| blob | 39c26feb24aacaa3447d3e0dbe4432562068d14e |
1 /* Optimize by combining instructions for GNU compiler.
2 Copyright (C) 1987-2017 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 /* This module is essentially the "combiner" phase of the U. of Arizona
21 Portable Optimizer, but redone to work on our list-structured
22 representation for RTL instead of their string representation.
24 The LOG_LINKS of each insn identify the most recent assignment
25 to each REG used in the insn. It is a list of previous insns,
26 each of which contains a SET for a REG that is used in this insn
27 and not used or set in between. LOG_LINKs never cross basic blocks.
28 They were set up by the preceding pass (lifetime analysis).
30 We try to combine each pair of insns joined by a logical link.
31 We also try to combine triplets of insns A, B and C when C has
32 a link back to B and B has a link back to A. Likewise for a
33 small number of quadruplets of insns A, B, C and D for which
34 there's high likelihood of success.
36 LOG_LINKS does not have links for use of the CC0. They don't
37 need to, because the insn that sets the CC0 is always immediately
38 before the insn that tests it. So we always regard a branch
39 insn as having a logical link to the preceding insn. The same is true
40 for an insn explicitly using CC0.
42 We check (with use_crosses_set_p) to avoid combining in such a way
43 as to move a computation to a place where its value would be different.
45 Combination is done by mathematically substituting the previous
46 insn(s) values for the regs they set into the expressions in
47 the later insns that refer to these regs. If the result is a valid insn
48 for our target machine, according to the machine description,
49 we install it, delete the earlier insns, and update the data flow
50 information (LOG_LINKS and REG_NOTES) for what we did.
52 There are a few exceptions where the dataflow information isn't
53 completely updated (however this is only a local issue since it is
54 regenerated before the next pass that uses it):
56 - reg_live_length is not updated
57 - reg_n_refs is not adjusted in the rare case when a register is
58 no longer required in a computation
59 - there are extremely rare cases (see distribute_notes) when a
60 REG_DEAD note is lost
61 - a LOG_LINKS entry that refers to an insn with multiple SETs may be
62 removed because there is no way to know which register it was
63 linking
65 To simplify substitution, we combine only when the earlier insn(s)
66 consist of only a single assignment. To simplify updating afterward,
67 we never combine when a subroutine call appears in the middle.
69 Since we do not represent assignments to CC0 explicitly except when that
70 is all an insn does, there is no LOG_LINKS entry in an insn that uses
71 the condition code for the insn that set the condition code.
72 Fortunately, these two insns must be consecutive.
73 Therefore, every JUMP_INSN is taken to have an implicit logical link
74 to the preceding insn. This is not quite right, since non-jumps can
75 also use the condition code; but in practice such insns would not
76 combine anyway. */
98 /* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
108 /* Number of attempts to combine instructions in this function. */
112 /* Number of attempts that got as far as substitution in this function. */
116 /* Number of instructions combined with added SETs in this function. */
120 /* Number of instructions combined in this function. */
124 /* Totals over entire compilation. */
128 /* combine_instructions may try to replace the right hand side of the
129 second instruction with the value of an associated REG_EQUAL note
130 before throwing it at try_combine. That is problematic when there
131 is a REG_DEAD note for a register used in the old right hand side
132 and can cause distribute_notes to do wrong things. This is the
133 second instruction if it has been so modified, null otherwise. */
137 /* When I2MOD is nonnull, this is a copy of the old right hand side. */
141 /* When I2MOD is nonnull, this is a copy of the new right hand side. */
144 \f
146 /* Record last point of death of (hard or pseudo) register n. */
149 /* Record last point of modification of (hard or pseudo) register n. */
152 /* The next group of fields allows the recording of the last value assigned
153 to (hard or pseudo) register n. We use this information to see if an
154 operation being processed is redundant given a prior operation performed
155 on the register. For example, an `and' with a constant is redundant if
156 all the zero bits are already known to be turned off.
158 We use an approach similar to that used by cse, but change it in the
159 following ways:
161 (1) We do not want to reinitialize at each label.
162 (2) It is useful, but not critical, to know the actual value assigned
163 to a register. Often just its form is helpful.
165 Therefore, we maintain the following fields:
167 last_set_value the last value assigned
168 last_set_label records the value of label_tick when the
169 register was assigned
170 last_set_table_tick records the value of label_tick when a
171 value using the register is assigned
172 last_set_invalid set to nonzero when it is not valid
173 to use the value of this register in some
174 register's value
176 To understand the usage of these tables, it is important to understand
177 the distinction between the value in last_set_value being valid and
178 the register being validly contained in some other expression in the
179 table.
181 (The next two parameters are out of date).
183 reg_stat[i].last_set_value is valid if it is nonzero, and either
184 reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
186 Register I may validly appear in any expression returned for the value
187 of another register if reg_n_sets[i] is 1. It may also appear in the
188 value for register J if reg_stat[j].last_set_invalid is zero, or
189 reg_stat[i].last_set_label < reg_stat[j].last_set_label.
191 If an expression is found in the table containing a register which may
192 not validly appear in an expression, the register is replaced by
193 something that won't match, (clobber (const_int 0)). */
195 /* Record last value assigned to (hard or pseudo) register n. */
197 rtx last_set_value;
199 /* Record the value of label_tick when an expression involving register n
200 is placed in last_set_value. */
204 /* Record the value of label_tick when the value for register n is placed in
205 last_set_value. */
209 /* These fields are maintained in parallel with last_set_value and are
210 used to store the mode in which the register was last set, the bits
211 that were known to be zero when it was last set, and the number of
212 sign bits copies it was known to have when it was last set. */
218 /* Set nonzero if references to register n in expressions should not be
219 used. last_set_invalid is set nonzero when this register is being
220 assigned to and last_set_table_tick == label_tick. */
224 /* Some registers that are set more than once and used in more than one
225 basic block are nevertheless always set in similar ways. For example,
226 a QImode register may be loaded from memory in two places on a machine
227 where byte loads zero extend.
229 We record in the following fields if a register has some leading bits
230 that are always equal to the sign bit, and what we know about the
231 nonzero bits of a register, specifically which bits are known to be
232 zero.
234 If an entry is zero, it means that we don't know anything special. */
240 /* Record the value of the label_tick when the last truncation
241 happened. The field truncated_to_mode is only valid if
242 truncation_label == label_tick. */
246 /* Record the last truncation seen for this register. If truncation
247 is not a nop to this mode we might be able to save an explicit
248 truncation if we know that value already contains a truncated
249 value. */
252 };
257 /* One plus the highest pseudo for which we track REG_N_SETS.
258 regstat_init_n_sets_and_refs allocates the array for REG_N_SETS just once,
259 but during combine_split_insns new pseudos can be created. As we don't have
260 updated DF information in that case, it is hard to initialize the array
261 after growing. The combiner only cares about REG_N_SETS (regno) == 1,
262 so instead of growing the arrays, just assume all newly created pseudos
263 during combine might be set multiple times. */
267 /* Record the luid of the last insn that invalidated memory
268 (anything that writes memory, and subroutine calls, but not pushes). */
272 /* Record the luid of the last CALL_INSN
273 so we can tell whether a potential combination crosses any calls. */
277 /* When `subst' is called, this is the insn that is being modified
278 (by combining in a previous insn). The PATTERN of this insn
279 is still the old pattern partially modified and it should not be
280 looked at, but this may be used to examine the successors of the insn
281 to judge whether a simplification is valid. */
285 /* This is the lowest LUID that `subst' is currently dealing with.
286 get_last_value will not return a value if the register was set at or
287 after this LUID. If not for this mechanism, we could get confused if
288 I2 or I1 in try_combine were an insn that used the old value of a register
289 to obtain a new value. In that case, we might erroneously get the
290 new value of the register when we wanted the old one. */
294 /* This contains any hard registers that are used in newpat; reg_dead_at_p
295 must consider all these registers to be always live. */
299 /* This is an insn to which a LOG_LINKS entry has been added. If this
300 insn is the earlier than I2 or I3, combine should rescan starting at
301 that location. */
305 /* Basic block in which we are performing combines. */
309 \f
310 /* Length of the currently allocated uid_insn_cost array. */
314 /* The following array records the insn_rtx_cost for every insn
315 in the instruction stream. */
319 /* The following array records the LOG_LINKS for every insn in the
320 instruction stream as struct insn_link pointers. */
326 };
332 {
336 }
338 #define INSN_COST(INSN) (uid_insn_cost[insn_uid_check (INSN)])
339 #define LOG_LINKS(INSN) (uid_log_links[insn_uid_check (INSN)])
341 #define FOR_EACH_LOG_LINK(L, INSN) \
342 for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
344 /* Links for LOG_LINKS are allocated from this obstack. */
348 /* Allocate a link. */
352 {
360 }
362 /* Incremented for each basic block. */
366 /* Reset to label_tick for each extended basic block in scanning order. */
370 /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
371 largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
375 /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
376 be safely used. It is zero while computing them and after combine has
377 completed. This former test prevents propagating values based on
378 previously set values, which can be incorrect if a variable is modified
379 in a loop. */
383 \f
384 /* Record one modification to rtl structure
385 to be undone by storing old_contents into *where. */
389 struct undo
390 {
395 };
397 /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
398 num_undo says how many are currently recorded.
400 other_insn is nonzero if we have modified some other insn in the process
401 of working on subst_insn. It must be verified too. */
403 struct undobuf
404 {
408 };
412 /* Number of times the pseudo being substituted for
413 was found and replaced. */
418 machine_mode,
422 machine_mode,
494 \f
496 /* It is not safe to use ordinary gen_lowpart in combine.
497 See comments in gen_lowpart_for_combine. */
498 #undef RTL_HOOKS_GEN_LOWPART
499 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
501 /* Our implementation of gen_lowpart never emits a new pseudo. */
502 #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
503 #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
505 #undef RTL_HOOKS_REG_NONZERO_REG_BITS
506 #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
508 #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
509 #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
511 #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
512 #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
516 \f
517 /* Convenience wrapper for the canonicalize_comparison target hook.
518 Target hooks cannot use enum rtx_code. */
522 {
526 }
528 /* Try to split PATTERN found in INSN. This returns NULL_RTX if
529 PATTERN can not be split. Otherwise, it returns an insn sequence.
530 This is a wrapper around split_insns which ensures that the
531 reg_stat vector is made larger if the splitter creates a new
532 register. */
536 {
545 }
547 /* This is used by find_single_use to locate an rtx in LOC that
548 contains exactly one use of DEST, which is typically either a REG
549 or CC0. It returns a pointer to the innermost rtx expression
550 containing DEST. Appearances of DEST that are being used to
551 totally replace it are not counted. */
555 {
564 {
568 CASE_CONST_ANY:
573 /* If the destination is anything other than CC0, PC, a REG or a SUBREG
574 of a REG that occupies all of the REG, the insn uses DEST if
575 it is mentioned in the destination or the source. Otherwise, we
576 need just check the source. */
596 }
598 /* If it wasn't one of the common cases above, check each expression and
599 vector of this code. Look for a unique usage of DEST. */
603 {
605 {
610 else
616 /* Duplicate usage. */
618 }
620 {
624 {
630 else
637 }
638 }
639 }
642 }
645 /* See if DEST, produced in INSN, is used only a single time in the
646 sequel. If so, return a pointer to the innermost rtx expression in which
647 it is used.
649 If PLOC is nonzero, *PLOC is set to the insn containing the single use.
651 If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
652 care about REG_DEAD notes or LOG_LINKS.
654 Otherwise, we find the single use by finding an insn that has a
655 LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
656 only referenced once in that insn, we know that it must be the first
657 and last insn referencing DEST. */
661 {
662 basic_block bb;
668 {
678 }
688 {
694 {
699 }
700 }
703 }
704 \f
705 /* Substitute NEWVAL, an rtx expression, into INTO, a place in some
706 insn. The substitution can be undone by undo_all. If INTO is already
707 set to NEWVAL, do not record this change. Because computing NEWVAL might
708 also call SUBST, we have to compute it before we put anything into
709 the undo table. */
711 static void
713 {
720 /* We'd like to catch as many invalid transformations here as
721 possible. Unfortunately, there are way too many mode changes
722 that are perfectly valid, so we'd waste too much effort for
723 little gain doing the checks here. Focus on catching invalid
724 transformations involving integer constants. */
727 {
728 /* Sanity check that we're replacing oldval with a CONST_INT
729 that is a valid sign-extension for the original mode. */
733 /* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
734 CONST_INT is not valid, because after the replacement, the
735 original mode would be gone. Unfortunately, we can't tell
736 when do_SUBST is called to replace the operand thereof, so we
737 perform this test on oldval instead, checking whether an
738 invalid replacement took place before we got here. */
743 }
747 else
756 }
758 #define SUBST(INTO, NEWVAL) do_SUBST (&(INTO), (NEWVAL))
760 /* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
761 for the value of a HOST_WIDE_INT value (including CONST_INT) is
762 not safe. */
764 static void
766 {
775 else
784 }
786 #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT (&(INTO), (NEWVAL))
788 /* Similar to SUBST, but just substitute the mode. This is used when
789 changing the mode of a pseudo-register, so that any other
790 references to the entry in the regno_reg_rtx array will change as
791 well. */
793 static void
795 {
804 else
813 }
815 #define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE (&(INTO), (NEWVAL))
817 /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
819 static void
821 {
830 else
839 }
841 #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
842 \f
843 /* Subroutine of try_combine. Determine whether the replacement patterns
844 NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to insn_rtx_cost
845 than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
846 that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
847 undobuf.other_insn may also both be NULL_RTX. Return false if the cost
848 of all the instructions can be estimated and the replacements are more
849 expensive than the original sequence. */
851 static bool
854 {
859 /* Lookup the original insn_rtx_costs. */
864 {
867 {
871 }
872 else
873 {
877 }
878 }
879 else
880 {
883 }
885 /* If we have split a PARALLEL I2 to I1,I2, we have counted its cost twice;
886 correct that. */
891 /* Calculate the replacement insn_rtx_costs. */
894 {
898 }
899 else
900 {
903 }
906 {
912 {
915 }
916 else
918 }
920 /* Disallow this combination if both new_cost and old_cost are greater than
921 zero, and new_cost is greater than old cost. */
925 {
944 else
946 }
951 /* Update the uid_insn_cost array with the replacement costs. */
955 {
959 }
962 }
965 /* Delete any insns that copy a register to itself. */
967 static void
969 {
971 basic_block bb;
974 {
976 {
979 {
984 }
985 }
986 }
987 }
989 \f
990 /* Return false if we do not want to (or cannot) combine DEF. */
991 static bool
993 {
994 /* Do not consider if it is pre/post modification in MEM. */
1000 /* Do not combine frame pointer adjustments. */
1003 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
1011 }
1013 /* Return false if we do not want to (or cannot) combine USE. */
1014 static bool
1016 {
1017 /* Do not consider the usage of the stack pointer by function call. */
1022 }
1024 /* Fill in log links field for all insns. */
1026 static void
1028 {
1029 basic_block bb;
1036 /* Pass through each block from the end, recording the uses of each
1037 register and establishing log links when def is encountered.
1038 Note that we do not clear next_use array in order to save time,
1039 so we have to test whether the use is in the same basic block as def.
1041 There are a few cases below when we do not consider the definition or
1042 usage -- these are taken from original flow.c did. Don't ask me why it is
1043 done this way; I don't know and if it works, I don't want to know. */
1046 {
1048 {
1052 /* Log links are created only once. */
1056 {
1072 /* flow.c claimed:
1074 We don't build a LOG_LINK for hard registers contained
1075 in ASM_OPERANDs. If these registers get replaced,
1076 we might wind up changing the semantics of the insn,
1077 even if reload can make what appear to be valid
1078 assignments later. */
1083 /* Don't add duplicate links between instructions. */
1092 }
1097 }
1098 }
1101 }
1103 /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1104 true if we found a LOG_LINK that proves that A feeds B. This only works
1105 if there are no instructions between A and B which could have a link
1106 depending on A, since in that case we would not record a link for B.
1107 We also check the implicit dependency created by a cc0 setter/user
1108 pair. */
1110 static bool
1112 {
1120 }
1121 \f
1122 /* Main entry point for combiner. F is the first insn of the function.
1123 NREGS is the first unused pseudo-reg number.
1125 Return nonzero if the combiner has turned an indirect jump
1126 instruction into a direct jump. */
1127 static int
1129 {
1134 basic_block last_bb;
1154 /* Allocate array for insn info. */
1162 /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1163 problems when, for example, we have j <<= 1 in a loop. */
1168 /* Scan all SETs and see if we can deduce anything about what
1169 bits are known to be zero for some registers and how many copies
1170 of the sign bit are known to exist for those registers.
1172 Also set any known values so that we can use it while searching
1173 for what bits are known to be set. */
1176 /* Allow the entry block and the first block to fall into the same EBB.
1177 Conceptually the incoming promotions are assigned to the entry block. */
1182 {
1187 label_tick++;
1195 {
1196 rtx links;
1202 insn);
1209 insn);
1211 /* Record the current insn_rtx_cost of this instruction. */
1214 optimize_this_for_speed_p);
1216 {
1219 }
1220 }
1221 }
1225 /* Now scan all the insns in forward order. */
1233 {
1239 label_tick++;
1249 {
1263 /* See if we know about function return values before this
1264 insn based upon SUBREG flags. */
1267 /* See if we can find hardregs and subreg of pseudos in
1268 narrower modes. This could help turning TRUNCATEs
1269 into SUBREGs. */
1272 /* Try this insn with each insn it links back to. */
1278 {
1281 }
1283 /* Try each sequence of three linked insns ending with this one. */
1287 {
1290 /* If the linked insn has been replaced by a note, then there
1291 is no point in pursuing this chain any further. */
1299 {
1302 }
1303 }
1305 /* Try to combine a jump insn that uses CC0
1306 with a preceding insn that sets CC0, and maybe with its
1307 logical predecessor as well.
1308 This is how we make decrement-and-branch insns.
1309 We need this special code because data flow connections
1310 via CC0 do not get entered in LOG_LINKS. */
1317 {
1328 }
1330 /* Do the same for an insn that explicitly references CC0. */
1337 {
1348 }
1350 /* Finally, see if any of the insns that this insn links to
1351 explicitly references CC0. If so, try this insn, that insn,
1352 and its predecessor if it sets CC0. */
1354 {
1366 }
1368 /* Try combining an insn with two different insns whose results it
1369 uses. */
1379 {
1382 }
1384 /* Try four-instruction combinations. */
1387 {
1391 /* If the linked insn has been replaced by a note, then there
1392 is no point in pursuing this chain any further. */
1397 {
1401 /* I0 -> I1 -> I2 -> I3. */
1407 {
1410 }
1411 /* I0, I1 -> I2, I2 -> I3. */
1418 {
1421 }
1422 }
1425 {
1429 /* I0 -> I2; I1, I2 -> I3. */
1435 {
1438 }
1439 /* I0 -> I1; I1, I2 -> I3. */
1445 {
1448 }
1449 }
1450 }
1452 /* Try this insn with each REG_EQUAL note it links back to. */
1454 {
1460 /* Avoid using a register that may already been marked
1461 dead by an earlier instruction. */
1469 {
1470 /* Temporarily replace the set's source with the
1471 contents of the REG_EQUAL note. The insn will
1472 be deleted or recognized by try_combine. */
1483 last_combined_insn);
1486 {
1489 }
1492 }
1493 }
1498 retry:
1499 ;
1500 }
1501 }
1508 /* Clean up. */
1514 {
1517 {
1520 }
1522 }
1532 /* Make recognizer allow volatile MEMs again. */
1536 }
1538 /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1540 static void
1542 {
1548 }
1549 \f
1550 /* Set up any promoted values for incoming argument registers. */
1552 static void
1554 {
1555 tree arg;
1560 {
1565 /* Only continue if the incoming argument is in a register. */
1569 /* Determine, if possible, whether all call sites of the current
1570 function lie within the current compilation unit. (This does
1571 take into account the exporting of a function via taking its
1572 address, and so forth.) */
1575 /* The mode and signedness of the argument before any promotions happen
1576 (equal to the mode of the pseudo holding it at that stage). */
1580 /* The mode and signedness of the argument after any source language and
1581 TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1585 /* The mode and signedness of the argument as it is actually passed,
1586 see assign_parm_setup_reg in function.c. */
1590 /* The mode of the register in which the argument is being passed. */
1593 /* Eliminate sign extensions in the callee when:
1594 (a) A mode promotion has occurred; */
1597 /* (b) The mode of the register is the same as the mode of
1598 the argument as it is passed; */
1601 /* (c) There's no language level extension; */
1603 ;
1604 /* (c.1) All callers are from the current compilation unit. If that's
1605 the case we don't have to rely on an ABI, we only have to know
1606 what we're generating right now, and we know that we will do the
1607 mode1 to mode2 promotion with the given sign. */
1610 /* (c.2) The combination of the two promotions is useful. This is
1611 true when the signs match, or if the first promotion is unsigned.
1612 In the later case, (sign_extend (zero_extend x)) is the same as
1613 (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
1619 /* Record that the value was promoted from mode1 to mode3,
1620 so that any sign extension at the head of the current
1621 function may be eliminated. */
1625 }
1626 }
1628 /* If MODE has a precision lower than PREC and SRC is a non-negative constant
1629 that would appear negative in MODE, sign-extend SRC for use in nonzero_bits
1630 because some machines (maybe most) will actually do the sign-extension and
1631 this is the conservative approach.
1633 ??? For 2.5, try to tighten up the MD files in this regard instead of this
1634 kludge. */
1636 static rtx
1638 {
1646 }
1648 /* Update RSP for pseudo-register X from INSN's REG_EQUAL note (if one exists)
1649 and SET. */
1651 static void
1653 rtx x)
1654 {
1664 {
1668 }
1670 /* Don't call nonzero_bits if it cannot change anything. */
1672 {
1677 }
1679 /* Don't call num_sign_bit_copies if it cannot change anything. */
1681 {
1684 {
1688 }
1691 }
1692 }
1694 /* Called via note_stores. If X is a pseudo that is narrower than
1695 HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1697 If we are setting only a portion of X and we can't figure out what
1698 portion, assume all bits will be used since we don't know what will
1699 be happening.
1701 Similarly, set how many bits of X are known to be copies of the sign bit
1702 at all locations in the function. This is the smallest number implied
1703 by any set of X. */
1705 static void
1707 {
1712 /* If this register is undefined at the start of the file, we can't
1713 say what its contents were. */
1714 && ! REGNO_REG_SET_P
1717 {
1721 {
1725 }
1727 /* If this register is being initialized using itself, and the
1728 register is uninitialized in this basic block, and there are
1729 no LOG_LINKS which set the register, then part of the
1730 register is uninitialized. In that case we can't assume
1731 anything about the number of nonzero bits.
1733 ??? We could do better if we checked this in
1734 reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1735 could avoid making assumptions about the insn which initially
1736 sets the register, while still using the information in other
1737 insns. We would have to be careful to check every insn
1738 involved in the combination. */
1744 {
1751 {
1755 }
1756 }
1758 /* If this is a complex assignment, see if we can convert it into a
1759 simple assignment. */
1762 /* If this is a simple assignment, or we have a paradoxical SUBREG,
1763 set what we know about X. */
1769 else
1770 {
1773 }
1774 }
1775 }
1776 \f
1777 /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1778 optionally insns that were previously combined into I3 or that will be
1779 combined into the merger of INSN and I3. The order is PRED, PRED2,
1780 INSN, SUCC, SUCC2, I3.
1782 Return 0 if the combination is not allowed for any reason.
1784 If the combination is allowed, *PDEST will be set to the single
1785 destination of INSN and *PSRC to the single source, and this function
1786 will return 1. */
1788 static int
1792 {
1797 rtx link;
1802 {
1804 {
1809 }
1814 }
1818 /* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
1819 or a PARALLEL consisting of such a SET and CLOBBERs.
1821 If INSN has CLOBBER parallel parts, ignore them for our processing.
1822 By definition, these happen during the execution of the insn. When it
1823 is merged with another insn, all bets are off. If they are, in fact,
1824 needed and aren't also supplied in I3, they may be added by
1825 recog_for_combine. Otherwise, it won't match.
1827 We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1828 note.
1830 Get the source and destination of INSN. If more than one, can't
1831 combine. */
1837 {
1839 {
1843 {
1844 /* This is important to combine floating point insns
1845 for the SH4 port. */
1847 /* Combining an isolated USE doesn't make sense.
1848 We depend here on combinable_i3pat to reject them. */
1849 /* The code below this loop only verifies that the inputs of
1850 the SET in INSN do not change. We call reg_set_between_p
1851 to verify that the REG in the USE does not change between
1852 I3 and INSN.
1853 If the USE in INSN was for a pseudo register, the matching
1854 insn pattern will likely match any register; combining this
1855 with any other USE would only be safe if we knew that the
1856 used registers have identical values, or if there was
1857 something to tell them apart, e.g. different modes. For
1858 now, we forgo such complicated tests and simply disallow
1859 combining of USES of pseudo registers with any other USE. */
1862 {
1867 do
1868 {
1878 }
1880 }
1883 /* We can ignore CLOBBERs. */
1888 /* Ignore SETs whose result isn't used but not those that
1889 have side-effects. */
1895 /* If we have already found a SET, this is a second one and
1896 so we cannot combine with this insn. */
1904 /* Anything else means we can't combine. */
1906 }
1907 }
1910 /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1911 so don't do anything with it. */
1914 }
1915 else
1921 /* The simplification in expand_field_assignment may call back to
1922 get_last_value, so set safe guard here. */
1928 /* Do not eliminate user-specified register if it is in an
1929 asm input because we may break the register asm usage defined
1930 in GCC manual if allow to do so.
1931 Be aware that this may cover more cases than we expect but this
1932 should be harmless. */
1937 /* Don't eliminate a store in the stack pointer. */
1939 /* Don't combine with an insn that sets a register to itself if it has
1940 a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1942 /* Can't merge an ASM_OPERANDS. */
1944 /* Can't merge a function call. */
1946 /* Don't eliminate a function call argument. */
1952 /* Don't substitute into an incremented register. */
1956 /* Don't substitute into a non-local goto, this confuses CFG. */
1958 /* Make sure that DEST is not used after INSN but before SUCC, or
1959 after SUCC and before SUCC2, or after SUCC2 but before I3. */
1960 || (!all_adjacent
1961 && ((succ2
1965 || (succ
1966 /* SUCC and SUCC2 can be split halves from a PARALLEL; in
1967 that case SUCC is not in the insn stream, so use SUCC2
1968 instead for this test. */
1970 succ2
1973 /* Make sure that the value that is to be substituted for the register
1974 does not use any registers whose values alter in between. However,
1975 If the insns are adjacent, a use can't cross a set even though we
1976 think it might (this can happen for a sequence of insns each setting
1977 the same destination; last_set of that register might point to
1978 a NOTE). If INSN has a REG_EQUIV note, the register is always
1979 equivalent to the memory so the substitution is valid even if there
1980 are intervening stores. Also, don't move a volatile asm or
1981 UNSPEC_VOLATILE across any other insns. */
1982 || (! all_adjacent
1988 /* Don't combine across a CALL_INSN, because that would possibly
1989 change whether the life span of some REGs crosses calls or not,
1990 and it is a pain to update that information.
1991 Exception: if source is a constant, moving it later can't hurt.
1992 Accept that as a special case. */
1996 /* DEST must either be a REG or CC0. */
1998 {
1999 /* If register alignment is being enforced for multi-word items in all
2000 cases except for parameters, it is possible to have a register copy
2001 insn referencing a hard register that is not allowed to contain the
2002 mode being copied and which would not be valid as an operand of most
2003 insns. Eliminate this problem by not combining with such an insn.
2005 Also, on some machines we don't want to extend the life of a hard
2006 register. */
2011 /* Don't extend the life of a hard register unless it is
2012 user variable (if we have few registers) or it can't
2013 fit into the desired register (meaning something special
2014 is going on).
2015 Also avoid substituting a return register into I3, because
2016 reload can't handle a conflict with constraints of other
2017 inputs. */
2021 }
2029 {
2032 /* If the clobber represents an earlyclobber operand, we must not
2033 substitute an expression containing the clobbered register.
2034 As we do not analyze the constraint strings here, we have to
2035 make the conservative assumption. However, if the register is
2036 a fixed hard reg, the clobber cannot represent any operand;
2037 we leave it up to the machine description to either accept or
2038 reject use-and-clobber patterns. */
2044 }
2046 /* If INSN contains anything volatile, or is an `asm' (whether volatile
2047 or not), reject, unless nothing volatile comes between it and I3 */
2050 {
2051 /* Make sure neither succ nor succ2 contains a volatile reference. */
2056 /* We'll check insns between INSN and I3 below. */
2057 }
2059 /* If INSN is an asm, and DEST is a hard register, reject, since it has
2060 to be an explicit register variable, and was chosen for a reason. */
2066 /* If INSN contains volatile references (specifically volatile MEMs),
2067 we cannot combine across any other volatile references.
2068 Even if INSN doesn't contain volatile references, any intervening
2069 volatile insn might affect machine state. */
2072 ? volatile_refs_p
2079 /* If INSN contains an autoincrement or autodecrement, make sure that
2080 register is not used between there and I3, and not already used in
2081 I3 either. Neither must it be used in PRED or SUCC, if they exist.
2082 Also insist that I3 not be a jump; if it were one
2083 and the incremented register were spilled, we would lose. */
2101 /* Don't combine an insn that follows a CC0-setting insn.
2102 An insn that uses CC0 must not be separated from the one that sets it.
2103 We do, however, allow I2 to follow a CC0-setting insn if that insn
2104 is passed as I1; in that case it will be deleted also.
2105 We also allow combining in this case if all the insns are adjacent
2106 because that would leave the two CC0 insns adjacent as well.
2107 It would be more logical to test whether CC0 occurs inside I1 or I2,
2108 but that would be much slower, and this ought to be equivalent. */
2111 {
2116 }
2118 /* If we get here, we have passed all the tests and the combination is
2119 to be allowed. */
2125 }
2126 \f
2127 /* LOC is the location within I3 that contains its pattern or the component
2128 of a PARALLEL of the pattern. We validate that it is valid for combining.
2130 One problem is if I3 modifies its output, as opposed to replacing it
2131 entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2132 doing so would produce an insn that is not equivalent to the original insns.
2134 Consider:
2136 (set (reg:DI 101) (reg:DI 100))
2137 (set (subreg:SI (reg:DI 101) 0) <foo>)
2139 This is NOT equivalent to:
2141 (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2142 (set (reg:DI 101) (reg:DI 100))])
2144 Not only does this modify 100 (in which case it might still be valid
2145 if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2147 We can also run into a problem if I2 sets a register that I1
2148 uses and I1 gets directly substituted into I3 (not via I2). In that
2149 case, we would be getting the wrong value of I2DEST into I3, so we
2150 must reject the combination. This case occurs when I2 and I1 both
2151 feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2152 If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2153 of a SET must prevent combination from occurring. The same situation
2154 can occur for I0, in which case I0_NOT_IN_SRC is set.
2156 Before doing the above check, we first try to expand a field assignment
2157 into a set of logical operations.
2159 If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2160 we place a register that is both set and used within I3. If more than one
2161 such register is detected, we fail.
2163 Return 1 if the combination is valid, zero otherwise. */
2165 static int
2168 {
2172 {
2177 rtx subdest;
2184 /* Check for the case where I3 modifies its output, as discussed
2185 above. We don't want to prevent pseudos from being combined
2186 into the address of a MEM, so only prevent the combination if
2187 i1 or i2 set the same MEM. */
2197 /* This is the same test done in can_combine_p except we can't test
2198 all_adjacent; we don't have to, since this instruction will stay
2199 in place, thus we are not considering increasing the lifetime of
2200 INNER_DEST.
2202 Also, if this insn sets a function argument, combining it with
2203 something that might need a spill could clobber a previous
2204 function argument; the all_adjacent test in can_combine_p also
2205 checks this; here, we do a more specific test for this case. */
2215 /* If DEST is used in I3, it is being killed in this insn, so
2216 record that for later. We have to consider paradoxical
2217 subregs here, since they kill the whole register, but we
2218 ignore partial subregs, STRICT_LOW_PART, etc.
2219 Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2220 STACK_POINTER_REGNUM, since these are always considered to be
2221 live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2231 && (HARD_FRAME_POINTER_IS_FRAME_POINTER
2237 {
2242 }
2243 }
2246 {
2253 }
2256 }
2257 \f
2258 /* Return 1 if X is an arithmetic expression that contains a multiplication
2259 and division. We don't count multiplications by powers of two here. */
2261 static int
2263 {
2265 {
2281 }
2282 }
2283 \f
2284 /* Determine whether INSN can be used in a combination. Return nonzero if
2285 not. This is used in try_combine to detect early some cases where we
2286 can't perform combinations. */
2288 static int
2290 {
2291 rtx set;
2294 /* If this isn't really an insn, we can't do anything.
2295 This can occur when flow deletes an insn that it has merged into an
2296 auto-increment address. */
2300 /* Never combine loads and stores involving hard regs that are likely
2301 to be spilled. The register allocator can usually handle such
2302 reg-reg moves by tying. If we allow the combiner to make
2303 substitutions of likely-spilled regs, reload might die.
2304 As an exception, we allow combinations involving fixed regs; these are
2305 not available to the register allocator so there's no risk involved. */
2326 }
2328 struct likely_spilled_retval_info
2329 {
2332 };
2334 /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2335 hard registers that are known to be written to / clobbered in full. */
2336 static void
2338 {
2355 else
2358 }
2360 /* Return nonzero iff part of the return value is live during INSN, and
2361 it is likely spilled. This can happen when more than one insn is needed
2362 to copy the return value, e.g. when we consider to combine into the
2363 second copy insn for a complex value. */
2365 static int
2367 {
2369 rtx reg;
2372 /* We assume here that no machine mode needs more than
2373 32 hard registers when the value overlaps with a register
2374 for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2389 /* Disregard parts of the return value that are set later. */
2398 /* Check if any of the (probably) live return value registers is
2399 likely spilled. */
2400 nregs --;
2401 do
2402 {
2408 }
2410 /* Adjust INSN after we made a change to its destination.
2412 Changing the destination can invalidate notes that say something about
2413 the results of the insn and a LOG_LINK pointing to the insn. */
2415 static void
2417 {
2418 /* For notes, be conservative and simply remove them. */
2421 /* The new insn will have a destination that was previously the destination
2422 of an insn just above it. Call distribute_links to make a LOG_LINK from
2423 the next use of that destination. */
2439 }
2441 /* Return TRUE if combine can reuse reg X in mode MODE.
2442 ADDED_SETS is nonzero if the original set is still required. */
2443 static bool
2445 {
2452 /* Allow hard registers if the new mode is legal, and occupies no more
2453 registers than the old mode. */
2458 /* Or a pseudo that is only used once. */
2461 && !added_sets
2463 }
2466 /* Check whether X, the destination of a set, refers to part of
2467 the register specified by REG. */
2469 static bool
2471 {
2472 /* Check that reg is an integer mode register. */
2483 }
2485 /* Delete the unconditional jump INSN and adjust the CFG correspondingly.
2486 Note that the INSN should be deleted *after* removing dead edges, so
2487 that the kept edge is the fallthrough edge for a (set (pc) (pc))
2488 but not for a (set (pc) (label_ref FOO)). */
2490 static void
2492 {
2500 {
2505 /* Remove barriers from the footer if there are any. */
2508 {
2511 else
2515 }
2518 }
2519 }
2521 /* Return whether PAT is a PARALLEL of exactly N register SETs followed
2522 by an arbitrary number of CLOBBERs. */
2523 static bool
2525 {
2544 }
2546 /* Return whether INSN, a PARALLEL of N register SETs (and maybe some
2547 CLOBBERs), can be split into individual SETs in that order, without
2548 changing semantics. */
2549 static bool
2551 {
2559 {
2568 }
2571 }
2573 /* Try to combine the insns I0, I1 and I2 into I3.
2574 Here I0, I1 and I2 appear earlier than I3.
2575 I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2576 I3.
2578 If we are combining more than two insns and the resulting insn is not
2579 recognized, try splitting it into two insns. If that happens, I2 and I3
2580 are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2581 Otherwise, I0, I1 and I2 are pseudo-deleted.
2583 Return 0 if the combination does not work. Then nothing is changed.
2584 If we did the combination, return the insn at which combine should
2585 resume scanning.
2587 Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
2588 new direct jump instruction.
2590 LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2591 been I3 passed to an earlier try_combine within the same basic
2592 block. */
2597 {
2598 /* New patterns for I3 and I2, respectively. */
2602 /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2603 dead. */
2605 /* Total number of SETs to put into I3. */
2607 /* Nonzero if I2's or I1's body now appears in I3. */
2609 /* INSN_CODEs for new I3, new I2, and user of condition code. */
2611 /* Contains I3 if the destination of I3 is used in its source, which means
2612 that the old life of I3 is being killed. If that usage is placed into
2613 I2 and not in I3, a REG_DEAD note must be made. */
2615 /* SET_DEST and SET_SRC of I2, I1 and I0. */
2617 /* Copy of SET_SRC of I1 and I0, if needed. */
2619 /* Set if I2DEST was reused as a scratch register. */
2621 /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2623 /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2628 /* Notes that must be added to REG_NOTES in I3 and I2. */
2630 /* Notes that we substituted I3 into I2 instead of the normal case. */
2632 /* Notes that I1, I2 or I3 is a MULT operation. */
2639 rtx temp_expr;
2642 rtx new_other_notes;
2645 /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
2646 never be). */
2650 /* Only try four-insn combinations when there's high likelihood of
2651 success. Look for simple insns, such as loads of constants or
2652 binary operations involving a constant. */
2654 {
2664 {
2667 rtx src;
2672 {
2675 }
2677 ngood++;
2680 nshift++;
2681 }
2683 /* If I0 loads a memory and I3 sets the same memory, then I1 and I2
2684 are likely manipulating its value. Ideally we'll be able to combine
2685 all four insns into a bitfield insertion of some kind.
2687 Note the source in I0 might be inside a sign/zero extension and the
2688 memory modes in I0 and I3 might be different. So extract the address
2689 from the destination of I3 and search for it in the source of I0.
2691 In the event that there's a match but the source/dest do not actually
2692 refer to the same memory, the worst that happens is we try some
2693 combinations that we wouldn't have otherwise. */
2695 /* Ensure the source of SET0 is a MEM, possibly buried inside
2696 an extension. */
2702 /* Ensure the destination of SET3 is a MEM. */
2704 /* Would it be better to extract the base address for the MEM
2705 in SET3 and look for that? I don't have cases where it matters
2706 but I could envision such cases. */
2712 }
2714 /* Exit early if one of the insns involved can't be used for
2715 combinations. */
2726 combine_attempts++;
2729 /* Reset the hard register usage information. */
2733 {
2740 else
2743 }
2745 /* If multiple insns feed into one of I2 or I3, they can be in any
2746 order. To simplify the code below, reorder them in sequence. */
2756 /* First check for one important special case that the code below will
2757 not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2758 and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2759 we may be able to replace that destination with the destination of I3.
2760 This occurs in the common code where we compute both a quotient and
2761 remainder into a structure, in which case we want to do the computation
2762 directly into the structure to avoid register-register copies.
2764 Note that this case handles both multiple sets in I2 and also cases
2765 where I2 has a number of CLOBBERs inside the PARALLEL.
2767 We make very conservative checks below and only try to handle the
2768 most common cases of this. For example, we only handle the case
2769 where I2 and I3 are adjacent to avoid making difficult register
2770 usage tests. */
2778 /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2779 below would need to check what is inside (and reg_overlap_mentioned_p
2780 doesn't support those codes anyway). Don't allow those destinations;
2781 the resulting insn isn't likely to be recognized anyway. */
2787 {
2790 /* Make sure that the destination of I3,
2791 which we are going to substitute into one output of I2,
2792 is not used within another output of I2. We must avoid making this:
2793 (parallel [(set (mem (reg 69)) ...)
2794 (set (reg 69) ...)])
2795 which is not well-defined as to order of actions.
2796 (Besides, reload can't handle output reloads for this.)
2798 The problem can also happen if the dest of I3 is a memory ref,
2799 if another dest in I2 is an indirect memory ref.
2801 Neither can this PARALLEL be an asm. We do not allow combining
2802 that usually (see can_combine_p), so do not here either. */
2805 {
2814 }
2820 {
2821 combine_merges++;
2831 /* Replace the dest in I2 with our dest and make the resulting
2832 insn the new pattern for I3. Then skip to where we validate
2833 the pattern. Everything was set up above. */
2838 }
2839 }
2841 /* If I2 is setting a pseudo to a constant and I3 is setting some
2842 sub-part of it to another constant, merge them by making a new
2843 constant. */
2850 {
2856 {
2859 {
2865 }
2866 }
2867 else
2868 {
2873 }
2876 {
2877 /* If this is the low part, we're done. */
2879 ;
2880 /* Handle the case where inner is twice the size of outer. */
2884 /* Otherwise give up for now. */
2885 else
2887 }
2890 {
2894 wide_int o
2899 combine_merges++;
2906 /* Replace the source in I2 with the new constant and make the
2907 resulting insn the new pattern for I3. Then skip to where we
2908 validate the pattern. Everything was set up above. */
2914 /* The dest of I3 has been replaced with the dest of I2. */
2917 }
2918 }
2920 /* If we have no I1 and I2 looks like:
2921 (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
2922 (set Y OP)])
2923 make up a dummy I1 that is
2924 (set Y OP)
2925 and change I2 to be
2926 (set (reg:CC X) (compare:CC Y (const_int 0)))
2928 (We can ignore any trailing CLOBBERs.)
2930 This undoes a previous combination and allows us to match a branch-and-
2931 decrement insn. */
2943 {
2944 /* We make I1 with the same INSN_UID as I2. This gives it
2945 the same DF_INSN_LUID for value tracking. Our fake I1 will
2946 never appear in the insn stream so giving it the same INSN_UID
2947 as I2 will not cause a problem. */
2960 }
2962 /* If I2 is a PARALLEL of two SETs of REGs (and perhaps some CLOBBERs),
2963 make those two SETs separate I1 and I2 insns, and make an I0 that is
2964 the original I1. */
2970 {
2971 /* If there is no I1, there is no I0 either. */
2974 /* We make I1 with the same INSN_UID as I2. This gives it
2975 the same DF_INSN_LUID for value tracking. Our fake I1 will
2976 never appear in the insn stream so giving it the same INSN_UID
2977 as I2 will not cause a problem. */
2985 }
2987 /* Verify that I2 and I1 are valid for combining. */
2993 {
2996 }
2998 /* Record whether I2DEST is used in I2SRC and similarly for the other
2999 cases. Knowing this will help in register status updating below. */
3010 /* For the earlier insns, determine which of the subsequent ones they
3011 feed. */
3018 /* Ensure that I3's pattern can be the destination of combines. */
3024 {
3027 }
3029 /* See if any of the insns is a MULT operation. Unless one is, we will
3030 reject a combination that is, since it must be slower. Be conservative
3031 here. */
3039 /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
3040 We used to do this EXCEPT in one case: I3 has a post-inc in an
3041 output operand. However, that exception can give rise to insns like
3042 mov r3,(r3)+
3043 which is a famous insn on the PDP-11 where the value of r3 used as the
3044 source was model-dependent. Avoid this sort of thing. */
3046 #if 0
3052 /* It's not the exception. */
3053 #endif
3055 {
3056 rtx link;
3062 {
3065 }
3066 }
3068 /* See if the SETs in I1 or I2 need to be kept around in the merged
3069 instruction: whenever the value set there is still needed past I3.
3070 For the SET in I2, this is easy: we see if I2DEST dies or is set in I3.
3072 For the SET in I1, we have two cases: if I1 and I2 independently feed
3073 into I3, the set in I1 needs to be kept around unless I1DEST dies
3074 or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
3075 in I1 needs to be kept around unless I1DEST dies or is set in either
3076 I2 or I3. The same considerations apply to I0. */
3083 else
3091 else
3094 /* We are about to copy insns for the case where they need to be kept
3095 around. Check that they can be copied in the merged instruction. */
3101 {
3104 }
3106 /* If the set in I2 needs to be kept around, we must make a copy of
3107 PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3108 PATTERN (I2), we are only substituting for the original I1DEST, not into
3109 an already-substituted copy. This also prevents making self-referential
3110 rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3111 I2DEST. */
3114 {
3117 else
3119 }
3122 {
3125 else
3127 }
3130 {
3133 else
3135 }
3137 combine_merges++;
3139 /* Substitute in the latest insn for the regs set by the earlier ones. */
3145 /* Many machines that don't use CC0 have insns that can both perform an
3146 arithmetic operation and set the condition code. These operations will
3147 be represented as a PARALLEL with the first element of the vector
3148 being a COMPARE of an arithmetic operation with the constant zero.
3149 The second element of the vector will set some pseudo to the result
3150 of the same arithmetic operation. If we simplify the COMPARE, we won't
3151 match such a pattern and so will generate an extra insn. Here we test
3152 for this case, where both the comparison and the operation result are
3153 needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3154 I2SRC. Later we will make the PARALLEL that contains I2. */
3160 {
3161 rtx newpat_dest;
3175 {
3180 }
3182 /* Do the rest only if op1 is const0_rtx, which may be the
3183 result of simplification. */
3185 {
3186 /* If a single use of the CC is found, prepare to modify it
3187 when SELECT_CC_MODE returns a new CC-class mode, or when
3188 the above simplify_compare_const() returned a new comparison
3189 operator. undobuf.other_insn is assigned the CC use insn
3190 when modifying it. */
3192 {
3193 #ifdef SELECT_CC_MODE
3194 machine_mode new_mode
3199 {
3204 else
3205 {
3208 }
3209 }
3210 #endif
3211 /* Cases for modifying the CC-using comparison. */
3213 /* ??? Do we need to verify the zero rtx? */
3215 {
3216 /* Replace cc_use_loc with entire new RTX. */
3221 }
3223 {
3224 /* Just replace the CC reg with a new mode. */
3227 }
3228 }
3230 /* Now we modify the current newpat:
3231 First, SET_DEST(newpat) is updated if the CC mode has been
3232 altered. For targets without SELECT_CC_MODE, this should be
3233 optimized away. */
3236 /* This is always done to propagate i2src into newpat. */
3239 /* Create new version of i2pat if needed; the below PARALLEL
3240 creation needs this to work correctly. */
3244 }
3245 }
3248 {
3249 /* It is possible that the source of I2 or I1 may be performing
3250 an unneeded operation, such as a ZERO_EXTEND of something
3251 that is known to have the high part zero. Handle that case
3252 by letting subst look at the inner insns.
3254 Another way to do this would be to have a function that tries
3255 to simplify a single insn instead of merging two or more
3256 insns. We don't do this because of the potential of infinite
3257 loops and because of the potential extra memory required.
3258 However, doing it the way we are is a bit of a kludge and
3259 doesn't catch all cases.
3261 But only do this if -fexpensive-optimizations since it slows
3262 things down and doesn't usually win.
3264 This is not done in the COMPARE case above because the
3265 unmodified I2PAT is used in the PARALLEL and so a pattern
3266 with a modified I2SRC would not match. */
3269 {
3270 /* Pass pc_rtx so no substitutions are done, just
3271 simplifications. */
3273 {
3276 }
3280 }
3285 /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3286 copy of I2SRC each time we substitute it, in order to avoid creating
3287 self-referential RTL when we will be substituting I1SRC for I1DEST
3288 later. Likewise if I0 feeds into I2, either directly or indirectly
3289 through I1, and I0DEST is in I0SRC. */
3296 /* Record whether I2's body now appears within I3's body. */
3298 }
3300 /* If we already got a failure, don't try to do more. Otherwise, try to
3301 substitute I1 if we have it. */
3304 {
3305 /* Check that an autoincrement side-effect on I1 has not been lost.
3306 This happens if I1DEST is mentioned in I2 and dies there, and
3307 has disappeared from the new pattern. */
3309 && i1_feeds_i2_n
3312 /* Before we can do this substitution, we must redo the test done
3313 above (see detailed comments there) that ensures I1DEST isn't
3314 mentioned in any SETs in NEWPAT that are field assignments. */
3317 {
3320 }
3325 /* If the following substitution will modify I1SRC, make a copy of it
3326 for the case where it is substituted for I1DEST in I2PAT later. */
3330 /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3331 copy of I1SRC each time we substitute it, in order to avoid creating
3332 self-referential RTL when we will be substituting I0SRC for I0DEST
3333 later. */
3338 /* Record whether I1's body now appears within I3's body. */
3340 }
3342 /* Likewise for I0 if we have it. */
3345 {
3352 {
3355 }
3357 /* If the following substitution will modify I0SRC, make a copy of it
3358 for the case where it is substituted for I0DEST in I1PAT later. */
3361 /* And a copy for I0DEST in I2PAT substitution. */
3370 }
3372 /* Fail if an autoincrement side-effect has been duplicated. Be careful
3373 to count all the ways that I2SRC and I1SRC can be used. */
3384 /* Fail if we tried to make a new register. */
3386 /* Fail if we couldn't do something and have a CLOBBER. */
3388 /* Fail if this new pattern is a MULT and we didn't have one before
3389 at the outer level. */
3392 {
3395 }
3397 /* If the actions of the earlier insns must be kept
3398 in addition to substituting them into the latest one,
3399 we must make a new PARALLEL for the latest insn
3400 to hold additional the SETs. */
3403 {
3405 combine_extras++;
3408 {
3414 }
3415 else
3416 {
3421 }
3427 {
3433 }
3435 {
3444 }
3445 }
3447 validate_replacement:
3449 /* Note which hard regs this insn has as inputs. */
3452 /* If recog_for_combine fails, it strips existing clobbers. If we'll
3453 consider splitting this pattern, we might need these clobbers. */
3456 {
3462 }
3464 /* We have recognized nothing yet. */
3467 /* See if this is a PARALLEL of two SETs where one SET's destination is
3468 a register that is unused and this isn't marked as an instruction that
3469 might trap in an EH region. In that case, we just need the other SET.
3470 We prefer this over the PARALLEL.
3472 This can occur when simplifying a divmod insn. We *must* test for this
3473 case here because the code below that splits two independent SETs doesn't
3474 handle this case correctly when it updates the register status.
3476 It's pointless doing this if we originally had two sets, one from
3477 i3, and one from i2. Combining then splitting the parallel results
3478 in the original i2 again plus an invalid insn (which we delete).
3479 The net effect is only to move instructions around, which makes
3480 debug info less accurate. */
3485 {
3496 {
3499 }
3508 {
3514 }
3518 }
3520 /* Is the result of combination a valid instruction? */
3524 /* If we were combining three insns and the result is a simple SET
3525 with no ASM_OPERANDS that wasn't recognized, try to split it into two
3526 insns. There are two ways to do this. It can be split using a
3527 machine-specific method (like when you have an addition of a large
3528 constant) or by combine in the function find_split_point. */
3532 {
3536 /* See if the MD file can split NEWPAT. If it can't, see if letting it
3537 use I2DEST as a scratch register will help. In the latter case,
3538 convert I2DEST to the mode of the source of NEWPAT if we can. */
3542 /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3543 inputs of NEWPAT. */
3545 /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3546 possible to try that as a scratch reg. This would require adding
3547 more code to make it work though. */
3550 {
3553 /* ??? Reusing i2dest without resetting the reg_stat entry for it
3554 (temporarily, until we are committed to this instruction
3555 combination) does not work: for example, any call to nonzero_bits
3556 on the register (from a splitter in the MD file, for example)
3557 will get the old information, which is invalid.
3559 Since nowadays we can create registers during combine just fine,
3560 we should just create a new one here, not reuse i2dest. */
3562 /* First try to split using the original register as a
3563 scratch register. */
3567 i2dest)));
3570 /* If that didn't work, try changing the mode of I2DEST if
3571 we can. */
3576 {
3578 rtx ni2dest;
3582 else
3583 {
3586 }
3588 parallel = (gen_rtx_PARALLEL
3592 ni2dest))));
3597 {
3605 }
3606 }
3609 }
3611 /* If recog_for_combine has discarded clobbers, try to use them
3612 again for the split. */
3614 {
3617 }
3620 {
3625 }
3629 {
3639 /* If I2 or I3 has multiple SETs, we won't know how to track
3640 register status, so don't use these insns. If I2's destination
3641 is used between I2 and I3, we also can't use these insns. */
3651 /* It is possible that both insns now set the destination of I3.
3652 If so, we must show an extra use of it. */
3655 {
3674 }
3675 }
3677 /* If we can split it and use I2DEST, go ahead and see if that
3678 helps things be recognized. Verify that none of the registers
3679 are set between I2 and I3. */
3683 /* We need I2DEST in the proper mode. If it is a hard register
3684 or the only use of a pseudo, we can change its mode.
3685 Make sure we don't change a hard register to have a mode that
3686 isn't valid for it, or change the number of registers. */
3693 /* We can't overwrite I2DEST if its value is still used by
3694 NEWPAT. */
3696 {
3705 /* *SPLIT may be part of I2SRC, so make sure we have the
3706 original expression around for later debug processing.
3707 We should not need I2SRC any more in other cases. */
3710 else
3713 /* Get NEWDEST as a register in the proper mode. We have already
3714 validated that we can do this. */
3716 {
3719 else
3720 {
3723 }
3724 }
3726 /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3727 an ASHIFT. This can occur if it was inside a PLUS and hence
3728 appeared to be a memory address. This is a kludge. */
3733 {
3736 /* Update split_code because we may not have a multiply
3737 anymore. */
3739 }
3741 /* Similarly for (plus (mult FOO (const_int pow2))). */
3747 {
3751 /* Update split_code because we may not have a multiply
3752 anymore. */
3754 }
3756 #ifdef INSN_SCHEDULING
3757 /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3758 be written as a ZERO_EXTEND. */
3760 {
3761 /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3762 what it really is. */
3767 else
3770 }
3771 #endif
3773 /* Attempt to split binary operators using arithmetic identities. */
3777 {
3784 /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3786 {
3791 }
3792 /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3799 {
3805 /* Split both "((X op Y) op X) op Y" and
3806 "((X op Y) op Y) op X" as "T op T" where T is
3807 "X op Y". */
3810 {
3815 }
3816 /* Split "((X op X) op Y) op Y)" as "T op T" where
3817 T is "X op Y". */
3819 {
3825 }
3826 }
3827 }
3830 {
3833 }
3837 /* recog_for_combine might have added CLOBBERs to newi2pat.
3838 Make sure NEWPAT does not depend on the clobbered regs. */
3842 {
3845 {
3848 }
3849 }
3851 /* If the split point was a MULT and we didn't have one before,
3852 don't use one now. */
3855 }
3856 }
3858 /* Check for a case where we loaded from memory in a narrow mode and
3859 then sign extended it, but we need both registers. In that case,
3860 we have a PARALLEL with both loads from the same memory location.
3861 We can split this into a load from memory followed by a register-register
3862 copy. This saves at least one insn, more if register allocation can
3863 eliminate the copy.
3865 We cannot do this if the destination of the first assignment is a
3866 condition code register or cc0. We eliminate this case by making sure
3867 the SET_DEST and SET_SRC have the same mode.
3869 We cannot do this if the destination of the second assignment is
3870 a register that we have already assumed is zero-extended. Similarly
3871 for a SUBREG of such a register. */
3906 {
3907 rtx ni2dest;
3921 }
3923 /* Similarly, check for a case where we have a PARALLEL of two independent
3924 SETs but we started with three insns. In this case, we can do the sets
3925 as two separate insns. This case occurs when some SET allows two
3926 other insns to combine, but the destination of that SET is still live.
3928 Also do this if we started with two insns and (at least) one of the
3929 resulting sets is a noop; this noop will be deleted later. */
3948 {
3952 /* Normally, it doesn't matter which of the two is done first,
3953 but the one that references cc0 can't be the second, and
3954 one which uses any regs/memory set in between i2 and i3 can't
3955 be first. The PARALLEL might also have been pre-existing in i3,
3956 so we need to make sure that we won't wrongly hoist a SET to i2
3957 that would conflict with a death note present in there. */
3965 /* If I3 is a jump, ensure that set0 is a jump so that
3966 we do not create invalid RTL. */
3968 )
3969 {
3972 }
3980 /* If I3 is a jump, ensure that set1 is a jump so that
3981 we do not create invalid RTL. */
3983 )
3984 {
3987 }
3988 else
3989 {
3992 }
3997 {
3998 /* recog_for_combine might have added CLOBBERs to newi2pat.
3999 Make sure NEWPAT does not depend on the clobbered regs. */
4001 {
4004 {
4007 {
4010 }
4011 }
4012 }
4015 }
4016 }
4018 /* If it still isn't recognized, fail and change things back the way they
4019 were. */
4021 /* Is the result a reasonable ASM_OPERANDS? */
4023 {
4026 }
4028 /* If we had to change another insn, make sure it is valid also. */
4030 {
4038 {
4041 }
4042 }
4044 /* If I2 is the CC0 setter and I3 is the CC0 user then check whether
4045 they are adjacent to each other or not. */
4047 {
4051 {
4054 }
4055 }
4057 /* Only allow this combination if insn_rtx_costs reports that the
4058 replacement instructions are cheaper than the originals. */
4060 {
4063 }
4066 {
4071 {
4076 /* Temporarily revert mode back. */
4080 {
4081 /* If we used i2dest as a scratch register with a
4082 different mode, substitute it for the original
4083 i2src while its original mode is temporarily
4084 restored, and then clear i2scratch so that we don't
4085 do it again later. */
4087 this_basic_block);
4089 /* Put back the new mode. */
4091 }
4092 else
4093 {
4098 {
4101 }
4102 else
4103 {
4110 }
4112 /* We're dealing with a reg that changed mode but not
4113 meaning, so we want to turn it into a subreg for
4114 the new mode. However, because of REG sharing and
4115 because its mode had already changed, we have to do
4116 it in two steps. First, replace any debug uses of
4117 reg, with its original mode temporarily restored,
4118 with this copy we have created; then, replace the
4119 copy with the SUBREG of the original shared reg,
4120 once again changed to the new mode. */
4122 this_basic_block);
4126 this_basic_block);
4127 }
4128 }
4129 }
4131 /* If we will be able to accept this, we have made a
4132 change to the destination of I3. This requires us to
4133 do a few adjustments. */
4136 {
4139 }
4141 /* We now know that we can do this combination. Merge the insns and
4142 update the status of registers and LOG_LINKS. */
4145 {
4150 /* If any of the notes in OTHER_INSN were REG_DEAD or REG_UNUSED,
4151 ensure that they are still valid. Then add any non-duplicate
4152 notes added by recog_for_combine. */
4154 {
4163 /* Simply drop equal note since it may be no longer valid
4164 for other_insn. It may be possible to record that CC
4165 register is changed and only discard those notes, but
4166 in practice it's unnecessary complication and doesn't
4167 give any meaningful improvement.
4169 See PR78559. */
4173 }
4177 NULL_RTX);
4178 }
4181 {
4184 rtx ni2dest;
4186 /* I3 now uses what used to be its destination and which is now
4187 I2's destination. This requires us to do a few adjustments. */
4191 /* We need a LOG_LINK from I3 to I2. But we used to have one,
4192 so we still will.
4194 However, some later insn might be using I2's dest and have
4195 a LOG_LINK pointing at I3. We must remove this link.
4196 The simplest way to remove the link is to point it at I1,
4197 which we know will be a NOTE. */
4199 /* newi2pat is usually a SET here; however, recog_for_combine might
4200 have added some clobbers. */
4203 else
4210 {
4213 {
4219 }
4220 }
4221 }
4223 {
4228 /* Compute which registers we expect to eliminate. newi2pat may be setting
4229 either i3dest or i2dest, so we must check it. */
4232 || !i2dest_killed
4234 /* For i1, we need to compute both local elimination and global
4235 elimination information with respect to newi2pat because i1dest
4236 may be the same as i3dest, in which case newi2pat may be setting
4237 i1dest. Global information is used when distributing REG_DEAD
4238 note for i2 and i3, in which case it does matter if newi2pat sets
4239 i1dest or not.
4241 Local information is used when distributing REG_DEAD note for i1,
4242 in which case it doesn't matter if newi2pat sets i1dest or not.
4243 See PR62151, if we have four insns combination:
4244 i0: r0 <- i0src
4245 i1: r1 <- i1src (using r0)
4246 REG_DEAD (r0)
4247 i2: r0 <- i2src (using r1)
4248 i3: r3 <- i3src (using r0)
4249 ix: using r0
4250 From i1's point of view, r0 is eliminated, no matter if it is set
4251 by newi2pat or not. In other words, REG_DEAD info for r0 in i1
4252 should be discarded.
4254 Note local information only affects cases in forms like "I1->I2->I3",
4255 "I0->I1->I2->I3" or "I0&I1->I2, I2->I3". For other cases like
4256 "I0->I1, I1&I2->I3" or "I1&I2->I3", newi2pat won't set i1dest or
4257 i0dest anyway. */
4259 || !i1dest_killed
4264 /* Same case as i1. */
4271 /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4272 clear them. */
4280 /* Ensure that we do not have something that should not be shared but
4281 occurs multiple times in the new insns. Check this by first
4282 resetting all the `used' flags and then copying anything is shared. */
4306 {
4309 {
4311 {
4312 /* I2SRC must still be meaningful at this point. Some
4313 splitting operations can invalidate I2SRC, but those
4314 operations do not apply to calls. */
4318 }
4325 }
4326 }
4331 /* We had one special case above where I2 had more than one set and
4332 we replaced a destination of one of those sets with the destination
4333 of I3. In that case, we have to update LOG_LINKS of insns later
4334 in this basic block. Note that this (expensive) case is rare.
4336 Also, in this case, we must pretend that all REG_NOTEs for I2
4337 actually came from I3, so that REG_UNUSED notes from I2 will be
4338 properly handled. */
4341 {
4350 temp_insn
4360 {
4365 }
4366 else
4369 }
4377 {
4380 this_basic_block);
4383 }
4384 else
4385 {
4388 this_basic_block);
4390 }
4393 {
4398 this_basic_block);
4400 }
4403 {
4408 this_basic_block);
4410 }
4412 /* Get death notes for everything that is now used in either I3 or
4413 I2 and used to die in a previous insn. If we built two new
4414 patterns, move from I1 to I2 then I2 to I3 so that we get the
4415 proper movement on registers that I2 modifies. */
4421 else
4427 /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4444 /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4445 know these are REG_UNUSED and want them to go to the desired insn,
4446 so we always pass it as i3. */
4450 NULL_RTX);
4454 NULL_RTX);
4456 /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4457 put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4458 I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4459 in that case, it might delete I2. Similarly for I2 and I1.
4460 Show an additional death due to the REG_DEAD note we make here. If
4461 we discard it in distribute_notes, we will decrement it again. */
4464 {
4469 else
4472 }
4475 {
4480 else
4483 }
4486 {
4491 else
4494 }
4497 {
4502 else
4505 }
4513 {
4518 /* The insn that used to set this register doesn't exist, and
4519 this life of the register may not exist either. See if one of
4520 I3's links points to an insn that sets I2DEST. If it does,
4521 that is now the last known value for I2DEST. If we don't update
4522 this and I2 set the register to a value that depended on its old
4523 contents, we will get confused. If this insn is used, thing
4524 will be set correctly in combine_instructions. */
4532 /* If the reg formerly set in I2 died only once and that was in I3,
4533 zero its use count so it won't make `reload' do any work. */
4536 && ! i2dest_in_i2src
4539 }
4542 {
4555 && ! i1dest_in_i1src
4558 }
4561 {
4574 && ! i0dest_in_i0src
4577 }
4579 /* Update reg_stat[].nonzero_bits et al for any changes that may have
4580 been made to this insn. The order is important, because newi2pat
4581 can affect nonzero_bits of newpat. */
4585 }
4588 {
4590 {
4593 }
4595 }
4598 {
4600 {
4603 }
4605 }
4608 {
4610 {
4613 }
4615 }
4618 {
4620 {
4623 }
4625 }
4628 {
4630 {
4633 }
4635 }
4637 /* Set new_direct_jump_p if a new return or simple jump instruction
4638 has been created. Adjust the CFG accordingly. */
4640 {
4644 }
4649 {
4652 }
4656 {
4662 }
4667 {
4673 }
4675 /* A noop might also need cleaning up of CFG, if it comes from the
4676 simplification of a jump. */
4681 {
4684 }
4691 {
4694 }
4696 combine_successes++;
4703 else
4705 }
4706 \f
4707 /* Get a marker for undoing to the current state. */
4711 {
4713 }
4715 /* Undo the modifications up to the marker. */
4717 static void
4719 {
4723 {
4728 {
4743 }
4747 }
4750 }
4752 /* Undo all the modifications recorded in undobuf. */
4754 static void
4756 {
4758 }
4760 /* We've committed to accepting the changes we made. Move all
4761 of the undos to the free list. */
4763 static void
4765 {
4769 {
4773 }
4775 }
4776 \f
4777 /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4778 where we have an arithmetic expression and return that point. LOC will
4779 be inside INSN.
4781 try_combine will call this function to see if an insn can be split into
4782 two insns. */
4786 {
4794 scalar_int_mode inner_mode;
4796 /* First special-case some codes. */
4798 {
4800 #ifdef INSN_SCHEDULING
4801 /* If we are making a paradoxical SUBREG invalid, it becomes a split
4802 point. */
4805 #endif
4809 /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4810 using LO_SUM and HIGH. */
4813 {
4821 }
4823 /* If we have a PLUS whose second operand is a constant and the
4824 address is not valid, perhaps will can split it up using
4825 the machine-specific way to split large constants. We use
4826 the first pseudo-reg (one of the virtual regs) as a placeholder;
4827 it will not remain in the result. */
4832 {
4835 subst_insn);
4837 /* This should have produced two insns, each of which sets our
4838 placeholder. If the source of the second is a valid address,
4839 we can make put both sources together and make a split point
4840 in the middle. */
4853 && memory_address_addr_space_p
4856 {
4860 /* Replace the placeholder in SRC2 with SRC1. If we can
4861 find where in SRC2 it was placed, that can become our
4862 split point and we can replace this address with SRC2.
4863 Just try two obvious places. */
4874 {
4877 }
4878 }
4880 /* If that didn't work, perhaps the first operand is complex and
4881 needs to be computed separately, so make a split point there.
4882 This will occur on machines that just support REG + CONST
4883 and have a constant moved through some previous computation. */
4889 }
4891 /* If we have a PLUS whose first operand is complex, try computing it
4892 separately by making a split there. */
4903 /* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
4904 ZERO_EXTRACT, the most likely reason why this doesn't match is that
4905 we need to put the operand into a register. So split at that
4906 point. */
4916 /* See if we can split SET_SRC as it stands. */
4921 /* See if we can split SET_DEST as it stands. */
4926 /* See if this is a bitfield assignment with everything constant. If
4927 so, this is an IOR of an AND, so split it into that. */
4937 {
4943 unsigned HOST_WIDE_INT mask
4945 rtx or_mask;
4954 else
4955 {
4961 or_mask));
4962 }
4969 }
4971 /* Otherwise, see if this is an operation that we can split into two.
4972 If so, try to split that. */
4976 {
4978 /* If we are AND'ing with a large constant that is only a single
4979 bit and the result is only being used in a context where we
4980 need to know if it is zero or nonzero, replace it with a bit
4981 extraction. This will avoid the large constant, which might
4982 have taken more than one insn to make. If the constant were
4983 not a valid argument to the AND but took only one insn to make,
4984 this is no worse, but if it took more than one insn, it will
4985 be better. */
4995 {
5000 {
5003 }
5004 }
5008 /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
5009 is known to be on, this can be converted into a NEG of a shift. */
5015 {
5027 }
5033 /* We can't optimize if either mode is a partial integer
5034 mode as we don't know how many bits are significant
5035 in those modes. */
5049 {
5057 }
5062 }
5066 {
5069 /* For unsigned, we have a choice of a shift followed by an
5070 AND or two shifts. Use two shifts for field sizes where the
5071 constant might be too large. We assume here that we can
5072 always at least get 8-bit constants in an AND insn, which is
5073 true for every current RISC. */
5076 {
5077 unsigned HOST_WIDE_INT mask
5081 gen_rtx_LSHIFTRT
5089 }
5090 else
5091 {
5093 gen_rtx_fmt_ee
5104 }
5105 }
5107 /* See if this is a simple operation with a constant as the second
5108 operand. It might be that this constant is out of range and hence
5109 could be used as a split point. */
5117 /* Finally, see if this is a simple operation with its first operand
5118 not in a register. The operation might require this operand in a
5119 register, so return it as a split point. We can always do this
5120 because if the first operand were another operation, we would have
5121 already found it as a split point. */
5130 /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
5131 it is better to write this as (not (ior A B)) so we can split it.
5132 Similarly for IOR. */
5134 {
5142 }
5144 /* Many RISC machines have a large set of logical insns. If the
5145 second operand is a NOT, put it first so we will try to split the
5146 other operand first. */
5148 {
5152 }
5157 /* Canonicalization can produce (minus A (mult B C)), where C is a
5158 constant. It may be better to try splitting (plus (mult B -C) A)
5159 instead if this isn't a multiply by a power of two. */
5163 {
5171 mode)),
5174 }
5176 /* Split at a multiply-accumulate instruction. However if this is
5177 the SET_SRC, we likely do not have such an instruction and it's
5178 worthless to try this split. */
5187 }
5189 /* Otherwise, select our actions depending on our rtx class. */
5191 {
5197 /* fall through */
5205 /* fall through */
5207 /* Some machines have (and (shift ...) ...) insns. If X is not
5208 an AND, but XEXP (X, 0) is, use it as our split point. */
5218 /* Otherwise, we don't have a split point. */
5220 }
5221 }
5222 \f
5223 /* Throughout X, replace FROM with TO, and return the result.
5224 The result is TO if X is FROM;
5225 otherwise the result is X, but its contents may have been modified.
5226 If they were modified, a record was made in undobuf so that
5227 undo_all will (among other things) return X to its original state.
5229 If the number of changes necessary is too much to record to undo,
5230 the excess changes are not made, so the result is invalid.
5231 The changes already made can still be undone.
5232 undobuf.num_undo is incremented for such changes, so by testing that
5233 the caller can tell whether the result is valid.
5235 `n_occurrences' is incremented each time FROM is replaced.
5237 IN_DEST is nonzero if we are processing the SET_DEST of a SET.
5239 IN_COND is nonzero if we are at the top level of a condition.
5241 UNIQUE_COPY is nonzero if each substitution must be unique. We do this
5242 by copying if `n_occurrences' is nonzero. */
5244 static rtx
5246 {
5251 rtx new_rtx;
5253 /* Two expressions are equal if they are identical copies of a shared
5254 RTX or if they are both registers with the same register number
5255 and mode. */
5257 #define COMBINE_RTX_EQUAL_P(X,Y) \
5258 ((X) == (Y) \
5259 || (REG_P (X) && REG_P (Y) \
5260 && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5262 /* Do not substitute into clobbers of regs -- this will never result in
5263 valid RTL. */
5268 {
5269 n_occurrences++;
5271 }
5273 /* If X and FROM are the same register but different modes, they
5274 will not have been seen as equal above. However, the log links code
5275 will make a LOG_LINKS entry for that case. If we do nothing, we
5276 will try to rerecognize our original insn and, when it succeeds,
5277 we will delete the feeding insn, which is incorrect.
5279 So force this insn not to match in this (rare) case. */
5284 /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5285 of which may contain things that can be combined. */
5289 /* It is possible to have a subexpression appear twice in the insn.
5290 Suppose that FROM is a register that appears within TO.
5291 Then, after that subexpression has been scanned once by `subst',
5292 the second time it is scanned, TO may be found. If we were
5293 to scan TO here, we would find FROM within it and create a
5294 self-referent rtl structure which is completely wrong. */
5298 /* Parallel asm_operands need special attention because all of the
5299 inputs are shared across the arms. Furthermore, unsharing the
5300 rtl results in recognition failures. Failure to handle this case
5301 specially can result in circular rtl.
5303 Solve this by doing a normal pass across the first entry of the
5304 parallel, and only processing the SET_DESTs of the subsequent
5305 entries. Ug. */
5310 {
5313 /* If this substitution failed, this whole thing fails. */
5321 {
5327 {
5330 /* If this substitution failed, this whole thing fails. */
5336 }
5337 }
5338 }
5339 else
5340 {
5344 /* We don't need to process a SET_DEST that is a register, CC0,
5345 or PC, so set up to skip this common case. All other cases
5346 where we want to suppress replacing something inside a
5347 SET_SRC are handled via the IN_DEST operand. */
5354 /* Trying to simplify the operands of a widening MULT is not likely
5355 to create RTL matching a machine insn. */
5367 /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5368 constant. */
5373 {
5375 {
5378 {
5380 {
5383 n_occurrences++;
5384 }
5385 else
5386 {
5388 unique_copy);
5390 /* If this substitution failed, this whole thing
5391 fails. */
5395 }
5398 }
5399 }
5401 {
5402 /* If this is a register being set, ignore it. */
5409 ;
5412 {
5413 /* In general, don't install a subreg involving two
5414 modes not tieable. It can worsen register
5415 allocation, and can even make invalid reload
5416 insns, since the reg inside may need to be copied
5417 from in the outside mode, and that may be invalid
5418 if it is an fp reg copied in integer mode.
5420 We allow two exceptions to this: It is valid if
5421 it is inside another SUBREG and the mode of that
5422 SUBREG and the mode of the inside of TO is
5423 tieable and it is valid if X is a SET that copies
5424 FROM to CC0. */
5432 && (!HAVE_cc0
5447 n_occurrences++;
5448 }
5449 else
5450 /* If we are in a SET_DEST, suppress most cases unless we
5451 have gone inside a MEM, in which case we want to
5452 simplify the address. We assume here that things that
5453 are actually part of the destination have their inner
5454 parts in the first expression. This is true for SUBREG,
5455 STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5456 things aside from REG and MEM that should appear in a
5457 SET_DEST. */
5459 (((in_dest
5465 unique_copy);
5467 /* If we found that we will have to reject this combination,
5468 indicate that by returning the CLOBBER ourselves, rather than
5469 an expression containing it. This will speed things up as
5470 well as prevent accidents where two CLOBBERs are considered
5471 to be equal, thus producing an incorrect simplification. */
5477 {
5485 }
5488 {
5492 }
5493 else
5495 }
5496 }
5497 }
5499 /* Check if we are loading something from the constant pool via float
5500 extension; in this case we would undo compress_float_constant
5501 optimization and degenerate constant load to an immediate value. */
5505 {
5509 }
5511 /* Try to simplify X. If the simplification changed the code, it is likely
5512 that further simplification will help, so loop, but limit the number
5513 of repetitions that will be performed. */
5516 {
5517 /* If X is sufficiently simple, don't bother trying to do anything
5518 with it. */
5527 /* We no longer know the original mode of operand 0 since we
5528 have changed the form of X) */
5530 }
5533 }
5534 \f
5535 /* If X is a commutative operation whose operands are not in the canonical
5536 order, use substitutions to swap them. */
5538 static void
5540 {
5543 {
5547 }
5548 }
5550 /* Simplify X, a piece of RTL. We just operate on the expression at the
5551 outer level; call `subst' to simplify recursively. Return the new
5552 expression.
5554 OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
5555 if we are inside a SET_DEST. IN_COND is nonzero if we are at the top level
5556 of a condition. */
5558 static rtx
5561 {
5564 scalar_int_mode int_mode;
5565 rtx temp;
5568 /* If this is a commutative operation, put a constant last and a complex
5569 expression first. We don't need to do this for comparisons here. */
5572 /* Try to fold this expression in case we have constants that weren't
5573 present before. */
5576 {
5584 {
5587 {
5591 }
5594 }
5607 }
5610 {
5615 }
5617 /* If this is a simple operation applied to an IF_THEN_ELSE, try
5618 applying it to the arms of the IF_THEN_ELSE. This often simplifies
5619 things. Check for cases where both arms are testing the same
5620 condition.
5622 Don't do anything if all operands are very simple. */
5635 {
5640 /* If everything is a comparison, what we have is highly unlikely
5641 to be simpler, so don't use it. */
5644 {
5651 /* Simplify the alternative arms; this may collapse the true and
5652 false arms to store-flag values. Be careful to use copy_rtx
5653 here since true_rtx or false_rtx might share RTL with x as a
5654 result of the if_then_else_cond call above. */
5658 /* If true_rtx and false_rtx are not general_operands, an if_then_else
5659 is unlikely to be simpler. */
5662 {
5665 /* Restarting if we generate a store-flag expression will cause
5666 us to loop. Just drop through in this case. */
5668 /* If the result values are STORE_FLAG_VALUE and zero, we can
5669 just make the comparison operation. */
5680 /* Likewise, we can make the negate of a comparison operation
5681 if the result values are - STORE_FLAG_VALUE and zero. */
5689 mode);
5700 mode);
5701 else
5704 mode,
5705 VOIDmode,
5706 cond,
5707 cop1),
5712 }
5713 }
5714 }
5716 /* First see if we can apply the inverse distributive law. */
5719 {
5723 }
5725 /* If CODE is an associative operation not otherwise handled, see if we
5726 can associate some operands. This can win if they are constants or
5727 if they are logically related (i.e. (a & b) & a). */
5733 {
5735 {
5739 rtx inner;
5741 /* Make sure we pass the constant operand if any as the second
5742 one if this is a commutative operation. */
5750 /* For commutative operations, try the other pair if that one
5751 didn't simplify. */
5753 {
5758 }
5762 }
5763 }
5765 /* A little bit of algebraic simplification here. */
5767 {
5769 /* Ensure that our address has any ASHIFTs converted to MULT in case
5770 address-recognizing predicates are called later. */
5779 /* See if this can be moved to simplify_subreg. */
5782 /* Don't call gen_lowpart if the inner mode
5783 is VOIDmode and we cannot simplify it, as SUBREG without
5784 inner mode is invalid. */
5791 {
5792 rtx temp;
5798 /* If op is known to have all lower bits zero, the result is zero. */
5808 }
5810 /* Don't change the mode of the MEM if that would change the meaning
5811 of the address. */
5818 /* Note that we cannot do any narrowing for non-constants since
5819 we might have been counting on using the fact that some bits were
5820 zero. We now do this in the SET. */
5827 /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
5828 replaced by (lshiftrt X C). This will convert
5829 (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
5837 /* If X has only a single bit that might be nonzero, say, bit I, convert
5838 (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
5839 MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
5840 (sign_extract X 1 Y). But only do this if TEMP isn't a register
5841 or a SUBREG of one since we'd be making the expression more
5842 complex if it was just a register. */
5848 {
5849 rtx temp1 = simplify_shift_const
5855 /* If all we did was surround TEMP with the two shifts, we
5856 haven't improved anything, so don't use it. Otherwise,
5857 we are better off with TEMP1. */
5862 }
5866 /* We can't handle truncation to a partial integer mode here
5867 because we don't know the real bitsize of the partial
5868 integer mode. */
5877 /* We can truncate a constant value and return it. */
5881 /* Similarly to what we do in simplify-rtx.c, a truncate of a register
5882 whose value is a comparison can be replaced with a subreg if
5883 STORE_FLAG_VALUE permits. */
5892 /* (const (const X)) can become (const X). Do it this way rather than
5893 returning the inner CONST since CONST can be shared with a
5894 REG_EQUAL note. */
5900 /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
5901 can add in an offset. find_split_point will split this address up
5902 again if it doesn't match. */
5909 /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
5910 when c is (const_int (pow2 + 1) / 2) is a sign extension of a
5911 bit-field and can be replaced by either a sign_extend or a
5912 sign_extract. The `and' may be a zero_extend and the two
5913 <c>, -<c> constants may be reversed. */
5928 return simplify_shift_const
5935 /* If only the low-order bit of X is possibly nonzero, (plus x -1)
5936 can become (ashiftrt (ashift (xor x 1) C) C) where C is
5937 the bitsize of the mode - 1. This allows simplification of
5938 "a = (b & 8) == 0;" */
5950 /* If we are adding two things that have no bits in common, convert
5951 the addition into an IOR. This will often be further simplified,
5952 for example in cases like ((a & 1) + (a & 2)), which can
5953 become a & 3. */
5958 {
5959 /* Try to simplify the expression further. */
5963 /* If we could, great. If not, do not go ahead with the IOR
5964 replacement, since PLUS appears in many special purpose
5965 address arithmetic instructions. */
5973 }
5975 /* Canonicalize x + x into x << 1. */
5984 /* (minus <foo> (and <foo> (const_int -pow2))) becomes
5985 (and <foo> (const_int pow2-1)) */
5995 /* If we have (mult (plus A B) C), apply the distributive law and then
5996 the inverse distributive law to see if things simplify. This
5997 occurs mostly in addresses, often when unrolling loops. */
6000 {
6004 }
6006 /* Try simplify a*(b/c) as (a*b)/c. */
6009 {
6015 }
6019 /* If this is a divide by a power of two, treat it as a shift if
6020 its first operand is a shift. */
6038 /* If the first operand is a condition code, we can't do anything
6039 with it. */
6043 {
6051 /* Simplify our comparison, if possible. */
6054 /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
6055 if only the low-order bit is possibly nonzero in X (such as when
6056 X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
6057 (xor X 1) or (minus 1 X); we use the former. Finally, if X is
6058 known to be either 0 or -1, NE becomes a NEG and EQ becomes
6059 (plus X 1).
6061 Remove any ZERO_EXTRACT we made when thinking this was a
6062 comparison. It may now be simpler to use, e.g., an AND. If a
6063 ZERO_EXTRACT is indeed appropriate, it will be placed back by
6064 the call to make_compound_operation in the SET case.
6066 Don't apply these optimizations if the caller would
6067 prefer a comparison rather than a value.
6068 E.g., for the condition in an IF_THEN_ELSE most targets need
6069 an explicit comparison. */
6072 ;
6090 {
6094 int_mode);
6095 }
6103 {
6107 const1_rtx);
6108 }
6117 {
6120 }
6122 /* If STORE_FLAG_VALUE is -1, we have cases similar to
6123 those above. */
6125 ;
6142 {
6146 int_mode);
6147 }
6156 {
6160 int_mode);
6161 }
6163 /* If X is 0/1, (eq X 0) is X-1. */
6170 {
6173 }
6175 /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
6176 one bit that might be nonzero, we can convert (ne x 0) to
6177 (ashift x c) where C puts the bit in the sign bit. Remove any
6178 AND with STORE_FLAG_VALUE when we are done, since we are only
6179 going to test the sign bit. */
6187 {
6193 else
6195 }
6197 /* If the code changed, return a whole new comparison.
6198 We also need to avoid using SUBST in cases where
6199 simplify_comparison has widened a comparison with a CONST_INT,
6200 since in that case the wider CONST_INT may fail the sanity
6201 checks in do_SUBST. */
6208 /* Otherwise, keep this operation, but maybe change its operands.
6209 This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
6212 }
6222 /* If we are processing SET_DEST, we are done. */
6240 /* If this is a shift by a constant amount, simplify it. */
6248 (HOST_WIDE_INT_1U
6256 }
6259 }
6260 \f
6261 /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6263 static rtx
6265 {
6272 rtx temp;
6275 rtx reversed;
6276 scalar_int_mode int_mode;
6278 /* Simplify storing of the truth value. */
6283 /* Also when the truth value has to be reversed. */
6289 /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6290 in it is being compared against certain values. Get the true and false
6291 comparisons and see if that says anything about the value of each arm. */
6297 {
6298 HOST_WIDE_INT nzb;
6304 /* If FALSE_CODE is EQ, swap the codes and arms. */
6307 {
6310 }
6312 /* If we are comparing against zero and the expression being tested has
6313 only a single bit that might be nonzero, that is its value when it is
6314 not equal to zero. Similarly if it is known to be -1 or 0. */
6318 {
6321 }
6325 {
6328 }
6330 /* Now simplify an arm if we know the value of the register in the
6331 branch and it is used in the arm. Be careful due to the potential
6332 of locally-shared RTL. */
6349 }
6351 /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6352 reversed, do so to avoid needing two sets of patterns for
6353 subtract-and-branch insns. Similarly if we have a constant in the true
6354 arm, the false arm is the same as the first operand of the comparison, or
6355 the false arm is more complicated than the true arm. */
6368 {
6377 /* It is possible that the conditional has been simplified out. */
6380 }
6382 /* If the two arms are identical, we don't need the comparison. */
6387 /* Convert a == b ? b : a to "a". */
6399 /* Look for cases where we have (abs x) or (neg (abs X)). */
6402 && comparison_p
6409 {
6415 return
6418 mode);
6421 }
6423 /* Look for MIN or MAX. */
6426 && comparison_p
6431 {
6446 }
6448 /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6449 second operand is zero, this can be done as (OP Z (mult COND C2)) where
6450 C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6451 SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6452 We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6453 neither 1 or -1, but it isn't worth checking for. */
6456 && comparison_p
6459 {
6475 /* If an identity-zero op is commutative, check whether there
6476 would be a match if we swapped the operands. */
6496 {
6500 }
6512 {
6516 }
6532 {
6536 }
6548 {
6552 }
6555 {
6561 const_true_rtx));
6569 }
6570 }
6572 /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6573 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6574 negation of a single bit, we can convert this operation to a shift. We
6575 can actually do this more generally, but it doesn't seem worth it. */
6584 return
6588 /* (IF_THEN_ELSE (NE A 0) C1 0) is A or a zero-extend of A if the only
6589 non-zero bit in A is C1. */
6596 {
6603 }
6606 }
6607 \f
6608 /* Simplify X, a SET expression. Return the new expression. */
6610 static rtx
6612 {
6615 machine_mode mode
6619 scalar_int_mode int_mode;
6621 /* (set (pc) (return)) gets written as (return). */
6625 /* Now that we know for sure which bits of SRC we are using, see if we can
6626 simplify the expression for the object knowing that we only need the
6627 low-order bits. */
6630 {
6633 }
6635 /* If we are setting CC0 or if the source is a COMPARE, look for the use of
6636 the comparison result and try to simplify it unless we already have used
6637 undobuf.other_insn. */
6645 {
6654 {
6657 {
6660 }
6661 }
6662 else
6670 {
6674 }
6675 else
6676 {
6681 /* Attempt to simplify CC user. */
6683 {
6687 }
6689 /* Convert X into a no-op move. */
6693 }
6695 /* Simplify our comparison, if possible. */
6698 #ifdef SELECT_CC_MODE
6699 /* If this machine has CC modes other than CCmode, check to see if we
6700 need to use a different CC mode here. */
6709 else
6712 /* If the mode changed, we have to change SET_DEST, the mode in the
6713 compare, and the mode in the place SET_DEST is used. If SET_DEST is
6714 a hard register, just build new versions with the proper mode. If it
6715 is a pseudo, we lose unless it is only time we set the pseudo, in
6716 which case we can safely change its mode. */
6718 {
6720 {
6722 rtx new_dest;
6726 else
6727 {
6730 }
6737 }
6738 }
6741 /* If the code changed, we have to build a new comparison in
6742 undobuf.other_insn. */
6744 {
6753 /* If the only change we made was to change an EQ into an NE or
6754 vice versa, OP0 has only one bit that might be nonzero, and OP1
6755 is zero, check if changing the user of the condition code will
6756 produce a valid insn. If it won't, we can keep the original code
6757 in that insn by surrounding our operation with an XOR. */
6764 {
6769 {
6776 }
6777 }
6778 }
6783 /* Don't generate a compare of a CC with 0, just use that CC. */
6785 {
6788 }
6789 /* Otherwise, if we didn't previously have the same COMPARE we
6790 want, create it from scratch. */
6793 {
6796 }
6797 }
6798 else
6799 {
6800 /* Get SET_SRC in a form where we have placed back any
6801 compound expressions. Then do the checks below. */
6804 }
6806 /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
6807 and X being a REG or (subreg (reg)), we may be able to convert this to
6808 (set (subreg:m2 x) (op)).
6810 We can always do this if M1 is narrower than M2 because that means that
6811 we only care about the low bits of the result.
6813 However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
6814 perform a narrower operation than requested since the high-order bits will
6815 be undefined. On machine where it is defined, this transformation is safe
6816 as long as M1 and M2 have the same number of words. */
6825 #ifdef CANNOT_CHANGE_MODE_CLASS
6830 #endif
6834 {
6837 dest));
6841 }
6843 /* If we have (set (cc0) (subreg ...)), we try to remove the subreg
6844 in SRC. */
6850 {
6854 /* Here we make sure that we don't have a sign bit on. */
6857 {
6860 }
6861 }
6863 /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
6864 would require a paradoxical subreg. Replace the subreg with a
6865 zero_extend to avoid the reload that would otherwise be required. */
6871 {
6876 }
6878 /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
6879 are comparing an item known to be 0 or -1 against 0, use a logical
6880 operation instead. Check for one of the arms being an IOR of the other
6881 arm with some value. We compute three terms to be IOR'ed together. In
6882 practice, at most two will be nonzero. Then we do the IOR's. */
6890 && (!HAVE_conditional_move
6895 {
6920 int_mode),
6921 false_rtx);
6927 term3));
6930 }
6932 /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
6933 whole thing fail. */
6938 else
6939 /* Convert this into a field assignment operation, if possible. */
6941 }
6942 \f
6943 /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
6944 result. */
6946 static rtx
6948 {
6954 {
6956 /* We can call simplify_and_const_int only if we don't lose
6957 any (sign) bits when converting INTVAL (op1) to
6958 "unsigned HOST_WIDE_INT". */
6962 {
6969 }
6971 /* If we have any of (and (ior A B) C) or (and (xor A B) C),
6972 apply the distributive law and then the inverse distributive
6973 law to see if things simplify. */
6975 {
6979 }
6981 {
6985 }
6989 /* If we have (ior (and A B) C), apply the distributive law and then
6990 the inverse distributive law to see if things simplify. */
6993 {
6997 }
7000 {
7004 }
7009 }
7012 }
7013 \f
7014 /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
7015 operations" because they can be replaced with two more basic operations.
7016 ZERO_EXTEND is also considered "compound" because it can be replaced with
7017 an AND operation, which is simpler, though only one operation.
7019 The function expand_compound_operation is called with an rtx expression
7020 and will convert it to the appropriate shifts and AND operations,
7021 simplifying at each stage.
7023 The function make_compound_operation is called to convert an expression
7024 consisting of shifts and ANDs into the equivalent compound expression.
7025 It is the inverse of this function, loosely speaking. */
7027 static rtx
7029 {
7033 rtx tem;
7036 {
7039 /* FALLTHRU */
7041 /* We can't necessarily use a const_int for a multiword mode;
7042 it depends on implicitly extending the value.
7043 Since we don't know the right way to extend it,
7044 we can't tell whether the implicit way is right.
7046 Even for a mode that is no wider than a const_int,
7047 we can't win, because we need to sign extend one of its bits through
7048 the rest of it, and we don't know which bit. */
7052 /* Return if (subreg:MODE FROM 0) is not a safe replacement for
7053 (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
7054 because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
7055 reloaded. If not for that, MEM's would very rarely be safe.
7057 Reject MODEs bigger than a word, because we might not be able
7058 to reference a two-register group starting with an arbitrary register
7059 (and currently gen_lowpart might crash for a SUBREG). */
7064 /* Reject MODEs that aren't scalar integers because turning vector
7065 or complex modes into shifts causes problems. */
7071 /* If the inner object has VOIDmode (the only way this can happen
7072 is if it is an ASM_OPERANDS), we can't do anything since we don't
7073 know how much masking to do. */
7082 /* fall through */
7085 /* If the operand is a CLOBBER, just return it. */
7094 /* Reject MODEs that aren't scalar integers because turning vector
7095 or complex modes into shifts causes problems. */
7103 /* This should stay within the object being extracted, fail otherwise. */
7114 }
7115 /* Convert sign extension to zero extension, if we know that the high
7116 bit is not set, as this is easier to optimize. It will be converted
7117 back to cheaper alternative in make_extraction. */
7125 {
7130 /* Make sure this is a profitable operation. */
7137 else
7139 }
7141 /* We can optimize some special cases of ZERO_EXTEND. */
7143 {
7144 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
7145 know that the last value didn't have any inappropriate bits
7146 set. */
7154 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7163 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
7164 is a comparison and STORE_FLAG_VALUE permits. This is like
7165 the first case, but it works even when GET_MODE (x) is larger
7166 than HOST_WIDE_INT. */
7175 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7185 }
7187 /* If we reach here, we want to return a pair of shifts. The inner
7188 shift is a left shift of BITSIZE - POS - LEN bits. The outer
7189 shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
7190 logical depending on the value of UNSIGNEDP.
7192 If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
7193 converted into an AND of a shift.
7195 We must check for the case where the left shift would have a negative
7196 count. This can happen in a case like (x >> 31) & 255 on machines
7197 that can't shift by a constant. On those machines, we would first
7198 combine the shift with the AND to produce a variable-position
7199 extraction. Then the constant of 31 would be substituted in
7200 to produce such a position. */
7204 {
7213 }
7220 else
7221 /* Any other cases we can't handle. */
7224 /* If we couldn't do this for some reason, return the original
7225 expression. */
7230 }
7231 \f
7232 /* X is a SET which contains an assignment of one object into
7233 a part of another (such as a bit-field assignment, STRICT_LOW_PART,
7234 or certain SUBREGS). If possible, convert it into a series of
7235 logical operations.
7237 We half-heartedly support variable positions, but do not at all
7238 support variable lengths. */
7240 static const_rtx
7242 {
7243 rtx inner;
7247 machine_mode compute_mode;
7249 /* Loop until we find something we can't simplify. */
7251 {
7254 {
7258 }
7261 {
7266 /* A constant position should stay within the width of INNER. */
7272 {
7280 /* If position is ADJUST - X, new position is X. */
7282 else
7283 {
7288 pos);
7289 }
7290 }
7291 }
7293 /* A SUBREG between two modes that occupy the same numbers of words
7294 can be done by moving the SUBREG to the source. */
7296 /* We need SUBREGs to compute nonzero_bits properly. */
7297 && nonzero_sign_valid
7302 {
7304 gen_lowpart
7308 }
7309 else
7317 /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7319 {
7320 /* Don't do anything for vector or complex integral types. */
7324 /* Try to find an integral mode to pun with. */
7330 }
7332 /* Compute a mask of LEN bits, if we can do this on the host machine. */
7336 /* Don't try to compute in too wide unsupported modes. */
7340 /* Now compute the equivalent expression. Make a copy of INNER
7341 for the SET_DEST in case it is a MEM into which we will substitute;
7342 we don't want shared RTL in that case. */
7344 compute_mode);
7348 compute_mode,
7350 compute_mode),
7351 inner);
7356 mask),
7357 pos);
7362 }
7365 }
7366 \f
7367 /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7368 it is an RTX that represents the (variable) starting position; otherwise,
7369 POS is the (constant) starting bit position. Both are counted from the LSB.
7371 UNSIGNEDP is nonzero for an unsigned reference and zero for a signed one.
7373 IN_DEST is nonzero if this is a reference in the destination of a SET.
7374 This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7375 a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7376 be used.
7378 IN_COMPARE is nonzero if we are in a COMPARE. This means that a
7379 ZERO_EXTRACT should be built even for bits starting at bit 0.
7381 MODE is the desired mode of the result (if IN_DEST == 0).
7383 The result is an RTX for the extraction or NULL_RTX if the target
7384 can't handle it. */
7386 static rtx
7390 {
7391 /* This mode describes the size of the storage area
7392 to fetch the overall value from. Within that, we
7393 ignore the POS lowest bits, etc. */
7395 machine_mode inner_mode;
7396 machine_mode wanted_inner_mode;
7402 HOST_WIDE_INT orig_pos;
7408 {
7409 /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7410 consider just the QI as the memory to extract from.
7411 The subreg adds or removes high bits; its mode is
7412 irrelevant to the meaning of this extraction,
7413 since POS and LEN count from the lsb. */
7417 }
7422 {
7423 /* We're extracting the least significant bits of an rtx
7424 (ashift X (const_int C)), where LEN > C. Extract the
7425 least significant (LEN - C) bits of X, giving an rtx
7426 whose mode is MODE, then shift it left C times. */
7432 }
7438 /* See if this can be done without an extraction. We never can if the
7439 width of the field is not the same as that of some integer mode. For
7440 registers, we can only avoid the extraction if the position is at the
7441 low-order bit and this is either not in the destination or we have the
7442 appropriate STRICT_LOW_PART operation available.
7444 For MEM, we can avoid an extract if the field starts on an appropriate
7445 boundary and we can change the mode of the memory reference. */
7447 scalar_int_mode tmode;
7456 && (! in_dest
7460 && (pos
7463 /* We can't do this if we are widening INNER_MODE (it
7464 may not be aligned, for one thing). */
7470 {
7471 /* If INNER is a MEM, make a new MEM that encompasses just the desired
7472 field. If the original and current mode are the same, we need not
7473 adjust the offset. Otherwise, we do if bytes big endian.
7475 If INNER is not a MEM, get a piece consisting of just the field
7476 of interest (in this case POS % BITS_PER_WORD must be 0). */
7479 {
7480 HOST_WIDE_INT offset;
7482 /* POS counts from lsb, but make OFFSET count in memory order. */
7485 else
7489 }
7491 {
7493 {
7494 /* We can't call gen_lowpart in a DEST since we
7495 always want a SUBREG (see below) and it would sometimes
7496 return a new hard register. */
7498 {
7499 unsigned int offset
7502 /* Avoid creating invalid subregs, for example when
7503 simplifying (x>>32)&255. */
7508 }
7509 else
7511 }
7512 else
7514 }
7515 else
7517 len >= HOST_BITS_PER_WIDE_INT
7518 ? HOST_WIDE_INT_M1U
7521 /* If this extraction is going into the destination of a SET,
7522 make a STRICT_LOW_PART unless we made a MEM. */
7537 /* If we know that no extraneous bits are set, and that the high
7538 bit is not set, convert the extraction to the cheaper of
7539 sign and zero extension, that are equivalent in these cases. */
7545 {
7549 /* Prefer ZERO_EXTENSION, since it gives more information to
7550 backends. */
7555 }
7557 /* Otherwise, sign- or zero-extend unless we already are in the
7558 proper mode. */
7562 }
7564 /* Unless this is a COMPARE or we have a funny memory reference,
7565 don't do anything with zero-extending field extracts starting at
7566 the low-order bit since they are simple AND operations. */
7571 /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7572 if the position is not a constant and the length is not 1. In all
7573 other cases, we would only be going outside our object in cases when
7574 an original shift would have been undefined. */
7583 /* If INNER is not from memory, we want it to have the mode of a register
7584 extraction pattern's structure operand, or word_mode if there is no
7585 such pattern. The same applies to extraction_mode and pos_mode
7586 and their respective operands.
7588 For memory, assume that the desired extraction_mode and pos_mode
7589 are the same as for a register operation, since at present we don't
7590 have named patterns for aligned memory structures. */
7594 {
7598 }
7600 /* Never narrow an object, since that might not be safe. */
7608 else
7609 {
7610 /* Be careful not to go beyond the extracted object and maintain the
7611 natural alignment of the memory. */
7616 }
7621 {
7622 /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7623 BITS_BIG_ENDIAN style. If position is constant, compute new
7624 position. Otherwise, build subtraction.
7625 Note that POS is relative to the mode of the original argument.
7626 If it's a MEM we need to recompute POS relative to that.
7627 However, if we're extracting from (or inserting into) a register,
7628 we want to recompute POS relative to wanted_inner_mode. */
7635 else
7636 pos_rtx
7639 pos_rtx);
7640 /* POS may be less than 0 now, but we check for that below.
7641 Note that it can only be less than 0 if !MEM_P (inner). */
7642 }
7644 /* If INNER has a wider mode, and this is a constant extraction, try to
7645 make it smaller and adjust the byte to point to the byte containing
7646 the value. */
7649 && ! pos_rtx
7654 {
7657 /* The computations below will be correct if the machine is big
7658 endian in both bits and bytes or little endian in bits and bytes.
7659 If it is mixed, we must adjust. */
7661 /* If bytes are big endian and we had a paradoxical SUBREG, we must
7662 adjust OFFSET to compensate. */
7667 /* We can now move to the desired byte. */
7678 }
7680 /* If INNER is not memory, get it into the proper mode. If we are changing
7681 its mode, POS must be a constant and smaller than the size of the new
7682 mode. */
7684 {
7685 /* On the LHS, don't create paradoxical subregs implicitely truncating
7686 the register unless TRULY_NOOP_TRUNCATION. */
7689 wanted_inner_mode))
7701 pos_rtx
7703 ? HOST_WIDE_INT_M1U
7707 }
7709 /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7710 have to zero extend. Otherwise, we can just use a SUBREG. */
7713 {
7717 /* If we know that no extraneous bits are set, and that the high
7718 bit is not set, convert extraction to cheaper one - either
7719 SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7720 cases. */
7728 {
7732 /* Prefer ZERO_EXTENSION, since it gives more information to
7733 backends. */
7737 }
7739 }
7741 /* Make POS_RTX unless we already have it and it is correct. If we don't
7742 have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7743 be a CONST_INT. */
7750 /* Make the required operation. See if we can use existing rtx. */
7757 }
7758 \f
7759 /* See if X contains an ASHIFT of COUNT or more bits that can be commuted
7760 with any other operations in X. Return X without that shift if so. */
7762 static rtx
7764 {
7767 rtx tem;
7770 {
7772 /* This is the shift itself. If it is wide enough, we will return
7773 either the value being shifted if the shift count is equal to
7774 COUNT or a shift for the difference. */
7788 /* If we can safely shift this constant and we find the inner shift,
7789 make a new operation. */
7794 {
7798 }
7803 }
7806 }
7807 \f
7808 /* Subroutine of make_compound_operation. *X_PTR is the rtx at the current
7809 level of the expression and MODE is its mode. IN_CODE is as for
7810 make_compound_operation. *NEXT_CODE_PTR is the value of IN_CODE
7811 that should be used when recursing on operands of *X_PTR.
7813 There are two possible actions:
7815 - Return null. This tells the caller to recurse on *X_PTR with IN_CODE
7816 equal to *NEXT_CODE_PTR, after which *X_PTR holds the final value.
7818 - Return a new rtx, which the caller returns directly. */
7820 static rtx
7824 {
7832 rtx tem;
7836 {
7839 }
7841 /* Process depending on the code of this operation. If NEW is set
7842 nonzero, it will be returned. */
7845 {
7847 /* Convert shifts by constants into multiplications if inside
7848 an address. */
7852 {
7858 {
7861 }
7864 }
7873 {
7877 }
7880 {
7884 mode));
7886 }
7887 else
7888 {
7891 }
7901 {
7905 }
7908 {
7912 mode));
7914 }
7915 else
7916 {
7920 }
7923 /* If the second operand is not a constant, we can't do anything
7924 with it. */
7928 /* If the constant is a power of two minus one and the first operand
7929 is a logical right shift, make an extraction. */
7932 {
7936 }
7938 /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
7943 {
7951 /* If we narrowed the mode when dropping the subreg, then we lose. */
7955 /* If that didn't give anything, see if the AND simplifies on
7956 its own. */
7958 {
7962 }
7963 }
7964 /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
7970 {
7971 /* Apply the distributive law, and then try to make extractions. */
7978 }
7980 /* If we are have (and (rotate X C) M) and C is larger than the number
7981 of bits in M, this is an extraction. */
7987 {
7993 }
7995 /* On machines without logical shifts, if the operand of the AND is
7996 a logical shift and our mask turns off all the propagated sign
7997 bits, we can replace the logical shift with an arithmetic shift. */
8005 {
8012 make_compound_operation
8015 }
8017 /* If the constant is one less than a power of two, this might be
8018 representable by an extraction even if no shift is present.
8019 If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
8020 we are in a COMPARE. */
8024 next_code),
8027 /* If we are in a comparison and this is an AND with a power of two,
8028 convert this into the appropriate bit extract. */
8034 next_code),
8037 /* If the one operand is a paradoxical subreg of a register or memory and
8038 the constant (limited to the smaller mode) has only zero bits where
8039 the sub expression has known zero bits, this can be expressed as
8040 a zero_extend. */
8042 {
8043 rtx sub;
8049 {
8053 /* original AND constant with all the known zero bits set */
8056 {
8061 }
8062 }
8063 }
8068 /* If the sign bit is known to be zero, replace this with an
8069 arithmetic shift. */
8074 {
8077 next_code),
8080 }
8082 /* fall through */
8088 /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
8089 this is a SIGN_EXTRACT. */
8096 {
8103 }
8105 /* See if we have operations between an ASHIFTRT and an ASHIFT.
8106 If so, try to merge the shifts into a SIGN_EXTEND. We could
8107 also do this for some cases of SIGN_EXTRACT, but it doesn't
8108 seem worth the effort; the case checked for occurs on Alpha. */
8125 /* Call ourselves recursively on the inner expression. If we are
8126 narrowing the object and it has a different RTL code from
8127 what it originally did, do this SUBREG as a force_to_mode. */
8128 {
8132 /* If the SUBREG is masking of a logical right shift,
8133 make an extraction. */
8140 {
8149 }
8151 /* If in_code is COMPARE, it isn't always safe to pass it through
8152 to the recursive make_compound_operation call. */
8156 /* (subreg:SI (and:DI (reg:DI) (const_int 0x800000000)) 0)
8157 is (const_int 0), rather than
8158 (subreg:SI (lshiftrt:DI (reg:DI) (const_int 35)) 0).
8159 Similarly (subreg:QI (and:SI (reg:SI) (const_int 0x80)) 0)
8160 for non-equality comparisons against 0 is not equivalent
8161 to (subreg:QI (lshiftrt:SI (reg:SI) (const_int 7)) 0). */
8171 simplified
8179 {
8180 rtx newer
8183 /* If we have something other than a SUBREG, we might have
8184 done an expansion, so rerun ourselves. */
8188 /* force_to_mode can expand compounds. If it just re-expanded the
8189 compound, use gen_lowpart to convert to the desired mode. */
8191 /* Likewise if it re-expanded the compound only partially.
8192 This happens for SUBREG of ZERO_EXTRACT if they extract
8193 the same number of bits. */
8202 }
8206 }
8211 }
8217 }
8219 /* Look at the expression rooted at X. Look for expressions
8220 equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
8221 Form these expressions.
8223 Return the new rtx, usually just X.
8225 Also, for machines like the VAX that don't have logical shift insns,
8226 try to convert logical to arithmetic shift operations in cases where
8227 they are equivalent. This undoes the canonicalizations to logical
8228 shifts done elsewhere.
8230 We try, as much as possible, to re-use rtl expressions to save memory.
8232 IN_CODE says what kind of expression we are processing. Normally, it is
8233 SET. In a memory address it is MEM. When processing the arguments of
8234 a comparison or a COMPARE against zero, it is COMPARE, or EQ if more
8235 precisely it is an equality comparison against zero. */
8237 rtx
8239 {
8246 /* Select the code to be used in recursive calls. Once we are inside an
8247 address, we stay there. If we have a comparison, set to COMPARE,
8248 but once inside, go back to our default of SET. */
8256 {
8262 }
8264 /* Now recursively process each operand of this operation. We need to
8265 handle ZERO_EXTEND specially so that we don't lose track of the
8266 inner mode. */
8268 {
8276 }
8281 {
8284 }
8287 {
8290 }
8294 }
8295 \f
8296 /* Given M see if it is a value that would select a field of bits
8297 within an item, but not the entire word. Return -1 if not.
8298 Otherwise, return the starting position of the field, where 0 is the
8299 low-order bit.
8301 *PLEN is set to the length of the field. */
8303 static int
8305 {
8306 /* Get the bit number of the first 1 bit from the right, -1 if none. */
8311 /* Now shift off the low-order zero bits and see if we have a
8312 power of two minus 1. */
8320 }
8321 \f
8322 /* If X refers to a register that equals REG in value, replace these
8323 references with REG. */
8324 static rtx
8326 {
8334 {
8367 /* FALLTHRU */
8371 {
8375 else
8377 }
8379 /* fall through */
8386 {
8389 {
8391 {
8394 }
8396 }
8397 }
8399 {
8402 {
8405 {
8407 {
8410 }
8412 }
8413 }
8414 }
8417 }
8420 }
8422 /* Return X converted to MODE. If the value is already truncated to
8423 MODE we can just return a subreg even though in the general case we
8424 would need an explicit truncation. */
8426 static rtx
8428 {
8433 {
8434 /* Bit-cast X into an integer mode. */
8439 }
8442 }
8444 /* See if X can be simplified knowing that we will only refer to it in
8445 MODE and will only refer to those bits that are nonzero in MASK.
8446 If other bits are being computed or if masking operations are done
8447 that select a superset of the bits in MASK, they can sometimes be
8448 ignored.
8450 Return a possibly simplified expression, but always convert X to
8451 MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8453 If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
8454 are all off in X. This is used when X will be complemented, by either
8455 NOT, NEG, or XOR. */
8457 static rtx
8460 {
8463 machine_mode op_mode;
8467 /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8468 code below will do the wrong thing since the mode of such an
8469 expression is VOIDmode.
8471 Also do nothing if X is a CLOBBER; this can happen if X was
8472 the return value from a call to gen_lowpart. */
8476 /* We want to perform the operation in its present mode unless we know
8477 that the operation is valid in MODE, in which case we do the operation
8478 in MODE. */
8483 /* It is not valid to do a right-shift in a narrower mode
8484 than the one it came in with. */
8489 /* Truncate MASK to fit OP_MODE. */
8493 /* When we have an arithmetic operation, or a shift whose count we
8494 do not know, we need to assume that all bits up to the highest-order
8495 bit in MASK will be needed. This is how we form such a mask. */
8498 else
8502 /* Determine what bits of X are guaranteed to be (non)zero. */
8505 /* If none of the bits in X are needed, return a zero. */
8509 /* If X is a CONST_INT, return a new one. Do this here since the
8510 test below will fail. */
8512 {
8515 else
8516 {
8519 }
8520 }
8522 /* If X is narrower than MODE and we want all the bits in X's mode, just
8523 get X in the proper mode. */
8528 /* We can ignore the effect of a SUBREG if it narrows the mode or
8529 if the constant masks to zero all the bits the mode doesn't have. */
8539 /* The arithmetic simplifications here only work for scalar integer modes. */
8544 {
8546 /* If X is a (clobber (const_int)), return it since we know we are
8547 generating something that won't match. */
8560 /* Similarly for a truncate. */
8564 /* If this is an AND with a constant, convert it into an AND
8565 whose constant is the AND of that constant with MASK. If it
8566 remains an AND of MASK, delete it since it is redundant. */
8569 {
8573 /* If X is still an AND, see if it is an AND with a mask that
8574 is just some low-order bits. If so, and it is MASK, we don't
8575 need it. */
8582 /* If it remains an AND, try making another AND with the bits
8583 in the mode mask that aren't in MASK turned on. If the
8584 constant in the AND is wide enough, this might make a
8585 cheaper constant. */
8590 {
8591 unsigned HOST_WIDE_INT cval
8594 rtx y;
8601 }
8604 }
8609 /* In (and (plus FOO C1) M), if M is a mask that just turns off
8610 low-order bits (as in an alignment operation) and FOO is already
8611 aligned to that boundary, mask C1 to that boundary as well.
8612 This may eliminate that PLUS and, later, the AND. */
8614 {
8618 /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8619 number, sign extend it. */
8632 }
8634 /* fall through */
8637 /* Substituting into the operands of a widening MULT is not likely to
8638 create RTL matching a machine insn. */
8648 /* For PLUS, MINUS and MULT, we need any bits less significant than the
8649 most significant bit in MASK since carries from those bits will
8650 affect the bits we are interested in. */
8655 /* If X is (minus C Y) where C's least set bit is larger than any bit
8656 in the mask, then we may replace with (neg Y). */
8659 {
8663 }
8665 /* Similarly, if C contains every bit in the fuller_mask, then we may
8666 replace with (not Y). */
8669 {
8673 }
8680 /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8681 LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8682 operation which may be a bitfield extraction. Ensure that the
8683 constant we form is not wider than the mode of X. */
8695 {
8704 }
8706 binop:
8707 /* For most binary operations, just propagate into the operation and
8708 change the mode if we have an operation of that mode. */
8713 /* If we ended up truncating both operands, truncate the result of the
8714 operation instead. */
8717 {
8720 }
8730 /* For left shifts, do the same, but just for the first operand.
8731 However, we cannot do anything with shifts where we cannot
8732 guarantee that the counts are smaller than the size of the mode
8733 because such a count will have a different meaning in a
8734 wider mode. */
8744 /* If the shift count is a constant and we can do arithmetic in
8745 the mode of the shift, refine which bits we need. Otherwise, use the
8746 conservative form of the mask. */
8752 else
8764 /* Here we can only do something if the shift count is a constant,
8765 this shift constant is valid for the host, and we can do arithmetic
8766 in OP_MODE. */
8772 {
8776 /* Select the mask of the bits we need for the shift operand. */
8779 /* We can only change the mode of the shift if we can do arithmetic
8780 in the mode of the shift and INNER_MASK is no wider than the
8781 width of X's mode. */
8789 }
8791 /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
8792 shift and AND produces only copies of the sign bit (C2 is one less
8793 than a power of two), we can do this with just a shift. */
8797 /* The shift puts one of the sign bit copies in the least significant
8798 bit. */
8803 /* Number of bits left after the shift must be more than the mask
8804 needs. */
8807 /* Must be more sign bit copies than the mask needs. */
8817 /* If we are just looking for the sign bit, we don't need this shift at
8818 all, even if it has a variable count. */
8822 /* If this is a shift by a constant, get a mask that contains those bits
8823 that are not copies of the sign bit. We then have two cases: If
8824 MASK only includes those bits, this can be a logical shift, which may
8825 allow simplifications. If MASK is a single-bit field not within
8826 those bits, we are requesting a copy of the sign bit and hence can
8827 shift the sign bit to the appropriate location. */
8831 {
8834 /* If the considered data is wider than HOST_WIDE_INT, we can't
8835 represent a mask for all its bits in a single scalar.
8836 But we only care about the lower bits, so calculate these. */
8839 {
8842 /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
8843 is the number of bits a full-width mask would have set.
8844 We need only shift if these are fewer than nonzero can
8845 hold. If not, we must keep all bits set in nonzero. */
8850 + HOST_BITS_PER_WIDE_INT
8852 }
8853 else
8854 {
8857 }
8860 {
8865 }
8868 {
8869 x = simplify_shift_const
8875 }
8876 }
8878 /* If MASK is 1, convert this to an LSHIFTRT. This can be done
8879 even if the shift count isn't a constant. */
8884 shiftrt:
8886 /* If this is a zero- or sign-extension operation that just affects bits
8887 we don't care about, remove it. Be sure the call above returned
8888 something that is still a shift. */
8898 next_select);
8904 /* If the shift count is constant and we can do computations
8905 in the mode of X, compute where the bits we care about are.
8906 Otherwise, we can't do anything. Don't change the mode of
8907 the shift or propagate MODE into the shift, though. */
8910 {
8920 }
8924 /* If we just want the low-order bit, the NEG isn't needed since it
8925 won't change the low-order bit. */
8929 /* We need any bits less significant than the most significant bit in
8930 MASK since carries from those bits will affect the bits we are
8931 interested in. */
8936 /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
8937 same as the XOR case above. Ensure that the constant we form is not
8938 wider than the mode of X. */
8946 {
8955 }
8957 /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
8958 use the full mask inside the NOT. */
8961 unop:
8964 next_select));
8970 /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
8971 in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
8972 which is equal to STORE_FLAG_VALUE. */
8984 /* We have no way of knowing if the IF_THEN_ELSE can itself be
8985 written in a narrower mode. We play it safe and do not do so. */
9001 }
9003 /* Ensure we return a value of the proper mode. */
9005 }
9006 \f
9007 /* Return nonzero if X is an expression that has one of two values depending on
9008 whether some other value is zero or nonzero. In that case, we return the
9009 value that is being tested, *PTRUE is set to the value if the rtx being
9010 returned has a nonzero value, and *PFALSE is set to the other alternative.
9012 If we return zero, we set *PTRUE and *PFALSE to X. */
9014 static rtx
9016 {
9022 /* If we are comparing a value against zero, we are done. */
9025 {
9029 }
9031 /* If this is a unary operation whose operand has one of two values, apply
9032 our opcode to compute those values. */
9035 {
9040 }
9042 /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
9043 make can't possibly match and would suppress other optimizations. */
9045 ;
9047 /* If this is a binary operation, see if either side has only one of two
9048 values. If either one does or if both do and they are conditional on
9049 the same value, compute the new true and false values. */
9051 {
9059 {
9060 /* Try to enable a simplification by undoing work done by
9061 if_then_else_cond if it converted a REG into something more
9062 complex. */
9064 {
9067 }
9068 else
9069 {
9072 }
9073 }
9077 {
9078 /* If if_then_else_cond returned zero, then true/false are the
9079 same rtl. We must copy one of them to prevent invalid rtl
9080 sharing. */
9087 {
9092 }
9093 else
9094 {
9097 }
9100 }
9102 /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
9103 operands is zero when the other is nonzero, and vice-versa,
9104 and STORE_FLAG_VALUE is 1 or -1. */
9110 {
9127 {
9134 const_true_rtx);
9136 }
9137 }
9139 /* Similarly for MULT, AND and UMIN, except that for these the result
9140 is always zero. */
9144 {
9158 {
9161 }
9162 }
9163 }
9166 {
9167 /* If we have IF_THEN_ELSE already, extract the condition and
9168 canonicalize it if it is NE or EQ. */
9174 {
9177 }
9178 else
9180 }
9182 /* If X is a SUBREG, we can narrow both the true and false values
9183 if the inner expression, if there is a condition. */
9187 {
9193 {
9197 }
9198 }
9200 /* If X is a constant, this isn't special and will cause confusions
9201 if we treat it as such. Likewise if it is equivalent to a constant. */
9204 ;
9206 /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
9207 will be least confusing to the rest of the compiler. */
9209 {
9212 }
9214 /* If X is known to be either 0 or -1, those are the true and
9215 false values when testing X. */
9219 {
9222 }
9224 /* Likewise for 0 or a single bit. */
9227 {
9230 }
9232 /* Otherwise fail; show no condition with true and false values the same. */
9235 }
9236 \f
9237 /* Return the value of expression X given the fact that condition COND
9238 is known to be true when applied to REG as its first operand and VAL
9239 as its second. X is known to not be shared and so can be modified in
9240 place.
9242 We only handle the simplest cases, and specifically those cases that
9243 arise with IF_THEN_ELSE expressions. */
9245 static rtx
9247 {
9255 /* If either operand of the condition is a floating point value,
9256 then we have to avoid collapsing an EQ comparison. */
9266 /* If X is (abs REG) and we know something about REG's relationship
9267 with zero, we may be able to simplify this. */
9271 {
9280 }
9282 /* The only other cases we handle are MIN, MAX, and comparisons if the
9283 operands are the same as REG and VAL. */
9286 {
9288 {
9291 }
9294 {
9296 {
9304 else
9306 }
9309 {
9312 /* Do not reverse the condition when it is NE or EQ.
9313 This is because we cannot conclude anything about
9314 the value of 'SMAX (x, y)' when x is not equal to y,
9315 but we can when x equals y. */
9321 {
9332 }
9333 }
9334 }
9335 }
9337 {
9342 {
9343 /* We must simplify subreg here, before we lose track of the
9344 original inner_mode. */
9349 else
9351 }
9354 }
9355 /* We don't have to handle SIGN_EXTEND here, because even in the
9356 case of replacing something with a modeless CONST_INT, a
9357 CONST_INT is already (supposed to be) a valid sign extension for
9358 its narrower mode, which implies it's already properly
9359 sign-extended for the wider mode. Now, for ZERO_EXTEND, the
9360 story is different. */
9362 {
9367 {
9368 /* We must simplify the zero_extend here, before we lose
9369 track of the original inner_mode. */
9374 else
9376 }
9379 }
9383 {
9390 }
9393 }
9394 \f
9395 /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9396 assignment as a field assignment. */
9398 static int
9400 {
9402 {
9407 /* For big endian, adjust the memory offset. */
9412 else
9414 }
9422 /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9423 Note that all SUBREGs of MEM are paradoxical; otherwise they
9424 would have been rewritten. */
9437 /* We used to see if get_last_value of X and Y were the same but that's
9438 not correct. In one direction, we'll cause the assignment to have
9439 the wrong destination and in the case, we'll import a register into this
9440 insn that might have already have been dead. So fail if none of the
9441 above cases are true. */
9443 }
9444 \f
9445 /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9446 Return that assignment if so.
9448 We only handle the most common cases. */
9450 static rtx
9452 {
9455 rtx assign;
9457 HOST_WIDE_INT c1;
9458 HOST_WIDE_INT pos;
9460 rtx other;
9461 machine_mode mode;
9463 /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9464 a clear of a one-bit field. We will have changed it to
9465 (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9466 for a SUBREG. */
9472 {
9478 }
9488 {
9495 }
9497 /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9498 one-bit field. */
9502 {
9508 }
9510 /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9511 SRC is an AND with all bits of that field set, then we can discard
9512 the AND. */
9517 {
9524 else
9527 /* Complete overlap. We can remove the source AND. */
9531 /* Partial overlap. We can reduce the source AND. */
9533 {
9538 }
9539 }
9541 /* The other case we handle is assignments into a constant-position
9542 field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9543 a mask that has all one bits except for a group of zero bits and
9544 OTHER is known to have zeros where C1 has ones, this is such an
9545 assignment. Compute the position and length from C1. Shift OTHER
9546 to the appropriate position, force it to the required mode, and
9547 make the extraction. Check for the AND in both operands. */
9549 /* One or more SUBREGs might obscure the constant-position field
9550 assignment. The first one we are likely to encounter is an outer
9551 narrowing SUBREG, which we can just strip for the purposes of
9552 identifying the constant-field assignment. */
9566 /* The second SUBREG that might get in the way is a paradoxical
9567 SUBREG around the first operand of the AND. We want to
9568 pretend the operand is as wide as the destination here. We
9569 do this by adjusting the MEM to wider mode for the sole
9570 purpose of the call to rtx_equal_for_field_assignment_p. Also
9571 note this trick only works for MEMs. */
9583 /* The second SUBREG that might get in the way is a paradoxical
9584 SUBREG around the first operand of the AND. We want to
9585 pretend the operand is as wide as the destination here. We
9586 do this by adjusting the MEM to wider mode for the sole
9587 purpose of the call to rtx_equal_for_field_assignment_p. Also
9588 note this trick only works for MEMs. */
9596 else
9609 /* The mode to use for the source is the mode of the assignment, or of
9610 what is inside a possible STRICT_LOW_PART. */
9614 /* Shift OTHER right POS places and make it the source, restricting it
9615 to the proper length and mode. */
9620 dest);
9622 len >= HOST_BITS_PER_WIDE_INT
9623 ? HOST_WIDE_INT_M1U
9627 /* If SRC is masked by an AND that does not make a difference in
9628 the value being stored, strip it. */
9639 }
9640 \f
9641 /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9642 if so. */
9644 static rtx
9646 {
9650 rtx tem;
9652 /* Distributivity is not true for floating point as it can change the
9653 value. So we don't do it unless -funsafe-math-optimizations. */
9658 /* The outer operation can only be one of the following: */
9666 /* If either operand is a primitive we can't do anything, so get out
9667 fast. */
9677 /* See if the inner and outer operations distribute. */
9679 {
9684 /* These all distribute except over PLUS. */
9695 /* This is also a multiply, so it distributes over everything. */
9698 /* This used to handle SUBREG, but this turned out to be counter-
9699 productive, since (subreg (op ...)) usually is not handled by
9700 insn patterns, and this "optimization" therefore transformed
9701 recognizable patterns into unrecognizable ones. Therefore the
9702 SUBREG case was removed from here.
9704 It is possible that distributing SUBREG over arithmetic operations
9705 leads to an intermediate result than can then be optimized further,
9706 e.g. by moving the outer SUBREG to the other side of a SET as done
9707 in simplify_set. This seems to have been the original intent of
9708 handling SUBREGs here.
9710 However, with current GCC this does not appear to actually happen,
9711 at least on major platforms. If some case is found where removing
9712 the SUBREG case here prevents follow-on optimizations, distributing
9713 SUBREGs ought to be re-added at that place, e.g. in simplify_set. */
9717 }
9719 /* Set LHS and RHS to the inner operands (A and B in the example
9720 above) and set OTHER to the common operand (C in the example).
9721 There is only one way to do this unless the inner operation is
9722 commutative. */
9734 else
9737 /* Form the new inner operation, seeing if it simplifies first. */
9740 /* There is one exception to the general way of distributing:
9741 (a | c) ^ (b | c) -> (a ^ b) & ~c */
9743 {
9746 }
9748 /* We may be able to continuing distributing the result, so call
9749 ourselves recursively on the inner operation before forming the
9750 outer operation, which we return. */
9753 }
9755 /* See if X is of the form (* (+ A B) C), and if so convert to
9756 (+ (* A C) (* B C)) and try to simplify.
9758 Most of the time, this results in no change. However, if some of
9759 the operands are the same or inverses of each other, simplifications
9760 will result.
9762 For example, (and (ior A B) (not B)) can occur as the result of
9763 expanding a bit field assignment. When we apply the distributive
9764 law to this, we get (ior (and (A (not B))) (and (B (not B)))),
9765 which then simplifies to (and (A (not B))).
9767 Note that no checks happen on the validity of applying the inverse
9768 distributive law. This is pointless since we can do it in the
9769 few places where this routine is called.
9771 N is the index of the term that is decomposed (the arithmetic operation,
9772 i.e. (+ A B) in the first example above). !N is the index of the term that
9773 is distributed, i.e. of C in the first example above. */
9774 static rtx
9776 {
9777 machine_mode mode;
9781 /* Distributivity is not true for floating point as it can change the
9782 value. So we don't do it unless -funsafe-math-optimizations. */
9799 /* Special case (and (xor B C) (not A)), which is equivalent to
9800 (xor (ior A B) (ior A C)) */
9802 {
9805 }
9808 {
9809 /* Distribute the second term. */
9812 }
9813 else
9814 {
9815 /* Distribute the first term. */
9818 }
9828 }
9829 \f
9830 /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
9831 in MODE. Return an equivalent form, if different from (and VAROP
9832 (const_int CONSTOP)). Otherwise, return NULL_RTX. */
9834 static rtx
9837 {
9840 rtx orig_varop;
9848 /* Simplify VAROP knowing that we will be only looking at some of the
9849 bits in it.
9851 Note by passing in CONSTOP, we guarantee that the bits not set in
9852 CONSTOP are not significant and will never be examined. We must
9853 ensure that is the case by explicitly masking out those bits
9854 before returning. */
9857 /* If VAROP is a CLOBBER, we will fail so return it. */
9861 /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
9862 to VAROP and return the new constant. */
9866 /* See what bits may be nonzero in VAROP. Unlike the general case of
9867 a call to nonzero_bits, here we don't care about bits outside
9868 MODE. */
9872 /* Turn off all bits in the constant that are known to already be zero.
9873 Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
9874 which is tested below. */
9878 /* If we don't have any bits left, return zero. */
9882 /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
9883 a power of two, we can replace this with an ASHIFT. */
9888 /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
9889 or XOR, then try to apply the distributive law. This may eliminate
9890 operations if either branch can be simplified because of the AND.
9891 It may also make some cases more complex, but those cases probably
9892 won't match a pattern either with or without this. */
9895 return
9896 gen_lowpart
9898 apply_distributive_law
9903 constop),
9907 constop))));
9909 /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
9910 the AND and see if one of the operands simplifies to zero. If so, we
9911 may eliminate it. */
9915 {
9924 }
9926 /* Make a SUBREG if necessary. If we can't make it, fail. */
9931 /* If we are only masking insignificant bits, return VAROP. */
9938 /* Otherwise, return an AND. */
9940 }
9943 /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
9944 in MODE.
9946 Return an equivalent form, if different from X. Otherwise, return X. If
9947 X is zero, we are to always construct the equivalent form. */
9949 static rtx
9952 {
9963 }
9964 \f
9965 /* Given a REG, X, compute which bits in X can be nonzero.
9966 We don't care about bits outside of those defined in MODE.
9968 For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
9969 a shift, AND, or zero_extract, we can do better. */
9971 static rtx
9973 const_rtx known_x ATTRIBUTE_UNUSED,
9974 machine_mode known_mode ATTRIBUTE_UNUSED,
9977 {
9978 rtx tem;
9981 /* If X is a register whose nonzero bits value is current, use it.
9982 Otherwise, if X is a register whose value we can find, use that
9983 value. Otherwise, use the previously-computed global nonzero bits
9984 for this register. */
9998 && !REGNO_REG_SET_P
10001 {
10002 /* Note that, even if the precision of last_set_mode is lower than that
10003 of mode, record_value_for_reg invoked nonzero_bits on the register
10004 with nonzero_bits_mode (because last_set_mode is necessarily integral
10005 and HWI_COMPUTABLE_MODE_P in this case) so bits in nonzero_bits_mode
10006 are all valid, hence in mode too since nonzero_bits_mode is defined
10007 to the largest HWI_COMPUTABLE_MODE_P mode. */
10010 }
10014 {
10020 }
10023 {
10027 /* We don't know anything about the upper bits. */
10031 }
10034 }
10036 /* Return the number of bits at the high-order end of X that are known to
10037 be equal to the sign bit. X will be used in mode MODE; if MODE is
10038 VOIDmode, X will be used in its own mode. The returned value will always
10039 be between 1 and the number of bits in MODE. */
10041 static rtx
10043 const_rtx known_x ATTRIBUTE_UNUSED,
10044 machine_mode known_mode
10045 ATTRIBUTE_UNUSED,
10048 {
10049 rtx tem;
10062 && !REGNO_REG_SET_P
10065 {
10068 }
10079 }
10080 \f
10081 /* Return the number of "extended" bits there are in X, when interpreted
10082 as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
10083 unsigned quantities, this is the number of high-order zero bits.
10084 For signed quantities, this is the number of copies of the sign bit
10085 minus 1. In both case, this function returns the number of "spare"
10086 bits. For example, if two quantities for which this function returns
10087 at least 1 are added, the addition is known not to overflow.
10089 This function will always return 0 unless called during combine, which
10090 implies that it must be called from a define_split. */
10092 unsigned int
10094 {
10104 }
10106 /* This function is called from `simplify_shift_const' to merge two
10107 outer operations. Specifically, we have already found that we need
10108 to perform operation *POP0 with constant *PCONST0 at the outermost
10109 position. We would now like to also perform OP1 with constant CONST1
10110 (with *POP0 being done last).
10112 Return 1 if we can do the operation and update *POP0 and *PCONST0 with
10113 the resulting operation. *PCOMP_P is set to 1 if we would need to
10114 complement the innermost operand, otherwise it is unchanged.
10116 MODE is the mode in which the operation will be done. No bits outside
10117 the width of this mode matter. It is assumed that the width of this mode
10118 is smaller than or equal to HOST_BITS_PER_WIDE_INT.
10120 If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
10121 IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
10122 result is simply *PCONST0.
10124 If the resulting operation cannot be expressed as one operation, we
10125 return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
10127 static int
10128 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)
10129 {
10136 /* If OP0 is an AND, clear unimportant bits in CONST1. */
10140 /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
10141 if OP0 is SET. */
10150 {
10152 {
10170 }
10171 }
10173 /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
10177 /* If the two constants aren't the same, we can't do anything. The
10178 remaining six cases can all be done. */
10182 else
10184 {
10187 /* (a & b) | b == b */
10190 /* (a ^ b) | b == a | b */
10191 {;}
10196 /* (a & b) ^ b == (~a) & b */
10199 /* (a | b) ^ b == a & ~b */
10205 /* (a | b) & b == b */
10208 /* (a ^ b) & b) == (~a) & b */
10213 }
10215 /* Check for NO-OP cases. */
10228 /* ??? Slightly redundant with the above mask, but not entirely.
10229 Moving this above means we'd have to sign-extend the mode mask
10230 for the final test. */
10235 }
10236 \f
10237 /* A helper to simplify_shift_const_1 to determine the mode we can perform
10238 the shift in. The original shift operation CODE is performed on OP in
10239 ORIG_MODE. Return the wider mode MODE if we can perform the operation
10240 in that mode. Return ORIG_MODE otherwise. We can also assume that the
10241 result of the shift is subject to operation OUTER_CODE with operand
10242 OUTER_CONST. */
10244 static machine_mode
10248 {
10253 /* In general we can't perform in wider mode for right shift and rotate. */
10255 {
10257 /* We can still widen if the bits brought in from the left are identical
10258 to the sign bit of ORIG_MODE. */
10266 /* Similarly here but with zero bits. */
10271 /* We can also widen if the bits brought in will be masked off. This
10272 operation is performed in ORIG_MODE. */
10274 {
10280 }
10281 /* fall through */
10291 }
10292 }
10294 /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
10295 of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
10296 if we cannot simplify it. Otherwise, return a simplified value.
10298 The shift is normally computed in the widest mode we find in VAROP, as
10299 long as it isn't a different number of words than RESULT_MODE. Exceptions
10300 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10302 static rtx
10305 {
10310 machine_mode shift_mode;
10312 unsigned int mode_words
10314 /* We form (outer_op (code varop count) (outer_const)). */
10320 /* Make sure and truncate the "natural" shift on the way in. We don't
10321 want to do this inside the loop as it makes it more difficult to
10322 combine shifts. */
10326 /* If we were given an invalid count, don't do anything except exactly
10327 what was requested. */
10334 /* Unless one of the branches of the `if' in this loop does a `continue',
10335 we will `break' the loop after the `if'. */
10338 {
10339 /* If we have an operand of (clobber (const_int 0)), fail. */
10343 /* Convert ROTATERT to ROTATE. */
10345 {
10349 }
10355 /* Handle cases where the count is greater than the size of the mode
10356 minus 1. For ASHIFT, use the size minus one as the count (this can
10357 occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
10358 take the count modulo the size. For other shifts, the result is
10359 zero.
10361 Since these shifts are being produced by the compiler by combining
10362 multiple operations, each of which are defined, we know what the
10363 result is supposed to be. */
10366 {
10371 else
10372 {
10373 /* We can't simply return zero because there may be an
10374 outer op. */
10378 }
10379 }
10381 /* If we discovered we had to complement VAROP, leave. Making a NOT
10382 here would cause an infinite loop. */
10387 {
10388 /* An arithmetic right shift of a quantity known to be -1 or 0
10389 is a no-op. */
10393 {
10396 }
10398 /* If we are doing an arithmetic right shift and discarding all but
10399 the sign bit copies, this is equivalent to doing a shift by the
10400 bitsize minus one. Convert it into that shift because it will
10401 often allow other simplifications. */
10408 /* We simplify the tests below and elsewhere by converting
10409 ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10410 `make_compound_operation' will convert it to an ASHIFTRT for
10411 those machines (such as VAX) that don't have an LSHIFTRT. */
10416 shift_unit_mode)))
10428 }
10431 {
10438 {
10441 }
10445 /* The following rules apply only to scalars. */
10449 /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10450 minus the width of a smaller mode, we can do this with a
10451 SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10458 {
10467 }
10471 /* The following rules apply only to scalars. */
10475 /* If VAROP is a SUBREG, strip it as long as the inner operand has
10476 the same number of words as what we've seen so far. Then store
10477 the widest mode in MODE. */
10483 == mode_words
10485 {
10490 }
10494 /* Some machines use MULT instead of ASHIFT because MULT
10495 is cheaper. But it is still better on those machines to
10496 merge two shifts into one. */
10499 {
10500 varop
10506 }
10510 /* Similar, for when divides are cheaper. */
10513 {
10514 varop
10520 }
10524 /* If we are extracting just the sign bit of an arithmetic
10525 right shift, that shift is not needed. However, the sign
10526 bit of a wider mode may be different from what would be
10527 interpreted as the sign bit in a narrower mode, so, if
10528 the result is narrower, don't discard the shift. */
10533 {
10536 }
10538 /* fall through */
10543 /* The following rules apply only to scalars. */
10547 /* Here we have two nested shifts. The result is usually the
10548 AND of a new shift with a mask. We compute the result below. */
10554 {
10558 rtx mask_rtx;
10560 /* We have one common special case. We can't do any merging if
10561 the inner code is an ASHIFTRT of a smaller mode. However, if
10562 we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10563 with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10564 we can convert it to
10565 (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10566 This simplifies certain SIGN_EXTEND operations. */
10570 {
10571 /* C3 has the low-order C1 bits zero. */
10583 }
10585 /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10586 than C1 high-order bits equal to the sign bit, we can convert
10587 this to either an ASHIFT or an ASHIFTRT depending on the
10588 two counts.
10590 We cannot do this if VAROP's mode is not SHIFT_MODE. */
10596 {
10600 {
10603 }
10606 }
10608 /* There are some cases we can't do. If CODE is ASHIFTRT,
10609 we can only do this if FIRST_CODE is also ASHIFTRT.
10611 We can't do the case when CODE is ROTATE and FIRST_CODE is
10612 ASHIFTRT.
10614 If the mode of this shift is not the mode of the outer shift,
10615 we can't do this if either shift is a right shift or ROTATE.
10617 Finally, we can't do any of these if the mode is too wide
10618 unless the codes are the same.
10620 Handle the case where the shift codes are the same
10621 first. */
10624 {
10633 }
10644 /* To compute the mask to apply after the shift, shift the
10645 nonzero bits of the inner shift the same way the
10646 outer shift will. */
10649 result_mode);
10651 mask_rtx
10655 /* Give up if we can't compute an outer operation to use. */
10663 /* If the shifts are in the same direction, we add the
10664 counts. Otherwise, we subtract them. */
10668 else
10671 /* If COUNT is positive, the new shift is usually CODE,
10672 except for the two exceptions below, in which case it is
10673 FIRST_CODE. If the count is negative, FIRST_CODE should
10674 always be used */
10684 }
10686 /* If we have (A << B << C) for any shift, we can convert this to
10687 (A << C << B). This wins if A is a constant. Only try this if
10688 B is not a constant. */
10693 {
10694 /* For ((unsigned) (cstULL >> count)) >> cst2 we have to make
10695 sure the result will be masked. See PR70222. */
10703 /* For ((int) (cstLL >> count)) >> cst2 just give up. Queuing
10704 up outer sign extension (often left and right shift) is
10705 hardly more efficient than the original. See PR70429. */
10715 }
10719 /* The following rules apply only to scalars. */
10723 /* Make this fit the case below. */
10730 /* The following rules apply only to scalars. */
10734 /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
10735 with C the size of VAROP - 1 and the shift is logical if
10736 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10737 we have an (le X 0) operation. If we have an arithmetic shift
10738 and STORE_FLAG_VALUE is 1 or we have a logical shift with
10739 STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
10747 {
10750 const0_rtx);
10756 }
10758 /* If we have (shift (logical)), move the logical to the outside
10759 to allow it to possibly combine with another logical and the
10760 shift to combine with another shift. This also canonicalizes to
10761 what a ZERO_EXTRACT looks like. Also, some machines have
10762 (and (shift)) insns. */
10765 /* We can't do this if we have (ashiftrt (xor)) and the
10766 constant has its sign bit set in shift_mode with shift_mode
10767 wider than result_mode. */
10771 shift_mode))
10779 {
10782 }
10784 /* If we can't do that, try to simplify the shift in each arm of the
10785 logical expression, make a new logical expression, and apply
10786 the inverse distributive law. This also can't be done for
10787 (ashiftrt (xor)) where we've widened the shift and the constant
10788 changes the sign bit. */
10793 shift_mode)))
10794 {
10806 }
10810 /* The following rules apply only to scalars. */
10814 /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
10815 says that the sign bit can be tested, FOO has mode MODE, C is
10816 GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
10817 that may be nonzero. */
10827 {
10831 }
10835 /* The following rules apply only to scalars. */
10839 /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
10840 than the number of bits in the mode is equivalent to A. */
10844 {
10848 }
10850 /* NEG commutes with ASHIFT since it is multiplication. Move the
10851 NEG outside to allow shifts to combine. */
10855 {
10858 }
10862 /* The following rules apply only to scalars. */
10866 /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
10867 is one less than the number of bits in the mode is
10868 equivalent to (xor A 1). */
10875 {
10879 }
10881 /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
10882 that might be nonzero in BAR are those being shifted out and those
10883 bits are known zero in FOO, we can replace the PLUS with FOO.
10884 Similarly in the other operand order. This code occurs when
10885 we are computing the size of a variable-size array. */
10892 {
10895 }
10903 result_mode)))
10904 {
10907 }
10909 /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
10919 {
10922 }
10924 /* Check for 'PLUS signbit', which is the canonical form of 'XOR
10925 signbit', and attempt to change the PLUS to an XOR and move it to
10926 the outer operation as is done above in the AND/IOR/XOR case
10927 leg for shift(logical). See details in logical handling above
10928 for reasoning in doing so. */
10939 {
10942 }
10947 /* The following rules apply only to scalars. */
10951 /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
10952 with C the size of VAROP - 1 and the shift is logical if
10953 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10954 we have a (gt X 0) operation. If the shift is arithmetic with
10955 STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
10956 we have a (neg (gt X 0)) operation. */
10965 {
10968 const0_rtx);
10974 }
10978 /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
10979 if the truncate does not affect the value. */
10986 {
10989 varop_inner
10992 GEN_INT
10997 }
11002 }
11005 }
11010 /* We have now finished analyzing the shift. The result should be
11011 a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
11012 OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
11013 to the result of the shift. OUTER_CONST is the relevant constant,
11014 but we must turn off all bits turned off in the shift. */
11022 /* Make a SUBREG if necessary. If we can't make it, fail. */
11027 /* If we have an outer operation and we just made a shift, it is
11028 possible that we could have simplified the shift were it not
11029 for the outer operation. So try to do the simplification
11030 recursively. */
11034 else
11040 /* If we were doing an LSHIFTRT in a wider mode than it was originally,
11041 turn off all the bits that the shift would have turned off. */
11046 /* Do the remainder of the processing in RESULT_MODE. */
11049 /* If COMPLEMENT_P is set, we have to complement X before doing the outer
11050 operation. */
11055 {
11063 {
11064 /* This means that we have determined that the result is
11065 equivalent to a constant. This should be rare. */
11068 }
11071 else
11074 }
11077 }
11079 /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
11080 The result of the shift is RESULT_MODE. If we cannot simplify it,
11081 return X or, if it is NULL, synthesize the expression with
11082 simplify_gen_binary. Otherwise, return a simplified value.
11084 The shift is normally computed in the widest mode we find in VAROP, as
11085 long as it isn't a different number of words than RESULT_MODE. Exceptions
11086 are ASHIFTRT and ROTATE, which are always done in their original mode. */
11088 static rtx
11091 {
11101 }
11103 \f
11104 /* A subroutine of recog_for_combine. See there for arguments and
11105 return value. */
11107 static int
11109 {
11111 rtx pat_without_clobbers;
11119 /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
11120 we use to indicate that something didn't match. If we find such a
11121 thing, force rejection. */
11135 {
11138 else
11141 }
11143 /* If it isn't, there is the possibility that we previously had an insn
11144 that clobbered some register as a side effect, but the combined
11145 insn doesn't need to do that. So try once more without the clobbers
11146 unless this represents an ASM insn. */
11150 {
11155 {
11158 pos++;
11159 }
11169 {
11172 else
11175 }
11176 }
11183 /* Recognize all noop sets, these will be killed by followup pass. */
11187 /* If we had any clobbers to add, make a new pattern than contains
11188 them. Then check to make sure that all of them are dead. */
11190 {
11200 else
11207 {
11212 {
11216 }
11217 }
11219 }
11223 {
11231 /* Allow targets to reject combined insn. */
11233 {
11236 dump_file);
11238 /* Callers expect recog_for_combine to strip
11239 clobbers from the pattern on failure. */
11244 }
11249 }
11255 }
11257 /* Change every ZERO_EXTRACT and ZERO_EXTEND of a SUBREG that can be
11258 expressed as an AND and maybe an LSHIFTRT, to that formulation.
11259 Return whether anything was so changed. */
11261 static bool
11263 {
11269 {
11280 {
11291 else
11295 }
11302 {
11307 }
11313 {
11316 }
11317 else
11323 {
11326 }
11330 }
11341 {
11356 else
11364 }
11367 }
11369 /* Like recog, but we receive the address of a pointer to a new pattern.
11370 We try to match the rtx that the pointer points to.
11371 If that fails, we may try to modify or replace the pattern,
11372 storing the replacement into the same pointer object.
11374 Modifications include deletion or addition of CLOBBERs. If the
11375 instruction will still not match, we change ZERO_EXTEND and ZERO_EXTRACT
11376 to the equivalent AND and perhaps LSHIFTRT patterns, and try with that
11377 (and undo if that fails).
11379 PNOTES is a pointer to a location where any REG_UNUSED notes added for
11380 the CLOBBERs are placed.
11382 The value is the final insn code from the pattern ultimately matched,
11383 or -1. */
11385 static int
11387 {
11399 {
11402 {
11406 }
11407 }
11410 {
11415 }
11418 }
11419 \f
11420 /* Like gen_lowpart_general but for use by combine. In combine it
11421 is not possible to create any new pseudoregs. However, it is
11422 safe to create invalid memory addresses, because combine will
11423 try to recognize them and all they will do is make the combine
11424 attempt fail.
11426 If for some reason this cannot do its job, an rtx
11427 (clobber (const_int 0)) is returned.
11428 An insn containing that will not be recognized. */
11430 static rtx
11432 {
11436 rtx result;
11441 /* We can only support MODE being wider than a word if X is a
11442 constant integer or has a mode the same size. */
11447 /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
11448 won't know what to do. So we will strip off the SUBREG here and
11449 process normally. */
11451 {
11454 /* For use in case we fall down into the address adjustments
11455 further below, we need to adjust the known mode and size of
11456 x; imode and isize, since we just adjusted x. */
11463 }
11471 {
11474 /* Refuse to work on a volatile memory ref or one with a mode-dependent
11475 address. */
11480 /* If we want to refer to something bigger than the original memref,
11481 generate a paradoxical subreg instead. That will force a reload
11482 of the original memref X. */
11489 /* Adjust the address so that the address-after-the-data is
11490 unchanged. */
11495 }
11497 /* If X is a comparison operator, rewrite it in a new mode. This
11498 probably won't match, but may allow further simplifications. */
11502 /* If we couldn't simplify X any other way, just enclose it in a
11503 SUBREG. Normally, this SUBREG won't match, but some patterns may
11504 include an explicit SUBREG or we may simplify it further in combine. */
11505 else
11506 {
11507 rtx res;
11510 {
11515 }
11519 }
11521 fail:
11523 }
11524 \f
11525 /* Try to simplify a comparison between OP0 and a constant OP1,
11526 where CODE is the comparison code that will be tested, into a
11527 (CODE OP0 const0_rtx) form.
11529 The result is a possibly different comparison code to use.
11530 *POP1 may be updated. */
11532 static enum rtx_code
11535 {
11539 /* Get the constant we are comparing against and turn off all bits
11540 not on in our mode. */
11544 /* If we are comparing against a constant power of two and the value
11545 being compared can only have that single bit nonzero (e.g., it was
11546 `and'ed with that bit), we can replace this with a comparison
11547 with zero. */
11555 {
11558 }
11560 /* Similarly, if we are comparing a value known to be either -1 or
11561 0 with -1, change it to the opposite comparison against zero. */
11566 {
11569 }
11571 /* Do some canonicalizations based on the comparison code. We prefer
11572 comparisons against zero and then prefer equality comparisons.
11573 If we can reduce the size of a constant, we will do that too. */
11575 {
11577 /* < C is equivalent to <= (C - 1) */
11579 {
11582 /* ... fall through to LE case below. */
11584 }
11585 else
11589 /* <= C is equivalent to < (C + 1); we do this for C < 0 */
11591 {
11594 }
11596 /* If we are doing a <= 0 comparison on a value known to have
11597 a zero sign bit, we can replace this with == 0. */
11607 /* >= C is equivalent to > (C - 1). */
11609 {
11612 /* ... fall through to GT below. */
11614 }
11615 else
11619 /* > C is equivalent to >= (C + 1); we do this for C < 0. */
11621 {
11624 }
11626 /* If we are doing a > 0 comparison on a value known to have
11627 a zero sign bit, we can replace this with != 0. */
11637 /* < C is equivalent to <= (C - 1). */
11639 {
11642 /* ... fall through ... */
11643 }
11644 /* (unsigned) < 0x80000000 is equivalent to >= 0. */
11648 {
11652 }
11653 else
11657 /* unsigned <= 0 is equivalent to == 0 */
11660 /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
11664 {
11667 }
11671 /* >= C is equivalent to > (C - 1). */
11673 {
11676 /* ... fall through ... */
11677 }
11679 /* (unsigned) >= 0x80000000 is equivalent to < 0. */
11683 {
11687 }
11688 else
11692 /* unsigned > 0 is equivalent to != 0 */
11695 /* (unsigned) > 0x7fffffff is equivalent to < 0. */
11699 {
11702 }
11707 }
11711 }
11712 \f
11713 /* Simplify a comparison between *POP0 and *POP1 where CODE is the
11714 comparison code that will be tested.
11716 The result is a possibly different comparison code to use. *POP0 and
11717 *POP1 may be updated.
11719 It is possible that we might detect that a comparison is either always
11720 true or always false. However, we do not perform general constant
11721 folding in combine, so this knowledge isn't useful. Such tautologies
11722 should have been detected earlier. Hence we ignore all such cases. */
11724 static enum rtx_code
11726 {
11732 machine_mode tmode;
11734 /* Try a few ways of applying the same transformation to both operands. */
11736 {
11737 /* The test below this one won't handle SIGN_EXTENDs on these machines,
11738 so check specially. */
11754 - (GET_MODE_PRECISION
11756 {
11759 }
11761 /* If both operands are the same constant shift, see if we can ignore the
11762 shift. We can if the shift is a rotate or if the bits shifted out of
11763 this shift are known to be zero for both inputs and if the type of
11764 comparison is compatible with the shift. */
11777 {
11790 else
11792 }
11794 /* If both operands are AND's of a paradoxical SUBREG by constant, the
11795 SUBREGs are of the same mode, and, in both cases, the AND would
11796 be redundant if the comparison was done in the narrower mode,
11797 do the comparison in the narrower mode (e.g., we are AND'ing with 1
11798 and the operand's possibly nonzero bits are 0xffffff01; in that case
11799 if we only care about QImode, we don't need the AND). This case
11800 occurs if the output mode of an scc insn is not SImode and
11801 STORE_FLAG_VALUE == 1 (e.g., the 386).
11803 Similarly, check for a case where the AND's are ZERO_EXTEND
11804 operations from some narrower mode even though a SUBREG is not
11805 present. */
11810 {
11827 {
11831 /* The resulting comparison is always unsigned since we masked
11832 off the original sign bit. */
11836 }
11841 {
11847 }
11851 }
11853 /* If both operands are NOT, we can strip off the outer operation
11854 and adjust the comparison code for swapped operands; similarly for
11855 NEG, except that this must be an equality comparison. */
11861 else
11863 }
11865 /* If the first operand is a constant, swap the operands and adjust the
11866 comparison code appropriately, but don't do this if the second operand
11867 is already a constant integer. */
11869 {
11872 }
11874 /* We now enter a loop during which we will try to simplify the comparison.
11875 For the most part, we only are concerned with comparisons with zero,
11876 but some things may really be comparisons with zero but not start
11877 out looking that way. */
11880 {
11887 HOST_WIDE_INT const_op;
11889 /* We only want to handle integral modes. This catches VOIDmode,
11890 CCmode, and the floating-point modes. An exception is that we
11891 can handle VOIDmode if OP0 is a COMPARE or a comparison
11892 operation. */
11899 /* Try to simplify the compare to constant, possibly changing the
11900 comparison op, and/or changing op1 to zero. */
11904 /* Compute some predicates to simplify code below. */
11911 /* If this is a sign bit comparison and we can do arithmetic in
11912 MODE, say that we will only be needing the sign bit of OP0. */
11915 HOST_WIDE_INT_1U
11919 /* Now try cases based on the opcode of OP0. If none of the cases
11920 does a "continue", we exit this loop immediately after the
11921 switch. */
11924 {
11926 /* If we are extracting a single bit from a variable position in
11927 a constant that has only a single bit set and are comparing it
11928 with zero, we can convert this into an equality comparison
11929 between the position and the location of the single bit. */
11930 /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
11931 have already reduced the shift count modulo the word size. */
11937 {
11945 /* Result is nonzero iff shift count is equal to I. */
11948 }
11950 /* fall through */
11955 {
11958 }
11962 /* If testing for equality, we can take the NOT of the constant. */
11965 {
11969 }
11971 /* If just looking at the sign bit, reverse the sense of the
11972 comparison. */
11974 {
11978 }
11982 /* If testing for equality, we can take the NEG of the constant. */
11985 {
11989 }
11991 /* The remaining cases only apply to comparisons with zero. */
11995 /* When X is ABS or is known positive,
11996 (neg X) is < 0 if and only if X != 0. */
12004 {
12008 }
12010 /* If we have NEG of something whose two high-order bits are the
12011 same, we know that "(-a) < 0" is equivalent to "a > 0". */
12013 {
12017 }
12021 /* If we are testing equality and our count is a constant, we
12022 can perform the inverse operation on our RHS. */
12026 {
12030 }
12032 /* If we are doing a < 0 or >= 0 comparison, it means we are testing
12033 a particular bit. Convert it to an AND of a constant of that
12034 bit. This will be converted into a ZERO_EXTRACT. */
12038 {
12040 (HOST_WIDE_INT_1U
12045 }
12047 /* Fall through. */
12050 /* ABS is ignorable inside an equality comparison with zero. */
12052 {
12055 }
12059 /* Can simplify (compare (zero/sign_extend FOO) CONST) to
12060 (compare FOO CONST) if CONST fits in FOO's mode and we
12061 are either testing inequality or have an unsigned
12062 comparison with ZERO_EXTEND or a signed comparison with
12063 SIGN_EXTEND. But don't do it if we don't have a compare
12064 insn of the given mode, since we'd have to revert it
12065 later on, and then we wouldn't know whether to sign- or
12066 zero-extend. */
12069 && ! unsigned_comparison_p
12073 {
12076 }
12080 /* Check for the case where we are comparing A - C1 with C2, that is
12082 (subreg:MODE (plus (A) (-C1))) op (C2)
12084 with C1 a constant, and try to lift the SUBREG, i.e. to do the
12085 comparison in the wider mode. One of the following two conditions
12086 must be true in order for this to be valid:
12088 1. The mode extension results in the same bit pattern being added
12089 on both sides and the comparison is equality or unsigned. As
12090 C2 has been truncated to fit in MODE, the pattern can only be
12091 all 0s or all 1s.
12093 2. The mode extension results in the sign bit being copied on
12094 each side.
12096 The difficulty here is that we have predicates for A but not for
12097 (A - C1) so we need to check that C1 is within proper bounds so
12098 as to perturbate A as little as possible. */
12105 {
12114 /* (A - C1) zero-extends if it is positive and sign-extends
12115 if it is negative, C2 both zero- and sign-extends. */
12119 /* (A - C1) sign-extends if it is positive and 1-extends
12120 if it is negative, C2 both sign- and 1-extends. */
12127 /* (A - C1) always sign-extends, like C2. */
12131 {
12134 }
12135 }
12137 /* If the inner mode is narrower and we are extracting the low part,
12138 we can treat the SUBREG as if it were a ZERO_EXTEND. */
12140 ;
12149 {
12150 /* Remove outer subregs that don't do anything. */
12155 {
12159 }
12161 }
12162 else
12165 /* FALLTHROUGH */
12175 {
12178 }
12182 /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
12183 this for equality comparisons due to pathological cases involving
12184 overflows. */
12188 {
12192 }
12194 /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
12197 {
12201 }
12205 /* We used to optimize signed comparisons against zero, but that
12206 was incorrect. Unsigned comparisons against zero (GTU, LEU)
12207 arrive here as equality comparisons, or (GEU, LTU) are
12208 optimized away. No need to special-case them. */
12210 /* (eq (minus A B) C) -> (eq A (plus B C)) or
12211 (eq B (minus A C)), whichever simplifies. We can only do
12212 this for equality comparisons due to pathological cases involving
12213 overflows. */
12217 {
12221 }
12226 {
12230 }
12232 /* The sign bit of (minus (ashiftrt X C) X), where C is the number
12233 of bits in X minus 1, is one iff X > 0. */
12238 {
12242 }
12246 /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
12247 if C is zero or B is a constant. */
12251 {
12255 }
12263 /* We can't do anything if OP0 is a condition code value, rather
12264 than an actual data value. */
12270 /* Get the two operands being compared. */
12273 else
12276 /* Check for the cases where we simply want the result of the
12277 earlier test or the opposite of that result. */
12281 {
12285 else
12289 {
12294 }
12295 }
12299 /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
12300 iff X <= 0. */
12304 {
12308 }
12312 /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
12313 will be converted to a ZERO_EXTRACT later. */
12317 {
12322 }
12324 /* If we are comparing (and (lshiftrt X C1) C2) for equality with
12325 zero and X is a comparison and C1 and C2 describe only bits set
12326 in STORE_FLAG_VALUE, we can compare with X. */
12334 {
12341 {
12344 }
12345 }
12347 /* If we are doing an equality comparison of an AND of a bit equal
12348 to the sign bit, replace this with a LT or GE comparison of
12349 the underlying value. */
12356 {
12360 }
12362 /* If this AND operation is really a ZERO_EXTEND from a narrower
12363 mode, the constant fits within that mode, and this is either an
12364 equality or unsigned comparison, try to do this comparison in
12365 the narrower mode.
12367 Note that in:
12369 (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
12370 -> (ne:DI (reg:SI 4) (const_int 0))
12372 unless TRULY_NOOP_TRUNCATION allows it or the register is
12373 known to hold a value of the required mode the
12374 transformation is invalid. */
12382 {
12385 }
12387 /* If this is (and:M1 (subreg:M1 X:M2 0) (const_int C1)) where C1
12388 fits in both M1 and M2 and the SUBREG is either paradoxical
12389 or represents the low part, permute the SUBREG and the AND
12390 and try again. */
12393 {
12396 /* Require an integral mode, to avoid creating something like
12397 (AND:SF ...). */
12399 /* It is unsafe to commute the AND into the SUBREG if the
12400 SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
12401 not defined. As originally written the upper bits
12402 have a defined value due to the AND operation.
12403 However, if we commute the AND inside the SUBREG then
12404 they no longer have defined values and the meaning of
12405 the code has been changed.
12406 Also C1 should not change value in the smaller mode,
12407 see PR67028 (a positive C1 can become negative in the
12408 smaller mode, so that the AND does no longer mask the
12409 upper bits). */
12410 && ((WORD_REGISTER_OPERATIONS
12422 {
12428 }
12429 }
12431 /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
12435 {
12440 }
12442 /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
12443 (eq (and (lshiftrt X) 1) 0).
12444 Also handle the case where (not X) is expressed using xor. */
12448 {
12458 == HOST_WIDE_INT_1U
12460 {
12461 op0
12466 }
12467 }
12471 /* If we have (compare (ashift FOO N) (const_int C)) and
12472 the high order N bits of FOO (N+1 if an inequality comparison)
12473 are known to be zero, we can do this by comparing FOO with C
12474 shifted right N bits so long as the low-order N bits of C are
12475 zero. */
12487 {
12488 /* We must perform a logical shift, not an arithmetic one,
12489 as we want the top N bits of C to be zero. */
12496 }
12498 /* If we are doing a sign bit comparison, it means we are testing
12499 a particular bit. Convert it to the appropriate AND. */
12502 {
12504 (HOST_WIDE_INT_1U
12509 }
12511 /* If this an equality comparison with zero and we are shifting
12512 the low bit to the sign bit, we can convert this to an AND of the
12513 low-order bit. */
12517 {
12520 }
12524 /* If this is an equality comparison with zero, we can do this
12525 as a logical shift, which might be much simpler. */
12528 {
12533 }
12535 /* If OP0 is a sign extension and CODE is not an unsigned comparison,
12536 do the comparison in a narrower mode. */
12546 {
12549 }
12551 /* Likewise if OP0 is a PLUS of a sign extension with a
12552 constant, which is usually represented with the PLUS
12553 between the shifts. */
12565 {
12573 new_const);
12575 }
12577 /* FALLTHROUGH */
12579 /* If we have (compare (xshiftrt FOO N) (const_int C)) and
12580 the low order N bits of FOO are known to be zero, we can do this
12581 by comparing FOO with C shifted left N bits so long as no
12582 overflow occurs. Even if the low order N bits of FOO aren't known
12583 to be zero, if the comparison is >= or < we can use the same
12584 optimization and for > or <= by setting all the low
12585 order N bits in the comparison constant. */
12596 {
12597 unsigned HOST_WIDE_INT low_bits
12599 & ((HOST_WIDE_INT_1U
12602 {
12603 /* If the shift was logical, then we must make the condition
12604 unsigned. */
12613 const_op
12618 }
12619 }
12621 /* If we are using this shift to extract just the sign bit, we
12622 can replace this with an LT or GE comparison. */
12627 {
12631 }
12636 }
12639 }
12641 /* Now make any compound operations involved in this comparison. Then,
12642 check for an outmost SUBREG on OP0 that is not doing anything or is
12643 paradoxical. The latter transformation must only be performed when
12644 it is known that the "extra" bits will be the same in op0 and op1 or
12645 that they don't matter. There are three cases to consider:
12647 1. SUBREG_REG (op0) is a register. In this case the bits are don't
12648 care bits and we can assume they have any convenient value. So
12649 making the transformation is safe.
12651 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is UNKNOWN.
12652 In this case the upper bits of op0 are undefined. We should not make
12653 the simplification in that case as we do not know the contents of
12654 those bits.
12656 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not UNKNOWN.
12657 In that case we know those bits are zeros or ones. We must also be
12658 sure that they are the same as the upper bits of op1.
12660 We can never remove a SUBREG for a non-equality comparison because
12661 the sign bit is in a different place in the underlying object. */
12674 {
12676 {
12677 /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
12678 implemented. */
12680 {
12683 }
12684 }
12688 {
12693 }
12694 }
12696 /* We now do the opposite procedure: Some machines don't have compare
12697 insns in all modes. If OP0's mode is an integer mode smaller than a
12698 word and we can't do a compare in that mode, see if there is a larger
12699 mode for which we can do the compare. There are a number of cases in
12700 which we can use the wider mode. */
12706 {
12710 {
12713 /* If this is a test for negative, we can make an explicit
12714 test of the sign bit. Test this first so we can use
12715 a paradoxical subreg to extend OP0. */
12719 {
12720 unsigned HOST_WIDE_INT sign
12727 }
12729 /* If the only nonzero bits in OP0 and OP1 are those in the
12730 narrower mode and this is an equality or unsigned comparison,
12731 we can use the wider mode. Similarly for sign-extended
12732 values, in which case it is true for all comparisons. */
12749 {
12750 /* If OP0 is an AND and we don't have an AND in MODE either,
12751 make a new AND in the proper mode. */
12759 else
12760 {
12762 {
12767 }
12768 else
12769 {
12774 }
12776 }
12777 }
12778 }
12779 }
12781 /* We may have changed the comparison operands. Re-canonicalize. */
12783 {
12786 }
12788 /* If this machine only supports a subset of valid comparisons, see if we
12789 can convert an unsupported one into a supported one. */
12796 }
12797 \f
12798 /* Utility function for record_value_for_reg. Count number of
12799 rtxs in X. */
12800 static int
12802 {
12809 {
12829 }
12840 }
12841 \f
12842 /* Utility function for following routine. Called when X is part of a value
12843 being stored into last_set_value. Sets last_set_table_tick
12844 for each register mentioned. Similar to mention_regs in cse.c */
12846 static void
12848 {
12854 {
12860 {
12863 }
12866 }
12870 {
12871 /* Check for identical subexpressions. If x contains
12872 identical subexpression we only have to traverse one of
12873 them. */
12875 {
12876 /* Note that at this point x1 has already been
12877 processed. */
12881 /* If x0 and x1 are identical then there is no need to
12882 process x0. */
12886 /* If x0 is identical to a subexpression of x1 then while
12887 processing x1, x0 has already been processed. Thus we
12888 are done with x. */
12893 /* If x1 is identical to a subexpression of x0 then we
12894 still have to process the rest of x0. */
12897 {
12900 }
12901 }
12904 }
12908 }
12910 /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
12911 are saying that the register is clobbered and we no longer know its
12912 value. If INSN is zero, don't update reg_stat[].last_set; this is
12913 only permitted with VALUE also zero and is used to invalidate the
12914 register. */
12916 static void
12918 {
12924 /* If VALUE contains REG and we have a previous value for REG, substitute
12925 the previous value. */
12927 {
12928 rtx tem;
12930 /* Set things up so get_last_value is allowed to see anything set up to
12931 our insn. */
12935 /* If TEM is simply a binary operation with two CLOBBERs as operands,
12936 it isn't going to be useful and will take a lot of time to process,
12937 so just use the CLOBBER. */
12940 {
12946 {
12947 /* If there are two or more occurrences of REG in VALUE,
12948 prevent the value from growing too much. */
12951 }
12954 }
12955 }
12957 /* For each register modified, show we don't know its value, that
12958 we don't know about its bitwise content, that its value has been
12959 updated, and that we don't know the location of the death of the
12960 register. */
12962 {
12974 }
12976 /* Mark registers that are being referenced in this value. */
12980 /* Now update the status of each register being set.
12981 If someone is using this register in this block, set this register
12982 to invalid since we will get confused between the two lives in this
12983 basic block. This makes using this register always invalid. In cse, we
12984 scan the table to invalidate all entries using this register, but this
12985 is too much work for us. */
12988 {
12994 else
12996 }
12998 /* The value being assigned might refer to X (like in "x++;"). In that
12999 case, we must replace it with (clobber (const_int 0)) to prevent
13000 infinite loops. */
13003 {
13007 }
13009 /* For the main register being modified, update the value, the mode, the
13010 nonzero bits, and the number of sign bit copies. */
13015 {
13023 rsp->last_set_sign_bit_copies
13025 }
13026 }
13028 /* Called via note_stores from record_dead_and_set_regs to handle one
13029 SET or CLOBBER in an insn. DATA is the instruction in which the
13030 set is occurring. */
13032 static void
13034 {
13041 {
13045 }
13048 {
13049 /* If we are setting the whole register, we know its value. Otherwise
13050 show that we don't know the value. We can handle SUBREG in
13051 some cases. */
13062 else
13064 }
13066 /* Ignore pushes, they clobber nothing. */
13069 }
13071 /* Update the records of when each REG was most recently set or killed
13072 for the things done by INSN. This is the last thing done in processing
13073 INSN in the combiner loop.
13075 We update reg_stat[], in particular fields last_set, last_set_value,
13076 last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
13077 last_death, and also the similar information mem_last_set (which insn
13078 most recently modified memory) and last_call_luid (which insn was the
13079 most recent subroutine call). */
13081 static void
13083 {
13084 rtx link;
13088 {
13091 {
13096 {
13101 }
13102 }
13105 }
13108 {
13109 hard_reg_set_iterator hrsi;
13111 {
13123 }
13127 /* We can't combine into a call pattern. Remember, though, that
13128 the return value register is set at this LUID. We could
13129 still replace a register with the return value from the
13130 wrong subroutine call! */
13132 }
13133 else
13135 }
13137 /* If a SUBREG has the promoted bit set, it is in fact a property of the
13138 register present in the SUBREG, so for each such SUBREG go back and
13139 adjust nonzero and sign bit information of the registers that are
13140 known to have some zero/sign bits set.
13142 This is needed because when combine blows the SUBREGs away, the
13143 information on zero/sign bits is lost and further combines can be
13144 missed because of that. */
13146 static void
13148 {
13150 rtx set;
13158 {
13167 {
13170 }
13174 {
13177 }
13180 {
13183 }
13184 else
13186 }
13187 }
13189 /* Check if X, a register, is known to contain a value already
13190 truncated to MODE. In this case we can use a subreg to refer to
13191 the truncated value even though in the generic case we would need
13192 an explicit truncation. */
13194 static bool
13196 {
13208 }
13210 /* If X is a hard reg or a subreg record the mode that the register is
13211 accessed in. For non-TRULY_NOOP_TRUNCATION targets we might be able
13212 to turn a truncate into a subreg using this information. Return true
13213 if traversing X is complete. */
13215 static bool
13217 {
13218 machine_mode truncated_mode;
13222 {
13233 }
13234 /* ??? For hard-regs we now record everything. We might be able to
13235 optimize this using last_set_mode. */
13238 else
13246 {
13249 }
13252 }
13254 /* Callback for note_uses. Find hardregs and subregs of pseudos and
13255 the modes they are used in. This can help truning TRUNCATEs into
13256 SUBREGs. */
13258 static void
13260 {
13265 }
13267 /* Scan X for promoted SUBREGs. For each one found,
13268 note what it implies to the registers used in it. */
13270 static void
13272 {
13277 else
13278 {
13284 {
13294 }
13295 }
13296 }
13297 \f
13298 /* Verify that all the registers and memory references mentioned in *LOC are
13299 still valid. *LOC was part of a value set in INSN when label_tick was
13300 equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
13301 the invalid references with (clobber (const_int 0)) and return 1. This
13302 replacement is useful because we often can get useful information about
13303 the form of a value (e.g., if it was produced by a shift that always
13304 produces -1 or 0) even though we don't know exactly what registers it
13305 was produced from. */
13307 static int
13309 {
13316 {
13322 {
13325 /* If this is a pseudo-register that was only set once and not
13326 live at the beginning of the function, it is always valid. */
13330 && (!REGNO_REG_SET_P
13332 regno)))
13334 {
13338 }
13339 }
13342 }
13343 /* If this is a memory reference, make sure that there were no stores after
13344 it that might have clobbered the value. We don't have alias info, so we
13345 assume any store invalidates it. Moreover, we only have local UIDs, so
13346 we also assume that there were stores in the intervening basic blocks. */
13349 {
13353 }
13356 {
13358 {
13359 /* Check for identical subexpressions. If x contains
13360 identical subexpression we only have to traverse one of
13361 them. */
13363 {
13364 /* Note that at this point x0 has already been checked
13365 and found valid. */
13369 /* If x0 and x1 are identical then x is also valid. */
13373 /* If x1 is identical to a subexpression of x0 then
13374 while checking x0, x1 has already been checked. Thus
13375 it is valid and so as x. */
13380 /* If x0 is identical to a subexpression of x1 then x is
13381 valid iff the rest of x1 is valid. */
13384 return
13388 }
13393 }
13399 }
13401 /* If we haven't found a reason for it to be invalid, it is valid. */
13403 }
13405 /* Get the last value assigned to X, if known. Some registers
13406 in the value may be replaced with (clobber (const_int 0)) if their value
13407 is known longer known reliably. */
13409 static rtx
13411 {
13413 rtx value;
13416 /* If this is a non-paradoxical SUBREG, get the value of its operand and
13417 then convert it to the desired mode. If this is a paradoxical SUBREG,
13418 we cannot predict what values the "extra" bits might have. */
13432 /* If we don't have a value, or if it isn't for this basic block and
13433 it's either a hard register, set more than once, or it's a live
13434 at the beginning of the function, return 0.
13436 Because if it's not live at the beginning of the function then the reg
13437 is always set before being used (is never used without being set).
13438 And, if it's set only once, and it's always set before use, then all
13439 uses must have the same last value, even if it's not from this basic
13440 block. */
13447 || REGNO_REG_SET_P
13451 /* If the value was set in a later insn than the ones we are processing,
13452 we can't use it even if the register was only set once. */
13457 /* If fewer bits were set than what we are asked for now, we cannot use
13458 the value. */
13463 /* If the value has all its registers valid, return it. */
13467 /* Otherwise, make a copy and replace any invalid register with
13468 (clobber (const_int 0)). If that fails for some reason, return 0. */
13475 }
13476 \f
13477 /* Return nonzero if expression X refers to a REG or to memory
13478 that is set in an instruction more recent than FROM_LUID. */
13480 static int
13482 {
13488 {
13492 #ifdef PUSH_ROUNDING
13493 /* Don't allow uses of the stack pointer to be moved,
13494 because we don't know whether the move crosses a push insn. */
13497 #endif
13499 {
13505 }
13507 }
13515 {
13517 {
13522 }
13526 }
13528 }
13529 \f
13530 /* Define three variables used for communication between the following
13531 routines. */
13536 /* Function called via note_stores from reg_dead_at_p.
13538 If DEST is within [reg_dead_regno, reg_dead_endregno), set
13539 reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
13541 static void
13543 {
13553 }
13555 /* Return nonzero if REG is known to be dead at INSN.
13557 We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
13558 referencing REG, it is dead. If we hit a SET referencing REG, it is
13559 live. Otherwise, see if it is live or dead at the start of the basic
13560 block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
13561 must be assumed to be always live. */
13563 static int
13565 {
13566 basic_block block;
13569 /* Set variables for reg_dead_at_p_1. */
13575 /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
13576 we allow the machine description to decide whether use-and-clobber
13577 patterns are OK. */
13579 {
13583 }
13585 /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
13586 beginning of basic block. */
13589 {
13591 {
13601 }
13607 }
13609 /* Look at live-in sets for the basic block that we were in. */
13615 }
13616 \f
13617 /* Note hard registers in X that are used. */
13619 static void
13621 {
13627 {
13631 CASE_CONST_ANY:
13636 /* CC0 must die in the insn after it is set, so we don't need to take
13637 special note of it here. */
13642 /* If we are clobbering a MEM, mark any hard registers inside the
13643 address as used. */
13650 /* A hard reg in a wide mode may really be multiple registers.
13651 If so, mark all of them just like the first. */
13653 {
13654 /* None of this applies to the stack, frame or arg pointers. */
13656 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
13664 }
13668 {
13669 /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
13670 the address. */
13682 }
13687 }
13689 /* Recursively scan the operands of this expression. */
13691 {
13695 {
13699 {
13704 }
13705 }
13706 }
13707 }
13708 \f
13709 /* Remove register number REGNO from the dead registers list of INSN.
13711 Return the note used to record the death, if there was one. */
13713 rtx
13715 {
13722 }
13724 /* For each register (hardware or pseudo) used within expression X, if its
13725 death is in an instruction with luid between FROM_LUID (inclusive) and
13726 TO_INSN (exclusive), put a REG_DEAD note for that register in the
13727 list headed by PNOTES.
13729 That said, don't move registers killed by maybe_kill_insn.
13731 This is done when X is being merged by combination into TO_INSN. These
13732 notes will then be distributed as needed. */
13734 static void
13737 {
13743 {
13747 /* Don't move the register if it gets killed in between from and to. */
13756 {
13759 /* It is possible for the call above to return 0. This can occur
13760 when last_death points to I2 or I1 that we combined with.
13761 In that case make a new note.
13763 We must also check for the case where X is a hard register
13764 and NOTE is a death note for a range of hard registers
13765 including X. In that case, we must put REG_DEAD notes for
13766 the remaining registers in place of NOTE. */
13771 {
13780 }
13782 /* If we didn't find any note, or if we found a REG_DEAD note that
13783 covers only part of the given reg, and we have a multi-reg hard
13784 register, then to be safe we must check for REG_DEAD notes
13785 for each register other than the first. They could have
13786 their own REG_DEAD notes lying around. */
13793 {
13800 else
13806 }
13809 {
13812 }
13813 else
13815 }
13818 }
13821 {
13826 /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
13827 that accesses one word of a multi-word item, some
13828 piece of everything register in the expression is used by
13829 this insn, so remove any old death. */
13830 /* ??? So why do we test for equality of the sizes? */
13839 {
13842 }
13844 /* If this is some other SUBREG, we know it replaces the entire
13845 value, so use that as the destination. */
13849 /* If this is a MEM, adjust deaths of anything used in the address.
13850 For a REG (the only other possibility), the entire value is
13851 being replaced so the old value is not used in this insn. */
13857 }
13866 {
13868 {
13873 }
13876 }
13877 }
13878 \f
13879 /* Return 1 if X is the target of a bit-field assignment in BODY, the
13880 pattern of an insn. X must be a REG. */
13882 static int
13884 {
13888 {
13890 rtx target;
13897 else
13914 }
13922 }
13923 \f
13924 /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
13925 as appropriate. I3 and I2 are the insns resulting from the combination
13926 insns including FROM (I2 may be zero).
13928 ELIM_I2 and ELIM_I1 are either zero or registers that we know will
13929 not need REG_DEAD notes because they are being substituted for. This
13930 saves searching in the most common cases.
13932 Each note in the list is either ignored or placed on some insns, depending
13933 on the type of note. */
13935 static void
13938 {
13940 rtx tem_note;
13944 {
13949 {
13952 /* Doesn't matter much where we put this, as long as it's somewhere.
13953 It is preferable to keep these notes on branches, which is most
13954 likely to be i3. */
13961 else
13962 {
13965 }
13969 /* These notes must remain with the call or trapping instruction. */
13974 else
13975 {
13981 /* ??? Otherwise assume we've combined things such that we
13982 can now prove that the instructions can't trap. Drop the
13983 note in this case. */
13984 }
13988 /* ??? How to distribute between i3-i1. Assume i3 contains the
13989 entire adjustment. Assert i3 contains at least some adjust. */
13991 {
13993 /* fixup_args_size_notes looks at REG_NORETURN note,
13994 so ensure the note is placed there first. */
13996 {
14000 {
14006 }
14007 }
14009 /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
14010 REG_ARGS_SIZE note to all noreturn calls, allow that here. */
14013 && !ACCUMULATE_OUTGOING_ARGS
14015 }
14022 /* These notes must remain with the call. It should not be
14023 possible for both I2 and I3 to be a call. */
14026 else
14027 {
14030 }
14034 /* Any clobbers for i3 may still exist, and so we must process
14035 REG_UNUSED notes from that insn.
14037 Any clobbers from i2 or i1 can only exist if they were added by
14038 recog_for_combine. In that case, recog_for_combine created the
14039 necessary REG_UNUSED notes. Trying to keep any original
14040 REG_UNUSED notes from these insns can cause incorrect output
14041 if it is for the same register as the original i3 dest.
14042 In that case, we will notice that the register is set in i3,
14043 and then add a REG_UNUSED note for the destination of i3, which
14044 is wrong. However, it is possible to have REG_UNUSED notes from
14045 i2 or i1 for register which were both used and clobbered, so
14046 we keep notes from i2 or i1 if they will turn into REG_DEAD
14047 notes. */
14049 /* If this register is set or clobbered in I3, put the note there
14050 unless there is one already. */
14052 {
14060 }
14061 /* Otherwise, if this register is used by I3, then this register
14062 now dies here, so we must put a REG_DEAD note here unless there
14063 is one already. */
14069 {
14072 }
14078 /* These notes say something about results of an insn. We can
14079 only support them if they used to be on I3 in which case they
14080 remain on I3. Otherwise they are ignored.
14082 If the note refers to an expression that is not a constant, we
14083 must also ignore the note since we cannot tell whether the
14084 equivalence is still true. It might be possible to do
14085 slightly better than this (we only have a problem if I2DEST
14086 or I1DEST is present in the expression), but it doesn't
14087 seem worth the trouble. */
14095 /* These notes say something about how a register is used. They must
14096 be present on any use of the register in I2 or I3. */
14101 {
14104 else
14106 }
14111 /* This can show up in several ways -- either directly in the
14112 pattern, or hidden off in the constant pool with (or without?)
14113 a REG_EQUAL note. */
14114 /* ??? Ignore the without-reg_equal-note problem for now. */
14126 {
14129 else
14131 }
14133 /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
14134 as a JUMP_LABEL or decrement LABEL_NUSES if it's already
14135 there. */
14140 {
14147 }
14153 {
14161 }
14165 /* This note says something about the value of a register prior
14166 to the execution of an insn. It is too much trouble to see
14167 if the note is still correct in all situations. It is better
14168 to simply delete it. */
14172 /* If we replaced the right hand side of FROM_INSN with a
14173 REG_EQUAL note, the original use of the dying register
14174 will not have been combined into I3 and I2. In such cases,
14175 FROM_INSN is guaranteed to be the first of the combined
14176 instructions, so we simply need to search back before
14177 FROM_INSN for the previous use or set of this register,
14178 then alter the notes there appropriately.
14180 If the register is used as an input in I3, it dies there.
14181 Similarly for I2, if it is nonzero and adjacent to I3.
14183 If the register is not used as an input in either I3 or I2
14184 and it is not one of the registers we were supposed to eliminate,
14185 there are two possibilities. We might have a non-adjacent I2
14186 or we might have somehow eliminated an additional register
14187 from a computation. For example, we might have had A & B where
14188 we discover that B will always be zero. In this case we will
14189 eliminate the reference to A.
14191 In both cases, we must search to see if we can find a previous
14192 use of A and put the death note there. */
14198 else
14199 {
14210 && !(i2mod
14212 i2mod_old_rhs)))
14217 /* If the new I2 sets the same register that is marked dead
14218 in the note, we do not know where to put the note.
14219 Give up. */
14222 }
14225 {
14229 {
14231 {
14235 }
14237 /* If the register is being set at TEM_INSN, see if that is all
14238 TEM_INSN is doing. If so, delete TEM_INSN. Otherwise, make this
14239 into a REG_UNUSED note instead. Don't delete sets to
14240 global register vars. */
14244 {
14255 ;
14257 /* Verify that it was the set, and not a clobber that
14258 modified the register.
14260 CC0 targets must be careful to maintain setter/user
14261 pairs. If we cannot delete the setter due to side
14262 effects, mark the user with an UNUSED note instead
14263 of deleting it. */
14267 && (!HAVE_cc0
14271 {
14272 /* Move the notes and links of TEM_INSN elsewhere.
14273 This might delete other dead insns recursively.
14274 First set the pattern to something that won't use
14275 any register. */
14294 /* Delete the setter too. */
14296 {
14309 }
14310 }
14311 else
14312 {
14315 /* If there isn't already a REG_UNUSED note, put one
14316 here. Do not place a REG_DEAD note, even if
14317 the register is also used here; that would not
14318 match the algorithm used in lifetime analysis
14319 and can cause the consistency check in the
14320 scheduler to fail. */
14325 }
14326 }
14330 {
14333 /* If we are doing a 3->2 combination, and we have a
14334 register which formerly died in i3 and was not used
14335 by i2, which now no longer dies in i3 and is used in
14336 i2 but does not die in i2, and place is between i2
14337 and i3, then we may need to move a link from place to
14338 i2. */
14340 && from_insn
14343 {
14347 }
14349 }
14353 }
14355 }
14357 /* If the register is set or already dead at PLACE, we needn't do
14358 anything with this note if it is still a REG_DEAD note.
14359 We check here if it is set at all, not if is it totally replaced,
14360 which is what `dead_or_set_p' checks, so also check for it being
14361 set partially. */
14364 {
14370 {
14371 /* Unless the register previously died in PLACE, clear
14372 last_death. [I no longer understand why this is
14373 being done.] */
14377 }
14378 else
14381 /* If this is a death note for a hard reg that is occupying
14382 multiple registers, ensure that we are still using all
14383 parts of the object. If we find a piece of the object
14384 that is unused, we must arrange for an appropriate REG_DEAD
14385 note to be added for it. However, we can't just emit a USE
14386 and tag the note to it, since the register might actually
14387 be dead; so we recourse, and the recursive call then finds
14388 the previous insn that used this register. */
14391 {
14400 {
14403 }
14406 {
14407 /* Put only REG_DEAD notes for pieces that are
14408 not already dead or set. */
14412 {
14419 {
14421 NULL_RTX);
14425 NULL_RTX);
14426 }
14431 {
14433 {
14437 }
14441 {
14444 }
14445 }
14446 }
14449 }
14450 }
14451 }
14455 /* Any other notes should not be present at this point in the
14456 compilation. */
14458 }
14461 {
14464 }
14468 }
14469 }
14470 \f
14471 /* Similarly to above, distribute the LOG_LINKS that used to be present on
14472 I3, I2, and I1 to new locations. This is also called to add a link
14473 pointing at I3 when I3's destination is changed. */
14475 static void
14477 {
14481 {
14488 /* If the insn that this link points to is a NOTE, ignore it. */
14497 {
14500 {
14516 }
14519 }
14520 else
14530 /* A LOG_LINK is defined as being placed on the first insn that uses
14531 a register and points to the insn that sets the register. Start
14532 searching at the next insn after the target of the link and stop
14533 when we reach a set of the register or the end of the basic block.
14535 Note that this correctly handles the link that used to point from
14536 I3 to I2. Also note that not much searching is typically done here
14537 since most links don't point very far away. */
14546 {
14550 }
14553 {
14556 }
14560 /* If we found a place to put the link, place it there unless there
14561 is already a link to the same insn as LINK at that point. */
14564 {
14572 {
14576 /* Set added_links_insn to the earliest insn we added a
14577 link to. */
14581 }
14582 }
14583 }
14584 }
14585 \f
14586 /* Check for any register or memory mentioned in EQUIV that is not
14587 mentioned in EXPR. This is used to restrict EQUIV to "specializations"
14588 of EXPR where some registers may have been replaced by constants. */
14590 static bool
14592 {
14595 {
14600 }
14602 }
14603 \f
14604 DEBUG_FUNCTION void
14606 {
14607 fprintf
14609 ";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
14611 }
14613 void
14615 {
14616 fprintf
14618 "\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
14620 }
14621 \f
14622 /* Try combining insns through substitution. */
14623 static unsigned int
14625 {
14635 rebuild_jump_labels_after_combine
14638 /* Combining insns may have turned an indirect jump into a
14639 direct jump. Rebuild the JUMP_LABEL fields of jumping
14640 instructions. */
14642 {
14649 }
14653 }
14658 {
14668 };
14671 {
14675 {}
14677 /* opt_pass methods: */
14680 {
14682 }
14688 rtl_opt_pass *
14690 {
14692 }