| blob | 3b96d86bdb3711aa2923440e1f8dadf5486bfb2c |
1 /* Optimize by combining instructions for GNU compiler.
2 Copyright (C) 1987-2017 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 /* This module is essentially the "combiner" phase of the U. of Arizona
21 Portable Optimizer, but redone to work on our list-structured
22 representation for RTL instead of their string representation.
24 The LOG_LINKS of each insn identify the most recent assignment
25 to each REG used in the insn. It is a list of previous insns,
26 each of which contains a SET for a REG that is used in this insn
27 and not used or set in between. LOG_LINKs never cross basic blocks.
28 They were set up by the preceding pass (lifetime analysis).
30 We try to combine each pair of insns joined by a logical link.
31 We also try to combine triplets of insns A, B and C when C has
32 a link back to B and B has a link back to A. Likewise for a
33 small number of quadruplets of insns A, B, C and D for which
34 there's high likelihood of success.
36 LOG_LINKS does not have links for use of the CC0. They don't
37 need to, because the insn that sets the CC0 is always immediately
38 before the insn that tests it. So we always regard a branch
39 insn as having a logical link to the preceding insn. The same is true
40 for an insn explicitly using CC0.
42 We check (with use_crosses_set_p) to avoid combining in such a way
43 as to move a computation to a place where its value would be different.
45 Combination is done by mathematically substituting the previous
46 insn(s) values for the regs they set into the expressions in
47 the later insns that refer to these regs. If the result is a valid insn
48 for our target machine, according to the machine description,
49 we install it, delete the earlier insns, and update the data flow
50 information (LOG_LINKS and REG_NOTES) for what we did.
52 There are a few exceptions where the dataflow information isn't
53 completely updated (however this is only a local issue since it is
54 regenerated before the next pass that uses it):
56 - reg_live_length is not updated
57 - reg_n_refs is not adjusted in the rare case when a register is
58 no longer required in a computation
59 - there are extremely rare cases (see distribute_notes) when a
60 REG_DEAD note is lost
61 - a LOG_LINKS entry that refers to an insn with multiple SETs may be
62 removed because there is no way to know which register it was
63 linking
65 To simplify substitution, we combine only when the earlier insn(s)
66 consist of only a single assignment. To simplify updating afterward,
67 we never combine when a subroutine call appears in the middle.
69 Since we do not represent assignments to CC0 explicitly except when that
70 is all an insn does, there is no LOG_LINKS entry in an insn that uses
71 the condition code for the insn that set the condition code.
72 Fortunately, these two insns must be consecutive.
73 Therefore, every JUMP_INSN is taken to have an implicit logical link
74 to the preceding insn. This is not quite right, since non-jumps can
75 also use the condition code; but in practice such insns would not
76 combine anyway. */
98 /* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
108 /* Number of attempts to combine instructions in this function. */
112 /* Number of attempts that got as far as substitution in this function. */
116 /* Number of instructions combined with added SETs in this function. */
120 /* Number of instructions combined in this function. */
124 /* Totals over entire compilation. */
128 /* combine_instructions may try to replace the right hand side of the
129 second instruction with the value of an associated REG_EQUAL note
130 before throwing it at try_combine. That is problematic when there
131 is a REG_DEAD note for a register used in the old right hand side
132 and can cause distribute_notes to do wrong things. This is the
133 second instruction if it has been so modified, null otherwise. */
137 /* When I2MOD is nonnull, this is a copy of the old right hand side. */
141 /* When I2MOD is nonnull, this is a copy of the new right hand side. */
144 \f
146 /* Record last point of death of (hard or pseudo) register n. */
149 /* Record last point of modification of (hard or pseudo) register n. */
152 /* The next group of fields allows the recording of the last value assigned
153 to (hard or pseudo) register n. We use this information to see if an
154 operation being processed is redundant given a prior operation performed
155 on the register. For example, an `and' with a constant is redundant if
156 all the zero bits are already known to be turned off.
158 We use an approach similar to that used by cse, but change it in the
159 following ways:
161 (1) We do not want to reinitialize at each label.
162 (2) It is useful, but not critical, to know the actual value assigned
163 to a register. Often just its form is helpful.
165 Therefore, we maintain the following fields:
167 last_set_value the last value assigned
168 last_set_label records the value of label_tick when the
169 register was assigned
170 last_set_table_tick records the value of label_tick when a
171 value using the register is assigned
172 last_set_invalid set to nonzero when it is not valid
173 to use the value of this register in some
174 register's value
176 To understand the usage of these tables, it is important to understand
177 the distinction between the value in last_set_value being valid and
178 the register being validly contained in some other expression in the
179 table.
181 (The next two parameters are out of date).
183 reg_stat[i].last_set_value is valid if it is nonzero, and either
184 reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
186 Register I may validly appear in any expression returned for the value
187 of another register if reg_n_sets[i] is 1. It may also appear in the
188 value for register J if reg_stat[j].last_set_invalid is zero, or
189 reg_stat[i].last_set_label < reg_stat[j].last_set_label.
191 If an expression is found in the table containing a register which may
192 not validly appear in an expression, the register is replaced by
193 something that won't match, (clobber (const_int 0)). */
195 /* Record last value assigned to (hard or pseudo) register n. */
197 rtx last_set_value;
199 /* Record the value of label_tick when an expression involving register n
200 is placed in last_set_value. */
204 /* Record the value of label_tick when the value for register n is placed in
205 last_set_value. */
209 /* These fields are maintained in parallel with last_set_value and are
210 used to store the mode in which the register was last set, the bits
211 that were known to be zero when it was last set, and the number of
212 sign bits copies it was known to have when it was last set. */
218 /* Set nonzero if references to register n in expressions should not be
219 used. last_set_invalid is set nonzero when this register is being
220 assigned to and last_set_table_tick == label_tick. */
224 /* Some registers that are set more than once and used in more than one
225 basic block are nevertheless always set in similar ways. For example,
226 a QImode register may be loaded from memory in two places on a machine
227 where byte loads zero extend.
229 We record in the following fields if a register has some leading bits
230 that are always equal to the sign bit, and what we know about the
231 nonzero bits of a register, specifically which bits are known to be
232 zero.
234 If an entry is zero, it means that we don't know anything special. */
240 /* Record the value of the label_tick when the last truncation
241 happened. The field truncated_to_mode is only valid if
242 truncation_label == label_tick. */
246 /* Record the last truncation seen for this register. If truncation
247 is not a nop to this mode we might be able to save an explicit
248 truncation if we know that value already contains a truncated
249 value. */
252 };
257 /* One plus the highest pseudo for which we track REG_N_SETS.
258 regstat_init_n_sets_and_refs allocates the array for REG_N_SETS just once,
259 but during combine_split_insns new pseudos can be created. As we don't have
260 updated DF information in that case, it is hard to initialize the array
261 after growing. The combiner only cares about REG_N_SETS (regno) == 1,
262 so instead of growing the arrays, just assume all newly created pseudos
263 during combine might be set multiple times. */
267 /* Record the luid of the last insn that invalidated memory
268 (anything that writes memory, and subroutine calls, but not pushes). */
272 /* Record the luid of the last CALL_INSN
273 so we can tell whether a potential combination crosses any calls. */
277 /* When `subst' is called, this is the insn that is being modified
278 (by combining in a previous insn). The PATTERN of this insn
279 is still the old pattern partially modified and it should not be
280 looked at, but this may be used to examine the successors of the insn
281 to judge whether a simplification is valid. */
285 /* This is the lowest LUID that `subst' is currently dealing with.
286 get_last_value will not return a value if the register was set at or
287 after this LUID. If not for this mechanism, we could get confused if
288 I2 or I1 in try_combine were an insn that used the old value of a register
289 to obtain a new value. In that case, we might erroneously get the
290 new value of the register when we wanted the old one. */
294 /* This contains any hard registers that are used in newpat; reg_dead_at_p
295 must consider all these registers to be always live. */
299 /* This is an insn to which a LOG_LINKS entry has been added. If this
300 insn is the earlier than I2 or I3, combine should rescan starting at
301 that location. */
305 /* Basic block in which we are performing combines. */
309 \f
310 /* Length of the currently allocated uid_insn_cost array. */
314 /* The following array records the insn_cost for every insn
315 in the instruction stream. */
319 /* The following array records the LOG_LINKS for every insn in the
320 instruction stream as struct insn_link pointers. */
326 };
332 {
336 }
338 #define INSN_COST(INSN) (uid_insn_cost[insn_uid_check (INSN)])
339 #define LOG_LINKS(INSN) (uid_log_links[insn_uid_check (INSN)])
341 #define FOR_EACH_LOG_LINK(L, INSN) \
342 for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
344 /* Links for LOG_LINKS are allocated from this obstack. */
348 /* Allocate a link. */
352 {
360 }
362 /* Incremented for each basic block. */
366 /* Reset to label_tick for each extended basic block in scanning order. */
370 /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
371 largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
375 /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
376 be safely used. It is zero while computing them and after combine has
377 completed. This former test prevents propagating values based on
378 previously set values, which can be incorrect if a variable is modified
379 in a loop. */
383 \f
384 /* Record one modification to rtl structure
385 to be undone by storing old_contents into *where. */
389 struct undo
390 {
395 };
397 /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
398 num_undo says how many are currently recorded.
400 other_insn is nonzero if we have modified some other insn in the process
401 of working on subst_insn. It must be verified too. */
403 struct undobuf
404 {
408 };
412 /* Number of times the pseudo being substituted for
413 was found and replaced. */
418 scalar_int_mode,
421 scalar_int_mode,
494 \f
496 /* It is not safe to use ordinary gen_lowpart in combine.
497 See comments in gen_lowpart_for_combine. */
498 #undef RTL_HOOKS_GEN_LOWPART
499 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
501 /* Our implementation of gen_lowpart never emits a new pseudo. */
502 #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
503 #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
505 #undef RTL_HOOKS_REG_NONZERO_REG_BITS
506 #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
508 #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
509 #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
511 #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
512 #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
516 \f
517 /* Convenience wrapper for the canonicalize_comparison target hook.
518 Target hooks cannot use enum rtx_code. */
522 {
526 }
528 /* Try to split PATTERN found in INSN. This returns NULL_RTX if
529 PATTERN can not be split. Otherwise, it returns an insn sequence.
530 This is a wrapper around split_insns which ensures that the
531 reg_stat vector is made larger if the splitter creates a new
532 register. */
536 {
545 }
547 /* This is used by find_single_use to locate an rtx in LOC that
548 contains exactly one use of DEST, which is typically either a REG
549 or CC0. It returns a pointer to the innermost rtx expression
550 containing DEST. Appearances of DEST that are being used to
551 totally replace it are not counted. */
555 {
564 {
568 CASE_CONST_ANY:
573 /* If the destination is anything other than CC0, PC, a REG or a SUBREG
574 of a REG that occupies all of the REG, the insn uses DEST if
575 it is mentioned in the destination or the source. Otherwise, we
576 need just check the source. */
596 }
598 /* If it wasn't one of the common cases above, check each expression and
599 vector of this code. Look for a unique usage of DEST. */
603 {
605 {
610 else
616 /* Duplicate usage. */
618 }
620 {
624 {
630 else
637 }
638 }
639 }
642 }
645 /* See if DEST, produced in INSN, is used only a single time in the
646 sequel. If so, return a pointer to the innermost rtx expression in which
647 it is used.
649 If PLOC is nonzero, *PLOC is set to the insn containing the single use.
651 If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
652 care about REG_DEAD notes or LOG_LINKS.
654 Otherwise, we find the single use by finding an insn that has a
655 LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
656 only referenced once in that insn, we know that it must be the first
657 and last insn referencing DEST. */
661 {
662 basic_block bb;
668 {
678 }
688 {
694 {
699 }
700 }
703 }
704 \f
705 /* Substitute NEWVAL, an rtx expression, into INTO, a place in some
706 insn. The substitution can be undone by undo_all. If INTO is already
707 set to NEWVAL, do not record this change. Because computing NEWVAL might
708 also call SUBST, we have to compute it before we put anything into
709 the undo table. */
711 static void
713 {
720 /* We'd like to catch as many invalid transformations here as
721 possible. Unfortunately, there are way too many mode changes
722 that are perfectly valid, so we'd waste too much effort for
723 little gain doing the checks here. Focus on catching invalid
724 transformations involving integer constants. */
727 {
728 /* Sanity check that we're replacing oldval with a CONST_INT
729 that is a valid sign-extension for the original mode. */
733 /* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
734 CONST_INT is not valid, because after the replacement, the
735 original mode would be gone. Unfortunately, we can't tell
736 when do_SUBST is called to replace the operand thereof, so we
737 perform this test on oldval instead, checking whether an
738 invalid replacement took place before we got here. */
743 }
747 else
756 }
758 #define SUBST(INTO, NEWVAL) do_SUBST (&(INTO), (NEWVAL))
760 /* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
761 for the value of a HOST_WIDE_INT value (including CONST_INT) is
762 not safe. */
764 static void
766 {
775 else
784 }
786 #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT (&(INTO), (NEWVAL))
788 /* Similar to SUBST, but just substitute the mode. This is used when
789 changing the mode of a pseudo-register, so that any other
790 references to the entry in the regno_reg_rtx array will change as
791 well. */
793 static void
795 {
804 else
813 }
815 #define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE (&(INTO), (NEWVAL))
817 /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
819 static void
821 {
830 else
839 }
841 #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
842 \f
843 /* Subroutine of try_combine. Determine whether the replacement patterns
844 NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to insn_cost
845 than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
846 that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
847 undobuf.other_insn may also both be NULL_RTX. Return false if the cost
848 of all the instructions can be estimated and the replacements are more
849 expensive than the original sequence. */
851 static bool
854 {
859 /* Lookup the original insn_costs. */
864 {
867 {
871 }
872 else
873 {
877 }
878 }
879 else
880 {
883 }
885 /* If we have split a PARALLEL I2 to I1,I2, we have counted its cost twice;
886 correct that. */
891 /* Calculate the replacement insn_costs. */
900 {
910 }
911 else
912 {
915 }
918 {
927 optimize_this_for_speed_p);
931 {
934 }
935 else
937 }
939 /* Disallow this combination if both new_cost and old_cost are greater than
940 zero, and new_cost is greater than old cost. */
944 {
963 else
965 }
970 /* Update the uid_insn_cost array with the replacement costs. */
974 {
978 }
981 }
984 /* Delete any insns that copy a register to itself. */
986 static void
988 {
990 basic_block bb;
993 {
995 {
998 {
1003 }
1004 }
1005 }
1006 }
1008 \f
1009 /* Return false if we do not want to (or cannot) combine DEF. */
1010 static bool
1012 {
1013 /* Do not consider if it is pre/post modification in MEM. */
1019 /* Do not combine frame pointer adjustments. */
1022 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
1030 }
1032 /* Return false if we do not want to (or cannot) combine USE. */
1033 static bool
1035 {
1036 /* Do not consider the usage of the stack pointer by function call. */
1041 }
1043 /* Fill in log links field for all insns. */
1045 static void
1047 {
1048 basic_block bb;
1055 /* Pass through each block from the end, recording the uses of each
1056 register and establishing log links when def is encountered.
1057 Note that we do not clear next_use array in order to save time,
1058 so we have to test whether the use is in the same basic block as def.
1060 There are a few cases below when we do not consider the definition or
1061 usage -- these are taken from original flow.c did. Don't ask me why it is
1062 done this way; I don't know and if it works, I don't want to know. */
1065 {
1067 {
1071 /* Log links are created only once. */
1075 {
1091 /* flow.c claimed:
1093 We don't build a LOG_LINK for hard registers contained
1094 in ASM_OPERANDs. If these registers get replaced,
1095 we might wind up changing the semantics of the insn,
1096 even if reload can make what appear to be valid
1097 assignments later. */
1102 /* Don't add duplicate links between instructions. */
1111 }
1116 }
1117 }
1120 }
1122 /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1123 true if we found a LOG_LINK that proves that A feeds B. This only works
1124 if there are no instructions between A and B which could have a link
1125 depending on A, since in that case we would not record a link for B.
1126 We also check the implicit dependency created by a cc0 setter/user
1127 pair. */
1129 static bool
1131 {
1139 }
1140 \f
1141 /* Main entry point for combiner. F is the first insn of the function.
1142 NREGS is the first unused pseudo-reg number.
1144 Return nonzero if the combiner has turned an indirect jump
1145 instruction into a direct jump. */
1146 static int
1148 {
1153 basic_block last_bb;
1173 /* Allocate array for insn info. */
1181 /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1182 problems when, for example, we have j <<= 1 in a loop. */
1187 /* Scan all SETs and see if we can deduce anything about what
1188 bits are known to be zero for some registers and how many copies
1189 of the sign bit are known to exist for those registers.
1191 Also set any known values so that we can use it while searching
1192 for what bits are known to be set. */
1195 /* Allow the entry block and the first block to fall into the same EBB.
1196 Conceptually the incoming promotions are assigned to the entry block. */
1201 {
1206 label_tick++;
1214 {
1215 rtx links;
1221 insn);
1228 insn);
1230 /* Record the current insn_cost of this instruction. */
1234 {
1237 }
1238 }
1239 }
1243 /* Now scan all the insns in forward order. */
1251 {
1254 /* Ignore instruction combination in basic blocks that are going to
1255 be removed as unreachable anyway. See PR82386. */
1263 label_tick++;
1273 {
1287 /* See if we know about function return values before this
1288 insn based upon SUBREG flags. */
1291 /* See if we can find hardregs and subreg of pseudos in
1292 narrower modes. This could help turning TRUNCATEs
1293 into SUBREGs. */
1296 /* Try this insn with each insn it links back to. */
1302 {
1305 }
1307 /* Try each sequence of three linked insns ending with this one. */
1311 {
1314 /* If the linked insn has been replaced by a note, then there
1315 is no point in pursuing this chain any further. */
1323 {
1326 }
1327 }
1329 /* Try to combine a jump insn that uses CC0
1330 with a preceding insn that sets CC0, and maybe with its
1331 logical predecessor as well.
1332 This is how we make decrement-and-branch insns.
1333 We need this special code because data flow connections
1334 via CC0 do not get entered in LOG_LINKS. */
1341 {
1352 }
1354 /* Do the same for an insn that explicitly references CC0. */
1361 {
1372 }
1374 /* Finally, see if any of the insns that this insn links to
1375 explicitly references CC0. If so, try this insn, that insn,
1376 and its predecessor if it sets CC0. */
1378 {
1390 }
1392 /* Try combining an insn with two different insns whose results it
1393 uses. */
1403 {
1406 }
1408 /* Try four-instruction combinations. */
1411 {
1415 /* If the linked insn has been replaced by a note, then there
1416 is no point in pursuing this chain any further. */
1421 {
1425 /* I0 -> I1 -> I2 -> I3. */
1431 {
1434 }
1435 /* I0, I1 -> I2, I2 -> I3. */
1442 {
1445 }
1446 }
1449 {
1453 /* I0 -> I2; I1, I2 -> I3. */
1459 {
1462 }
1463 /* I0 -> I1; I1, I2 -> I3. */
1469 {
1472 }
1473 }
1474 }
1476 /* Try this insn with each REG_EQUAL note it links back to. */
1478 {
1484 /* Avoid using a register that may already been marked
1485 dead by an earlier instruction. */
1493 {
1494 /* Temporarily replace the set's source with the
1495 contents of the REG_EQUAL note. The insn will
1496 be deleted or recognized by try_combine. */
1507 last_combined_insn);
1510 {
1513 }
1516 }
1517 }
1522 retry:
1523 ;
1524 }
1525 }
1532 /* Clean up. */
1538 {
1541 {
1544 }
1546 }
1556 /* Make recognizer allow volatile MEMs again. */
1560 }
1562 /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1564 static void
1566 {
1572 }
1573 \f
1574 /* Set up any promoted values for incoming argument registers. */
1576 static void
1578 {
1579 tree arg;
1584 {
1589 /* Only continue if the incoming argument is in a register. */
1593 /* Determine, if possible, whether all call sites of the current
1594 function lie within the current compilation unit. (This does
1595 take into account the exporting of a function via taking its
1596 address, and so forth.) */
1599 /* The mode and signedness of the argument before any promotions happen
1600 (equal to the mode of the pseudo holding it at that stage). */
1604 /* The mode and signedness of the argument after any source language and
1605 TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1609 /* The mode and signedness of the argument as it is actually passed,
1610 see assign_parm_setup_reg in function.c. */
1614 /* The mode of the register in which the argument is being passed. */
1617 /* Eliminate sign extensions in the callee when:
1618 (a) A mode promotion has occurred; */
1621 /* (b) The mode of the register is the same as the mode of
1622 the argument as it is passed; */
1625 /* (c) There's no language level extension; */
1627 ;
1628 /* (c.1) All callers are from the current compilation unit. If that's
1629 the case we don't have to rely on an ABI, we only have to know
1630 what we're generating right now, and we know that we will do the
1631 mode1 to mode2 promotion with the given sign. */
1634 /* (c.2) The combination of the two promotions is useful. This is
1635 true when the signs match, or if the first promotion is unsigned.
1636 In the later case, (sign_extend (zero_extend x)) is the same as
1637 (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
1643 /* Record that the value was promoted from mode1 to mode3,
1644 so that any sign extension at the head of the current
1645 function may be eliminated. */
1649 }
1650 }
1652 /* If MODE has a precision lower than PREC and SRC is a non-negative constant
1653 that would appear negative in MODE, sign-extend SRC for use in nonzero_bits
1654 because some machines (maybe most) will actually do the sign-extension and
1655 this is the conservative approach.
1657 ??? For 2.5, try to tighten up the MD files in this regard instead of this
1658 kludge. */
1660 static rtx
1662 {
1663 scalar_int_mode int_mode;
1672 }
1674 /* Update RSP for pseudo-register X from INSN's REG_EQUAL note (if one exists)
1675 and SET. */
1677 static void
1679 rtx x)
1680 {
1690 {
1694 }
1696 /* Don't call nonzero_bits if it cannot change anything. */
1698 {
1703 }
1705 /* Don't call num_sign_bit_copies if it cannot change anything. */
1707 {
1710 {
1714 }
1717 }
1718 }
1720 /* Called via note_stores. If X is a pseudo that is narrower than
1721 HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1723 If we are setting only a portion of X and we can't figure out what
1724 portion, assume all bits will be used since we don't know what will
1725 be happening.
1727 Similarly, set how many bits of X are known to be copies of the sign bit
1728 at all locations in the function. This is the smallest number implied
1729 by any set of X. */
1731 static void
1733 {
1735 scalar_int_mode mode;
1739 /* If this register is undefined at the start of the file, we can't
1740 say what its contents were. */
1741 && ! REGNO_REG_SET_P
1745 {
1749 {
1753 }
1755 /* If this register is being initialized using itself, and the
1756 register is uninitialized in this basic block, and there are
1757 no LOG_LINKS which set the register, then part of the
1758 register is uninitialized. In that case we can't assume
1759 anything about the number of nonzero bits.
1761 ??? We could do better if we checked this in
1762 reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1763 could avoid making assumptions about the insn which initially
1764 sets the register, while still using the information in other
1765 insns. We would have to be careful to check every insn
1766 involved in the combination. */
1772 {
1779 {
1783 }
1784 }
1786 /* If this is a complex assignment, see if we can convert it into a
1787 simple assignment. */
1790 /* If this is a simple assignment, or we have a paradoxical SUBREG,
1791 set what we know about X. */
1797 else
1798 {
1801 }
1802 }
1803 }
1804 \f
1805 /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1806 optionally insns that were previously combined into I3 or that will be
1807 combined into the merger of INSN and I3. The order is PRED, PRED2,
1808 INSN, SUCC, SUCC2, I3.
1810 Return 0 if the combination is not allowed for any reason.
1812 If the combination is allowed, *PDEST will be set to the single
1813 destination of INSN and *PSRC to the single source, and this function
1814 will return 1. */
1816 static int
1820 {
1825 rtx link;
1830 {
1832 {
1837 }
1842 }
1846 /* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
1847 or a PARALLEL consisting of such a SET and CLOBBERs.
1849 If INSN has CLOBBER parallel parts, ignore them for our processing.
1850 By definition, these happen during the execution of the insn. When it
1851 is merged with another insn, all bets are off. If they are, in fact,
1852 needed and aren't also supplied in I3, they may be added by
1853 recog_for_combine. Otherwise, it won't match.
1855 We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1856 note.
1858 Get the source and destination of INSN. If more than one, can't
1859 combine. */
1865 {
1867 {
1871 {
1872 /* This is important to combine floating point insns
1873 for the SH4 port. */
1875 /* Combining an isolated USE doesn't make sense.
1876 We depend here on combinable_i3pat to reject them. */
1877 /* The code below this loop only verifies that the inputs of
1878 the SET in INSN do not change. We call reg_set_between_p
1879 to verify that the REG in the USE does not change between
1880 I3 and INSN.
1881 If the USE in INSN was for a pseudo register, the matching
1882 insn pattern will likely match any register; combining this
1883 with any other USE would only be safe if we knew that the
1884 used registers have identical values, or if there was
1885 something to tell them apart, e.g. different modes. For
1886 now, we forgo such complicated tests and simply disallow
1887 combining of USES of pseudo registers with any other USE. */
1890 {
1895 do
1896 {
1906 }
1908 }
1911 /* We can ignore CLOBBERs. */
1916 /* Ignore SETs whose result isn't used but not those that
1917 have side-effects. */
1923 /* If we have already found a SET, this is a second one and
1924 so we cannot combine with this insn. */
1932 /* Anything else means we can't combine. */
1934 }
1935 }
1938 /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1939 so don't do anything with it. */
1942 }
1943 else
1949 /* The simplification in expand_field_assignment may call back to
1950 get_last_value, so set safe guard here. */
1956 /* Do not eliminate user-specified register if it is in an
1957 asm input because we may break the register asm usage defined
1958 in GCC manual if allow to do so.
1959 Be aware that this may cover more cases than we expect but this
1960 should be harmless. */
1965 /* Don't eliminate a store in the stack pointer. */
1967 /* Don't combine with an insn that sets a register to itself if it has
1968 a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1970 /* Can't merge an ASM_OPERANDS. */
1972 /* Can't merge a function call. */
1974 /* Don't eliminate a function call argument. */
1980 /* Don't substitute into an incremented register. */
1984 /* Don't substitute into a non-local goto, this confuses CFG. */
1986 /* Make sure that DEST is not used after INSN but before SUCC, or
1987 after SUCC and before SUCC2, or after SUCC2 but before I3. */
1988 || (!all_adjacent
1989 && ((succ2
1993 || (succ
1994 /* SUCC and SUCC2 can be split halves from a PARALLEL; in
1995 that case SUCC is not in the insn stream, so use SUCC2
1996 instead for this test. */
1998 succ2
2001 /* Make sure that the value that is to be substituted for the register
2002 does not use any registers whose values alter in between. However,
2003 If the insns are adjacent, a use can't cross a set even though we
2004 think it might (this can happen for a sequence of insns each setting
2005 the same destination; last_set of that register might point to
2006 a NOTE). If INSN has a REG_EQUIV note, the register is always
2007 equivalent to the memory so the substitution is valid even if there
2008 are intervening stores. Also, don't move a volatile asm or
2009 UNSPEC_VOLATILE across any other insns. */
2010 || (! all_adjacent
2016 /* Don't combine across a CALL_INSN, because that would possibly
2017 change whether the life span of some REGs crosses calls or not,
2018 and it is a pain to update that information.
2019 Exception: if source is a constant, moving it later can't hurt.
2020 Accept that as a special case. */
2024 /* DEST must either be a REG or CC0. */
2026 {
2027 /* If register alignment is being enforced for multi-word items in all
2028 cases except for parameters, it is possible to have a register copy
2029 insn referencing a hard register that is not allowed to contain the
2030 mode being copied and which would not be valid as an operand of most
2031 insns. Eliminate this problem by not combining with such an insn.
2033 Also, on some machines we don't want to extend the life of a hard
2034 register. */
2039 /* Don't extend the life of a hard register unless it is
2040 user variable (if we have few registers) or it can't
2041 fit into the desired register (meaning something special
2042 is going on).
2043 Also avoid substituting a return register into I3, because
2044 reload can't handle a conflict with constraints of other
2045 inputs. */
2050 }
2058 {
2061 /* If the clobber represents an earlyclobber operand, we must not
2062 substitute an expression containing the clobbered register.
2063 As we do not analyze the constraint strings here, we have to
2064 make the conservative assumption. However, if the register is
2065 a fixed hard reg, the clobber cannot represent any operand;
2066 we leave it up to the machine description to either accept or
2067 reject use-and-clobber patterns. */
2073 }
2075 /* If INSN contains anything volatile, or is an `asm' (whether volatile
2076 or not), reject, unless nothing volatile comes between it and I3 */
2079 {
2080 /* Make sure neither succ nor succ2 contains a volatile reference. */
2085 /* We'll check insns between INSN and I3 below. */
2086 }
2088 /* If INSN is an asm, and DEST is a hard register, reject, since it has
2089 to be an explicit register variable, and was chosen for a reason. */
2095 /* If INSN contains volatile references (specifically volatile MEMs),
2096 we cannot combine across any other volatile references.
2097 Even if INSN doesn't contain volatile references, any intervening
2098 volatile insn might affect machine state. */
2101 ? volatile_refs_p
2108 /* If INSN contains an autoincrement or autodecrement, make sure that
2109 register is not used between there and I3, and not already used in
2110 I3 either. Neither must it be used in PRED or SUCC, if they exist.
2111 Also insist that I3 not be a jump; if it were one
2112 and the incremented register were spilled, we would lose. */
2130 /* Don't combine an insn that follows a CC0-setting insn.
2131 An insn that uses CC0 must not be separated from the one that sets it.
2132 We do, however, allow I2 to follow a CC0-setting insn if that insn
2133 is passed as I1; in that case it will be deleted also.
2134 We also allow combining in this case if all the insns are adjacent
2135 because that would leave the two CC0 insns adjacent as well.
2136 It would be more logical to test whether CC0 occurs inside I1 or I2,
2137 but that would be much slower, and this ought to be equivalent. */
2140 {
2145 }
2147 /* If we get here, we have passed all the tests and the combination is
2148 to be allowed. */
2154 }
2155 \f
2156 /* LOC is the location within I3 that contains its pattern or the component
2157 of a PARALLEL of the pattern. We validate that it is valid for combining.
2159 One problem is if I3 modifies its output, as opposed to replacing it
2160 entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2161 doing so would produce an insn that is not equivalent to the original insns.
2163 Consider:
2165 (set (reg:DI 101) (reg:DI 100))
2166 (set (subreg:SI (reg:DI 101) 0) <foo>)
2168 This is NOT equivalent to:
2170 (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2171 (set (reg:DI 101) (reg:DI 100))])
2173 Not only does this modify 100 (in which case it might still be valid
2174 if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2176 We can also run into a problem if I2 sets a register that I1
2177 uses and I1 gets directly substituted into I3 (not via I2). In that
2178 case, we would be getting the wrong value of I2DEST into I3, so we
2179 must reject the combination. This case occurs when I2 and I1 both
2180 feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2181 If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2182 of a SET must prevent combination from occurring. The same situation
2183 can occur for I0, in which case I0_NOT_IN_SRC is set.
2185 Before doing the above check, we first try to expand a field assignment
2186 into a set of logical operations.
2188 If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2189 we place a register that is both set and used within I3. If more than one
2190 such register is detected, we fail.
2192 Return 1 if the combination is valid, zero otherwise. */
2194 static int
2197 {
2201 {
2206 rtx subdest;
2213 /* Check for the case where I3 modifies its output, as discussed
2214 above. We don't want to prevent pseudos from being combined
2215 into the address of a MEM, so only prevent the combination if
2216 i1 or i2 set the same MEM. */
2226 /* This is the same test done in can_combine_p except we can't test
2227 all_adjacent; we don't have to, since this instruction will stay
2228 in place, thus we are not considering increasing the lifetime of
2229 INNER_DEST.
2231 Also, if this insn sets a function argument, combining it with
2232 something that might need a spill could clobber a previous
2233 function argument; the all_adjacent test in can_combine_p also
2234 checks this; here, we do a more specific test for this case. */
2244 /* If DEST is used in I3, it is being killed in this insn, so
2245 record that for later. We have to consider paradoxical
2246 subregs here, since they kill the whole register, but we
2247 ignore partial subregs, STRICT_LOW_PART, etc.
2248 Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2249 STACK_POINTER_REGNUM, since these are always considered to be
2250 live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2258 && (HARD_FRAME_POINTER_IS_FRAME_POINTER
2264 {
2269 }
2270 }
2273 {
2280 }
2283 }
2284 \f
2285 /* Return 1 if X is an arithmetic expression that contains a multiplication
2286 and division. We don't count multiplications by powers of two here. */
2288 static int
2290 {
2292 {
2308 }
2309 }
2310 \f
2311 /* Determine whether INSN can be used in a combination. Return nonzero if
2312 not. This is used in try_combine to detect early some cases where we
2313 can't perform combinations. */
2315 static int
2317 {
2318 rtx set;
2321 /* If this isn't really an insn, we can't do anything.
2322 This can occur when flow deletes an insn that it has merged into an
2323 auto-increment address. */
2327 /* Never combine loads and stores involving hard regs that are likely
2328 to be spilled. The register allocator can usually handle such
2329 reg-reg moves by tying. If we allow the combiner to make
2330 substitutions of likely-spilled regs, reload might die.
2331 As an exception, we allow combinations involving fixed regs; these are
2332 not available to the register allocator so there's no risk involved. */
2353 }
2355 struct likely_spilled_retval_info
2356 {
2359 };
2361 /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2362 hard registers that are known to be written to / clobbered in full. */
2363 static void
2365 {
2382 else
2385 }
2387 /* Return nonzero iff part of the return value is live during INSN, and
2388 it is likely spilled. This can happen when more than one insn is needed
2389 to copy the return value, e.g. when we consider to combine into the
2390 second copy insn for a complex value. */
2392 static int
2394 {
2396 rtx reg;
2399 /* We assume here that no machine mode needs more than
2400 32 hard registers when the value overlaps with a register
2401 for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2416 /* Disregard parts of the return value that are set later. */
2425 /* Check if any of the (probably) live return value registers is
2426 likely spilled. */
2427 nregs --;
2428 do
2429 {
2435 }
2437 /* Adjust INSN after we made a change to its destination.
2439 Changing the destination can invalidate notes that say something about
2440 the results of the insn and a LOG_LINK pointing to the insn. */
2442 static void
2444 {
2445 /* For notes, be conservative and simply remove them. */
2448 /* The new insn will have a destination that was previously the destination
2449 of an insn just above it. Call distribute_links to make a LOG_LINK from
2450 the next use of that destination. */
2466 }
2468 /* Return TRUE if combine can reuse reg X in mode MODE.
2469 ADDED_SETS is nonzero if the original set is still required. */
2470 static bool
2472 {
2478 /* Don't change between modes with different underlying register sizes,
2479 since this could lead to invalid subregs. */
2485 /* Allow hard registers if the new mode is legal, and occupies no more
2486 registers than the old mode. */
2491 /* Or a pseudo that is only used once. */
2494 && !added_sets
2496 }
2499 /* Check whether X, the destination of a set, refers to part of
2500 the register specified by REG. */
2502 static bool
2504 {
2505 /* Check that reg is an integer mode register. */
2516 }
2518 /* Delete the unconditional jump INSN and adjust the CFG correspondingly.
2519 Note that the INSN should be deleted *after* removing dead edges, so
2520 that the kept edge is the fallthrough edge for a (set (pc) (pc))
2521 but not for a (set (pc) (label_ref FOO)). */
2523 static void
2525 {
2533 {
2538 /* Remove barriers from the footer if there are any. */
2541 {
2544 else
2548 }
2551 }
2552 }
2554 /* Return whether PAT is a PARALLEL of exactly N register SETs followed
2555 by an arbitrary number of CLOBBERs. */
2556 static bool
2558 {
2577 }
2579 /* Return whether INSN, a PARALLEL of N register SETs (and maybe some
2580 CLOBBERs), can be split into individual SETs in that order, without
2581 changing semantics. */
2582 static bool
2584 {
2592 {
2601 }
2604 }
2606 /* Try to combine the insns I0, I1 and I2 into I3.
2607 Here I0, I1 and I2 appear earlier than I3.
2608 I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2609 I3.
2611 If we are combining more than two insns and the resulting insn is not
2612 recognized, try splitting it into two insns. If that happens, I2 and I3
2613 are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2614 Otherwise, I0, I1 and I2 are pseudo-deleted.
2616 Return 0 if the combination does not work. Then nothing is changed.
2617 If we did the combination, return the insn at which combine should
2618 resume scanning.
2620 Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
2621 new direct jump instruction.
2623 LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2624 been I3 passed to an earlier try_combine within the same basic
2625 block. */
2630 {
2631 /* New patterns for I3 and I2, respectively. */
2635 /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2636 dead. */
2638 /* Total number of SETs to put into I3. */
2640 /* Nonzero if I2's or I1's body now appears in I3. */
2642 /* INSN_CODEs for new I3, new I2, and user of condition code. */
2644 /* Contains I3 if the destination of I3 is used in its source, which means
2645 that the old life of I3 is being killed. If that usage is placed into
2646 I2 and not in I3, a REG_DEAD note must be made. */
2648 /* SET_DEST and SET_SRC of I2, I1 and I0. */
2650 /* Copy of SET_SRC of I1 and I0, if needed. */
2652 /* Set if I2DEST was reused as a scratch register. */
2654 /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2656 /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2661 /* Notes that must be added to REG_NOTES in I3 and I2. */
2663 /* Notes that we substituted I3 into I2 instead of the normal case. */
2665 /* Notes that I1, I2 or I3 is a MULT operation. */
2672 rtx temp_expr;
2675 rtx new_other_notes;
2679 /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
2680 never be). */
2684 /* Only try four-insn combinations when there's high likelihood of
2685 success. Look for simple insns, such as loads of constants or
2686 binary operations involving a constant. */
2688 {
2698 {
2701 rtx src;
2706 {
2709 }
2711 ngood++;
2714 nshift++;
2715 }
2717 /* If I0 loads a memory and I3 sets the same memory, then I1 and I2
2718 are likely manipulating its value. Ideally we'll be able to combine
2719 all four insns into a bitfield insertion of some kind.
2721 Note the source in I0 might be inside a sign/zero extension and the
2722 memory modes in I0 and I3 might be different. So extract the address
2723 from the destination of I3 and search for it in the source of I0.
2725 In the event that there's a match but the source/dest do not actually
2726 refer to the same memory, the worst that happens is we try some
2727 combinations that we wouldn't have otherwise. */
2729 /* Ensure the source of SET0 is a MEM, possibly buried inside
2730 an extension. */
2736 /* Ensure the destination of SET3 is a MEM. */
2738 /* Would it be better to extract the base address for the MEM
2739 in SET3 and look for that? I don't have cases where it matters
2740 but I could envision such cases. */
2746 }
2748 /* Exit early if one of the insns involved can't be used for
2749 combinations. */
2760 combine_attempts++;
2763 /* Reset the hard register usage information. */
2767 {
2774 else
2777 }
2779 /* If multiple insns feed into one of I2 or I3, they can be in any
2780 order. To simplify the code below, reorder them in sequence. */
2790 /* First check for one important special case that the code below will
2791 not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2792 and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2793 we may be able to replace that destination with the destination of I3.
2794 This occurs in the common code where we compute both a quotient and
2795 remainder into a structure, in which case we want to do the computation
2796 directly into the structure to avoid register-register copies.
2798 Note that this case handles both multiple sets in I2 and also cases
2799 where I2 has a number of CLOBBERs inside the PARALLEL.
2801 We make very conservative checks below and only try to handle the
2802 most common cases of this. For example, we only handle the case
2803 where I2 and I3 are adjacent to avoid making difficult register
2804 usage tests. */
2812 /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2813 below would need to check what is inside (and reg_overlap_mentioned_p
2814 doesn't support those codes anyway). Don't allow those destinations;
2815 the resulting insn isn't likely to be recognized anyway. */
2821 {
2824 /* Make sure that the destination of I3,
2825 which we are going to substitute into one output of I2,
2826 is not used within another output of I2. We must avoid making this:
2827 (parallel [(set (mem (reg 69)) ...)
2828 (set (reg 69) ...)])
2829 which is not well-defined as to order of actions.
2830 (Besides, reload can't handle output reloads for this.)
2832 The problem can also happen if the dest of I3 is a memory ref,
2833 if another dest in I2 is an indirect memory ref.
2835 Neither can this PARALLEL be an asm. We do not allow combining
2836 that usually (see can_combine_p), so do not here either. */
2839 {
2848 }
2854 {
2855 combine_merges++;
2865 /* Replace the dest in I2 with our dest and make the resulting
2866 insn the new pattern for I3. Then skip to where we validate
2867 the pattern. Everything was set up above. */
2872 }
2873 }
2875 /* If I2 is setting a pseudo to a constant and I3 is setting some
2876 sub-part of it to another constant, merge them by making a new
2877 constant. */
2885 {
2892 {
2897 {
2903 }
2904 }
2905 else
2906 {
2910 {
2913 }
2914 }
2917 {
2918 /* If this is the low part, we're done. */
2920 ;
2921 /* Handle the case where inner is twice the size of outer. */
2925 /* Otherwise give up for now. */
2926 else
2928 }
2931 {
2939 combine_merges++;
2946 /* Replace the source in I2 with the new constant and make the
2947 resulting insn the new pattern for I3. Then skip to where we
2948 validate the pattern. Everything was set up above. */
2954 /* The dest of I3 has been replaced with the dest of I2. */
2957 }
2958 }
2960 /* If we have no I1 and I2 looks like:
2961 (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
2962 (set Y OP)])
2963 make up a dummy I1 that is
2964 (set Y OP)
2965 and change I2 to be
2966 (set (reg:CC X) (compare:CC Y (const_int 0)))
2968 (We can ignore any trailing CLOBBERs.)
2970 This undoes a previous combination and allows us to match a branch-and-
2971 decrement insn. */
2983 {
2984 /* We make I1 with the same INSN_UID as I2. This gives it
2985 the same DF_INSN_LUID for value tracking. Our fake I1 will
2986 never appear in the insn stream so giving it the same INSN_UID
2987 as I2 will not cause a problem. */
3000 }
3002 /* If I2 is a PARALLEL of two SETs of REGs (and perhaps some CLOBBERs),
3003 make those two SETs separate I1 and I2 insns, and make an I0 that is
3004 the original I1. */
3010 {
3011 /* If there is no I1, there is no I0 either. */
3014 /* We make I1 with the same INSN_UID as I2. This gives it
3015 the same DF_INSN_LUID for value tracking. Our fake I1 will
3016 never appear in the insn stream so giving it the same INSN_UID
3017 as I2 will not cause a problem. */
3025 }
3027 /* Verify that I2 and I1 are valid for combining. */
3033 {
3036 }
3038 /* Record whether I2DEST is used in I2SRC and similarly for the other
3039 cases. Knowing this will help in register status updating below. */
3050 /* For the earlier insns, determine which of the subsequent ones they
3051 feed. */
3058 /* Ensure that I3's pattern can be the destination of combines. */
3064 {
3067 }
3069 /* See if any of the insns is a MULT operation. Unless one is, we will
3070 reject a combination that is, since it must be slower. Be conservative
3071 here. */
3079 /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
3080 We used to do this EXCEPT in one case: I3 has a post-inc in an
3081 output operand. However, that exception can give rise to insns like
3082 mov r3,(r3)+
3083 which is a famous insn on the PDP-11 where the value of r3 used as the
3084 source was model-dependent. Avoid this sort of thing. */
3086 #if 0
3092 /* It's not the exception. */
3093 #endif
3095 {
3096 rtx link;
3102 {
3105 }
3106 }
3108 /* See if the SETs in I1 or I2 need to be kept around in the merged
3109 instruction: whenever the value set there is still needed past I3.
3110 For the SET in I2, this is easy: we see if I2DEST dies or is set in I3.
3112 For the SET in I1, we have two cases: if I1 and I2 independently feed
3113 into I3, the set in I1 needs to be kept around unless I1DEST dies
3114 or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
3115 in I1 needs to be kept around unless I1DEST dies or is set in either
3116 I2 or I3. The same considerations apply to I0. */
3123 else
3131 else
3134 /* We are about to copy insns for the case where they need to be kept
3135 around. Check that they can be copied in the merged instruction. */
3141 {
3144 }
3146 /* If the set in I2 needs to be kept around, we must make a copy of
3147 PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3148 PATTERN (I2), we are only substituting for the original I1DEST, not into
3149 an already-substituted copy. This also prevents making self-referential
3150 rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3151 I2DEST. */
3154 {
3157 else
3159 }
3162 {
3165 else
3167 }
3170 {
3173 else
3175 }
3177 combine_merges++;
3179 /* Substitute in the latest insn for the regs set by the earlier ones. */
3185 /* Many machines that don't use CC0 have insns that can both perform an
3186 arithmetic operation and set the condition code. These operations will
3187 be represented as a PARALLEL with the first element of the vector
3188 being a COMPARE of an arithmetic operation with the constant zero.
3189 The second element of the vector will set some pseudo to the result
3190 of the same arithmetic operation. If we simplify the COMPARE, we won't
3191 match such a pattern and so will generate an extra insn. Here we test
3192 for this case, where both the comparison and the operation result are
3193 needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3194 I2SRC. Later we will make the PARALLEL that contains I2. */
3200 {
3201 rtx newpat_dest;
3207 scalar_int_mode mode;
3216 {
3222 }
3224 /* Do the rest only if op1 is const0_rtx, which may be the
3225 result of simplification. */
3227 {
3228 /* If a single use of the CC is found, prepare to modify it
3229 when SELECT_CC_MODE returns a new CC-class mode, or when
3230 the above simplify_compare_const() returned a new comparison
3231 operator. undobuf.other_insn is assigned the CC use insn
3232 when modifying it. */
3234 {
3235 #ifdef SELECT_CC_MODE
3236 machine_mode new_mode
3241 {
3246 else
3247 {
3250 }
3251 }
3252 #endif
3253 /* Cases for modifying the CC-using comparison. */
3255 /* ??? Do we need to verify the zero rtx? */
3257 {
3258 /* Replace cc_use_loc with entire new RTX. */
3263 }
3265 {
3266 /* Just replace the CC reg with a new mode. */
3269 }
3270 }
3272 /* Now we modify the current newpat:
3273 First, SET_DEST(newpat) is updated if the CC mode has been
3274 altered. For targets without SELECT_CC_MODE, this should be
3275 optimized away. */
3278 /* This is always done to propagate i2src into newpat. */
3281 /* Create new version of i2pat if needed; the below PARALLEL
3282 creation needs this to work correctly. */
3286 }
3287 }
3290 {
3291 /* It is possible that the source of I2 or I1 may be performing
3292 an unneeded operation, such as a ZERO_EXTEND of something
3293 that is known to have the high part zero. Handle that case
3294 by letting subst look at the inner insns.
3296 Another way to do this would be to have a function that tries
3297 to simplify a single insn instead of merging two or more
3298 insns. We don't do this because of the potential of infinite
3299 loops and because of the potential extra memory required.
3300 However, doing it the way we are is a bit of a kludge and
3301 doesn't catch all cases.
3303 But only do this if -fexpensive-optimizations since it slows
3304 things down and doesn't usually win.
3306 This is not done in the COMPARE case above because the
3307 unmodified I2PAT is used in the PARALLEL and so a pattern
3308 with a modified I2SRC would not match. */
3311 {
3312 /* Pass pc_rtx so no substitutions are done, just
3313 simplifications. */
3315 {
3318 }
3322 }
3327 /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3328 copy of I2SRC each time we substitute it, in order to avoid creating
3329 self-referential RTL when we will be substituting I1SRC for I1DEST
3330 later. Likewise if I0 feeds into I2, either directly or indirectly
3331 through I1, and I0DEST is in I0SRC. */
3338 /* Record whether I2's body now appears within I3's body. */
3340 }
3342 /* If we already got a failure, don't try to do more. Otherwise, try to
3343 substitute I1 if we have it. */
3346 {
3347 /* Check that an autoincrement side-effect on I1 has not been lost.
3348 This happens if I1DEST is mentioned in I2 and dies there, and
3349 has disappeared from the new pattern. */
3351 && i1_feeds_i2_n
3354 /* Before we can do this substitution, we must redo the test done
3355 above (see detailed comments there) that ensures I1DEST isn't
3356 mentioned in any SETs in NEWPAT that are field assignments. */
3359 {
3362 }
3367 /* If the following substitution will modify I1SRC, make a copy of it
3368 for the case where it is substituted for I1DEST in I2PAT later. */
3372 /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3373 copy of I1SRC each time we substitute it, in order to avoid creating
3374 self-referential RTL when we will be substituting I0SRC for I0DEST
3375 later. */
3380 /* Record whether I1's body now appears within I3's body. */
3382 }
3384 /* Likewise for I0 if we have it. */
3387 {
3394 {
3397 }
3399 /* If the following substitution will modify I0SRC, make a copy of it
3400 for the case where it is substituted for I0DEST in I1PAT later. */
3403 /* And a copy for I0DEST in I2PAT substitution. */
3412 }
3414 /* Fail if an autoincrement side-effect has been duplicated. Be careful
3415 to count all the ways that I2SRC and I1SRC can be used. */
3426 /* Fail if we tried to make a new register. */
3428 /* Fail if we couldn't do something and have a CLOBBER. */
3430 /* Fail if this new pattern is a MULT and we didn't have one before
3431 at the outer level. */
3434 {
3437 }
3439 /* If the actions of the earlier insns must be kept
3440 in addition to substituting them into the latest one,
3441 we must make a new PARALLEL for the latest insn
3442 to hold additional the SETs. */
3445 {
3447 combine_extras++;
3450 {
3456 }
3457 else
3458 {
3463 }
3469 {
3475 }
3477 {
3486 }
3487 }
3489 validate_replacement:
3491 /* Note which hard regs this insn has as inputs. */
3494 /* If recog_for_combine fails, it strips existing clobbers. If we'll
3495 consider splitting this pattern, we might need these clobbers. */
3498 {
3504 }
3506 /* We have recognized nothing yet. */
3509 /* See if this is a PARALLEL of two SETs where one SET's destination is
3510 a register that is unused and this isn't marked as an instruction that
3511 might trap in an EH region. In that case, we just need the other SET.
3512 We prefer this over the PARALLEL.
3514 This can occur when simplifying a divmod insn. We *must* test for this
3515 case here because the code below that splits two independent SETs doesn't
3516 handle this case correctly when it updates the register status.
3518 It's pointless doing this if we originally had two sets, one from
3519 i3, and one from i2. Combining then splitting the parallel results
3520 in the original i2 again plus an invalid insn (which we delete).
3521 The net effect is only to move instructions around, which makes
3522 debug info less accurate.
3524 If the remaining SET came from I2 its destination should not be used
3525 between I2 and I3. See PR82024. */
3530 {
3541 {
3544 }
3553 {
3558 {
3564 }
3565 }
3569 }
3571 /* Is the result of combination a valid instruction? */
3575 /* If we were combining three insns and the result is a simple SET
3576 with no ASM_OPERANDS that wasn't recognized, try to split it into two
3577 insns. There are two ways to do this. It can be split using a
3578 machine-specific method (like when you have an addition of a large
3579 constant) or by combine in the function find_split_point. */
3583 {
3587 /* See if the MD file can split NEWPAT. If it can't, see if letting it
3588 use I2DEST as a scratch register will help. In the latter case,
3589 convert I2DEST to the mode of the source of NEWPAT if we can. */
3593 /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3594 inputs of NEWPAT. */
3596 /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3597 possible to try that as a scratch reg. This would require adding
3598 more code to make it work though. */
3601 {
3604 /* ??? Reusing i2dest without resetting the reg_stat entry for it
3605 (temporarily, until we are committed to this instruction
3606 combination) does not work: for example, any call to nonzero_bits
3607 on the register (from a splitter in the MD file, for example)
3608 will get the old information, which is invalid.
3610 Since nowadays we can create registers during combine just fine,
3611 we should just create a new one here, not reuse i2dest. */
3613 /* First try to split using the original register as a
3614 scratch register. */
3618 i2dest)));
3621 /* If that didn't work, try changing the mode of I2DEST if
3622 we can. */
3627 {
3629 rtx ni2dest;
3633 else
3634 {
3637 }
3639 parallel = (gen_rtx_PARALLEL
3643 ni2dest))));
3648 {
3656 }
3657 }
3660 }
3662 /* If recog_for_combine has discarded clobbers, try to use them
3663 again for the split. */
3665 {
3668 }
3671 {
3676 }
3680 {
3690 /* If I2 or I3 has multiple SETs, we won't know how to track
3691 register status, so don't use these insns. If I2's destination
3692 is used between I2 and I3, we also can't use these insns. */
3702 /* It is possible that both insns now set the destination of I3.
3703 If so, we must show an extra use of it. */
3706 {
3725 }
3726 }
3728 /* If we can split it and use I2DEST, go ahead and see if that
3729 helps things be recognized. Verify that none of the registers
3730 are set between I2 and I3. */
3734 /* We need I2DEST in the proper mode. If it is a hard register
3735 or the only use of a pseudo, we can change its mode.
3736 Make sure we don't change a hard register to have a mode that
3737 isn't valid for it, or change the number of registers. */
3744 /* We can't overwrite I2DEST if its value is still used by
3745 NEWPAT. */
3747 {
3756 /* *SPLIT may be part of I2SRC, so make sure we have the
3757 original expression around for later debug processing.
3758 We should not need I2SRC any more in other cases. */
3761 else
3764 /* Get NEWDEST as a register in the proper mode. We have already
3765 validated that we can do this. */
3767 {
3770 else
3771 {
3774 }
3775 }
3777 /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3778 an ASHIFT. This can occur if it was inside a PLUS and hence
3779 appeared to be a memory address. This is a kludge. */
3784 {
3787 /* Update split_code because we may not have a multiply
3788 anymore. */
3790 }
3792 /* Similarly for (plus (mult FOO (const_int pow2))). */
3798 {
3802 /* Update split_code because we may not have a multiply
3803 anymore. */
3805 }
3807 #ifdef INSN_SCHEDULING
3808 /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3809 be written as a ZERO_EXTEND. */
3811 {
3812 /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3813 what it really is. */
3818 else
3821 }
3822 #endif
3824 /* Attempt to split binary operators using arithmetic identities. */
3828 {
3835 /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3837 {
3842 }
3843 /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3850 {
3856 /* Split both "((X op Y) op X) op Y" and
3857 "((X op Y) op Y) op X" as "T op T" where T is
3858 "X op Y". */
3861 {
3866 }
3867 /* Split "((X op X) op Y) op Y)" as "T op T" where
3868 T is "X op Y". */
3870 {
3876 }
3877 }
3878 }
3881 {
3884 }
3888 /* recog_for_combine might have added CLOBBERs to newi2pat.
3889 Make sure NEWPAT does not depend on the clobbered regs. */
3893 {
3896 {
3899 }
3900 }
3902 /* If the split point was a MULT and we didn't have one before,
3903 don't use one now. */
3906 }
3907 }
3909 /* Check for a case where we loaded from memory in a narrow mode and
3910 then sign extended it, but we need both registers. In that case,
3911 we have a PARALLEL with both loads from the same memory location.
3912 We can split this into a load from memory followed by a register-register
3913 copy. This saves at least one insn, more if register allocation can
3914 eliminate the copy.
3916 We cannot do this if the destination of the first assignment is a
3917 condition code register or cc0. We eliminate this case by making sure
3918 the SET_DEST and SET_SRC have the same mode.
3920 We cannot do this if the destination of the second assignment is
3921 a register that we have already assumed is zero-extended. Similarly
3922 for a SUBREG of such a register. */
3957 {
3958 rtx ni2dest;
3972 }
3974 /* Similarly, check for a case where we have a PARALLEL of two independent
3975 SETs but we started with three insns. In this case, we can do the sets
3976 as two separate insns. This case occurs when some SET allows two
3977 other insns to combine, but the destination of that SET is still live.
3979 Also do this if we started with two insns and (at least) one of the
3980 resulting sets is a noop; this noop will be deleted later. */
3999 {
4003 /* Normally, it doesn't matter which of the two is done first,
4004 but the one that references cc0 can't be the second, and
4005 one which uses any regs/memory set in between i2 and i3 can't
4006 be first. The PARALLEL might also have been pre-existing in i3,
4007 so we need to make sure that we won't wrongly hoist a SET to i2
4008 that would conflict with a death note present in there. */
4016 /* If I3 is a jump, ensure that set0 is a jump so that
4017 we do not create invalid RTL. */
4019 )
4020 {
4023 }
4031 /* If I3 is a jump, ensure that set1 is a jump so that
4032 we do not create invalid RTL. */
4034 )
4035 {
4038 }
4039 else
4040 {
4043 }
4048 {
4049 /* recog_for_combine might have added CLOBBERs to newi2pat.
4050 Make sure NEWPAT does not depend on the clobbered regs. */
4052 {
4055 {
4058 {
4061 }
4062 }
4063 }
4066 }
4067 }
4069 /* If it still isn't recognized, fail and change things back the way they
4070 were. */
4072 /* Is the result a reasonable ASM_OPERANDS? */
4074 {
4077 }
4079 /* If we had to change another insn, make sure it is valid also. */
4081 {
4089 {
4092 }
4093 }
4095 /* If I2 is the CC0 setter and I3 is the CC0 user then check whether
4096 they are adjacent to each other or not. */
4098 {
4102 {
4105 }
4106 }
4108 /* Only allow this combination if insn_cost reports that the
4109 replacement instructions are cheaper than the originals. */
4111 {
4114 }
4117 {
4122 {
4127 /* Temporarily revert mode back. */
4131 {
4132 /* If we used i2dest as a scratch register with a
4133 different mode, substitute it for the original
4134 i2src while its original mode is temporarily
4135 restored, and then clear i2scratch so that we don't
4136 do it again later. */
4138 this_basic_block);
4140 /* Put back the new mode. */
4142 }
4143 else
4144 {
4149 {
4152 }
4153 else
4154 {
4161 }
4163 /* We're dealing with a reg that changed mode but not
4164 meaning, so we want to turn it into a subreg for
4165 the new mode. However, because of REG sharing and
4166 because its mode had already changed, we have to do
4167 it in two steps. First, replace any debug uses of
4168 reg, with its original mode temporarily restored,
4169 with this copy we have created; then, replace the
4170 copy with the SUBREG of the original shared reg,
4171 once again changed to the new mode. */
4173 this_basic_block);
4177 this_basic_block);
4178 }
4179 }
4180 }
4182 /* If we will be able to accept this, we have made a
4183 change to the destination of I3. This requires us to
4184 do a few adjustments. */
4187 {
4190 }
4192 /* We now know that we can do this combination. Merge the insns and
4193 update the status of registers and LOG_LINKS. */
4196 {
4201 /* If any of the notes in OTHER_INSN were REG_DEAD or REG_UNUSED,
4202 ensure that they are still valid. Then add any non-duplicate
4203 notes added by recog_for_combine. */
4205 {
4214 /* Simply drop equal note since it may be no longer valid
4215 for other_insn. It may be possible to record that CC
4216 register is changed and only discard those notes, but
4217 in practice it's unnecessary complication and doesn't
4218 give any meaningful improvement.
4220 See PR78559. */
4224 }
4228 NULL_RTX);
4229 }
4232 {
4235 rtx ni2dest;
4237 /* I3 now uses what used to be its destination and which is now
4238 I2's destination. This requires us to do a few adjustments. */
4242 /* We need a LOG_LINK from I3 to I2. But we used to have one,
4243 so we still will.
4245 However, some later insn might be using I2's dest and have
4246 a LOG_LINK pointing at I3. We must remove this link.
4247 The simplest way to remove the link is to point it at I1,
4248 which we know will be a NOTE. */
4250 /* newi2pat is usually a SET here; however, recog_for_combine might
4251 have added some clobbers. */
4254 else
4261 {
4264 {
4270 }
4271 }
4272 }
4274 {
4279 /* Compute which registers we expect to eliminate. newi2pat may be setting
4280 either i3dest or i2dest, so we must check it. */
4283 || !i2dest_killed
4285 /* For i1, we need to compute both local elimination and global
4286 elimination information with respect to newi2pat because i1dest
4287 may be the same as i3dest, in which case newi2pat may be setting
4288 i1dest. Global information is used when distributing REG_DEAD
4289 note for i2 and i3, in which case it does matter if newi2pat sets
4290 i1dest or not.
4292 Local information is used when distributing REG_DEAD note for i1,
4293 in which case it doesn't matter if newi2pat sets i1dest or not.
4294 See PR62151, if we have four insns combination:
4295 i0: r0 <- i0src
4296 i1: r1 <- i1src (using r0)
4297 REG_DEAD (r0)
4298 i2: r0 <- i2src (using r1)
4299 i3: r3 <- i3src (using r0)
4300 ix: using r0
4301 From i1's point of view, r0 is eliminated, no matter if it is set
4302 by newi2pat or not. In other words, REG_DEAD info for r0 in i1
4303 should be discarded.
4305 Note local information only affects cases in forms like "I1->I2->I3",
4306 "I0->I1->I2->I3" or "I0&I1->I2, I2->I3". For other cases like
4307 "I0->I1, I1&I2->I3" or "I1&I2->I3", newi2pat won't set i1dest or
4308 i0dest anyway. */
4310 || !i1dest_killed
4315 /* Same case as i1. */
4322 /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4323 clear them. */
4331 /* Ensure that we do not have something that should not be shared but
4332 occurs multiple times in the new insns. Check this by first
4333 resetting all the `used' flags and then copying anything is shared. */
4357 {
4360 {
4362 {
4363 /* I2SRC must still be meaningful at this point. Some
4364 splitting operations can invalidate I2SRC, but those
4365 operations do not apply to calls. */
4369 }
4376 }
4377 }
4382 /* We had one special case above where I2 had more than one set and
4383 we replaced a destination of one of those sets with the destination
4384 of I3. In that case, we have to update LOG_LINKS of insns later
4385 in this basic block. Note that this (expensive) case is rare.
4387 Also, in this case, we must pretend that all REG_NOTEs for I2
4388 actually came from I3, so that REG_UNUSED notes from I2 will be
4389 properly handled. */
4392 {
4401 temp_insn
4411 {
4416 }
4417 else
4420 }
4428 {
4431 this_basic_block);
4434 }
4435 else
4436 {
4439 this_basic_block);
4441 }
4444 {
4449 this_basic_block);
4451 }
4454 {
4459 this_basic_block);
4461 }
4463 /* Get death notes for everything that is now used in either I3 or
4464 I2 and used to die in a previous insn. If we built two new
4465 patterns, move from I1 to I2 then I2 to I3 so that we get the
4466 proper movement on registers that I2 modifies. */
4472 else
4478 /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4495 /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4496 know these are REG_UNUSED and want them to go to the desired insn,
4497 so we always pass it as i3. */
4501 NULL_RTX);
4505 NULL_RTX);
4507 /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4508 put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4509 I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4510 in that case, it might delete I2. Similarly for I2 and I1.
4511 Show an additional death due to the REG_DEAD note we make here. If
4512 we discard it in distribute_notes, we will decrement it again. */
4515 {
4520 else
4523 }
4526 {
4531 else
4534 }
4537 {
4542 else
4545 }
4548 {
4553 else
4556 }
4564 {
4569 /* The insn that used to set this register doesn't exist, and
4570 this life of the register may not exist either. See if one of
4571 I3's links points to an insn that sets I2DEST. If it does,
4572 that is now the last known value for I2DEST. If we don't update
4573 this and I2 set the register to a value that depended on its old
4574 contents, we will get confused. If this insn is used, thing
4575 will be set correctly in combine_instructions. */
4583 /* If the reg formerly set in I2 died only once and that was in I3,
4584 zero its use count so it won't make `reload' do any work. */
4587 && ! i2dest_in_i2src
4590 }
4593 {
4606 && ! i1dest_in_i1src
4609 }
4612 {
4625 && ! i0dest_in_i0src
4628 }
4630 /* Update reg_stat[].nonzero_bits et al for any changes that may have
4631 been made to this insn. The order is important, because newi2pat
4632 can affect nonzero_bits of newpat. */
4636 }
4639 {
4641 {
4644 }
4646 }
4649 {
4651 {
4654 }
4656 }
4659 {
4661 {
4664 }
4666 }
4669 {
4671 {
4674 }
4676 }
4679 {
4681 {
4684 }
4686 }
4688 /* Set new_direct_jump_p if a new return or simple jump instruction
4689 has been created. Adjust the CFG accordingly. */
4691 {
4695 }
4700 {
4703 }
4707 {
4713 }
4718 {
4724 }
4726 /* A noop might also need cleaning up of CFG, if it comes from the
4727 simplification of a jump. */
4732 {
4735 }
4742 {
4745 }
4747 combine_successes++;
4754 else
4756 }
4757 \f
4758 /* Get a marker for undoing to the current state. */
4762 {
4764 }
4766 /* Undo the modifications up to the marker. */
4768 static void
4770 {
4774 {
4779 {
4794 }
4798 }
4801 }
4803 /* Undo all the modifications recorded in undobuf. */
4805 static void
4807 {
4809 }
4811 /* We've committed to accepting the changes we made. Move all
4812 of the undos to the free list. */
4814 static void
4816 {
4820 {
4824 }
4826 }
4827 \f
4828 /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4829 where we have an arithmetic expression and return that point. LOC will
4830 be inside INSN.
4832 try_combine will call this function to see if an insn can be split into
4833 two insns. */
4837 {
4847 /* First special-case some codes. */
4849 {
4851 #ifdef INSN_SCHEDULING
4852 /* If we are making a paradoxical SUBREG invalid, it becomes a split
4853 point. */
4856 #endif
4860 /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4861 using LO_SUM and HIGH. */
4864 {
4872 }
4874 /* If we have a PLUS whose second operand is a constant and the
4875 address is not valid, perhaps will can split it up using
4876 the machine-specific way to split large constants. We use
4877 the first pseudo-reg (one of the virtual regs) as a placeholder;
4878 it will not remain in the result. */
4883 {
4886 subst_insn);
4888 /* This should have produced two insns, each of which sets our
4889 placeholder. If the source of the second is a valid address,
4890 we can make put both sources together and make a split point
4891 in the middle. */
4904 && memory_address_addr_space_p
4907 {
4911 /* Replace the placeholder in SRC2 with SRC1. If we can
4912 find where in SRC2 it was placed, that can become our
4913 split point and we can replace this address with SRC2.
4914 Just try two obvious places. */
4925 {
4928 }
4929 }
4931 /* If that didn't work, perhaps the first operand is complex and
4932 needs to be computed separately, so make a split point there.
4933 This will occur on machines that just support REG + CONST
4934 and have a constant moved through some previous computation. */
4940 }
4942 /* If we have a PLUS whose first operand is complex, try computing it
4943 separately by making a split there. */
4954 /* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
4955 ZERO_EXTRACT, the most likely reason why this doesn't match is that
4956 we need to put the operand into a register. So split at that
4957 point. */
4967 /* See if we can split SET_SRC as it stands. */
4972 /* See if we can split SET_DEST as it stands. */
4977 /* See if this is a bitfield assignment with everything constant. If
4978 so, this is an IOR of an AND, so split it into that. */
4990 {
4995 unsigned HOST_WIDE_INT mask
4997 rtx or_mask;
5006 else
5007 {
5013 or_mask));
5014 }
5021 }
5023 /* Otherwise, see if this is an operation that we can split into two.
5024 If so, try to split that. */
5028 {
5030 /* If we are AND'ing with a large constant that is only a single
5031 bit and the result is only being used in a context where we
5032 need to know if it is zero or nonzero, replace it with a bit
5033 extraction. This will avoid the large constant, which might
5034 have taken more than one insn to make. If the constant were
5035 not a valid argument to the AND but took only one insn to make,
5036 this is no worse, but if it took more than one insn, it will
5037 be better. */
5047 {
5052 {
5055 }
5056 }
5060 /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
5061 is known to be on, this can be converted into a NEG of a shift. */
5067 {
5079 }
5085 /* We can't optimize if either mode is a partial integer
5086 mode as we don't know how many bits are significant
5087 in those modes. */
5103 {
5111 }
5116 }
5121 {
5122 /* For unsigned, we have a choice of a shift followed by an
5123 AND or two shifts. Use two shifts for field sizes where the
5124 constant might be too large. We assume here that we can
5125 always at least get 8-bit constants in an AND insn, which is
5126 true for every current RISC. */
5129 {
5130 unsigned HOST_WIDE_INT mask
5134 gen_rtx_LSHIFTRT
5142 }
5143 else
5144 {
5146 gen_rtx_fmt_ee
5157 }
5158 }
5160 /* See if this is a simple operation with a constant as the second
5161 operand. It might be that this constant is out of range and hence
5162 could be used as a split point. */
5170 /* Finally, see if this is a simple operation with its first operand
5171 not in a register. The operation might require this operand in a
5172 register, so return it as a split point. We can always do this
5173 because if the first operand were another operation, we would have
5174 already found it as a split point. */
5183 /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
5184 it is better to write this as (not (ior A B)) so we can split it.
5185 Similarly for IOR. */
5187 {
5195 }
5197 /* Many RISC machines have a large set of logical insns. If the
5198 second operand is a NOT, put it first so we will try to split the
5199 other operand first. */
5201 {
5205 }
5210 /* Canonicalization can produce (minus A (mult B C)), where C is a
5211 constant. It may be better to try splitting (plus (mult B -C) A)
5212 instead if this isn't a multiply by a power of two. */
5216 {
5224 mode)),
5227 }
5229 /* Split at a multiply-accumulate instruction. However if this is
5230 the SET_SRC, we likely do not have such an instruction and it's
5231 worthless to try this split. */
5240 }
5242 /* Otherwise, select our actions depending on our rtx class. */
5244 {
5250 /* fall through */
5258 /* fall through */
5260 /* Some machines have (and (shift ...) ...) insns. If X is not
5261 an AND, but XEXP (X, 0) is, use it as our split point. */
5271 /* Otherwise, we don't have a split point. */
5273 }
5274 }
5275 \f
5276 /* Throughout X, replace FROM with TO, and return the result.
5277 The result is TO if X is FROM;
5278 otherwise the result is X, but its contents may have been modified.
5279 If they were modified, a record was made in undobuf so that
5280 undo_all will (among other things) return X to its original state.
5282 If the number of changes necessary is too much to record to undo,
5283 the excess changes are not made, so the result is invalid.
5284 The changes already made can still be undone.
5285 undobuf.num_undo is incremented for such changes, so by testing that
5286 the caller can tell whether the result is valid.
5288 `n_occurrences' is incremented each time FROM is replaced.
5290 IN_DEST is nonzero if we are processing the SET_DEST of a SET.
5292 IN_COND is nonzero if we are at the top level of a condition.
5294 UNIQUE_COPY is nonzero if each substitution must be unique. We do this
5295 by copying if `n_occurrences' is nonzero. */
5297 static rtx
5299 {
5304 rtx new_rtx;
5306 /* Two expressions are equal if they are identical copies of a shared
5307 RTX or if they are both registers with the same register number
5308 and mode. */
5310 #define COMBINE_RTX_EQUAL_P(X,Y) \
5311 ((X) == (Y) \
5312 || (REG_P (X) && REG_P (Y) \
5313 && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5315 /* Do not substitute into clobbers of regs -- this will never result in
5316 valid RTL. */
5321 {
5322 n_occurrences++;
5324 }
5326 /* If X and FROM are the same register but different modes, they
5327 will not have been seen as equal above. However, the log links code
5328 will make a LOG_LINKS entry for that case. If we do nothing, we
5329 will try to rerecognize our original insn and, when it succeeds,
5330 we will delete the feeding insn, which is incorrect.
5332 So force this insn not to match in this (rare) case. */
5337 /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5338 of which may contain things that can be combined. */
5342 /* It is possible to have a subexpression appear twice in the insn.
5343 Suppose that FROM is a register that appears within TO.
5344 Then, after that subexpression has been scanned once by `subst',
5345 the second time it is scanned, TO may be found. If we were
5346 to scan TO here, we would find FROM within it and create a
5347 self-referent rtl structure which is completely wrong. */
5351 /* Parallel asm_operands need special attention because all of the
5352 inputs are shared across the arms. Furthermore, unsharing the
5353 rtl results in recognition failures. Failure to handle this case
5354 specially can result in circular rtl.
5356 Solve this by doing a normal pass across the first entry of the
5357 parallel, and only processing the SET_DESTs of the subsequent
5358 entries. Ug. */
5363 {
5366 /* If this substitution failed, this whole thing fails. */
5374 {
5380 {
5383 /* If this substitution failed, this whole thing fails. */
5389 }
5390 }
5391 }
5392 else
5393 {
5397 /* We don't need to process a SET_DEST that is a register, CC0,
5398 or PC, so set up to skip this common case. All other cases
5399 where we want to suppress replacing something inside a
5400 SET_SRC are handled via the IN_DEST operand. */
5407 /* Trying to simplify the operands of a widening MULT is not likely
5408 to create RTL matching a machine insn. */
5420 /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5421 constant. */
5426 {
5428 {
5431 {
5433 {
5436 n_occurrences++;
5437 }
5438 else
5439 {
5441 unique_copy);
5443 /* If this substitution failed, this whole thing
5444 fails. */
5448 }
5451 }
5452 }
5454 {
5455 /* If this is a register being set, ignore it. */
5462 ;
5465 {
5466 /* In general, don't install a subreg involving two
5467 modes not tieable. It can worsen register
5468 allocation, and can even make invalid reload
5469 insns, since the reg inside may need to be copied
5470 from in the outside mode, and that may be invalid
5471 if it is an fp reg copied in integer mode.
5473 We allow two exceptions to this: It is valid if
5474 it is inside another SUBREG and the mode of that
5475 SUBREG and the mode of the inside of TO is
5476 tieable and it is valid if X is a SET that copies
5477 FROM to CC0. */
5485 && (!HAVE_cc0
5500 n_occurrences++;
5501 }
5502 else
5503 /* If we are in a SET_DEST, suppress most cases unless we
5504 have gone inside a MEM, in which case we want to
5505 simplify the address. We assume here that things that
5506 are actually part of the destination have their inner
5507 parts in the first expression. This is true for SUBREG,
5508 STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5509 things aside from REG and MEM that should appear in a
5510 SET_DEST. */
5512 (((in_dest
5518 unique_copy);
5520 /* If we found that we will have to reject this combination,
5521 indicate that by returning the CLOBBER ourselves, rather than
5522 an expression containing it. This will speed things up as
5523 well as prevent accidents where two CLOBBERs are considered
5524 to be equal, thus producing an incorrect simplification. */
5530 {
5538 }
5541 {
5545 }
5546 else
5548 }
5549 }
5550 }
5552 /* Check if we are loading something from the constant pool via float
5553 extension; in this case we would undo compress_float_constant
5554 optimization and degenerate constant load to an immediate value. */
5558 {
5562 }
5564 /* Try to simplify X. If the simplification changed the code, it is likely
5565 that further simplification will help, so loop, but limit the number
5566 of repetitions that will be performed. */
5569 {
5570 /* If X is sufficiently simple, don't bother trying to do anything
5571 with it. */
5580 /* We no longer know the original mode of operand 0 since we
5581 have changed the form of X) */
5583 }
5586 }
5587 \f
5588 /* If X is a commutative operation whose operands are not in the canonical
5589 order, use substitutions to swap them. */
5591 static void
5593 {
5596 {
5600 }
5601 }
5603 /* Simplify X, a piece of RTL. We just operate on the expression at the
5604 outer level; call `subst' to simplify recursively. Return the new
5605 expression.
5607 OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
5608 if we are inside a SET_DEST. IN_COND is nonzero if we are at the top level
5609 of a condition. */
5611 static rtx
5614 {
5617 scalar_int_mode int_mode;
5618 rtx temp;
5621 /* If this is a commutative operation, put a constant last and a complex
5622 expression first. We don't need to do this for comparisons here. */
5625 /* Try to fold this expression in case we have constants that weren't
5626 present before. */
5629 {
5637 {
5640 {
5644 }
5647 }
5660 }
5663 {
5668 }
5670 /* If this is a simple operation applied to an IF_THEN_ELSE, try
5671 applying it to the arms of the IF_THEN_ELSE. This often simplifies
5672 things. Check for cases where both arms are testing the same
5673 condition.
5675 Don't do anything if all operands are very simple. */
5688 {
5693 /* If everything is a comparison, what we have is highly unlikely
5694 to be simpler, so don't use it. */
5697 {
5704 /* Simplify the alternative arms; this may collapse the true and
5705 false arms to store-flag values. Be careful to use copy_rtx
5706 here since true_rtx or false_rtx might share RTL with x as a
5707 result of the if_then_else_cond call above. */
5711 /* If true_rtx and false_rtx are not general_operands, an if_then_else
5712 is unlikely to be simpler. */
5715 {
5718 /* Restarting if we generate a store-flag expression will cause
5719 us to loop. Just drop through in this case. */
5721 /* If the result values are STORE_FLAG_VALUE and zero, we can
5722 just make the comparison operation. */
5733 /* Likewise, we can make the negate of a comparison operation
5734 if the result values are - STORE_FLAG_VALUE and zero. */
5742 mode);
5753 mode);
5754 else
5757 mode,
5758 VOIDmode,
5759 cond,
5760 cop1),
5765 }
5766 }
5767 }
5769 /* First see if we can apply the inverse distributive law. */
5772 {
5776 }
5778 /* If CODE is an associative operation not otherwise handled, see if we
5779 can associate some operands. This can win if they are constants or
5780 if they are logically related (i.e. (a & b) & a). */
5786 {
5788 {
5792 rtx inner;
5794 /* Make sure we pass the constant operand if any as the second
5795 one if this is a commutative operation. */
5803 /* For commutative operations, try the other pair if that one
5804 didn't simplify. */
5806 {
5811 }
5815 }
5816 }
5818 /* A little bit of algebraic simplification here. */
5820 {
5822 /* Ensure that our address has any ASHIFTs converted to MULT in case
5823 address-recognizing predicates are called later. */
5832 /* See if this can be moved to simplify_subreg. */
5835 /* Don't call gen_lowpart if the inner mode
5836 is VOIDmode and we cannot simplify it, as SUBREG without
5837 inner mode is invalid. */
5844 {
5845 rtx temp;
5851 /* If op is known to have all lower bits zero, the result is zero. */
5864 }
5866 /* Don't change the mode of the MEM if that would change the meaning
5867 of the address. */
5874 /* Note that we cannot do any narrowing for non-constants since
5875 we might have been counting on using the fact that some bits were
5876 zero. We now do this in the SET. */
5883 /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
5884 replaced by (lshiftrt X C). This will convert
5885 (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
5893 /* If X has only a single bit that might be nonzero, say, bit I, convert
5894 (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
5895 MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
5896 (sign_extract X 1 Y). But only do this if TEMP isn't a register
5897 or a SUBREG of one since we'd be making the expression more
5898 complex if it was just a register. */
5905 {
5906 rtx temp1 = simplify_shift_const
5912 /* If all we did was surround TEMP with the two shifts, we
5913 haven't improved anything, so don't use it. Otherwise,
5914 we are better off with TEMP1. */
5919 }
5923 /* We can't handle truncation to a partial integer mode here
5924 because we don't know the real bitsize of the partial
5925 integer mode. */
5934 /* We can truncate a constant value and return it. */
5938 /* Similarly to what we do in simplify-rtx.c, a truncate of a register
5939 whose value is a comparison can be replaced with a subreg if
5940 STORE_FLAG_VALUE permits. */
5949 /* (const (const X)) can become (const X). Do it this way rather than
5950 returning the inner CONST since CONST can be shared with a
5951 REG_EQUAL note. */
5957 /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
5958 can add in an offset. find_split_point will split this address up
5959 again if it doesn't match. */
5966 /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
5967 when c is (const_int (pow2 + 1) / 2) is a sign extension of a
5968 bit-field and can be replaced by either a sign_extend or a
5969 sign_extract. The `and' may be a zero_extend and the two
5970 <c>, -<c> constants may be reversed. */
5986 return simplify_shift_const
5993 /* If only the low-order bit of X is possibly nonzero, (plus x -1)
5994 can become (ashiftrt (ashift (xor x 1) C) C) where C is
5995 the bitsize of the mode - 1. This allows simplification of
5996 "a = (b & 8) == 0;" */
6003 return simplify_shift_const
6007 const1_rtx),
6011 /* If we are adding two things that have no bits in common, convert
6012 the addition into an IOR. This will often be further simplified,
6013 for example in cases like ((a & 1) + (a & 2)), which can
6014 become a & 3. */
6019 {
6020 /* Try to simplify the expression further. */
6024 /* If we could, great. If not, do not go ahead with the IOR
6025 replacement, since PLUS appears in many special purpose
6026 address arithmetic instructions. */
6034 }
6036 /* Canonicalize x + x into x << 1. */
6045 /* (minus <foo> (and <foo> (const_int -pow2))) becomes
6046 (and <foo> (const_int pow2-1)) */
6057 /* If we have (mult (plus A B) C), apply the distributive law and then
6058 the inverse distributive law to see if things simplify. This
6059 occurs mostly in addresses, often when unrolling loops. */
6062 {
6066 }
6068 /* Try simplify a*(b/c) as (a*b)/c. */
6071 {
6077 }
6081 /* If this is a divide by a power of two, treat it as a shift if
6082 its first operand is a shift. */
6102 /* If the first operand is a condition code, we can't do anything
6103 with it. */
6107 {
6115 /* Simplify our comparison, if possible. */
6118 /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
6119 if only the low-order bit is possibly nonzero in X (such as when
6120 X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
6121 (xor X 1) or (minus 1 X); we use the former. Finally, if X is
6122 known to be either 0 or -1, NE becomes a NEG and EQ becomes
6123 (plus X 1).
6125 Remove any ZERO_EXTRACT we made when thinking this was a
6126 comparison. It may now be simpler to use, e.g., an AND. If a
6127 ZERO_EXTRACT is indeed appropriate, it will be placed back by
6128 the call to make_compound_operation in the SET case.
6130 Don't apply these optimizations if the caller would
6131 prefer a comparison rather than a value.
6132 E.g., for the condition in an IF_THEN_ELSE most targets need
6133 an explicit comparison. */
6136 ;
6154 {
6158 int_mode);
6159 }
6167 {
6171 const1_rtx);
6172 }
6181 {
6184 }
6186 /* If STORE_FLAG_VALUE is -1, we have cases similar to
6187 those above. */
6189 ;
6206 {
6210 int_mode);
6211 }
6220 {
6224 int_mode);
6225 }
6227 /* If X is 0/1, (eq X 0) is X-1. */
6234 {
6237 }
6239 /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
6240 one bit that might be nonzero, we can convert (ne x 0) to
6241 (ashift x c) where C puts the bit in the sign bit. Remove any
6242 AND with STORE_FLAG_VALUE when we are done, since we are only
6243 going to test the sign bit. */
6251 {
6257 else
6259 }
6261 /* If the code changed, return a whole new comparison.
6262 We also need to avoid using SUBST in cases where
6263 simplify_comparison has widened a comparison with a CONST_INT,
6264 since in that case the wider CONST_INT may fail the sanity
6265 checks in do_SUBST. */
6272 /* Otherwise, keep this operation, but maybe change its operands.
6273 This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
6276 }
6286 /* If we are processing SET_DEST, we are done. */
6304 /* If this is a shift by a constant amount, simplify it. */
6312 (HOST_WIDE_INT_1U
6321 }
6324 }
6325 \f
6326 /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6328 static rtx
6330 {
6337 rtx temp;
6340 rtx reversed;
6343 /* Simplify storing of the truth value. */
6348 /* Also when the truth value has to be reversed. */
6354 /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6355 in it is being compared against certain values. Get the true and false
6356 comparisons and see if that says anything about the value of each arm. */
6362 {
6363 HOST_WIDE_INT nzb;
6369 /* If FALSE_CODE is EQ, swap the codes and arms. */
6372 {
6375 }
6377 scalar_int_mode from_mode;
6379 {
6380 /* If we are comparing against zero and the expression being
6381 tested has only a single bit that might be nonzero, that is
6382 its value when it is not equal to zero. Similarly if it is
6383 known to be -1 or 0. */
6387 {
6390 }
6395 {
6398 }
6399 }
6401 /* Now simplify an arm if we know the value of the register in the
6402 branch and it is used in the arm. Be careful due to the potential
6403 of locally-shared RTL. */
6420 }
6422 /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6423 reversed, do so to avoid needing two sets of patterns for
6424 subtract-and-branch insns. Similarly if we have a constant in the true
6425 arm, the false arm is the same as the first operand of the comparison, or
6426 the false arm is more complicated than the true arm. */
6439 {
6448 /* It is possible that the conditional has been simplified out. */
6451 }
6453 /* If the two arms are identical, we don't need the comparison. */
6458 /* Convert a == b ? b : a to "a". */
6470 /* Look for cases where we have (abs x) or (neg (abs X)). */
6473 && comparison_p
6480 {
6486 return
6489 mode);
6492 }
6494 /* Look for MIN or MAX. */
6497 && comparison_p
6502 {
6517 }
6519 /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6520 second operand is zero, this can be done as (OP Z (mult COND C2)) where
6521 C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6522 SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6523 We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6524 neither 1 or -1, but it isn't worth checking for. */
6527 && comparison_p
6530 {
6546 /* If an identity-zero op is commutative, check whether there
6547 would be a match if we swapped the operands. */
6568 {
6572 }
6585 {
6589 }
6606 {
6610 }
6623 {
6627 }
6630 {
6636 const_true_rtx));
6644 }
6645 }
6647 /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6648 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6649 negation of a single bit, we can convert this operation to a shift. We
6650 can actually do this more generally, but it doesn't seem worth it. */
6662 return
6666 /* (IF_THEN_ELSE (NE A 0) C1 0) is A or a zero-extend of A if the only
6667 non-zero bit in A is C1. */
6675 {
6681 }
6684 }
6685 \f
6686 /* Simplify X, a SET expression. Return the new expression. */
6688 static rtx
6690 {
6693 machine_mode mode
6697 scalar_int_mode int_mode;
6699 /* (set (pc) (return)) gets written as (return). */
6703 /* Now that we know for sure which bits of SRC we are using, see if we can
6704 simplify the expression for the object knowing that we only need the
6705 low-order bits. */
6708 {
6711 }
6713 /* If we are setting CC0 or if the source is a COMPARE, look for the use of
6714 the comparison result and try to simplify it unless we already have used
6715 undobuf.other_insn. */
6723 {
6732 {
6735 {
6738 }
6739 }
6740 else
6748 {
6752 }
6753 else
6754 {
6759 /* Attempt to simplify CC user. */
6761 {
6765 }
6767 /* Convert X into a no-op move. */
6771 }
6773 /* Simplify our comparison, if possible. */
6776 #ifdef SELECT_CC_MODE
6777 /* If this machine has CC modes other than CCmode, check to see if we
6778 need to use a different CC mode here. */
6787 else
6790 /* If the mode changed, we have to change SET_DEST, the mode in the
6791 compare, and the mode in the place SET_DEST is used. If SET_DEST is
6792 a hard register, just build new versions with the proper mode. If it
6793 is a pseudo, we lose unless it is only time we set the pseudo, in
6794 which case we can safely change its mode. */
6796 {
6798 {
6800 rtx new_dest;
6804 else
6805 {
6808 }
6815 }
6816 }
6819 /* If the code changed, we have to build a new comparison in
6820 undobuf.other_insn. */
6822 {
6831 /* If the only change we made was to change an EQ into an NE or
6832 vice versa, OP0 has only one bit that might be nonzero, and OP1
6833 is zero, check if changing the user of the condition code will
6834 produce a valid insn. If it won't, we can keep the original code
6835 in that insn by surrounding our operation with an XOR. */
6842 {
6847 {
6854 }
6855 }
6856 }
6861 /* Don't generate a compare of a CC with 0, just use that CC. */
6863 {
6866 }
6867 /* Otherwise, if we didn't previously have the same COMPARE we
6868 want, create it from scratch. */
6871 {
6874 }
6875 }
6876 else
6877 {
6878 /* Get SET_SRC in a form where we have placed back any
6879 compound expressions. Then do the checks below. */
6882 }
6884 /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
6885 and X being a REG or (subreg (reg)), we may be able to convert this to
6886 (set (subreg:m2 x) (op)).
6888 We can always do this if M1 is narrower than M2 because that means that
6889 we only care about the low bits of the result.
6891 However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
6892 perform a narrower operation than requested since the high-order bits will
6893 be undefined. On machine where it is defined, this transformation is safe
6894 as long as M1 and M2 have the same number of words. */
6910 {
6913 dest));
6917 }
6919 /* If we have (set (cc0) (subreg ...)), we try to remove the subreg
6920 in SRC. */
6924 {
6928 /* Here we make sure that we don't have a sign bit on. */
6931 {
6934 }
6935 }
6937 /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
6938 would require a paradoxical subreg. Replace the subreg with a
6939 zero_extend to avoid the reload that would otherwise be required. */
6945 {
6950 }
6952 /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
6953 are comparing an item known to be 0 or -1 against 0, use a logical
6954 operation instead. Check for one of the arms being an IOR of the other
6955 arm with some value. We compute three terms to be IOR'ed together. In
6956 practice, at most two will be nonzero. Then we do the IOR's. */
6964 && (!HAVE_conditional_move
6969 {
6994 int_mode),
6995 false_rtx);
7001 term3));
7004 }
7006 /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
7007 whole thing fail. */
7012 else
7013 /* Convert this into a field assignment operation, if possible. */
7015 }
7016 \f
7017 /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
7018 result. */
7020 static rtx
7022 {
7025 scalar_int_mode mode;
7028 {
7030 /* We can call simplify_and_const_int only if we don't lose
7031 any (sign) bits when converting INTVAL (op1) to
7032 "unsigned HOST_WIDE_INT". */
7037 {
7044 }
7046 /* If we have any of (and (ior A B) C) or (and (xor A B) C),
7047 apply the distributive law and then the inverse distributive
7048 law to see if things simplify. */
7050 {
7054 }
7056 {
7060 }
7064 /* If we have (ior (and A B) C), apply the distributive law and then
7065 the inverse distributive law to see if things simplify. */
7068 {
7072 }
7075 {
7079 }
7084 }
7087 }
7088 \f
7089 /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
7090 operations" because they can be replaced with two more basic operations.
7091 ZERO_EXTEND is also considered "compound" because it can be replaced with
7092 an AND operation, which is simpler, though only one operation.
7094 The function expand_compound_operation is called with an rtx expression
7095 and will convert it to the appropriate shifts and AND operations,
7096 simplifying at each stage.
7098 The function make_compound_operation is called to convert an expression
7099 consisting of shifts and ANDs into the equivalent compound expression.
7100 It is the inverse of this function, loosely speaking. */
7102 static rtx
7104 {
7108 rtx tem;
7109 scalar_int_mode inner_mode;
7112 {
7115 /* FALLTHRU */
7117 /* We can't necessarily use a const_int for a multiword mode;
7118 it depends on implicitly extending the value.
7119 Since we don't know the right way to extend it,
7120 we can't tell whether the implicit way is right.
7122 Even for a mode that is no wider than a const_int,
7123 we can't win, because we need to sign extend one of its bits through
7124 the rest of it, and we don't know which bit. */
7128 /* Reject modes that aren't scalar integers because turning vector
7129 or complex modes into shifts causes problems. */
7133 /* Return if (subreg:MODE FROM 0) is not a safe replacement for
7134 (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
7135 because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
7136 reloaded. If not for that, MEM's would very rarely be safe.
7138 Reject modes bigger than a word, because we might not be able
7139 to reference a two-register group starting with an arbitrary register
7140 (and currently gen_lowpart might crash for a SUBREG). */
7146 /* If the inner object has VOIDmode (the only way this can happen
7147 is if it is an ASM_OPERANDS), we can't do anything since we don't
7148 know how much masking to do. */
7157 /* fall through */
7160 /* If the operand is a CLOBBER, just return it. */
7168 /* Reject modes that aren't scalar integers because turning vector
7169 or complex modes into shifts causes problems. */
7176 /* This should stay within the object being extracted, fail otherwise. */
7187 }
7189 /* We've rejected non-scalar operations by now. */
7192 /* Convert sign extension to zero extension, if we know that the high
7193 bit is not set, as this is easier to optimize. It will be converted
7194 back to cheaper alternative in make_extraction. */
7200 {
7204 /* Make sure this is a profitable operation. */
7211 else
7213 }
7215 /* We can optimize some special cases of ZERO_EXTEND. */
7217 {
7218 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
7219 know that the last value didn't have any inappropriate bits
7220 set. */
7228 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7237 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
7238 is a comparison and STORE_FLAG_VALUE permits. This is like
7239 the first case, but it works even when MODE is larger
7240 than HOST_WIDE_INT. */
7248 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7257 }
7259 /* If we reach here, we want to return a pair of shifts. The inner
7260 shift is a left shift of BITSIZE - POS - LEN bits. The outer
7261 shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
7262 logical depending on the value of UNSIGNEDP.
7264 If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
7265 converted into an AND of a shift.
7267 We must check for the case where the left shift would have a negative
7268 count. This can happen in a case like (x >> 31) & 255 on machines
7269 that can't shift by a constant. On those machines, we would first
7270 combine the shift with the AND to produce a variable-position
7271 extraction. Then the constant of 31 would be substituted in
7272 to produce such a position. */
7276 {
7284 }
7289 pos),
7291 else
7292 /* Any other cases we can't handle. */
7295 /* If we couldn't do this for some reason, return the original
7296 expression. */
7301 }
7302 \f
7303 /* X is a SET which contains an assignment of one object into
7304 a part of another (such as a bit-field assignment, STRICT_LOW_PART,
7305 or certain SUBREGS). If possible, convert it into a series of
7306 logical operations.
7308 We half-heartedly support variable positions, but do not at all
7309 support variable lengths. */
7311 static const_rtx
7313 {
7314 rtx inner;
7318 scalar_int_mode compute_mode;
7320 /* Loop until we find something we can't simplify. */
7322 {
7325 {
7329 }
7332 {
7337 /* A constant position should stay within the width of INNER. */
7343 {
7351 /* If position is ADJUST - X, new position is X. */
7353 else
7354 {
7359 pos);
7360 }
7361 }
7362 }
7364 /* A SUBREG between two modes that occupy the same numbers of words
7365 can be done by moving the SUBREG to the source. */
7367 /* We need SUBREGs to compute nonzero_bits properly. */
7368 && nonzero_sign_valid
7373 {
7375 gen_lowpart
7379 }
7380 else
7386 /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7388 {
7389 /* Don't do anything for vector or complex integral types. */
7393 /* Try to find an integral mode to pun with. */
7399 }
7401 /* Compute a mask of LEN bits, if we can do this on the host machine. */
7405 /* Don't try to compute in too wide unsupported modes. */
7409 /* Now compute the equivalent expression. Make a copy of INNER
7410 for the SET_DEST in case it is a MEM into which we will substitute;
7411 we don't want shared RTL in that case. */
7413 compute_mode);
7417 compute_mode,
7419 compute_mode),
7420 inner);
7425 mask),
7426 pos);
7431 }
7434 }
7435 \f
7436 /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7437 it is an RTX that represents the (variable) starting position; otherwise,
7438 POS is the (constant) starting bit position. Both are counted from the LSB.
7440 UNSIGNEDP is nonzero for an unsigned reference and zero for a signed one.
7442 IN_DEST is nonzero if this is a reference in the destination of a SET.
7443 This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7444 a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7445 be used.
7447 IN_COMPARE is nonzero if we are in a COMPARE. This means that a
7448 ZERO_EXTRACT should be built even for bits starting at bit 0.
7450 MODE is the desired mode of the result (if IN_DEST == 0).
7452 The result is an RTX for the extraction or NULL_RTX if the target
7453 can't handle it. */
7455 static rtx
7459 {
7460 /* This mode describes the size of the storage area
7461 to fetch the overall value from. Within that, we
7462 ignore the POS lowest bits, etc. */
7464 machine_mode inner_mode;
7465 scalar_int_mode wanted_inner_mode;
7471 HOST_WIDE_INT orig_pos;
7479 /* If trying or potentionally trying to extract
7480 bits outside of is_mode, don't look through
7481 non-paradoxical SUBREGs. See PR82192. */
7484 {
7485 /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7486 consider just the QI as the memory to extract from.
7487 The subreg adds or removes high bits; its mode is
7488 irrelevant to the meaning of this extraction,
7489 since POS and LEN count from the lsb. */
7493 }
7498 {
7499 /* We're extracting the least significant bits of an rtx
7500 (ashift X (const_int C)), where LEN > C. Extract the
7501 least significant (LEN - C) bits of X, giving an rtx
7502 whose mode is MODE, then shift it left C times. */
7508 }
7510 /* If trying or potentionally trying to extract
7511 bits outside of is_mode, don't look through
7512 TRUNCATE. See PR82192. */
7519 /* See if this can be done without an extraction. We never can if the
7520 width of the field is not the same as that of some integer mode. For
7521 registers, we can only avoid the extraction if the position is at the
7522 low-order bit and this is either not in the destination or we have the
7523 appropriate STRICT_LOW_PART operation available.
7525 For MEM, we can avoid an extract if the field starts on an appropriate
7526 boundary and we can change the mode of the memory reference. */
7528 scalar_int_mode tmode;
7537 && (! in_dest
7541 && (pos
7544 /* We can't do this if we are widening INNER_MODE (it
7545 may not be aligned, for one thing). */
7551 {
7552 /* If INNER is a MEM, make a new MEM that encompasses just the desired
7553 field. If the original and current mode are the same, we need not
7554 adjust the offset. Otherwise, we do if bytes big endian.
7556 If INNER is not a MEM, get a piece consisting of just the field
7557 of interest (in this case POS % BITS_PER_WORD must be 0). */
7560 {
7561 HOST_WIDE_INT offset;
7563 /* POS counts from lsb, but make OFFSET count in memory order. */
7566 else
7570 }
7572 {
7574 {
7575 /* We can't call gen_lowpart in a DEST since we
7576 always want a SUBREG (see below) and it would sometimes
7577 return a new hard register. */
7579 {
7580 unsigned int offset
7583 /* Avoid creating invalid subregs, for example when
7584 simplifying (x>>32)&255. */
7589 }
7590 else
7592 }
7593 else
7595 }
7596 else
7598 len >= HOST_BITS_PER_WIDE_INT
7599 ? HOST_WIDE_INT_M1U
7602 /* If this extraction is going into the destination of a SET,
7603 make a STRICT_LOW_PART unless we made a MEM. */
7618 /* If we know that no extraneous bits are set, and that the high
7619 bit is not set, convert the extraction to the cheaper of
7620 sign and zero extension, that are equivalent in these cases. */
7626 {
7630 /* Prefer ZERO_EXTENSION, since it gives more information to
7631 backends. */
7636 }
7638 /* Otherwise, sign- or zero-extend unless we already are in the
7639 proper mode. */
7643 }
7645 /* Unless this is a COMPARE or we have a funny memory reference,
7646 don't do anything with zero-extending field extracts starting at
7647 the low-order bit since they are simple AND operations. */
7652 /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7653 if the position is not a constant and the length is not 1. In all
7654 other cases, we would only be going outside our object in cases when
7655 an original shift would have been undefined. */
7664 /* If INNER is not from memory, we want it to have the mode of a register
7665 extraction pattern's structure operand, or word_mode if there is no
7666 such pattern. The same applies to extraction_mode and pos_mode
7667 and their respective operands.
7669 For memory, assume that the desired extraction_mode and pos_mode
7670 are the same as for a register operation, since at present we don't
7671 have named patterns for aligned memory structures. */
7675 {
7679 }
7681 /* Never narrow an object, since that might not be safe. */
7689 else
7690 {
7691 /* Be careful not to go beyond the extracted object and maintain the
7692 natural alignment of the memory. */
7697 }
7702 {
7703 /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7704 BITS_BIG_ENDIAN style. If position is constant, compute new
7705 position. Otherwise, build subtraction.
7706 Note that POS is relative to the mode of the original argument.
7707 If it's a MEM we need to recompute POS relative to that.
7708 However, if we're extracting from (or inserting into) a register,
7709 we want to recompute POS relative to wanted_inner_mode. */
7716 else
7717 pos_rtx
7720 pos_rtx);
7721 /* POS may be less than 0 now, but we check for that below.
7722 Note that it can only be less than 0 if !MEM_P (inner). */
7723 }
7725 /* If INNER has a wider mode, and this is a constant extraction, try to
7726 make it smaller and adjust the byte to point to the byte containing
7727 the value. */
7730 && ! pos_rtx
7735 {
7738 /* The computations below will be correct if the machine is big
7739 endian in both bits and bytes or little endian in bits and bytes.
7740 If it is mixed, we must adjust. */
7742 /* If bytes are big endian and we had a paradoxical SUBREG, we must
7743 adjust OFFSET to compensate. */
7748 /* We can now move to the desired byte. */
7759 }
7761 /* If INNER is not memory, get it into the proper mode. If we are changing
7762 its mode, POS must be a constant and smaller than the size of the new
7763 mode. */
7765 {
7766 /* On the LHS, don't create paradoxical subregs implicitely truncating
7767 the register unless TARGET_TRULY_NOOP_TRUNCATION. */
7770 wanted_inner_mode))
7782 pos_rtx
7784 ? HOST_WIDE_INT_M1U
7788 }
7790 /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7791 have to zero extend. Otherwise, we can just use a SUBREG.
7793 We dealt with constant rtxes earlier, so pos_rtx cannot
7794 have VOIDmode at this point. */
7798 {
7802 /* If we know that no extraneous bits are set, and that the high
7803 bit is not set, convert extraction to cheaper one - either
7804 SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7805 cases. */
7813 {
7817 /* Prefer ZERO_EXTENSION, since it gives more information to
7818 backends. */
7822 }
7824 }
7826 /* Make POS_RTX unless we already have it and it is correct. If we don't
7827 have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7828 be a CONST_INT. */
7835 /* Make the required operation. See if we can use existing rtx. */
7842 }
7843 \f
7844 /* See if X (of mode MODE) contains an ASHIFT of COUNT or more bits that
7845 can be commuted with any other operations in X. Return X without
7846 that shift if so. */
7848 static rtx
7850 {
7852 rtx tem;
7855 {
7857 /* This is the shift itself. If it is wide enough, we will return
7858 either the value being shifted if the shift count is equal to
7859 COUNT or a shift for the difference. */
7873 /* If we can safely shift this constant and we find the inner shift,
7874 make a new operation. */
7879 {
7883 }
7888 }
7891 }
7892 \f
7893 /* Subroutine of make_compound_operation. *X_PTR is the rtx at the current
7894 level of the expression and MODE is its mode. IN_CODE is as for
7895 make_compound_operation. *NEXT_CODE_PTR is the value of IN_CODE
7896 that should be used when recursing on operands of *X_PTR.
7898 There are two possible actions:
7900 - Return null. This tells the caller to recurse on *X_PTR with IN_CODE
7901 equal to *NEXT_CODE_PTR, after which *X_PTR holds the final value.
7903 - Return a new rtx, which the caller returns directly. */
7905 static rtx
7909 {
7917 rtx tem;
7918 scalar_int_mode inner_mode;
7922 {
7925 }
7927 /* Process depending on the code of this operation. If NEW is set
7928 nonzero, it will be returned. */
7931 {
7933 /* Convert shifts by constants into multiplications if inside
7934 an address. */
7938 {
7944 {
7947 }
7950 }
7959 {
7963 }
7966 {
7970 mode));
7972 }
7973 else
7974 {
7977 }
7987 {
7991 }
7994 {
7998 mode));
8000 }
8001 else
8002 {
8006 }
8009 /* If the second operand is not a constant, we can't do anything
8010 with it. */
8014 /* If the constant is a power of two minus one and the first operand
8015 is a logical right shift, make an extraction. */
8018 {
8022 }
8024 /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
8031 {
8038 /* If we narrowed the mode when dropping the subreg, then we lose. */
8042 /* If that didn't give anything, see if the AND simplifies on
8043 its own. */
8045 {
8049 }
8050 }
8051 /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
8057 {
8058 /* Apply the distributive law, and then try to make extractions. */
8065 }
8067 /* If we are have (and (rotate X C) M) and C is larger than the number
8068 of bits in M, this is an extraction. */
8074 {
8080 }
8082 /* On machines without logical shifts, if the operand of the AND is
8083 a logical shift and our mask turns off all the propagated sign
8084 bits, we can replace the logical shift with an arithmetic shift. */
8092 {
8102 next_code),
8104 }
8106 /* If the constant is one less than a power of two, this might be
8107 representable by an extraction even if no shift is present.
8108 If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
8109 we are in a COMPARE. */
8113 next_code),
8116 /* If we are in a comparison and this is an AND with a power of two,
8117 convert this into the appropriate bit extract. */
8123 next_code),
8126 /* If the one operand is a paradoxical subreg of a register or memory and
8127 the constant (limited to the smaller mode) has only zero bits where
8128 the sub expression has known zero bits, this can be expressed as
8129 a zero_extend. */
8131 {
8132 rtx sub;
8138 {
8142 /* original AND constant with all the known zero bits set */
8145 {
8150 }
8151 }
8152 }
8157 /* If the sign bit is known to be zero, replace this with an
8158 arithmetic shift. */
8163 {
8166 next_code),
8169 }
8171 /* fall through */
8177 /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
8178 this is a SIGN_EXTRACT. */
8185 {
8192 }
8194 /* See if we have operations between an ASHIFTRT and an ASHIFT.
8195 If so, try to merge the shifts into a SIGN_EXTEND. We could
8196 also do this for some cases of SIGN_EXTRACT, but it doesn't
8197 seem worth the effort; the case checked for occurs on Alpha. */
8208 next_code),
8215 /* Call ourselves recursively on the inner expression. If we are
8216 narrowing the object and it has a different RTL code from
8217 what it originally did, do this SUBREG as a force_to_mode. */
8218 {
8222 /* If the SUBREG is masking of a logical right shift,
8223 make an extraction. */
8230 {
8239 }
8241 /* If in_code is COMPARE, it isn't always safe to pass it through
8242 to the recursive make_compound_operation call. */
8246 /* (subreg:SI (and:DI (reg:DI) (const_int 0x800000000)) 0)
8247 is (const_int 0), rather than
8248 (subreg:SI (lshiftrt:DI (reg:DI) (const_int 35)) 0).
8249 Similarly (subreg:QI (and:SI (reg:SI) (const_int 0x80)) 0)
8250 for non-equality comparisons against 0 is not equivalent
8251 to (subreg:QI (lshiftrt:SI (reg:SI) (const_int 7)) 0). */
8261 simplified
8269 {
8270 rtx newer
8273 /* If we have something other than a SUBREG, we might have
8274 done an expansion, so rerun ourselves. */
8278 /* force_to_mode can expand compounds. If it just re-expanded
8279 the compound, use gen_lowpart to convert to the desired
8280 mode. */
8282 /* Likewise if it re-expanded the compound only partially.
8283 This happens for SUBREG of ZERO_EXTRACT if they extract
8284 the same number of bits. */
8293 }
8297 }
8302 }
8308 }
8310 /* Look at the expression rooted at X. Look for expressions
8311 equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
8312 Form these expressions.
8314 Return the new rtx, usually just X.
8316 Also, for machines like the VAX that don't have logical shift insns,
8317 try to convert logical to arithmetic shift operations in cases where
8318 they are equivalent. This undoes the canonicalizations to logical
8319 shifts done elsewhere.
8321 We try, as much as possible, to re-use rtl expressions to save memory.
8323 IN_CODE says what kind of expression we are processing. Normally, it is
8324 SET. In a memory address it is MEM. When processing the arguments of
8325 a comparison or a COMPARE against zero, it is COMPARE, or EQ if more
8326 precisely it is an equality comparison against zero. */
8328 rtx
8330 {
8337 /* Select the code to be used in recursive calls. Once we are inside an
8338 address, we stay there. If we have a comparison, set to COMPARE,
8339 but once inside, go back to our default of SET. */
8346 scalar_int_mode mode;
8348 {
8354 }
8356 /* Now recursively process each operand of this operation. We need to
8357 handle ZERO_EXTEND specially so that we don't lose track of the
8358 inner mode. */
8360 {
8368 }
8373 {
8376 }
8379 {
8382 }
8386 }
8387 \f
8388 /* Given M see if it is a value that would select a field of bits
8389 within an item, but not the entire word. Return -1 if not.
8390 Otherwise, return the starting position of the field, where 0 is the
8391 low-order bit.
8393 *PLEN is set to the length of the field. */
8395 static int
8397 {
8398 /* Get the bit number of the first 1 bit from the right, -1 if none. */
8403 /* Now shift off the low-order zero bits and see if we have a
8404 power of two minus 1. */
8412 }
8413 \f
8414 /* If X refers to a register that equals REG in value, replace these
8415 references with REG. */
8416 static rtx
8418 {
8426 {
8459 /* FALLTHRU */
8463 {
8467 else
8469 }
8471 /* fall through */
8478 {
8481 {
8483 {
8486 }
8488 }
8489 }
8491 {
8494 {
8497 {
8499 {
8502 }
8504 }
8505 }
8506 }
8509 }
8512 }
8514 /* Return X converted to MODE. If the value is already truncated to
8515 MODE we can just return a subreg even though in the general case we
8516 would need an explicit truncation. */
8518 static rtx
8520 {
8525 {
8526 /* Bit-cast X into an integer mode. */
8531 }
8534 }
8536 /* See if X can be simplified knowing that we will only refer to it in
8537 MODE and will only refer to those bits that are nonzero in MASK.
8538 If other bits are being computed or if masking operations are done
8539 that select a superset of the bits in MASK, they can sometimes be
8540 ignored.
8542 Return a possibly simplified expression, but always convert X to
8543 MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8545 If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
8546 are all off in X. This is used when X will be complemented, by either
8547 NOT, NEG, or XOR. */
8549 static rtx
8552 {
8555 machine_mode op_mode;
8558 /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8559 code below will do the wrong thing since the mode of such an
8560 expression is VOIDmode.
8562 Also do nothing if X is a CLOBBER; this can happen if X was
8563 the return value from a call to gen_lowpart. */
8567 /* We want to perform the operation in its present mode unless we know
8568 that the operation is valid in MODE, in which case we do the operation
8569 in MODE. */
8574 /* It is not valid to do a right-shift in a narrower mode
8575 than the one it came in with. */
8580 /* Truncate MASK to fit OP_MODE. */
8584 /* Determine what bits of X are guaranteed to be (non)zero. */
8587 /* If none of the bits in X are needed, return a zero. */
8591 /* If X is a CONST_INT, return a new one. Do this here since the
8592 test below will fail. */
8594 {
8597 else
8598 {
8601 }
8602 }
8604 /* If X is narrower than MODE and we want all the bits in X's mode, just
8605 get X in the proper mode. */
8610 /* We can ignore the effect of a SUBREG if it narrows the mode or
8611 if the constant masks to zero all the bits the mode doesn't have. */
8623 /* OP_MODE is either MODE or XMODE, so it must be a scalar
8624 integer too. */
8630 }
8632 /* Subroutine of force_to_mode that handles cases in which both X and
8633 the result are scalar integers. MODE is the mode of the result,
8634 XMODE is the mode of X, and OP_MODE says which of MODE or XMODE
8635 is preferred for simplified versions of X. The other arguments
8636 are as for force_to_mode. */
8638 static rtx
8642 {
8648 /* When we have an arithmetic operation, or a shift whose count we
8649 do not know, we need to assume that all bits up to the highest-order
8650 bit in MASK will be needed. This is how we form such a mask. */
8653 else
8658 {
8660 /* If X is a (clobber (const_int)), return it since we know we are
8661 generating something that won't match. */
8674 /* Similarly for a truncate. */
8678 /* If this is an AND with a constant, convert it into an AND
8679 whose constant is the AND of that constant with MASK. If it
8680 remains an AND of MASK, delete it since it is redundant. */
8683 {
8688 /* If X is still an AND, see if it is an AND with a mask that
8689 is just some low-order bits. If so, and it is MASK, we don't
8690 need it. */
8696 /* If it remains an AND, try making another AND with the bits
8697 in the mode mask that aren't in MASK turned on. If the
8698 constant in the AND is wide enough, this might make a
8699 cheaper constant. */
8704 {
8705 unsigned HOST_WIDE_INT cval
8707 rtx y;
8714 }
8717 }
8722 /* In (and (plus FOO C1) M), if M is a mask that just turns off
8723 low-order bits (as in an alignment operation) and FOO is already
8724 aligned to that boundary, mask C1 to that boundary as well.
8725 This may eliminate that PLUS and, later, the AND. */
8727 {
8731 /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8732 number, sign extend it. */
8745 }
8747 /* fall through */
8750 /* Substituting into the operands of a widening MULT is not likely to
8751 create RTL matching a machine insn. */
8761 /* For PLUS, MINUS and MULT, we need any bits less significant than the
8762 most significant bit in MASK since carries from those bits will
8763 affect the bits we are interested in. */
8768 /* If X is (minus C Y) where C's least set bit is larger than any bit
8769 in the mask, then we may replace with (neg Y). */
8772 {
8775 }
8777 /* Similarly, if C contains every bit in the fuller_mask, then we may
8778 replace with (not Y). */
8781 {
8784 }
8791 /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8792 LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8793 operation which may be a bitfield extraction. Ensure that the
8794 constant we form is not wider than the mode of X. */
8806 {
8809 xmode);
8815 }
8817 binop:
8818 /* For most binary operations, just propagate into the operation and
8819 change the mode if we have an operation of that mode. */
8824 /* If we ended up truncating both operands, truncate the result of the
8825 operation instead. */
8828 {
8831 }
8837 {
8840 }
8844 /* For left shifts, do the same, but just for the first operand.
8845 However, we cannot do anything with shifts where we cannot
8846 guarantee that the counts are smaller than the size of the mode
8847 because such a count will have a different meaning in a
8848 wider mode. */
8858 /* If the shift count is a constant and we can do arithmetic in
8859 the mode of the shift, refine which bits we need. Otherwise, use the
8860 conservative form of the mask. */
8866 else
8874 {
8877 }
8881 /* Here we can only do something if the shift count is a constant,
8882 this shift constant is valid for the host, and we can do arithmetic
8883 in OP_MODE. */
8889 {
8893 /* Select the mask of the bits we need for the shift operand. */
8896 /* We can only change the mode of the shift if we can do arithmetic
8897 in the mode of the shift and INNER_MASK is no wider than the
8898 width of X's mode. */
8905 {
8908 }
8909 }
8911 /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
8912 shift and AND produces only copies of the sign bit (C2 is one less
8913 than a power of two), we can do this with just a shift. */
8917 /* The shift puts one of the sign bit copies in the least significant
8918 bit. */
8923 /* Number of bits left after the shift must be more than the mask
8924 needs. */
8927 /* Must be more sign bit copies than the mask needs. */
8937 /* If we are just looking for the sign bit, we don't need this shift at
8938 all, even if it has a variable count. */
8942 /* If this is a shift by a constant, get a mask that contains those bits
8943 that are not copies of the sign bit. We then have two cases: If
8944 MASK only includes those bits, this can be a logical shift, which may
8945 allow simplifications. If MASK is a single-bit field not within
8946 those bits, we are requesting a copy of the sign bit and hence can
8947 shift the sign bit to the appropriate location. */
8951 {
8955 /* If the considered data is wider than HOST_WIDE_INT, we can't
8956 represent a mask for all its bits in a single scalar.
8957 But we only care about the lower bits, so calculate these. */
8960 {
8963 /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
8964 is the number of bits a full-width mask would have set.
8965 We need only shift if these are fewer than nonzero can
8966 hold. If not, we must keep all bits set in nonzero. */
8971 + HOST_BITS_PER_WIDE_INT
8973 }
8974 else
8975 {
8978 }
8981 {
8986 }
8989 {
8990 x = simplify_shift_const
8996 }
8997 }
8999 /* If MASK is 1, convert this to an LSHIFTRT. This can be done
9000 even if the shift count isn't a constant. */
9004 shiftrt:
9006 /* If this is a zero- or sign-extension operation that just affects bits
9007 we don't care about, remove it. Be sure the call above returned
9008 something that is still a shift. */
9018 next_select);
9024 /* If the shift count is constant and we can do computations
9025 in the mode of X, compute where the bits we care about are.
9026 Otherwise, we can't do anything. Don't change the mode of
9027 the shift or propagate MODE into the shift, though. */
9030 {
9039 }
9043 /* If we just want the low-order bit, the NEG isn't needed since it
9044 won't change the low-order bit. */
9048 /* We need any bits less significant than the most significant bit in
9049 MASK since carries from those bits will affect the bits we are
9050 interested in. */
9055 /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
9056 same as the XOR case above. Ensure that the constant we form is not
9057 wider than the mode of X. */
9065 {
9072 }
9074 /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
9075 use the full mask inside the NOT. */
9078 unop:
9081 next_select));
9083 {
9086 }
9090 /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
9091 in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
9092 which is equal to STORE_FLAG_VALUE. */
9104 /* We have no way of knowing if the IF_THEN_ELSE can itself be
9105 written in a narrower mode. We play it safe and do not do so. */
9121 }
9123 /* Ensure we return a value of the proper mode. */
9125 }
9126 \f
9127 /* Return nonzero if X is an expression that has one of two values depending on
9128 whether some other value is zero or nonzero. In that case, we return the
9129 value that is being tested, *PTRUE is set to the value if the rtx being
9130 returned has a nonzero value, and *PFALSE is set to the other alternative.
9132 If we return zero, we set *PTRUE and *PFALSE to X. */
9134 static rtx
9136 {
9141 scalar_int_mode int_mode;
9143 /* If we are comparing a value against zero, we are done. */
9146 {
9150 }
9152 /* If this is a unary operation whose operand has one of two values, apply
9153 our opcode to compute those values. */
9156 {
9161 }
9163 /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
9164 make can't possibly match and would suppress other optimizations. */
9166 ;
9168 /* If this is a binary operation, see if either side has only one of two
9169 values. If either one does or if both do and they are conditional on
9170 the same value, compute the new true and false values. */
9172 {
9180 {
9181 /* Try to enable a simplification by undoing work done by
9182 if_then_else_cond if it converted a REG into something more
9183 complex. */
9185 {
9188 }
9189 else
9190 {
9193 }
9194 }
9198 {
9199 /* If if_then_else_cond returned zero, then true/false are the
9200 same rtl. We must copy one of them to prevent invalid rtl
9201 sharing. */
9208 {
9213 }
9214 else
9215 {
9218 }
9221 }
9223 /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
9224 operands is zero when the other is nonzero, and vice-versa,
9225 and STORE_FLAG_VALUE is 1 or -1. */
9231 {
9248 {
9255 const_true_rtx);
9257 }
9258 }
9260 /* Similarly for MULT, AND and UMIN, except that for these the result
9261 is always zero. */
9265 {
9279 {
9282 }
9283 }
9284 }
9287 {
9288 /* If we have IF_THEN_ELSE already, extract the condition and
9289 canonicalize it if it is NE or EQ. */
9295 {
9298 }
9299 else
9301 }
9303 /* If X is a SUBREG, we can narrow both the true and false values
9304 if the inner expression, if there is a condition. */
9308 {
9314 {
9318 }
9319 }
9321 /* If X is a constant, this isn't special and will cause confusions
9322 if we treat it as such. Likewise if it is equivalent to a constant. */
9325 ;
9327 /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
9328 will be least confusing to the rest of the compiler. */
9330 {
9333 }
9335 /* If X is known to be either 0 or -1, those are the true and
9336 false values when testing X. */
9341 {
9344 }
9346 /* Likewise for 0 or a single bit. */
9349 {
9352 }
9354 /* Otherwise fail; show no condition with true and false values the same. */
9357 }
9358 \f
9359 /* Return the value of expression X given the fact that condition COND
9360 is known to be true when applied to REG as its first operand and VAL
9361 as its second. X is known to not be shared and so can be modified in
9362 place.
9364 We only handle the simplest cases, and specifically those cases that
9365 arise with IF_THEN_ELSE expressions. */
9367 static rtx
9369 {
9377 /* If either operand of the condition is a floating point value,
9378 then we have to avoid collapsing an EQ comparison. */
9388 /* If X is (abs REG) and we know something about REG's relationship
9389 with zero, we may be able to simplify this. */
9393 {
9402 }
9404 /* The only other cases we handle are MIN, MAX, and comparisons if the
9405 operands are the same as REG and VAL. */
9408 {
9410 {
9413 }
9416 {
9418 {
9426 else
9428 }
9431 {
9434 /* Do not reverse the condition when it is NE or EQ.
9435 This is because we cannot conclude anything about
9436 the value of 'SMAX (x, y)' when x is not equal to y,
9437 but we can when x equals y. */
9443 {
9454 }
9455 }
9456 }
9457 }
9459 {
9464 {
9465 /* We must simplify subreg here, before we lose track of the
9466 original inner_mode. */
9471 else
9473 }
9476 }
9477 /* We don't have to handle SIGN_EXTEND here, because even in the
9478 case of replacing something with a modeless CONST_INT, a
9479 CONST_INT is already (supposed to be) a valid sign extension for
9480 its narrower mode, which implies it's already properly
9481 sign-extended for the wider mode. Now, for ZERO_EXTEND, the
9482 story is different. */
9484 {
9489 {
9490 /* We must simplify the zero_extend here, before we lose
9491 track of the original inner_mode. */
9496 else
9498 }
9501 }
9505 {
9512 }
9515 }
9516 \f
9517 /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9518 assignment as a field assignment. */
9520 static int
9522 {
9524 {
9529 /* For big endian, adjust the memory offset. */
9534 else
9536 }
9544 /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9545 Note that all SUBREGs of MEM are paradoxical; otherwise they
9546 would have been rewritten. */
9559 /* We used to see if get_last_value of X and Y were the same but that's
9560 not correct. In one direction, we'll cause the assignment to have
9561 the wrong destination and in the case, we'll import a register into this
9562 insn that might have already have been dead. So fail if none of the
9563 above cases are true. */
9565 }
9566 \f
9567 /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9568 Return that assignment if so.
9570 We only handle the most common cases. */
9572 static rtx
9574 {
9577 rtx assign;
9579 HOST_WIDE_INT c1;
9580 HOST_WIDE_INT pos;
9582 rtx other;
9584 /* All the rules in this function are specific to scalar integers. */
9585 scalar_int_mode mode;
9589 /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9590 a clear of a one-bit field. We will have changed it to
9591 (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9592 for a SUBREG. */
9598 {
9604 }
9613 {
9620 }
9622 /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9623 one-bit field. */
9627 {
9633 }
9635 /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9636 SRC is an AND with all bits of that field set, then we can discard
9637 the AND. */
9642 {
9649 else
9652 /* Complete overlap. We can remove the source AND. */
9656 /* Partial overlap. We can reduce the source AND. */
9658 {
9662 }
9663 }
9665 /* The other case we handle is assignments into a constant-position
9666 field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9667 a mask that has all one bits except for a group of zero bits and
9668 OTHER is known to have zeros where C1 has ones, this is such an
9669 assignment. Compute the position and length from C1. Shift OTHER
9670 to the appropriate position, force it to the required mode, and
9671 make the extraction. Check for the AND in both operands. */
9673 /* One or more SUBREGs might obscure the constant-position field
9674 assignment. The first one we are likely to encounter is an outer
9675 narrowing SUBREG, which we can just strip for the purposes of
9676 identifying the constant-field assignment. */
9693 /* The second SUBREG that might get in the way is a paradoxical
9694 SUBREG around the first operand of the AND. We want to
9695 pretend the operand is as wide as the destination here. We
9696 do this by adjusting the MEM to wider mode for the sole
9697 purpose of the call to rtx_equal_for_field_assignment_p. Also
9698 note this trick only works for MEMs. */
9710 /* The second SUBREG that might get in the way is a paradoxical
9711 SUBREG around the first operand of the AND. We want to
9712 pretend the operand is as wide as the destination here. We
9713 do this by adjusting the MEM to wider mode for the sole
9714 purpose of the call to rtx_equal_for_field_assignment_p. Also
9715 note this trick only works for MEMs. */
9723 else
9737 /* The mode to use for the source is the mode of the assignment, or of
9738 what is inside a possible STRICT_LOW_PART. */
9742 /* Shift OTHER right POS places and make it the source, restricting it
9743 to the proper length and mode. */
9747 dest);
9749 len >= HOST_BITS_PER_WIDE_INT
9750 ? HOST_WIDE_INT_M1U
9754 /* If SRC is masked by an AND that does not make a difference in
9755 the value being stored, strip it. */
9766 }
9767 \f
9768 /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9769 if so. */
9771 static rtx
9773 {
9777 rtx tem;
9779 /* Distributivity is not true for floating point as it can change the
9780 value. So we don't do it unless -funsafe-math-optimizations. */
9785 /* The outer operation can only be one of the following: */
9793 /* If either operand is a primitive we can't do anything, so get out
9794 fast. */
9804 /* See if the inner and outer operations distribute. */
9806 {
9811 /* These all distribute except over PLUS. */
9822 /* This is also a multiply, so it distributes over everything. */
9825 /* This used to handle SUBREG, but this turned out to be counter-
9826 productive, since (subreg (op ...)) usually is not handled by
9827 insn patterns, and this "optimization" therefore transformed
9828 recognizable patterns into unrecognizable ones. Therefore the
9829 SUBREG case was removed from here.
9831 It is possible that distributing SUBREG over arithmetic operations
9832 leads to an intermediate result than can then be optimized further,
9833 e.g. by moving the outer SUBREG to the other side of a SET as done
9834 in simplify_set. This seems to have been the original intent of
9835 handling SUBREGs here.
9837 However, with current GCC this does not appear to actually happen,
9838 at least on major platforms. If some case is found where removing
9839 the SUBREG case here prevents follow-on optimizations, distributing
9840 SUBREGs ought to be re-added at that place, e.g. in simplify_set. */
9844 }
9846 /* Set LHS and RHS to the inner operands (A and B in the example
9847 above) and set OTHER to the common operand (C in the example).
9848 There is only one way to do this unless the inner operation is
9849 commutative. */
9861 else
9864 /* Form the new inner operation, seeing if it simplifies first. */
9867 /* There is one exception to the general way of distributing:
9868 (a | c) ^ (b | c) -> (a ^ b) & ~c */
9870 {
9873 }
9875 /* We may be able to continuing distributing the result, so call
9876 ourselves recursively on the inner operation before forming the
9877 outer operation, which we return. */
9880 }
9882 /* See if X is of the form (* (+ A B) C), and if so convert to
9883 (+ (* A C) (* B C)) and try to simplify.
9885 Most of the time, this results in no change. However, if some of
9886 the operands are the same or inverses of each other, simplifications
9887 will result.
9889 For example, (and (ior A B) (not B)) can occur as the result of
9890 expanding a bit field assignment. When we apply the distributive
9891 law to this, we get (ior (and (A (not B))) (and (B (not B)))),
9892 which then simplifies to (and (A (not B))).
9894 Note that no checks happen on the validity of applying the inverse
9895 distributive law. This is pointless since we can do it in the
9896 few places where this routine is called.
9898 N is the index of the term that is decomposed (the arithmetic operation,
9899 i.e. (+ A B) in the first example above). !N is the index of the term that
9900 is distributed, i.e. of C in the first example above. */
9901 static rtx
9903 {
9904 machine_mode mode;
9908 /* Distributivity is not true for floating point as it can change the
9909 value. So we don't do it unless -funsafe-math-optimizations. */
9926 /* Special case (and (xor B C) (not A)), which is equivalent to
9927 (xor (ior A B) (ior A C)) */
9929 {
9932 }
9935 {
9936 /* Distribute the second term. */
9939 }
9940 else
9941 {
9942 /* Distribute the first term. */
9945 }
9955 }
9956 \f
9957 /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
9958 in MODE. Return an equivalent form, if different from (and VAROP
9959 (const_int CONSTOP)). Otherwise, return NULL_RTX. */
9961 static rtx
9964 {
9967 rtx orig_varop;
9975 /* Simplify VAROP knowing that we will be only looking at some of the
9976 bits in it.
9978 Note by passing in CONSTOP, we guarantee that the bits not set in
9979 CONSTOP are not significant and will never be examined. We must
9980 ensure that is the case by explicitly masking out those bits
9981 before returning. */
9984 /* If VAROP is a CLOBBER, we will fail so return it. */
9988 /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
9989 to VAROP and return the new constant. */
9993 /* See what bits may be nonzero in VAROP. Unlike the general case of
9994 a call to nonzero_bits, here we don't care about bits outside
9995 MODE. */
9999 /* Turn off all bits in the constant that are known to already be zero.
10000 Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
10001 which is tested below. */
10005 /* If we don't have any bits left, return zero. */
10009 /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
10010 a power of two, we can replace this with an ASHIFT. */
10015 /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
10016 or XOR, then try to apply the distributive law. This may eliminate
10017 operations if either branch can be simplified because of the AND.
10018 It may also make some cases more complex, but those cases probably
10019 won't match a pattern either with or without this. */
10022 {
10024 return
10025 gen_lowpart
10027 apply_distributive_law
10031 constop),
10034 constop))));
10035 }
10037 /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
10038 the AND and see if one of the operands simplifies to zero. If so, we
10039 may eliminate it. */
10043 {
10052 }
10054 /* Make a SUBREG if necessary. If we can't make it, fail. */
10059 /* If we are only masking insignificant bits, return VAROP. */
10066 /* Otherwise, return an AND. */
10068 }
10071 /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
10072 in MODE.
10074 Return an equivalent form, if different from X. Otherwise, return X. If
10075 X is zero, we are to always construct the equivalent form. */
10077 static rtx
10080 {
10091 }
10092 \f
10093 /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
10094 We don't care about bits outside of those defined in MODE.
10096 For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
10097 a shift, AND, or zero_extract, we can do better. */
10099 static rtx
10101 scalar_int_mode mode,
10103 {
10104 rtx tem;
10107 /* If X is a register whose nonzero bits value is current, use it.
10108 Otherwise, if X is a register whose value we can find, use that
10109 value. Otherwise, use the previously-computed global nonzero bits
10110 for this register. */
10124 && !REGNO_REG_SET_P
10127 {
10128 /* Note that, even if the precision of last_set_mode is lower than that
10129 of mode, record_value_for_reg invoked nonzero_bits on the register
10130 with nonzero_bits_mode (because last_set_mode is necessarily integral
10131 and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
10132 are all valid, hence in mode too since nonzero_bits_mode is defined
10133 to the largest HWI_COMPUTABLE_MODE_P mode. */
10136 }
10140 {
10145 }
10148 {
10152 /* We don't know anything about the upper bits. */
10156 }
10159 }
10161 /* Given a reg X of mode XMODE, return the number of bits at the high-order
10162 end of X that are known to be equal to the sign bit. X will be used
10163 in mode MODE; the returned value will always be between 1 and the
10164 number of bits in MODE. */
10166 static rtx
10168 scalar_int_mode mode,
10170 {
10171 rtx tem;
10184 && !REGNO_REG_SET_P
10187 {
10190 }
10201 }
10202 \f
10203 /* Return the number of "extended" bits there are in X, when interpreted
10204 as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
10205 unsigned quantities, this is the number of high-order zero bits.
10206 For signed quantities, this is the number of copies of the sign bit
10207 minus 1. In both case, this function returns the number of "spare"
10208 bits. For example, if two quantities for which this function returns
10209 at least 1 are added, the addition is known not to overflow.
10211 This function will always return 0 unless called during combine, which
10212 implies that it must be called from a define_split. */
10214 unsigned int
10216 {
10220 scalar_int_mode int_mode;
10228 }
10230 /* This function is called from `simplify_shift_const' to merge two
10231 outer operations. Specifically, we have already found that we need
10232 to perform operation *POP0 with constant *PCONST0 at the outermost
10233 position. We would now like to also perform OP1 with constant CONST1
10234 (with *POP0 being done last).
10236 Return 1 if we can do the operation and update *POP0 and *PCONST0 with
10237 the resulting operation. *PCOMP_P is set to 1 if we would need to
10238 complement the innermost operand, otherwise it is unchanged.
10240 MODE is the mode in which the operation will be done. No bits outside
10241 the width of this mode matter. It is assumed that the width of this mode
10242 is smaller than or equal to HOST_BITS_PER_WIDE_INT.
10244 If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
10245 IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
10246 result is simply *PCONST0.
10248 If the resulting operation cannot be expressed as one operation, we
10249 return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
10251 static int
10252 merge_outer_ops (enum rtx_code *pop0, HOST_WIDE_INT *pconst0, enum rtx_code op1, HOST_WIDE_INT const1, machine_mode mode, int *pcomp_p)
10253 {
10260 /* If OP0 is an AND, clear unimportant bits in CONST1. */
10264 /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
10265 if OP0 is SET. */
10274 {
10276 {
10294 }
10295 }
10297 /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
10301 /* If the two constants aren't the same, we can't do anything. The
10302 remaining six cases can all be done. */
10306 else
10308 {
10311 /* (a & b) | b == b */
10314 /* (a ^ b) | b == a | b */
10315 {;}
10320 /* (a & b) ^ b == (~a) & b */
10323 /* (a | b) ^ b == a & ~b */
10329 /* (a | b) & b == b */
10332 /* (a ^ b) & b) == (~a) & b */
10337 }
10339 /* Check for NO-OP cases. */
10352 /* ??? Slightly redundant with the above mask, but not entirely.
10353 Moving this above means we'd have to sign-extend the mode mask
10354 for the final test. */
10359 }
10360 \f
10361 /* A helper to simplify_shift_const_1 to determine the mode we can perform
10362 the shift in. The original shift operation CODE is performed on OP in
10363 ORIG_MODE. Return the wider mode MODE if we can perform the operation
10364 in that mode. Return ORIG_MODE otherwise. We can also assume that the
10365 result of the shift is subject to operation OUTER_CODE with operand
10366 OUTER_CONST. */
10368 static scalar_int_mode
10372 {
10375 /* In general we can't perform in wider mode for right shift and rotate. */
10377 {
10379 /* We can still widen if the bits brought in from the left are identical
10380 to the sign bit of ORIG_MODE. */
10388 /* Similarly here but with zero bits. */
10393 /* We can also widen if the bits brought in will be masked off. This
10394 operation is performed in ORIG_MODE. */
10396 {
10402 }
10403 /* fall through */
10413 }
10414 }
10416 /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
10417 of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
10418 if we cannot simplify it. Otherwise, return a simplified value.
10420 The shift is normally computed in the widest mode we find in VAROP, as
10421 long as it isn't a different number of words than RESULT_MODE. Exceptions
10422 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10424 static rtx
10427 {
10432 machine_mode shift_mode;
10434 unsigned int mode_words
10436 /* We form (outer_op (code varop count) (outer_const)). */
10442 /* Make sure and truncate the "natural" shift on the way in. We don't
10443 want to do this inside the loop as it makes it more difficult to
10444 combine shifts. */
10448 /* If we were given an invalid count, don't do anything except exactly
10449 what was requested. */
10456 /* Unless one of the branches of the `if' in this loop does a `continue',
10457 we will `break' the loop after the `if'. */
10460 {
10461 /* If we have an operand of (clobber (const_int 0)), fail. */
10465 /* Convert ROTATERT to ROTATE. */
10467 {
10471 }
10475 {
10476 /* We only change the modes of scalar shifts. */
10482 }
10484 scalar_int_mode shift_unit_mode
10487 /* Handle cases where the count is greater than the size of the mode
10488 minus 1. For ASHIFT, use the size minus one as the count (this can
10489 occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
10490 take the count modulo the size. For other shifts, the result is
10491 zero.
10493 Since these shifts are being produced by the compiler by combining
10494 multiple operations, each of which are defined, we know what the
10495 result is supposed to be. */
10498 {
10503 else
10504 {
10505 /* We can't simply return zero because there may be an
10506 outer op. */
10510 }
10511 }
10513 /* If we discovered we had to complement VAROP, leave. Making a NOT
10514 here would cause an infinite loop. */
10519 {
10520 /* An arithmetic right shift of a quantity known to be -1 or 0
10521 is a no-op. */
10525 {
10528 }
10530 /* If we are doing an arithmetic right shift and discarding all but
10531 the sign bit copies, this is equivalent to doing a shift by the
10532 bitsize minus one. Convert it into that shift because it will
10533 often allow other simplifications. */
10540 /* We simplify the tests below and elsewhere by converting
10541 ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10542 `make_compound_operation' will convert it to an ASHIFTRT for
10543 those machines (such as VAX) that don't have an LSHIFTRT. */
10548 shift_unit_mode)))
10560 }
10563 {
10570 {
10573 }
10577 /* The following rules apply only to scalars. */
10582 /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10583 minus the width of a smaller mode, we can do this with a
10584 SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10591 {
10600 }
10604 /* The following rules apply only to scalars. */
10610 /* If VAROP is a SUBREG, strip it as long as the inner operand has
10611 the same number of words as what we've seen so far. Then store
10612 the widest mode in MODE. */
10618 == mode_words
10620 {
10625 }
10629 /* Some machines use MULT instead of ASHIFT because MULT
10630 is cheaper. But it is still better on those machines to
10631 merge two shifts into one. */
10634 {
10635 varop
10641 }
10645 /* Similar, for when divides are cheaper. */
10648 {
10649 varop
10655 }
10659 /* If we are extracting just the sign bit of an arithmetic
10660 right shift, that shift is not needed. However, the sign
10661 bit of a wider mode may be different from what would be
10662 interpreted as the sign bit in a narrower mode, so, if
10663 the result is narrower, don't discard the shift. */
10668 {
10671 }
10673 /* fall through */
10678 /* The following rules apply only to scalars. */
10685 /* Here we have two nested shifts. The result is usually the
10686 AND of a new shift with a mask. We compute the result below. */
10692 {
10696 rtx mask_rtx;
10698 /* We have one common special case. We can't do any merging if
10699 the inner code is an ASHIFTRT of a smaller mode. However, if
10700 we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10701 with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10702 we can convert it to
10703 (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10704 This simplifies certain SIGN_EXTEND operations. */
10708 {
10709 /* C3 has the low-order C1 bits zero. */
10721 }
10723 /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10724 than C1 high-order bits equal to the sign bit, we can convert
10725 this to either an ASHIFT or an ASHIFTRT depending on the
10726 two counts.
10728 We cannot do this if VAROP's mode is not SHIFT_UNIT_MODE. */
10734 {
10738 {
10741 }
10744 }
10746 /* There are some cases we can't do. If CODE is ASHIFTRT,
10747 we can only do this if FIRST_CODE is also ASHIFTRT.
10749 We can't do the case when CODE is ROTATE and FIRST_CODE is
10750 ASHIFTRT.
10752 If the mode of this shift is not the mode of the outer shift,
10753 we can't do this if either shift is a right shift or ROTATE.
10755 Finally, we can't do any of these if the mode is too wide
10756 unless the codes are the same.
10758 Handle the case where the shift codes are the same
10759 first. */
10762 {
10771 }
10782 /* To compute the mask to apply after the shift, shift the
10783 nonzero bits of the inner shift the same way the
10784 outer shift will. */
10787 int_result_mode);
10789 mask_rtx
10793 /* Give up if we can't compute an outer operation to use. */
10801 /* If the shifts are in the same direction, we add the
10802 counts. Otherwise, we subtract them. */
10806 else
10809 /* If COUNT is positive, the new shift is usually CODE,
10810 except for the two exceptions below, in which case it is
10811 FIRST_CODE. If the count is negative, FIRST_CODE should
10812 always be used */
10822 }
10824 /* If we have (A << B << C) for any shift, we can convert this to
10825 (A << C << B). This wins if A is a constant. Only try this if
10826 B is not a constant. */
10831 {
10832 /* For ((unsigned) (cstULL >> count)) >> cst2 we have to make
10833 sure the result will be masked. See PR70222. */
10841 /* For ((int) (cstLL >> count)) >> cst2 just give up. Queuing
10842 up outer sign extension (often left and right shift) is
10843 hardly more efficient than the original. See PR70429. */
10853 }
10857 /* The following rules apply only to scalars. */
10861 /* Make this fit the case below. */
10868 /* The following rules apply only to scalars. */
10874 /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
10875 with C the size of VAROP - 1 and the shift is logical if
10876 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10877 we have an (le X 0) operation. If we have an arithmetic shift
10878 and STORE_FLAG_VALUE is 1 or we have a logical shift with
10879 STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
10887 {
10890 const0_rtx);
10896 }
10898 /* If we have (shift (logical)), move the logical to the outside
10899 to allow it to possibly combine with another logical and the
10900 shift to combine with another shift. This also canonicalizes to
10901 what a ZERO_EXTRACT looks like. Also, some machines have
10902 (and (shift)) insns. */
10905 /* We can't do this if we have (ashiftrt (xor)) and the
10906 constant has its sign bit set in shift_unit_mode with
10907 shift_unit_mode wider than result_mode. */
10911 shift_unit_mode))
10920 {
10923 }
10925 /* If we can't do that, try to simplify the shift in each arm of the
10926 logical expression, make a new logical expression, and apply
10927 the inverse distributive law. This also can't be done for
10928 (ashiftrt (xor)) where we've widened the shift and the constant
10929 changes the sign bit. */
10934 shift_unit_mode)))
10935 {
10947 }
10951 /* The following rules apply only to scalars. */
10956 /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
10957 says that the sign bit can be tested, FOO has mode MODE, C is
10958 GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
10959 that may be nonzero. */
10969 {
10973 }
10977 /* The following rules apply only to scalars. */
10982 /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
10983 than the number of bits in the mode is equivalent to A. */
10987 {
10991 }
10993 /* NEG commutes with ASHIFT since it is multiplication. Move the
10994 NEG outside to allow shifts to combine. */
10998 {
11001 }
11005 /* The following rules apply only to scalars. */
11010 /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
11011 is one less than the number of bits in the mode is
11012 equivalent to (xor A 1). */
11019 {
11023 }
11025 /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
11026 that might be nonzero in BAR are those being shifted out and those
11027 bits are known zero in FOO, we can replace the PLUS with FOO.
11028 Similarly in the other operand order. This code occurs when
11029 we are computing the size of a variable-size array. */
11036 {
11039 }
11047 {
11050 }
11052 /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
11063 {
11066 }
11068 /* Check for 'PLUS signbit', which is the canonical form of 'XOR
11069 signbit', and attempt to change the PLUS to an XOR and move it to
11070 the outer operation as is done above in the AND/IOR/XOR case
11071 leg for shift(logical). See details in logical handling above
11072 for reasoning in doing so. */
11084 {
11087 }
11092 /* The following rules apply only to scalars. */
11097 /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
11098 with C the size of VAROP - 1 and the shift is logical if
11099 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
11100 we have a (gt X 0) operation. If the shift is arithmetic with
11101 STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
11102 we have a (neg (gt X 0)) operation. */
11111 {
11114 const0_rtx);
11120 }
11124 /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
11125 if the truncate does not affect the value. */
11132 {
11135 varop_inner
11138 GEN_INT
11143 }
11148 }
11151 }
11155 {
11156 /* We only change the modes of scalar shifts. */
11161 }
11163 /* We have now finished analyzing the shift. The result should be
11164 a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
11165 OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
11166 to the result of the shift. OUTER_CONST is the relevant constant,
11167 but we must turn off all bits turned off in the shift. */
11175 /* Make a SUBREG if necessary. If we can't make it, fail. */
11180 /* If we have an outer operation and we just made a shift, it is
11181 possible that we could have simplified the shift were it not
11182 for the outer operation. So try to do the simplification
11183 recursively. */
11187 else
11193 /* If we were doing an LSHIFTRT in a wider mode than it was originally,
11194 turn off all the bits that the shift would have turned off. */
11196 /* We only change the modes of scalar shifts. */
11200 /* Do the remainder of the processing in RESULT_MODE. */
11203 /* If COMPLEMENT_P is set, we have to complement X before doing the outer
11204 operation. */
11209 {
11219 {
11220 /* This means that we have determined that the result is
11221 equivalent to a constant. This should be rare. */
11224 }
11227 else
11230 }
11233 }
11235 /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
11236 The result of the shift is RESULT_MODE. If we cannot simplify it,
11237 return X or, if it is NULL, synthesize the expression with
11238 simplify_gen_binary. Otherwise, return a simplified value.
11240 The shift is normally computed in the widest mode we find in VAROP, as
11241 long as it isn't a different number of words than RESULT_MODE. Exceptions
11242 are ASHIFTRT and ROTATE, which are always done in their original mode. */
11244 static rtx
11247 {
11257 }
11259 \f
11260 /* A subroutine of recog_for_combine. See there for arguments and
11261 return value. */
11263 static int
11265 {
11267 rtx pat_without_clobbers;
11275 /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
11276 we use to indicate that something didn't match. If we find such a
11277 thing, force rejection. */
11291 {
11294 else
11297 }
11299 /* If it isn't, there is the possibility that we previously had an insn
11300 that clobbered some register as a side effect, but the combined
11301 insn doesn't need to do that. So try once more without the clobbers
11302 unless this represents an ASM insn. */
11306 {
11311 {
11314 pos++;
11315 }
11325 {
11328 else
11331 }
11332 }
11339 /* Recognize all noop sets, these will be killed by followup pass. */
11343 /* If we had any clobbers to add, make a new pattern than contains
11344 them. Then check to make sure that all of them are dead. */
11346 {
11356 else
11363 {
11368 {
11372 }
11373 }
11375 }
11379 {
11387 /* Allow targets to reject combined insn. */
11389 {
11392 dump_file);
11394 /* Callers expect recog_for_combine to strip
11395 clobbers from the pattern on failure. */
11400 }
11405 }
11411 }
11413 /* Change every ZERO_EXTRACT and ZERO_EXTEND of a SUBREG that can be
11414 expressed as an AND and maybe an LSHIFTRT, to that formulation.
11415 Return whether anything was so changed. */
11417 static bool
11419 {
11425 {
11437 {
11446 else
11450 }
11456 {
11462 }
11467 {
11471 }
11472 else
11478 {
11481 }
11485 }
11492 scalar_int_mode mode;
11498 {
11512 else
11520 }
11523 }
11525 /* Like recog, but we receive the address of a pointer to a new pattern.
11526 We try to match the rtx that the pointer points to.
11527 If that fails, we may try to modify or replace the pattern,
11528 storing the replacement into the same pointer object.
11530 Modifications include deletion or addition of CLOBBERs. If the
11531 instruction will still not match, we change ZERO_EXTEND and ZERO_EXTRACT
11532 to the equivalent AND and perhaps LSHIFTRT patterns, and try with that
11533 (and undo if that fails).
11535 PNOTES is a pointer to a location where any REG_UNUSED notes added for
11536 the CLOBBERs are placed.
11538 The value is the final insn code from the pattern ultimately matched,
11539 or -1. */
11541 static int
11543 {
11555 {
11558 {
11562 }
11563 }
11566 {
11571 }
11574 }
11575 \f
11576 /* Like gen_lowpart_general but for use by combine. In combine it
11577 is not possible to create any new pseudoregs. However, it is
11578 safe to create invalid memory addresses, because combine will
11579 try to recognize them and all they will do is make the combine
11580 attempt fail.
11582 If for some reason this cannot do its job, an rtx
11583 (clobber (const_int 0)) is returned.
11584 An insn containing that will not be recognized. */
11586 static rtx
11588 {
11592 rtx result;
11597 /* We can only support MODE being wider than a word if X is a
11598 constant integer or has a mode the same size. */
11603 /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
11604 won't know what to do. So we will strip off the SUBREG here and
11605 process normally. */
11607 {
11610 /* For use in case we fall down into the address adjustments
11611 further below, we need to adjust the known mode and size of
11612 x; imode and isize, since we just adjusted x. */
11619 }
11627 {
11628 /* Refuse to work on a volatile memory ref or one with a mode-dependent
11629 address. */
11634 /* If we want to refer to something bigger than the original memref,
11635 generate a paradoxical subreg instead. That will force a reload
11636 of the original memref X. */
11642 }
11644 /* If X is a comparison operator, rewrite it in a new mode. This
11645 probably won't match, but may allow further simplifications. */
11649 /* If we couldn't simplify X any other way, just enclose it in a
11650 SUBREG. Normally, this SUBREG won't match, but some patterns may
11651 include an explicit SUBREG or we may simplify it further in combine. */
11652 else
11653 {
11654 rtx res;
11657 {
11662 }
11666 }
11668 fail:
11670 }
11671 \f
11672 /* Try to simplify a comparison between OP0 and a constant OP1,
11673 where CODE is the comparison code that will be tested, into a
11674 (CODE OP0 const0_rtx) form.
11676 The result is a possibly different comparison code to use.
11677 *POP1 may be updated. */
11679 static enum rtx_code
11682 {
11683 scalar_int_mode int_mode;
11686 /* Get the constant we are comparing against and turn off all bits
11687 not on in our mode. */
11691 /* If we are comparing against a constant power of two and the value
11692 being compared can only have that single bit nonzero (e.g., it was
11693 `and'ed with that bit), we can replace this with a comparison
11694 with zero. */
11703 {
11706 }
11708 /* Similarly, if we are comparing a value known to be either -1 or
11709 0 with -1, change it to the opposite comparison against zero. */
11715 {
11718 }
11720 /* Do some canonicalizations based on the comparison code. We prefer
11721 comparisons against zero and then prefer equality comparisons.
11722 If we can reduce the size of a constant, we will do that too. */
11724 {
11726 /* < C is equivalent to <= (C - 1) */
11728 {
11731 /* ... fall through to LE case below. */
11733 }
11734 else
11738 /* <= C is equivalent to < (C + 1); we do this for C < 0 */
11740 {
11743 }
11745 /* If we are doing a <= 0 comparison on a value known to have
11746 a zero sign bit, we can replace this with == 0. */
11757 /* >= C is equivalent to > (C - 1). */
11759 {
11762 /* ... fall through to GT below. */
11764 }
11765 else
11769 /* > C is equivalent to >= (C + 1); we do this for C < 0. */
11771 {
11774 }
11776 /* If we are doing a > 0 comparison on a value known to have
11777 a zero sign bit, we can replace this with != 0. */
11788 /* < C is equivalent to <= (C - 1). */
11790 {
11793 /* ... fall through ... */
11794 }
11795 /* (unsigned) < 0x80000000 is equivalent to >= 0. */
11800 {
11804 }
11805 else
11809 /* unsigned <= 0 is equivalent to == 0 */
11812 /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
11816 == ((HOST_WIDE_INT_1U
11818 {
11821 }
11825 /* >= C is equivalent to > (C - 1). */
11827 {
11830 /* ... fall through ... */
11831 }
11833 /* (unsigned) >= 0x80000000 is equivalent to < 0. */
11838 {
11842 }
11843 else
11847 /* unsigned > 0 is equivalent to != 0 */
11850 /* (unsigned) > 0x7fffffff is equivalent to < 0. */
11854 == (HOST_WIDE_INT_1U
11856 {
11859 }
11864 }
11868 }
11869 \f
11870 /* Simplify a comparison between *POP0 and *POP1 where CODE is the
11871 comparison code that will be tested.
11873 The result is a possibly different comparison code to use. *POP0 and
11874 *POP1 may be updated.
11876 It is possible that we might detect that a comparison is either always
11877 true or always false. However, we do not perform general constant
11878 folding in combine, so this knowledge isn't useful. Such tautologies
11879 should have been detected earlier. Hence we ignore all such cases. */
11881 static enum rtx_code
11883 {
11889 opt_scalar_int_mode tmode_iter;
11891 /* Try a few ways of applying the same transformation to both operands. */
11893 {
11894 /* The test below this one won't handle SIGN_EXTENDs on these machines,
11895 so check specially. */
11914 {
11917 }
11919 /* If both operands are the same constant shift, see if we can ignore the
11920 shift. We can if the shift is a rotate or if the bits shifted out of
11921 this shift are known to be zero for both inputs and if the type of
11922 comparison is compatible with the shift. */
11935 {
11948 else
11950 }
11952 /* If both operands are AND's of a paradoxical SUBREG by constant, the
11953 SUBREGs are of the same mode, and, in both cases, the AND would
11954 be redundant if the comparison was done in the narrower mode,
11955 do the comparison in the narrower mode (e.g., we are AND'ing with 1
11956 and the operand's possibly nonzero bits are 0xffffff01; in that case
11957 if we only care about QImode, we don't need the AND). This case
11958 occurs if the output mode of an scc insn is not SImode and
11959 STORE_FLAG_VALUE == 1 (e.g., the 386).
11961 Similarly, check for a case where the AND's are ZERO_EXTEND
11962 operations from some narrower mode even though a SUBREG is not
11963 present. */
11968 {
11985 {
11989 /* The resulting comparison is always unsigned since we masked
11990 off the original sign bit. */
11994 }
12000 {
12006 }
12010 }
12012 /* If both operands are NOT, we can strip off the outer operation
12013 and adjust the comparison code for swapped operands; similarly for
12014 NEG, except that this must be an equality comparison. */
12020 else
12022 }
12024 /* If the first operand is a constant, swap the operands and adjust the
12025 comparison code appropriately, but don't do this if the second operand
12026 is already a constant integer. */
12028 {
12031 }
12033 /* We now enter a loop during which we will try to simplify the comparison.
12034 For the most part, we only are concerned with comparisons with zero,
12035 but some things may really be comparisons with zero but not start
12036 out looking that way. */
12039 {
12041 scalar_int_mode int_mode;
12045 HOST_WIDE_INT const_op;
12047 /* We only want to handle integral modes. This catches VOIDmode,
12048 CCmode, and the floating-point modes. An exception is that we
12049 can handle VOIDmode if OP0 is a COMPARE or a comparison
12050 operation. */
12057 /* Try to simplify the compare to constant, possibly changing the
12058 comparison op, and/or changing op1 to zero. */
12062 /* Compute some predicates to simplify code below. */
12069 /* If this is a sign bit comparison and we can do arithmetic in
12070 MODE, say that we will only be needing the sign bit of OP0. */
12075 HOST_WIDE_INT_1U
12080 {
12081 /* We can't do anything if OP0 is a condition code value, rather
12082 than an actual data value. */
12088 /* Get the two operands being compared. */
12091 else
12094 /* Check for the cases where we simply want the result of the
12095 earlier test or the opposite of that result. */
12099 {
12103 else
12107 {
12112 }
12113 }
12115 }
12121 /* Now try cases based on the opcode of OP0. If none of the cases
12122 does a "continue", we exit this loop immediately after the
12123 switch. */
12128 {
12130 /* If we are extracting a single bit from a variable position in
12131 a constant that has only a single bit set and are comparing it
12132 with zero, we can convert this into an equality comparison
12133 between the position and the location of the single bit. */
12134 /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
12135 have already reduced the shift count modulo the word size. */
12141 {
12149 /* Result is nonzero iff shift count is equal to I. */
12152 }
12154 /* fall through */
12159 {
12162 }
12166 /* If testing for equality, we can take the NOT of the constant. */
12169 {
12173 }
12175 /* If just looking at the sign bit, reverse the sense of the
12176 comparison. */
12178 {
12182 }
12186 /* If testing for equality, we can take the NEG of the constant. */
12189 {
12193 }
12195 /* The remaining cases only apply to comparisons with zero. */
12199 /* When X is ABS or is known positive,
12200 (neg X) is < 0 if and only if X != 0. */
12208 {
12212 }
12214 /* If we have NEG of something whose two high-order bits are the
12215 same, we know that "(-a) < 0" is equivalent to "a > 0". */
12217 {
12221 }
12225 /* If we are testing equality and our count is a constant, we
12226 can perform the inverse operation on our RHS. */
12230 {
12234 }
12236 /* If we are doing a < 0 or >= 0 comparison, it means we are testing
12237 a particular bit. Convert it to an AND of a constant of that
12238 bit. This will be converted into a ZERO_EXTRACT. */
12242 {
12244 (HOST_WIDE_INT_1U
12249 }
12251 /* Fall through. */
12254 /* ABS is ignorable inside an equality comparison with zero. */
12256 {
12259 }
12263 /* Can simplify (compare (zero/sign_extend FOO) CONST) to
12264 (compare FOO CONST) if CONST fits in FOO's mode and we
12265 are either testing inequality or have an unsigned
12266 comparison with ZERO_EXTEND or a signed comparison with
12267 SIGN_EXTEND. But don't do it if we don't have a compare
12268 insn of the given mode, since we'd have to revert it
12269 later on, and then we wouldn't know whether to sign- or
12270 zero-extend. */
12272 && ! unsigned_comparison_p
12276 {
12279 }
12283 /* Check for the case where we are comparing A - C1 with C2, that is
12285 (subreg:MODE (plus (A) (-C1))) op (C2)
12287 with C1 a constant, and try to lift the SUBREG, i.e. to do the
12288 comparison in the wider mode. One of the following two conditions
12289 must be true in order for this to be valid:
12291 1. The mode extension results in the same bit pattern being added
12292 on both sides and the comparison is equality or unsigned. As
12293 C2 has been truncated to fit in MODE, the pattern can only be
12294 all 0s or all 1s.
12296 2. The mode extension results in the sign bit being copied on
12297 each side.
12299 The difficulty here is that we have predicates for A but not for
12300 (A - C1) so we need to check that C1 is within proper bounds so
12301 as to perturbate A as little as possible. */
12310 {
12318 /* (A - C1) zero-extends if it is positive and sign-extends
12319 if it is negative, C2 both zero- and sign-extends. */
12323 /* (A - C1) sign-extends if it is positive and 1-extends
12324 if it is negative, C2 both sign- and 1-extends. */
12331 /* (A - C1) always sign-extends, like C2. */
12335 {
12338 }
12339 }
12341 /* If the inner mode is narrower and we are extracting the low part,
12342 we can treat the SUBREG as if it were a ZERO_EXTEND. */
12344 ;
12353 {
12354 /* Remove outer subregs that don't do anything. */
12359 {
12363 }
12365 }
12366 else
12369 /* FALLTHROUGH */
12378 {
12381 }
12385 /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
12386 this for equality comparisons due to pathological cases involving
12387 overflows. */
12391 {
12395 }
12397 /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
12400 {
12404 }
12408 /* We used to optimize signed comparisons against zero, but that
12409 was incorrect. Unsigned comparisons against zero (GTU, LEU)
12410 arrive here as equality comparisons, or (GEU, LTU) are
12411 optimized away. No need to special-case them. */
12413 /* (eq (minus A B) C) -> (eq A (plus B C)) or
12414 (eq B (minus A C)), whichever simplifies. We can only do
12415 this for equality comparisons due to pathological cases involving
12416 overflows. */
12420 {
12424 }
12429 {
12433 }
12435 /* The sign bit of (minus (ashiftrt X C) X), where C is the number
12436 of bits in X minus 1, is one iff X > 0. */
12441 {
12445 }
12449 /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
12450 if C is zero or B is a constant. */
12454 {
12458 }
12463 /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
12464 iff X <= 0. */
12468 {
12472 }
12476 /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
12477 will be converted to a ZERO_EXTRACT later. */
12481 {
12486 }
12488 /* If we are comparing (and (lshiftrt X C1) C2) for equality with
12489 zero and X is a comparison and C1 and C2 describe only bits set
12490 in STORE_FLAG_VALUE, we can compare with X. */
12498 {
12505 {
12508 }
12509 }
12511 /* If we are doing an equality comparison of an AND of a bit equal
12512 to the sign bit, replace this with a LT or GE comparison of
12513 the underlying value. */
12520 {
12524 }
12526 /* If this AND operation is really a ZERO_EXTEND from a narrower
12527 mode, the constant fits within that mode, and this is either an
12528 equality or unsigned comparison, try to do this comparison in
12529 the narrower mode.
12531 Note that in:
12533 (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
12534 -> (ne:DI (reg:SI 4) (const_int 0))
12536 unless TARGET_TRULY_NOOP_TRUNCATION allows it or the register is
12537 known to hold a value of the required mode the
12538 transformation is invalid. */
12546 {
12549 }
12551 /* If this is (and:M1 (subreg:M1 X:M2 0) (const_int C1)) where C1
12552 fits in both M1 and M2 and the SUBREG is either paradoxical
12553 or represents the low part, permute the SUBREG and the AND
12554 and try again. */
12557 {
12559 /* Require an integral mode, to avoid creating something like
12560 (AND:SF ...). */
12563 /* It is unsafe to commute the AND into the SUBREG if the
12564 SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
12565 not defined. As originally written the upper bits
12566 have a defined value due to the AND operation.
12567 However, if we commute the AND inside the SUBREG then
12568 they no longer have defined values and the meaning of
12569 the code has been changed.
12570 Also C1 should not change value in the smaller mode,
12571 see PR67028 (a positive C1 can become negative in the
12572 smaller mode, so that the AND does no longer mask the
12573 upper bits). */
12574 && ((WORD_REGISTER_OPERATIONS
12586 {
12592 }
12593 }
12595 /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
12599 {
12604 }
12606 /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
12607 (eq (and (lshiftrt X) 1) 0).
12608 Also handle the case where (not X) is expressed using xor. */
12612 {
12622 == HOST_WIDE_INT_1U
12624 {
12625 op0
12630 }
12631 }
12635 /* If we have (compare (ashift FOO N) (const_int C)) and
12636 the high order N bits of FOO (N+1 if an inequality comparison)
12637 are known to be zero, we can do this by comparing FOO with C
12638 shifted right N bits so long as the low-order N bits of C are
12639 zero. */
12651 {
12652 /* We must perform a logical shift, not an arithmetic one,
12653 as we want the top N bits of C to be zero. */
12660 }
12662 /* If we are doing a sign bit comparison, it means we are testing
12663 a particular bit. Convert it to the appropriate AND. */
12666 {
12668 (HOST_WIDE_INT_1U
12673 }
12675 /* If this an equality comparison with zero and we are shifting
12676 the low bit to the sign bit, we can convert this to an AND of the
12677 low-order bit. */
12681 {
12684 }
12688 /* If this is an equality comparison with zero, we can do this
12689 as a logical shift, which might be much simpler. */
12692 {
12697 }
12699 /* If OP0 is a sign extension and CODE is not an unsigned comparison,
12700 do the comparison in a narrower mode. */
12710 {
12713 }
12715 /* Likewise if OP0 is a PLUS of a sign extension with a
12716 constant, which is usually represented with the PLUS
12717 between the shifts. */
12729 {
12737 new_const);
12739 }
12741 /* FALLTHROUGH */
12743 /* If we have (compare (xshiftrt FOO N) (const_int C)) and
12744 the low order N bits of FOO are known to be zero, we can do this
12745 by comparing FOO with C shifted left N bits so long as no
12746 overflow occurs. Even if the low order N bits of FOO aren't known
12747 to be zero, if the comparison is >= or < we can use the same
12748 optimization and for > or <= by setting all the low
12749 order N bits in the comparison constant. */
12760 {
12761 unsigned HOST_WIDE_INT low_bits
12763 & ((HOST_WIDE_INT_1U
12766 {
12767 /* If the shift was logical, then we must make the condition
12768 unsigned. */
12777 const_op
12782 }
12783 }
12785 /* If we are using this shift to extract just the sign bit, we
12786 can replace this with an LT or GE comparison. */
12791 {
12795 }
12800 }
12803 }
12805 /* Now make any compound operations involved in this comparison. Then,
12806 check for an outmost SUBREG on OP0 that is not doing anything or is
12807 paradoxical. The latter transformation must only be performed when
12808 it is known that the "extra" bits will be the same in op0 and op1 or
12809 that they don't matter. There are three cases to consider:
12811 1. SUBREG_REG (op0) is a register. In this case the bits are don't
12812 care bits and we can assume they have any convenient value. So
12813 making the transformation is safe.
12815 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is UNKNOWN.
12816 In this case the upper bits of op0 are undefined. We should not make
12817 the simplification in that case as we do not know the contents of
12818 those bits.
12820 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not UNKNOWN.
12821 In that case we know those bits are zeros or ones. We must also be
12822 sure that they are the same as the upper bits of op1.
12824 We can never remove a SUBREG for a non-equality comparison because
12825 the sign bit is in a different place in the underlying object. */
12838 {
12840 {
12841 /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
12842 implemented. */
12844 {
12847 }
12848 }
12852 {
12857 }
12858 }
12860 /* We now do the opposite procedure: Some machines don't have compare
12861 insns in all modes. If OP0's mode is an integer mode smaller than a
12862 word and we can't do a compare in that mode, see if there is a larger
12863 mode for which we can do the compare. There are a number of cases in
12864 which we can use the wider mode. */
12870 {
12875 {
12878 /* If this is a test for negative, we can make an explicit
12879 test of the sign bit. Test this first so we can use
12880 a paradoxical subreg to extend OP0. */
12884 {
12885 unsigned HOST_WIDE_INT sign
12892 }
12894 /* If the only nonzero bits in OP0 and OP1 are those in the
12895 narrower mode and this is an equality or unsigned comparison,
12896 we can use the wider mode. Similarly for sign-extended
12897 values, in which case it is true for all comparisons. */
12914 {
12915 /* If OP0 is an AND and we don't have an AND in MODE either,
12916 make a new AND in the proper mode. */
12924 else
12925 {
12927 {
12932 }
12933 else
12934 {
12939 }
12941 }
12942 }
12943 }
12944 }
12946 /* We may have changed the comparison operands. Re-canonicalize. */
12948 {
12951 }
12953 /* If this machine only supports a subset of valid comparisons, see if we
12954 can convert an unsupported one into a supported one. */
12961 }
12962 \f
12963 /* Utility function for record_value_for_reg. Count number of
12964 rtxs in X. */
12965 static int
12967 {
12974 {
12994 }
13005 }
13006 \f
13007 /* Utility function for following routine. Called when X is part of a value
13008 being stored into last_set_value. Sets last_set_table_tick
13009 for each register mentioned. Similar to mention_regs in cse.c */
13011 static void
13013 {
13019 {
13025 {
13028 }
13031 }
13035 {
13036 /* Check for identical subexpressions. If x contains
13037 identical subexpression we only have to traverse one of
13038 them. */
13040 {
13041 /* Note that at this point x1 has already been
13042 processed. */
13046 /* If x0 and x1 are identical then there is no need to
13047 process x0. */
13051 /* If x0 is identical to a subexpression of x1 then while
13052 processing x1, x0 has already been processed. Thus we
13053 are done with x. */
13058 /* If x1 is identical to a subexpression of x0 then we
13059 still have to process the rest of x0. */
13062 {
13065 }
13066 }
13069 }
13073 }
13075 /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
13076 are saying that the register is clobbered and we no longer know its
13077 value. If INSN is zero, don't update reg_stat[].last_set; this is
13078 only permitted with VALUE also zero and is used to invalidate the
13079 register. */
13081 static void
13083 {
13089 /* If VALUE contains REG and we have a previous value for REG, substitute
13090 the previous value. */
13092 {
13093 rtx tem;
13095 /* Set things up so get_last_value is allowed to see anything set up to
13096 our insn. */
13100 /* If TEM is simply a binary operation with two CLOBBERs as operands,
13101 it isn't going to be useful and will take a lot of time to process,
13102 so just use the CLOBBER. */
13105 {
13111 {
13112 /* If there are two or more occurrences of REG in VALUE,
13113 prevent the value from growing too much. */
13116 }
13119 }
13120 }
13122 /* For each register modified, show we don't know its value, that
13123 we don't know about its bitwise content, that its value has been
13124 updated, and that we don't know the location of the death of the
13125 register. */
13127 {
13139 }
13141 /* Mark registers that are being referenced in this value. */
13145 /* Now update the status of each register being set.
13146 If someone is using this register in this block, set this register
13147 to invalid since we will get confused between the two lives in this
13148 basic block. This makes using this register always invalid. In cse, we
13149 scan the table to invalidate all entries using this register, but this
13150 is too much work for us. */
13153 {
13159 else
13161 }
13163 /* The value being assigned might refer to X (like in "x++;"). In that
13164 case, we must replace it with (clobber (const_int 0)) to prevent
13165 infinite loops. */
13168 {
13172 }
13174 /* For the main register being modified, update the value, the mode, the
13175 nonzero bits, and the number of sign bit copies. */
13180 {
13188 rsp->last_set_sign_bit_copies
13190 }
13191 }
13193 /* Called via note_stores from record_dead_and_set_regs to handle one
13194 SET or CLOBBER in an insn. DATA is the instruction in which the
13195 set is occurring. */
13197 static void
13199 {
13206 {
13210 }
13213 {
13214 /* If we are setting the whole register, we know its value. Otherwise
13215 show that we don't know the value. We can handle SUBREG in
13216 some cases. */
13227 else
13229 }
13231 /* Ignore pushes, they clobber nothing. */
13234 }
13236 /* Update the records of when each REG was most recently set or killed
13237 for the things done by INSN. This is the last thing done in processing
13238 INSN in the combiner loop.
13240 We update reg_stat[], in particular fields last_set, last_set_value,
13241 last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
13242 last_death, and also the similar information mem_last_set (which insn
13243 most recently modified memory) and last_call_luid (which insn was the
13244 most recent subroutine call). */
13246 static void
13248 {
13249 rtx link;
13253 {
13256 {
13261 {
13266 }
13267 }
13270 }
13273 {
13274 hard_reg_set_iterator hrsi;
13276 {
13288 }
13292 /* We can't combine into a call pattern. Remember, though, that
13293 the return value register is set at this LUID. We could
13294 still replace a register with the return value from the
13295 wrong subroutine call! */
13297 }
13298 else
13300 }
13302 /* If a SUBREG has the promoted bit set, it is in fact a property of the
13303 register present in the SUBREG, so for each such SUBREG go back and
13304 adjust nonzero and sign bit information of the registers that are
13305 known to have some zero/sign bits set.
13307 This is needed because when combine blows the SUBREGs away, the
13308 information on zero/sign bits is lost and further combines can be
13309 missed because of that. */
13311 static void
13313 {
13315 rtx set;
13323 {
13332 {
13335 }
13339 {
13342 }
13345 {
13348 }
13349 else
13351 }
13352 }
13354 /* Check if X, a register, is known to contain a value already
13355 truncated to MODE. In this case we can use a subreg to refer to
13356 the truncated value even though in the generic case we would need
13357 an explicit truncation. */
13359 static bool
13361 {
13373 }
13375 /* If X is a hard reg or a subreg record the mode that the register is
13376 accessed in. For non-TARGET_TRULY_NOOP_TRUNCATION targets we might be
13377 able to turn a truncate into a subreg using this information. Return true
13378 if traversing X is complete. */
13380 static bool
13382 {
13383 machine_mode truncated_mode;
13387 {
13399 }
13400 /* ??? For hard-regs we now record everything. We might be able to
13401 optimize this using last_set_mode. */
13404 else
13411 {
13414 }
13417 }
13419 /* Callback for note_uses. Find hardregs and subregs of pseudos and
13420 the modes they are used in. This can help truning TRUNCATEs into
13421 SUBREGs. */
13423 static void
13425 {
13430 }
13432 /* Scan X for promoted SUBREGs. For each one found,
13433 note what it implies to the registers used in it. */
13435 static void
13437 {
13442 else
13443 {
13449 {
13459 }
13460 }
13461 }
13462 \f
13463 /* Verify that all the registers and memory references mentioned in *LOC are
13464 still valid. *LOC was part of a value set in INSN when label_tick was
13465 equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
13466 the invalid references with (clobber (const_int 0)) and return 1. This
13467 replacement is useful because we often can get useful information about
13468 the form of a value (e.g., if it was produced by a shift that always
13469 produces -1 or 0) even though we don't know exactly what registers it
13470 was produced from. */
13472 static int
13474 {
13481 {
13487 {
13490 /* If this is a pseudo-register that was only set once and not
13491 live at the beginning of the function, it is always valid. */
13495 && (!REGNO_REG_SET_P
13497 regno)))
13499 {
13503 }
13504 }
13507 }
13508 /* If this is a memory reference, make sure that there were no stores after
13509 it that might have clobbered the value. We don't have alias info, so we
13510 assume any store invalidates it. Moreover, we only have local UIDs, so
13511 we also assume that there were stores in the intervening basic blocks. */
13514 {
13518 }
13521 {
13523 {
13524 /* Check for identical subexpressions. If x contains
13525 identical subexpression we only have to traverse one of
13526 them. */
13528 {
13529 /* Note that at this point x0 has already been checked
13530 and found valid. */
13534 /* If x0 and x1 are identical then x is also valid. */
13538 /* If x1 is identical to a subexpression of x0 then
13539 while checking x0, x1 has already been checked. Thus
13540 it is valid and so as x. */
13545 /* If x0 is identical to a subexpression of x1 then x is
13546 valid iff the rest of x1 is valid. */
13549 return
13553 }
13558 }
13564 }
13566 /* If we haven't found a reason for it to be invalid, it is valid. */
13568 }
13570 /* Get the last value assigned to X, if known. Some registers
13571 in the value may be replaced with (clobber (const_int 0)) if their value
13572 is known longer known reliably. */
13574 static rtx
13576 {
13578 rtx value;
13581 /* If this is a non-paradoxical SUBREG, get the value of its operand and
13582 then convert it to the desired mode. If this is a paradoxical SUBREG,
13583 we cannot predict what values the "extra" bits might have. */
13597 /* If we don't have a value, or if it isn't for this basic block and
13598 it's either a hard register, set more than once, or it's a live
13599 at the beginning of the function, return 0.
13601 Because if it's not live at the beginning of the function then the reg
13602 is always set before being used (is never used without being set).
13603 And, if it's set only once, and it's always set before use, then all
13604 uses must have the same last value, even if it's not from this basic
13605 block. */
13612 || REGNO_REG_SET_P
13616 /* If the value was set in a later insn than the ones we are processing,
13617 we can't use it even if the register was only set once. */
13622 /* If fewer bits were set than what we are asked for now, we cannot use
13623 the value. */
13628 /* If the value has all its registers valid, return it. */
13632 /* Otherwise, make a copy and replace any invalid register with
13633 (clobber (const_int 0)). If that fails for some reason, return 0. */
13640 }
13641 \f
13642 /* Return nonzero if expression X refers to a REG or to memory
13643 that is set in an instruction more recent than FROM_LUID. */
13645 static int
13647 {
13653 {
13657 #ifdef PUSH_ROUNDING
13658 /* Don't allow uses of the stack pointer to be moved,
13659 because we don't know whether the move crosses a push insn. */
13662 #endif
13664 {
13670 }
13672 }
13680 {
13682 {
13687 }
13691 }
13693 }
13694 \f
13695 /* Define three variables used for communication between the following
13696 routines. */
13701 /* Function called via note_stores from reg_dead_at_p.
13703 If DEST is within [reg_dead_regno, reg_dead_endregno), set
13704 reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
13706 static void
13708 {
13718 }
13720 /* Return nonzero if REG is known to be dead at INSN.
13722 We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
13723 referencing REG, it is dead. If we hit a SET referencing REG, it is
13724 live. Otherwise, see if it is live or dead at the start of the basic
13725 block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
13726 must be assumed to be always live. */
13728 static int
13730 {
13731 basic_block block;
13734 /* Set variables for reg_dead_at_p_1. */
13740 /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
13741 we allow the machine description to decide whether use-and-clobber
13742 patterns are OK. */
13744 {
13748 }
13750 /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
13751 beginning of basic block. */
13754 {
13756 {
13766 }
13772 }
13774 /* Look at live-in sets for the basic block that we were in. */
13780 }
13781 \f
13782 /* Note hard registers in X that are used. */
13784 static void
13786 {
13792 {
13796 CASE_CONST_ANY:
13801 /* CC0 must die in the insn after it is set, so we don't need to take
13802 special note of it here. */
13807 /* If we are clobbering a MEM, mark any hard registers inside the
13808 address as used. */
13815 /* A hard reg in a wide mode may really be multiple registers.
13816 If so, mark all of them just like the first. */
13818 {
13819 /* None of this applies to the stack, frame or arg pointers. */
13821 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
13829 }
13833 {
13834 /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
13835 the address. */
13847 }
13852 }
13854 /* Recursively scan the operands of this expression. */
13856 {
13860 {
13864 {
13869 }
13870 }
13871 }
13872 }
13873 \f
13874 /* Remove register number REGNO from the dead registers list of INSN.
13876 Return the note used to record the death, if there was one. */
13878 rtx
13880 {
13887 }
13889 /* For each register (hardware or pseudo) used within expression X, if its
13890 death is in an instruction with luid between FROM_LUID (inclusive) and
13891 TO_INSN (exclusive), put a REG_DEAD note for that register in the
13892 list headed by PNOTES.
13894 That said, don't move registers killed by maybe_kill_insn.
13896 This is done when X is being merged by combination into TO_INSN. These
13897 notes will then be distributed as needed. */
13899 static void
13902 {
13908 {
13912 /* Don't move the register if it gets killed in between from and to. */
13921 {
13924 /* It is possible for the call above to return 0. This can occur
13925 when last_death points to I2 or I1 that we combined with.
13926 In that case make a new note.
13928 We must also check for the case where X is a hard register
13929 and NOTE is a death note for a range of hard registers
13930 including X. In that case, we must put REG_DEAD notes for
13931 the remaining registers in place of NOTE. */
13935 {
13944 }
13946 /* If we didn't find any note, or if we found a REG_DEAD note that
13947 covers only part of the given reg, and we have a multi-reg hard
13948 register, then to be safe we must check for REG_DEAD notes
13949 for each register other than the first. They could have
13950 their own REG_DEAD notes lying around. */
13957 {
13964 else
13970 }
13973 {
13976 }
13977 else
13979 }
13982 }
13985 {
13990 /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
13991 that accesses one word of a multi-word item, some
13992 piece of everything register in the expression is used by
13993 this insn, so remove any old death. */
13994 /* ??? So why do we test for equality of the sizes? */
14003 {
14006 }
14008 /* If this is some other SUBREG, we know it replaces the entire
14009 value, so use that as the destination. */
14013 /* If this is a MEM, adjust deaths of anything used in the address.
14014 For a REG (the only other possibility), the entire value is
14015 being replaced so the old value is not used in this insn. */
14021 }
14030 {
14032 {
14037 }
14040 }
14041 }
14042 \f
14043 /* Return 1 if X is the target of a bit-field assignment in BODY, the
14044 pattern of an insn. X must be a REG. */
14046 static int
14048 {
14052 {
14054 rtx target;
14061 else
14078 }
14086 }
14087 \f
14088 /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
14089 as appropriate. I3 and I2 are the insns resulting from the combination
14090 insns including FROM (I2 may be zero).
14092 ELIM_I2 and ELIM_I1 are either zero or registers that we know will
14093 not need REG_DEAD notes because they are being substituted for. This
14094 saves searching in the most common cases.
14096 Each note in the list is either ignored or placed on some insns, depending
14097 on the type of note. */
14099 static void
14102 {
14104 rtx tem_note;
14108 {
14113 {
14116 /* Doesn't matter much where we put this, as long as it's somewhere.
14117 It is preferable to keep these notes on branches, which is most
14118 likely to be i3. */
14125 else
14126 {
14129 }
14133 /* These notes must remain with the call or trapping instruction. */
14138 else
14139 {
14145 /* ??? Otherwise assume we've combined things such that we
14146 can now prove that the instructions can't trap. Drop the
14147 note in this case. */
14148 }
14152 /* ??? How to distribute between i3-i1. Assume i3 contains the
14153 entire adjustment. Assert i3 contains at least some adjust. */
14155 {
14157 /* fixup_args_size_notes looks at REG_NORETURN note,
14158 so ensure the note is placed there first. */
14160 {
14164 {
14170 }
14171 }
14173 /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
14174 REG_ARGS_SIZE note to all noreturn calls, allow that here. */
14177 && !ACCUMULATE_OUTGOING_ARGS
14179 }
14186 /* These notes must remain with the call. It should not be
14187 possible for both I2 and I3 to be a call. */
14190 else
14191 {
14194 }
14198 /* Any clobbers for i3 may still exist, and so we must process
14199 REG_UNUSED notes from that insn.
14201 Any clobbers from i2 or i1 can only exist if they were added by
14202 recog_for_combine. In that case, recog_for_combine created the
14203 necessary REG_UNUSED notes. Trying to keep any original
14204 REG_UNUSED notes from these insns can cause incorrect output
14205 if it is for the same register as the original i3 dest.
14206 In that case, we will notice that the register is set in i3,
14207 and then add a REG_UNUSED note for the destination of i3, which
14208 is wrong. However, it is possible to have REG_UNUSED notes from
14209 i2 or i1 for register which were both used and clobbered, so
14210 we keep notes from i2 or i1 if they will turn into REG_DEAD
14211 notes. */
14213 /* If this register is set or clobbered in I3, put the note there
14214 unless there is one already. */
14216 {
14224 }
14225 /* Otherwise, if this register is used by I3, then this register
14226 now dies here, so we must put a REG_DEAD note here unless there
14227 is one already. */
14233 {
14236 }
14242 /* These notes say something about results of an insn. We can
14243 only support them if they used to be on I3 in which case they
14244 remain on I3. Otherwise they are ignored.
14246 If the note refers to an expression that is not a constant, we
14247 must also ignore the note since we cannot tell whether the
14248 equivalence is still true. It might be possible to do
14249 slightly better than this (we only have a problem if I2DEST
14250 or I1DEST is present in the expression), but it doesn't
14251 seem worth the trouble. */
14259 /* These notes say something about how a register is used. They must
14260 be present on any use of the register in I2 or I3. */
14265 {
14268 else
14270 }
14275 /* This can show up in several ways -- either directly in the
14276 pattern, or hidden off in the constant pool with (or without?)
14277 a REG_EQUAL note. */
14278 /* ??? Ignore the without-reg_equal-note problem for now. */
14290 {
14293 else
14295 }
14297 /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
14298 as a JUMP_LABEL or decrement LABEL_NUSES if it's already
14299 there. */
14304 {
14311 }
14317 {
14325 }
14329 /* This note says something about the value of a register prior
14330 to the execution of an insn. It is too much trouble to see
14331 if the note is still correct in all situations. It is better
14332 to simply delete it. */
14336 /* If we replaced the right hand side of FROM_INSN with a
14337 REG_EQUAL note, the original use of the dying register
14338 will not have been combined into I3 and I2. In such cases,
14339 FROM_INSN is guaranteed to be the first of the combined
14340 instructions, so we simply need to search back before
14341 FROM_INSN for the previous use or set of this register,
14342 then alter the notes there appropriately.
14344 If the register is used as an input in I3, it dies there.
14345 Similarly for I2, if it is nonzero and adjacent to I3.
14347 If the register is not used as an input in either I3 or I2
14348 and it is not one of the registers we were supposed to eliminate,
14349 there are two possibilities. We might have a non-adjacent I2
14350 or we might have somehow eliminated an additional register
14351 from a computation. For example, we might have had A & B where
14352 we discover that B will always be zero. In this case we will
14353 eliminate the reference to A.
14355 In both cases, we must search to see if we can find a previous
14356 use of A and put the death note there. */
14362 else
14363 {
14374 && !(i2mod
14376 i2mod_old_rhs)))
14381 /* If the new I2 sets the same register that is marked dead
14382 in the note, we do not know where to put the note.
14383 Give up. */
14386 }
14389 {
14393 {
14395 {
14399 }
14401 /* If the register is being set at TEM_INSN, see if that is all
14402 TEM_INSN is doing. If so, delete TEM_INSN. Otherwise, make this
14403 into a REG_UNUSED note instead. Don't delete sets to
14404 global register vars. */
14408 {
14419 ;
14421 /* Verify that it was the set, and not a clobber that
14422 modified the register.
14424 CC0 targets must be careful to maintain setter/user
14425 pairs. If we cannot delete the setter due to side
14426 effects, mark the user with an UNUSED note instead
14427 of deleting it. */
14431 && (!HAVE_cc0
14435 {
14436 /* Move the notes and links of TEM_INSN elsewhere.
14437 This might delete other dead insns recursively.
14438 First set the pattern to something that won't use
14439 any register. */
14458 /* Delete the setter too. */
14460 {
14473 }
14474 }
14475 else
14476 {
14479 /* If there isn't already a REG_UNUSED note, put one
14480 here. Do not place a REG_DEAD note, even if
14481 the register is also used here; that would not
14482 match the algorithm used in lifetime analysis
14483 and can cause the consistency check in the
14484 scheduler to fail. */
14489 }
14490 }
14494 {
14497 /* If we are doing a 3->2 combination, and we have a
14498 register which formerly died in i3 and was not used
14499 by i2, which now no longer dies in i3 and is used in
14500 i2 but does not die in i2, and place is between i2
14501 and i3, then we may need to move a link from place to
14502 i2. */
14504 && from_insn
14507 {
14511 }
14513 }
14517 }
14519 }
14521 /* If the register is set or already dead at PLACE, we needn't do
14522 anything with this note if it is still a REG_DEAD note.
14523 We check here if it is set at all, not if is it totally replaced,
14524 which is what `dead_or_set_p' checks, so also check for it being
14525 set partially. */
14528 {
14534 {
14535 /* Unless the register previously died in PLACE, clear
14536 last_death. [I no longer understand why this is
14537 being done.] */
14541 }
14542 else
14545 /* If this is a death note for a hard reg that is occupying
14546 multiple registers, ensure that we are still using all
14547 parts of the object. If we find a piece of the object
14548 that is unused, we must arrange for an appropriate REG_DEAD
14549 note to be added for it. However, we can't just emit a USE
14550 and tag the note to it, since the register might actually
14551 be dead; so we recourse, and the recursive call then finds
14552 the previous insn that used this register. */
14555 {
14564 {
14567 }
14570 {
14571 /* Put only REG_DEAD notes for pieces that are
14572 not already dead or set. */
14576 {
14583 {
14585 NULL_RTX);
14589 NULL_RTX);
14590 }
14595 {
14597 {
14601 }
14605 {
14608 }
14609 }
14610 }
14613 }
14614 }
14615 }
14619 /* Any other notes should not be present at this point in the
14620 compilation. */
14622 }
14625 {
14628 }
14632 }
14633 }
14634 \f
14635 /* Similarly to above, distribute the LOG_LINKS that used to be present on
14636 I3, I2, and I1 to new locations. This is also called to add a link
14637 pointing at I3 when I3's destination is changed. */
14639 static void
14641 {
14645 {
14652 /* If the insn that this link points to is a NOTE, ignore it. */
14661 {
14664 {
14680 }
14683 }
14684 else
14694 /* A LOG_LINK is defined as being placed on the first insn that uses
14695 a register and points to the insn that sets the register. Start
14696 searching at the next insn after the target of the link and stop
14697 when we reach a set of the register or the end of the basic block.
14699 Note that this correctly handles the link that used to point from
14700 I3 to I2. Also note that not much searching is typically done here
14701 since most links don't point very far away. */
14710 {
14714 }
14717 {
14720 }
14724 /* If we found a place to put the link, place it there unless there
14725 is already a link to the same insn as LINK at that point. */
14728 {
14736 {
14740 /* Set added_links_insn to the earliest insn we added a
14741 link to. */
14745 }
14746 }
14747 }
14748 }
14749 \f
14750 /* Check for any register or memory mentioned in EQUIV that is not
14751 mentioned in EXPR. This is used to restrict EQUIV to "specializations"
14752 of EXPR where some registers may have been replaced by constants. */
14754 static bool
14756 {
14759 {
14764 }
14766 }
14767 \f
14768 DEBUG_FUNCTION void
14770 {
14771 fprintf
14773 ";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
14775 }
14777 void
14779 {
14780 fprintf
14782 "\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
14784 }
14785 \f
14786 /* Try combining insns through substitution. */
14787 static unsigned int
14789 {
14799 rebuild_jump_labels_after_combine
14802 /* Combining insns may have turned an indirect jump into a
14803 direct jump. Rebuild the JUMP_LABEL fields of jumping
14804 instructions. */
14806 {
14813 }
14817 }
14822 {
14832 };
14835 {
14839 {}
14841 /* opt_pass methods: */
14844 {
14846 }
14852 rtl_opt_pass *
14854 {
14856 }