 | {
|
---|
7029 | for (q = PREV_INSN (JUMP_LABEL (p)); q; q = PREV_INSN (q))
|
---|
7030 | if ((GET_CODE (q) != NOTE
|
---|
7031 | || NOTE_LINE_NUMBER (q) == NOTE_INSN_LOOP_END
|
---|
7032 | || (PREV_INSN (q) && GET_CODE (PREV_INSN (q)) == CALL_INSN
|
---|
7033 | && find_reg_note (PREV_INSN (q), REG_SETJMP, NULL)))
|
---|
7034 | && (GET_CODE (q) != CODE_LABEL || LABEL_NUSES (q) != 0))
|
---|
7035 | break;
|
---|
7036 |
|
---|
7037 | /* If we ran into a BARRIER, this code is an extension of the
|
---|
7038 | basic block when the branch is taken. */
|
---|
7039 | if (follow_jumps && q != 0 && GET_CODE (q) == BARRIER)
|
---|
7040 | {
|
---|
7041 | /* Don't allow ourself to keep walking around an
|
---|
7042 | always-executed loop. */
|
---|
7043 | if (next_real_insn (q) == next)
|
---|
7044 | {
|
---|
7045 | p = NEXT_INSN (p);
|
---|
7046 | continue;
|
---|
7047 | }
|
---|
7048 |
|
---|
7049 | /* Similarly, don't put a branch in our path more than once. */
|
---|
7050 | for (i = 0; i < path_entry; i++)
|
---|
7051 | if (data->path[i].branch == p)
|
---|
7052 | break;
|
---|
7053 |
|
---|
7054 | if (i != path_entry)
|
---|
7055 | break;
|
---|
7056 |
|
---|
7057 | data->path[path_entry].branch = p;
|
---|
7058 | data->path[path_entry++].status = TAKEN;
|
---|
7059 |
|
---|
7060 | /* This branch now ends our path. It was possible that we
|
---|
7061 | didn't see this branch the last time around (when the
|
---|
7062 | insn in front of the target was a JUMP_INSN that was
|
---|
7063 | turned into a no-op). */
|
---|
7064 | path_size = path_entry;
|
---|
7065 |
|
---|
7066 | p = JUMP_LABEL (p);
|
---|
7067 | /* Mark block so we won't scan it again later. */
|
---|
7068 | PUT_MODE (NEXT_INSN (p), QImode);
|
---|
7069 | }
|
---|
7070 | /* Detect a branch around a block of code. */
|
---|
7071 | else if (skip_blocks && q != 0 && GET_CODE (q) != CODE_LABEL)
|
---|
7072 | {
|
---|
7073 | rtx tmp;
|
---|
7074 |
|
---|
7075 | if (next_real_insn (q) == next)
|
---|
7076 | {
|
---|
7077 | p = NEXT_INSN (p);
|
---|
7078 | continue;
|
---|
7079 | }
|
---|
7080 |
|
---|
7081 | for (i = 0; i < path_entry; i++)
|
---|
7082 | if (data->path[i].branch == p)
|
---|
7083 | break;
|
---|
7084 |
|
---|
7085 | if (i != path_entry)
|
---|
7086 | break;
|
---|
7087 |
|
---|
7088 | /* This is no_labels_between_p (p, q) with an added check for
|
---|
7089 | reaching the end of a function (in case Q precedes P). */
|
---|
7090 | for (tmp = NEXT_INSN (p); tmp && tmp != q; tmp = NEXT_INSN (tmp))
|
---|
7091 | if (GET_CODE (tmp) == CODE_LABEL)
|
---|
7092 | break;
|
---|
7093 |
|
---|
7094 | if (tmp == q)
|
---|
7095 | {
|
---|
7096 | data->path[path_entry].branch = p;
|
---|
7097 | data->path[path_entry++].status = AROUND;
|
---|
7098 |
|
---|
7099 | path_size = path_entry;
|
---|
7100 |
|
---|
7101 | p = JUMP_LABEL (p);
|
---|
7102 | /* Mark block so we won't scan it again later. */
|
---|
7103 | PUT_MODE (NEXT_INSN (p), QImode);
|
---|
7104 | }
|
---|
7105 | }
|
---|
7106 | }
|
---|
7107 | p = NEXT_INSN (p);
|
---|
7108 | }
|
---|
7109 |
|
---|
7110 | data->low_cuid = low_cuid;
|
---|
7111 | data->high_cuid = high_cuid;
|
---|
7112 | data->nsets = nsets;
|
---|
7113 | data->last = p;
|
---|
7114 |
|
---|
7115 | /* If all jumps in the path are not taken, set our path length to zero
|
---|
7116 | so a rescan won't be done. */
|
---|
7117 | for (i = path_size - 1; i >= 0; i--)
|
---|
7118 | if (data->path[i].status != NOT_TAKEN)
|
---|
7119 | break;
|
---|
7120 |
|
---|
7121 | if (i == -1)
|
---|
7122 | data->path_size = 0;
|
---|
7123 | else
|
---|
7124 | data->path_size = path_size;
|
---|
7125 |
|
---|
7126 | /* End the current branch path. */
|
---|
7127 | data->path[path_size].branch = 0;
|
---|
7128 | }
|
---|
7129 | |
---|
7130 |
|
---|
7131 | /* Perform cse on the instructions of a function.
|
---|
7132 | F is the first instruction.
|
---|
7133 | NREGS is one plus the highest pseudo-reg number used in the instruction.
|
---|
7134 |
|
---|
7135 | AFTER_LOOP is 1 if this is the cse call done after loop optimization
|
---|
7136 | (only if -frerun-cse-after-loop).
|
---|
7137 |
|
---|
7138 | Returns 1 if jump_optimize should be redone due to simplifications
|
---|
7139 | in conditional jump instructions. */
|
---|
7140 |
|
---|
7141 | int
|
---|
7142 | cse_main (f, nregs, after_loop, file)
|
---|
7143 | rtx f;
|
---|
7144 | int nregs;
|
---|
7145 | int after_loop;
|
---|
7146 | FILE *file;
|
---|
7147 | {
|
---|
7148 | struct cse_basic_block_data val;
|
---|
7149 | rtx insn = f;
|
---|
7150 | int i;
|
---|
7151 |
|
---|
7152 | cse_jumps_altered = 0;
|
---|
7153 | recorded_label_ref = 0;
|
---|
7154 | constant_pool_entries_cost = 0;
|
---|
7155 | val.path_size = 0;
|
---|
7156 |
|
---|
7157 | init_recog ();
|
---|
7158 | init_alias_analysis ();
|
---|
7159 |
|
---|
7160 | max_reg = nregs;
|
---|
7161 |
|
---|
7162 | max_insn_uid = get_max_uid ();
|
---|
7163 |
|
---|
7164 | reg_eqv_table = (struct reg_eqv_elem *)
|
---|
7165 | xmalloc (nregs * sizeof (struct reg_eqv_elem));
|
---|
7166 |
|
---|
7167 | #ifdef LOAD_EXTEND_OP
|
---|
7168 |
|
---|
7169 | /* Allocate scratch rtl here. cse_insn will fill in the memory reference
|
---|
7170 | and change the code and mode as appropriate. */
|
---|
7171 | memory_extend_rtx = gen_rtx_ZERO_EXTEND (VOIDmode, NULL_RTX);
|
---|
7172 | #endif
|
---|
7173 |
|
---|
7174 | /* Reset the counter indicating how many elements have been made
|
---|
7175 | thus far. */
|
---|
7176 | n_elements_made = 0;
|
---|
7177 |
|
---|
7178 | /* Find the largest uid. */
|
---|
7179 |
|
---|
7180 | max_uid = get_max_uid ();
|
---|
7181 | uid_cuid = (int *) xcalloc (max_uid + 1, sizeof (int));
|
---|
7182 |
|
---|
7183 | /* Compute the mapping from uids to cuids.
|
---|
7184 | CUIDs are numbers assigned to insns, like uids,
|
---|
7185 | except that cuids increase monotonically through the code.
|
---|
7186 | Don't assign cuids to line-number NOTEs, so that the distance in cuids
|
---|
7187 | between two insns is not affected by -g. */
|
---|
7188 |
|
---|
7189 | for (insn = f, i = 0; insn; insn = NEXT_INSN (insn))
|
---|
7190 | {
|
---|
7191 | if (GET_CODE (insn) != NOTE
|
---|
7192 | || NOTE_LINE_NUMBER (insn) < 0)
|
---|
7193 | INSN_CUID (insn) = ++i;
|
---|
7194 | else
|
---|
7195 | /* Give a line number note the same cuid as preceding insn. */
|
---|
7196 | INSN_CUID (insn) = i;
|
---|
7197 | }
|
---|
7198 |
|
---|
7199 | ggc_push_context ();
|
---|
7200 |
|
---|
7201 | /* Loop over basic blocks.
|
---|
7202 | Compute the maximum number of qty's needed for each basic block
|
---|
7203 | (which is 2 for each SET). */
|
---|
7204 | insn = f;
|
---|
7205 | while (insn)
|
---|
7206 | {
|
---|
7207 | cse_altered = 0;
|
---|
7208 | cse_end_of_basic_block (insn, &val, flag_cse_follow_jumps, after_loop,
|
---|
7209 | flag_cse_skip_blocks);
|
---|
7210 |
|
---|
7211 | /* If this basic block was already processed or has no sets, skip it. */
|
---|
7212 | if (val.nsets == 0 || GET_MODE (insn) == QImode)
|
---|
7213 | {
|
---|
7214 | PUT_MODE (insn, VOIDmode);
|
---|
7215 | insn = (val.last ? NEXT_INSN (val.last) : 0);
|
---|
7216 | val.path_size = 0;
|
---|
7217 | continue;
|
---|
7218 | }
|
---|
7219 |
|
---|
7220 | cse_basic_block_start = val.low_cuid;
|
---|
7221 | cse_basic_block_end = val.high_cuid;
|
---|
7222 | max_qty = val.nsets * 2;
|
---|
7223 |
|
---|
7224 | if (file)
|
---|
7225 | fnotice (file, ";; Processing block from %d to %d, %d sets.\n",
|
---|
7226 | INSN_UID (insn), val.last ? INSN_UID (val.last) : 0,
|
---|
7227 | val.nsets);
|
---|
7228 |
|
---|
7229 | /* Make MAX_QTY bigger to give us room to optimize
|
---|
7230 | past the end of this basic block, if that should prove useful. */
|
---|
7231 | if (max_qty < 500)
|
---|
7232 | max_qty = 500;
|
---|
7233 |
|
---|
7234 | max_qty += max_reg;
|
---|
7235 |
|
---|
7236 | /* If this basic block is being extended by following certain jumps,
|
---|
7237 | (see `cse_end_of_basic_block'), we reprocess the code from the start.
|
---|
7238 | Otherwise, we start after this basic block. */
|
---|
7239 | if (val.path_size > 0)
|
---|
7240 | cse_basic_block (insn, val.last, val.path, 0);
|
---|
7241 | else
|
---|
7242 | {
|
---|
7243 | int old_cse_jumps_altered = cse_jumps_altered;
|
---|
7244 | rtx temp;
|
---|
7245 |
|
---|
7246 | /* When cse changes a conditional jump to an unconditional
|
---|
7247 | jump, we want to reprocess the block, since it will give
|
---|
7248 | us a new branch path to investigate. */
|
---|
7249 | cse_jumps_altered = 0;
|
---|
7250 | temp = cse_basic_block (insn, val.last, val.path, ! after_loop);
|
---|
7251 | if (cse_jumps_altered == 0
|
---|
7252 | || (flag_cse_follow_jumps == 0 && flag_cse_skip_blocks == 0))
|
---|
7253 | insn = temp;
|
---|
7254 |
|
---|
7255 | cse_jumps_altered |= old_cse_jumps_altered;
|
---|
7256 | }
|
---|
7257 |
|
---|
7258 | if (cse_altered)
|
---|
7259 | ggc_collect ();
|
---|
7260 |
|
---|
7261 | #ifdef USE_C_ALLOCA
|
---|
7262 | alloca (0);
|
---|
7263 | #endif
|
---|
7264 | }
|
---|
7265 |
|
---|
7266 | ggc_pop_context ();
|
---|
7267 |
|
---|
7268 | if (max_elements_made < n_elements_made)
|
---|
7269 | max_elements_made = n_elements_made;
|
---|
7270 |
|
---|
7271 | /* Clean up. */
|
---|
7272 | end_alias_analysis ();
|
---|
7273 | free (uid_cuid);
|
---|
7274 | free (reg_eqv_table);
|
---|
7275 |
|
---|
7276 | return cse_jumps_altered || recorded_label_ref;
|
---|
7277 | }
|
---|
7278 |
|
---|
7279 | /* Process a single basic block. FROM and TO and the limits of the basic
|
---|
7280 | block. NEXT_BRANCH points to the branch path when following jumps or
|
---|
7281 | a null path when not following jumps.
|
---|
7282 |
|
---|
7283 | AROUND_LOOP is nonzero if we are to try to cse around to the start of a
|
---|
7284 | loop. This is true when we are being called for the last time on a
|
---|
7285 | block and this CSE pass is before loop.c. */
|
---|
7286 |
|
---|
7287 | static rtx
|
---|
7288 | cse_basic_block (from, to, next_branch, around_loop)
|
---|
7289 | rtx from, to;
|
---|
7290 | struct branch_path *next_branch;
|
---|
7291 | int around_loop;
|
---|
7292 | {
|
---|
7293 | rtx insn;
|
---|
7294 | int to_usage = 0;
|
---|
7295 | rtx libcall_insn = NULL_RTX;
|
---|
7296 | int num_insns = 0;
|
---|
7297 |
|
---|
7298 | /* This array is undefined before max_reg, so only allocate
|
---|
7299 | the space actually needed and adjust the start. */
|
---|
7300 |
|
---|
7301 | qty_table
|
---|
7302 | = (struct qty_table_elem *) xmalloc ((max_qty - max_reg)
|
---|
7303 | * sizeof (struct qty_table_elem));
|
---|
7304 | qty_table -= max_reg;
|
---|
7305 |
|
---|
7306 | new_basic_block ();
|
---|
7307 |
|
---|
7308 | /* TO might be a label. If so, protect it from being deleted. */
|
---|
7309 | if (to != 0 && GET_CODE (to) == CODE_LABEL)
|
---|
7310 | ++LABEL_NUSES (to);
|
---|
7311 |
|
---|
7312 | for (insn = from; insn != to; insn = NEXT_INSN (insn))
|
---|
7313 | {
|
---|
7314 | enum rtx_code code = GET_CODE (insn);
|
---|
7315 |
|
---|
7316 | /* If we have processed 1,000 insns, flush the hash table to
|
---|
7317 | avoid extreme quadratic behavior. We must not include NOTEs
|
---|
7318 | in the count since there may be more of them when generating
|
---|
7319 | debugging information. If we clear the table at different
|
---|
7320 | times, code generated with -g -O might be different than code
|
---|
7321 | generated with -O but not -g.
|
---|
7322 |
|
---|
7323 | ??? This is a real kludge and needs to be done some other way.
|
---|
7324 | Perhaps for 2.9. */
|
---|
7325 | if (code != NOTE && num_insns++ > 1000)
|
---|
7326 | {
|
---|
7327 | flush_hash_table ();
|
---|
7328 | num_insns = 0;
|
---|
7329 | }
|
---|
7330 |
|
---|
7331 | /* See if this is a branch that is part of the path. If so, and it is
|
---|
7332 | to be taken, do so. */
|
---|
7333 | if (next_branch->branch == insn)
|
---|
7334 | {
|
---|
7335 | enum taken status = next_branch++->status;
|
---|
7336 | if (status != NOT_TAKEN)
|
---|
7337 | {
|
---|
7338 | if (status == TAKEN)
|
---|
7339 | record_jump_equiv (insn, 1);
|
---|
7340 | else
|
---|
7341 | invalidate_skipped_block (NEXT_INSN (insn));
|
---|
7342 |
|
---|
7343 | /* Set the last insn as the jump insn; it doesn't affect cc0.
|
---|
7344 | Then follow this branch. */
|
---|
7345 | #ifdef HAVE_cc0
|
---|
7346 | prev_insn_cc0 = 0;
|
---|
7347 | #endif
|
---|
7348 | prev_insn = insn;
|
---|
7349 | insn = JUMP_LABEL (insn);
|
---|
7350 | continue;
|
---|
7351 | }
|
---|
7352 | }
|
---|
7353 |
|
---|
7354 | if (GET_MODE (insn) == QImode)
|
---|
7355 | PUT_MODE (insn, VOIDmode);
|
---|
7356 |
|
---|
7357 | if (GET_RTX_CLASS (code) == 'i')
|
---|
7358 | {
|
---|
7359 | rtx p;
|
---|
7360 |
|
---|
7361 | /* Process notes first so we have all notes in canonical forms when
|
---|
7362 | looking for duplicate operations. */
|
---|
7363 |
|
---|
7364 | if (REG_NOTES (insn))
|
---|
7365 | REG_NOTES (insn) = cse_process_notes (REG_NOTES (insn), NULL_RTX);
|
---|
7366 |
|
---|
7367 | /* Track when we are inside in LIBCALL block. Inside such a block,
|
---|
7368 | we do not want to record destinations. The last insn of a
|
---|
7369 | LIBCALL block is not considered to be part of the block, since
|
---|
7370 | its destination is the result of the block and hence should be
|
---|
7371 | recorded. */
|
---|
7372 |
|
---|
7373 | if (REG_NOTES (insn) != 0)
|
---|
7374 | {
|
---|
7375 | if ((p = find_reg_note (insn, REG_LIBCALL, NULL_RTX)))
|
---|
7376 | libcall_insn = XEXP (p, 0);
|
---|
7377 | else if (find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
---|
7378 | libcall_insn = 0;
|
---|
7379 | }
|
---|
7380 |
|
---|
7381 | cse_insn (insn, libcall_insn);
|
---|
7382 |
|
---|
7383 | /* If we haven't already found an insn where we added a LABEL_REF,
|
---|
7384 | check this one. */
|
---|
7385 | if (GET_CODE (insn) == INSN && ! recorded_label_ref
|
---|
7386 | && for_each_rtx (&PATTERN (insn), check_for_label_ref,
|
---|
7387 | (void *) insn))
|
---|
7388 | recorded_label_ref = 1;
|
---|
7389 | }
|
---|
7390 |
|
---|
7391 | /* If INSN is now an unconditional jump, skip to the end of our
|
---|
7392 | basic block by pretending that we just did the last insn in the
|
---|
7393 | basic block. If we are jumping to the end of our block, show
|
---|
7394 | that we can have one usage of TO. */
|
---|
7395 |
|
---|
7396 | if (any_uncondjump_p (insn))
|
---|
7397 | {
|
---|
7398 | if (to == 0)
|
---|
7399 | {
|
---|
7400 | free (qty_table + max_reg);
|
---|
7401 | return 0;
|
---|
7402 | }
|
---|
7403 |
|
---|
7404 | if (JUMP_LABEL (insn) == to)
|
---|
7405 | to_usage = 1;
|
---|
7406 |
|
---|
7407 | /* Maybe TO was deleted because the jump is unconditional.
|
---|
7408 | If so, there is nothing left in this basic block. */
|
---|
7409 | /* ??? Perhaps it would be smarter to set TO
|
---|
7410 | to whatever follows this insn,
|
---|
7411 | and pretend the basic block had always ended here. */
|
---|
7412 | if (INSN_DELETED_P (to))
|
---|
7413 | break;
|
---|
7414 |
|
---|
7415 | insn = PREV_INSN (to);
|
---|
7416 | }
|
---|
7417 |
|
---|
7418 | /* See if it is ok to keep on going past the label
|
---|
7419 | which used to end our basic block. Remember that we incremented
|
---|
7420 | the count of that label, so we decrement it here. If we made
|
---|
7421 | a jump unconditional, TO_USAGE will be one; in that case, we don't
|
---|
7422 | want to count the use in that jump. */
|
---|
7423 |
|
---|
7424 | if (to != 0 && NEXT_INSN (insn) == to
|
---|
7425 | && GET_CODE (to) == CODE_LABEL && --LABEL_NUSES (to) == to_usage)
|
---|
7426 | {
|
---|
7427 | struct cse_basic_block_data val;
|
---|
7428 | rtx prev;
|
---|
7429 |
|
---|
7430 | insn = NEXT_INSN (to);
|
---|
7431 |
|
---|
7432 | /* If TO was the last insn in the function, we are done. */
|
---|
7433 | if (insn == 0)
|
---|
7434 | {
|
---|
7435 | free (qty_table + max_reg);
|
---|
7436 | return 0;
|
---|
7437 | }
|
---|
7438 |
|
---|
7439 | /* If TO was preceded by a BARRIER we are done with this block
|
---|
7440 | because it has no continuation. */
|
---|
7441 | prev = prev_nonnote_insn (to);
|
---|
7442 | if (prev && GET_CODE (prev) == BARRIER)
|
---|
7443 | {
|
---|
7444 | free (qty_table + max_reg);
|
---|
7445 | return insn;
|
---|
7446 | }
|
---|
7447 |
|
---|
7448 | /* Find the end of the following block. Note that we won't be
|
---|
7449 | following branches in this case. */
|
---|
7450 | to_usage = 0;
|
---|
7451 | val.path_size = 0;
|
---|
7452 | cse_end_of_basic_block (insn, &val, 0, 0, 0);
|
---|
7453 |
|
---|
7454 | /* If the tables we allocated have enough space left
|
---|
7455 | to handle all the SETs in the next basic block,
|
---|
7456 | continue through it. Otherwise, return,
|
---|
7457 | and that block will be scanned individually. */
|
---|
7458 | if (val.nsets * 2 + next_qty > max_qty)
|
---|
7459 | break;
|
---|
7460 |
|
---|
7461 | cse_basic_block_start = val.low_cuid;
|
---|
7462 | cse_basic_block_end = val.high_cuid;
|
---|
7463 | to = val.last;
|
---|
7464 |
|
---|
7465 | /* Prevent TO from being deleted if it is a label. */
|
---|
7466 | if (to != 0 && GET_CODE (to) == CODE_LABEL)
|
---|
7467 | ++LABEL_NUSES (to);
|
---|
7468 |
|
---|
7469 | /* Back up so we process the first insn in the extension. */
|
---|
7470 | insn = PREV_INSN (insn);
|
---|
7471 | }
|
---|
7472 | }
|
---|
7473 |
|
---|
7474 | if (next_qty > max_qty)
|
---|
7475 | abort ();
|
---|
7476 |
|
---|
7477 | /* If we are running before loop.c, we stopped on a NOTE_INSN_LOOP_END, and
|
---|
7478 | the previous insn is the only insn that branches to the head of a loop,
|
---|
7479 | we can cse into the loop. Don't do this if we changed the jump
|
---|
7480 | structure of a loop unless we aren't going to be following jumps. */
|
---|
7481 |
|
---|
7482 | insn = prev_nonnote_insn (to);
|
---|
7483 | if ((cse_jumps_altered == 0
|
---|
7484 | || (flag_cse_follow_jumps == 0 && flag_cse_skip_blocks == 0))
|
---|
7485 | && around_loop && to != 0
|
---|
7486 | && GET_CODE (to) == NOTE && NOTE_LINE_NUMBER (to) == NOTE_INSN_LOOP_END
|
---|
7487 | && GET_CODE (insn) == JUMP_INSN
|
---|
7488 | && JUMP_LABEL (insn) != 0
|
---|
7489 | && LABEL_NUSES (JUMP_LABEL (insn)) == 1)
|
---|
7490 | cse_around_loop (JUMP_LABEL (insn));
|
---|
7491 |
|
---|
7492 | free (qty_table + max_reg);
|
---|
7493 |
|
---|
7494 | return to ? NEXT_INSN (to) : 0;
|
---|
7495 | }
|
---|
7496 | |
---|
7497 |
|
---|
7498 | /* Called via for_each_rtx to see if an insn is using a LABEL_REF for which
|
---|
7499 | there isn't a REG_LABEL note. Return one if so. DATA is the insn. */
|
---|
7500 |
|
---|
7501 | static int
|
---|
7502 | check_for_label_ref (rtl, data)
|
---|
7503 | rtx *rtl;
|
---|
7504 | void *data;
|
---|
7505 | {
|
---|
7506 | rtx insn = (rtx) data;
|
---|
7507 |
|
---|
7508 | /* If this insn uses a LABEL_REF and there isn't a REG_LABEL note for it,
|
---|
7509 | we must rerun jump since it needs to place the note. If this is a
|
---|
7510 | LABEL_REF for a CODE_LABEL that isn't in the insn chain, don't do this
|
---|
7511 | since no REG_LABEL will be added. */
|
---|
7512 | return (GET_CODE (*rtl) == LABEL_REF
|
---|
7513 | && ! LABEL_REF_NONLOCAL_P (*rtl)
|
---|
7514 | && LABEL_P (XEXP (*rtl, 0))
|
---|
7515 | && INSN_UID (XEXP (*rtl, 0)) != 0
|
---|
7516 | && ! find_reg_note (insn, REG_LABEL, XEXP (*rtl, 0)));
|
---|
7517 | }
|
---|
7518 | |
---|
7519 |
|
---|
7520 | /* Count the number of times registers are used (not set) in X.
|
---|
7521 | COUNTS is an array in which we accumulate the count, INCR is how much
|
---|
7522 | we count each register usage.
|
---|
7523 |
|
---|
7524 | Don't count a usage of DEST, which is the SET_DEST of a SET which
|
---|
7525 | contains X in its SET_SRC. This is because such a SET does not
|
---|
7526 | modify the liveness of DEST. */
|
---|
7527 |
|
---|
7528 | static void
|
---|
7529 | count_reg_usage (x, counts, dest, incr)
|
---|
7530 | rtx x;
|
---|
7531 | int *counts;
|
---|
7532 | rtx dest;
|
---|
7533 | int incr;
|
---|
7534 | {
|
---|
7535 | enum rtx_code code;
|
---|
7536 | rtx note;
|
---|
7537 | const char *fmt;
|
---|
7538 | int i, j;
|
---|
7539 |
|
---|
7540 | if (x == 0)
|
---|
7541 | return;
|
---|
7542 |
|
---|
7543 | switch (code = GET_CODE (x))
|
---|
7544 | {
|
---|
7545 | case REG:
|
---|
7546 | if (x != dest)
|
---|
7547 | counts[REGNO (x)] += incr;
|
---|
7548 | return;
|
---|
7549 |
|
---|
7550 | case PC:
|
---|
7551 | case CC0:
|
---|
7552 | case CONST:
|
---|
7553 | case CONST_INT:
|
---|
7554 | case CONST_DOUBLE:
|
---|
7555 | case CONST_VECTOR:
|
---|
7556 | case SYMBOL_REF:
|
---|
7557 | case LABEL_REF:
|
---|
7558 | return;
|
---|
7559 |
|
---|
7560 | case CLOBBER:
|
---|
7561 | /* If we are clobbering a MEM, mark any registers inside the address
|
---|
7562 | as being used. */
|
---|
7563 | if (GET_CODE (XEXP (x, 0)) == MEM)
|
---|
7564 | count_reg_usage (XEXP (XEXP (x, 0), 0), counts, NULL_RTX, incr);
|
---|
7565 | return;
|
---|
7566 |
|
---|
7567 | case SET:
|
---|
7568 | /* Unless we are setting a REG, count everything in SET_DEST. */
|
---|
7569 | if (GET_CODE (SET_DEST (x)) != REG)
|
---|
7570 | count_reg_usage (SET_DEST (x), counts, NULL_RTX, incr);
|
---|
7571 |
|
---|
7572 | /* If SRC has side-effects, then we can't delete this insn, so the
|
---|
7573 | usage of SET_DEST inside SRC counts.
|
---|
7574 |
|
---|
7575 | ??? Strictly-speaking, we might be preserving this insn
|
---|
7576 | because some other SET has side-effects, but that's hard
|
---|
7577 | to do and can't happen now. */
|
---|
7578 | count_reg_usage (SET_SRC (x), counts,
|
---|
7579 | side_effects_p (SET_SRC (x)) ? NULL_RTX : SET_DEST (x),
|
---|
7580 | incr);
|
---|
7581 | return;
|
---|
7582 |
|
---|
7583 | case CALL_INSN:
|
---|
7584 | count_reg_usage (CALL_INSN_FUNCTION_USAGE (x), counts, NULL_RTX, incr);
|
---|
7585 | /* Fall through. */
|
---|
7586 |
|
---|
7587 | case INSN:
|
---|
7588 | case JUMP_INSN:
|
---|
7589 | count_reg_usage (PATTERN (x), counts, NULL_RTX, incr);
|
---|
7590 |
|
---|
7591 | /* Things used in a REG_EQUAL note aren't dead since loop may try to
|
---|
7592 | use them. */
|
---|
7593 |
|
---|
7594 | note = find_reg_equal_equiv_note (x);
|
---|
7595 | if (note)
|
---|
7596 | count_reg_usage (XEXP (note, 0), counts, NULL_RTX, incr);
|
---|
7597 | return;
|
---|
7598 |
|
---|
7599 | case INSN_LIST:
|
---|
7600 | abort ();
|
---|
7601 |
|
---|
7602 | default:
|
---|
7603 | break;
|
---|
7604 | }
|
---|
7605 |
|
---|
7606 | fmt = GET_RTX_FORMAT (code);
|
---|
7607 | for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
---|
7608 | {
|
---|
7609 | if (fmt[i] == 'e')
|
---|
7610 | count_reg_usage (XEXP (x, i), counts, dest, incr);
|
---|
7611 | else if (fmt[i] == 'E')
|
---|
7612 | for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
---|
7613 | count_reg_usage (XVECEXP (x, i, j), counts, dest, incr);
|
---|
7614 | }
|
---|
7615 | }
|
---|
7616 | |
---|
7617 |
|
---|
7618 | /* Return true if set is live. */
|
---|
7619 | static bool
|
---|
7620 | set_live_p (set, insn, counts)
|
---|
7621 | rtx set;
|
---|
7622 | rtx insn ATTRIBUTE_UNUSED; /* Only used with HAVE_cc0. */
|
---|
7623 | int *counts;
|
---|
7624 | {
|
---|
7625 | #ifdef HAVE_cc0
|
---|
7626 | rtx tem;
|
---|
7627 | #endif
|
---|
7628 |
|
---|
7629 | if (set_noop_p (set))
|
---|
7630 | ;
|
---|
7631 |
|
---|
7632 | #ifdef HAVE_cc0
|
---|
7633 | else if (GET_CODE (SET_DEST (set)) == CC0
|
---|
7634 | && !side_effects_p (SET_SRC (set))
|
---|
7635 | && ((tem = next_nonnote_insn (insn)) == 0
|
---|
7636 | || !INSN_P (tem)
|
---|
7637 | || !reg_referenced_p (cc0_rtx, PATTERN (tem))))
|
---|
7638 | return false;
|
---|
7639 | #endif
|
---|
7640 | else if (GET_CODE (SET_DEST (set)) != REG
|
---|
7641 | || REGNO (SET_DEST (set)) < FIRST_PSEUDO_REGISTER
|
---|
7642 | || counts[REGNO (SET_DEST (set))] != 0
|
---|
7643 | || side_effects_p (SET_SRC (set))
|
---|
7644 | /* An ADDRESSOF expression can turn into a use of the
|
---|
7645 | internal arg pointer, so always consider the
|
---|
7646 | internal arg pointer live. If it is truly dead,
|
---|
7647 | flow will delete the initializing insn. */
|
---|
7648 | || (SET_DEST (set) == current_function_internal_arg_pointer))
|
---|
7649 | return true;
|
---|
7650 | return false;
|
---|
7651 | }
|
---|
7652 |
|
---|
7653 | /* Return true if insn is live. */
|
---|
7654 |
|
---|
7655 | static bool
|
---|
7656 | insn_live_p (insn, counts)
|
---|
7657 | rtx insn;
|
---|
7658 | int *counts;
|
---|
7659 | {
|
---|
7660 | int i;
|
---|
7661 | if (flag_non_call_exceptions && may_trap_p (PATTERN (insn)))
|
---|
7662 | return true;
|
---|
7663 | else if (GET_CODE (PATTERN (insn)) == SET)
|
---|
7664 | return set_live_p (PATTERN (insn), insn, counts);
|
---|
7665 | else if (GET_CODE (PATTERN (insn)) == PARALLEL)
|
---|
7666 | {
|
---|
7667 | for (i = XVECLEN (PATTERN (insn), 0) - 1; i >= 0; i--)
|
---|
7668 | {
|
---|
7669 | rtx elt = XVECEXP (PATTERN (insn), 0, i);
|
---|
7670 |
|
---|
7671 | if (GET_CODE (elt) == SET)
|
---|
7672 | {
|
---|
7673 | if (set_live_p (elt, insn, counts))
|
---|
7674 | return true;
|
---|
7675 | }
|
---|
7676 | else if (GET_CODE (elt) != CLOBBER && GET_CODE (elt) != USE)
|
---|
7677 | return true;
|
---|
7678 | }
|
---|
7679 | return false;
|
---|
7680 | }
|
---|
7681 | else
|
---|
7682 | return true;
|
---|
7683 | }
|
---|
7684 |
|
---|
7685 | /* Return true if libcall is dead as a whole. */
|
---|
7686 |
|
---|
7687 | static bool
|
---|
7688 | dead_libcall_p (insn, counts)
|
---|
7689 | rtx insn;
|
---|
7690 | int *counts;
|
---|
7691 | {
|
---|
7692 | rtx note;
|
---|
7693 | /* See if there's a REG_EQUAL note on this insn and try to
|
---|
7694 | replace the source with the REG_EQUAL expression.
|
---|
7695 |
|
---|
7696 | We assume that insns with REG_RETVALs can only be reg->reg
|
---|
7697 | copies at this point. */
|
---|
7698 | note = find_reg_note (insn, REG_EQUAL, NULL_RTX);
|
---|
7699 | if (note)
|
---|
7700 | {
|
---|
7701 | rtx set = single_set (insn);
|
---|
7702 | rtx new = simplify_rtx (XEXP (note, 0));
|
---|
7703 |
|
---|
7704 | if (!new)
|
---|
7705 | new = XEXP (note, 0);
|
---|
7706 |
|
---|
7707 | /* While changing insn, we must update the counts accordingly. */
|
---|
7708 | count_reg_usage (insn, counts, NULL_RTX, -1);
|
---|
7709 |
|
---|
7710 | if (set && validate_change (insn, &SET_SRC (set), new, 0))
|
---|
7711 | {
|
---|
7712 | count_reg_usage (insn, counts, NULL_RTX, 1);
|
---|
7713 | remove_note (insn, find_reg_note (insn, REG_RETVAL, NULL_RTX));
|
---|
7714 | remove_note (insn, note);
|
---|
7715 | return true;
|
---|
7716 | }
|
---|
7717 | count_reg_usage (insn, counts, NULL_RTX, 1);
|
---|
7718 | }
|
---|
7719 | return false;
|
---|
7720 | }
|
---|
7721 |
|
---|
7722 | /* Scan all the insns and delete any that are dead; i.e., they store a register
|
---|
7723 | that is never used or they copy a register to itself.
|
---|
7724 |
|
---|
7725 | This is used to remove insns made obviously dead by cse, loop or other
|
---|
7726 | optimizations. It improves the heuristics in loop since it won't try to
|
---|
7727 | move dead invariants out of loops or make givs for dead quantities. The
|
---|
7728 | remaining passes of the compilation are also sped up. */
|
---|
7729 |
|
---|
7730 | int
|
---|
7731 | delete_trivially_dead_insns (insns, nreg)
|
---|
7732 | rtx insns;
|
---|
7733 | int nreg;
|
---|
7734 | {
|
---|
7735 | int *counts;
|
---|
7736 | rtx insn, prev;
|
---|
7737 | int in_libcall = 0, dead_libcall = 0;
|
---|
7738 | int ndead = 0, nlastdead, niterations = 0;
|
---|
7739 |
|
---|
7740 | timevar_push (TV_DELETE_TRIVIALLY_DEAD);
|
---|
7741 | /* First count the number of times each register is used. */
|
---|
7742 | counts = (int *) xcalloc (nreg, sizeof (int));
|
---|
7743 | for (insn = next_real_insn (insns); insn; insn = next_real_insn (insn))
|
---|
7744 | count_reg_usage (insn, counts, NULL_RTX, 1);
|
---|
7745 |
|
---|
7746 | do
|
---|
7747 | {
|
---|
7748 | nlastdead = ndead;
|
---|
7749 | niterations++;
|
---|
7750 | /* Go from the last insn to the first and delete insns that only set unused
|
---|
7751 | registers or copy a register to itself. As we delete an insn, remove
|
---|
7752 | usage counts for registers it uses.
|
---|
7753 |
|
---|
7754 | The first jump optimization pass may leave a real insn as the last
|
---|
7755 | insn in the function. We must not skip that insn or we may end
|
---|
7756 | up deleting code that is not really dead. */
|
---|
7757 | insn = get_last_insn ();
|
---|
7758 | if (! INSN_P (insn))
|
---|
7759 | insn = prev_real_insn (insn);
|
---|
7760 |
|
---|
7761 | for (; insn; insn = prev)
|
---|
7762 | {
|
---|
7763 | int live_insn = 0;
|
---|
7764 |
|
---|
7765 | prev = prev_real_insn (insn);
|
---|
7766 |
|
---|
7767 | /* Don't delete any insns that are part of a libcall block unless
|
---|
7768 | we can delete the whole libcall block.
|
---|
7769 |
|
---|
7770 | Flow or loop might get confused if we did that. Remember
|
---|
7771 | that we are scanning backwards. */
|
---|
7772 | if (find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
---|
7773 | {
|
---|
7774 | in_libcall = 1;
|
---|
7775 | live_insn = 1;
|
---|
7776 | dead_libcall = dead_libcall_p (insn, counts);
|
---|
7777 | }
|
---|
7778 | else if (in_libcall)
|
---|
7779 | live_insn = ! dead_libcall;
|
---|
7780 | else
|
---|
7781 | live_insn = insn_live_p (insn, counts);
|
---|
7782 |
|
---|
7783 | /* If this is a dead insn, delete it and show registers in it aren't
|
---|
7784 | being used. */
|
---|
7785 |
|
---|
7786 | if (! live_insn)
|
---|
7787 | {
|
---|
7788 | count_reg_usage (insn, counts, NULL_RTX, -1);
|
---|
7789 | delete_insn_and_edges (insn);
|
---|
7790 | ndead++;
|
---|
7791 | }
|
---|
7792 |
|
---|
7793 | if (find_reg_note (insn, REG_LIBCALL, NULL_RTX))
|
---|
7794 | {
|
---|
7795 | in_libcall = 0;
|
---|
7796 | dead_libcall = 0;
|
---|
7797 | }
|
---|
7798 | }
|
---|
7799 | }
|
---|
7800 | while (ndead != nlastdead);
|
---|
7801 |
|
---|
7802 | if (rtl_dump_file && ndead)
|
---|
7803 | fprintf (rtl_dump_file, "Deleted %i trivially dead insns; %i iterations\n",
|
---|
7804 | ndead, niterations);
|
---|
7805 | /* Clean up. */
|
---|
7806 | free (counts);
|
---|
7807 | timevar_pop (TV_DELETE_TRIVIALLY_DEAD);
|
---|
7808 | return ndead;
|
---|
7809 | }
|
---|