| blob | 58366a6d3316eb10c96a810e1275f90a6c878f03 |
1 /* Optimize by combining instructions for GNU compiler.
2 Copyright (C) 1987-2020 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 modified_between_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. */
109 /* Number of attempts to combine instructions in this function. */
113 /* Number of attempts that got as far as substitution in this function. */
117 /* Number of instructions combined with added SETs in this function. */
121 /* Number of instructions combined in this function. */
125 /* Totals over entire compilation. */
129 /* combine_instructions may try to replace the right hand side of the
130 second instruction with the value of an associated REG_EQUAL note
131 before throwing it at try_combine. That is problematic when there
132 is a REG_DEAD note for a register used in the old right hand side
133 and can cause distribute_notes to do wrong things. This is the
134 second instruction if it has been so modified, null otherwise. */
138 /* When I2MOD is nonnull, this is a copy of the old right hand side. */
142 /* When I2MOD is nonnull, this is a copy of the new right hand side. */
145 \f
147 /* Record last point of death of (hard or pseudo) register n. */
150 /* Record last point of modification of (hard or pseudo) register n. */
153 /* The next group of fields allows the recording of the last value assigned
154 to (hard or pseudo) register n. We use this information to see if an
155 operation being processed is redundant given a prior operation performed
156 on the register. For example, an `and' with a constant is redundant if
157 all the zero bits are already known to be turned off.
159 We use an approach similar to that used by cse, but change it in the
160 following ways:
162 (1) We do not want to reinitialize at each label.
163 (2) It is useful, but not critical, to know the actual value assigned
164 to a register. Often just its form is helpful.
166 Therefore, we maintain the following fields:
168 last_set_value the last value assigned
169 last_set_label records the value of label_tick when the
170 register was assigned
171 last_set_table_tick records the value of label_tick when a
172 value using the register is assigned
173 last_set_invalid set to nonzero when it is not valid
174 to use the value of this register in some
175 register's value
177 To understand the usage of these tables, it is important to understand
178 the distinction between the value in last_set_value being valid and
179 the register being validly contained in some other expression in the
180 table.
182 (The next two parameters are out of date).
184 reg_stat[i].last_set_value is valid if it is nonzero, and either
185 reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
187 Register I may validly appear in any expression returned for the value
188 of another register if reg_n_sets[i] is 1. It may also appear in the
189 value for register J if reg_stat[j].last_set_invalid is zero, or
190 reg_stat[i].last_set_label < reg_stat[j].last_set_label.
192 If an expression is found in the table containing a register which may
193 not validly appear in an expression, the register is replaced by
194 something that won't match, (clobber (const_int 0)). */
196 /* Record last value assigned to (hard or pseudo) register n. */
198 rtx last_set_value;
200 /* Record the value of label_tick when an expression involving register n
201 is placed in last_set_value. */
205 /* Record the value of label_tick when the value for register n is placed in
206 last_set_value. */
210 /* These fields are maintained in parallel with last_set_value and are
211 used to store the mode in which the register was last set, the bits
212 that were known to be zero when it was last set, and the number of
213 sign bits copies it was known to have when it was last set. */
219 /* Set nonzero if references to register n in expressions should not be
220 used. last_set_invalid is set nonzero when this register is being
221 assigned to and last_set_table_tick == label_tick. */
225 /* Some registers that are set more than once and used in more than one
226 basic block are nevertheless always set in similar ways. For example,
227 a QImode register may be loaded from memory in two places on a machine
228 where byte loads zero extend.
230 We record in the following fields if a register has some leading bits
231 that are always equal to the sign bit, and what we know about the
232 nonzero bits of a register, specifically which bits are known to be
233 zero.
235 If an entry is zero, it means that we don't know anything special. */
241 /* Record the value of the label_tick when the last truncation
242 happened. The field truncated_to_mode is only valid if
243 truncation_label == label_tick. */
247 /* Record the last truncation seen for this register. If truncation
248 is not a nop to this mode we might be able to save an explicit
249 truncation if we know that value already contains a truncated
250 value. */
253 };
258 /* One plus the highest pseudo for which we track REG_N_SETS.
259 regstat_init_n_sets_and_refs allocates the array for REG_N_SETS just once,
260 but during combine_split_insns new pseudos can be created. As we don't have
261 updated DF information in that case, it is hard to initialize the array
262 after growing. The combiner only cares about REG_N_SETS (regno) == 1,
263 so instead of growing the arrays, just assume all newly created pseudos
264 during combine might be set multiple times. */
268 /* Record the luid of the last insn that invalidated memory
269 (anything that writes memory, and subroutine calls, but not pushes). */
273 /* Record the luid of the last CALL_INSN
274 so we can tell whether a potential combination crosses any calls. */
278 /* When `subst' is called, this is the insn that is being modified
279 (by combining in a previous insn). The PATTERN of this insn
280 is still the old pattern partially modified and it should not be
281 looked at, but this may be used to examine the successors of the insn
282 to judge whether a simplification is valid. */
286 /* This is the lowest LUID that `subst' is currently dealing with.
287 get_last_value will not return a value if the register was set at or
288 after this LUID. If not for this mechanism, we could get confused if
289 I2 or I1 in try_combine were an insn that used the old value of a register
290 to obtain a new value. In that case, we might erroneously get the
291 new value of the register when we wanted the old one. */
295 /* This contains any hard registers that are used in newpat; reg_dead_at_p
296 must consider all these registers to be always live. */
300 /* This is an insn to which a LOG_LINKS entry has been added. If this
301 insn is the earlier than I2 or I3, combine should rescan starting at
302 that location. */
306 /* And similarly, for notes. */
310 /* Basic block in which we are performing combines. */
314 \f
315 /* Length of the currently allocated uid_insn_cost array. */
319 /* The following array records the insn_cost for every insn
320 in the instruction stream. */
324 /* The following array records the LOG_LINKS for every insn in the
325 instruction stream as struct insn_link pointers. */
331 };
337 {
341 }
343 #define INSN_COST(INSN) (uid_insn_cost[insn_uid_check (INSN)])
344 #define LOG_LINKS(INSN) (uid_log_links[insn_uid_check (INSN)])
346 #define FOR_EACH_LOG_LINK(L, INSN) \
347 for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
349 /* Links for LOG_LINKS are allocated from this obstack. */
353 /* Allocate a link. */
357 {
365 }
367 /* Incremented for each basic block. */
371 /* Reset to label_tick for each extended basic block in scanning order. */
375 /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
376 largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
380 /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
381 be safely used. It is zero while computing them and after combine has
382 completed. This former test prevents propagating values based on
383 previously set values, which can be incorrect if a variable is modified
384 in a loop. */
388 \f
389 /* Record one modification to rtl structure
390 to be undone by storing old_contents into *where. */
394 struct undo
395 {
400 };
402 /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
403 num_undo says how many are currently recorded.
405 other_insn is nonzero if we have modified some other insn in the process
406 of working on subst_insn. It must be verified too. */
408 struct undobuf
409 {
413 };
417 /* Number of times the pseudo being substituted for
418 was found and replaced. */
423 scalar_int_mode,
426 scalar_int_mode,
498 \f
500 /* It is not safe to use ordinary gen_lowpart in combine.
501 See comments in gen_lowpart_for_combine. */
502 #undef RTL_HOOKS_GEN_LOWPART
503 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
505 /* Our implementation of gen_lowpart never emits a new pseudo. */
506 #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
507 #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
509 #undef RTL_HOOKS_REG_NONZERO_REG_BITS
510 #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
512 #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
513 #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
515 #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
516 #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
520 \f
521 /* Convenience wrapper for the canonicalize_comparison target hook.
522 Target hooks cannot use enum rtx_code. */
526 {
530 }
532 /* Try to split PATTERN found in INSN. This returns NULL_RTX if
533 PATTERN cannot be split. Otherwise, it returns an insn sequence.
534 This is a wrapper around split_insns which ensures that the
535 reg_stat vector is made larger if the splitter creates a new
536 register. */
540 {
549 }
551 /* This is used by find_single_use to locate an rtx in LOC that
552 contains exactly one use of DEST, which is typically either a REG
553 or CC0. It returns a pointer to the innermost rtx expression
554 containing DEST. Appearances of DEST that are being used to
555 totally replace it are not counted. */
559 {
568 {
572 CASE_CONST_ANY:
577 /* If the destination is anything other than CC0, PC, a REG or a SUBREG
578 of a REG that occupies all of the REG, the insn uses DEST if
579 it is mentioned in the destination or the source. Otherwise, we
580 need just check the source. */
597 }
599 /* If it wasn't one of the common cases above, check each expression and
600 vector of this code. Look for a unique usage of DEST. */
604 {
606 {
611 else
617 /* Duplicate usage. */
619 }
621 {
625 {
631 else
638 }
639 }
640 }
643 }
646 /* See if DEST, produced in INSN, is used only a single time in the
647 sequel. If so, return a pointer to the innermost rtx expression in which
648 it is used.
650 If PLOC is nonzero, *PLOC is set to the insn containing the single use.
652 If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
653 care about REG_DEAD notes or LOG_LINKS.
655 Otherwise, we find the single use by finding an insn that has a
656 LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
657 only referenced once in that insn, we know that it must be the first
658 and last insn referencing DEST. */
662 {
663 basic_block bb;
669 {
679 }
689 {
695 {
700 }
701 }
704 }
705 \f
706 /* Substitute NEWVAL, an rtx expression, into INTO, a place in some
707 insn. The substitution can be undone by undo_all. If INTO is already
708 set to NEWVAL, do not record this change. Because computing NEWVAL might
709 also call SUBST, we have to compute it before we put anything into
710 the undo table. */
712 static void
714 {
721 /* We'd like to catch as many invalid transformations here as
722 possible. Unfortunately, there are way too many mode changes
723 that are perfectly valid, so we'd waste too much effort for
724 little gain doing the checks here. Focus on catching invalid
725 transformations involving integer constants. */
728 {
729 /* Sanity check that we're replacing oldval with a CONST_INT
730 that is a valid sign-extension for the original mode. */
734 /* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
735 CONST_INT is not valid, because after the replacement, the
736 original mode would be gone. Unfortunately, we can't tell
737 when do_SUBST is called to replace the operand thereof, so we
738 perform this test on oldval instead, checking whether an
739 invalid replacement took place before we got here. */
744 }
748 else
757 }
759 #define SUBST(INTO, NEWVAL) do_SUBST (&(INTO), (NEWVAL))
761 /* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
762 for the value of a HOST_WIDE_INT value (including CONST_INT) is
763 not safe. */
765 static void
767 {
776 else
785 }
787 #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT (&(INTO), (NEWVAL))
789 /* Similar to SUBST, but just substitute the mode. This is used when
790 changing the mode of a pseudo-register, so that any other
791 references to the entry in the regno_reg_rtx array will change as
792 well. */
794 static void
796 {
805 else
814 }
816 #define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE (&(INTO), (NEWVAL))
818 /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
820 static void
822 {
831 else
840 }
842 #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
843 \f
844 /* Subroutine of try_combine. Determine whether the replacement patterns
845 NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to insn_cost
846 than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
847 that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
848 undobuf.other_insn may also both be NULL_RTX. Return false if the cost
849 of all the instructions can be estimated and the replacements are more
850 expensive than the original sequence. */
852 static bool
855 {
860 /* Lookup the original insn_costs. */
865 {
868 {
872 }
873 else
874 {
878 }
879 }
880 else
881 {
884 }
886 /* If we have split a PARALLEL I2 to I1,I2, we have counted its cost twice;
887 correct that. */
892 /* Calculate the replacement insn_costs. */
901 {
911 }
912 else
913 {
916 }
919 {
928 optimize_this_for_speed_p);
932 {
935 }
936 else
938 }
940 /* Disallow this combination if both new_cost and old_cost are greater than
941 zero, and new_cost is greater than old cost. */
945 {
964 else
966 }
971 /* Update the uid_insn_cost array with the replacement costs. */
975 {
979 }
982 }
985 /* Delete any insns that copy a register to itself.
986 Return true if the CFG was changed. */
988 static bool
990 {
992 basic_block bb;
997 {
999 {
1002 {
1007 }
1008 }
1009 }
1012 }
1014 \f
1015 /* Return false if we do not want to (or cannot) combine DEF. */
1016 static bool
1018 {
1019 /* Do not consider if it is pre/post modification in MEM. */
1025 /* Do not combine frame pointer adjustments. */
1028 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
1036 }
1038 /* Return false if we do not want to (or cannot) combine USE. */
1039 static bool
1041 {
1042 /* Do not consider the usage of the stack pointer by function call. */
1047 }
1049 /* Fill in log links field for all insns. */
1051 static void
1053 {
1054 basic_block bb;
1061 /* Pass through each block from the end, recording the uses of each
1062 register and establishing log links when def is encountered.
1063 Note that we do not clear next_use array in order to save time,
1064 so we have to test whether the use is in the same basic block as def.
1066 There are a few cases below when we do not consider the definition or
1067 usage -- these are taken from original flow.c did. Don't ask me why it is
1068 done this way; I don't know and if it works, I don't want to know. */
1071 {
1073 {
1077 /* Log links are created only once. */
1081 {
1097 /* flow.c claimed:
1099 We don't build a LOG_LINK for hard registers contained
1100 in ASM_OPERANDs. If these registers get replaced,
1101 we might wind up changing the semantics of the insn,
1102 even if reload can make what appear to be valid
1103 assignments later. */
1108 /* Don't add duplicate links between instructions. */
1117 }
1122 }
1123 }
1126 }
1128 /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1129 true if we found a LOG_LINK that proves that A feeds B. This only works
1130 if there are no instructions between A and B which could have a link
1131 depending on A, since in that case we would not record a link for B.
1132 We also check the implicit dependency created by a cc0 setter/user
1133 pair. */
1135 static bool
1137 {
1145 }
1146 \f
1147 /* Main entry point for combiner. F is the first insn of the function.
1148 NREGS is the first unused pseudo-reg number.
1150 Return nonzero if the CFG was changed (e.g. if the combiner has
1151 turned an indirect jump instruction into a direct jump). */
1152 static int
1154 {
1159 basic_block last_bb;
1179 /* Allocate array for insn info. */
1187 /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1188 problems when, for example, we have j <<= 1 in a loop. */
1193 /* Scan all SETs and see if we can deduce anything about what
1194 bits are known to be zero for some registers and how many copies
1195 of the sign bit are known to exist for those registers.
1197 Also set any known values so that we can use it while searching
1198 for what bits are known to be set. */
1201 /* Allow the entry block and the first block to fall into the same EBB.
1202 Conceptually the incoming promotions are assigned to the entry block. */
1207 {
1212 label_tick++;
1220 {
1221 rtx links;
1233 insn);
1235 /* Record the current insn_cost of this instruction. */
1238 {
1241 }
1242 }
1243 }
1247 /* Now scan all the insns in forward order. */
1255 {
1258 /* Ignore instruction combination in basic blocks that are going to
1259 be removed as unreachable anyway. See PR82386. */
1267 label_tick++;
1277 {
1291 /* See if we know about function return values before this
1292 insn based upon SUBREG flags. */
1295 /* See if we can find hardregs and subreg of pseudos in
1296 narrower modes. This could help turning TRUNCATEs
1297 into SUBREGs. */
1300 /* Try this insn with each insn it links back to. */
1306 {
1309 }
1311 /* Try each sequence of three linked insns ending with this one. */
1315 {
1318 /* If the linked insn has been replaced by a note, then there
1319 is no point in pursuing this chain any further. */
1327 {
1330 }
1331 }
1333 /* Try to combine a jump insn that uses CC0
1334 with a preceding insn that sets CC0, and maybe with its
1335 logical predecessor as well.
1336 This is how we make decrement-and-branch insns.
1337 We need this special code because data flow connections
1338 via CC0 do not get entered in LOG_LINKS. */
1345 {
1356 }
1358 /* Do the same for an insn that explicitly references CC0. */
1365 {
1376 }
1378 /* Finally, see if any of the insns that this insn links to
1379 explicitly references CC0. If so, try this insn, that insn,
1380 and its predecessor if it sets CC0. */
1382 {
1394 }
1396 /* Try combining an insn with two different insns whose results it
1397 uses. */
1407 {
1410 }
1412 /* Try four-instruction combinations. */
1415 {
1419 /* If the linked insn has been replaced by a note, then there
1420 is no point in pursuing this chain any further. */
1425 {
1429 /* I0 -> I1 -> I2 -> I3. */
1435 {
1438 }
1439 /* I0, I1 -> I2, I2 -> I3. */
1446 {
1449 }
1450 }
1453 {
1457 /* I0 -> I2; I1, I2 -> I3. */
1463 {
1466 }
1467 /* I0 -> I1; I1, I2 -> I3. */
1473 {
1476 }
1477 }
1478 }
1480 /* Try this insn with each REG_EQUAL note it links back to. */
1482 {
1488 /* Avoid using a register that may already been marked
1489 dead by an earlier instruction. */
1497 {
1498 /* Temporarily replace the set's source with the
1499 contents of the REG_EQUAL note. The insn will
1500 be deleted or recognized by try_combine. */
1511 last_combined_insn);
1514 {
1517 }
1520 }
1521 }
1526 retry:
1527 ;
1528 }
1529 }
1536 /* Clean up. */
1542 {
1545 {
1548 }
1550 }
1560 /* Make recognizer allow volatile MEMs again. */
1564 }
1566 /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1568 static void
1570 {
1576 }
1577 \f
1578 /* Set up any promoted values for incoming argument registers. */
1580 static void
1582 {
1583 tree arg;
1588 {
1593 /* Only continue if the incoming argument is in a register. */
1597 /* Determine, if possible, whether all call sites of the current
1598 function lie within the current compilation unit. (This does
1599 take into account the exporting of a function via taking its
1600 address, and so forth.) */
1601 strictly_local
1604 /* The mode and signedness of the argument before any promotions happen
1605 (equal to the mode of the pseudo holding it at that stage). */
1609 /* The mode and signedness of the argument after any source language and
1610 TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1614 /* The mode and signedness of the argument as it is actually passed,
1615 see assign_parm_setup_reg in function.c. */
1619 /* The mode of the register in which the argument is being passed. */
1622 /* Eliminate sign extensions in the callee when:
1623 (a) A mode promotion has occurred; */
1626 /* (b) The mode of the register is the same as the mode of
1627 the argument as it is passed; */
1630 /* (c) There's no language level extension; */
1632 ;
1633 /* (c.1) All callers are from the current compilation unit. If that's
1634 the case we don't have to rely on an ABI, we only have to know
1635 what we're generating right now, and we know that we will do the
1636 mode1 to mode2 promotion with the given sign. */
1639 /* (c.2) The combination of the two promotions is useful. This is
1640 true when the signs match, or if the first promotion is unsigned.
1641 In the later case, (sign_extend (zero_extend x)) is the same as
1642 (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
1648 /* Record that the value was promoted from mode1 to mode3,
1649 so that any sign extension at the head of the current
1650 function may be eliminated. */
1654 }
1655 }
1657 /* If MODE has a precision lower than PREC and SRC is a non-negative constant
1658 that would appear negative in MODE, sign-extend SRC for use in nonzero_bits
1659 because some machines (maybe most) will actually do the sign-extension and
1660 this is the conservative approach.
1662 ??? For 2.5, try to tighten up the MD files in this regard instead of this
1663 kludge. */
1665 static rtx
1667 {
1668 scalar_int_mode int_mode;
1677 }
1679 /* Update RSP for pseudo-register X from INSN's REG_EQUAL note (if one exists)
1680 and SET. */
1682 static void
1684 rtx x)
1685 {
1695 {
1699 }
1701 /* Don't call nonzero_bits if it cannot change anything. */
1703 {
1712 }
1714 /* Don't call num_sign_bit_copies if it cannot change anything. */
1716 {
1719 {
1723 }
1726 }
1727 }
1729 /* Called via note_stores. If X is a pseudo that is narrower than
1730 HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1732 If we are setting only a portion of X and we can't figure out what
1733 portion, assume all bits will be used since we don't know what will
1734 be happening.
1736 Similarly, set how many bits of X are known to be copies of the sign bit
1737 at all locations in the function. This is the smallest number implied
1738 by any set of X. */
1740 static void
1742 {
1744 scalar_int_mode mode;
1748 /* If this register is undefined at the start of the file, we can't
1749 say what its contents were. */
1750 && ! REGNO_REG_SET_P
1754 {
1758 {
1762 }
1764 /* If this register is being initialized using itself, and the
1765 register is uninitialized in this basic block, and there are
1766 no LOG_LINKS which set the register, then part of the
1767 register is uninitialized. In that case we can't assume
1768 anything about the number of nonzero bits.
1770 ??? We could do better if we checked this in
1771 reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1772 could avoid making assumptions about the insn which initially
1773 sets the register, while still using the information in other
1774 insns. We would have to be careful to check every insn
1775 involved in the combination. */
1781 {
1788 {
1792 }
1793 }
1795 /* If this is a complex assignment, see if we can convert it into a
1796 simple assignment. */
1799 /* If this is a simple assignment, or we have a paradoxical SUBREG,
1800 set what we know about X. */
1806 else
1807 {
1810 }
1811 }
1812 }
1813 \f
1814 /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1815 optionally insns that were previously combined into I3 or that will be
1816 combined into the merger of INSN and I3. The order is PRED, PRED2,
1817 INSN, SUCC, SUCC2, I3.
1819 Return 0 if the combination is not allowed for any reason.
1821 If the combination is allowed, *PDEST will be set to the single
1822 destination of INSN and *PSRC to the single source, and this function
1823 will return 1. */
1825 static int
1829 {
1834 rtx link;
1839 {
1841 {
1846 }
1851 }
1855 /* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
1856 or a PARALLEL consisting of such a SET and CLOBBERs.
1858 If INSN has CLOBBER parallel parts, ignore them for our processing.
1859 By definition, these happen during the execution of the insn. When it
1860 is merged with another insn, all bets are off. If they are, in fact,
1861 needed and aren't also supplied in I3, they may be added by
1862 recog_for_combine. Otherwise, it won't match.
1864 We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1865 note.
1867 Get the source and destination of INSN. If more than one, can't
1868 combine. */
1874 {
1876 {
1880 {
1881 /* This is important to combine floating point insns
1882 for the SH4 port. */
1884 /* Combining an isolated USE doesn't make sense.
1885 We depend here on combinable_i3pat to reject them. */
1886 /* The code below this loop only verifies that the inputs of
1887 the SET in INSN do not change. We call reg_set_between_p
1888 to verify that the REG in the USE does not change between
1889 I3 and INSN.
1890 If the USE in INSN was for a pseudo register, the matching
1891 insn pattern will likely match any register; combining this
1892 with any other USE would only be safe if we knew that the
1893 used registers have identical values, or if there was
1894 something to tell them apart, e.g. different modes. For
1895 now, we forgo such complicated tests and simply disallow
1896 combining of USES of pseudo registers with any other USE. */
1899 {
1904 do
1905 {
1915 }
1917 }
1920 /* We can ignore CLOBBERs. */
1925 /* Ignore SETs whose result isn't used but not those that
1926 have side-effects. */
1932 /* If we have already found a SET, this is a second one and
1933 so we cannot combine with this insn. */
1941 /* Anything else means we can't combine. */
1943 }
1944 }
1947 /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1948 so don't do anything with it. */
1951 }
1952 else
1958 /* The simplification in expand_field_assignment may call back to
1959 get_last_value, so set safe guard here. */
1965 /* Do not eliminate user-specified register if it is in an
1966 asm input because we may break the register asm usage defined
1967 in GCC manual if allow to do so.
1968 Be aware that this may cover more cases than we expect but this
1969 should be harmless. */
1974 /* Don't eliminate a store in the stack pointer. */
1976 /* Don't combine with an insn that sets a register to itself if it has
1977 a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1979 /* Can't merge an ASM_OPERANDS. */
1981 /* Can't merge a function call. */
1983 /* Don't eliminate a function call argument. */
1989 /* Don't substitute into an incremented register. */
1993 /* Don't substitute into a non-local goto, this confuses CFG. */
1995 /* Make sure that DEST is not used after INSN but before SUCC, or
1996 after SUCC and before SUCC2, or after SUCC2 but before I3. */
1997 || (!all_adjacent
1998 && ((succ2
2003 || (succ
2004 /* SUCC and SUCC2 can be split halves from a PARALLEL; in
2005 that case SUCC is not in the insn stream, so use SUCC2
2006 instead for this test. */
2008 succ2
2011 /* Make sure that the value that is to be substituted for the register
2012 does not use any registers whose values alter in between. However,
2013 If the insns are adjacent, a use can't cross a set even though we
2014 think it might (this can happen for a sequence of insns each setting
2015 the same destination; last_set of that register might point to
2016 a NOTE). If INSN has a REG_EQUIV note, the register is always
2017 equivalent to the memory so the substitution is valid even if there
2018 are intervening stores. Also, don't move a volatile asm or
2019 UNSPEC_VOLATILE across any other insns. */
2020 || (! all_adjacent
2026 /* Don't combine across a CALL_INSN, because that would possibly
2027 change whether the life span of some REGs crosses calls or not,
2028 and it is a pain to update that information.
2029 Exception: if source is a constant, moving it later can't hurt.
2030 Accept that as a special case. */
2034 /* DEST must either be a REG or CC0. */
2036 {
2037 /* If register alignment is being enforced for multi-word items in all
2038 cases except for parameters, it is possible to have a register copy
2039 insn referencing a hard register that is not allowed to contain the
2040 mode being copied and which would not be valid as an operand of most
2041 insns. Eliminate this problem by not combining with such an insn.
2043 Also, on some machines we don't want to extend the life of a hard
2044 register. */
2049 /* Don't extend the life of a hard register unless it is
2050 user variable (if we have few registers) or it can't
2051 fit into the desired register (meaning something special
2052 is going on).
2053 Also avoid substituting a return register into I3, because
2054 reload can't handle a conflict with constraints of other
2055 inputs. */
2060 }
2068 {
2071 /* If the clobber represents an earlyclobber operand, we must not
2072 substitute an expression containing the clobbered register.
2073 As we do not analyze the constraint strings here, we have to
2074 make the conservative assumption. However, if the register is
2075 a fixed hard reg, the clobber cannot represent any operand;
2076 we leave it up to the machine description to either accept or
2077 reject use-and-clobber patterns. */
2083 }
2085 /* If INSN contains anything volatile, or is an `asm' (whether volatile
2086 or not), reject, unless nothing volatile comes between it and I3 */
2089 {
2090 /* Make sure neither succ nor succ2 contains a volatile reference. */
2095 /* We'll check insns between INSN and I3 below. */
2096 }
2098 /* If INSN is an asm, and DEST is a hard register, reject, since it has
2099 to be an explicit register variable, and was chosen for a reason. */
2105 /* If INSN contains volatile references (specifically volatile MEMs),
2106 we cannot combine across any other volatile references.
2107 Even if INSN doesn't contain volatile references, any intervening
2108 volatile insn might affect machine state. */
2111 ? volatile_refs_p
2118 /* If INSN contains an autoincrement or autodecrement, make sure that
2119 register is not used between there and I3, and not already used in
2120 I3 either. Neither must it be used in PRED or SUCC, if they exist.
2121 Also insist that I3 not be a jump if using LRA; if it were one
2122 and the incremented register were spilled, we would lose.
2123 Reload handles this correctly. */
2141 /* Don't combine an insn that follows a CC0-setting insn.
2142 An insn that uses CC0 must not be separated from the one that sets it.
2143 We do, however, allow I2 to follow a CC0-setting insn if that insn
2144 is passed as I1; in that case it will be deleted also.
2145 We also allow combining in this case if all the insns are adjacent
2146 because that would leave the two CC0 insns adjacent as well.
2147 It would be more logical to test whether CC0 occurs inside I1 or I2,
2148 but that would be much slower, and this ought to be equivalent. */
2151 {
2156 }
2158 /* If we get here, we have passed all the tests and the combination is
2159 to be allowed. */
2165 }
2166 \f
2167 /* LOC is the location within I3 that contains its pattern or the component
2168 of a PARALLEL of the pattern. We validate that it is valid for combining.
2170 One problem is if I3 modifies its output, as opposed to replacing it
2171 entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2172 doing so would produce an insn that is not equivalent to the original insns.
2174 Consider:
2176 (set (reg:DI 101) (reg:DI 100))
2177 (set (subreg:SI (reg:DI 101) 0) <foo>)
2179 This is NOT equivalent to:
2181 (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2182 (set (reg:DI 101) (reg:DI 100))])
2184 Not only does this modify 100 (in which case it might still be valid
2185 if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2187 We can also run into a problem if I2 sets a register that I1
2188 uses and I1 gets directly substituted into I3 (not via I2). In that
2189 case, we would be getting the wrong value of I2DEST into I3, so we
2190 must reject the combination. This case occurs when I2 and I1 both
2191 feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2192 If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2193 of a SET must prevent combination from occurring. The same situation
2194 can occur for I0, in which case I0_NOT_IN_SRC is set.
2196 Before doing the above check, we first try to expand a field assignment
2197 into a set of logical operations.
2199 If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2200 we place a register that is both set and used within I3. If more than one
2201 such register is detected, we fail.
2203 Return 1 if the combination is valid, zero otherwise. */
2205 static int
2208 {
2212 {
2217 rtx subdest;
2224 /* Check for the case where I3 modifies its output, as discussed
2225 above. We don't want to prevent pseudos from being combined
2226 into the address of a MEM, so only prevent the combination if
2227 i1 or i2 set the same MEM. */
2237 /* This is the same test done in can_combine_p except we can't test
2238 all_adjacent; we don't have to, since this instruction will stay
2239 in place, thus we are not considering increasing the lifetime of
2240 INNER_DEST.
2242 Also, if this insn sets a function argument, combining it with
2243 something that might need a spill could clobber a previous
2244 function argument; the all_adjacent test in can_combine_p also
2245 checks this; here, we do a more specific test for this case. */
2255 /* If DEST is used in I3, it is being killed in this insn, so
2256 record that for later. We have to consider paradoxical
2257 subregs here, since they kill the whole register, but we
2258 ignore partial subregs, STRICT_LOW_PART, etc.
2259 Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2260 STACK_POINTER_REGNUM, since these are always considered to be
2261 live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2269 && (HARD_FRAME_POINTER_IS_FRAME_POINTER
2275 {
2280 }
2281 }
2284 {
2291 }
2294 }
2295 \f
2296 /* Return 1 if X is an arithmetic expression that contains a multiplication
2297 and division. We don't count multiplications by powers of two here. */
2299 static int
2301 {
2303 {
2319 }
2320 }
2321 \f
2322 /* Determine whether INSN can be used in a combination. Return nonzero if
2323 not. This is used in try_combine to detect early some cases where we
2324 can't perform combinations. */
2326 static int
2328 {
2329 rtx set;
2332 /* If this isn't really an insn, we can't do anything.
2333 This can occur when flow deletes an insn that it has merged into an
2334 auto-increment address. */
2338 /* Never combine loads and stores involving hard regs that are likely
2339 to be spilled. The register allocator can usually handle such
2340 reg-reg moves by tying. If we allow the combiner to make
2341 substitutions of likely-spilled regs, reload might die.
2342 As an exception, we allow combinations involving fixed regs; these are
2343 not available to the register allocator so there's no risk involved. */
2357 #ifdef LEAF_REGISTERS
2359 #else
2360 )
2361 #endif
2368 }
2370 struct likely_spilled_retval_info
2371 {
2374 };
2376 /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2377 hard registers that are known to be written to / clobbered in full. */
2378 static void
2380 {
2397 else
2400 }
2402 /* Return nonzero iff part of the return value is live during INSN, and
2403 it is likely spilled. This can happen when more than one insn is needed
2404 to copy the return value, e.g. when we consider to combine into the
2405 second copy insn for a complex value. */
2407 static int
2409 {
2411 rtx reg;
2414 /* We assume here that no machine mode needs more than
2415 32 hard registers when the value overlaps with a register
2416 for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2431 /* Disregard parts of the return value that are set later. */
2440 /* Check if any of the (probably) live return value registers is
2441 likely spilled. */
2442 nregs --;
2443 do
2444 {
2450 }
2452 /* Adjust INSN after we made a change to its destination.
2454 Changing the destination can invalidate notes that say something about
2455 the results of the insn and a LOG_LINK pointing to the insn. */
2457 static void
2459 {
2460 /* For notes, be conservative and simply remove them. */
2463 /* The new insn will have a destination that was previously the destination
2464 of an insn just above it. Call distribute_links to make a LOG_LINK from
2465 the next use of that destination. */
2481 }
2483 /* Return TRUE if combine can reuse reg X in mode MODE.
2484 ADDED_SETS is nonzero if the original set is still required. */
2485 static bool
2487 {
2493 /* Don't change between modes with different underlying register sizes,
2494 since this could lead to invalid subregs. */
2500 /* Allow hard registers if the new mode is legal, and occupies no more
2501 registers than the old mode. */
2506 /* Or a pseudo that is only used once. */
2509 && !added_sets
2511 }
2514 /* Check whether X, the destination of a set, refers to part of
2515 the register specified by REG. */
2517 static bool
2519 {
2520 /* Check that reg is an integer mode register. */
2531 }
2533 /* Delete the unconditional jump INSN and adjust the CFG correspondingly.
2534 Note that the INSN should be deleted *after* removing dead edges, so
2535 that the kept edge is the fallthrough edge for a (set (pc) (pc))
2536 but not for a (set (pc) (label_ref FOO)). */
2538 static void
2540 {
2548 {
2553 /* Remove barriers from the footer if there are any. */
2556 {
2559 else
2563 }
2566 }
2567 }
2569 /* Return whether PAT is a PARALLEL of exactly N register SETs followed
2570 by an arbitrary number of CLOBBERs. */
2571 static bool
2573 {
2588 {
2595 }
2597 }
2599 /* Return whether INSN, a PARALLEL of N register SETs (and maybe some
2600 CLOBBERs), can be split into individual SETs in that order, without
2601 changing semantics. */
2602 static bool
2604 {
2612 {
2621 }
2624 }
2626 /* Return whether X is just a single set, with the source
2627 a general_operand. */
2628 static bool
2630 {
2635 }
2637 /* Callback function to count autoincs. */
2639 static int
2641 {
2645 }
2647 /* Try to combine the insns I0, I1 and I2 into I3.
2648 Here I0, I1 and I2 appear earlier than I3.
2649 I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2650 I3.
2652 If we are combining more than two insns and the resulting insn is not
2653 recognized, try splitting it into two insns. If that happens, I2 and I3
2654 are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2655 Otherwise, I0, I1 and I2 are pseudo-deleted.
2657 Return 0 if the combination does not work. Then nothing is changed.
2658 If we did the combination, return the insn at which combine should
2659 resume scanning.
2661 Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
2662 new direct jump instruction.
2664 LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2665 been I3 passed to an earlier try_combine within the same basic
2666 block. */
2671 {
2672 /* New patterns for I3 and I2, respectively. */
2676 /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2677 dead. */
2679 /* Total number of SETs to put into I3. */
2681 /* Nonzero if I2's or I1's body now appears in I3. */
2683 /* INSN_CODEs for new I3, new I2, and user of condition code. */
2685 /* Contains I3 if the destination of I3 is used in its source, which means
2686 that the old life of I3 is being killed. If that usage is placed into
2687 I2 and not in I3, a REG_DEAD note must be made. */
2689 /* SET_DEST and SET_SRC of I2, I1 and I0. */
2691 /* Copy of SET_SRC of I1 and I0, if needed. */
2693 /* Set if I2DEST was reused as a scratch register. */
2695 /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2697 /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2702 /* Notes that must be added to REG_NOTES in I3 and I2. */
2704 /* Notes that we substituted I3 into I2 instead of the normal case. */
2706 /* Notes that I1, I2 or I3 is a MULT operation. */
2716 rtx temp_expr;
2719 rtx new_other_notes;
2723 /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
2724 never be). */
2728 /* Only try four-insn combinations when there's high likelihood of
2729 success. Look for simple insns, such as loads of constants or
2730 binary operations involving a constant. */
2732 {
2742 {
2745 rtx src;
2750 {
2753 }
2755 ngood++;
2758 nshift++;
2759 }
2761 /* If I0 loads a memory and I3 sets the same memory, then I1 and I2
2762 are likely manipulating its value. Ideally we'll be able to combine
2763 all four insns into a bitfield insertion of some kind.
2765 Note the source in I0 might be inside a sign/zero extension and the
2766 memory modes in I0 and I3 might be different. So extract the address
2767 from the destination of I3 and search for it in the source of I0.
2769 In the event that there's a match but the source/dest do not actually
2770 refer to the same memory, the worst that happens is we try some
2771 combinations that we wouldn't have otherwise. */
2773 /* Ensure the source of SET0 is a MEM, possibly buried inside
2774 an extension. */
2780 /* Ensure the destination of SET3 is a MEM. */
2782 /* Would it be better to extract the base address for the MEM
2783 in SET3 and look for that? I don't have cases where it matters
2784 but I could envision such cases. */
2790 }
2792 /* Exit early if one of the insns involved can't be used for
2793 combinations. */
2804 combine_attempts++;
2807 /* Reset the hard register usage information. */
2811 {
2818 else
2828 }
2830 /* If multiple insns feed into one of I2 or I3, they can be in any
2831 order. To simplify the code below, reorder them in sequence. */
2842 /* First check for one important special case that the code below will
2843 not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2844 and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2845 we may be able to replace that destination with the destination of I3.
2846 This occurs in the common code where we compute both a quotient and
2847 remainder into a structure, in which case we want to do the computation
2848 directly into the structure to avoid register-register copies.
2850 Note that this case handles both multiple sets in I2 and also cases
2851 where I2 has a number of CLOBBERs inside the PARALLEL.
2853 We make very conservative checks below and only try to handle the
2854 most common cases of this. For example, we only handle the case
2855 where I2 and I3 are adjacent to avoid making difficult register
2856 usage tests. */
2864 /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2865 below would need to check what is inside (and reg_overlap_mentioned_p
2866 doesn't support those codes anyway). Don't allow those destinations;
2867 the resulting insn isn't likely to be recognized anyway. */
2873 {
2876 /* Make sure that the destination of I3,
2877 which we are going to substitute into one output of I2,
2878 is not used within another output of I2. We must avoid making this:
2879 (parallel [(set (mem (reg 69)) ...)
2880 (set (reg 69) ...)])
2881 which is not well-defined as to order of actions.
2882 (Besides, reload can't handle output reloads for this.)
2884 The problem can also happen if the dest of I3 is a memory ref,
2885 if another dest in I2 is an indirect memory ref.
2887 Neither can this PARALLEL be an asm. We do not allow combining
2888 that usually (see can_combine_p), so do not here either. */
2891 {
2900 }
2906 {
2907 combine_merges++;
2917 /* Replace the dest in I2 with our dest and make the resulting
2918 insn the new pattern for I3. Then skip to where we validate
2919 the pattern. Everything was set up above. */
2924 }
2925 }
2927 /* If I2 is setting a pseudo to a constant and I3 is setting some
2928 sub-part of it to another constant, merge them by making a new
2929 constant. */
2937 {
2944 {
2949 {
2955 }
2956 }
2957 else
2958 {
2962 {
2965 }
2966 }
2969 {
2970 /* If this is the low part, we're done. */
2972 ;
2973 /* Handle the case where inner is twice the size of outer. */
2977 /* Otherwise give up for now. */
2978 else
2980 }
2983 {
2991 combine_merges++;
2998 /* Replace the source in I2 with the new constant and make the
2999 resulting insn the new pattern for I3. Then skip to where we
3000 validate the pattern. Everything was set up above. */
3006 /* The dest of I3 has been replaced with the dest of I2. */
3009 }
3010 }
3012 /* If we have no I1 and I2 looks like:
3013 (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
3014 (set Y OP)])
3015 make up a dummy I1 that is
3016 (set Y OP)
3017 and change I2 to be
3018 (set (reg:CC X) (compare:CC Y (const_int 0)))
3020 (We can ignore any trailing CLOBBERs.)
3022 This undoes a previous combination and allows us to match a branch-and-
3023 decrement insn. */
3035 {
3036 /* We make I1 with the same INSN_UID as I2. This gives it
3037 the same DF_INSN_LUID for value tracking. Our fake I1 will
3038 never appear in the insn stream so giving it the same INSN_UID
3039 as I2 will not cause a problem. */
3052 }
3054 /* If I2 is a PARALLEL of two SETs of REGs (and perhaps some CLOBBERs),
3055 make those two SETs separate I1 and I2 insns, and make an I0 that is
3056 the original I1. */
3064 {
3065 /* If there is no I1, there is no I0 either. */
3068 /* We make I1 with the same INSN_UID as I2. This gives it
3069 the same DF_INSN_LUID for value tracking. Our fake I1 will
3070 never appear in the insn stream so giving it the same INSN_UID
3071 as I2 will not cause a problem. */
3079 }
3081 /* Verify that I2 and maybe I1 and I0 can be combined into I3. */
3083 {
3088 }
3090 {
3095 }
3097 {
3102 }
3104 /* Record whether i2 and i3 are trivial moves. */
3108 /* Record whether I2DEST is used in I2SRC and similarly for the other
3109 cases. Knowing this will help in register status updating below. */
3120 /* For the earlier insns, determine which of the subsequent ones they
3121 feed. */
3128 /* Ensure that I3's pattern can be the destination of combines. */
3134 {
3137 }
3139 /* See if any of the insns is a MULT operation. Unless one is, we will
3140 reject a combination that is, since it must be slower. Be conservative
3141 here. */
3149 /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
3150 We used to do this EXCEPT in one case: I3 has a post-inc in an
3151 output operand. However, that exception can give rise to insns like
3152 mov r3,(r3)+
3153 which is a famous insn on the PDP-11 where the value of r3 used as the
3154 source was model-dependent. Avoid this sort of thing. */
3156 #if 0
3162 /* It's not the exception. */
3163 #endif
3165 {
3166 rtx link;
3172 {
3175 }
3176 }
3178 /* See if the SETs in I1 or I2 need to be kept around in the merged
3179 instruction: whenever the value set there is still needed past I3.
3180 For the SET in I2, this is easy: we see if I2DEST dies or is set in I3.
3182 For the SET in I1, we have two cases: if I1 and I2 independently feed
3183 into I3, the set in I1 needs to be kept around unless I1DEST dies
3184 or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
3185 in I1 needs to be kept around unless I1DEST dies or is set in either
3186 I2 or I3. The same considerations apply to I0. */
3193 else
3201 else
3204 /* We are about to copy insns for the case where they need to be kept
3205 around. Check that they can be copied in the merged instruction. */
3211 {
3214 }
3216 /* Count how many auto_inc expressions there were in the original insns;
3217 we need to have the same number in the resulting patterns. */
3226 /* If the set in I2 needs to be kept around, we must make a copy of
3227 PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3228 PATTERN (I2), we are only substituting for the original I1DEST, not into
3229 an already-substituted copy. This also prevents making self-referential
3230 rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3231 I2DEST. */
3234 {
3237 else
3239 }
3242 {
3245 else
3247 }
3250 {
3253 else
3255 }
3257 combine_merges++;
3259 /* Substitute in the latest insn for the regs set by the earlier ones. */
3265 /* Many machines that don't use CC0 have insns that can both perform an
3266 arithmetic operation and set the condition code. These operations will
3267 be represented as a PARALLEL with the first element of the vector
3268 being a COMPARE of an arithmetic operation with the constant zero.
3269 The second element of the vector will set some pseudo to the result
3270 of the same arithmetic operation. If we simplify the COMPARE, we won't
3271 match such a pattern and so will generate an extra insn. Here we test
3272 for this case, where both the comparison and the operation result are
3273 needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3274 I2SRC. Later we will make the PARALLEL that contains I2. */
3280 {
3281 rtx newpat_dest;
3287 scalar_int_mode mode;
3296 {
3302 }
3304 /* Do the rest only if op1 is const0_rtx, which may be the
3305 result of simplification. */
3307 {
3308 /* If a single use of the CC is found, prepare to modify it
3309 when SELECT_CC_MODE returns a new CC-class mode, or when
3310 the above simplify_compare_const() returned a new comparison
3311 operator. undobuf.other_insn is assigned the CC use insn
3312 when modifying it. */
3314 {
3315 #ifdef SELECT_CC_MODE
3316 machine_mode new_mode
3321 {
3326 else
3327 {
3330 }
3331 }
3332 #endif
3333 /* Cases for modifying the CC-using comparison. */
3335 /* ??? Do we need to verify the zero rtx? */
3337 {
3338 /* Replace cc_use_loc with entire new RTX. */
3343 }
3345 {
3346 /* Just replace the CC reg with a new mode. */
3349 }
3350 }
3352 /* Now we modify the current newpat:
3353 First, SET_DEST(newpat) is updated if the CC mode has been
3354 altered. For targets without SELECT_CC_MODE, this should be
3355 optimized away. */
3358 /* This is always done to propagate i2src into newpat. */
3361 /* Create new version of i2pat if needed; the below PARALLEL
3362 creation needs this to work correctly. */
3366 }
3367 }
3370 {
3371 /* It is possible that the source of I2 or I1 may be performing
3372 an unneeded operation, such as a ZERO_EXTEND of something
3373 that is known to have the high part zero. Handle that case
3374 by letting subst look at the inner insns.
3376 Another way to do this would be to have a function that tries
3377 to simplify a single insn instead of merging two or more
3378 insns. We don't do this because of the potential of infinite
3379 loops and because of the potential extra memory required.
3380 However, doing it the way we are is a bit of a kludge and
3381 doesn't catch all cases.
3383 But only do this if -fexpensive-optimizations since it slows
3384 things down and doesn't usually win.
3386 This is not done in the COMPARE case above because the
3387 unmodified I2PAT is used in the PARALLEL and so a pattern
3388 with a modified I2SRC would not match. */
3391 {
3392 /* Pass pc_rtx so no substitutions are done, just
3393 simplifications. */
3395 {
3398 }
3402 }
3407 /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3408 copy of I2SRC each time we substitute it, in order to avoid creating
3409 self-referential RTL when we will be substituting I1SRC for I1DEST
3410 later. Likewise if I0 feeds into I2, either directly or indirectly
3411 through I1, and I0DEST is in I0SRC. */
3418 /* Record whether I2's body now appears within I3's body. */
3420 }
3422 /* If we already got a failure, don't try to do more. Otherwise, try to
3423 substitute I1 if we have it. */
3426 {
3427 /* Before we can do this substitution, we must redo the test done
3428 above (see detailed comments there) that ensures I1DEST isn't
3429 mentioned in any SETs in NEWPAT that are field assignments. */
3432 {
3435 }
3440 /* If the following substitution will modify I1SRC, make a copy of it
3441 for the case where it is substituted for I1DEST in I2PAT later. */
3445 /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3446 copy of I1SRC each time we substitute it, in order to avoid creating
3447 self-referential RTL when we will be substituting I0SRC for I0DEST
3448 later. */
3453 /* Record whether I1's body now appears within I3's body. */
3455 }
3457 /* Likewise for I0 if we have it. */
3460 {
3463 {
3466 }
3468 /* If the following substitution will modify I0SRC, make a copy of it
3469 for the case where it is substituted for I0DEST in I1PAT later. */
3472 /* And a copy for I0DEST in I2PAT substitution. */
3481 }
3484 {
3489 {
3494 }
3495 }
3497 /* Fail if an autoincrement side-effect has been duplicated. Be careful
3498 to count all the ways that I2SRC and I1SRC can be used. */
3509 /* Fail if we tried to make a new register. */
3511 /* Fail if we couldn't do something and have a CLOBBER. */
3513 /* Fail if this new pattern is a MULT and we didn't have one before
3514 at the outer level. */
3517 {
3520 }
3522 /* If the actions of the earlier insns must be kept
3523 in addition to substituting them into the latest one,
3524 we must make a new PARALLEL for the latest insn
3525 to hold additional the SETs. */
3528 {
3530 combine_extras++;
3533 {
3539 }
3540 else
3541 {
3546 }
3552 {
3558 }
3560 {
3569 }
3570 }
3572 validate_replacement:
3574 /* Note which hard regs this insn has as inputs. */
3577 /* If recog_for_combine fails, it strips existing clobbers. If we'll
3578 consider splitting this pattern, we might need these clobbers. */
3581 {
3587 }
3589 /* We have recognized nothing yet. */
3592 /* See if this is a PARALLEL of two SETs where one SET's destination is
3593 a register that is unused and this isn't marked as an instruction that
3594 might trap in an EH region. In that case, we just need the other SET.
3595 We prefer this over the PARALLEL.
3597 This can occur when simplifying a divmod insn. We *must* test for this
3598 case here because the code below that splits two independent SETs doesn't
3599 handle this case correctly when it updates the register status.
3601 It's pointless doing this if we originally had two sets, one from
3602 i3, and one from i2. Combining then splitting the parallel results
3603 in the original i2 again plus an invalid insn (which we delete).
3604 The net effect is only to move instructions around, which makes
3605 debug info less accurate.
3607 If the remaining SET came from I2 its destination should not be used
3608 between I2 and I3. See PR82024. */
3613 {
3624 {
3627 }
3636 {
3641 {
3647 }
3648 }
3652 }
3654 /* Is the result of combination a valid instruction? */
3658 /* If we were combining three insns and the result is a simple SET
3659 with no ASM_OPERANDS that wasn't recognized, try to split it into two
3660 insns. There are two ways to do this. It can be split using a
3661 machine-specific method (like when you have an addition of a large
3662 constant) or by combine in the function find_split_point. */
3666 {
3670 /* See if the MD file can split NEWPAT. If it can't, see if letting it
3671 use I2DEST as a scratch register will help. In the latter case,
3672 convert I2DEST to the mode of the source of NEWPAT if we can. */
3676 /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3677 inputs of NEWPAT. */
3679 /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3680 possible to try that as a scratch reg. This would require adding
3681 more code to make it work though. */
3684 {
3687 /* ??? Reusing i2dest without resetting the reg_stat entry for it
3688 (temporarily, until we are committed to this instruction
3689 combination) does not work: for example, any call to nonzero_bits
3690 on the register (from a splitter in the MD file, for example)
3691 will get the old information, which is invalid.
3693 Since nowadays we can create registers during combine just fine,
3694 we should just create a new one here, not reuse i2dest. */
3696 /* First try to split using the original register as a
3697 scratch register. */
3701 i2dest)));
3704 /* If that didn't work, try changing the mode of I2DEST if
3705 we can. */
3710 {
3712 rtx ni2dest;
3716 else
3717 {
3720 }
3722 parallel = (gen_rtx_PARALLEL
3726 ni2dest))));
3731 {
3739 }
3740 }
3743 }
3745 /* If recog_for_combine has discarded clobbers, try to use them
3746 again for the split. */
3748 {
3751 }
3754 {
3759 }
3763 {
3773 /* If I2 or I3 has multiple SETs, we won't know how to track
3774 register status, so don't use these insns. If I2's destination
3775 is used between I2 and I3, we also can't use these insns. */
3785 /* It is possible that both insns now set the destination of I3.
3786 If so, we must show an extra use of it. */
3789 {
3808 }
3809 }
3811 /* If we can split it and use I2DEST, go ahead and see if that
3812 helps things be recognized. Verify that none of the registers
3813 are set between I2 and I3. */
3817 /* We need I2DEST in the proper mode. If it is a hard register
3818 or the only use of a pseudo, we can change its mode.
3819 Make sure we don't change a hard register to have a mode that
3820 isn't valid for it, or change the number of registers. */
3827 /* We can't overwrite I2DEST if its value is still used by
3828 NEWPAT. */
3830 {
3839 /* *SPLIT may be part of I2SRC, so make sure we have the
3840 original expression around for later debug processing.
3841 We should not need I2SRC any more in other cases. */
3844 else
3847 /* Get NEWDEST as a register in the proper mode. We have already
3848 validated that we can do this. */
3850 {
3853 else
3854 {
3857 }
3858 }
3860 /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3861 an ASHIFT. This can occur if it was inside a PLUS and hence
3862 appeared to be a memory address. This is a kludge. */
3867 {
3871 /* Update split_code because we may not have a multiply
3872 anymore. */
3874 }
3876 /* Similarly for (plus (mult FOO (const_int pow2))). */
3882 {
3887 i_rtx));
3888 /* Update split_code because we may not have a multiply
3889 anymore. */
3891 }
3893 #ifdef INSN_SCHEDULING
3894 /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3895 be written as a ZERO_EXTEND. */
3897 {
3898 /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3899 what it really is. */
3904 else
3907 }
3908 #endif
3910 /* Attempt to split binary operators using arithmetic identities. */
3914 {
3921 /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3923 {
3928 }
3929 /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3936 {
3942 /* Split both "((X op Y) op X) op Y" and
3943 "((X op Y) op Y) op X" as "T op T" where T is
3944 "X op Y". */
3947 {
3952 }
3953 /* Split "((X op X) op Y) op Y)" as "T op T" where
3954 T is "X op Y". */
3956 {
3962 }
3963 }
3964 }
3967 {
3970 }
3974 /* recog_for_combine might have added CLOBBERs to newi2pat.
3975 Make sure NEWPAT does not depend on the clobbered regs. */
3979 {
3982 {
3985 }
3986 }
3988 /* If the split point was a MULT and we didn't have one before,
3989 don't use one now. */
3992 }
3993 }
3995 /* Check for a case where we loaded from memory in a narrow mode and
3996 then sign extended it, but we need both registers. In that case,
3997 we have a PARALLEL with both loads from the same memory location.
3998 We can split this into a load from memory followed by a register-register
3999 copy. This saves at least one insn, more if register allocation can
4000 eliminate the copy.
4002 We cannot do this if the destination of the first assignment is a
4003 condition code register or cc0. We eliminate this case by making sure
4004 the SET_DEST and SET_SRC have the same mode.
4006 We cannot do this if the destination of the second assignment is
4007 a register that we have already assumed is zero-extended. Similarly
4008 for a SUBREG of such a register. */
4027 BITS_PER_WORD)
4029 HOST_BITS_PER_INT)
4037 BITS_PER_WORD)
4039 HOST_BITS_PER_INT)
4046 {
4047 rtx ni2dest;
4061 }
4063 /* Similarly, check for a case where we have a PARALLEL of two independent
4064 SETs but we started with three insns. In this case, we can do the sets
4065 as two separate insns. This case occurs when some SET allows two
4066 other insns to combine, but the destination of that SET is still live.
4068 Also do this if we started with two insns and (at least) one of the
4069 resulting sets is a noop; this noop will be deleted later.
4071 Also do this if we started with two insns neither of which was a simple
4072 move. */
4079 && (i1
4093 {
4097 /* Normally, it doesn't matter which of the two is done first,
4098 but the one that references cc0 can't be the second, and
4099 one which uses any regs/memory set in between i2 and i3 can't
4100 be first. The PARALLEL might also have been pre-existing in i3,
4101 so we need to make sure that we won't wrongly hoist a SET to i2
4102 that would conflict with a death note present in there, or would
4103 have its dest modified between i2 and i3. */
4112 /* If I3 is a jump, ensure that set0 is a jump so that
4113 we do not create invalid RTL. */
4115 )
4116 {
4119 }
4128 /* If I3 is a jump, ensure that set1 is a jump so that
4129 we do not create invalid RTL. */
4131 )
4132 {
4135 }
4136 else
4137 {
4140 }
4145 {
4146 /* recog_for_combine might have added CLOBBERs to newi2pat.
4147 Make sure NEWPAT does not depend on the clobbered regs. */
4149 {
4152 {
4155 {
4158 }
4159 }
4160 }
4166 }
4167 }
4169 /* If it still isn't recognized, fail and change things back the way they
4170 were. */
4172 /* Is the result a reasonable ASM_OPERANDS? */
4174 {
4177 }
4179 /* If we had to change another insn, make sure it is valid also. */
4181 {
4189 {
4192 }
4193 }
4195 /* If I2 is the CC0 setter and I3 is the CC0 user then check whether
4196 they are adjacent to each other or not. */
4198 {
4202 {
4205 }
4206 }
4208 /* Only allow this combination if insn_cost reports that the
4209 replacement instructions are cheaper than the originals. */
4211 {
4214 }
4217 {
4222 {
4227 /* Temporarily revert mode back. */
4231 {
4232 /* If we used i2dest as a scratch register with a
4233 different mode, substitute it for the original
4234 i2src while its original mode is temporarily
4235 restored, and then clear i2scratch so that we don't
4236 do it again later. */
4238 this_basic_block);
4240 /* Put back the new mode. */
4242 }
4243 else
4244 {
4249 {
4252 }
4253 else
4254 {
4261 }
4263 /* We're dealing with a reg that changed mode but not
4264 meaning, so we want to turn it into a subreg for
4265 the new mode. However, because of REG sharing and
4266 because its mode had already changed, we have to do
4267 it in two steps. First, replace any debug uses of
4268 reg, with its original mode temporarily restored,
4269 with this copy we have created; then, replace the
4270 copy with the SUBREG of the original shared reg,
4271 once again changed to the new mode. */
4273 this_basic_block);
4277 this_basic_block);
4278 }
4279 }
4280 }
4282 /* If we will be able to accept this, we have made a
4283 change to the destination of I3. This requires us to
4284 do a few adjustments. */
4287 {
4290 }
4292 /* We now know that we can do this combination. Merge the insns and
4293 update the status of registers and LOG_LINKS. */
4296 {
4301 /* If any of the notes in OTHER_INSN were REG_DEAD or REG_UNUSED,
4302 ensure that they are still valid. Then add any non-duplicate
4303 notes added by recog_for_combine. */
4305 {
4314 /* Simply drop equal note since it may be no longer valid
4315 for other_insn. It may be possible to record that CC
4316 register is changed and only discard those notes, but
4317 in practice it's unnecessary complication and doesn't
4318 give any meaningful improvement.
4320 See PR78559. */
4324 }
4328 NULL_RTX);
4329 }
4332 {
4333 /* I3 now uses what used to be its destination and which is now
4334 I2's destination. This requires us to do a few adjustments. */
4337 }
4340 {
4341 /* We might need a LOG_LINK from I3 to I2. But then we used to
4342 have one, so we still will.
4344 However, some later insn might be using I2's dest and have
4345 a LOG_LINK pointing at I3. We should change it to point at
4346 I2 instead. */
4348 /* newi2pat is usually a SET here; however, recog_for_combine might
4349 have added some clobbers. */
4354 /* It can only be a SET of a REG or of a SUBREG of a REG. */
4359 !done
4360 && insn
4364 {
4368 {
4372 }
4373 }
4374 }
4376 {
4381 /* Compute which registers we expect to eliminate. newi2pat may be setting
4382 either i3dest or i2dest, so we must check it. */
4385 || !i2dest_killed
4387 /* For i1, we need to compute both local elimination and global
4388 elimination information with respect to newi2pat because i1dest
4389 may be the same as i3dest, in which case newi2pat may be setting
4390 i1dest. Global information is used when distributing REG_DEAD
4391 note for i2 and i3, in which case it does matter if newi2pat sets
4392 i1dest or not.
4394 Local information is used when distributing REG_DEAD note for i1,
4395 in which case it doesn't matter if newi2pat sets i1dest or not.
4396 See PR62151, if we have four insns combination:
4397 i0: r0 <- i0src
4398 i1: r1 <- i1src (using r0)
4399 REG_DEAD (r0)
4400 i2: r0 <- i2src (using r1)
4401 i3: r3 <- i3src (using r0)
4402 ix: using r0
4403 From i1's point of view, r0 is eliminated, no matter if it is set
4404 by newi2pat or not. In other words, REG_DEAD info for r0 in i1
4405 should be discarded.
4407 Note local information only affects cases in forms like "I1->I2->I3",
4408 "I0->I1->I2->I3" or "I0&I1->I2, I2->I3". For other cases like
4409 "I0->I1, I1&I2->I3" or "I1&I2->I3", newi2pat won't set i1dest or
4410 i0dest anyway. */
4412 || !i1dest_killed
4417 /* Same case as i1. */
4424 /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4425 clear them. */
4433 /* Ensure that we do not have something that should not be shared but
4434 occurs multiple times in the new insns. Check this by first
4435 resetting all the `used' flags and then copying anything is shared. */
4459 {
4462 {
4464 {
4465 /* I2SRC must still be meaningful at this point. Some
4466 splitting operations can invalidate I2SRC, but those
4467 operations do not apply to calls. */
4471 }
4478 }
4479 }
4484 /* We had one special case above where I2 had more than one set and
4485 we replaced a destination of one of those sets with the destination
4486 of I3. In that case, we have to update LOG_LINKS of insns later
4487 in this basic block. Note that this (expensive) case is rare.
4489 Also, in this case, we must pretend that all REG_NOTEs for I2
4490 actually came from I3, so that REG_UNUSED notes from I2 will be
4491 properly handled. */
4494 {
4503 temp_insn
4513 {
4518 }
4519 else
4522 }
4530 {
4533 this_basic_block);
4536 }
4537 else
4538 {
4541 this_basic_block);
4543 }
4546 {
4551 this_basic_block);
4553 }
4556 {
4561 this_basic_block);
4563 }
4565 /* Get death notes for everything that is now used in either I3 or
4566 I2 and used to die in a previous insn. If we built two new
4567 patterns, move from I1 to I2 then I2 to I3 so that we get the
4568 proper movement on registers that I2 modifies. */
4574 else
4580 /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4597 /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4598 know these are REG_UNUSED and want them to go to the desired insn,
4599 so we always pass it as i3. */
4603 NULL_RTX);
4607 NULL_RTX);
4609 /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4610 put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4611 I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4612 in that case, it might delete I2. Similarly for I2 and I1.
4613 Show an additional death due to the REG_DEAD note we make here. If
4614 we discard it in distribute_notes, we will decrement it again. */
4617 {
4622 else
4625 }
4628 {
4633 else
4636 }
4639 {
4644 else
4647 }
4650 {
4655 else
4658 }
4666 {
4671 /* The insn that used to set this register doesn't exist, and
4672 this life of the register may not exist either. See if one of
4673 I3's links points to an insn that sets I2DEST. If it does,
4674 that is now the last known value for I2DEST. If we don't update
4675 this and I2 set the register to a value that depended on its old
4676 contents, we will get confused. If this insn is used, thing
4677 will be set correctly in combine_instructions. */
4685 /* If the reg formerly set in I2 died only once and that was in I3,
4686 zero its use count so it won't make `reload' do any work. */
4689 && ! i2dest_in_i2src
4692 }
4695 {
4708 && ! i1dest_in_i1src
4711 }
4714 {
4727 && ! i0dest_in_i0src
4730 }
4732 /* Update reg_stat[].nonzero_bits et al for any changes that may have
4733 been made to this insn. The order is important, because newi2pat
4734 can affect nonzero_bits of newpat. */
4738 }
4741 {
4743 {
4746 }
4748 }
4751 {
4753 {
4756 }
4758 }
4761 {
4763 {
4766 }
4768 }
4771 {
4773 {
4776 }
4778 }
4781 {
4783 {
4786 }
4788 }
4790 /* Set new_direct_jump_p if a new return or simple jump instruction
4791 has been created. Adjust the CFG accordingly. */
4793 {
4797 }
4802 {
4805 }
4809 {
4815 }
4820 {
4826 }
4828 /* A noop might also need cleaning up of CFG, if it comes from the
4829 simplification of a jump. */
4834 {
4837 }
4844 {
4847 }
4849 combine_successes++;
4859 }
4860 \f
4861 /* Get a marker for undoing to the current state. */
4865 {
4867 }
4869 /* Undo the modifications up to the marker. */
4871 static void
4873 {
4877 {
4882 {
4897 }
4901 }
4904 }
4906 /* Undo all the modifications recorded in undobuf. */
4908 static void
4910 {
4912 }
4914 /* We've committed to accepting the changes we made. Move all
4915 of the undos to the free list. */
4917 static void
4919 {
4923 {
4927 }
4929 }
4930 \f
4931 /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4932 where we have an arithmetic expression and return that point. LOC will
4933 be inside INSN.
4935 try_combine will call this function to see if an insn can be split into
4936 two insns. */
4940 {
4950 /* First special-case some codes. */
4952 {
4954 #ifdef INSN_SCHEDULING
4955 /* If we are making a paradoxical SUBREG invalid, it becomes a split
4956 point. */
4959 #endif
4963 /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4964 using LO_SUM and HIGH. */
4967 {
4975 }
4977 /* If we have a PLUS whose second operand is a constant and the
4978 address is not valid, perhaps we can split it up using
4979 the machine-specific way to split large constants. We use
4980 the first pseudo-reg (one of the virtual regs) as a placeholder;
4981 it will not remain in the result. */
4986 {
4989 subst_insn);
4991 /* This should have produced two insns, each of which sets our
4992 placeholder. If the source of the second is a valid address,
4993 we can put both sources together and make a split point
4994 in the middle. */
5007 && memory_address_addr_space_p
5010 {
5014 /* Replace the placeholder in SRC2 with SRC1. If we can
5015 find where in SRC2 it was placed, that can become our
5016 split point and we can replace this address with SRC2.
5017 Just try two obvious places. */
5028 {
5031 }
5032 }
5034 /* If that didn't work and we have a nested plus, like:
5035 ((REG1 * CONST1) + REG2) + CONST2 and (REG1 + REG2) + CONST2
5036 is valid address, try to split (REG1 * CONST1). */
5043 {
5048 {
5051 }
5053 }
5060 {
5065 {
5068 }
5070 }
5072 /* If that didn't work, perhaps the first operand is complex and
5073 needs to be computed separately, so make a split point there.
5074 This will occur on machines that just support REG + CONST
5075 and have a constant moved through some previous computation. */
5080 }
5082 /* If we have a PLUS whose first operand is complex, try computing it
5083 separately by making a split there. */
5094 /* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
5095 ZERO_EXTRACT, the most likely reason why this doesn't match is that
5096 we need to put the operand into a register. So split at that
5097 point. */
5107 /* See if we can split SET_SRC as it stands. */
5112 /* See if we can split SET_DEST as it stands. */
5117 /* See if this is a bitfield assignment with everything constant. If
5118 so, this is an IOR of an AND, so split it into that. */
5130 {
5136 rtx or_mask;
5145 else
5146 {
5152 or_mask));
5153 }
5160 }
5162 /* Otherwise, see if this is an operation that we can split into two.
5163 If so, try to split that. */
5167 {
5169 /* If we are AND'ing with a large constant that is only a single
5170 bit and the result is only being used in a context where we
5171 need to know if it is zero or nonzero, replace it with a bit
5172 extraction. This will avoid the large constant, which might
5173 have taken more than one insn to make. If the constant were
5174 not a valid argument to the AND but took only one insn to make,
5175 this is no worse, but if it took more than one insn, it will
5176 be better. */
5186 {
5191 {
5194 }
5195 }
5199 /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
5200 is known to be on, this can be converted into a NEG of a shift. */
5206 {
5213 pos_rtx)));
5218 }
5224 /* We can't optimize if either mode is a partial integer
5225 mode as we don't know how many bits are significant
5226 in those modes. */
5242 {
5250 }
5255 }
5261 {
5262 /* For unsigned, we have a choice of a shift followed by an
5263 AND or two shifts. Use two shifts for field sizes where the
5264 constant might be too large. We assume here that we can
5265 always at least get 8-bit constants in an AND insn, which is
5266 true for every current RISC. */
5269 {
5270 unsigned HOST_WIDE_INT mask
5275 gen_rtx_LSHIFTRT
5282 }
5283 else
5284 {
5288 gen_rtx_fmt_ee
5298 }
5299 }
5301 /* See if this is a simple operation with a constant as the second
5302 operand. It might be that this constant is out of range and hence
5303 could be used as a split point. */
5311 /* Finally, see if this is a simple operation with its first operand
5312 not in a register. The operation might require this operand in a
5313 register, so return it as a split point. We can always do this
5314 because if the first operand were another operation, we would have
5315 already found it as a split point. */
5324 /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
5325 it is better to write this as (not (ior A B)) so we can split it.
5326 Similarly for IOR. */
5328 {
5336 }
5338 /* Many RISC machines have a large set of logical insns. If the
5339 second operand is a NOT, put it first so we will try to split the
5340 other operand first. */
5342 {
5346 }
5351 /* Canonicalization can produce (minus A (mult B C)), where C is a
5352 constant. It may be better to try splitting (plus (mult B -C) A)
5353 instead if this isn't a multiply by a power of two. */
5357 {
5365 mode)),
5368 }
5370 /* Split at a multiply-accumulate instruction. However if this is
5371 the SET_SRC, we likely do not have such an instruction and it's
5372 worthless to try this split. */
5381 }
5383 /* Otherwise, select our actions depending on our rtx class. */
5385 {
5391 /* fall through */
5399 /* fall through */
5401 /* Some machines have (and (shift ...) ...) insns. If X is not
5402 an AND, but XEXP (X, 0) is, use it as our split point. */
5412 /* Otherwise, we don't have a split point. */
5414 }
5415 }
5416 \f
5417 /* Throughout X, replace FROM with TO, and return the result.
5418 The result is TO if X is FROM;
5419 otherwise the result is X, but its contents may have been modified.
5420 If they were modified, a record was made in undobuf so that
5421 undo_all will (among other things) return X to its original state.
5423 If the number of changes necessary is too much to record to undo,
5424 the excess changes are not made, so the result is invalid.
5425 The changes already made can still be undone.
5426 undobuf.num_undo is incremented for such changes, so by testing that
5427 the caller can tell whether the result is valid.
5429 `n_occurrences' is incremented each time FROM is replaced.
5431 IN_DEST is nonzero if we are processing the SET_DEST of a SET.
5433 IN_COND is nonzero if we are at the top level of a condition.
5435 UNIQUE_COPY is nonzero if each substitution must be unique. We do this
5436 by copying if `n_occurrences' is nonzero. */
5438 static rtx
5440 {
5445 rtx new_rtx;
5447 /* Two expressions are equal if they are identical copies of a shared
5448 RTX or if they are both registers with the same register number
5449 and mode. */
5451 #define COMBINE_RTX_EQUAL_P(X,Y) \
5452 ((X) == (Y) \
5453 || (REG_P (X) && REG_P (Y) \
5454 && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5456 /* Do not substitute into clobbers of regs -- this will never result in
5457 valid RTL. */
5462 {
5463 n_occurrences++;
5465 }
5467 /* If X and FROM are the same register but different modes, they
5468 will not have been seen as equal above. However, the log links code
5469 will make a LOG_LINKS entry for that case. If we do nothing, we
5470 will try to rerecognize our original insn and, when it succeeds,
5471 we will delete the feeding insn, which is incorrect.
5473 So force this insn not to match in this (rare) case. */
5478 /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5479 of which may contain things that can be combined. */
5483 /* It is possible to have a subexpression appear twice in the insn.
5484 Suppose that FROM is a register that appears within TO.
5485 Then, after that subexpression has been scanned once by `subst',
5486 the second time it is scanned, TO may be found. If we were
5487 to scan TO here, we would find FROM within it and create a
5488 self-referent rtl structure which is completely wrong. */
5492 /* Parallel asm_operands need special attention because all of the
5493 inputs are shared across the arms. Furthermore, unsharing the
5494 rtl results in recognition failures. Failure to handle this case
5495 specially can result in circular rtl.
5497 Solve this by doing a normal pass across the first entry of the
5498 parallel, and only processing the SET_DESTs of the subsequent
5499 entries. Ug. */
5504 {
5507 /* If this substitution failed, this whole thing fails. */
5515 {
5521 {
5524 /* If this substitution failed, this whole thing fails. */
5530 }
5531 }
5532 }
5533 else
5534 {
5538 /* We don't need to process a SET_DEST that is a register, CC0,
5539 or PC, so set up to skip this common case. All other cases
5540 where we want to suppress replacing something inside a
5541 SET_SRC are handled via the IN_DEST operand. */
5548 /* Trying to simplify the operands of a widening MULT is not likely
5549 to create RTL matching a machine insn. */
5561 /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5562 constant. */
5567 {
5569 {
5572 {
5574 {
5577 n_occurrences++;
5578 }
5579 else
5580 {
5582 unique_copy);
5584 /* If this substitution failed, this whole thing
5585 fails. */
5589 }
5592 }
5593 }
5595 {
5596 /* If this is a register being set, ignore it. */
5603 ;
5606 {
5607 /* In general, don't install a subreg involving two
5608 modes not tieable. It can worsen register
5609 allocation, and can even make invalid reload
5610 insns, since the reg inside may need to be copied
5611 from in the outside mode, and that may be invalid
5612 if it is an fp reg copied in integer mode.
5614 We allow two exceptions to this: It is valid if
5615 it is inside another SUBREG and the mode of that
5616 SUBREG and the mode of the inside of TO is
5617 tieable and it is valid if X is a SET that copies
5618 FROM to CC0. */
5626 && (!HAVE_cc0
5641 n_occurrences++;
5642 }
5643 else
5644 /* If we are in a SET_DEST, suppress most cases unless we
5645 have gone inside a MEM, in which case we want to
5646 simplify the address. We assume here that things that
5647 are actually part of the destination have their inner
5648 parts in the first expression. This is true for SUBREG,
5649 STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5650 things aside from REG and MEM that should appear in a
5651 SET_DEST. */
5653 (((in_dest
5659 unique_copy);
5661 /* If we found that we will have to reject this combination,
5662 indicate that by returning the CLOBBER ourselves, rather than
5663 an expression containing it. This will speed things up as
5664 well as prevent accidents where two CLOBBERs are considered
5665 to be equal, thus producing an incorrect simplification. */
5671 {
5679 }
5685 {
5687 new_rtx,
5691 }
5692 else
5694 }
5695 }
5696 }
5698 /* Check if we are loading something from the constant pool via float
5699 extension; in this case we would undo compress_float_constant
5700 optimization and degenerate constant load to an immediate value. */
5704 {
5708 }
5710 /* Try to simplify X. If the simplification changed the code, it is likely
5711 that further simplification will help, so loop, but limit the number
5712 of repetitions that will be performed. */
5715 {
5716 /* If X is sufficiently simple, don't bother trying to do anything
5717 with it. */
5726 /* We no longer know the original mode of operand 0 since we
5727 have changed the form of X) */
5729 }
5732 }
5733 \f
5734 /* If X is a commutative operation whose operands are not in the canonical
5735 order, use substitutions to swap them. */
5737 static void
5739 {
5742 {
5746 }
5747 }
5749 /* Simplify X, a piece of RTL. We just operate on the expression at the
5750 outer level; call `subst' to simplify recursively. Return the new
5751 expression.
5753 OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
5754 if we are inside a SET_DEST. IN_COND is nonzero if we are at the top level
5755 of a condition. */
5757 static rtx
5760 {
5763 scalar_int_mode int_mode;
5764 rtx temp;
5767 /* If this is a commutative operation, put a constant last and a complex
5768 expression first. We don't need to do this for comparisons here. */
5771 /* Try to fold this expression in case we have constants that weren't
5772 present before. */
5775 {
5783 {
5786 {
5790 }
5793 }
5806 }
5809 {
5814 }
5816 /* If this is a simple operation applied to an IF_THEN_ELSE, try
5817 applying it to the arms of the IF_THEN_ELSE. This often simplifies
5818 things. Check for cases where both arms are testing the same
5819 condition.
5821 Don't do anything if all operands are very simple. */
5834 {
5839 /* If everything is a comparison, what we have is highly unlikely
5840 to be simpler, so don't use it. */
5843 /* Similarly, if we end up with one of the expressions the same
5844 as the original, it is certainly not simpler. */
5847 {
5854 /* Simplify the alternative arms; this may collapse the true and
5855 false arms to store-flag values. Be careful to use copy_rtx
5856 here since true_rtx or false_rtx might share RTL with x as a
5857 result of the if_then_else_cond call above. */
5861 /* If true_rtx and false_rtx are not general_operands, an if_then_else
5862 is unlikely to be simpler. */
5865 {
5868 /* Restarting if we generate a store-flag expression will cause
5869 us to loop. Just drop through in this case. */
5871 /* If the result values are STORE_FLAG_VALUE and zero, we can
5872 just make the comparison operation. */
5883 /* Likewise, we can make the negate of a comparison operation
5884 if the result values are - STORE_FLAG_VALUE and zero. */
5892 mode);
5903 mode);
5907 }
5908 }
5909 }
5911 /* First see if we can apply the inverse distributive law. */
5914 {
5918 }
5920 /* If CODE is an associative operation not otherwise handled, see if we
5921 can associate some operands. This can win if they are constants or
5922 if they are logically related (i.e. (a & b) & a). */
5928 {
5930 {
5934 rtx inner;
5936 /* Make sure we pass the constant operand if any as the second
5937 one if this is a commutative operation. */
5945 /* For commutative operations, try the other pair if that one
5946 didn't simplify. */
5948 {
5953 }
5957 }
5958 }
5960 /* A little bit of algebraic simplification here. */
5962 {
5964 /* Ensure that our address has any ASHIFTs converted to MULT in case
5965 address-recognizing predicates are called later. */
5974 /* See if this can be moved to simplify_subreg. */
5977 /* Don't call gen_lowpart if the inner mode
5978 is VOIDmode and we cannot simplify it, as SUBREG without
5979 inner mode is invalid. */
5986 {
5987 rtx temp;
5993 /* If op is known to have all lower bits zero, the result is zero. */
6007 }
6009 /* Don't change the mode of the MEM if that would change the meaning
6010 of the address. */
6017 /* Note that we cannot do any narrowing for non-constants since
6018 we might have been counting on using the fact that some bits were
6019 zero. We now do this in the SET. */
6026 /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
6027 replaced by (lshiftrt X C). This will convert
6028 (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
6036 /* If X has only a single bit that might be nonzero, say, bit I, convert
6037 (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
6038 MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
6039 (sign_extract X 1 Y). But only do this if TEMP isn't a register
6040 or a SUBREG of one since we'd be making the expression more
6041 complex if it was just a register. */
6048 {
6049 rtx temp1 = simplify_shift_const
6055 /* If all we did was surround TEMP with the two shifts, we
6056 haven't improved anything, so don't use it. Otherwise,
6057 we are better off with TEMP1. */
6062 }
6066 /* We can't handle truncation to a partial integer mode here
6067 because we don't know the real bitsize of the partial
6068 integer mode. */
6077 /* We can truncate a constant value and return it. */
6078 {
6079 poly_int64 c;
6082 }
6084 /* Similarly to what we do in simplify-rtx.c, a truncate of a register
6085 whose value is a comparison can be replaced with a subreg if
6086 STORE_FLAG_VALUE permits. */
6095 /* (const (const X)) can become (const X). Do it this way rather than
6096 returning the inner CONST since CONST can be shared with a
6097 REG_EQUAL note. */
6103 /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
6104 can add in an offset. find_split_point will split this address up
6105 again if it doesn't match. */
6112 /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
6113 when c is (const_int (pow2 + 1) / 2) is a sign extension of a
6114 bit-field and can be replaced by either a sign_extend or a
6115 sign_extract. The `and' may be a zero_extend and the two
6116 <c>, -<c> constants may be reversed. */
6133 return simplify_shift_const
6140 /* If only the low-order bit of X is possibly nonzero, (plus x -1)
6141 can become (ashiftrt (ashift (xor x 1) C) C) where C is
6142 the bitsize of the mode - 1. This allows simplification of
6143 "a = (b & 8) == 0;" */
6150 return simplify_shift_const
6154 const1_rtx),
6158 /* If we are adding two things that have no bits in common, convert
6159 the addition into an IOR. This will often be further simplified,
6160 for example in cases like ((a & 1) + (a & 2)), which can
6161 become a & 3. */
6166 {
6167 /* Try to simplify the expression further. */
6171 /* If we could, great. If not, do not go ahead with the IOR
6172 replacement, since PLUS appears in many special purpose
6173 address arithmetic instructions. */
6181 }
6183 /* Canonicalize x + x into x << 1. */
6192 /* (minus <foo> (and <foo> (const_int -pow2))) becomes
6193 (and <foo> (const_int pow2-1)) */
6204 /* If we have (mult (plus A B) C), apply the distributive law and then
6205 the inverse distributive law to see if things simplify. This
6206 occurs mostly in addresses, often when unrolling loops. */
6209 {
6213 }
6215 /* Try simplify a*(b/c) as (a*b)/c. */
6218 {
6224 }
6228 /* If this is a divide by a power of two, treat it as a shift if
6229 its first operand is a shift. */
6249 /* If the first operand is a condition code, we can't do anything
6250 with it. */
6254 {
6262 /* Simplify our comparison, if possible. */
6265 /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
6266 if only the low-order bit is possibly nonzero in X (such as when
6267 X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
6268 (xor X 1) or (minus 1 X); we use the former. Finally, if X is
6269 known to be either 0 or -1, NE becomes a NEG and EQ becomes
6270 (plus X 1).
6272 Remove any ZERO_EXTRACT we made when thinking this was a
6273 comparison. It may now be simpler to use, e.g., an AND. If a
6274 ZERO_EXTRACT is indeed appropriate, it will be placed back by
6275 the call to make_compound_operation in the SET case.
6277 Don't apply these optimizations if the caller would
6278 prefer a comparison rather than a value.
6279 E.g., for the condition in an IF_THEN_ELSE most targets need
6280 an explicit comparison. */
6283 ;
6301 {
6305 int_mode);
6306 }
6314 {
6318 const1_rtx);
6319 }
6328 {
6331 }
6333 /* If STORE_FLAG_VALUE is -1, we have cases similar to
6334 those above. */
6336 ;
6353 {
6357 int_mode);
6358 }
6367 {
6371 int_mode);
6372 }
6374 /* If X is 0/1, (eq X 0) is X-1. */
6381 {
6384 }
6386 /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
6387 one bit that might be nonzero, we can convert (ne x 0) to
6388 (ashift x c) where C puts the bit in the sign bit. Remove any
6389 AND with STORE_FLAG_VALUE when we are done, since we are only
6390 going to test the sign bit. */
6398 {
6404 else
6406 }
6408 /* If the code changed, return a whole new comparison.
6409 We also need to avoid using SUBST in cases where
6410 simplify_comparison has widened a comparison with a CONST_INT,
6411 since in that case the wider CONST_INT may fail the sanity
6412 checks in do_SUBST. */
6419 /* Otherwise, keep this operation, but maybe change its operands.
6420 This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
6423 }
6433 /* If we are processing SET_DEST, we are done. */
6451 /* If this is a shift by a constant amount, simplify it. */
6459 (HOST_WIDE_INT_1U
6468 }
6471 }
6472 \f
6473 /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6475 static rtx
6477 {
6484 rtx temp;
6487 rtx reversed;
6490 /* Simplify storing of the truth value. */
6495 /* Also when the truth value has to be reversed. */
6501 /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6502 in it is being compared against certain values. Get the true and false
6503 comparisons and see if that says anything about the value of each arm. */
6509 {
6510 HOST_WIDE_INT nzb;
6516 /* If FALSE_CODE is EQ, swap the codes and arms. */
6519 {
6522 }
6524 scalar_int_mode from_mode;
6526 {
6527 /* If we are comparing against zero and the expression being
6528 tested has only a single bit that might be nonzero, that is
6529 its value when it is not equal to zero. Similarly if it is
6530 known to be -1 or 0. */
6534 {
6537 }
6542 {
6545 }
6546 }
6548 /* Now simplify an arm if we know the value of the register in the
6549 branch and it is used in the arm. Be careful due to the potential
6550 of locally-shared RTL. */
6567 }
6569 /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6570 reversed, do so to avoid needing two sets of patterns for
6571 subtract-and-branch insns. Similarly if we have a constant in the true
6572 arm, the false arm is the same as the first operand of the comparison, or
6573 the false arm is more complicated than the true arm. */
6586 {
6594 /* It is possible that the conditional has been simplified out. */
6597 }
6599 /* If the two arms are identical, we don't need the comparison. */
6604 /* Convert a == b ? b : a to "a". */
6616 /* Look for cases where we have (abs x) or (neg (abs X)). */
6619 && comparison_p
6626 {
6632 return
6635 mode);
6638 }
6640 /* Look for MIN or MAX. */
6643 && comparison_p
6648 {
6663 }
6665 /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6666 second operand is zero, this can be done as (OP Z (mult COND C2)) where
6667 C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6668 SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6669 We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6670 neither 1 or -1, but it isn't worth checking for. */
6673 && comparison_p
6676 {
6692 /* If an identity-zero op is commutative, check whether there
6693 would be a match if we swapped the operands. */
6714 {
6718 }
6731 {
6735 }
6752 {
6756 }
6769 {
6773 }
6776 {
6786 const_true_rtx));
6794 }
6795 }
6797 /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6798 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6799 negation of a single bit, we can convert this operation to a shift. We
6800 can actually do this more generally, but it doesn't seem worth it. */
6812 return
6816 /* (IF_THEN_ELSE (NE A 0) C1 0) is A or a zero-extend of A if the only
6817 non-zero bit in A is C1. */
6825 {
6831 }
6834 }
6835 \f
6836 /* Simplify X, a SET expression. Return the new expression. */
6838 static rtx
6840 {
6843 machine_mode mode
6847 scalar_int_mode int_mode;
6849 /* (set (pc) (return)) gets written as (return). */
6853 /* Now that we know for sure which bits of SRC we are using, see if we can
6854 simplify the expression for the object knowing that we only need the
6855 low-order bits. */
6858 {
6861 }
6863 /* If we are setting CC0 or if the source is a COMPARE, look for the use of
6864 the comparison result and try to simplify it unless we already have used
6865 undobuf.other_insn. */
6873 {
6882 {
6885 {
6888 }
6889 }
6890 else
6898 {
6902 }
6903 else
6904 {
6909 /* Attempt to simplify CC user. */
6911 {
6915 }
6917 /* Convert X into a no-op move. */
6921 }
6923 /* Simplify our comparison, if possible. */
6926 #ifdef SELECT_CC_MODE
6927 /* If this machine has CC modes other than CCmode, check to see if we
6928 need to use a different CC mode here. */
6937 else
6940 /* If the mode changed, we have to change SET_DEST, the mode in the
6941 compare, and the mode in the place SET_DEST is used. If SET_DEST is
6942 a hard register, just build new versions with the proper mode. If it
6943 is a pseudo, we lose unless it is only time we set the pseudo, in
6944 which case we can safely change its mode. */
6946 {
6948 {
6950 rtx new_dest;
6954 else
6955 {
6958 }
6965 }
6966 }
6969 /* If the code changed, we have to build a new comparison in
6970 undobuf.other_insn. */
6972 {
6981 /* If the only change we made was to change an EQ into an NE or
6982 vice versa, OP0 has only one bit that might be nonzero, and OP1
6983 is zero, check if changing the user of the condition code will
6984 produce a valid insn. If it won't, we can keep the original code
6985 in that insn by surrounding our operation with an XOR. */
6992 {
6997 {
7004 }
7005 }
7006 }
7011 /* Don't generate a compare of a CC with 0, just use that CC. */
7013 {
7016 }
7017 /* Otherwise, if we didn't previously have the same COMPARE we
7018 want, create it from scratch. */
7021 {
7024 }
7025 }
7026 else
7027 {
7028 /* Get SET_SRC in a form where we have placed back any
7029 compound expressions. Then do the checks below. */
7032 }
7034 /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
7035 and X being a REG or (subreg (reg)), we may be able to convert this to
7036 (set (subreg:m2 x) (op)).
7038 We can always do this if M1 is narrower than M2 because that means that
7039 we only care about the low bits of the result.
7041 However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
7042 perform a narrower operation than requested since the high-order bits will
7043 be undefined. On machine where it is defined, this transformation is safe
7044 as long as M1 and M2 have the same number of words. */
7048 && (known_equal_after_align_up
7051 UNITS_PER_WORD))
7060 {
7063 dest));
7067 }
7069 /* If we have (set (cc0) (subreg ...)), we try to remove the subreg
7070 in SRC. */
7074 {
7078 /* Here we make sure that we don't have a sign bit on. */
7081 {
7084 }
7085 }
7087 /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
7088 would require a paradoxical subreg. Replace the subreg with a
7089 zero_extend to avoid the reload that would otherwise be required.
7090 Don't do this unless we have a scalar integer mode, otherwise the
7091 transformation is incorrect. */
7098 {
7103 }
7105 /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
7106 are comparing an item known to be 0 or -1 against 0, use a logical
7107 operation instead. Check for one of the arms being an IOR of the other
7108 arm with some value. We compute three terms to be IOR'ed together. In
7109 practice, at most two will be nonzero. Then we do the IOR's. */
7117 && (!HAVE_conditional_move
7122 {
7147 int_mode),
7148 false_rtx);
7154 term3));
7157 }
7159 /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
7160 whole thing fail. */
7165 else
7166 /* Convert this into a field assignment operation, if possible. */
7168 }
7169 \f
7170 /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
7171 result. */
7173 static rtx
7175 {
7178 scalar_int_mode mode;
7181 {
7183 /* We can call simplify_and_const_int only if we don't lose
7184 any (sign) bits when converting INTVAL (op1) to
7185 "unsigned HOST_WIDE_INT". */
7190 {
7197 }
7199 /* If we have any of (and (ior A B) C) or (and (xor A B) C),
7200 apply the distributive law and then the inverse distributive
7201 law to see if things simplify. */
7203 {
7207 }
7209 {
7213 }
7217 /* If we have (ior (and A B) C), apply the distributive law and then
7218 the inverse distributive law to see if things simplify. */
7221 {
7225 }
7228 {
7232 }
7237 }
7240 }
7241 \f
7242 /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
7243 operations" because they can be replaced with two more basic operations.
7244 ZERO_EXTEND is also considered "compound" because it can be replaced with
7245 an AND operation, which is simpler, though only one operation.
7247 The function expand_compound_operation is called with an rtx expression
7248 and will convert it to the appropriate shifts and AND operations,
7249 simplifying at each stage.
7251 The function make_compound_operation is called to convert an expression
7252 consisting of shifts and ANDs into the equivalent compound expression.
7253 It is the inverse of this function, loosely speaking. */
7255 static rtx
7257 {
7261 rtx tem;
7262 scalar_int_mode inner_mode;
7265 {
7268 /* FALLTHRU */
7270 /* We can't necessarily use a const_int for a multiword mode;
7271 it depends on implicitly extending the value.
7272 Since we don't know the right way to extend it,
7273 we can't tell whether the implicit way is right.
7275 Even for a mode that is no wider than a const_int,
7276 we can't win, because we need to sign extend one of its bits through
7277 the rest of it, and we don't know which bit. */
7281 /* Reject modes that aren't scalar integers because turning vector
7282 or complex modes into shifts causes problems. */
7286 /* Return if (subreg:MODE FROM 0) is not a safe replacement for
7287 (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
7288 because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
7289 reloaded. If not for that, MEM's would very rarely be safe.
7291 Reject modes bigger than a word, because we might not be able
7292 to reference a two-register group starting with an arbitrary register
7293 (and currently gen_lowpart might crash for a SUBREG). */
7299 /* If the inner object has VOIDmode (the only way this can happen
7300 is if it is an ASM_OPERANDS), we can't do anything since we don't
7301 know how much masking to do. */
7310 /* fall through */
7313 /* If the operand is a CLOBBER, just return it. */
7321 /* Reject modes that aren't scalar integers because turning vector
7322 or complex modes into shifts causes problems. */
7329 /* This should stay within the object being extracted, fail otherwise. */
7340 }
7342 /* We've rejected non-scalar operations by now. */
7345 /* Convert sign extension to zero extension, if we know that the high
7346 bit is not set, as this is easier to optimize. It will be converted
7347 back to cheaper alternative in make_extraction. */
7353 {
7357 /* Make sure this is a profitable operation. */
7364 else
7366 }
7368 /* We can optimize some special cases of ZERO_EXTEND. */
7370 {
7371 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
7372 know that the last value didn't have any inappropriate bits
7373 set. */
7381 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7390 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
7391 is a comparison and STORE_FLAG_VALUE permits. This is like
7392 the first case, but it works even when MODE is larger
7393 than HOST_WIDE_INT. */
7401 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7410 }
7412 /* If we reach here, we want to return a pair of shifts. The inner
7413 shift is a left shift of BITSIZE - POS - LEN bits. The outer
7414 shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
7415 logical depending on the value of UNSIGNEDP.
7417 If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
7418 converted into an AND of a shift.
7420 We must check for the case where the left shift would have a negative
7421 count. This can happen in a case like (x >> 31) & 255 on machines
7422 that can't shift by a constant. On those machines, we would first
7423 combine the shift with the AND to produce a variable-position
7424 extraction. Then the constant of 31 would be substituted in
7425 to produce such a position. */
7429 {
7437 }
7442 pos),
7444 else
7445 /* Any other cases we can't handle. */
7448 /* If we couldn't do this for some reason, return the original
7449 expression. */
7454 }
7455 \f
7456 /* X is a SET which contains an assignment of one object into
7457 a part of another (such as a bit-field assignment, STRICT_LOW_PART,
7458 or certain SUBREGS). If possible, convert it into a series of
7459 logical operations.
7461 We half-heartedly support variable positions, but do not at all
7462 support variable lengths. */
7464 static const_rtx
7466 {
7467 rtx inner;
7471 scalar_int_mode compute_mode;
7473 /* Loop until we find something we can't simplify. */
7475 {
7478 {
7484 MAX_MODE_INT);
7485 }
7488 {
7496 /* A constant position should stay within the width of INNER. */
7501 {
7507 /* If position is ADJUST - X, new position is X. */
7509 else
7513 pos);
7514 }
7515 }
7517 /* If the destination is a subreg that overwrites the whole of the inner
7518 register, we can move the subreg to the source. */
7520 /* We need SUBREGs to compute nonzero_bits properly. */
7521 && nonzero_sign_valid
7523 {
7525 gen_lowpart
7529 }
7530 else
7536 /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7538 {
7539 /* Don't do anything for vector or complex integral types. */
7543 /* Try to find an integral mode to pun with. */
7549 }
7551 /* Compute a mask of LEN bits, if we can do this on the host machine. */
7555 /* Don't try to compute in too wide unsupported modes. */
7559 /* Now compute the equivalent expression. Make a copy of INNER
7560 for the SET_DEST in case it is a MEM into which we will substitute;
7561 we don't want shared RTL in that case. */
7563 compute_mode);
7567 compute_mode,
7569 compute_mode),
7570 inner);
7575 mask),
7576 pos);
7581 }
7584 }
7585 \f
7586 /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7587 it is an RTX that represents the (variable) starting position; otherwise,
7588 POS is the (constant) starting bit position. Both are counted from the LSB.
7590 UNSIGNEDP is nonzero for an unsigned reference and zero for a signed one.
7592 IN_DEST is nonzero if this is a reference in the destination of a SET.
7593 This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7594 a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7595 be used.
7597 IN_COMPARE is nonzero if we are in a COMPARE. This means that a
7598 ZERO_EXTRACT should be built even for bits starting at bit 0.
7600 MODE is the desired mode of the result (if IN_DEST == 0).
7602 The result is an RTX for the extraction or NULL_RTX if the target
7603 can't handle it. */
7605 static rtx
7609 {
7610 /* This mode describes the size of the storage area
7611 to fetch the overall value from. Within that, we
7612 ignore the POS lowest bits, etc. */
7614 machine_mode inner_mode;
7615 scalar_int_mode wanted_inner_mode;
7621 HOST_WIDE_INT orig_pos;
7629 /* If trying or potentionally trying to extract
7630 bits outside of is_mode, don't look through
7631 non-paradoxical SUBREGs. See PR82192. */
7634 {
7635 /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7636 consider just the QI as the memory to extract from.
7637 The subreg adds or removes high bits; its mode is
7638 irrelevant to the meaning of this extraction,
7639 since POS and LEN count from the lsb. */
7643 }
7648 {
7649 /* We're extracting the least significant bits of an rtx
7650 (ashift X (const_int C)), where LEN > C. Extract the
7651 least significant (LEN - C) bits of X, giving an rtx
7652 whose mode is MODE, then shift it left C times. */
7658 }
7660 /* If trying or potentionally trying to extract
7661 bits outside of is_mode, don't look through
7662 TRUNCATE. See PR82192. */
7669 /* See if this can be done without an extraction. We never can if the
7670 width of the field is not the same as that of some integer mode. For
7671 registers, we can only avoid the extraction if the position is at the
7672 low-order bit and this is either not in the destination or we have the
7673 appropriate STRICT_LOW_PART operation available.
7675 For MEM, we can avoid an extract if the field starts on an appropriate
7676 boundary and we can change the mode of the memory reference. */
7678 scalar_int_mode tmode;
7687 && (! in_dest
7691 && (pos
7694 /* We can't do this if we are widening INNER_MODE (it
7695 may not be aligned, for one thing). */
7702 {
7703 /* If INNER is a MEM, make a new MEM that encompasses just the desired
7704 field. If the original and current mode are the same, we need not
7705 adjust the offset. Otherwise, we do if bytes big endian.
7707 If INNER is not a MEM, get a piece consisting of just the field
7708 of interest (in this case POS % BITS_PER_WORD must be 0). */
7711 {
7712 poly_int64 offset;
7714 /* POS counts from lsb, but make OFFSET count in memory order. */
7718 else
7722 }
7724 {
7726 {
7727 /* We can't call gen_lowpart in a DEST since we
7728 always want a SUBREG (see below) and it would sometimes
7729 return a new hard register. */
7731 {
7732 poly_uint64 offset
7735 /* Avoid creating invalid subregs, for example when
7736 simplifying (x>>32)&255. */
7741 }
7742 else
7744 }
7745 else
7747 }
7748 else
7750 len >= HOST_BITS_PER_WIDE_INT
7751 ? HOST_WIDE_INT_M1U
7754 /* If this extraction is going into the destination of a SET,
7755 make a STRICT_LOW_PART unless we made a MEM. */
7770 /* If we know that no extraneous bits are set, and that the high
7771 bit is not set, convert the extraction to the cheaper of
7772 sign and zero extension, that are equivalent in these cases. */
7778 {
7782 /* Prefer ZERO_EXTENSION, since it gives more information to
7783 backends. */
7788 }
7790 /* Otherwise, sign- or zero-extend unless we already are in the
7791 proper mode. */
7795 }
7797 /* Unless this is a COMPARE or we have a funny memory reference,
7798 don't do anything with zero-extending field extracts starting at
7799 the low-order bit since they are simple AND operations. */
7804 /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7805 if the position is not a constant and the length is not 1. In all
7806 other cases, we would only be going outside our object in cases when
7807 an original shift would have been undefined. */
7816 /* If INNER is not from memory, we want it to have the mode of a register
7817 extraction pattern's structure operand, or word_mode if there is no
7818 such pattern. The same applies to extraction_mode and pos_mode
7819 and their respective operands.
7821 For memory, assume that the desired extraction_mode and pos_mode
7822 are the same as for a register operation, since at present we don't
7823 have named patterns for aligned memory structures. */
7828 {
7832 }
7834 /* Never narrow an object, since that might not be safe. */
7840 /* Punt if len is too large for extraction_mode. */
7846 else
7847 {
7848 /* Be careful not to go beyond the extracted object and maintain the
7849 natural alignment of the memory. */
7854 }
7859 {
7860 /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7861 BITS_BIG_ENDIAN style. If position is constant, compute new
7862 position. Otherwise, build subtraction.
7863 Note that POS is relative to the mode of the original argument.
7864 If it's a MEM we need to recompute POS relative to that.
7865 However, if we're extracting from (or inserting into) a register,
7866 we want to recompute POS relative to wanted_inner_mode. */
7875 else
7876 pos_rtx
7879 pos_rtx);
7880 /* POS may be less than 0 now, but we check for that below.
7881 Note that it can only be less than 0 if !MEM_P (inner). */
7882 }
7884 /* If INNER has a wider mode, and this is a constant extraction, try to
7885 make it smaller and adjust the byte to point to the byte containing
7886 the value. */
7889 && ! pos_rtx
7894 {
7897 /* The computations below will be correct if the machine is big
7898 endian in both bits and bytes or little endian in bits and bytes.
7899 If it is mixed, we must adjust. */
7901 /* If bytes are big endian and we had a paradoxical SUBREG, we must
7902 adjust OFFSET to compensate. */
7907 /* We can now move to the desired byte. */
7918 }
7920 /* If INNER is not memory, get it into the proper mode. If we are changing
7921 its mode, POS must be a constant and smaller than the size of the new
7922 mode. */
7924 {
7925 /* On the LHS, don't create paradoxical subregs implicitely truncating
7926 the register unless TARGET_TRULY_NOOP_TRUNCATION. */
7929 wanted_inner_mode))
7941 pos_rtx
7943 ? HOST_WIDE_INT_M1U
7947 }
7949 /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7950 have to zero extend. Otherwise, we can just use a SUBREG.
7952 We dealt with constant rtxes earlier, so pos_rtx cannot
7953 have VOIDmode at this point. */
7957 {
7961 /* If we know that no extraneous bits are set, and that the high
7962 bit is not set, convert extraction to cheaper one - either
7963 SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7964 cases. */
7972 {
7976 /* Prefer ZERO_EXTENSION, since it gives more information to
7977 backends. */
7981 }
7983 }
7985 /* Make POS_RTX unless we already have it and it is correct. If we don't
7986 have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7987 be a CONST_INT. */
7994 /* Make the required operation. See if we can use existing rtx. */
8001 }
8002 \f
8003 /* See if X (of mode MODE) contains an ASHIFT of COUNT or more bits that
8004 can be commuted with any other operations in X. Return X without
8005 that shift if so. */
8007 static rtx
8009 {
8011 rtx tem;
8014 {
8016 /* This is the shift itself. If it is wide enough, we will return
8017 either the value being shifted if the shift count is equal to
8018 COUNT or a shift for the difference. */
8032 /* If we can safely shift this constant and we find the inner shift,
8033 make a new operation. */
8038 {
8042 }
8047 }
8050 }
8051 \f
8052 /* Subroutine of make_compound_operation. *X_PTR is the rtx at the current
8053 level of the expression and MODE is its mode. IN_CODE is as for
8054 make_compound_operation. *NEXT_CODE_PTR is the value of IN_CODE
8055 that should be used when recursing on operands of *X_PTR.
8057 There are two possible actions:
8059 - Return null. This tells the caller to recurse on *X_PTR with IN_CODE
8060 equal to *NEXT_CODE_PTR, after which *X_PTR holds the final value.
8062 - Return a new rtx, which the caller returns directly. */
8064 static rtx
8068 {
8076 rtx tem;
8077 scalar_int_mode inner_mode;
8081 {
8084 }
8086 /* Process depending on the code of this operation. If NEW is set
8087 nonzero, it will be returned. */
8090 {
8092 /* Convert shifts by constants into multiplications if inside
8093 an address. */
8097 {
8103 {
8106 }
8109 }
8118 {
8122 }
8125 {
8129 mode));
8131 }
8132 else
8133 {
8136 }
8146 {
8150 }
8153 {
8157 mode));
8159 }
8160 else
8161 {
8165 }
8168 /* If the second operand is not a constant, we can't do anything
8169 with it. */
8173 /* If the constant is a power of two minus one and the first operand
8174 is a logical right shift, make an extraction. */
8177 {
8181 }
8183 /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
8190 {
8197 /* If we narrowed the mode when dropping the subreg, then we lose. */
8201 /* If that didn't give anything, see if the AND simplifies on
8202 its own. */
8204 {
8208 }
8209 }
8210 /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
8216 {
8217 /* Apply the distributive law, and then try to make extractions. */
8224 }
8226 /* If we are have (and (rotate X C) M) and C is larger than the number
8227 of bits in M, this is an extraction. */
8233 {
8239 }
8241 /* On machines without logical shifts, if the operand of the AND is
8242 a logical shift and our mask turns off all the propagated sign
8243 bits, we can replace the logical shift with an arithmetic shift. */
8251 {
8261 next_code),
8263 }
8265 /* If the constant is one less than a power of two, this might be
8266 representable by an extraction even if no shift is present.
8267 If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
8268 we are in a COMPARE. */
8272 next_code),
8275 /* If we are in a comparison and this is an AND with a power of two,
8276 convert this into the appropriate bit extract. */
8282 next_code),
8285 /* If the one operand is a paradoxical subreg of a register or memory and
8286 the constant (limited to the smaller mode) has only zero bits where
8287 the sub expression has known zero bits, this can be expressed as
8288 a zero_extend. */
8290 {
8291 rtx sub;
8299 {
8303 /* original AND constant with all the known zero bits set */
8306 {
8310 }
8311 }
8312 }
8317 /* If the sign bit is known to be zero, replace this with an
8318 arithmetic shift. */
8323 {
8326 next_code),
8329 }
8331 /* fall through */
8337 /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
8338 this is a SIGN_EXTRACT. */
8345 {
8352 }
8354 /* See if we have operations between an ASHIFTRT and an ASHIFT.
8355 If so, try to merge the shifts into a SIGN_EXTEND. We could
8356 also do this for some cases of SIGN_EXTRACT, but it doesn't
8357 seem worth the effort; the case checked for occurs on Alpha. */
8368 next_code),
8375 /* Call ourselves recursively on the inner expression. If we are
8376 narrowing the object and it has a different RTL code from
8377 what it originally did, do this SUBREG as a force_to_mode. */
8378 {
8382 /* If the SUBREG is masking of a logical right shift,
8383 make an extraction. */
8390 {
8399 }
8401 /* If in_code is COMPARE, it isn't always safe to pass it through
8402 to the recursive make_compound_operation call. */
8406 /* (subreg:SI (and:DI (reg:DI) (const_int 0x800000000)) 0)
8407 is (const_int 0), rather than
8408 (subreg:SI (lshiftrt:DI (reg:DI) (const_int 35)) 0).
8409 Similarly (subreg:QI (and:SI (reg:SI) (const_int 0x80)) 0)
8410 for non-equality comparisons against 0 is not equivalent
8411 to (subreg:QI (lshiftrt:SI (reg:SI) (const_int 7)) 0). */
8421 simplified
8429 {
8430 rtx newer
8433 /* If we have something other than a SUBREG, we might have
8434 done an expansion, so rerun ourselves. */
8438 /* force_to_mode can expand compounds. If it just re-expanded
8439 the compound, use gen_lowpart to convert to the desired
8440 mode. */
8442 /* Likewise if it re-expanded the compound only partially.
8443 This happens for SUBREG of ZERO_EXTRACT if they extract
8444 the same number of bits. */
8453 }
8457 }
8462 }
8468 }
8470 /* Look at the expression rooted at X. Look for expressions
8471 equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
8472 Form these expressions.
8474 Return the new rtx, usually just X.
8476 Also, for machines like the VAX that don't have logical shift insns,
8477 try to convert logical to arithmetic shift operations in cases where
8478 they are equivalent. This undoes the canonicalizations to logical
8479 shifts done elsewhere.
8481 We try, as much as possible, to re-use rtl expressions to save memory.
8483 IN_CODE says what kind of expression we are processing. Normally, it is
8484 SET. In a memory address it is MEM. When processing the arguments of
8485 a comparison or a COMPARE against zero, it is COMPARE, or EQ if more
8486 precisely it is an equality comparison against zero. */
8488 rtx
8490 {
8497 /* Select the code to be used in recursive calls. Once we are inside an
8498 address, we stay there. If we have a comparison, set to COMPARE,
8499 but once inside, go back to our default of SET. */
8506 scalar_int_mode mode;
8508 {
8514 }
8516 /* Now recursively process each operand of this operation. We need to
8517 handle ZERO_EXTEND specially so that we don't lose track of the
8518 inner mode. */
8520 {
8528 }
8533 {
8536 }
8539 {
8542 }
8546 }
8547 \f
8548 /* Given M see if it is a value that would select a field of bits
8549 within an item, but not the entire word. Return -1 if not.
8550 Otherwise, return the starting position of the field, where 0 is the
8551 low-order bit.
8553 *PLEN is set to the length of the field. */
8555 static int
8557 {
8558 /* Get the bit number of the first 1 bit from the right, -1 if none. */
8563 /* Now shift off the low-order zero bits and see if we have a
8564 power of two minus 1. */
8572 }
8573 \f
8574 /* If X refers to a register that equals REG in value, replace these
8575 references with REG. */
8576 static rtx
8578 {
8586 {
8619 /* FALLTHRU */
8623 {
8627 else
8629 }
8631 /* fall through */
8638 {
8641 {
8643 {
8646 }
8648 }
8649 }
8651 {
8654 {
8657 {
8659 {
8662 }
8664 }
8665 }
8666 }
8669 }
8672 }
8674 /* Return X converted to MODE. If the value is already truncated to
8675 MODE we can just return a subreg even though in the general case we
8676 would need an explicit truncation. */
8678 static rtx
8680 {
8685 {
8686 /* Bit-cast X into an integer mode. */
8691 }
8694 }
8696 /* See if X can be simplified knowing that we will only refer to it in
8697 MODE and will only refer to those bits that are nonzero in MASK.
8698 If other bits are being computed or if masking operations are done
8699 that select a superset of the bits in MASK, they can sometimes be
8700 ignored.
8702 Return a possibly simplified expression, but always convert X to
8703 MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8705 If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
8706 are all off in X. This is used when X will be complemented, by either
8707 NOT, NEG, or XOR. */
8709 static rtx
8712 {
8715 machine_mode op_mode;
8718 /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8719 code below will do the wrong thing since the mode of such an
8720 expression is VOIDmode.
8722 Also do nothing if X is a CLOBBER; this can happen if X was
8723 the return value from a call to gen_lowpart. */
8727 /* We want to perform the operation in its present mode unless we know
8728 that the operation is valid in MODE, in which case we do the operation
8729 in MODE. */
8734 /* It is not valid to do a right-shift in a narrower mode
8735 than the one it came in with. */
8740 /* Truncate MASK to fit OP_MODE. */
8744 /* Determine what bits of X are guaranteed to be (non)zero. */
8747 /* If none of the bits in X are needed, return a zero. */
8751 /* If X is a CONST_INT, return a new one. Do this here since the
8752 test below will fail. */
8754 {
8757 else
8758 {
8761 }
8762 }
8764 /* If X is narrower than MODE and we want all the bits in X's mode, just
8765 get X in the proper mode. */
8770 /* We can ignore the effect of a SUBREG if it narrows the mode or
8771 if the constant masks to zero all the bits the mode doesn't have. */
8775 || (mask
8783 /* OP_MODE is either MODE or XMODE, so it must be a scalar
8784 integer too. */
8790 }
8792 /* Subroutine of force_to_mode that handles cases in which both X and
8793 the result are scalar integers. MODE is the mode of the result,
8794 XMODE is the mode of X, and OP_MODE says which of MODE or XMODE
8795 is preferred for simplified versions of X. The other arguments
8796 are as for force_to_mode. */
8798 static rtx
8802 {
8807 poly_int64 const_op0;
8809 /* When we have an arithmetic operation, or a shift whose count we
8810 do not know, we need to assume that all bits up to the highest-order
8811 bit in MASK will be needed. This is how we form such a mask. */
8814 else
8819 {
8821 /* If X is a (clobber (const_int)), return it since we know we are
8822 generating something that won't match. */
8835 /* Similarly for a truncate. */
8839 /* If this is an AND with a constant, convert it into an AND
8840 whose constant is the AND of that constant with MASK. If it
8841 remains an AND of MASK, delete it since it is redundant. */
8844 {
8849 /* If X is still an AND, see if it is an AND with a mask that
8850 is just some low-order bits. If so, and it is MASK, we don't
8851 need it. */
8857 /* If it remains an AND, try making another AND with the bits
8858 in the mode mask that aren't in MASK turned on. If the
8859 constant in the AND is wide enough, this might make a
8860 cheaper constant. */
8865 {
8866 unsigned HOST_WIDE_INT cval
8868 rtx y;
8875 }
8878 }
8883 /* In (and (plus FOO C1) M), if M is a mask that just turns off
8884 low-order bits (as in an alignment operation) and FOO is already
8885 aligned to that boundary, mask C1 to that boundary as well.
8886 This may eliminate that PLUS and, later, the AND. */
8888 {
8892 /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8893 number, sign extend it. */
8906 }
8908 /* fall through */
8911 /* Substituting into the operands of a widening MULT is not likely to
8912 create RTL matching a machine insn. */
8922 /* For PLUS, MINUS and MULT, we need any bits less significant than the
8923 most significant bit in MASK since carries from those bits will
8924 affect the bits we are interested in. */
8929 /* If X is (minus C Y) where C's least set bit is larger than any bit
8930 in the mask, then we may replace with (neg Y). */
8933 {
8936 }
8938 /* Similarly, if C contains every bit in the fuller_mask, then we may
8939 replace with (not Y). */
8942 {
8945 }
8952 /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8953 LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8954 operation which may be a bitfield extraction. Ensure that the
8955 constant we form is not wider than the mode of X. */
8967 {
8970 xmode);
8976 }
8978 binop:
8979 /* For most binary operations, just propagate into the operation and
8980 change the mode if we have an operation of that mode. */
8985 /* If we ended up truncating both operands, truncate the result of the
8986 operation instead. */
8989 {
8992 }
8998 {
9001 }
9005 /* For left shifts, do the same, but just for the first operand.
9006 However, we cannot do anything with shifts where we cannot
9007 guarantee that the counts are smaller than the size of the mode
9008 because such a count will have a different meaning in a
9009 wider mode. */
9019 /* If the shift count is a constant and we can do arithmetic in
9020 the mode of the shift, refine which bits we need. Otherwise, use the
9021 conservative form of the mask. */
9027 else
9035 {
9038 }
9042 /* Here we can only do something if the shift count is a constant,
9043 this shift constant is valid for the host, and we can do arithmetic
9044 in OP_MODE. */
9050 {
9054 /* Select the mask of the bits we need for the shift operand. */
9057 /* We can only change the mode of the shift if we can do arithmetic
9058 in the mode of the shift and INNER_MASK is no wider than the
9059 width of X's mode. */
9066 {
9069 }
9070 }
9072 /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
9073 shift and AND produces only copies of the sign bit (C2 is one less
9074 than a power of two), we can do this with just a shift. */
9078 /* The shift puts one of the sign bit copies in the least significant
9079 bit. */
9084 /* Number of bits left after the shift must be more than the mask
9085 needs. */
9088 /* Must be more sign bit copies than the mask needs. */
9091 {
9095 }
9099 /* If we are just looking for the sign bit, we don't need this shift at
9100 all, even if it has a variable count. */
9104 /* If this is a shift by a constant, get a mask that contains those bits
9105 that are not copies of the sign bit. We then have two cases: If
9106 MASK only includes those bits, this can be a logical shift, which may
9107 allow simplifications. If MASK is a single-bit field not within
9108 those bits, we are requesting a copy of the sign bit and hence can
9109 shift the sign bit to the appropriate location. */
9113 {
9117 /* If the considered data is wider than HOST_WIDE_INT, we can't
9118 represent a mask for all its bits in a single scalar.
9119 But we only care about the lower bits, so calculate these. */
9122 {
9125 /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
9126 is the number of bits a full-width mask would have set.
9127 We need only shift if these are fewer than nonzero can
9128 hold. If not, we must keep all bits set in nonzero. */
9133 + HOST_BITS_PER_WIDE_INT
9135 }
9136 else
9137 {
9140 }
9143 {
9148 }
9151 {
9152 x = simplify_shift_const
9158 }
9159 }
9161 /* If MASK is 1, convert this to an LSHIFTRT. This can be done
9162 even if the shift count isn't a constant. */
9166 shiftrt:
9168 /* If this is a zero- or sign-extension operation that just affects bits
9169 we don't care about, remove it. Be sure the call above returned
9170 something that is still a shift. */
9180 next_select);
9186 /* If the shift count is constant and we can do computations
9187 in the mode of X, compute where the bits we care about are.
9188 Otherwise, we can't do anything. Don't change the mode of
9189 the shift or propagate MODE into the shift, though. */
9192 {
9201 }
9205 /* If we just want the low-order bit, the NEG isn't needed since it
9206 won't change the low-order bit. */
9210 /* We need any bits less significant than the most significant bit in
9211 MASK since carries from those bits will affect the bits we are
9212 interested in. */
9217 /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
9218 same as the XOR case above. Ensure that the constant we form is not
9219 wider than the mode of X. */
9227 {
9234 }
9236 /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
9237 use the full mask inside the NOT. */
9240 unop:
9243 next_select));
9245 {
9248 }
9252 /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
9253 in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
9254 which is equal to STORE_FLAG_VALUE. */
9266 /* We have no way of knowing if the IF_THEN_ELSE can itself be
9267 written in a narrower mode. We play it safe and do not do so. */
9283 }
9285 /* Ensure we return a value of the proper mode. */
9287 }
9288 \f
9289 /* Return nonzero if X is an expression that has one of two values depending on
9290 whether some other value is zero or nonzero. In that case, we return the
9291 value that is being tested, *PTRUE is set to the value if the rtx being
9292 returned has a nonzero value, and *PFALSE is set to the other alternative.
9294 If we return zero, we set *PTRUE and *PFALSE to X. */
9296 static rtx
9298 {
9303 scalar_int_mode int_mode;
9305 /* If we are comparing a value against zero, we are done. */
9308 {
9312 }
9314 /* If this is a unary operation whose operand has one of two values, apply
9315 our opcode to compute those values. */
9318 {
9323 }
9325 /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
9326 make can't possibly match and would suppress other optimizations. */
9328 ;
9330 /* If this is a binary operation, see if either side has only one of two
9331 values. If either one does or if both do and they are conditional on
9332 the same value, compute the new true and false values. */
9334 {
9342 {
9343 /* Try to enable a simplification by undoing work done by
9344 if_then_else_cond if it converted a REG into something more
9345 complex. */
9347 {
9350 }
9351 else
9352 {
9355 }
9356 }
9360 {
9361 /* If if_then_else_cond returned zero, then true/false are the
9362 same rtl. We must copy one of them to prevent invalid rtl
9363 sharing. */
9370 {
9375 }
9376 else
9377 {
9380 }
9383 }
9385 /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
9386 operands is zero when the other is nonzero, and vice-versa,
9387 and STORE_FLAG_VALUE is 1 or -1. */
9393 {
9411 {
9418 const_true_rtx);
9420 }
9421 }
9423 /* Similarly for MULT, AND and UMIN, except that for these the result
9424 is always zero. */
9428 {
9442 {
9445 }
9446 }
9447 }
9450 {
9451 /* If we have IF_THEN_ELSE already, extract the condition and
9452 canonicalize it if it is NE or EQ. */
9458 {
9461 }
9462 else
9464 }
9466 /* If X is a SUBREG, we can narrow both the true and false values
9467 if the inner expression, if there is a condition. */
9471 {
9477 {
9481 }
9482 }
9484 /* If X is a constant, this isn't special and will cause confusions
9485 if we treat it as such. Likewise if it is equivalent to a constant. */
9488 ;
9490 /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
9491 will be least confusing to the rest of the compiler. */
9493 {
9496 }
9498 /* If X is known to be either 0 or -1, those are the true and
9499 false values when testing X. */
9504 {
9507 }
9509 /* Likewise for 0 or a single bit. */
9512 {
9515 }
9517 /* Otherwise fail; show no condition with true and false values the same. */
9520 }
9521 \f
9522 /* Return the value of expression X given the fact that condition COND
9523 is known to be true when applied to REG as its first operand and VAL
9524 as its second. X is known to not be shared and so can be modified in
9525 place.
9527 We only handle the simplest cases, and specifically those cases that
9528 arise with IF_THEN_ELSE expressions. */
9530 static rtx
9532 {
9540 /* If either operand of the condition is a floating point value,
9541 then we have to avoid collapsing an EQ comparison. */
9551 /* If X is (abs REG) and we know something about REG's relationship
9552 with zero, we may be able to simplify this. */
9556 {
9565 }
9567 /* The only other cases we handle are MIN, MAX, and comparisons if the
9568 operands are the same as REG and VAL. */
9571 {
9573 {
9576 }
9579 {
9581 {
9589 else
9591 }
9594 {
9597 /* Do not reverse the condition when it is NE or EQ.
9598 This is because we cannot conclude anything about
9599 the value of 'SMAX (x, y)' when x is not equal to y,
9600 but we can when x equals y. */
9606 {
9617 }
9618 }
9619 }
9620 }
9622 {
9627 {
9628 /* We must simplify subreg here, before we lose track of the
9629 original inner_mode. */
9634 else
9636 }
9639 }
9640 /* We don't have to handle SIGN_EXTEND here, because even in the
9641 case of replacing something with a modeless CONST_INT, a
9642 CONST_INT is already (supposed to be) a valid sign extension for
9643 its narrower mode, which implies it's already properly
9644 sign-extended for the wider mode. Now, for ZERO_EXTEND, the
9645 story is different. */
9647 {
9652 {
9653 /* We must simplify the zero_extend here, before we lose
9654 track of the original inner_mode. */
9659 else
9661 }
9664 }
9668 {
9675 }
9678 }
9679 \f
9680 /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9681 assignment as a field assignment. */
9683 static int
9685 {
9687 {
9695 }
9703 /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9704 Note that all SUBREGs of MEM are paradoxical; otherwise they
9705 would have been rewritten. */
9718 /* We used to see if get_last_value of X and Y were the same but that's
9719 not correct. In one direction, we'll cause the assignment to have
9720 the wrong destination and in the case, we'll import a register into this
9721 insn that might have already have been dead. So fail if none of the
9722 above cases are true. */
9724 }
9725 \f
9726 /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9727 Return that assignment if so.
9729 We only handle the most common cases. */
9731 static rtx
9733 {
9736 rtx assign;
9738 HOST_WIDE_INT c1;
9739 HOST_WIDE_INT pos;
9741 rtx other;
9743 /* All the rules in this function are specific to scalar integers. */
9744 scalar_int_mode mode;
9748 /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9749 a clear of a one-bit field. We will have changed it to
9750 (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9751 for a SUBREG. */
9757 {
9763 }
9772 {
9779 }
9781 /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9782 one-bit field. */
9786 {
9792 }
9794 /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9795 SRC is an AND with all bits of that field set, then we can discard
9796 the AND. */
9801 {
9808 else
9811 /* Complete overlap. We can remove the source AND. */
9815 /* Partial overlap. We can reduce the source AND. */
9817 {
9821 }
9822 }
9824 /* The other case we handle is assignments into a constant-position
9825 field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9826 a mask that has all one bits except for a group of zero bits and
9827 OTHER is known to have zeros where C1 has ones, this is such an
9828 assignment. Compute the position and length from C1. Shift OTHER
9829 to the appropriate position, force it to the required mode, and
9830 make the extraction. Check for the AND in both operands. */
9832 /* One or more SUBREGs might obscure the constant-position field
9833 assignment. The first one we are likely to encounter is an outer
9834 narrowing SUBREG, which we can just strip for the purposes of
9835 identifying the constant-field assignment. */
9852 /* The second SUBREG that might get in the way is a paradoxical
9853 SUBREG around the first operand of the AND. We want to
9854 pretend the operand is as wide as the destination here. We
9855 do this by adjusting the MEM to wider mode for the sole
9856 purpose of the call to rtx_equal_for_field_assignment_p. Also
9857 note this trick only works for MEMs. */
9869 /* The second SUBREG that might get in the way is a paradoxical
9870 SUBREG around the first operand of the AND. We want to
9871 pretend the operand is as wide as the destination here. We
9872 do this by adjusting the MEM to wider mode for the sole
9873 purpose of the call to rtx_equal_for_field_assignment_p. Also
9874 note this trick only works for MEMs. */
9882 else
9896 /* The mode to use for the source is the mode of the assignment, or of
9897 what is inside a possible STRICT_LOW_PART. */
9901 /* Shift OTHER right POS places and make it the source, restricting it
9902 to the proper length and mode. */
9906 dest);
9908 len >= HOST_BITS_PER_WIDE_INT
9909 ? HOST_WIDE_INT_M1U
9913 /* If SRC is masked by an AND that does not make a difference in
9914 the value being stored, strip it. */
9925 }
9926 \f
9927 /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9928 if so. */
9930 static rtx
9932 {
9936 rtx tem;
9938 /* Distributivity is not true for floating point as it can change the
9939 value. So we don't do it unless -funsafe-math-optimizations. */
9944 /* The outer operation can only be one of the following: */
9952 /* If either operand is a primitive we can't do anything, so get out
9953 fast. */
9963 /* See if the inner and outer operations distribute. */
9965 {
9970 /* These all distribute except over PLUS. */
9981 /* This is also a multiply, so it distributes over everything. */
9984 /* This used to handle SUBREG, but this turned out to be counter-
9985 productive, since (subreg (op ...)) usually is not handled by
9986 insn patterns, and this "optimization" therefore transformed
9987 recognizable patterns into unrecognizable ones. Therefore the
9988 SUBREG case was removed from here.
9990 It is possible that distributing SUBREG over arithmetic operations
9991 leads to an intermediate result than can then be optimized further,
9992 e.g. by moving the outer SUBREG to the other side of a SET as done
9993 in simplify_set. This seems to have been the original intent of
9994 handling SUBREGs here.
9996 However, with current GCC this does not appear to actually happen,
9997 at least on major platforms. If some case is found where removing
9998 the SUBREG case here prevents follow-on optimizations, distributing
9999 SUBREGs ought to be re-added at that place, e.g. in simplify_set. */
10003 }
10005 /* Set LHS and RHS to the inner operands (A and B in the example
10006 above) and set OTHER to the common operand (C in the example).
10007 There is only one way to do this unless the inner operation is
10008 commutative. */
10020 else
10023 /* Form the new inner operation, seeing if it simplifies first. */
10026 /* There is one exception to the general way of distributing:
10027 (a | c) ^ (b | c) -> (a ^ b) & ~c */
10029 {
10032 }
10034 /* We may be able to continuing distributing the result, so call
10035 ourselves recursively on the inner operation before forming the
10036 outer operation, which we return. */
10039 }
10041 /* See if X is of the form (* (+ A B) C), and if so convert to
10042 (+ (* A C) (* B C)) and try to simplify.
10044 Most of the time, this results in no change. However, if some of
10045 the operands are the same or inverses of each other, simplifications
10046 will result.
10048 For example, (and (ior A B) (not B)) can occur as the result of
10049 expanding a bit field assignment. When we apply the distributive
10050 law to this, we get (ior (and (A (not B))) (and (B (not B)))),
10051 which then simplifies to (and (A (not B))).
10053 Note that no checks happen on the validity of applying the inverse
10054 distributive law. This is pointless since we can do it in the
10055 few places where this routine is called.
10057 N is the index of the term that is decomposed (the arithmetic operation,
10058 i.e. (+ A B) in the first example above). !N is the index of the term that
10059 is distributed, i.e. of C in the first example above. */
10060 static rtx
10062 {
10063 machine_mode mode;
10067 /* Distributivity is not true for floating point as it can change the
10068 value. So we don't do it unless -funsafe-math-optimizations. */
10085 /* Special case (and (xor B C) (not A)), which is equivalent to
10086 (xor (ior A B) (ior A C)) */
10088 {
10091 }
10094 {
10095 /* Distribute the second term. */
10098 }
10099 else
10100 {
10101 /* Distribute the first term. */
10104 }
10114 }
10115 \f
10116 /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
10117 in MODE. Return an equivalent form, if different from (and VAROP
10118 (const_int CONSTOP)). Otherwise, return NULL_RTX. */
10120 static rtx
10123 {
10126 rtx orig_varop;
10134 /* Simplify VAROP knowing that we will be only looking at some of the
10135 bits in it.
10137 Note by passing in CONSTOP, we guarantee that the bits not set in
10138 CONSTOP are not significant and will never be examined. We must
10139 ensure that is the case by explicitly masking out those bits
10140 before returning. */
10143 /* If VAROP is a CLOBBER, we will fail so return it. */
10147 /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
10148 to VAROP and return the new constant. */
10152 /* See what bits may be nonzero in VAROP. Unlike the general case of
10153 a call to nonzero_bits, here we don't care about bits outside
10154 MODE. */
10158 /* Turn off all bits in the constant that are known to already be zero.
10159 Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
10160 which is tested below. */
10164 /* If we don't have any bits left, return zero. */
10168 /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
10169 a power of two, we can replace this with an ASHIFT. */
10174 /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
10175 or XOR, then try to apply the distributive law. This may eliminate
10176 operations if either branch can be simplified because of the AND.
10177 It may also make some cases more complex, but those cases probably
10178 won't match a pattern either with or without this. */
10181 {
10183 return
10184 gen_lowpart
10186 apply_distributive_law
10190 constop),
10193 constop))));
10194 }
10196 /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
10197 the AND and see if one of the operands simplifies to zero. If so, we
10198 may eliminate it. */
10202 {
10211 }
10213 /* Make a SUBREG if necessary. If we can't make it, fail. */
10218 /* If we are only masking insignificant bits, return VAROP. */
10225 /* Otherwise, return an AND. */
10227 }
10230 /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
10231 in MODE.
10233 Return an equivalent form, if different from X. Otherwise, return X. If
10234 X is zero, we are to always construct the equivalent form. */
10236 static rtx
10239 {
10250 }
10251 \f
10252 /* Given a REG X of mode XMODE, compute which bits in X can be nonzero.
10253 We don't care about bits outside of those defined in MODE.
10254 We DO care about all the bits in MODE, even if XMODE is smaller than MODE.
10256 For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
10257 a shift, AND, or zero_extract, we can do better. */
10259 static rtx
10261 scalar_int_mode mode,
10263 {
10264 rtx tem;
10267 /* If X is a register whose nonzero bits value is current, use it.
10268 Otherwise, if X is a register whose value we can find, use that
10269 value. Otherwise, use the previously-computed global nonzero bits
10270 for this register. */
10285 && !REGNO_REG_SET_P
10288 {
10289 /* Note that, even if the precision of last_set_mode is lower than that
10290 of mode, record_value_for_reg invoked nonzero_bits on the register
10291 with nonzero_bits_mode (because last_set_mode is necessarily integral
10292 and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
10293 are all valid, hence in mode too since nonzero_bits_mode is defined
10294 to the largest HWI_COMPUTABLE_MODE_P mode. */
10297 }
10301 {
10306 }
10309 {
10313 /* We don't know anything about the upper bits. */
10317 }
10320 }
10322 /* Given a reg X of mode XMODE, return the number of bits at the high-order
10323 end of X that are known to be equal to the sign bit. X will be used
10324 in mode MODE; the returned value will always be between 1 and the
10325 number of bits in MODE. */
10327 static rtx
10329 scalar_int_mode mode,
10331 {
10332 rtx tem;
10345 && !REGNO_REG_SET_P
10348 {
10351 }
10362 }
10363 \f
10364 /* Return the number of "extended" bits there are in X, when interpreted
10365 as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
10366 unsigned quantities, this is the number of high-order zero bits.
10367 For signed quantities, this is the number of copies of the sign bit
10368 minus 1. In both case, this function returns the number of "spare"
10369 bits. For example, if two quantities for which this function returns
10370 at least 1 are added, the addition is known not to overflow.
10372 This function will always return 0 unless called during combine, which
10373 implies that it must be called from a define_split. */
10375 unsigned int
10377 {
10381 scalar_int_mode int_mode;
10389 }
10391 /* This function is called from `simplify_shift_const' to merge two
10392 outer operations. Specifically, we have already found that we need
10393 to perform operation *POP0 with constant *PCONST0 at the outermost
10394 position. We would now like to also perform OP1 with constant CONST1
10395 (with *POP0 being done last).
10397 Return 1 if we can do the operation and update *POP0 and *PCONST0 with
10398 the resulting operation. *PCOMP_P is set to 1 if we would need to
10399 complement the innermost operand, otherwise it is unchanged.
10401 MODE is the mode in which the operation will be done. No bits outside
10402 the width of this mode matter. It is assumed that the width of this mode
10403 is smaller than or equal to HOST_BITS_PER_WIDE_INT.
10405 If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
10406 IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
10407 result is simply *PCONST0.
10409 If the resulting operation cannot be expressed as one operation, we
10410 return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
10412 static int
10413 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)
10414 {
10421 /* If OP0 is an AND, clear unimportant bits in CONST1. */
10425 /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
10426 if OP0 is SET. */
10435 {
10437 {
10455 }
10456 }
10458 /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
10462 /* If the two constants aren't the same, we can't do anything. The
10463 remaining six cases can all be done. */
10467 else
10469 {
10472 /* (a & b) | b == b */
10475 /* (a ^ b) | b == a | b */
10476 {;}
10481 /* (a & b) ^ b == (~a) & b */
10484 /* (a | b) ^ b == a & ~b */
10490 /* (a | b) & b == b */
10493 /* (a ^ b) & b) == (~a) & b */
10498 }
10500 /* Check for NO-OP cases. */
10513 /* ??? Slightly redundant with the above mask, but not entirely.
10514 Moving this above means we'd have to sign-extend the mode mask
10515 for the final test. */
10520 }
10521 \f
10522 /* A helper to simplify_shift_const_1 to determine the mode we can perform
10523 the shift in. The original shift operation CODE is performed on OP in
10524 ORIG_MODE. Return the wider mode MODE if we can perform the operation
10525 in that mode. Return ORIG_MODE otherwise. We can also assume that the
10526 result of the shift is subject to operation OUTER_CODE with operand
10527 OUTER_CONST. */
10529 static scalar_int_mode
10533 {
10536 /* In general we can't perform in wider mode for right shift and rotate. */
10538 {
10540 /* We can still widen if the bits brought in from the left are identical
10541 to the sign bit of ORIG_MODE. */
10549 /* Similarly here but with zero bits. */
10554 /* We can also widen if the bits brought in will be masked off. This
10555 operation is performed in ORIG_MODE. */
10557 {
10563 }
10564 /* fall through */
10574 }
10575 }
10577 /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
10578 of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
10579 if we cannot simplify it. Otherwise, return a simplified value.
10581 The shift is normally computed in the widest mode we find in VAROP, as
10582 long as it isn't a different number of words than RESULT_MODE. Exceptions
10583 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10585 static rtx
10588 {
10593 machine_mode shift_mode;
10595 /* We form (outer_op (code varop count) (outer_const)). */
10601 /* Make sure and truncate the "natural" shift on the way in. We don't
10602 want to do this inside the loop as it makes it more difficult to
10603 combine shifts. */
10607 /* If we were given an invalid count, don't do anything except exactly
10608 what was requested. */
10615 /* Unless one of the branches of the `if' in this loop does a `continue',
10616 we will `break' the loop after the `if'. */
10619 {
10620 /* If we have an operand of (clobber (const_int 0)), fail. */
10624 /* Convert ROTATERT to ROTATE. */
10626 {
10630 }
10634 {
10635 /* We only change the modes of scalar shifts. */
10641 }
10643 scalar_int_mode shift_unit_mode
10646 /* Handle cases where the count is greater than the size of the mode
10647 minus 1. For ASHIFT, use the size minus one as the count (this can
10648 occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
10649 take the count modulo the size. For other shifts, the result is
10650 zero.
10652 Since these shifts are being produced by the compiler by combining
10653 multiple operations, each of which are defined, we know what the
10654 result is supposed to be. */
10657 {
10662 else
10663 {
10664 /* We can't simply return zero because there may be an
10665 outer op. */
10669 }
10670 }
10672 /* If we discovered we had to complement VAROP, leave. Making a NOT
10673 here would cause an infinite loop. */
10678 {
10679 /* An arithmetic right shift of a quantity known to be -1 or 0
10680 is a no-op. */
10684 {
10687 }
10689 /* If we are doing an arithmetic right shift and discarding all but
10690 the sign bit copies, this is equivalent to doing a shift by the
10691 bitsize minus one. Convert it into that shift because it will
10692 often allow other simplifications. */
10699 /* We simplify the tests below and elsewhere by converting
10700 ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10701 `make_compound_operation' will convert it to an ASHIFTRT for
10702 those machines (such as VAX) that don't have an LSHIFTRT. */
10707 shift_unit_mode)))
10719 }
10722 {
10729 {
10732 }
10736 /* The following rules apply only to scalars. */
10741 /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10742 minus the width of a smaller mode, we can do this with a
10743 SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10750 {
10759 }
10763 /* The following rules apply only to scalars. */
10769 /* If VAROP is a SUBREG, strip it as long as the inner operand has
10770 the same number of words as what we've seen so far. Then store
10771 the widest mode in MODE. */
10778 {
10783 }
10787 /* Some machines use MULT instead of ASHIFT because MULT
10788 is cheaper. But it is still better on those machines to
10789 merge two shifts into one. */
10792 {
10797 }
10801 /* Similar, for when divides are cheaper. */
10804 {
10809 }
10813 /* If we are extracting just the sign bit of an arithmetic
10814 right shift, that shift is not needed. However, the sign
10815 bit of a wider mode may be different from what would be
10816 interpreted as the sign bit in a narrower mode, so, if
10817 the result is narrower, don't discard the shift. */
10822 {
10825 }
10827 /* fall through */
10832 /* The following rules apply only to scalars. */
10839 /* Here we have two nested shifts. The result is usually the
10840 AND of a new shift with a mask. We compute the result below. */
10846 {
10850 rtx mask_rtx;
10852 /* We have one common special case. We can't do any merging if
10853 the inner code is an ASHIFTRT of a smaller mode. However, if
10854 we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10855 with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10856 we can convert it to
10857 (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10858 This simplifies certain SIGN_EXTEND operations. */
10862 {
10863 /* C3 has the low-order C1 bits zero. */
10875 }
10877 /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10878 than C1 high-order bits equal to the sign bit, we can convert
10879 this to either an ASHIFT or an ASHIFTRT depending on the
10880 two counts.
10882 We cannot do this if VAROP's mode is not SHIFT_UNIT_MODE. */
10888 {
10892 {
10895 }
10898 }
10900 /* There are some cases we can't do. If CODE is ASHIFTRT,
10901 we can only do this if FIRST_CODE is also ASHIFTRT.
10903 We can't do the case when CODE is ROTATE and FIRST_CODE is
10904 ASHIFTRT.
10906 If the mode of this shift is not the mode of the outer shift,
10907 we can't do this if either shift is a right shift or ROTATE.
10909 Finally, we can't do any of these if the mode is too wide
10910 unless the codes are the same.
10912 Handle the case where the shift codes are the same
10913 first. */
10916 {
10925 }
10936 /* To compute the mask to apply after the shift, shift the
10937 nonzero bits of the inner shift the same way the
10938 outer shift will. */
10941 int_result_mode);
10943 mask_rtx
10947 /* Give up if we can't compute an outer operation to use. */
10955 /* If the shifts are in the same direction, we add the
10956 counts. Otherwise, we subtract them. */
10960 else
10963 /* If COUNT is positive, the new shift is usually CODE,
10964 except for the two exceptions below, in which case it is
10965 FIRST_CODE. If the count is negative, FIRST_CODE should
10966 always be used */
10976 }
10978 /* If we have (A << B << C) for any shift, we can convert this to
10979 (A << C << B). This wins if A is a constant. Only try this if
10980 B is not a constant. */
10985 {
10986 /* For ((unsigned) (cstULL >> count)) >> cst2 we have to make
10987 sure the result will be masked. See PR70222. */
10995 /* For ((int) (cstLL >> count)) >> cst2 just give up. Queuing
10996 up outer sign extension (often left and right shift) is
10997 hardly more efficient than the original. See PR70429. */
11004 count_rtx);
11008 }
11012 /* The following rules apply only to scalars. */
11016 /* Make this fit the case below. */
11023 /* The following rules apply only to scalars. */
11029 /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
11030 with C the size of VAROP - 1 and the shift is logical if
11031 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
11032 we have an (le X 0) operation. If we have an arithmetic shift
11033 and STORE_FLAG_VALUE is 1 or we have a logical shift with
11034 STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
11042 {
11045 const0_rtx);
11051 }
11053 /* If we have (shift (logical)), move the logical to the outside
11054 to allow it to possibly combine with another logical and the
11055 shift to combine with another shift. This also canonicalizes to
11056 what a ZERO_EXTRACT looks like. Also, some machines have
11057 (and (shift)) insns. */
11060 /* We can't do this if we have (ashiftrt (xor)) and the
11061 constant has its sign bit set in shift_unit_mode with
11062 shift_unit_mode wider than result_mode. */
11075 {
11078 }
11080 /* If we can't do that, try to simplify the shift in each arm of the
11081 logical expression, make a new logical expression, and apply
11082 the inverse distributive law. This also can't be done for
11083 (ashiftrt (xor)) where we've widened the shift and the constant
11084 changes the sign bit. */
11090 {
11102 }
11106 /* The following rules apply only to scalars. */
11111 /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
11112 says that the sign bit can be tested, FOO has mode MODE, C is
11113 GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
11114 that may be nonzero. */
11124 {
11128 }
11132 /* The following rules apply only to scalars. */
11137 /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
11138 than the number of bits in the mode is equivalent to A. */
11142 {
11146 }
11148 /* NEG commutes with ASHIFT since it is multiplication. Move the
11149 NEG outside to allow shifts to combine. */
11153 {
11156 }
11160 /* The following rules apply only to scalars. */
11165 /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
11166 is one less than the number of bits in the mode is
11167 equivalent to (xor A 1). */
11174 {
11178 }
11180 /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
11181 that might be nonzero in BAR are those being shifted out and those
11182 bits are known zero in FOO, we can replace the PLUS with FOO.
11183 Similarly in the other operand order. This code occurs when
11184 we are computing the size of a variable-size array. */
11191 {
11194 }
11202 {
11205 }
11207 /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
11218 {
11221 }
11223 /* Check for 'PLUS signbit', which is the canonical form of 'XOR
11224 signbit', and attempt to change the PLUS to an XOR and move it to
11225 the outer operation as is done above in the AND/IOR/XOR case
11226 leg for shift(logical). See details in logical handling above
11227 for reasoning in doing so. */
11239 {
11242 }
11247 /* The following rules apply only to scalars. */
11252 /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
11253 with C the size of VAROP - 1 and the shift is logical if
11254 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
11255 we have a (gt X 0) operation. If the shift is arithmetic with
11256 STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
11257 we have a (neg (gt X 0)) operation. */
11266 {
11269 const0_rtx);
11275 }
11279 /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
11280 if the truncate does not affect the value. */
11287 {
11291 new_count);
11294 new_count_rtx);
11298 }
11303 }
11306 }
11310 {
11311 /* We only change the modes of scalar shifts. */
11316 }
11318 /* We have now finished analyzing the shift. The result should be
11319 a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
11320 OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
11321 to the result of the shift. OUTER_CONST is the relevant constant,
11322 but we must turn off all bits turned off in the shift. */
11330 /* Make a SUBREG if necessary. If we can't make it, fail. */
11335 /* If we have an outer operation and we just made a shift, it is
11336 possible that we could have simplified the shift were it not
11337 for the outer operation. So try to do the simplification
11338 recursively. */
11342 else
11349 /* If we were doing an LSHIFTRT in a wider mode than it was originally,
11350 turn off all the bits that the shift would have turned off. */
11352 /* We only change the modes of scalar shifts. */
11356 /* Do the remainder of the processing in RESULT_MODE. */
11359 /* If COMPLEMENT_P is set, we have to complement X before doing the outer
11360 operation. */
11365 {
11375 {
11376 /* This means that we have determined that the result is
11377 equivalent to a constant. This should be rare. */
11380 }
11383 else
11386 }
11389 }
11391 /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
11392 The result of the shift is RESULT_MODE. If we cannot simplify it,
11393 return X or, if it is NULL, synthesize the expression with
11394 simplify_gen_binary. Otherwise, return a simplified value.
11396 The shift is normally computed in the widest mode we find in VAROP, as
11397 long as it isn't a different number of words than RESULT_MODE. Exceptions
11398 are ASHIFTRT and ROTATE, which are always done in their original mode. */
11400 static rtx
11403 {
11414 }
11416 \f
11417 /* A subroutine of recog_for_combine. See there for arguments and
11418 return value. */
11420 static int
11422 {
11424 rtx pat_without_clobbers;
11432 /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
11433 we use to indicate that something didn't match. If we find such a
11434 thing, force rejection. */
11448 {
11451 else
11454 }
11456 /* If it isn't, there is the possibility that we previously had an insn
11457 that clobbered some register as a side effect, but the combined
11458 insn doesn't need to do that. So try once more without the clobbers
11459 unless this represents an ASM insn. */
11463 {
11468 {
11471 pos++;
11472 }
11482 {
11485 else
11488 }
11489 }
11496 /* Recognize all noop sets, these will be killed by followup pass. */
11500 /* If we had any clobbers to add, make a new pattern than contains
11501 them. Then check to make sure that all of them are dead. */
11503 {
11513 else
11520 {
11525 {
11529 }
11530 }
11532 }
11536 {
11544 /* Allow targets to reject combined insn. */
11546 {
11549 dump_file);
11551 /* Callers expect recog_for_combine to strip
11552 clobbers from the pattern on failure. */
11557 }
11562 }
11568 }
11570 /* Change every ZERO_EXTRACT and ZERO_EXTEND of a SUBREG that can be
11571 expressed as an AND and maybe an LSHIFTRT, to that formulation.
11572 Return whether anything was so changed. */
11574 static bool
11576 {
11582 {
11594 {
11604 else
11608 {
11614 }
11615 }
11621 {
11626 {
11632 }
11633 }
11638 {
11642 }
11643 else
11649 {
11652 }
11656 }
11663 scalar_int_mode mode;
11669 {
11683 else
11691 }
11694 }
11696 /* Like recog, but we receive the address of a pointer to a new pattern.
11697 We try to match the rtx that the pointer points to.
11698 If that fails, we may try to modify or replace the pattern,
11699 storing the replacement into the same pointer object.
11701 Modifications include deletion or addition of CLOBBERs. If the
11702 instruction will still not match, we change ZERO_EXTEND and ZERO_EXTRACT
11703 to the equivalent AND and perhaps LSHIFTRT patterns, and try with that
11704 (and undo if that fails).
11706 PNOTES is a pointer to a location where any REG_UNUSED notes added for
11707 the CLOBBERs are placed.
11709 The value is the final insn code from the pattern ultimately matched,
11710 or -1. */
11712 static int
11714 {
11726 {
11729 {
11733 }
11734 }
11737 {
11742 }
11745 }
11746 \f
11747 /* Like gen_lowpart_general but for use by combine. In combine it
11748 is not possible to create any new pseudoregs. However, it is
11749 safe to create invalid memory addresses, because combine will
11750 try to recognize them and all they will do is make the combine
11751 attempt fail.
11753 If for some reason this cannot do its job, an rtx
11754 (clobber (const_int 0)) is returned.
11755 An insn containing that will not be recognized. */
11757 static rtx
11759 {
11761 rtx result;
11766 /* We can only support MODE being wider than a word if X is a
11767 constant integer or has a mode the same size. */
11773 /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
11774 won't know what to do. So we will strip off the SUBREG here and
11775 process normally. */
11777 {
11780 /* For use in case we fall down into the address adjustments
11781 further below, we need to adjust the known mode and size of
11782 x; imode and isize, since we just adjusted x. */
11787 }
11795 {
11796 /* Refuse to work on a volatile memory ref or one with a mode-dependent
11797 address. */
11802 /* If we want to refer to something bigger than the original memref,
11803 generate a paradoxical subreg instead. That will force a reload
11804 of the original memref X. */
11810 }
11812 /* If X is a comparison operator, rewrite it in a new mode. This
11813 probably won't match, but may allow further simplifications. */
11819 /* If we couldn't simplify X any other way, just enclose it in a
11820 SUBREG. Normally, this SUBREG won't match, but some patterns may
11821 include an explicit SUBREG or we may simplify it further in combine. */
11822 else
11823 {
11824 rtx res;
11827 {
11832 }
11836 }
11838 fail:
11840 }
11841 \f
11842 /* Try to simplify a comparison between OP0 and a constant OP1,
11843 where CODE is the comparison code that will be tested, into a
11844 (CODE OP0 const0_rtx) form.
11846 The result is a possibly different comparison code to use.
11847 *POP1 may be updated. */
11849 static enum rtx_code
11852 {
11853 scalar_int_mode int_mode;
11856 /* Get the constant we are comparing against and turn off all bits
11857 not on in our mode. */
11861 /* If we are comparing against a constant power of two and the value
11862 being compared can only have that single bit nonzero (e.g., it was
11863 `and'ed with that bit), we can replace this with a comparison
11864 with zero. */
11873 {
11876 }
11878 /* Similarly, if we are comparing a value known to be either -1 or
11879 0 with -1, change it to the opposite comparison against zero. */
11885 {
11888 }
11890 /* Do some canonicalizations based on the comparison code. We prefer
11891 comparisons against zero and then prefer equality comparisons.
11892 If we can reduce the size of a constant, we will do that too. */
11894 {
11896 /* < C is equivalent to <= (C - 1) */
11898 {
11901 /* ... fall through to LE case below. */
11903 }
11904 else
11908 /* <= C is equivalent to < (C + 1); we do this for C < 0 */
11910 {
11913 }
11915 /* If we are doing a <= 0 comparison on a value known to have
11916 a zero sign bit, we can replace this with == 0. */
11927 /* >= C is equivalent to > (C - 1). */
11929 {
11932 /* ... fall through to GT below. */
11934 }
11935 else
11939 /* > C is equivalent to >= (C + 1); we do this for C < 0. */
11941 {
11944 }
11946 /* If we are doing a > 0 comparison on a value known to have
11947 a zero sign bit, we can replace this with != 0. */
11958 /* < C is equivalent to <= (C - 1). */
11960 {
11963 /* ... fall through ... */
11965 }
11966 /* (unsigned) < 0x80000000 is equivalent to >= 0. */
11971 {
11975 }
11976 else
11980 /* unsigned <= 0 is equivalent to == 0 */
11983 /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
11987 == ((HOST_WIDE_INT_1U
11989 {
11992 }
11996 /* >= C is equivalent to > (C - 1). */
11998 {
12001 /* ... fall through ... */
12003 }
12005 /* (unsigned) >= 0x80000000 is equivalent to < 0. */
12010 {
12014 }
12015 else
12019 /* unsigned > 0 is equivalent to != 0 */
12022 /* (unsigned) > 0x7fffffff is equivalent to < 0. */
12026 == (HOST_WIDE_INT_1U
12028 {
12031 }
12036 }
12040 }
12041 \f
12042 /* Simplify a comparison between *POP0 and *POP1 where CODE is the
12043 comparison code that will be tested.
12045 The result is a possibly different comparison code to use. *POP0 and
12046 *POP1 may be updated.
12048 It is possible that we might detect that a comparison is either always
12049 true or always false. However, we do not perform general constant
12050 folding in combine, so this knowledge isn't useful. Such tautologies
12051 should have been detected earlier. Hence we ignore all such cases. */
12053 static enum rtx_code
12055 {
12061 opt_scalar_int_mode tmode_iter;
12063 /* Try a few ways of applying the same transformation to both operands. */
12065 {
12066 /* The test below this one won't handle SIGN_EXTENDs on these machines,
12067 so check specially. */
12086 {
12089 }
12091 /* If both operands are the same constant shift, see if we can ignore the
12092 shift. We can if the shift is a rotate or if the bits shifted out of
12093 this shift are known to be zero for both inputs and if the type of
12094 comparison is compatible with the shift. */
12107 {
12120 else
12122 }
12124 /* If both operands are AND's of a paradoxical SUBREG by constant, the
12125 SUBREGs are of the same mode, and, in both cases, the AND would
12126 be redundant if the comparison was done in the narrower mode,
12127 do the comparison in the narrower mode (e.g., we are AND'ing with 1
12128 and the operand's possibly nonzero bits are 0xffffff01; in that case
12129 if we only care about QImode, we don't need the AND). This case
12130 occurs if the output mode of an scc insn is not SImode and
12131 STORE_FLAG_VALUE == 1 (e.g., the 386).
12133 Similarly, check for a case where the AND's are ZERO_EXTEND
12134 operations from some narrower mode even though a SUBREG is not
12135 present. */
12140 {
12156 {
12160 /* The resulting comparison is always unsigned since we masked
12161 off the original sign bit. */
12165 }
12171 {
12177 }
12181 }
12183 /* If both operands are NOT, we can strip off the outer operation
12184 and adjust the comparison code for swapped operands; similarly for
12185 NEG, except that this must be an equality comparison. */
12191 else
12193 }
12195 /* If the first operand is a constant, swap the operands and adjust the
12196 comparison code appropriately, but don't do this if the second operand
12197 is already a constant integer. */
12199 {
12202 }
12204 /* We now enter a loop during which we will try to simplify the comparison.
12205 For the most part, we only are concerned with comparisons with zero,
12206 but some things may really be comparisons with zero but not start
12207 out looking that way. */
12210 {
12212 scalar_int_mode int_mode;
12216 HOST_WIDE_INT const_op;
12218 /* We only want to handle integral modes. This catches VOIDmode,
12219 CCmode, and the floating-point modes. An exception is that we
12220 can handle VOIDmode if OP0 is a COMPARE or a comparison
12221 operation. */
12228 /* Try to simplify the compare to constant, possibly changing the
12229 comparison op, and/or changing op1 to zero. */
12233 /* Compute some predicates to simplify code below. */
12240 /* If this is a sign bit comparison and we can do arithmetic in
12241 MODE, say that we will only be needing the sign bit of OP0. */
12246 HOST_WIDE_INT_1U
12251 {
12252 /* We can't do anything if OP0 is a condition code value, rather
12253 than an actual data value. */
12259 /* Get the two operands being compared. */
12262 else
12265 /* Check for the cases where we simply want the result of the
12266 earlier test or the opposite of that result. */
12270 {
12274 else
12278 {
12283 }
12284 }
12286 }
12292 /* Now try cases based on the opcode of OP0. If none of the cases
12293 does a "continue", we exit this loop immediately after the
12294 switch. */
12299 {
12301 /* If we are extracting a single bit from a variable position in
12302 a constant that has only a single bit set and are comparing it
12303 with zero, we can convert this into an equality comparison
12304 between the position and the location of the single bit. */
12305 /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
12306 have already reduced the shift count modulo the word size. */
12312 {
12320 /* Result is nonzero iff shift count is equal to I. */
12323 }
12325 /* fall through */
12330 {
12333 }
12337 /* If testing for equality, we can take the NOT of the constant. */
12340 {
12344 }
12346 /* If just looking at the sign bit, reverse the sense of the
12347 comparison. */
12349 {
12353 }
12357 /* If testing for equality, we can take the NEG of the constant. */
12360 {
12364 }
12366 /* The remaining cases only apply to comparisons with zero. */
12370 /* When X is ABS or is known positive,
12371 (neg X) is < 0 if and only if X != 0. */
12379 {
12383 }
12385 /* If we have NEG of something whose two high-order bits are the
12386 same, we know that "(-a) < 0" is equivalent to "a > 0". */
12388 {
12392 }
12396 /* If we are testing equality and our count is a constant, we
12397 can perform the inverse operation on our RHS. */
12401 {
12405 }
12407 /* If we are doing a < 0 or >= 0 comparison, it means we are testing
12408 a particular bit. Convert it to an AND of a constant of that
12409 bit. This will be converted into a ZERO_EXTRACT. */
12414 {
12416 (HOST_WIDE_INT_1U
12421 }
12423 /* Fall through. */
12426 /* ABS is ignorable inside an equality comparison with zero. */
12428 {
12431 }
12435 /* Can simplify (compare (zero/sign_extend FOO) CONST) to
12436 (compare FOO CONST) if CONST fits in FOO's mode and we
12437 are either testing inequality or have an unsigned
12438 comparison with ZERO_EXTEND or a signed comparison with
12439 SIGN_EXTEND. But don't do it if we don't have a compare
12440 insn of the given mode, since we'd have to revert it
12441 later on, and then we wouldn't know whether to sign- or
12442 zero-extend. */
12444 && ! unsigned_comparison_p
12448 {
12451 }
12455 /* Check for the case where we are comparing A - C1 with C2, that is
12457 (subreg:MODE (plus (A) (-C1))) op (C2)
12459 with C1 a constant, and try to lift the SUBREG, i.e. to do the
12460 comparison in the wider mode. One of the following two conditions
12461 must be true in order for this to be valid:
12463 1. The mode extension results in the same bit pattern being added
12464 on both sides and the comparison is equality or unsigned. As
12465 C2 has been truncated to fit in MODE, the pattern can only be
12466 all 0s or all 1s.
12468 2. The mode extension results in the sign bit being copied on
12469 each side.
12471 The difficulty here is that we have predicates for A but not for
12472 (A - C1) so we need to check that C1 is within proper bounds so
12473 as to perturbate A as little as possible. */
12482 {
12490 /* (A - C1) zero-extends if it is positive and sign-extends
12491 if it is negative, C2 both zero- and sign-extends. */
12495 /* (A - C1) sign-extends if it is positive and 1-extends
12496 if it is negative, C2 both sign- and 1-extends. */
12503 /* (A - C1) always sign-extends, like C2. */
12507 {
12510 }
12511 }
12513 /* If the inner mode is narrower and we are extracting the low part,
12514 we can treat the SUBREG as if it were a ZERO_EXTEND. */
12516 ;
12525 {
12526 /* Remove outer subregs that don't do anything. */
12531 {
12535 }
12537 }
12538 else
12541 /* FALLTHROUGH */
12550 {
12553 }
12557 /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
12558 this for equality comparisons due to pathological cases involving
12559 overflows. */
12563 {
12567 }
12569 /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
12572 {
12576 }
12580 /* We used to optimize signed comparisons against zero, but that
12581 was incorrect. Unsigned comparisons against zero (GTU, LEU)
12582 arrive here as equality comparisons, or (GEU, LTU) are
12583 optimized away. No need to special-case them. */
12585 /* (eq (minus A B) C) -> (eq A (plus B C)) or
12586 (eq B (minus A C)), whichever simplifies. We can only do
12587 this for equality comparisons due to pathological cases involving
12588 overflows. */
12592 {
12596 }
12601 {
12605 }
12607 /* The sign bit of (minus (ashiftrt X C) X), where C is the number
12608 of bits in X minus 1, is one iff X > 0. */
12613 {
12617 }
12621 /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
12622 if C is zero or B is a constant. */
12626 {
12630 }
12635 /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
12636 iff X <= 0. */
12640 {
12644 }
12648 /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
12649 will be converted to a ZERO_EXTRACT later. */
12653 {
12658 }
12660 /* If we are comparing (and (lshiftrt X C1) C2) for equality with
12661 zero and X is a comparison and C1 and C2 describe only bits set
12662 in STORE_FLAG_VALUE, we can compare with X. */
12670 {
12677 {
12680 }
12681 }
12683 /* If we are doing an equality comparison of an AND of a bit equal
12684 to the sign bit, replace this with a LT or GE comparison of
12685 the underlying value. */
12692 {
12696 }
12698 /* If this AND operation is really a ZERO_EXTEND from a narrower
12699 mode, the constant fits within that mode, and this is either an
12700 equality or unsigned comparison, try to do this comparison in
12701 the narrower mode.
12703 Note that in:
12705 (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
12706 -> (ne:DI (reg:SI 4) (const_int 0))
12708 unless TARGET_TRULY_NOOP_TRUNCATION allows it or the register is
12709 known to hold a value of the required mode the
12710 transformation is invalid. */
12718 {
12721 }
12723 /* If this is (and:M1 (subreg:M1 X:M2 0) (const_int C1)) where C1
12724 fits in both M1 and M2 and the SUBREG is either paradoxical
12725 or represents the low part, permute the SUBREG and the AND
12726 and try again. */
12729 {
12731 /* Require an integral mode, to avoid creating something like
12732 (AND:SF ...). */
12735 /* It is unsafe to commute the AND into the SUBREG if the
12736 SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
12737 not defined. As originally written the upper bits
12738 have a defined value due to the AND operation.
12739 However, if we commute the AND inside the SUBREG then
12740 they no longer have defined values and the meaning of
12741 the code has been changed.
12742 Also C1 should not change value in the smaller mode,
12743 see PR67028 (a positive C1 can become negative in the
12744 smaller mode, so that the AND does no longer mask the
12745 upper bits). */
12746 && ((WORD_REGISTER_OPERATIONS
12758 {
12764 }
12765 }
12767 /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
12771 {
12776 }
12778 /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
12779 (eq (and (lshiftrt X) 1) 0).
12780 Also handle the case where (not X) is expressed using xor. */
12784 {
12794 == HOST_WIDE_INT_1U
12796 {
12797 op0
12802 }
12803 }
12807 /* If we have (compare (ashift FOO N) (const_int C)) and
12808 the high order N bits of FOO (N+1 if an inequality comparison)
12809 are known to be zero, we can do this by comparing FOO with C
12810 shifted right N bits so long as the low-order N bits of C are
12811 zero. */
12823 {
12824 /* We must perform a logical shift, not an arithmetic one,
12825 as we want the top N bits of C to be zero. */
12832 }
12834 /* If we are doing a sign bit comparison, it means we are testing
12835 a particular bit. Convert it to the appropriate AND. */
12838 {
12840 (HOST_WIDE_INT_1U
12845 }
12847 /* If this an equality comparison with zero and we are shifting
12848 the low bit to the sign bit, we can convert this to an AND of the
12849 low-order bit. */
12853 {
12856 }
12860 /* If this is an equality comparison with zero, we can do this
12861 as a logical shift, which might be much simpler. */
12864 {
12869 }
12871 /* If OP0 is a sign extension and CODE is not an unsigned comparison,
12872 do the comparison in a narrower mode. */
12882 {
12885 }
12887 /* Likewise if OP0 is a PLUS of a sign extension with a
12888 constant, which is usually represented with the PLUS
12889 between the shifts. */
12901 {
12909 new_const);
12911 }
12913 /* FALLTHROUGH */
12915 /* If we have (compare (xshiftrt FOO N) (const_int C)) and
12916 the low order N bits of FOO are known to be zero, we can do this
12917 by comparing FOO with C shifted left N bits so long as no
12918 overflow occurs. Even if the low order N bits of FOO aren't known
12919 to be zero, if the comparison is >= or < we can use the same
12920 optimization and for > or <= by setting all the low
12921 order N bits in the comparison constant. */
12932 {
12933 unsigned HOST_WIDE_INT low_bits
12935 & ((HOST_WIDE_INT_1U
12938 {
12939 /* If the shift was logical, then we must make the condition
12940 unsigned. */
12949 const_op
12954 }
12955 }
12957 /* If we are using this shift to extract just the sign bit, we
12958 can replace this with an LT or GE comparison. */
12963 {
12967 }
12972 }
12975 }
12977 /* Now make any compound operations involved in this comparison. Then,
12978 check for an outmost SUBREG on OP0 that is not doing anything or is
12979 paradoxical. The latter transformation must only be performed when
12980 it is known that the "extra" bits will be the same in op0 and op1 or
12981 that they don't matter. There are three cases to consider:
12983 1. SUBREG_REG (op0) is a register. In this case the bits are don't
12984 care bits and we can assume they have any convenient value. So
12985 making the transformation is safe.
12987 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is UNKNOWN.
12988 In this case the upper bits of op0 are undefined. We should not make
12989 the simplification in that case as we do not know the contents of
12990 those bits.
12992 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not UNKNOWN.
12993 In that case we know those bits are zeros or ones. We must also be
12994 sure that they are the same as the upper bits of op1.
12996 We can never remove a SUBREG for a non-equality comparison because
12997 the sign bit is in a different place in the underlying object. */
13010 {
13012 {
13013 /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
13014 implemented. */
13016 {
13019 }
13020 }
13024 {
13029 }
13030 }
13032 /* We now do the opposite procedure: Some machines don't have compare
13033 insns in all modes. If OP0's mode is an integer mode smaller than a
13034 word and we can't do a compare in that mode, see if there is a larger
13035 mode for which we can do the compare. There are a number of cases in
13036 which we can use the wider mode. */
13042 {
13047 {
13050 /* If this is a test for negative, we can make an explicit
13051 test of the sign bit. Test this first so we can use
13052 a paradoxical subreg to extend OP0. */
13056 {
13057 unsigned HOST_WIDE_INT sign
13064 }
13066 /* If the only nonzero bits in OP0 and OP1 are those in the
13067 narrower mode and this is an equality or unsigned comparison,
13068 we can use the wider mode. Similarly for sign-extended
13069 values, in which case it is true for all comparisons. */
13086 {
13087 /* If OP0 is an AND and we don't have an AND in MODE either,
13088 make a new AND in the proper mode. */
13096 else
13097 {
13099 {
13104 }
13105 else
13106 {
13111 }
13113 }
13114 }
13115 }
13116 }
13118 /* We may have changed the comparison operands. Re-canonicalize. */
13120 {
13123 }
13125 /* If this machine only supports a subset of valid comparisons, see if we
13126 can convert an unsupported one into a supported one. */
13133 }
13134 \f
13135 /* Utility function for record_value_for_reg. Count number of
13136 rtxs in X. */
13137 static int
13139 {
13146 {
13166 }
13177 }
13178 \f
13179 /* Utility function for following routine. Called when X is part of a value
13180 being stored into last_set_value. Sets last_set_table_tick
13181 for each register mentioned. Similar to mention_regs in cse.c */
13183 static void
13185 {
13191 {
13197 {
13200 }
13203 }
13207 {
13208 /* Check for identical subexpressions. If x contains
13209 identical subexpression we only have to traverse one of
13210 them. */
13212 {
13213 /* Note that at this point x1 has already been
13214 processed. */
13218 /* If x0 and x1 are identical then there is no need to
13219 process x0. */
13223 /* If x0 is identical to a subexpression of x1 then while
13224 processing x1, x0 has already been processed. Thus we
13225 are done with x. */
13230 /* If x1 is identical to a subexpression of x0 then we
13231 still have to process the rest of x0. */
13234 {
13237 }
13238 }
13241 }
13245 }
13247 /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
13248 are saying that the register is clobbered and we no longer know its
13249 value. If INSN is zero, don't update reg_stat[].last_set; this is
13250 only permitted with VALUE also zero and is used to invalidate the
13251 register. */
13253 static void
13255 {
13261 /* If VALUE contains REG and we have a previous value for REG, substitute
13262 the previous value. */
13264 {
13265 rtx tem;
13267 /* Set things up so get_last_value is allowed to see anything set up to
13268 our insn. */
13272 /* If TEM is simply a binary operation with two CLOBBERs as operands,
13273 it isn't going to be useful and will take a lot of time to process,
13274 so just use the CLOBBER. */
13277 {
13283 {
13284 /* If there are two or more occurrences of REG in VALUE,
13285 prevent the value from growing too much. */
13288 }
13291 }
13292 }
13294 /* For each register modified, show we don't know its value, that
13295 we don't know about its bitwise content, that its value has been
13296 updated, and that we don't know the location of the death of the
13297 register. */
13299 {
13311 }
13313 /* Mark registers that are being referenced in this value. */
13317 /* Now update the status of each register being set.
13318 If someone is using this register in this block, set this register
13319 to invalid since we will get confused between the two lives in this
13320 basic block. This makes using this register always invalid. In cse, we
13321 scan the table to invalidate all entries using this register, but this
13322 is too much work for us. */
13325 {
13331 else
13333 }
13335 /* The value being assigned might refer to X (like in "x++;"). In that
13336 case, we must replace it with (clobber (const_int 0)) to prevent
13337 infinite loops. */
13340 {
13344 }
13346 /* For the main register being modified, update the value, the mode, the
13347 nonzero bits, and the number of sign bit copies. */
13352 {
13360 rsp->last_set_sign_bit_copies
13362 }
13363 }
13365 /* Called via note_stores from record_dead_and_set_regs to handle one
13366 SET or CLOBBER in an insn. DATA is the instruction in which the
13367 set is occurring. */
13369 static void
13371 {
13378 {
13382 }
13385 {
13386 /* If we are setting the whole register, we know its value. Otherwise
13387 show that we don't know the value. We can handle a SUBREG if it's
13388 the low part, but we must be careful with paradoxical SUBREGs on
13389 RISC architectures because we cannot strip e.g. an extension around
13390 a load and record the naked load since the RTL middle-end considers
13391 that the upper bits are defined according to LOAD_EXTEND_OP. */
13398 BITS_PER_WORD)
13401 WORD_REGISTER_OPERATIONS
13407 else
13409 }
13411 /* Ignore pushes, they clobber nothing. */
13414 }
13416 /* Update the records of when each REG was most recently set or killed
13417 for the things done by INSN. This is the last thing done in processing
13418 INSN in the combiner loop.
13420 We update reg_stat[], in particular fields last_set, last_set_value,
13421 last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
13422 last_death, and also the similar information mem_last_set (which insn
13423 most recently modified memory) and last_call_luid (which insn was the
13424 most recent subroutine call). */
13426 static void
13428 {
13429 rtx link;
13433 {
13436 {
13441 {
13446 }
13447 }
13450 }
13453 {
13454 HARD_REG_SET callee_clobbers
13456 hard_reg_set_iterator hrsi;
13458 {
13461 /* ??? We could try to preserve some information from the last
13462 set of register I if the call doesn't actually clobber
13463 (reg:last_set_mode I), which might be true for ABIs with
13464 partial clobbers. However, it would be difficult to
13465 update last_set_nonzero_bits and last_sign_bit_copies
13466 to account for the part of I that actually was clobbered.
13467 It wouldn't help much anyway, since we rarely see this
13468 situation before RA. */
13478 }
13482 /* We can't combine into a call pattern. Remember, though, that
13483 the return value register is set at this LUID. We could
13484 still replace a register with the return value from the
13485 wrong subroutine call! */
13487 }
13488 else
13490 }
13492 /* If a SUBREG has the promoted bit set, it is in fact a property of the
13493 register present in the SUBREG, so for each such SUBREG go back and
13494 adjust nonzero and sign bit information of the registers that are
13495 known to have some zero/sign bits set.
13497 This is needed because when combine blows the SUBREGs away, the
13498 information on zero/sign bits is lost and further combines can be
13499 missed because of that. */
13501 static void
13503 {
13505 rtx set;
13513 {
13522 {
13525 }
13529 {
13532 }
13535 {
13538 }
13539 else
13541 }
13542 }
13544 /* Check if X, a register, is known to contain a value already
13545 truncated to MODE. In this case we can use a subreg to refer to
13546 the truncated value even though in the generic case we would need
13547 an explicit truncation. */
13549 static bool
13551 {
13563 }
13565 /* If X is a hard reg or a subreg record the mode that the register is
13566 accessed in. For non-TARGET_TRULY_NOOP_TRUNCATION targets we might be
13567 able to turn a truncate into a subreg using this information. Return true
13568 if traversing X is complete. */
13570 static bool
13572 {
13573 machine_mode truncated_mode;
13577 {
13589 }
13590 /* ??? For hard-regs we now record everything. We might be able to
13591 optimize this using last_set_mode. */
13594 else
13601 {
13604 }
13607 }
13609 /* Callback for note_uses. Find hardregs and subregs of pseudos and
13610 the modes they are used in. This can help truning TRUNCATEs into
13611 SUBREGs. */
13613 static void
13615 {
13620 }
13622 /* Scan X for promoted SUBREGs. For each one found,
13623 note what it implies to the registers used in it. */
13625 static void
13627 {
13632 else
13633 {
13639 {
13649 }
13650 }
13651 }
13652 \f
13653 /* Verify that all the registers and memory references mentioned in *LOC are
13654 still valid. *LOC was part of a value set in INSN when label_tick was
13655 equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
13656 the invalid references with (clobber (const_int 0)) and return 1. This
13657 replacement is useful because we often can get useful information about
13658 the form of a value (e.g., if it was produced by a shift that always
13659 produces -1 or 0) even though we don't know exactly what registers it
13660 was produced from. */
13662 static int
13664 {
13671 {
13677 {
13680 /* If this is a pseudo-register that was only set once and not
13681 live at the beginning of the function, it is always valid. */
13685 && (!REGNO_REG_SET_P
13687 regno)))
13689 {
13693 }
13694 }
13697 }
13698 /* If this is a memory reference, make sure that there were no stores after
13699 it that might have clobbered the value. We don't have alias info, so we
13700 assume any store invalidates it. Moreover, we only have local UIDs, so
13701 we also assume that there were stores in the intervening basic blocks. */
13704 {
13708 }
13711 {
13713 {
13714 /* Check for identical subexpressions. If x contains
13715 identical subexpression we only have to traverse one of
13716 them. */
13718 {
13719 /* Note that at this point x0 has already been checked
13720 and found valid. */
13724 /* If x0 and x1 are identical then x is also valid. */
13728 /* If x1 is identical to a subexpression of x0 then
13729 while checking x0, x1 has already been checked. Thus
13730 it is valid and so as x. */
13735 /* If x0 is identical to a subexpression of x1 then x is
13736 valid iff the rest of x1 is valid. */
13739 return
13743 }
13748 }
13754 }
13756 /* If we haven't found a reason for it to be invalid, it is valid. */
13758 }
13760 /* Get the last value assigned to X, if known. Some registers
13761 in the value may be replaced with (clobber (const_int 0)) if their value
13762 is known longer known reliably. */
13764 static rtx
13766 {
13768 rtx value;
13771 /* If this is a non-paradoxical SUBREG, get the value of its operand and
13772 then convert it to the desired mode. If this is a paradoxical SUBREG,
13773 we cannot predict what values the "extra" bits might have. */
13787 /* If we don't have a value, or if it isn't for this basic block and
13788 it's either a hard register, set more than once, or it's a live
13789 at the beginning of the function, return 0.
13791 Because if it's not live at the beginning of the function then the reg
13792 is always set before being used (is never used without being set).
13793 And, if it's set only once, and it's always set before use, then all
13794 uses must have the same last value, even if it's not from this basic
13795 block. */
13802 || REGNO_REG_SET_P
13806 /* If the value was set in a later insn than the ones we are processing,
13807 we can't use it even if the register was only set once. */
13812 /* If fewer bits were set than what we are asked for now, we cannot use
13813 the value. */
13818 /* If the value has all its registers valid, return it. */
13822 /* Otherwise, make a copy and replace any invalid register with
13823 (clobber (const_int 0)). If that fails for some reason, return 0. */
13830 }
13831 \f
13832 /* Define three variables used for communication between the following
13833 routines. */
13837 rtx reg_dead_reg;
13839 /* Function called via note_stores from reg_dead_at_p.
13841 If DEST is within [reg_dead_regno, reg_dead_endregno), set
13842 reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
13844 static void
13846 {
13856 }
13858 /* Return nonzero if REG is known to be dead at INSN.
13860 We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
13861 referencing REG, it is dead. If we hit a SET referencing REG, it is
13862 live. Otherwise, see if it is live or dead at the start of the basic
13863 block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
13864 must be assumed to be always live. */
13866 static int
13868 {
13869 basic_block block;
13872 /* Set variables for reg_dead_at_p_1. */
13879 /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
13880 we allow the machine description to decide whether use-and-clobber
13881 patterns are OK. */
13883 {
13887 }
13889 /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
13890 beginning of basic block. */
13893 {
13895 {
13905 }
13911 }
13913 /* Look at live-in sets for the basic block that we were in. */
13919 }
13920 \f
13921 /* Note hard registers in X that are used. */
13923 static void
13925 {
13931 {
13935 CASE_CONST_ANY:
13940 /* CC0 must die in the insn after it is set, so we don't need to take
13941 special note of it here. */
13946 /* If we are clobbering a MEM, mark any hard registers inside the
13947 address as used. */
13954 /* A hard reg in a wide mode may really be multiple registers.
13955 If so, mark all of them just like the first. */
13957 {
13958 /* None of this applies to the stack, frame or arg pointers. */
13960 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
13968 }
13972 {
13973 /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
13974 the address. */
13986 }
13991 }
13993 /* Recursively scan the operands of this expression. */
13995 {
13999 {
14003 {
14008 }
14009 }
14010 }
14011 }
14012 \f
14013 /* Remove register number REGNO from the dead registers list of INSN.
14015 Return the note used to record the death, if there was one. */
14017 rtx
14019 {
14026 }
14028 /* For each register (hardware or pseudo) used within expression X, if its
14029 death is in an instruction with luid between FROM_LUID (inclusive) and
14030 TO_INSN (exclusive), put a REG_DEAD note for that register in the
14031 list headed by PNOTES.
14033 That said, don't move registers killed by maybe_kill_insn.
14035 This is done when X is being merged by combination into TO_INSN. These
14036 notes will then be distributed as needed. */
14038 static void
14041 {
14047 {
14051 /* If we do not know where the register died, it may still die between
14052 FROM_LUID and TO_INSN. If so, find it. This is PR83304. */
14054 {
14059 {
14061 {
14065 }
14068 }
14069 }
14071 /* Don't move the register if it gets killed in between from and to. */
14080 {
14083 /* It is possible for the call above to return 0. This can occur
14084 when last_death points to I2 or I1 that we combined with.
14085 In that case make a new note.
14087 We must also check for the case where X is a hard register
14088 and NOTE is a death note for a range of hard registers
14089 including X. In that case, we must put REG_DEAD notes for
14090 the remaining registers in place of NOTE. */
14094 {
14103 }
14105 /* If we didn't find any note, or if we found a REG_DEAD note that
14106 covers only part of the given reg, and we have a multi-reg hard
14107 register, then to be safe we must check for REG_DEAD notes
14108 for each register other than the first. They could have
14109 their own REG_DEAD notes lying around. */
14116 {
14123 else
14129 }
14132 {
14135 }
14136 else
14138 }
14141 }
14144 {
14149 /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
14150 that accesses one word of a multi-word item, some
14151 piece of everything register in the expression is used by
14152 this insn, so remove any old death. */
14153 /* ??? So why do we test for equality of the sizes? */
14159 {
14162 }
14164 /* If this is some other SUBREG, we know it replaces the entire
14165 value, so use that as the destination. */
14169 /* If this is a MEM, adjust deaths of anything used in the address.
14170 For a REG (the only other possibility), the entire value is
14171 being replaced so the old value is not used in this insn. */
14177 }
14186 {
14188 {
14193 }
14196 }
14197 }
14198 \f
14199 /* Return 1 if X is the target of a bit-field assignment in BODY, the
14200 pattern of an insn. X must be a REG. */
14202 static int
14204 {
14208 {
14210 rtx target;
14217 else
14234 }
14242 }
14243 \f
14244 /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
14245 as appropriate. I3 and I2 are the insns resulting from the combination
14246 insns including FROM (I2 may be zero).
14248 ELIM_I2 and ELIM_I1 are either zero or registers that we know will
14249 not need REG_DEAD notes because they are being substituted for. This
14250 saves searching in the most common cases.
14252 Each note in the list is either ignored or placed on some insns, depending
14253 on the type of note. */
14255 static void
14258 {
14260 rtx tem_note;
14264 {
14269 {
14272 /* Doesn't matter much where we put this, as long as it's somewhere.
14273 It is preferable to keep these notes on branches, which is most
14274 likely to be i3. */
14281 else
14282 {
14285 }
14289 /* These notes must remain with the call or trapping instruction. */
14294 else
14295 {
14301 /* ??? Otherwise assume we've combined things such that we
14302 can now prove that the instructions can't trap. Drop the
14303 note in this case. */
14304 }
14308 /* ??? How to distribute between i3-i1. Assume i3 contains the
14309 entire adjustment. Assert i3 contains at least some adjust. */
14311 {
14313 /* fixup_args_size_notes looks at REG_NORETURN note,
14314 so ensure the note is placed there first. */
14316 {
14320 {
14326 }
14327 }
14329 /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
14330 REG_ARGS_SIZE note to all noreturn calls, allow that here. */
14333 && !ACCUMULATE_OUTGOING_ARGS
14335 }
14343 /* These notes must remain with the call. It should not be
14344 possible for both I2 and I3 to be a call. */
14347 else
14348 {
14351 }
14355 /* Any clobbers for i3 may still exist, and so we must process
14356 REG_UNUSED notes from that insn.
14358 Any clobbers from i2 or i1 can only exist if they were added by
14359 recog_for_combine. In that case, recog_for_combine created the
14360 necessary REG_UNUSED notes. Trying to keep any original
14361 REG_UNUSED notes from these insns can cause incorrect output
14362 if it is for the same register as the original i3 dest.
14363 In that case, we will notice that the register is set in i3,
14364 and then add a REG_UNUSED note for the destination of i3, which
14365 is wrong. However, it is possible to have REG_UNUSED notes from
14366 i2 or i1 for register which were both used and clobbered, so
14367 we keep notes from i2 or i1 if they will turn into REG_DEAD
14368 notes. */
14370 /* If this register is set or clobbered in I3, put the note there
14371 unless there is one already. */
14373 {
14381 }
14382 /* Otherwise, if this register is used by I3, then this register
14383 now dies here, so we must put a REG_DEAD note here unless there
14384 is one already. */
14390 {
14393 }
14395 /* A SET or CLOBBER of the REG_UNUSED reg has been removed,
14396 but we can't tell which at this point. We must reset any
14397 expectations we had about the value that was previously
14398 stored in the reg. ??? Ideally, we'd adjust REG_N_SETS
14399 and, if appropriate, restore its previous value, but we
14400 don't have enough information for that at this point. */
14401 else
14402 {
14405 /* Otherwise, if this register is now referenced in i2
14406 then the register used to be modified in one of the
14407 original insns. If it was i3 (say, in an unused
14408 parallel), it's now completely gone, so the note can
14409 be discarded. But if it was modified in i2, i1 or i0
14410 and we still reference it in i2, then we're
14411 referencing the previous value, and since the
14412 register was modified and REG_UNUSED, we know that
14413 the previous value is now dead. So, if we only
14414 reference the register in i2, we change the note to
14415 REG_DEAD, to reflect the previous value. However, if
14416 we're also setting or clobbering the register as
14417 scratch, we know (because the register was not
14418 referenced in i3) that it's unused, just as it was
14419 unused before, and we place the note in i2. */
14422 {
14431 }
14432 }
14439 /* These notes say something about results of an insn. We can
14440 only support them if they used to be on I3 in which case they
14441 remain on I3. Otherwise they are ignored.
14443 If the note refers to an expression that is not a constant, we
14444 must also ignore the note since we cannot tell whether the
14445 equivalence is still true. It might be possible to do
14446 slightly better than this (we only have a problem if I2DEST
14447 or I1DEST is present in the expression), but it doesn't
14448 seem worth the trouble. */
14456 /* These notes say something about how a register is used. They must
14457 be present on any use of the register in I2 or I3. */
14462 {
14465 else
14467 }
14472 /* This can show up in several ways -- either directly in the
14473 pattern, or hidden off in the constant pool with (or without?)
14474 a REG_EQUAL note. */
14475 /* ??? Ignore the without-reg_equal-note problem for now. */
14487 {
14490 else
14492 }
14494 /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
14495 as a JUMP_LABEL or decrement LABEL_NUSES if it's already
14496 there. */
14501 {
14508 }
14514 {
14522 }
14526 /* This note says something about the value of a register prior
14527 to the execution of an insn. It is too much trouble to see
14528 if the note is still correct in all situations. It is better
14529 to simply delete it. */
14533 /* If we replaced the right hand side of FROM_INSN with a
14534 REG_EQUAL note, the original use of the dying register
14535 will not have been combined into I3 and I2. In such cases,
14536 FROM_INSN is guaranteed to be the first of the combined
14537 instructions, so we simply need to search back before
14538 FROM_INSN for the previous use or set of this register,
14539 then alter the notes there appropriately.
14541 If the register is used as an input in I3, it dies there.
14542 Similarly for I2, if it is nonzero and adjacent to I3.
14544 If the register is not used as an input in either I3 or I2
14545 and it is not one of the registers we were supposed to eliminate,
14546 there are two possibilities. We might have a non-adjacent I2
14547 or we might have somehow eliminated an additional register
14548 from a computation. For example, we might have had A & B where
14549 we discover that B will always be zero. In this case we will
14550 eliminate the reference to A.
14552 In both cases, we must search to see if we can find a previous
14553 use of A and put the death note there. */
14559 else
14560 {
14566 {
14567 /* If the new I2 sets the same register that is marked
14568 dead in the note, we do not in general know where to
14569 put the note. One important case we _can_ handle is
14570 when the note comes from I3. */
14573 else
14575 }
14582 && !(i2mod
14584 i2mod_old_rhs)))
14589 }
14592 {
14596 {
14598 {
14602 }
14604 /* If the register is being set at TEM_INSN, see if that is all
14605 TEM_INSN is doing. If so, delete TEM_INSN. Otherwise, make this
14606 into a REG_UNUSED note instead. Don't delete sets to
14607 global register vars. */
14611 {
14622 ;
14624 /* Verify that it was the set, and not a clobber that
14625 modified the register.
14627 CC0 targets must be careful to maintain setter/user
14628 pairs. If we cannot delete the setter due to side
14629 effects, mark the user with an UNUSED note instead
14630 of deleting it. */
14634 && (!HAVE_cc0
14638 {
14639 /* Move the notes and links of TEM_INSN elsewhere.
14640 This might delete other dead insns recursively.
14641 First set the pattern to something that won't use
14642 any register. */
14661 /* Delete the setter too. */
14663 {
14676 }
14677 }
14678 else
14679 {
14682 /* If there isn't already a REG_UNUSED note, put one
14683 here. Do not place a REG_DEAD note, even if
14684 the register is also used here; that would not
14685 match the algorithm used in lifetime analysis
14686 and can cause the consistency check in the
14687 scheduler to fail. */
14692 }
14693 }
14697 {
14700 /* If we are doing a 3->2 combination, and we have a
14701 register which formerly died in i3 and was not used
14702 by i2, which now no longer dies in i3 and is used in
14703 i2 but does not die in i2, and place is between i2
14704 and i3, then we may need to move a link from place to
14705 i2. */
14707 && from_insn
14710 {
14714 }
14716 }
14720 }
14722 }
14724 /* If the register is set or already dead at PLACE, we needn't do
14725 anything with this note if it is still a REG_DEAD note.
14726 We check here if it is set at all, not if is it totally replaced,
14727 which is what `dead_or_set_p' checks, so also check for it being
14728 set partially. */
14731 {
14737 {
14738 /* Unless the register previously died in PLACE, clear
14739 last_death. [I no longer understand why this is
14740 being done.] */
14744 }
14745 else
14748 /* If this is a death note for a hard reg that is occupying
14749 multiple registers, ensure that we are still using all
14750 parts of the object. If we find a piece of the object
14751 that is unused, we must arrange for an appropriate REG_DEAD
14752 note to be added for it. However, we can't just emit a USE
14753 and tag the note to it, since the register might actually
14754 be dead; so we recourse, and the recursive call then finds
14755 the previous insn that used this register. */
14758 {
14767 {
14770 }
14773 {
14774 /* Put only REG_DEAD notes for pieces that are
14775 not already dead or set. */
14779 {
14786 {
14788 NULL_RTX);
14792 NULL_RTX);
14793 }
14798 {
14800 {
14804 }
14808 {
14811 }
14812 }
14813 }
14816 }
14817 }
14818 }
14822 /* Any other notes should not be present at this point in the
14823 compilation. */
14825 }
14828 {
14832 /* Set added_notes_insn to the earliest insn we added a note to. */
14836 }
14839 {
14842 /* Set added_notes_insn to the earliest insn we added a note to. */
14846 }
14847 }
14848 }
14849 \f
14850 /* Similarly to above, distribute the LOG_LINKS that used to be present on
14851 I3, I2, and I1 to new locations. This is also called to add a link
14852 pointing at I3 when I3's destination is changed. */
14854 static void
14856 {
14860 {
14867 /* If the insn that this link points to is a NOTE, ignore it. */
14876 {
14879 {
14895 }
14898 }
14899 else
14912 /* A LOG_LINK is defined as being placed on the first insn that uses
14913 a register and points to the insn that sets the register. Start
14914 searching at the next insn after the target of the link and stop
14915 when we reach a set of the register or the end of the basic block.
14917 Note that this correctly handles the link that used to point from
14918 I3 to I2. Also note that not much searching is typically done here
14919 since most links don't point very far away. */
14928 {
14932 }
14935 {
14938 }
14942 /* If we found a place to put the link, place it there unless there
14943 is already a link to the same insn as LINK at that point. */
14946 {
14954 {
14958 /* Set added_links_insn to the earliest insn we added a
14959 link to. */
14963 }
14964 }
14965 }
14966 }
14967 \f
14968 /* Check for any register or memory mentioned in EQUIV that is not
14969 mentioned in EXPR. This is used to restrict EQUIV to "specializations"
14970 of EXPR where some registers may have been replaced by constants. */
14972 static bool
14974 {
14977 {
14982 }
14984 }
14985 \f
14986 DEBUG_FUNCTION void
14988 {
14989 fprintf
14991 ";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
14993 }
14995 void
14997 {
14998 fprintf
15000 "\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
15002 }
15003 \f
15004 /* Make pseudo-to-pseudo copies after every hard-reg-to-pseudo-copy, because
15005 the reg-to-reg copy can usefully combine with later instructions, but we
15006 do not want to combine the hard reg into later instructions, for that
15007 restricts register allocation. */
15008 static void
15010 {
15011 basic_block bb;
15014 {
15018 {
15041 }
15042 }
15043 }
15045 /* Try combining insns through substitution. */
15046 static unsigned int
15048 {
15058 int rebuild_jump_labels_after_combine
15061 /* Combining insns may have turned an indirect jump into a
15062 direct jump. Rebuild the JUMP_LABEL fields of jumping
15063 instructions. */
15065 {
15072 }
15076 }
15081 {
15091 };
15094 {
15098 {}
15100 /* opt_pass methods: */
15103 {
15105 }
15111 rtl_opt_pass *
15113 {
15115 }