| blob | 75c6229766c00b410ad263963c54d8b7d8eee18f |
1 /* Optimize by combining instructions for GNU compiler.
2 Copyright (C) 1987-2016 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. */
96 /* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
106 #ifndef LOAD_EXTEND_OP
107 #define LOAD_EXTEND_OP(M) UNKNOWN
108 #endif
110 /* Number of attempts to combine instructions in this function. */
114 /* Number of attempts that got as far as substitution in this function. */
118 /* Number of instructions combined with added SETs in this function. */
122 /* Number of instructions combined in this function. */
126 /* Totals over entire compilation. */
130 /* combine_instructions may try to replace the right hand side of the
131 second instruction with the value of an associated REG_EQUAL note
132 before throwing it at try_combine. That is problematic when there
133 is a REG_DEAD note for a register used in the old right hand side
134 and can cause distribute_notes to do wrong things. This is the
135 second instruction if it has been so modified, null otherwise. */
139 /* When I2MOD is nonnull, this is a copy of the old right hand side. */
143 /* When I2MOD is nonnull, this is a copy of the new right hand side. */
146 \f
148 /* Record last point of death of (hard or pseudo) register n. */
151 /* Record last point of modification of (hard or pseudo) register n. */
154 /* The next group of fields allows the recording of the last value assigned
155 to (hard or pseudo) register n. We use this information to see if an
156 operation being processed is redundant given a prior operation performed
157 on the register. For example, an `and' with a constant is redundant if
158 all the zero bits are already known to be turned off.
160 We use an approach similar to that used by cse, but change it in the
161 following ways:
163 (1) We do not want to reinitialize at each label.
164 (2) It is useful, but not critical, to know the actual value assigned
165 to a register. Often just its form is helpful.
167 Therefore, we maintain the following fields:
169 last_set_value the last value assigned
170 last_set_label records the value of label_tick when the
171 register was assigned
172 last_set_table_tick records the value of label_tick when a
173 value using the register is assigned
174 last_set_invalid set to nonzero when it is not valid
175 to use the value of this register in some
176 register's value
178 To understand the usage of these tables, it is important to understand
179 the distinction between the value in last_set_value being valid and
180 the register being validly contained in some other expression in the
181 table.
183 (The next two parameters are out of date).
185 reg_stat[i].last_set_value is valid if it is nonzero, and either
186 reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
188 Register I may validly appear in any expression returned for the value
189 of another register if reg_n_sets[i] is 1. It may also appear in the
190 value for register J if reg_stat[j].last_set_invalid is zero, or
191 reg_stat[i].last_set_label < reg_stat[j].last_set_label.
193 If an expression is found in the table containing a register which may
194 not validly appear in an expression, the register is replaced by
195 something that won't match, (clobber (const_int 0)). */
197 /* Record last value assigned to (hard or pseudo) register n. */
199 rtx last_set_value;
201 /* Record the value of label_tick when an expression involving register n
202 is placed in last_set_value. */
206 /* Record the value of label_tick when the value for register n is placed in
207 last_set_value. */
211 /* These fields are maintained in parallel with last_set_value and are
212 used to store the mode in which the register was last set, the bits
213 that were known to be zero when it was last set, and the number of
214 sign bits copies it was known to have when it was last set. */
220 /* Set nonzero if references to register n in expressions should not be
221 used. last_set_invalid is set nonzero when this register is being
222 assigned to and last_set_table_tick == label_tick. */
226 /* Some registers that are set more than once and used in more than one
227 basic block are nevertheless always set in similar ways. For example,
228 a QImode register may be loaded from memory in two places on a machine
229 where byte loads zero extend.
231 We record in the following fields if a register has some leading bits
232 that are always equal to the sign bit, and what we know about the
233 nonzero bits of a register, specifically which bits are known to be
234 zero.
236 If an entry is zero, it means that we don't know anything special. */
242 /* Record the value of the label_tick when the last truncation
243 happened. The field truncated_to_mode is only valid if
244 truncation_label == label_tick. */
248 /* Record the last truncation seen for this register. If truncation
249 is not a nop to this mode we might be able to save an explicit
250 truncation if we know that value already contains a truncated
251 value. */
254 };
259 /* One plus the highest pseudo for which we track REG_N_SETS.
260 regstat_init_n_sets_and_refs allocates the array for REG_N_SETS just once,
261 but during combine_split_insns new pseudos can be created. As we don't have
262 updated DF information in that case, it is hard to initialize the array
263 after growing. The combiner only cares about REG_N_SETS (regno) == 1,
264 so instead of growing the arrays, just assume all newly created pseudos
265 during combine might be set multiple times. */
269 /* Record the luid of the last insn that invalidated memory
270 (anything that writes memory, and subroutine calls, but not pushes). */
274 /* Record the luid of the last CALL_INSN
275 so we can tell whether a potential combination crosses any calls. */
279 /* When `subst' is called, this is the insn that is being modified
280 (by combining in a previous insn). The PATTERN of this insn
281 is still the old pattern partially modified and it should not be
282 looked at, but this may be used to examine the successors of the insn
283 to judge whether a simplification is valid. */
287 /* This is the lowest LUID that `subst' is currently dealing with.
288 get_last_value will not return a value if the register was set at or
289 after this LUID. If not for this mechanism, we could get confused if
290 I2 or I1 in try_combine were an insn that used the old value of a register
291 to obtain a new value. In that case, we might erroneously get the
292 new value of the register when we wanted the old one. */
296 /* This contains any hard registers that are used in newpat; reg_dead_at_p
297 must consider all these registers to be always live. */
301 /* This is an insn to which a LOG_LINKS entry has been added. If this
302 insn is the earlier than I2 or I3, combine should rescan starting at
303 that location. */
307 /* Basic block in which we are performing combines. */
311 \f
312 /* Length of the currently allocated uid_insn_cost array. */
316 /* The following array records the insn_rtx_cost for every insn
317 in the instruction stream. */
321 /* The following array records the LOG_LINKS for every insn in the
322 instruction stream as struct insn_link pointers. */
328 };
332 #define INSN_COST(INSN) (uid_insn_cost[INSN_UID (INSN)])
333 #define LOG_LINKS(INSN) (uid_log_links[INSN_UID (INSN)])
335 #define FOR_EACH_LOG_LINK(L, INSN) \
336 for ((L) = LOG_LINKS (INSN); (L); (L) = (L)->next)
338 /* Links for LOG_LINKS are allocated from this obstack. */
342 /* Allocate a link. */
346 {
354 }
356 /* Incremented for each basic block. */
360 /* Reset to label_tick for each extended basic block in scanning order. */
364 /* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
365 largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
369 /* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
370 be safely used. It is zero while computing them and after combine has
371 completed. This former test prevents propagating values based on
372 previously set values, which can be incorrect if a variable is modified
373 in a loop. */
377 \f
378 /* Record one modification to rtl structure
379 to be undone by storing old_contents into *where. */
383 struct undo
384 {
389 };
391 /* Record a bunch of changes to be undone, up to MAX_UNDO of them.
392 num_undo says how many are currently recorded.
394 other_insn is nonzero if we have modified some other insn in the process
395 of working on subst_insn. It must be verified too. */
397 struct undobuf
398 {
402 };
406 /* Number of times the pseudo being substituted for
407 was found and replaced. */
412 machine_mode,
416 machine_mode,
488 \f
490 /* It is not safe to use ordinary gen_lowpart in combine.
491 See comments in gen_lowpart_for_combine. */
492 #undef RTL_HOOKS_GEN_LOWPART
493 #define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
495 /* Our implementation of gen_lowpart never emits a new pseudo. */
496 #undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
497 #define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
499 #undef RTL_HOOKS_REG_NONZERO_REG_BITS
500 #define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
502 #undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
503 #define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
505 #undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
506 #define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
510 \f
511 /* Convenience wrapper for the canonicalize_comparison target hook.
512 Target hooks cannot use enum rtx_code. */
516 {
520 }
522 /* Try to split PATTERN found in INSN. This returns NULL_RTX if
523 PATTERN can not be split. Otherwise, it returns an insn sequence.
524 This is a wrapper around split_insns which ensures that the
525 reg_stat vector is made larger if the splitter creates a new
526 register. */
530 {
539 }
541 /* This is used by find_single_use to locate an rtx in LOC that
542 contains exactly one use of DEST, which is typically either a REG
543 or CC0. It returns a pointer to the innermost rtx expression
544 containing DEST. Appearances of DEST that are being used to
545 totally replace it are not counted. */
549 {
558 {
562 CASE_CONST_ANY:
567 /* If the destination is anything other than CC0, PC, a REG or a SUBREG
568 of a REG that occupies all of the REG, the insn uses DEST if
569 it is mentioned in the destination or the source. Otherwise, we
570 need just check the source. */
590 }
592 /* If it wasn't one of the common cases above, check each expression and
593 vector of this code. Look for a unique usage of DEST. */
597 {
599 {
604 else
610 /* Duplicate usage. */
612 }
614 {
618 {
624 else
631 }
632 }
633 }
636 }
639 /* See if DEST, produced in INSN, is used only a single time in the
640 sequel. If so, return a pointer to the innermost rtx expression in which
641 it is used.
643 If PLOC is nonzero, *PLOC is set to the insn containing the single use.
645 If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
646 care about REG_DEAD notes or LOG_LINKS.
648 Otherwise, we find the single use by finding an insn that has a
649 LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
650 only referenced once in that insn, we know that it must be the first
651 and last insn referencing DEST. */
655 {
656 basic_block bb;
662 {
672 }
682 {
688 {
693 }
694 }
697 }
698 \f
699 /* Substitute NEWVAL, an rtx expression, into INTO, a place in some
700 insn. The substitution can be undone by undo_all. If INTO is already
701 set to NEWVAL, do not record this change. Because computing NEWVAL might
702 also call SUBST, we have to compute it before we put anything into
703 the undo table. */
705 static void
707 {
714 /* We'd like to catch as many invalid transformations here as
715 possible. Unfortunately, there are way too many mode changes
716 that are perfectly valid, so we'd waste too much effort for
717 little gain doing the checks here. Focus on catching invalid
718 transformations involving integer constants. */
721 {
722 /* Sanity check that we're replacing oldval with a CONST_INT
723 that is a valid sign-extension for the original mode. */
727 /* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
728 CONST_INT is not valid, because after the replacement, the
729 original mode would be gone. Unfortunately, we can't tell
730 when do_SUBST is called to replace the operand thereof, so we
731 perform this test on oldval instead, checking whether an
732 invalid replacement took place before we got here. */
737 }
741 else
750 }
752 #define SUBST(INTO, NEWVAL) do_SUBST (&(INTO), (NEWVAL))
754 /* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
755 for the value of a HOST_WIDE_INT value (including CONST_INT) is
756 not safe. */
758 static void
760 {
769 else
778 }
780 #define SUBST_INT(INTO, NEWVAL) do_SUBST_INT (&(INTO), (NEWVAL))
782 /* Similar to SUBST, but just substitute the mode. This is used when
783 changing the mode of a pseudo-register, so that any other
784 references to the entry in the regno_reg_rtx array will change as
785 well. */
787 static void
789 {
798 else
807 }
809 #define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE (&(INTO), (NEWVAL))
811 /* Similar to SUBST, but NEWVAL is a LOG_LINKS expression. */
813 static void
815 {
824 else
833 }
835 #define SUBST_LINK(oldval, newval) do_SUBST_LINK (&oldval, newval)
836 \f
837 /* Subroutine of try_combine. Determine whether the replacement patterns
838 NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to insn_rtx_cost
839 than the original sequence I0, I1, I2, I3 and undobuf.other_insn. Note
840 that I0, I1 and/or NEWI2PAT may be NULL_RTX. Similarly, NEWOTHERPAT and
841 undobuf.other_insn may also both be NULL_RTX. Return false if the cost
842 of all the instructions can be estimated and the replacements are more
843 expensive than the original sequence. */
845 static bool
848 {
853 /* Lookup the original insn_rtx_costs. */
858 {
861 {
865 }
866 else
867 {
871 }
872 }
873 else
874 {
877 }
879 /* If we have split a PARALLEL I2 to I1,I2, we have counted its cost twice;
880 correct that. */
885 /* Calculate the replacement insn_rtx_costs. */
888 {
892 }
893 else
894 {
897 }
900 {
906 {
909 }
910 else
912 }
914 /* Disallow this combination if both new_cost and old_cost are greater than
915 zero, and new_cost is greater than old cost. */
919 {
938 else
940 }
945 /* Update the uid_insn_cost array with the replacement costs. */
949 {
953 }
956 }
959 /* Delete any insns that copy a register to itself. */
961 static void
963 {
965 basic_block bb;
968 {
970 {
973 {
978 }
979 }
980 }
981 }
983 \f
984 /* Return false if we do not want to (or cannot) combine DEF. */
985 static bool
987 {
988 /* Do not consider if it is pre/post modification in MEM. */
994 /* Do not combine frame pointer adjustments. */
997 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
1005 }
1007 /* Return false if we do not want to (or cannot) combine USE. */
1008 static bool
1010 {
1011 /* Do not consider the usage of the stack pointer by function call. */
1016 }
1018 /* Fill in log links field for all insns. */
1020 static void
1022 {
1023 basic_block bb;
1030 /* Pass through each block from the end, recording the uses of each
1031 register and establishing log links when def is encountered.
1032 Note that we do not clear next_use array in order to save time,
1033 so we have to test whether the use is in the same basic block as def.
1035 There are a few cases below when we do not consider the definition or
1036 usage -- these are taken from original flow.c did. Don't ask me why it is
1037 done this way; I don't know and if it works, I don't want to know. */
1040 {
1042 {
1046 /* Log links are created only once. */
1050 {
1066 /* flow.c claimed:
1068 We don't build a LOG_LINK for hard registers contained
1069 in ASM_OPERANDs. If these registers get replaced,
1070 we might wind up changing the semantics of the insn,
1071 even if reload can make what appear to be valid
1072 assignments later. */
1077 /* Don't add duplicate links between instructions. */
1086 }
1091 }
1092 }
1095 }
1097 /* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
1098 true if we found a LOG_LINK that proves that A feeds B. This only works
1099 if there are no instructions between A and B which could have a link
1100 depending on A, since in that case we would not record a link for B.
1101 We also check the implicit dependency created by a cc0 setter/user
1102 pair. */
1104 static bool
1106 {
1114 }
1115 \f
1116 /* Main entry point for combiner. F is the first insn of the function.
1117 NREGS is the first unused pseudo-reg number.
1119 Return nonzero if the combiner has turned an indirect jump
1120 instruction into a direct jump. */
1121 static int
1123 {
1128 basic_block last_bb;
1148 /* Allocate array for insn info. */
1156 /* Don't use reg_stat[].nonzero_bits when computing it. This can cause
1157 problems when, for example, we have j <<= 1 in a loop. */
1162 /* Scan all SETs and see if we can deduce anything about what
1163 bits are known to be zero for some registers and how many copies
1164 of the sign bit are known to exist for those registers.
1166 Also set any known values so that we can use it while searching
1167 for what bits are known to be set. */
1170 /* Allow the entry block and the first block to fall into the same EBB.
1171 Conceptually the incoming promotions are assigned to the entry block. */
1176 {
1181 label_tick++;
1189 {
1190 rtx links;
1196 insn);
1203 insn);
1205 /* Record the current insn_rtx_cost of this instruction. */
1208 optimize_this_for_speed_p);
1212 }
1213 }
1217 /* Now scan all the insns in forward order. */
1225 {
1231 label_tick++;
1241 {
1255 /* See if we know about function return values before this
1256 insn based upon SUBREG flags. */
1259 /* See if we can find hardregs and subreg of pseudos in
1260 narrower modes. This could help turning TRUNCATEs
1261 into SUBREGs. */
1264 /* Try this insn with each insn it links back to. */
1270 {
1273 }
1275 /* Try each sequence of three linked insns ending with this one. */
1279 {
1282 /* If the linked insn has been replaced by a note, then there
1283 is no point in pursuing this chain any further. */
1291 {
1294 }
1295 }
1297 /* Try to combine a jump insn that uses CC0
1298 with a preceding insn that sets CC0, and maybe with its
1299 logical predecessor as well.
1300 This is how we make decrement-and-branch insns.
1301 We need this special code because data flow connections
1302 via CC0 do not get entered in LOG_LINKS. */
1309 {
1320 }
1322 /* Do the same for an insn that explicitly references CC0. */
1329 {
1340 }
1342 /* Finally, see if any of the insns that this insn links to
1343 explicitly references CC0. If so, try this insn, that insn,
1344 and its predecessor if it sets CC0. */
1346 {
1358 }
1360 /* Try combining an insn with two different insns whose results it
1361 uses. */
1371 {
1374 }
1376 /* Try four-instruction combinations. */
1379 {
1383 /* If the linked insn has been replaced by a note, then there
1384 is no point in pursuing this chain any further. */
1389 {
1393 /* I0 -> I1 -> I2 -> I3. */
1399 {
1402 }
1403 /* I0, I1 -> I2, I2 -> I3. */
1410 {
1413 }
1414 }
1417 {
1421 /* I0 -> I2; I1, I2 -> I3. */
1427 {
1430 }
1431 /* I0 -> I1; I1, I2 -> I3. */
1437 {
1440 }
1441 }
1442 }
1444 /* Try this insn with each REG_EQUAL note it links back to. */
1446 {
1452 /* Avoid using a register that may already been marked
1453 dead by an earlier instruction. */
1461 {
1462 /* Temporarily replace the set's source with the
1463 contents of the REG_EQUAL note. The insn will
1464 be deleted or recognized by try_combine. */
1475 last_combined_insn);
1478 {
1481 }
1484 }
1485 }
1490 retry:
1491 ;
1492 }
1493 }
1500 /* Clean up. */
1506 {
1509 {
1512 }
1514 }
1524 /* Make recognizer allow volatile MEMs again. */
1528 }
1530 /* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
1532 static void
1534 {
1540 }
1541 \f
1542 /* Set up any promoted values for incoming argument registers. */
1544 static void
1546 {
1547 tree arg;
1552 {
1557 /* Only continue if the incoming argument is in a register. */
1561 /* Determine, if possible, whether all call sites of the current
1562 function lie within the current compilation unit. (This does
1563 take into account the exporting of a function via taking its
1564 address, and so forth.) */
1567 /* The mode and signedness of the argument before any promotions happen
1568 (equal to the mode of the pseudo holding it at that stage). */
1572 /* The mode and signedness of the argument after any source language and
1573 TARGET_PROMOTE_PROTOTYPES-driven promotions. */
1577 /* The mode and signedness of the argument as it is actually passed,
1578 see assign_parm_setup_reg in function.c. */
1582 /* The mode of the register in which the argument is being passed. */
1585 /* Eliminate sign extensions in the callee when:
1586 (a) A mode promotion has occurred; */
1589 /* (b) The mode of the register is the same as the mode of
1590 the argument as it is passed; */
1593 /* (c) There's no language level extension; */
1595 ;
1596 /* (c.1) All callers are from the current compilation unit. If that's
1597 the case we don't have to rely on an ABI, we only have to know
1598 what we're generating right now, and we know that we will do the
1599 mode1 to mode2 promotion with the given sign. */
1602 /* (c.2) The combination of the two promotions is useful. This is
1603 true when the signs match, or if the first promotion is unsigned.
1604 In the later case, (sign_extend (zero_extend x)) is the same as
1605 (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
1611 /* Record that the value was promoted from mode1 to mode3,
1612 so that any sign extension at the head of the current
1613 function may be eliminated. */
1617 }
1618 }
1620 /* If MODE has a precision lower than PREC and SRC is a non-negative constant
1621 that would appear negative in MODE, sign-extend SRC for use in nonzero_bits
1622 because some machines (maybe most) will actually do the sign-extension and
1623 this is the conservative approach.
1625 ??? For 2.5, try to tighten up the MD files in this regard instead of this
1626 kludge. */
1628 static rtx
1630 {
1638 }
1640 /* Update RSP for pseudo-register X from INSN's REG_EQUAL note (if one exists)
1641 and SET. */
1643 static void
1645 rtx x)
1646 {
1656 {
1660 }
1662 /* Don't call nonzero_bits if it cannot change anything. */
1664 {
1669 }
1671 /* Don't call num_sign_bit_copies if it cannot change anything. */
1673 {
1676 {
1680 }
1683 }
1684 }
1686 /* Called via note_stores. If X is a pseudo that is narrower than
1687 HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
1689 If we are setting only a portion of X and we can't figure out what
1690 portion, assume all bits will be used since we don't know what will
1691 be happening.
1693 Similarly, set how many bits of X are known to be copies of the sign bit
1694 at all locations in the function. This is the smallest number implied
1695 by any set of X. */
1697 static void
1699 {
1704 /* If this register is undefined at the start of the file, we can't
1705 say what its contents were. */
1706 && ! REGNO_REG_SET_P
1709 {
1713 {
1717 }
1719 /* If this register is being initialized using itself, and the
1720 register is uninitialized in this basic block, and there are
1721 no LOG_LINKS which set the register, then part of the
1722 register is uninitialized. In that case we can't assume
1723 anything about the number of nonzero bits.
1725 ??? We could do better if we checked this in
1726 reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
1727 could avoid making assumptions about the insn which initially
1728 sets the register, while still using the information in other
1729 insns. We would have to be careful to check every insn
1730 involved in the combination. */
1736 {
1743 {
1747 }
1748 }
1750 /* If this is a complex assignment, see if we can convert it into a
1751 simple assignment. */
1754 /* If this is a simple assignment, or we have a paradoxical SUBREG,
1755 set what we know about X. */
1761 else
1762 {
1765 }
1766 }
1767 }
1768 \f
1769 /* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
1770 optionally insns that were previously combined into I3 or that will be
1771 combined into the merger of INSN and I3. The order is PRED, PRED2,
1772 INSN, SUCC, SUCC2, I3.
1774 Return 0 if the combination is not allowed for any reason.
1776 If the combination is allowed, *PDEST will be set to the single
1777 destination of INSN and *PSRC to the single source, and this function
1778 will return 1. */
1780 static int
1784 {
1789 rtx link;
1794 {
1796 {
1801 }
1806 }
1810 /* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
1811 or a PARALLEL consisting of such a SET and CLOBBERs.
1813 If INSN has CLOBBER parallel parts, ignore them for our processing.
1814 By definition, these happen during the execution of the insn. When it
1815 is merged with another insn, all bets are off. If they are, in fact,
1816 needed and aren't also supplied in I3, they may be added by
1817 recog_for_combine. Otherwise, it won't match.
1819 We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
1820 note.
1822 Get the source and destination of INSN. If more than one, can't
1823 combine. */
1829 {
1831 {
1835 {
1836 /* This is important to combine floating point insns
1837 for the SH4 port. */
1839 /* Combining an isolated USE doesn't make sense.
1840 We depend here on combinable_i3pat to reject them. */
1841 /* The code below this loop only verifies that the inputs of
1842 the SET in INSN do not change. We call reg_set_between_p
1843 to verify that the REG in the USE does not change between
1844 I3 and INSN.
1845 If the USE in INSN was for a pseudo register, the matching
1846 insn pattern will likely match any register; combining this
1847 with any other USE would only be safe if we knew that the
1848 used registers have identical values, or if there was
1849 something to tell them apart, e.g. different modes. For
1850 now, we forgo such complicated tests and simply disallow
1851 combining of USES of pseudo registers with any other USE. */
1854 {
1859 do
1860 {
1870 }
1872 }
1875 /* We can ignore CLOBBERs. */
1880 /* Ignore SETs whose result isn't used but not those that
1881 have side-effects. */
1887 /* If we have already found a SET, this is a second one and
1888 so we cannot combine with this insn. */
1896 /* Anything else means we can't combine. */
1898 }
1899 }
1902 /* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
1903 so don't do anything with it. */
1906 }
1907 else
1913 /* The simplification in expand_field_assignment may call back to
1914 get_last_value, so set safe guard here. */
1920 /* Do not eliminate user-specified register if it is in an
1921 asm input because we may break the register asm usage defined
1922 in GCC manual if allow to do so.
1923 Be aware that this may cover more cases than we expect but this
1924 should be harmless. */
1929 /* Don't eliminate a store in the stack pointer. */
1931 /* Don't combine with an insn that sets a register to itself if it has
1932 a REG_EQUAL note. This may be part of a LIBCALL sequence. */
1934 /* Can't merge an ASM_OPERANDS. */
1936 /* Can't merge a function call. */
1938 /* Don't eliminate a function call argument. */
1944 /* Don't substitute into an incremented register. */
1948 /* Don't substitute into a non-local goto, this confuses CFG. */
1950 /* Make sure that DEST is not used after SUCC but before I3. */
1951 || (!all_adjacent
1952 && ((succ2
1956 /* Make sure that the value that is to be substituted for the register
1957 does not use any registers whose values alter in between. However,
1958 If the insns are adjacent, a use can't cross a set even though we
1959 think it might (this can happen for a sequence of insns each setting
1960 the same destination; last_set of that register might point to
1961 a NOTE). If INSN has a REG_EQUIV note, the register is always
1962 equivalent to the memory so the substitution is valid even if there
1963 are intervening stores. Also, don't move a volatile asm or
1964 UNSPEC_VOLATILE across any other insns. */
1965 || (! all_adjacent
1971 /* Don't combine across a CALL_INSN, because that would possibly
1972 change whether the life span of some REGs crosses calls or not,
1973 and it is a pain to update that information.
1974 Exception: if source is a constant, moving it later can't hurt.
1975 Accept that as a special case. */
1979 /* DEST must either be a REG or CC0. */
1981 {
1982 /* If register alignment is being enforced for multi-word items in all
1983 cases except for parameters, it is possible to have a register copy
1984 insn referencing a hard register that is not allowed to contain the
1985 mode being copied and which would not be valid as an operand of most
1986 insns. Eliminate this problem by not combining with such an insn.
1988 Also, on some machines we don't want to extend the life of a hard
1989 register. */
1994 /* Don't extend the life of a hard register unless it is
1995 user variable (if we have few registers) or it can't
1996 fit into the desired register (meaning something special
1997 is going on).
1998 Also avoid substituting a return register into I3, because
1999 reload can't handle a conflict with constraints of other
2000 inputs. */
2004 }
2012 {
2015 /* If the clobber represents an earlyclobber operand, we must not
2016 substitute an expression containing the clobbered register.
2017 As we do not analyze the constraint strings here, we have to
2018 make the conservative assumption. However, if the register is
2019 a fixed hard reg, the clobber cannot represent any operand;
2020 we leave it up to the machine description to either accept or
2021 reject use-and-clobber patterns. */
2027 }
2029 /* If INSN contains anything volatile, or is an `asm' (whether volatile
2030 or not), reject, unless nothing volatile comes between it and I3 */
2033 {
2034 /* Make sure neither succ nor succ2 contains a volatile reference. */
2039 /* We'll check insns between INSN and I3 below. */
2040 }
2042 /* If INSN is an asm, and DEST is a hard register, reject, since it has
2043 to be an explicit register variable, and was chosen for a reason. */
2049 /* If INSN contains volatile references (specifically volatile MEMs),
2050 we cannot combine across any other volatile references.
2051 Even if INSN doesn't contain volatile references, any intervening
2052 volatile insn might affect machine state. */
2055 ? volatile_refs_p
2062 /* If INSN contains an autoincrement or autodecrement, make sure that
2063 register is not used between there and I3, and not already used in
2064 I3 either. Neither must it be used in PRED or SUCC, if they exist.
2065 Also insist that I3 not be a jump; if it were one
2066 and the incremented register were spilled, we would lose. */
2084 /* Don't combine an insn that follows a CC0-setting insn.
2085 An insn that uses CC0 must not be separated from the one that sets it.
2086 We do, however, allow I2 to follow a CC0-setting insn if that insn
2087 is passed as I1; in that case it will be deleted also.
2088 We also allow combining in this case if all the insns are adjacent
2089 because that would leave the two CC0 insns adjacent as well.
2090 It would be more logical to test whether CC0 occurs inside I1 or I2,
2091 but that would be much slower, and this ought to be equivalent. */
2094 {
2099 }
2101 /* If we get here, we have passed all the tests and the combination is
2102 to be allowed. */
2108 }
2109 \f
2110 /* LOC is the location within I3 that contains its pattern or the component
2111 of a PARALLEL of the pattern. We validate that it is valid for combining.
2113 One problem is if I3 modifies its output, as opposed to replacing it
2114 entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
2115 doing so would produce an insn that is not equivalent to the original insns.
2117 Consider:
2119 (set (reg:DI 101) (reg:DI 100))
2120 (set (subreg:SI (reg:DI 101) 0) <foo>)
2122 This is NOT equivalent to:
2124 (parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
2125 (set (reg:DI 101) (reg:DI 100))])
2127 Not only does this modify 100 (in which case it might still be valid
2128 if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
2130 We can also run into a problem if I2 sets a register that I1
2131 uses and I1 gets directly substituted into I3 (not via I2). In that
2132 case, we would be getting the wrong value of I2DEST into I3, so we
2133 must reject the combination. This case occurs when I2 and I1 both
2134 feed into I3, rather than when I1 feeds into I2, which feeds into I3.
2135 If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
2136 of a SET must prevent combination from occurring. The same situation
2137 can occur for I0, in which case I0_NOT_IN_SRC is set.
2139 Before doing the above check, we first try to expand a field assignment
2140 into a set of logical operations.
2142 If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
2143 we place a register that is both set and used within I3. If more than one
2144 such register is detected, we fail.
2146 Return 1 if the combination is valid, zero otherwise. */
2148 static int
2151 {
2155 {
2160 rtx subdest;
2167 /* Check for the case where I3 modifies its output, as discussed
2168 above. We don't want to prevent pseudos from being combined
2169 into the address of a MEM, so only prevent the combination if
2170 i1 or i2 set the same MEM. */
2180 /* This is the same test done in can_combine_p except we can't test
2181 all_adjacent; we don't have to, since this instruction will stay
2182 in place, thus we are not considering increasing the lifetime of
2183 INNER_DEST.
2185 Also, if this insn sets a function argument, combining it with
2186 something that might need a spill could clobber a previous
2187 function argument; the all_adjacent test in can_combine_p also
2188 checks this; here, we do a more specific test for this case. */
2198 /* If DEST is used in I3, it is being killed in this insn, so
2199 record that for later. We have to consider paradoxical
2200 subregs here, since they kill the whole register, but we
2201 ignore partial subregs, STRICT_LOW_PART, etc.
2202 Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
2203 STACK_POINTER_REGNUM, since these are always considered to be
2204 live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
2214 && (HARD_FRAME_POINTER_IS_FRAME_POINTER
2220 {
2225 }
2226 }
2229 {
2236 }
2239 }
2240 \f
2241 /* Return 1 if X is an arithmetic expression that contains a multiplication
2242 and division. We don't count multiplications by powers of two here. */
2244 static int
2246 {
2248 {
2264 }
2265 }
2266 \f
2267 /* Determine whether INSN can be used in a combination. Return nonzero if
2268 not. This is used in try_combine to detect early some cases where we
2269 can't perform combinations. */
2271 static int
2273 {
2274 rtx set;
2277 /* If this isn't really an insn, we can't do anything.
2278 This can occur when flow deletes an insn that it has merged into an
2279 auto-increment address. */
2283 /* Never combine loads and stores involving hard regs that are likely
2284 to be spilled. The register allocator can usually handle such
2285 reg-reg moves by tying. If we allow the combiner to make
2286 substitutions of likely-spilled regs, reload might die.
2287 As an exception, we allow combinations involving fixed regs; these are
2288 not available to the register allocator so there's no risk involved. */
2309 }
2311 struct likely_spilled_retval_info
2312 {
2315 };
2317 /* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
2318 hard registers that are known to be written to / clobbered in full. */
2319 static void
2321 {
2338 else
2341 }
2343 /* Return nonzero iff part of the return value is live during INSN, and
2344 it is likely spilled. This can happen when more than one insn is needed
2345 to copy the return value, e.g. when we consider to combine into the
2346 second copy insn for a complex value. */
2348 static int
2350 {
2352 rtx reg;
2355 /* We assume here that no machine mode needs more than
2356 32 hard registers when the value overlaps with a register
2357 for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
2372 /* Disregard parts of the return value that are set later. */
2381 /* Check if any of the (probably) live return value registers is
2382 likely spilled. */
2383 nregs --;
2384 do
2385 {
2391 }
2393 /* Adjust INSN after we made a change to its destination.
2395 Changing the destination can invalidate notes that say something about
2396 the results of the insn and a LOG_LINK pointing to the insn. */
2398 static void
2400 {
2401 /* For notes, be conservative and simply remove them. */
2404 /* The new insn will have a destination that was previously the destination
2405 of an insn just above it. Call distribute_links to make a LOG_LINK from
2406 the next use of that destination. */
2422 }
2424 /* Return TRUE if combine can reuse reg X in mode MODE.
2425 ADDED_SETS is nonzero if the original set is still required. */
2426 static bool
2428 {
2435 /* Allow hard registers if the new mode is legal, and occupies no more
2436 registers than the old mode. */
2441 /* Or a pseudo that is only used once. */
2444 && !added_sets
2446 }
2449 /* Check whether X, the destination of a set, refers to part of
2450 the register specified by REG. */
2452 static bool
2454 {
2455 /* Check that reg is an integer mode register. */
2466 }
2468 /* Delete the unconditional jump INSN and adjust the CFG correspondingly.
2469 Note that the INSN should be deleted *after* removing dead edges, so
2470 that the kept edge is the fallthrough edge for a (set (pc) (pc))
2471 but not for a (set (pc) (label_ref FOO)). */
2473 static void
2475 {
2483 {
2488 /* Remove barriers from the footer if there are any. */
2491 {
2494 else
2498 }
2501 }
2502 }
2504 /* Return whether PAT is a PARALLEL of exactly N register SETs followed
2505 by an arbitrary number of CLOBBERs. */
2506 static bool
2508 {
2527 }
2529 /* Return whether INSN, a PARALLEL of N register SETs (and maybe some
2530 CLOBBERs), can be split into individual SETs in that order, without
2531 changing semantics. */
2532 static bool
2534 {
2542 {
2551 }
2554 }
2556 /* Try to combine the insns I0, I1 and I2 into I3.
2557 Here I0, I1 and I2 appear earlier than I3.
2558 I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
2559 I3.
2561 If we are combining more than two insns and the resulting insn is not
2562 recognized, try splitting it into two insns. If that happens, I2 and I3
2563 are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
2564 Otherwise, I0, I1 and I2 are pseudo-deleted.
2566 Return 0 if the combination does not work. Then nothing is changed.
2567 If we did the combination, return the insn at which combine should
2568 resume scanning.
2570 Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
2571 new direct jump instruction.
2573 LAST_COMBINED_INSN is either I3, or some insn after I3 that has
2574 been I3 passed to an earlier try_combine within the same basic
2575 block. */
2580 {
2581 /* New patterns for I3 and I2, respectively. */
2585 /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
2586 dead. */
2588 /* Total number of SETs to put into I3. */
2590 /* Nonzero if I2's or I1's body now appears in I3. */
2592 /* INSN_CODEs for new I3, new I2, and user of condition code. */
2594 /* Contains I3 if the destination of I3 is used in its source, which means
2595 that the old life of I3 is being killed. If that usage is placed into
2596 I2 and not in I3, a REG_DEAD note must be made. */
2598 /* SET_DEST and SET_SRC of I2, I1 and I0. */
2600 /* Copy of SET_SRC of I1 and I0, if needed. */
2602 /* Set if I2DEST was reused as a scratch register. */
2604 /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
2606 /* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
2611 /* Notes that must be added to REG_NOTES in I3 and I2. */
2613 /* Notes that we substituted I3 into I2 instead of the normal case. */
2615 /* Notes that I1, I2 or I3 is a MULT operation. */
2622 rtx temp_expr;
2625 rtx new_other_notes;
2628 /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
2629 never be). */
2633 /* Only try four-insn combinations when there's high likelihood of
2634 success. Look for simple insns, such as loads of constants or
2635 binary operations involving a constant. */
2637 {
2647 {
2650 rtx src;
2655 {
2658 }
2660 ngood++;
2663 nshift++;
2664 }
2666 /* If I0 loads a memory and I3 sets the same memory, then I1 and I2
2667 are likely manipulating its value. Ideally we'll be able to combine
2668 all four insns into a bitfield insertion of some kind.
2670 Note the source in I0 might be inside a sign/zero extension and the
2671 memory modes in I0 and I3 might be different. So extract the address
2672 from the destination of I3 and search for it in the source of I0.
2674 In the event that there's a match but the source/dest do not actually
2675 refer to the same memory, the worst that happens is we try some
2676 combinations that we wouldn't have otherwise. */
2678 /* Ensure the source of SET0 is a MEM, possibly buried inside
2679 an extension. */
2685 /* Ensure the destination of SET3 is a MEM. */
2687 /* Would it be better to extract the base address for the MEM
2688 in SET3 and look for that? I don't have cases where it matters
2689 but I could envision such cases. */
2695 }
2697 /* Exit early if one of the insns involved can't be used for
2698 combinations. */
2709 combine_attempts++;
2712 /* Reset the hard register usage information. */
2716 {
2723 else
2726 }
2728 /* If multiple insns feed into one of I2 or I3, they can be in any
2729 order. To simplify the code below, reorder them in sequence. */
2739 /* First check for one important special case that the code below will
2740 not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
2741 and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
2742 we may be able to replace that destination with the destination of I3.
2743 This occurs in the common code where we compute both a quotient and
2744 remainder into a structure, in which case we want to do the computation
2745 directly into the structure to avoid register-register copies.
2747 Note that this case handles both multiple sets in I2 and also cases
2748 where I2 has a number of CLOBBERs inside the PARALLEL.
2750 We make very conservative checks below and only try to handle the
2751 most common cases of this. For example, we only handle the case
2752 where I2 and I3 are adjacent to avoid making difficult register
2753 usage tests. */
2761 /* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
2762 below would need to check what is inside (and reg_overlap_mentioned_p
2763 doesn't support those codes anyway). Don't allow those destinations;
2764 the resulting insn isn't likely to be recognized anyway. */
2770 {
2773 /* Make sure that the destination of I3,
2774 which we are going to substitute into one output of I2,
2775 is not used within another output of I2. We must avoid making this:
2776 (parallel [(set (mem (reg 69)) ...)
2777 (set (reg 69) ...)])
2778 which is not well-defined as to order of actions.
2779 (Besides, reload can't handle output reloads for this.)
2781 The problem can also happen if the dest of I3 is a memory ref,
2782 if another dest in I2 is an indirect memory ref. */
2790 /* Make sure this PARALLEL is not an asm. We do not allow combining
2791 that usually (see can_combine_p), so do not here either. */
2801 {
2802 combine_merges++;
2812 /* Replace the dest in I2 with our dest and make the resulting
2813 insn the new pattern for I3. Then skip to where we validate
2814 the pattern. Everything was set up above. */
2819 }
2820 }
2822 /* If I2 is setting a pseudo to a constant and I3 is setting some
2823 sub-part of it to another constant, merge them by making a new
2824 constant. */
2831 {
2837 {
2840 {
2846 }
2847 }
2848 else
2849 {
2854 }
2857 {
2858 /* If this is the low part, we're done. */
2860 ;
2861 /* Handle the case where inner is twice the size of outer. */
2865 /* Otherwise give up for now. */
2866 else
2868 }
2871 {
2875 wide_int o
2880 combine_merges++;
2887 /* Replace the source in I2 with the new constant and make the
2888 resulting insn the new pattern for I3. Then skip to where we
2889 validate the pattern. Everything was set up above. */
2895 /* The dest of I3 has been replaced with the dest of I2. */
2898 }
2899 }
2901 /* If we have no I1 and I2 looks like:
2902 (parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
2903 (set Y OP)])
2904 make up a dummy I1 that is
2905 (set Y OP)
2906 and change I2 to be
2907 (set (reg:CC X) (compare:CC Y (const_int 0)))
2909 (We can ignore any trailing CLOBBERs.)
2911 This undoes a previous combination and allows us to match a branch-and-
2912 decrement insn. */
2924 {
2925 /* We make I1 with the same INSN_UID as I2. This gives it
2926 the same DF_INSN_LUID for value tracking. Our fake I1 will
2927 never appear in the insn stream so giving it the same INSN_UID
2928 as I2 will not cause a problem. */
2941 }
2943 /* If I2 is a PARALLEL of two SETs of REGs (and perhaps some CLOBBERs),
2944 make those two SETs separate I1 and I2 insns, and make an I0 that is
2945 the original I1. */
2951 {
2952 /* If there is no I1, there is no I0 either. */
2955 /* We make I1 with the same INSN_UID as I2. This gives it
2956 the same DF_INSN_LUID for value tracking. Our fake I1 will
2957 never appear in the insn stream so giving it the same INSN_UID
2958 as I2 will not cause a problem. */
2966 }
2968 /* Verify that I2 and I1 are valid for combining. */
2974 {
2977 }
2979 /* Record whether I2DEST is used in I2SRC and similarly for the other
2980 cases. Knowing this will help in register status updating below. */
2991 /* For the earlier insns, determine which of the subsequent ones they
2992 feed. */
2999 /* Ensure that I3's pattern can be the destination of combines. */
3005 {
3008 }
3010 /* See if any of the insns is a MULT operation. Unless one is, we will
3011 reject a combination that is, since it must be slower. Be conservative
3012 here. */
3020 /* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
3021 We used to do this EXCEPT in one case: I3 has a post-inc in an
3022 output operand. However, that exception can give rise to insns like
3023 mov r3,(r3)+
3024 which is a famous insn on the PDP-11 where the value of r3 used as the
3025 source was model-dependent. Avoid this sort of thing. */
3027 #if 0
3033 /* It's not the exception. */
3034 #endif
3036 {
3037 rtx link;
3043 {
3046 }
3047 }
3049 /* See if the SETs in I1 or I2 need to be kept around in the merged
3050 instruction: whenever the value set there is still needed past I3.
3051 For the SET in I2, this is easy: we see if I2DEST dies or is set in I3.
3053 For the SET in I1, we have two cases: if I1 and I2 independently feed
3054 into I3, the set in I1 needs to be kept around unless I1DEST dies
3055 or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
3056 in I1 needs to be kept around unless I1DEST dies or is set in either
3057 I2 or I3. The same considerations apply to I0. */
3064 else
3072 else
3075 /* We are about to copy insns for the case where they need to be kept
3076 around. Check that they can be copied in the merged instruction. */
3082 {
3085 }
3087 /* If the set in I2 needs to be kept around, we must make a copy of
3088 PATTERN (I2), so that when we substitute I1SRC for I1DEST in
3089 PATTERN (I2), we are only substituting for the original I1DEST, not into
3090 an already-substituted copy. This also prevents making self-referential
3091 rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
3092 I2DEST. */
3095 {
3098 else
3100 }
3103 {
3106 else
3108 }
3111 {
3114 else
3116 }
3118 combine_merges++;
3120 /* Substitute in the latest insn for the regs set by the earlier ones. */
3126 /* Many machines that don't use CC0 have insns that can both perform an
3127 arithmetic operation and set the condition code. These operations will
3128 be represented as a PARALLEL with the first element of the vector
3129 being a COMPARE of an arithmetic operation with the constant zero.
3130 The second element of the vector will set some pseudo to the result
3131 of the same arithmetic operation. If we simplify the COMPARE, we won't
3132 match such a pattern and so will generate an extra insn. Here we test
3133 for this case, where both the comparison and the operation result are
3134 needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
3135 I2SRC. Later we will make the PARALLEL that contains I2. */
3141 {
3142 rtx newpat_dest;
3156 {
3161 }
3163 /* Do the rest only if op1 is const0_rtx, which may be the
3164 result of simplification. */
3166 {
3167 /* If a single use of the CC is found, prepare to modify it
3168 when SELECT_CC_MODE returns a new CC-class mode, or when
3169 the above simplify_compare_const() returned a new comparison
3170 operator. undobuf.other_insn is assigned the CC use insn
3171 when modifying it. */
3173 {
3174 #ifdef SELECT_CC_MODE
3175 machine_mode new_mode
3180 {
3185 else
3186 {
3189 }
3190 }
3191 #endif
3192 /* Cases for modifying the CC-using comparison. */
3194 /* ??? Do we need to verify the zero rtx? */
3196 {
3197 /* Replace cc_use_loc with entire new RTX. */
3202 }
3204 {
3205 /* Just replace the CC reg with a new mode. */
3208 }
3209 }
3211 /* Now we modify the current newpat:
3212 First, SET_DEST(newpat) is updated if the CC mode has been
3213 altered. For targets without SELECT_CC_MODE, this should be
3214 optimized away. */
3217 /* This is always done to propagate i2src into newpat. */
3220 /* Create new version of i2pat if needed; the below PARALLEL
3221 creation needs this to work correctly. */
3225 }
3226 }
3229 {
3230 /* It is possible that the source of I2 or I1 may be performing
3231 an unneeded operation, such as a ZERO_EXTEND of something
3232 that is known to have the high part zero. Handle that case
3233 by letting subst look at the inner insns.
3235 Another way to do this would be to have a function that tries
3236 to simplify a single insn instead of merging two or more
3237 insns. We don't do this because of the potential of infinite
3238 loops and because of the potential extra memory required.
3239 However, doing it the way we are is a bit of a kludge and
3240 doesn't catch all cases.
3242 But only do this if -fexpensive-optimizations since it slows
3243 things down and doesn't usually win.
3245 This is not done in the COMPARE case above because the
3246 unmodified I2PAT is used in the PARALLEL and so a pattern
3247 with a modified I2SRC would not match. */
3250 {
3251 /* Pass pc_rtx so no substitutions are done, just
3252 simplifications. */
3254 {
3257 }
3261 }
3266 /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a unique
3267 copy of I2SRC each time we substitute it, in order to avoid creating
3268 self-referential RTL when we will be substituting I1SRC for I1DEST
3269 later. Likewise if I0 feeds into I2, either directly or indirectly
3270 through I1, and I0DEST is in I0SRC. */
3277 /* Record whether I2's body now appears within I3's body. */
3279 }
3281 /* If we already got a failure, don't try to do more. Otherwise, try to
3282 substitute I1 if we have it. */
3285 {
3286 /* Check that an autoincrement side-effect on I1 has not been lost.
3287 This happens if I1DEST is mentioned in I2 and dies there, and
3288 has disappeared from the new pattern. */
3290 && i1_feeds_i2_n
3293 /* Before we can do this substitution, we must redo the test done
3294 above (see detailed comments there) that ensures I1DEST isn't
3295 mentioned in any SETs in NEWPAT that are field assignments. */
3298 {
3301 }
3306 /* If the following substitution will modify I1SRC, make a copy of it
3307 for the case where it is substituted for I1DEST in I2PAT later. */
3311 /* If I0 feeds into I1 and I0DEST is in I0SRC, we need to make a unique
3312 copy of I1SRC each time we substitute it, in order to avoid creating
3313 self-referential RTL when we will be substituting I0SRC for I0DEST
3314 later. */
3319 /* Record whether I1's body now appears within I3's body. */
3321 }
3323 /* Likewise for I0 if we have it. */
3326 {
3333 {
3336 }
3338 /* If the following substitution will modify I0SRC, make a copy of it
3339 for the case where it is substituted for I0DEST in I1PAT later. */
3342 /* And a copy for I0DEST in I2PAT substitution. */
3351 }
3353 /* Fail if an autoincrement side-effect has been duplicated. Be careful
3354 to count all the ways that I2SRC and I1SRC can be used. */
3365 /* Fail if we tried to make a new register. */
3367 /* Fail if we couldn't do something and have a CLOBBER. */
3369 /* Fail if this new pattern is a MULT and we didn't have one before
3370 at the outer level. */
3373 {
3376 }
3378 /* If the actions of the earlier insns must be kept
3379 in addition to substituting them into the latest one,
3380 we must make a new PARALLEL for the latest insn
3381 to hold additional the SETs. */
3384 {
3386 combine_extras++;
3389 {
3395 }
3396 else
3397 {
3402 }
3408 {
3414 }
3416 {
3425 }
3426 }
3428 validate_replacement:
3430 /* Note which hard regs this insn has as inputs. */
3433 /* If recog_for_combine fails, it strips existing clobbers. If we'll
3434 consider splitting this pattern, we might need these clobbers. */
3437 {
3443 }
3445 /* We have recognized nothing yet. */
3448 /* See if this is a PARALLEL of two SETs where one SET's destination is
3449 a register that is unused and this isn't marked as an instruction that
3450 might trap in an EH region. In that case, we just need the other SET.
3451 We prefer this over the PARALLEL.
3453 This can occur when simplifying a divmod insn. We *must* test for this
3454 case here because the code below that splits two independent SETs doesn't
3455 handle this case correctly when it updates the register status.
3457 It's pointless doing this if we originally had two sets, one from
3458 i3, and one from i2. Combining then splitting the parallel results
3459 in the original i2 again plus an invalid insn (which we delete).
3460 The net effect is only to move instructions around, which makes
3461 debug info less accurate. */
3466 {
3477 {
3480 }
3489 {
3495 }
3499 }
3501 /* Is the result of combination a valid instruction? */
3505 /* If we were combining three insns and the result is a simple SET
3506 with no ASM_OPERANDS that wasn't recognized, try to split it into two
3507 insns. There are two ways to do this. It can be split using a
3508 machine-specific method (like when you have an addition of a large
3509 constant) or by combine in the function find_split_point. */
3513 {
3517 /* See if the MD file can split NEWPAT. If it can't, see if letting it
3518 use I2DEST as a scratch register will help. In the latter case,
3519 convert I2DEST to the mode of the source of NEWPAT if we can. */
3523 /* We can only use I2DEST as a scratch reg if it doesn't overlap any
3524 inputs of NEWPAT. */
3526 /* ??? If I2DEST is not safe, and I1DEST exists, then it would be
3527 possible to try that as a scratch reg. This would require adding
3528 more code to make it work though. */
3531 {
3534 /* First try to split using the original register as a
3535 scratch register. */
3539 i2dest)));
3542 /* If that didn't work, try changing the mode of I2DEST if
3543 we can. */
3548 {
3550 rtx ni2dest;
3554 else
3555 {
3558 }
3560 parallel = (gen_rtx_PARALLEL
3564 ni2dest))));
3569 {
3577 }
3578 }
3581 }
3583 /* If recog_for_combine has discarded clobbers, try to use them
3584 again for the split. */
3586 {
3589 }
3592 {
3597 }
3601 {
3611 /* If I2 or I3 has multiple SETs, we won't know how to track
3612 register status, so don't use these insns. If I2's destination
3613 is used between I2 and I3, we also can't use these insns. */
3623 /* It is possible that both insns now set the destination of I3.
3624 If so, we must show an extra use of it. */
3627 {
3646 }
3647 }
3649 /* If we can split it and use I2DEST, go ahead and see if that
3650 helps things be recognized. Verify that none of the registers
3651 are set between I2 and I3. */
3655 /* We need I2DEST in the proper mode. If it is a hard register
3656 or the only use of a pseudo, we can change its mode.
3657 Make sure we don't change a hard register to have a mode that
3658 isn't valid for it, or change the number of registers. */
3665 /* We can't overwrite I2DEST if its value is still used by
3666 NEWPAT. */
3668 {
3677 /* *SPLIT may be part of I2SRC, so make sure we have the
3678 original expression around for later debug processing.
3679 We should not need I2SRC any more in other cases. */
3682 else
3685 /* Get NEWDEST as a register in the proper mode. We have already
3686 validated that we can do this. */
3688 {
3691 else
3692 {
3695 }
3696 }
3698 /* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
3699 an ASHIFT. This can occur if it was inside a PLUS and hence
3700 appeared to be a memory address. This is a kludge. */
3705 {
3708 /* Update split_code because we may not have a multiply
3709 anymore. */
3711 }
3713 /* Similarly for (plus (mult FOO (const_int pow2))). */
3719 {
3723 /* Update split_code because we may not have a multiply
3724 anymore. */
3726 }
3728 #ifdef INSN_SCHEDULING
3729 /* If *SPLIT is a paradoxical SUBREG, when we split it, it should
3730 be written as a ZERO_EXTEND. */
3732 {
3733 /* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
3734 what it really is. */
3739 else
3742 }
3743 #endif
3745 /* Attempt to split binary operators using arithmetic identities. */
3749 {
3756 /* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
3758 {
3763 }
3764 /* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
3771 {
3777 /* Split both "((X op Y) op X) op Y" and
3778 "((X op Y) op Y) op X" as "T op T" where T is
3779 "X op Y". */
3782 {
3787 }
3788 /* Split "((X op X) op Y) op Y)" as "T op T" where
3789 T is "X op Y". */
3791 {
3797 }
3798 }
3799 }
3802 {
3805 }
3809 /* recog_for_combine might have added CLOBBERs to newi2pat.
3810 Make sure NEWPAT does not depend on the clobbered regs. */
3814 {
3817 {
3820 }
3821 }
3823 /* If the split point was a MULT and we didn't have one before,
3824 don't use one now. */
3827 }
3828 }
3830 /* Check for a case where we loaded from memory in a narrow mode and
3831 then sign extended it, but we need both registers. In that case,
3832 we have a PARALLEL with both loads from the same memory location.
3833 We can split this into a load from memory followed by a register-register
3834 copy. This saves at least one insn, more if register allocation can
3835 eliminate the copy.
3837 We cannot do this if the destination of the first assignment is a
3838 condition code register or cc0. We eliminate this case by making sure
3839 the SET_DEST and SET_SRC have the same mode.
3841 We cannot do this if the destination of the second assignment is
3842 a register that we have already assumed is zero-extended. Similarly
3843 for a SUBREG of such a register. */
3878 {
3879 rtx ni2dest;
3893 }
3895 /* Similarly, check for a case where we have a PARALLEL of two independent
3896 SETs but we started with three insns. In this case, we can do the sets
3897 as two separate insns. This case occurs when some SET allows two
3898 other insns to combine, but the destination of that SET is still live.
3900 Also do this if we started with two insns and (at least) one of the
3901 resulting sets is a noop; this noop will be deleted later. */
3920 {
3924 /* Normally, it doesn't matter which of the two is done first,
3925 but the one that references cc0 can't be the second, and
3926 one which uses any regs/memory set in between i2 and i3 can't
3927 be first. The PARALLEL might also have been pre-existing in i3,
3928 so we need to make sure that we won't wrongly hoist a SET to i2
3929 that would conflict with a death note present in there. */
3937 /* If I3 is a jump, ensure that set0 is a jump so that
3938 we do not create invalid RTL. */
3940 )
3941 {
3944 }
3952 /* If I3 is a jump, ensure that set1 is a jump so that
3953 we do not create invalid RTL. */
3955 )
3956 {
3959 }
3960 else
3961 {
3964 }
3969 {
3970 /* recog_for_combine might have added CLOBBERs to newi2pat.
3971 Make sure NEWPAT does not depend on the clobbered regs. */
3973 {
3976 {
3979 {
3982 }
3983 }
3984 }
3987 }
3988 }
3990 /* If it still isn't recognized, fail and change things back the way they
3991 were. */
3993 /* Is the result a reasonable ASM_OPERANDS? */
3995 {
3998 }
4000 /* If we had to change another insn, make sure it is valid also. */
4002 {
4010 {
4013 }
4014 }
4016 /* If I2 is the CC0 setter and I3 is the CC0 user then check whether
4017 they are adjacent to each other or not. */
4019 {
4023 {
4026 }
4027 }
4029 /* Only allow this combination if insn_rtx_costs reports that the
4030 replacement instructions are cheaper than the originals. */
4032 {
4035 }
4038 {
4043 {
4048 /* Temporarily revert mode back. */
4052 {
4053 /* If we used i2dest as a scratch register with a
4054 different mode, substitute it for the original
4055 i2src while its original mode is temporarily
4056 restored, and then clear i2scratch so that we don't
4057 do it again later. */
4059 this_basic_block);
4061 /* Put back the new mode. */
4063 }
4064 else
4065 {
4070 {
4073 }
4074 else
4075 {
4082 }
4084 /* We're dealing with a reg that changed mode but not
4085 meaning, so we want to turn it into a subreg for
4086 the new mode. However, because of REG sharing and
4087 because its mode had already changed, we have to do
4088 it in two steps. First, replace any debug uses of
4089 reg, with its original mode temporarily restored,
4090 with this copy we have created; then, replace the
4091 copy with the SUBREG of the original shared reg,
4092 once again changed to the new mode. */
4094 this_basic_block);
4098 this_basic_block);
4099 }
4100 }
4101 }
4103 /* If we will be able to accept this, we have made a
4104 change to the destination of I3. This requires us to
4105 do a few adjustments. */
4108 {
4111 }
4113 /* We now know that we can do this combination. Merge the insns and
4114 update the status of registers and LOG_LINKS. */
4117 {
4122 /* If any of the notes in OTHER_INSN were REG_DEAD or REG_UNUSED,
4123 ensure that they are still valid. Then add any non-duplicate
4124 notes added by recog_for_combine. */
4126 {
4136 }
4140 NULL_RTX);
4141 }
4144 {
4147 rtx ni2dest;
4149 /* I3 now uses what used to be its destination and which is now
4150 I2's destination. This requires us to do a few adjustments. */
4154 /* We need a LOG_LINK from I3 to I2. But we used to have one,
4155 so we still will.
4157 However, some later insn might be using I2's dest and have
4158 a LOG_LINK pointing at I3. We must remove this link.
4159 The simplest way to remove the link is to point it at I1,
4160 which we know will be a NOTE. */
4162 /* newi2pat is usually a SET here; however, recog_for_combine might
4163 have added some clobbers. */
4166 else
4173 {
4175 {
4181 }
4182 }
4183 }
4185 {
4190 /* Compute which registers we expect to eliminate. newi2pat may be setting
4191 either i3dest or i2dest, so we must check it. */
4194 || !i2dest_killed
4196 /* For i1, we need to compute both local elimination and global
4197 elimination information with respect to newi2pat because i1dest
4198 may be the same as i3dest, in which case newi2pat may be setting
4199 i1dest. Global information is used when distributing REG_DEAD
4200 note for i2 and i3, in which case it does matter if newi2pat sets
4201 i1dest or not.
4203 Local information is used when distributing REG_DEAD note for i1,
4204 in which case it doesn't matter if newi2pat sets i1dest or not.
4205 See PR62151, if we have four insns combination:
4206 i0: r0 <- i0src
4207 i1: r1 <- i1src (using r0)
4208 REG_DEAD (r0)
4209 i2: r0 <- i2src (using r1)
4210 i3: r3 <- i3src (using r0)
4211 ix: using r0
4212 From i1's point of view, r0 is eliminated, no matter if it is set
4213 by newi2pat or not. In other words, REG_DEAD info for r0 in i1
4214 should be discarded.
4216 Note local information only affects cases in forms like "I1->I2->I3",
4217 "I0->I1->I2->I3" or "I0&I1->I2, I2->I3". For other cases like
4218 "I0->I1, I1&I2->I3" or "I1&I2->I3", newi2pat won't set i1dest or
4219 i0dest anyway. */
4221 || !i1dest_killed
4226 /* Same case as i1. */
4233 /* Get the old REG_NOTES and LOG_LINKS from all our insns and
4234 clear them. */
4242 /* Ensure that we do not have something that should not be shared but
4243 occurs multiple times in the new insns. Check this by first
4244 resetting all the `used' flags and then copying anything is shared. */
4268 {
4275 {
4276 /* I2SRC must still be meaningful at this point. Some splitting
4277 operations can invalidate I2SRC, but those operations do not
4278 apply to calls. */
4281 }
4289 }
4294 /* We had one special case above where I2 had more than one set and
4295 we replaced a destination of one of those sets with the destination
4296 of I3. In that case, we have to update LOG_LINKS of insns later
4297 in this basic block. Note that this (expensive) case is rare.
4299 Also, in this case, we must pretend that all REG_NOTEs for I2
4300 actually came from I3, so that REG_UNUSED notes from I2 will be
4301 properly handled. */
4304 {
4313 temp_insn
4323 {
4328 }
4329 else
4332 }
4340 {
4343 this_basic_block);
4346 }
4347 else
4348 {
4351 this_basic_block);
4353 }
4356 {
4361 this_basic_block);
4363 }
4366 {
4371 this_basic_block);
4373 }
4375 /* Get death notes for everything that is now used in either I3 or
4376 I2 and used to die in a previous insn. If we built two new
4377 patterns, move from I1 to I2 then I2 to I3 so that we get the
4378 proper movement on registers that I2 modifies. */
4384 else
4390 /* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
4407 /* Distribute any notes added to I2 or I3 by recog_for_combine. We
4408 know these are REG_UNUSED and want them to go to the desired insn,
4409 so we always pass it as i3. */
4413 NULL_RTX);
4417 NULL_RTX);
4419 /* If I3DEST was used in I3SRC, it really died in I3. We may need to
4420 put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
4421 I3DEST, the death must be somewhere before I2, not I3. If we passed I3
4422 in that case, it might delete I2. Similarly for I2 and I1.
4423 Show an additional death due to the REG_DEAD note we make here. If
4424 we discard it in distribute_notes, we will decrement it again. */
4427 {
4432 else
4435 }
4438 {
4443 else
4446 }
4449 {
4454 else
4457 }
4460 {
4465 else
4468 }
4476 {
4481 /* The insn that used to set this register doesn't exist, and
4482 this life of the register may not exist either. See if one of
4483 I3's links points to an insn that sets I2DEST. If it does,
4484 that is now the last known value for I2DEST. If we don't update
4485 this and I2 set the register to a value that depended on its old
4486 contents, we will get confused. If this insn is used, thing
4487 will be set correctly in combine_instructions. */
4495 /* If the reg formerly set in I2 died only once and that was in I3,
4496 zero its use count so it won't make `reload' do any work. */
4499 && ! i2dest_in_i2src
4502 }
4505 {
4518 && ! i1dest_in_i1src
4521 }
4524 {
4537 && ! i0dest_in_i0src
4540 }
4542 /* Update reg_stat[].nonzero_bits et al for any changes that may have
4543 been made to this insn. The order is important, because newi2pat
4544 can affect nonzero_bits of newpat. */
4548 }
4551 {
4553 {
4556 }
4558 }
4561 {
4563 {
4566 }
4568 }
4571 {
4573 {
4576 }
4578 }
4581 {
4583 {
4586 }
4588 }
4591 {
4593 {
4596 }
4598 }
4600 /* Set new_direct_jump_p if a new return or simple jump instruction
4601 has been created. Adjust the CFG accordingly. */
4603 {
4607 }
4612 {
4615 }
4617 /* A noop might also need cleaning up of CFG, if it comes from the
4618 simplification of a jump. */
4623 {
4626 }
4633 {
4636 }
4638 combine_successes++;
4645 else
4647 }
4648 \f
4649 /* Get a marker for undoing to the current state. */
4653 {
4655 }
4657 /* Undo the modifications up to the marker. */
4659 static void
4661 {
4665 {
4670 {
4685 }
4689 }
4692 }
4694 /* Undo all the modifications recorded in undobuf. */
4696 static void
4698 {
4700 }
4702 /* We've committed to accepting the changes we made. Move all
4703 of the undos to the free list. */
4705 static void
4707 {
4711 {
4715 }
4717 }
4718 \f
4719 /* Find the innermost point within the rtx at LOC, possibly LOC itself,
4720 where we have an arithmetic expression and return that point. LOC will
4721 be inside INSN.
4723 try_combine will call this function to see if an insn can be split into
4724 two insns. */
4728 {
4737 /* First special-case some codes. */
4739 {
4741 #ifdef INSN_SCHEDULING
4742 /* If we are making a paradoxical SUBREG invalid, it becomes a split
4743 point. */
4746 #endif
4750 /* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
4751 using LO_SUM and HIGH. */
4754 {
4762 }
4764 /* If we have a PLUS whose second operand is a constant and the
4765 address is not valid, perhaps will can split it up using
4766 the machine-specific way to split large constants. We use
4767 the first pseudo-reg (one of the virtual regs) as a placeholder;
4768 it will not remain in the result. */
4773 {
4776 subst_insn);
4778 /* This should have produced two insns, each of which sets our
4779 placeholder. If the source of the second is a valid address,
4780 we can make put both sources together and make a split point
4781 in the middle. */
4794 && memory_address_addr_space_p
4797 {
4801 /* Replace the placeholder in SRC2 with SRC1. If we can
4802 find where in SRC2 it was placed, that can become our
4803 split point and we can replace this address with SRC2.
4804 Just try two obvious places. */
4815 {
4818 }
4819 }
4821 /* If that didn't work, perhaps the first operand is complex and
4822 needs to be computed separately, so make a split point there.
4823 This will occur on machines that just support REG + CONST
4824 and have a constant moved through some previous computation. */
4830 }
4832 /* If we have a PLUS whose first operand is complex, try computing it
4833 separately by making a split there. */
4844 /* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
4845 ZERO_EXTRACT, the most likely reason why this doesn't match is that
4846 we need to put the operand into a register. So split at that
4847 point. */
4857 /* See if we can split SET_SRC as it stands. */
4862 /* See if we can split SET_DEST as it stands. */
4867 /* See if this is a bitfield assignment with everything constant. If
4868 so, this is an IOR of an AND, so split it into that. */
4878 {
4884 unsigned HOST_WIDE_INT mask
4886 rtx or_mask;
4895 else
4896 {
4902 or_mask));
4903 }
4910 }
4912 /* Otherwise, see if this is an operation that we can split into two.
4913 If so, try to split that. */
4917 {
4919 /* If we are AND'ing with a large constant that is only a single
4920 bit and the result is only being used in a context where we
4921 need to know if it is zero or nonzero, replace it with a bit
4922 extraction. This will avoid the large constant, which might
4923 have taken more than one insn to make. If the constant were
4924 not a valid argument to the AND but took only one insn to make,
4925 this is no worse, but if it took more than one insn, it will
4926 be better. */
4936 {
4941 {
4944 }
4945 }
4949 /* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
4950 is known to be on, this can be converted into a NEG of a shift. */
4956 {
4968 }
4974 /* We can't optimize if either mode is a partial integer
4975 mode as we don't know how many bits are significant
4976 in those modes. */
4990 {
4998 }
5003 }
5007 {
5010 /* For unsigned, we have a choice of a shift followed by an
5011 AND or two shifts. Use two shifts for field sizes where the
5012 constant might be too large. We assume here that we can
5013 always at least get 8-bit constants in an AND insn, which is
5014 true for every current RISC. */
5017 {
5018 unsigned HOST_WIDE_INT mask
5022 gen_rtx_LSHIFTRT
5030 }
5031 else
5032 {
5034 gen_rtx_fmt_ee
5045 }
5046 }
5048 /* See if this is a simple operation with a constant as the second
5049 operand. It might be that this constant is out of range and hence
5050 could be used as a split point. */
5058 /* Finally, see if this is a simple operation with its first operand
5059 not in a register. The operation might require this operand in a
5060 register, so return it as a split point. We can always do this
5061 because if the first operand were another operation, we would have
5062 already found it as a split point. */
5071 /* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
5072 it is better to write this as (not (ior A B)) so we can split it.
5073 Similarly for IOR. */
5075 {
5083 }
5085 /* Many RISC machines have a large set of logical insns. If the
5086 second operand is a NOT, put it first so we will try to split the
5087 other operand first. */
5089 {
5093 }
5098 /* Canonicalization can produce (minus A (mult B C)), where C is a
5099 constant. It may be better to try splitting (plus (mult B -C) A)
5100 instead if this isn't a multiply by a power of two. */
5104 {
5112 mode)),
5115 }
5117 /* Split at a multiply-accumulate instruction. However if this is
5118 the SET_SRC, we likely do not have such an instruction and it's
5119 worthless to try this split. */
5128 }
5130 /* Otherwise, select our actions depending on our rtx class. */
5132 {
5138 /* ... fall through ... */
5146 /* ... fall through ... */
5148 /* Some machines have (and (shift ...) ...) insns. If X is not
5149 an AND, but XEXP (X, 0) is, use it as our split point. */
5159 /* Otherwise, we don't have a split point. */
5161 }
5162 }
5163 \f
5164 /* Throughout X, replace FROM with TO, and return the result.
5165 The result is TO if X is FROM;
5166 otherwise the result is X, but its contents may have been modified.
5167 If they were modified, a record was made in undobuf so that
5168 undo_all will (among other things) return X to its original state.
5170 If the number of changes necessary is too much to record to undo,
5171 the excess changes are not made, so the result is invalid.
5172 The changes already made can still be undone.
5173 undobuf.num_undo is incremented for such changes, so by testing that
5174 the caller can tell whether the result is valid.
5176 `n_occurrences' is incremented each time FROM is replaced.
5178 IN_DEST is nonzero if we are processing the SET_DEST of a SET.
5180 IN_COND is nonzero if we are at the top level of a condition.
5182 UNIQUE_COPY is nonzero if each substitution must be unique. We do this
5183 by copying if `n_occurrences' is nonzero. */
5185 static rtx
5187 {
5192 rtx new_rtx;
5194 /* Two expressions are equal if they are identical copies of a shared
5195 RTX or if they are both registers with the same register number
5196 and mode. */
5198 #define COMBINE_RTX_EQUAL_P(X,Y) \
5199 ((X) == (Y) \
5200 || (REG_P (X) && REG_P (Y) \
5201 && REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
5203 /* Do not substitute into clobbers of regs -- this will never result in
5204 valid RTL. */
5209 {
5210 n_occurrences++;
5212 }
5214 /* If X and FROM are the same register but different modes, they
5215 will not have been seen as equal above. However, the log links code
5216 will make a LOG_LINKS entry for that case. If we do nothing, we
5217 will try to rerecognize our original insn and, when it succeeds,
5218 we will delete the feeding insn, which is incorrect.
5220 So force this insn not to match in this (rare) case. */
5225 /* If this is an object, we are done unless it is a MEM or LO_SUM, both
5226 of which may contain things that can be combined. */
5230 /* It is possible to have a subexpression appear twice in the insn.
5231 Suppose that FROM is a register that appears within TO.
5232 Then, after that subexpression has been scanned once by `subst',
5233 the second time it is scanned, TO may be found. If we were
5234 to scan TO here, we would find FROM within it and create a
5235 self-referent rtl structure which is completely wrong. */
5239 /* Parallel asm_operands need special attention because all of the
5240 inputs are shared across the arms. Furthermore, unsharing the
5241 rtl results in recognition failures. Failure to handle this case
5242 specially can result in circular rtl.
5244 Solve this by doing a normal pass across the first entry of the
5245 parallel, and only processing the SET_DESTs of the subsequent
5246 entries. Ug. */
5251 {
5254 /* If this substitution failed, this whole thing fails. */
5262 {
5268 {
5271 /* If this substitution failed, this whole thing fails. */
5277 }
5278 }
5279 }
5280 else
5281 {
5285 /* We don't need to process a SET_DEST that is a register, CC0,
5286 or PC, so set up to skip this common case. All other cases
5287 where we want to suppress replacing something inside a
5288 SET_SRC are handled via the IN_DEST operand. */
5295 /* Trying to simplify the operands of a widening MULT is not likely
5296 to create RTL matching a machine insn. */
5308 /* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
5309 constant. */
5314 {
5316 {
5319 {
5321 {
5324 n_occurrences++;
5325 }
5326 else
5327 {
5329 unique_copy);
5331 /* If this substitution failed, this whole thing
5332 fails. */
5336 }
5339 }
5340 }
5342 {
5343 /* If this is a register being set, ignore it. */
5350 ;
5353 {
5354 /* In general, don't install a subreg involving two
5355 modes not tieable. It can worsen register
5356 allocation, and can even make invalid reload
5357 insns, since the reg inside may need to be copied
5358 from in the outside mode, and that may be invalid
5359 if it is an fp reg copied in integer mode.
5361 We allow two exceptions to this: It is valid if
5362 it is inside another SUBREG and the mode of that
5363 SUBREG and the mode of the inside of TO is
5364 tieable and it is valid if X is a SET that copies
5365 FROM to CC0. */
5373 && (!HAVE_cc0
5388 n_occurrences++;
5389 }
5390 else
5391 /* If we are in a SET_DEST, suppress most cases unless we
5392 have gone inside a MEM, in which case we want to
5393 simplify the address. We assume here that things that
5394 are actually part of the destination have their inner
5395 parts in the first expression. This is true for SUBREG,
5396 STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
5397 things aside from REG and MEM that should appear in a
5398 SET_DEST. */
5400 (((in_dest
5406 unique_copy);
5408 /* If we found that we will have to reject this combination,
5409 indicate that by returning the CLOBBER ourselves, rather than
5410 an expression containing it. This will speed things up as
5411 well as prevent accidents where two CLOBBERs are considered
5412 to be equal, thus producing an incorrect simplification. */
5418 {
5426 }
5429 {
5433 }
5434 else
5436 }
5437 }
5438 }
5440 /* Check if we are loading something from the constant pool via float
5441 extension; in this case we would undo compress_float_constant
5442 optimization and degenerate constant load to an immediate value. */
5446 {
5450 }
5452 /* Try to simplify X. If the simplification changed the code, it is likely
5453 that further simplification will help, so loop, but limit the number
5454 of repetitions that will be performed. */
5457 {
5458 /* If X is sufficiently simple, don't bother trying to do anything
5459 with it. */
5468 /* We no longer know the original mode of operand 0 since we
5469 have changed the form of X) */
5471 }
5474 }
5475 \f
5476 /* Simplify X, a piece of RTL. We just operate on the expression at the
5477 outer level; call `subst' to simplify recursively. Return the new
5478 expression.
5480 OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
5481 if we are inside a SET_DEST. IN_COND is nonzero if we are at the top level
5482 of a condition. */
5484 static rtx
5487 {
5490 rtx temp;
5493 /* If this is a commutative operation, put a constant last and a complex
5494 expression first. We don't need to do this for comparisons here. */
5497 {
5501 }
5503 /* Try to fold this expression in case we have constants that weren't
5504 present before. */
5507 {
5515 {
5518 {
5522 }
5525 }
5538 }
5541 {
5546 }
5548 /* If this is a simple operation applied to an IF_THEN_ELSE, try
5549 applying it to the arms of the IF_THEN_ELSE. This often simplifies
5550 things. Check for cases where both arms are testing the same
5551 condition.
5553 Don't do anything if all operands are very simple. */
5566 {
5571 /* If everything is a comparison, what we have is highly unlikely
5572 to be simpler, so don't use it. */
5575 {
5582 /* Simplify the alternative arms; this may collapse the true and
5583 false arms to store-flag values. Be careful to use copy_rtx
5584 here since true_rtx or false_rtx might share RTL with x as a
5585 result of the if_then_else_cond call above. */
5589 /* If true_rtx and false_rtx are not general_operands, an if_then_else
5590 is unlikely to be simpler. */
5593 {
5596 /* Restarting if we generate a store-flag expression will cause
5597 us to loop. Just drop through in this case. */
5599 /* If the result values are STORE_FLAG_VALUE and zero, we can
5600 just make the comparison operation. */
5611 /* Likewise, we can make the negate of a comparison operation
5612 if the result values are - STORE_FLAG_VALUE and zero. */
5620 mode);
5631 mode);
5632 else
5635 mode,
5636 VOIDmode,
5637 cond,
5638 cop1),
5643 }
5644 }
5645 }
5647 /* First see if we can apply the inverse distributive law. */
5650 {
5654 }
5656 /* If CODE is an associative operation not otherwise handled, see if we
5657 can associate some operands. This can win if they are constants or
5658 if they are logically related (i.e. (a & b) & a). */
5664 {
5666 {
5670 rtx inner;
5672 /* Make sure we pass the constant operand if any as the second
5673 one if this is a commutative operation. */
5681 /* For commutative operations, try the other pair if that one
5682 didn't simplify. */
5684 {
5689 }
5693 }
5694 }
5696 /* A little bit of algebraic simplification here. */
5698 {
5700 /* Ensure that our address has any ASHIFTs converted to MULT in case
5701 address-recognizing predicates are called later. */
5710 /* See if this can be moved to simplify_subreg. */
5713 /* Don't call gen_lowpart if the inner mode
5714 is VOIDmode and we cannot simplify it, as SUBREG without
5715 inner mode is invalid. */
5722 {
5723 rtx temp;
5729 /* If op is known to have all lower bits zero, the result is zero. */
5739 }
5741 /* Don't change the mode of the MEM if that would change the meaning
5742 of the address. */
5749 /* Note that we cannot do any narrowing for non-constants since
5750 we might have been counting on using the fact that some bits were
5751 zero. We now do this in the SET. */
5758 /* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
5759 replaced by (lshiftrt X C). This will convert
5760 (neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
5768 /* If X has only a single bit that might be nonzero, say, bit I, convert
5769 (neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
5770 MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
5771 (sign_extract X 1 Y). But only do this if TEMP isn't a register
5772 or a SUBREG of one since we'd be making the expression more
5773 complex if it was just a register. */
5779 {
5780 rtx temp1 = simplify_shift_const
5786 /* If all we did was surround TEMP with the two shifts, we
5787 haven't improved anything, so don't use it. Otherwise,
5788 we are better off with TEMP1. */
5793 }
5797 /* We can't handle truncation to a partial integer mode here
5798 because we don't know the real bitsize of the partial
5799 integer mode. */
5808 /* We can truncate a constant value and return it. */
5812 /* Similarly to what we do in simplify-rtx.c, a truncate of a register
5813 whose value is a comparison can be replaced with a subreg if
5814 STORE_FLAG_VALUE permits. */
5823 /* (const (const X)) can become (const X). Do it this way rather than
5824 returning the inner CONST since CONST can be shared with a
5825 REG_EQUAL note. */
5831 /* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
5832 can add in an offset. find_split_point will split this address up
5833 again if it doesn't match. */
5840 /* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
5841 when c is (const_int (pow2 + 1) / 2) is a sign extension of a
5842 bit-field and can be replaced by either a sign_extend or a
5843 sign_extract. The `and' may be a zero_extend and the two
5844 <c>, -<c> constants may be reversed. */
5859 return simplify_shift_const
5866 /* If only the low-order bit of X is possibly nonzero, (plus x -1)
5867 can become (ashiftrt (ashift (xor x 1) C) C) where C is
5868 the bitsize of the mode - 1. This allows simplification of
5869 "a = (b & 8) == 0;" */
5881 /* If we are adding two things that have no bits in common, convert
5882 the addition into an IOR. This will often be further simplified,
5883 for example in cases like ((a & 1) + (a & 2)), which can
5884 become a & 3. */
5889 {
5890 /* Try to simplify the expression further. */
5894 /* If we could, great. If not, do not go ahead with the IOR
5895 replacement, since PLUS appears in many special purpose
5896 address arithmetic instructions. */
5904 }
5906 /* Canonicalize x + x into x << 1. */
5915 /* (minus <foo> (and <foo> (const_int -pow2))) becomes
5916 (and <foo> (const_int pow2-1)) */
5926 /* If we have (mult (plus A B) C), apply the distributive law and then
5927 the inverse distributive law to see if things simplify. This
5928 occurs mostly in addresses, often when unrolling loops. */
5931 {
5935 }
5937 /* Try simplify a*(b/c) as (a*b)/c. */
5940 {
5946 }
5950 /* If this is a divide by a power of two, treat it as a shift if
5951 its first operand is a shift. */
5969 /* If the first operand is a condition code, we can't do anything
5970 with it. */
5974 {
5982 /* Simplify our comparison, if possible. */
5985 /* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
5986 if only the low-order bit is possibly nonzero in X (such as when
5987 X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
5988 (xor X 1) or (minus 1 X); we use the former. Finally, if X is
5989 known to be either 0 or -1, NE becomes a NEG and EQ becomes
5990 (plus X 1).
5992 Remove any ZERO_EXTRACT we made when thinking this was a
5993 comparison. It may now be simpler to use, e.g., an AND. If a
5994 ZERO_EXTRACT is indeed appropriate, it will be placed back by
5995 the call to make_compound_operation in the SET case.
5997 Don't apply these optimizations if the caller would
5998 prefer a comparison rather than a value.
5999 E.g., for the condition in an IF_THEN_ELSE most targets need
6000 an explicit comparison. */
6003 ;
6019 {
6023 mode);
6024 }
6031 {
6035 const1_rtx);
6036 }
6044 {
6047 }
6049 /* If STORE_FLAG_VALUE is -1, we have cases similar to
6050 those above. */
6052 ;
6068 {
6072 mode);
6073 }
6081 {
6085 mode);
6086 }
6088 /* If X is 0/1, (eq X 0) is X-1. */
6094 {
6097 }
6099 /* If STORE_FLAG_VALUE says to just test the sign bit and X has just
6100 one bit that might be nonzero, we can convert (ne x 0) to
6101 (ashift x c) where C puts the bit in the sign bit. Remove any
6102 AND with STORE_FLAG_VALUE when we are done, since we are only
6103 going to test the sign bit. */
6110 {
6116 else
6118 }
6120 /* If the code changed, return a whole new comparison.
6121 We also need to avoid using SUBST in cases where
6122 simplify_comparison has widened a comparison with a CONST_INT,
6123 since in that case the wider CONST_INT may fail the sanity
6124 checks in do_SUBST. */
6131 /* Otherwise, keep this operation, but maybe change its operands.
6132 This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
6135 }
6145 /* If we are processing SET_DEST, we are done. */
6163 /* If this is a shift by a constant amount, simplify it. */
6179 }
6182 }
6183 \f
6184 /* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
6186 static rtx
6188 {
6195 rtx temp;
6198 rtx reversed;
6200 /* Simplify storing of the truth value. */
6205 /* Also when the truth value has to be reversed. */
6211 /* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
6212 in it is being compared against certain values. Get the true and false
6213 comparisons and see if that says anything about the value of each arm. */
6219 {
6220 HOST_WIDE_INT nzb;
6226 /* If FALSE_CODE is EQ, swap the codes and arms. */
6229 {
6232 }
6234 /* If we are comparing against zero and the expression being tested has
6235 only a single bit that might be nonzero, that is its value when it is
6236 not equal to zero. Similarly if it is known to be -1 or 0. */
6240 {
6243 }
6247 {
6250 }
6252 /* Now simplify an arm if we know the value of the register in the
6253 branch and it is used in the arm. Be careful due to the potential
6254 of locally-shared RTL. */
6271 }
6273 /* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
6274 reversed, do so to avoid needing two sets of patterns for
6275 subtract-and-branch insns. Similarly if we have a constant in the true
6276 arm, the false arm is the same as the first operand of the comparison, or
6277 the false arm is more complicated than the true arm. */
6290 {
6299 /* It is possible that the conditional has been simplified out. */
6302 }
6304 /* If the two arms are identical, we don't need the comparison. */
6309 /* Convert a == b ? b : a to "a". */
6321 /* Look for cases where we have (abs x) or (neg (abs X)). */
6324 && comparison_p
6331 {
6337 return
6340 mode);
6343 }
6345 /* Look for MIN or MAX. */
6348 && comparison_p
6353 {
6368 }
6370 /* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
6371 second operand is zero, this can be done as (OP Z (mult COND C2)) where
6372 C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
6373 SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
6374 We can do this kind of thing in some cases when STORE_FLAG_VALUE is
6375 neither 1 or -1, but it isn't worth checking for. */
6378 && comparison_p
6381 {
6397 /* If an identity-zero op is commutative, check whether there
6398 would be a match if we swapped the operands. */
6418 {
6422 }
6434 {
6438 }
6454 {
6458 }
6470 {
6474 }
6477 {
6483 const_true_rtx));
6491 }
6492 }
6494 /* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
6495 1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
6496 negation of a single bit, we can convert this operation to a shift. We
6497 can actually do this more generally, but it doesn't seem worth it. */
6506 return
6510 /* (IF_THEN_ELSE (NE REG 0) (0) (8)) is REG for nonzero_bits (REG) == 8. */
6520 }
6521 \f
6522 /* Simplify X, a SET expression. Return the new expression. */
6524 static rtx
6526 {
6529 machine_mode mode
6534 /* (set (pc) (return)) gets written as (return). */
6538 /* Now that we know for sure which bits of SRC we are using, see if we can
6539 simplify the expression for the object knowing that we only need the
6540 low-order bits. */
6543 {
6546 }
6548 /* If we are setting CC0 or if the source is a COMPARE, look for the use of
6549 the comparison result and try to simplify it unless we already have used
6550 undobuf.other_insn. */
6558 {
6567 {
6570 {
6573 }
6574 }
6575 else
6583 {
6587 }
6588 else
6589 {
6594 /* Attempt to simplify CC user. */
6596 {
6600 }
6602 /* Convert X into a no-op move. */
6606 }
6608 /* Simplify our comparison, if possible. */
6611 #ifdef SELECT_CC_MODE
6612 /* If this machine has CC modes other than CCmode, check to see if we
6613 need to use a different CC mode here. */
6622 else
6625 /* If the mode changed, we have to change SET_DEST, the mode in the
6626 compare, and the mode in the place SET_DEST is used. If SET_DEST is
6627 a hard register, just build new versions with the proper mode. If it
6628 is a pseudo, we lose unless it is only time we set the pseudo, in
6629 which case we can safely change its mode. */
6631 {
6633 {
6635 rtx new_dest;
6639 else
6640 {
6643 }
6650 }
6651 }
6654 /* If the code changed, we have to build a new comparison in
6655 undobuf.other_insn. */
6657 {
6666 /* If the only change we made was to change an EQ into an NE or
6667 vice versa, OP0 has only one bit that might be nonzero, and OP1
6668 is zero, check if changing the user of the condition code will
6669 produce a valid insn. If it won't, we can keep the original code
6670 in that insn by surrounding our operation with an XOR. */
6677 {
6682 {
6689 }
6690 }
6691 }
6696 /* Don't generate a compare of a CC with 0, just use that CC. */
6698 {
6701 }
6702 /* Otherwise, if we didn't previously have the same COMPARE we
6703 want, create it from scratch. */
6706 {
6709 }
6710 }
6711 else
6712 {
6713 /* Get SET_SRC in a form where we have placed back any
6714 compound expressions. Then do the checks below. */
6717 }
6719 /* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
6720 and X being a REG or (subreg (reg)), we may be able to convert this to
6721 (set (subreg:m2 x) (op)).
6723 We can always do this if M1 is narrower than M2 because that means that
6724 we only care about the low bits of the result.
6726 However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
6727 perform a narrower operation than requested since the high-order bits will
6728 be undefined. On machine where it is defined, this transformation is safe
6729 as long as M1 and M2 have the same number of words. */
6737 && (WORD_REGISTER_OPERATIONS
6740 #ifdef CANNOT_CHANGE_MODE_CLASS
6745 #endif
6749 {
6752 dest));
6756 }
6758 /* If we have (set (cc0) (subreg ...)), we try to remove the subreg
6759 in SRC. */
6765 {
6769 /* Here we make sure that we don't have a sign bit on. */
6772 {
6775 }
6776 }
6778 /* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
6779 would require a paradoxical subreg. Replace the subreg with a
6780 zero_extend to avoid the reload that would otherwise be required. */
6788 {
6794 }
6796 /* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
6797 are comparing an item known to be 0 or -1 against 0, use a logical
6798 operation instead. Check for one of the arms being an IOR of the other
6799 arm with some value. We compute three terms to be IOR'ed together. In
6800 practice, at most two will be nonzero. Then we do the IOR's. */
6808 && (!HAVE_conditional_move
6814 {
6840 false_rtx);
6846 term3));
6849 }
6851 /* If either SRC or DEST is a CLOBBER of (const_int 0), make this
6852 whole thing fail. */
6857 else
6858 /* Convert this into a field assignment operation, if possible. */
6860 }
6861 \f
6862 /* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
6863 result. */
6865 static rtx
6867 {
6873 {
6875 /* We can call simplify_and_const_int only if we don't lose
6876 any (sign) bits when converting INTVAL (op1) to
6877 "unsigned HOST_WIDE_INT". */
6881 {
6888 }
6890 /* If we have any of (and (ior A B) C) or (and (xor A B) C),
6891 apply the distributive law and then the inverse distributive
6892 law to see if things simplify. */
6894 {
6898 }
6900 {
6904 }
6908 /* If we have (ior (and A B) C), apply the distributive law and then
6909 the inverse distributive law to see if things simplify. */
6912 {
6916 }
6919 {
6923 }
6928 }
6931 }
6932 \f
6933 /* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
6934 operations" because they can be replaced with two more basic operations.
6935 ZERO_EXTEND is also considered "compound" because it can be replaced with
6936 an AND operation, which is simpler, though only one operation.
6938 The function expand_compound_operation is called with an rtx expression
6939 and will convert it to the appropriate shifts and AND operations,
6940 simplifying at each stage.
6942 The function make_compound_operation is called to convert an expression
6943 consisting of shifts and ANDs into the equivalent compound expression.
6944 It is the inverse of this function, loosely speaking. */
6946 static rtx
6948 {
6952 rtx tem;
6955 {
6959 /* We can't necessarily use a const_int for a multiword mode;
6960 it depends on implicitly extending the value.
6961 Since we don't know the right way to extend it,
6962 we can't tell whether the implicit way is right.
6964 Even for a mode that is no wider than a const_int,
6965 we can't win, because we need to sign extend one of its bits through
6966 the rest of it, and we don't know which bit. */
6970 /* Return if (subreg:MODE FROM 0) is not a safe replacement for
6971 (zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
6972 because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
6973 reloaded. If not for that, MEM's would very rarely be safe.
6975 Reject MODEs bigger than a word, because we might not be able
6976 to reference a two-register group starting with an arbitrary register
6977 (and currently gen_lowpart might crash for a SUBREG). */
6982 /* Reject MODEs that aren't scalar integers because turning vector
6983 or complex modes into shifts causes problems. */
6989 /* If the inner object has VOIDmode (the only way this can happen
6990 is if it is an ASM_OPERANDS), we can't do anything since we don't
6991 know how much masking to do. */
7000 /* ... fall through ... */
7003 /* If the operand is a CLOBBER, just return it. */
7012 /* Reject MODEs that aren't scalar integers because turning vector
7013 or complex modes into shifts causes problems. */
7021 /* This should stay within the object being extracted, fail otherwise. */
7032 }
7033 /* Convert sign extension to zero extension, if we know that the high
7034 bit is not set, as this is easier to optimize. It will be converted
7035 back to cheaper alternative in make_extraction. */
7043 {
7048 /* Make sure this is a profitable operation. */
7055 else
7057 }
7059 /* We can optimize some special cases of ZERO_EXTEND. */
7061 {
7062 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
7063 know that the last value didn't have any inappropriate bits
7064 set. */
7072 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7081 /* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
7082 is a comparison and STORE_FLAG_VALUE permits. This is like
7083 the first case, but it works even when GET_MODE (x) is larger
7084 than HOST_WIDE_INT. */
7093 /* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
7103 }
7105 /* If we reach here, we want to return a pair of shifts. The inner
7106 shift is a left shift of BITSIZE - POS - LEN bits. The outer
7107 shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
7108 logical depending on the value of UNSIGNEDP.
7110 If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
7111 converted into an AND of a shift.
7113 We must check for the case where the left shift would have a negative
7114 count. This can happen in a case like (x >> 31) & 255 on machines
7115 that can't shift by a constant. On those machines, we would first
7116 combine the shift with the AND to produce a variable-position
7117 extraction. Then the constant of 31 would be substituted in
7118 to produce such a position. */
7122 {
7131 }
7138 else
7139 /* Any other cases we can't handle. */
7142 /* If we couldn't do this for some reason, return the original
7143 expression. */
7148 }
7149 \f
7150 /* X is a SET which contains an assignment of one object into
7151 a part of another (such as a bit-field assignment, STRICT_LOW_PART,
7152 or certain SUBREGS). If possible, convert it into a series of
7153 logical operations.
7155 We half-heartedly support variable positions, but do not at all
7156 support variable lengths. */
7158 static const_rtx
7160 {
7161 rtx inner;
7165 machine_mode compute_mode;
7167 /* Loop until we find something we can't simplify. */
7169 {
7172 {
7176 }
7179 {
7184 /* A constant position should stay within the width of INNER. */
7190 {
7198 /* If position is ADJUST - X, new position is X. */
7200 else
7201 {
7206 pos);
7207 }
7208 }
7209 }
7211 /* A SUBREG between two modes that occupy the same numbers of words
7212 can be done by moving the SUBREG to the source. */
7214 /* We need SUBREGs to compute nonzero_bits properly. */
7215 && nonzero_sign_valid
7220 {
7222 gen_lowpart
7226 }
7227 else
7235 /* Don't attempt bitwise arithmetic on non scalar integer modes. */
7237 {
7238 machine_mode imode;
7240 /* Don't do anything for vector or complex integral types. */
7244 /* Try to find an integral mode to pun with. */
7251 }
7253 /* Compute a mask of LEN bits, if we can do this on the host machine. */
7257 /* Don't try to compute in too wide unsupported modes. */
7261 /* Now compute the equivalent expression. Make a copy of INNER
7262 for the SET_DEST in case it is a MEM into which we will substitute;
7263 we don't want shared RTL in that case. */
7265 compute_mode);
7269 compute_mode,
7271 compute_mode),
7272 inner);
7277 mask),
7278 pos);
7283 }
7286 }
7287 \f
7288 /* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
7289 it is an RTX that represents the (variable) starting position; otherwise,
7290 POS is the (constant) starting bit position. Both are counted from the LSB.
7292 UNSIGNEDP is nonzero for an unsigned reference and zero for a signed one.
7294 IN_DEST is nonzero if this is a reference in the destination of a SET.
7295 This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
7296 a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
7297 be used.
7299 IN_COMPARE is nonzero if we are in a COMPARE. This means that a
7300 ZERO_EXTRACT should be built even for bits starting at bit 0.
7302 MODE is the desired mode of the result (if IN_DEST == 0).
7304 The result is an RTX for the extraction or NULL_RTX if the target
7305 can't handle it. */
7307 static rtx
7311 {
7312 /* This mode describes the size of the storage area
7313 to fetch the overall value from. Within that, we
7314 ignore the POS lowest bits, etc. */
7316 machine_mode inner_mode;
7317 machine_mode wanted_inner_mode;
7324 HOST_WIDE_INT orig_pos;
7330 {
7331 /* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
7332 consider just the QI as the memory to extract from.
7333 The subreg adds or removes high bits; its mode is
7334 irrelevant to the meaning of this extraction,
7335 since POS and LEN count from the lsb. */
7339 }
7344 {
7345 /* We're extracting the least significant bits of an rtx
7346 (ashift X (const_int C)), where LEN > C. Extract the
7347 least significant (LEN - C) bits of X, giving an rtx
7348 whose mode is MODE, then shift it left C times. */
7354 }
7360 /* See if this can be done without an extraction. We never can if the
7361 width of the field is not the same as that of some integer mode. For
7362 registers, we can only avoid the extraction if the position is at the
7363 low-order bit and this is either not in the destination or we have the
7364 appropriate STRICT_LOW_PART operation available.
7366 For MEM, we can avoid an extract if the field starts on an appropriate
7367 boundary and we can change the mode of the memory reference. */
7377 && (! in_dest
7381 && (pos
7384 /* We can't do this if we are widening INNER_MODE (it
7385 may not be aligned, for one thing). */
7391 {
7392 /* If INNER is a MEM, make a new MEM that encompasses just the desired
7393 field. If the original and current mode are the same, we need not
7394 adjust the offset. Otherwise, we do if bytes big endian.
7396 If INNER is not a MEM, get a piece consisting of just the field
7397 of interest (in this case POS % BITS_PER_WORD must be 0). */
7400 {
7401 HOST_WIDE_INT offset;
7403 /* POS counts from lsb, but make OFFSET count in memory order. */
7406 else
7410 }
7412 {
7414 {
7415 /* We can't call gen_lowpart in a DEST since we
7416 always want a SUBREG (see below) and it would sometimes
7417 return a new hard register. */
7419 {
7434 /* Avoid creating invalid subregs, for example when
7435 simplifying (x>>32)&255. */
7440 }
7441 else
7443 }
7444 else
7446 }
7447 else
7449 len >= HOST_BITS_PER_WIDE_INT
7453 /* If this extraction is going into the destination of a SET,
7454 make a STRICT_LOW_PART unless we made a MEM. */
7469 /* If we know that no extraneous bits are set, and that the high
7470 bit is not set, convert the extraction to the cheaper of
7471 sign and zero extension, that are equivalent in these cases. */
7477 {
7481 /* Prefer ZERO_EXTENSION, since it gives more information to
7482 backends. */
7487 }
7489 /* Otherwise, sign- or zero-extend unless we already are in the
7490 proper mode. */
7494 }
7496 /* Unless this is a COMPARE or we have a funny memory reference,
7497 don't do anything with zero-extending field extracts starting at
7498 the low-order bit since they are simple AND operations. */
7503 /* Unless INNER is not MEM, reject this if we would be spanning bytes or
7504 if the position is not a constant and the length is not 1. In all
7505 other cases, we would only be going outside our object in cases when
7506 an original shift would have been undefined. */
7515 /* If INNER is not from memory, we want it to have the mode of a register
7516 extraction pattern's structure operand, or word_mode if there is no
7517 such pattern. The same applies to extraction_mode and pos_mode
7518 and their respective operands.
7520 For memory, assume that the desired extraction_mode and pos_mode
7521 are the same as for a register operation, since at present we don't
7522 have named patterns for aligned memory structures. */
7526 {
7530 }
7532 /* Never narrow an object, since that might not be safe. */
7540 else
7541 {
7542 /* Be careful not to go beyond the extracted object and maintain the
7543 natural alignment of the memory. */
7547 {
7550 }
7551 }
7556 {
7557 /* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
7558 BITS_BIG_ENDIAN style. If position is constant, compute new
7559 position. Otherwise, build subtraction.
7560 Note that POS is relative to the mode of the original argument.
7561 If it's a MEM we need to recompute POS relative to that.
7562 However, if we're extracting from (or inserting into) a register,
7563 we want to recompute POS relative to wanted_inner_mode. */
7570 else
7571 pos_rtx
7574 pos_rtx);
7575 /* POS may be less than 0 now, but we check for that below.
7576 Note that it can only be less than 0 if !MEM_P (inner). */
7577 }
7579 /* If INNER has a wider mode, and this is a constant extraction, try to
7580 make it smaller and adjust the byte to point to the byte containing
7581 the value. */
7584 && ! pos_rtx
7589 {
7592 /* The computations below will be correct if the machine is big
7593 endian in both bits and bytes or little endian in bits and bytes.
7594 If it is mixed, we must adjust. */
7596 /* If bytes are big endian and we had a paradoxical SUBREG, we must
7597 adjust OFFSET to compensate. */
7602 /* We can now move to the desired byte. */
7613 }
7615 /* If INNER is not memory, get it into the proper mode. If we are changing
7616 its mode, POS must be a constant and smaller than the size of the new
7617 mode. */
7619 {
7620 /* On the LHS, don't create paradoxical subregs implicitely truncating
7621 the register unless TRULY_NOOP_TRUNCATION. */
7624 wanted_inner_mode))
7636 pos_rtx
7642 }
7644 /* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
7645 have to zero extend. Otherwise, we can just use a SUBREG. */
7648 {
7652 /* If we know that no extraneous bits are set, and that the high
7653 bit is not set, convert extraction to cheaper one - either
7654 SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
7655 cases. */
7663 {
7667 /* Prefer ZERO_EXTENSION, since it gives more information to
7668 backends. */
7672 }
7674 }
7676 /* Make POS_RTX unless we already have it and it is correct. If we don't
7677 have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
7678 be a CONST_INT. */
7685 /* Make the required operation. See if we can use existing rtx. */
7692 }
7693 \f
7694 /* See if X contains an ASHIFT of COUNT or more bits that can be commuted
7695 with any other operations in X. Return X without that shift if so. */
7697 static rtx
7699 {
7702 rtx tem;
7705 {
7707 /* This is the shift itself. If it is wide enough, we will return
7708 either the value being shifted if the shift count is equal to
7709 COUNT or a shift for the difference. */
7723 /* If we can safely shift this constant and we find the inner shift,
7724 make a new operation. */
7729 {
7733 }
7738 }
7741 }
7742 \f
7743 /* Look at the expression rooted at X. Look for expressions
7744 equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
7745 Form these expressions.
7747 Return the new rtx, usually just X.
7749 Also, for machines like the VAX that don't have logical shift insns,
7750 try to convert logical to arithmetic shift operations in cases where
7751 they are equivalent. This undoes the canonicalizations to logical
7752 shifts done elsewhere.
7754 We try, as much as possible, to re-use rtl expressions to save memory.
7756 IN_CODE says what kind of expression we are processing. Normally, it is
7757 SET. In a memory address it is MEM. When processing the arguments of
7758 a comparison or a COMPARE against zero, it is COMPARE, or EQ if more
7759 precisely it is an equality comparison against zero. */
7761 rtx
7763 {
7771 rtx tem;
7775 /* Select the code to be used in recursive calls. Once we are inside an
7776 address, we stay there. If we have a comparison, set to COMPARE,
7777 but once inside, go back to our default of SET. */
7780 {
7783 }
7789 /* Process depending on the code of this operation. If NEW is set
7790 nonzero, it will be returned. */
7793 {
7795 /* Convert shifts by constants into multiplications if inside
7796 an address. */
7801 {
7807 {
7810 }
7813 }
7823 {
7827 }
7830 {
7834 mode));
7836 }
7837 else
7838 {
7842 }
7853 {
7857 }
7860 {
7864 mode));
7866 }
7867 else
7868 {
7872 }
7876 /* If the second operand is not a constant, we can't do anything
7877 with it. */
7881 /* If the constant is a power of two minus one and the first operand
7882 is a logical right shift, make an extraction. */
7885 {
7889 }
7891 /* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
7896 {
7905 {
7906 /* If we narrowed the mode when dropping the subreg, then
7907 we must zero-extend to keep the semantics of the AND. */
7909 ;
7913 else
7915 }
7917 /* If that didn't give anything, see if the AND simplifies on
7918 its own. */
7920 {
7924 }
7925 }
7926 /* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
7932 {
7933 /* Apply the distributive law, and then try to make extractions. */
7940 }
7942 /* If we are have (and (rotate X C) M) and C is larger than the number
7943 of bits in M, this is an extraction. */
7949 {
7955 }
7957 /* On machines without logical shifts, if the operand of the AND is
7958 a logical shift and our mask turns off all the propagated sign
7959 bits, we can replace the logical shift with an arithmetic shift. */
7967 {
7974 make_compound_operation
7977 }
7979 /* If the constant is one less than a power of two, this might be
7980 representable by an extraction even if no shift is present.
7981 If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
7982 we are in a COMPARE. */
7986 next_code),
7989 /* If we are in a comparison and this is an AND with a power of two,
7990 convert this into the appropriate bit extract. */
7996 next_code),
8002 /* If the sign bit is known to be zero, replace this with an
8003 arithmetic shift. */
8008 {
8011 next_code),
8014 }
8016 /* ... fall through ... */
8022 /* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
8023 this is a SIGN_EXTRACT. */
8030 {
8037 }
8039 /* See if we have operations between an ASHIFTRT and an ASHIFT.
8040 If so, try to merge the shifts into a SIGN_EXTEND. We could
8041 also do this for some cases of SIGN_EXTRACT, but it doesn't
8042 seem worth the effort; the case checked for occurs on Alpha. */
8058 /* Call ourselves recursively on the inner expression. If we are
8059 narrowing the object and it has a different RTL code from
8060 what it originally did, do this SUBREG as a force_to_mode. */
8061 {
8065 /* If in_code is COMPARE, it isn't always safe to pass it through
8066 to the recursive make_compound_operation call. */
8070 /* (subreg:SI (and:DI (reg:DI) (const_int 0x800000000)) 0)
8071 is (const_int 0), rather than
8072 (subreg:SI (lshiftrt:DI (reg:DI) (const_int 35)) 0). */
8082 simplified
8090 {
8091 rtx newer
8094 /* If we have something other than a SUBREG, we might have
8095 done an expansion, so rerun ourselves. */
8099 /* force_to_mode can expand compounds. If it just re-expanded the
8100 compound, use gen_lowpart to convert to the desired mode. */
8102 /* Likewise if it re-expanded the compound only partially.
8103 This happens for SUBREG of ZERO_EXTRACT if they extract
8104 the same number of bits. */
8113 }
8117 }
8122 }
8125 {
8128 }
8130 /* Now recursively process each operand of this operation. We need to
8131 handle ZERO_EXTEND specially so that we don't lose track of the
8132 inner mode. */
8134 {
8142 }
8147 {
8150 }
8153 {
8156 }
8158 maybe_swap:
8159 /* If this is a commutative operation, the changes to the operands
8160 may have made it noncanonical. */
8163 {
8167 }
8170 }
8171 \f
8172 /* Given M see if it is a value that would select a field of bits
8173 within an item, but not the entire word. Return -1 if not.
8174 Otherwise, return the starting position of the field, where 0 is the
8175 low-order bit.
8177 *PLEN is set to the length of the field. */
8179 static int
8181 {
8182 /* Get the bit number of the first 1 bit from the right, -1 if none. */
8187 /* Now shift off the low-order zero bits and see if we have a
8188 power of two minus 1. */
8196 }
8197 \f
8198 /* If X refers to a register that equals REG in value, replace these
8199 references with REG. */
8200 static rtx
8202 {
8210 {
8246 {
8250 else
8252 }
8254 /* fall through */
8261 {
8264 {
8266 {
8269 }
8271 }
8272 }
8274 {
8277 {
8280 {
8282 {
8285 }
8287 }
8288 }
8289 }
8292 }
8295 }
8297 /* Return X converted to MODE. If the value is already truncated to
8298 MODE we can just return a subreg even though in the general case we
8299 would need an explicit truncation. */
8301 static rtx
8303 {
8308 {
8309 /* Bit-cast X into an integer mode. */
8314 }
8317 }
8319 /* See if X can be simplified knowing that we will only refer to it in
8320 MODE and will only refer to those bits that are nonzero in MASK.
8321 If other bits are being computed or if masking operations are done
8322 that select a superset of the bits in MASK, they can sometimes be
8323 ignored.
8325 Return a possibly simplified expression, but always convert X to
8326 MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
8328 If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
8329 are all off in X. This is used when X will be complemented, by either
8330 NOT, NEG, or XOR. */
8332 static rtx
8335 {
8338 machine_mode op_mode;
8342 /* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
8343 code below will do the wrong thing since the mode of such an
8344 expression is VOIDmode.
8346 Also do nothing if X is a CLOBBER; this can happen if X was
8347 the return value from a call to gen_lowpart. */
8351 /* We want to perform the operation in its present mode unless we know
8352 that the operation is valid in MODE, in which case we do the operation
8353 in MODE. */
8358 /* It is not valid to do a right-shift in a narrower mode
8359 than the one it came in with. */
8364 /* Truncate MASK to fit OP_MODE. */
8368 /* When we have an arithmetic operation, or a shift whose count we
8369 do not know, we need to assume that all bits up to the highest-order
8370 bit in MASK will be needed. This is how we form such a mask. */
8373 else
8377 /* Determine what bits of X are guaranteed to be (non)zero. */
8380 /* If none of the bits in X are needed, return a zero. */
8384 /* If X is a CONST_INT, return a new one. Do this here since the
8385 test below will fail. */
8387 {
8390 else
8391 {
8394 }
8395 }
8397 /* If X is narrower than MODE and we want all the bits in X's mode, just
8398 get X in the proper mode. */
8403 /* We can ignore the effect of a SUBREG if it narrows the mode or
8404 if the constant masks to zero all the bits the mode doesn't have. */
8414 /* The arithmetic simplifications here only work for scalar integer modes. */
8419 {
8421 /* If X is a (clobber (const_int)), return it since we know we are
8422 generating something that won't match. */
8435 /* Similarly for a truncate. */
8439 /* If this is an AND with a constant, convert it into an AND
8440 whose constant is the AND of that constant with MASK. If it
8441 remains an AND of MASK, delete it since it is redundant. */
8444 {
8448 /* If X is still an AND, see if it is an AND with a mask that
8449 is just some low-order bits. If so, and it is MASK, we don't
8450 need it. */
8457 /* If it remains an AND, try making another AND with the bits
8458 in the mode mask that aren't in MASK turned on. If the
8459 constant in the AND is wide enough, this might make a
8460 cheaper constant. */
8465 {
8466 unsigned HOST_WIDE_INT cval
8469 rtx y;
8476 }
8479 }
8484 /* In (and (plus FOO C1) M), if M is a mask that just turns off
8485 low-order bits (as in an alignment operation) and FOO is already
8486 aligned to that boundary, mask C1 to that boundary as well.
8487 This may eliminate that PLUS and, later, the AND. */
8489 {
8493 /* If MODE is narrower than HOST_WIDE_INT and mask is a negative
8494 number, sign extend it. */
8507 }
8509 /* ... fall through ... */
8512 /* Substituting into the operands of a widening MULT is not likely to
8513 create RTL matching a machine insn. */
8523 /* For PLUS, MINUS and MULT, we need any bits less significant than the
8524 most significant bit in MASK since carries from those bits will
8525 affect the bits we are interested in. */
8530 /* If X is (minus C Y) where C's least set bit is larger than any bit
8531 in the mask, then we may replace with (neg Y). */
8534 {
8538 }
8540 /* Similarly, if C contains every bit in the fuller_mask, then we may
8541 replace with (not Y). */
8544 {
8548 }
8555 /* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
8556 LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
8557 operation which may be a bitfield extraction. Ensure that the
8558 constant we form is not wider than the mode of X. */
8570 {
8579 }
8581 binop:
8582 /* For most binary operations, just propagate into the operation and
8583 change the mode if we have an operation of that mode. */
8588 /* If we ended up truncating both operands, truncate the result of the
8589 operation instead. */
8592 {
8595 }
8605 /* For left shifts, do the same, but just for the first operand.
8606 However, we cannot do anything with shifts where we cannot
8607 guarantee that the counts are smaller than the size of the mode
8608 because such a count will have a different meaning in a
8609 wider mode. */
8619 /* If the shift count is a constant and we can do arithmetic in
8620 the mode of the shift, refine which bits we need. Otherwise, use the
8621 conservative form of the mask. */
8627 else
8639 /* Here we can only do something if the shift count is a constant,
8640 this shift constant is valid for the host, and we can do arithmetic
8641 in OP_MODE. */
8647 {
8651 /* Select the mask of the bits we need for the shift operand. */
8654 /* We can only change the mode of the shift if we can do arithmetic
8655 in the mode of the shift and INNER_MASK is no wider than the
8656 width of X's mode. */
8664 }
8666 /* If we have (and (lshiftrt FOO C1) C2) where the combination of the
8667 shift and AND produces only copies of the sign bit (C2 is one less
8668 than a power of two), we can do this with just a shift. */
8672 /* The shift puts one of the sign bit copies in the least significant
8673 bit. */
8678 /* Number of bits left after the shift must be more than the mask
8679 needs. */
8682 /* Must be more sign bit copies than the mask needs. */
8692 /* If we are just looking for the sign bit, we don't need this shift at
8693 all, even if it has a variable count. */
8697 /* If this is a shift by a constant, get a mask that contains those bits
8698 that are not copies of the sign bit. We then have two cases: If
8699 MASK only includes those bits, this can be a logical shift, which may
8700 allow simplifications. If MASK is a single-bit field not within
8701 those bits, we are requesting a copy of the sign bit and hence can
8702 shift the sign bit to the appropriate location. */
8706 {
8709 /* If the considered data is wider than HOST_WIDE_INT, we can't
8710 represent a mask for all its bits in a single scalar.
8711 But we only care about the lower bits, so calculate these. */
8714 {
8717 /* GET_MODE_PRECISION (GET_MODE (x)) - INTVAL (XEXP (x, 1))
8718 is the number of bits a full-width mask would have set.
8719 We need only shift if these are fewer than nonzero can
8720 hold. If not, we must keep all bits set in nonzero. */
8725 + HOST_BITS_PER_WIDE_INT
8727 }
8728 else
8729 {
8732 }
8735 {
8740 }
8743 {
8744 x = simplify_shift_const
8750 }
8751 }
8753 /* If MASK is 1, convert this to an LSHIFTRT. This can be done
8754 even if the shift count isn't a constant. */
8759 shiftrt:
8761 /* If this is a zero- or sign-extension operation that just affects bits
8762 we don't care about, remove it. Be sure the call above returned
8763 something that is still a shift. */
8773 next_select);
8779 /* If the shift count is constant and we can do computations
8780 in the mode of X, compute where the bits we care about are.
8781 Otherwise, we can't do anything. Don't change the mode of
8782 the shift or propagate MODE into the shift, though. */
8785 {
8795 }
8799 /* If we just want the low-order bit, the NEG isn't needed since it
8800 won't change the low-order bit. */
8804 /* We need any bits less significant than the most significant bit in
8805 MASK since carries from those bits will affect the bits we are
8806 interested in. */
8811 /* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
8812 same as the XOR case above. Ensure that the constant we form is not
8813 wider than the mode of X. */
8821 {
8830 }
8832 /* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
8833 use the full mask inside the NOT. */
8836 unop:
8839 next_select));
8845 /* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
8846 in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
8847 which is equal to STORE_FLAG_VALUE. */
8859 /* We have no way of knowing if the IF_THEN_ELSE can itself be
8860 written in a narrower mode. We play it safe and do not do so. */
8876 }
8878 /* Ensure we return a value of the proper mode. */
8880 }
8881 \f
8882 /* Return nonzero if X is an expression that has one of two values depending on
8883 whether some other value is zero or nonzero. In that case, we return the
8884 value that is being tested, *PTRUE is set to the value if the rtx being
8885 returned has a nonzero value, and *PFALSE is set to the other alternative.
8887 If we return zero, we set *PTRUE and *PFALSE to X. */
8889 static rtx
8891 {
8897 /* If we are comparing a value against zero, we are done. */
8900 {
8904 }
8906 /* If this is a unary operation whose operand has one of two values, apply
8907 our opcode to compute those values. */
8910 {
8915 }
8917 /* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
8918 make can't possibly match and would suppress other optimizations. */
8920 ;
8922 /* If this is a binary operation, see if either side has only one of two
8923 values. If either one does or if both do and they are conditional on
8924 the same value, compute the new true and false values. */
8926 {
8932 {
8933 /* If if_then_else_cond returned zero, then true/false are the
8934 same rtl. We must copy one of them to prevent invalid rtl
8935 sharing. */
8942 {
8947 }
8948 else
8949 {
8952 }
8955 }
8957 /* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
8958 operands is zero when the other is nonzero, and vice-versa,
8959 and STORE_FLAG_VALUE is 1 or -1. */
8965 {
8982 {
8989 const_true_rtx);
8991 }
8992 }
8994 /* Similarly for MULT, AND and UMIN, except that for these the result
8995 is always zero. */
8999 {
9013 {
9016 }
9017 }
9018 }
9021 {
9022 /* If we have IF_THEN_ELSE already, extract the condition and
9023 canonicalize it if it is NE or EQ. */
9029 {
9032 }
9033 else
9035 }
9037 /* If X is a SUBREG, we can narrow both the true and false values
9038 if the inner expression, if there is a condition. */
9042 {
9048 {
9052 }
9053 }
9055 /* If X is a constant, this isn't special and will cause confusions
9056 if we treat it as such. Likewise if it is equivalent to a constant. */
9059 ;
9061 /* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
9062 will be least confusing to the rest of the compiler. */
9064 {
9067 }
9069 /* If X is known to be either 0 or -1, those are the true and
9070 false values when testing X. */
9074 {
9077 }
9079 /* Likewise for 0 or a single bit. */
9082 {
9085 }
9087 /* Otherwise fail; show no condition with true and false values the same. */
9090 }
9091 \f
9092 /* Return the value of expression X given the fact that condition COND
9093 is known to be true when applied to REG as its first operand and VAL
9094 as its second. X is known to not be shared and so can be modified in
9095 place.
9097 We only handle the simplest cases, and specifically those cases that
9098 arise with IF_THEN_ELSE expressions. */
9100 static rtx
9102 {
9110 /* If either operand of the condition is a floating point value,
9111 then we have to avoid collapsing an EQ comparison. */
9121 /* If X is (abs REG) and we know something about REG's relationship
9122 with zero, we may be able to simplify this. */
9126 {
9135 }
9137 /* The only other cases we handle are MIN, MAX, and comparisons if the
9138 operands are the same as REG and VAL. */
9141 {
9143 {
9146 }
9149 {
9151 {
9159 else
9161 }
9164 {
9167 /* Do not reverse the condition when it is NE or EQ.
9168 This is because we cannot conclude anything about
9169 the value of 'SMAX (x, y)' when x is not equal to y,
9170 but we can when x equals y. */
9176 {
9187 }
9188 }
9189 }
9190 }
9192 {
9197 {
9198 /* We must simplify subreg here, before we lose track of the
9199 original inner_mode. */
9204 else
9206 }
9209 }
9210 /* We don't have to handle SIGN_EXTEND here, because even in the
9211 case of replacing something with a modeless CONST_INT, a
9212 CONST_INT is already (supposed to be) a valid sign extension for
9213 its narrower mode, which implies it's already properly
9214 sign-extended for the wider mode. Now, for ZERO_EXTEND, the
9215 story is different. */
9217 {
9222 {
9223 /* We must simplify the zero_extend here, before we lose
9224 track of the original inner_mode. */
9229 else
9231 }
9234 }
9238 {
9245 }
9248 }
9249 \f
9250 /* See if X and Y are equal for the purposes of seeing if we can rewrite an
9251 assignment as a field assignment. */
9253 static int
9255 {
9257 {
9262 /* For big endian, adjust the memory offset. */
9267 else
9269 }
9277 /* Check for a paradoxical SUBREG of a MEM compared with the MEM.
9278 Note that all SUBREGs of MEM are paradoxical; otherwise they
9279 would have been rewritten. */
9292 /* We used to see if get_last_value of X and Y were the same but that's
9293 not correct. In one direction, we'll cause the assignment to have
9294 the wrong destination and in the case, we'll import a register into this
9295 insn that might have already have been dead. So fail if none of the
9296 above cases are true. */
9298 }
9299 \f
9300 /* See if X, a SET operation, can be rewritten as a bit-field assignment.
9301 Return that assignment if so.
9303 We only handle the most common cases. */
9305 static rtx
9307 {
9310 rtx assign;
9312 HOST_WIDE_INT c1;
9313 HOST_WIDE_INT pos;
9315 rtx other;
9316 machine_mode mode;
9318 /* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
9319 a clear of a one-bit field. We will have changed it to
9320 (and (rotate (const_int -2) POS) DEST), so check for that. Also check
9321 for a SUBREG. */
9327 {
9333 }
9343 {
9350 }
9352 /* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
9353 one-bit field. */
9357 {
9363 }
9365 /* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
9366 SRC is an AND with all bits of that field set, then we can discard
9367 the AND. */
9372 {
9379 else
9382 /* Complete overlap. We can remove the source AND. */
9386 /* Partial overlap. We can reduce the source AND. */
9388 {
9393 }
9394 }
9396 /* The other case we handle is assignments into a constant-position
9397 field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
9398 a mask that has all one bits except for a group of zero bits and
9399 OTHER is known to have zeros where C1 has ones, this is such an
9400 assignment. Compute the position and length from C1. Shift OTHER
9401 to the appropriate position, force it to the required mode, and
9402 make the extraction. Check for the AND in both operands. */
9404 /* One or more SUBREGs might obscure the constant-position field
9405 assignment. The first one we are likely to encounter is an outer
9406 narrowing SUBREG, which we can just strip for the purposes of
9407 identifying the constant-field assignment. */
9421 /* The second SUBREG that might get in the way is a paradoxical
9422 SUBREG around the first operand of the AND. We want to
9423 pretend the operand is as wide as the destination here. We
9424 do this by adjusting the MEM to wider mode for the sole
9425 purpose of the call to rtx_equal_for_field_assignment_p. Also
9426 note this trick only works for MEMs. */
9438 /* The second SUBREG that might get in the way is a paradoxical
9439 SUBREG around the first operand of the AND. We want to
9440 pretend the operand is as wide as the destination here. We
9441 do this by adjusting the MEM to wider mode for the sole
9442 purpose of the call to rtx_equal_for_field_assignment_p. Also
9443 note this trick only works for MEMs. */
9451 else
9464 /* The mode to use for the source is the mode of the assignment, or of
9465 what is inside a possible STRICT_LOW_PART. */
9469 /* Shift OTHER right POS places and make it the source, restricting it
9470 to the proper length and mode. */
9475 dest);
9482 /* If SRC is masked by an AND that does not make a difference in
9483 the value being stored, strip it. */
9494 }
9495 \f
9496 /* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
9497 if so. */
9499 static rtx
9501 {
9505 rtx tem;
9507 /* Distributivity is not true for floating point as it can change the
9508 value. So we don't do it unless -funsafe-math-optimizations. */
9513 /* The outer operation can only be one of the following: */
9521 /* If either operand is a primitive we can't do anything, so get out
9522 fast. */
9532 /* See if the inner and outer operations distribute. */
9534 {
9539 /* These all distribute except over PLUS. */
9550 /* This is also a multiply, so it distributes over everything. */
9553 /* This used to handle SUBREG, but this turned out to be counter-
9554 productive, since (subreg (op ...)) usually is not handled by
9555 insn patterns, and this "optimization" therefore transformed
9556 recognizable patterns into unrecognizable ones. Therefore the
9557 SUBREG case was removed from here.
9559 It is possible that distributing SUBREG over arithmetic operations
9560 leads to an intermediate result than can then be optimized further,
9561 e.g. by moving the outer SUBREG to the other side of a SET as done
9562 in simplify_set. This seems to have been the original intent of
9563 handling SUBREGs here.
9565 However, with current GCC this does not appear to actually happen,
9566 at least on major platforms. If some case is found where removing
9567 the SUBREG case here prevents follow-on optimizations, distributing
9568 SUBREGs ought to be re-added at that place, e.g. in simplify_set. */
9572 }
9574 /* Set LHS and RHS to the inner operands (A and B in the example
9575 above) and set OTHER to the common operand (C in the example).
9576 There is only one way to do this unless the inner operation is
9577 commutative. */
9589 else
9592 /* Form the new inner operation, seeing if it simplifies first. */
9595 /* There is one exception to the general way of distributing:
9596 (a | c) ^ (b | c) -> (a ^ b) & ~c */
9598 {
9601 }
9603 /* We may be able to continuing distributing the result, so call
9604 ourselves recursively on the inner operation before forming the
9605 outer operation, which we return. */
9608 }
9610 /* See if X is of the form (* (+ A B) C), and if so convert to
9611 (+ (* A C) (* B C)) and try to simplify.
9613 Most of the time, this results in no change. However, if some of
9614 the operands are the same or inverses of each other, simplifications
9615 will result.
9617 For example, (and (ior A B) (not B)) can occur as the result of
9618 expanding a bit field assignment. When we apply the distributive
9619 law to this, we get (ior (and (A (not B))) (and (B (not B)))),
9620 which then simplifies to (and (A (not B))).
9622 Note that no checks happen on the validity of applying the inverse
9623 distributive law. This is pointless since we can do it in the
9624 few places where this routine is called.
9626 N is the index of the term that is decomposed (the arithmetic operation,
9627 i.e. (+ A B) in the first example above). !N is the index of the term that
9628 is distributed, i.e. of C in the first example above. */
9629 static rtx
9631 {
9632 machine_mode mode;
9636 /* Distributivity is not true for floating point as it can change the
9637 value. So we don't do it unless -funsafe-math-optimizations. */
9654 /* Special case (and (xor B C) (not A)), which is equivalent to
9655 (xor (ior A B) (ior A C)) */
9657 {
9660 }
9663 {
9664 /* Distribute the second term. */
9667 }
9668 else
9669 {
9670 /* Distribute the first term. */
9673 }
9683 }
9684 \f
9685 /* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
9686 in MODE. Return an equivalent form, if different from (and VAROP
9687 (const_int CONSTOP)). Otherwise, return NULL_RTX. */
9689 static rtx
9692 {
9695 rtx orig_varop;
9703 /* Simplify VAROP knowing that we will be only looking at some of the
9704 bits in it.
9706 Note by passing in CONSTOP, we guarantee that the bits not set in
9707 CONSTOP are not significant and will never be examined. We must
9708 ensure that is the case by explicitly masking out those bits
9709 before returning. */
9712 /* If VAROP is a CLOBBER, we will fail so return it. */
9716 /* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
9717 to VAROP and return the new constant. */
9721 /* See what bits may be nonzero in VAROP. Unlike the general case of
9722 a call to nonzero_bits, here we don't care about bits outside
9723 MODE. */
9727 /* Turn off all bits in the constant that are known to already be zero.
9728 Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
9729 which is tested below. */
9733 /* If we don't have any bits left, return zero. */
9737 /* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
9738 a power of two, we can replace this with an ASHIFT. */
9743 /* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
9744 or XOR, then try to apply the distributive law. This may eliminate
9745 operations if either branch can be simplified because of the AND.
9746 It may also make some cases more complex, but those cases probably
9747 won't match a pattern either with or without this. */
9750 return
9751 gen_lowpart
9753 apply_distributive_law
9758 constop),
9762 constop))));
9764 /* If VAROP is PLUS, and the constant is a mask of low bits, distribute
9765 the AND and see if one of the operands simplifies to zero. If so, we
9766 may eliminate it. */
9770 {
9779 }
9781 /* Make a SUBREG if necessary. If we can't make it, fail. */
9786 /* If we are only masking insignificant bits, return VAROP. */
9793 /* Otherwise, return an AND. */
9795 }
9798 /* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
9799 in MODE.
9801 Return an equivalent form, if different from X. Otherwise, return X. If
9802 X is zero, we are to always construct the equivalent form. */
9804 static rtx
9807 {
9818 }
9819 \f
9820 /* Given a REG, X, compute which bits in X can be nonzero.
9821 We don't care about bits outside of those defined in MODE.
9823 For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
9824 a shift, AND, or zero_extract, we can do better. */
9826 static rtx
9828 const_rtx known_x ATTRIBUTE_UNUSED,
9829 machine_mode known_mode ATTRIBUTE_UNUSED,
9832 {
9833 rtx tem;
9836 /* If X is a register whose nonzero bits value is current, use it.
9837 Otherwise, if X is a register whose value we can find, use that
9838 value. Otherwise, use the previously-computed global nonzero bits
9839 for this register. */
9853 && !REGNO_REG_SET_P
9856 {
9860 /* We don't know anything about the upper bits. */
9865 }
9870 {
9876 }
9878 {
9882 /* We don't know anything about the upper bits. */
9886 }
9889 }
9891 /* Return the number of bits at the high-order end of X that are known to
9892 be equal to the sign bit. X will be used in mode MODE; if MODE is
9893 VOIDmode, X will be used in its own mode. The returned value will always
9894 be between 1 and the number of bits in MODE. */
9896 static rtx
9898 const_rtx known_x ATTRIBUTE_UNUSED,
9899 machine_mode known_mode
9900 ATTRIBUTE_UNUSED,
9903 {
9904 rtx tem;
9917 && !REGNO_REG_SET_P
9920 {
9923 }
9934 }
9935 \f
9936 /* Return the number of "extended" bits there are in X, when interpreted
9937 as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
9938 unsigned quantities, this is the number of high-order zero bits.
9939 For signed quantities, this is the number of copies of the sign bit
9940 minus 1. In both case, this function returns the number of "spare"
9941 bits. For example, if two quantities for which this function returns
9942 at least 1 are added, the addition is known not to overflow.
9944 This function will always return 0 unless called during combine, which
9945 implies that it must be called from a define_split. */
9947 unsigned int
9949 {
9959 }
9961 /* This function is called from `simplify_shift_const' to merge two
9962 outer operations. Specifically, we have already found that we need
9963 to perform operation *POP0 with constant *PCONST0 at the outermost
9964 position. We would now like to also perform OP1 with constant CONST1
9965 (with *POP0 being done last).
9967 Return 1 if we can do the operation and update *POP0 and *PCONST0 with
9968 the resulting operation. *PCOMP_P is set to 1 if we would need to
9969 complement the innermost operand, otherwise it is unchanged.
9971 MODE is the mode in which the operation will be done. No bits outside
9972 the width of this mode matter. It is assumed that the width of this mode
9973 is smaller than or equal to HOST_BITS_PER_WIDE_INT.
9975 If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
9976 IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
9977 result is simply *PCONST0.
9979 If the resulting operation cannot be expressed as one operation, we
9980 return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
9982 static int
9983 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)
9984 {
9991 /* If OP0 is an AND, clear unimportant bits in CONST1. */
9995 /* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
9996 if OP0 is SET. */
10005 {
10007 {
10025 }
10026 }
10028 /* Otherwise, if either is a PLUS or NEG, we can't do anything. */
10032 /* If the two constants aren't the same, we can't do anything. The
10033 remaining six cases can all be done. */
10037 else
10039 {
10042 /* (a & b) | b == b */
10045 /* (a ^ b) | b == a | b */
10046 {;}
10051 /* (a & b) ^ b == (~a) & b */
10054 /* (a | b) ^ b == a & ~b */
10060 /* (a | b) & b == b */
10063 /* (a ^ b) & b) == (~a) & b */
10068 }
10070 /* Check for NO-OP cases. */
10083 /* ??? Slightly redundant with the above mask, but not entirely.
10084 Moving this above means we'd have to sign-extend the mode mask
10085 for the final test. */
10090 }
10091 \f
10092 /* A helper to simplify_shift_const_1 to determine the mode we can perform
10093 the shift in. The original shift operation CODE is performed on OP in
10094 ORIG_MODE. Return the wider mode MODE if we can perform the operation
10095 in that mode. Return ORIG_MODE otherwise. We can also assume that the
10096 result of the shift is subject to operation OUTER_CODE with operand
10097 OUTER_CONST. */
10099 static machine_mode
10103 {
10108 /* In general we can't perform in wider mode for right shift and rotate. */
10110 {
10112 /* We can still widen if the bits brought in from the left are identical
10113 to the sign bit of ORIG_MODE. */
10121 /* Similarly here but with zero bits. */
10126 /* We can also widen if the bits brought in will be masked off. This
10127 operation is performed in ORIG_MODE. */
10129 {
10135 }
10136 /* fall through */
10146 }
10147 }
10149 /* Simplify a shift of VAROP by ORIG_COUNT bits. CODE says what kind
10150 of shift. The result of the shift is RESULT_MODE. Return NULL_RTX
10151 if we cannot simplify it. Otherwise, return a simplified value.
10153 The shift is normally computed in the widest mode we find in VAROP, as
10154 long as it isn't a different number of words than RESULT_MODE. Exceptions
10155 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10157 static rtx
10160 {
10166 unsigned int mode_words
10168 /* We form (outer_op (code varop count) (outer_const)). */
10174 /* Make sure and truncate the "natural" shift on the way in. We don't
10175 want to do this inside the loop as it makes it more difficult to
10176 combine shifts. */
10180 /* If we were given an invalid count, don't do anything except exactly
10181 what was requested. */
10188 /* Unless one of the branches of the `if' in this loop does a `continue',
10189 we will `break' the loop after the `if'. */
10192 {
10193 /* If we have an operand of (clobber (const_int 0)), fail. */
10197 /* Convert ROTATERT to ROTATE. */
10199 {
10204 else
10206 }
10211 /* Handle cases where the count is greater than the size of the mode
10212 minus 1. For ASHIFT, use the size minus one as the count (this can
10213 occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
10214 take the count modulo the size. For other shifts, the result is
10215 zero.
10217 Since these shifts are being produced by the compiler by combining
10218 multiple operations, each of which are defined, we know what the
10219 result is supposed to be. */
10222 {
10227 else
10228 {
10229 /* We can't simply return zero because there may be an
10230 outer op. */
10234 }
10235 }
10237 /* If we discovered we had to complement VAROP, leave. Making a NOT
10238 here would cause an infinite loop. */
10242 /* An arithmetic right shift of a quantity known to be -1 or 0
10243 is a no-op. */
10247 {
10250 }
10252 /* If we are doing an arithmetic right shift and discarding all but
10253 the sign bit copies, this is equivalent to doing a shift by the
10254 bitsize minus one. Convert it into that shift because it will often
10255 allow other simplifications. */
10262 /* We simplify the tests below and elsewhere by converting
10263 ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
10264 `make_compound_operation' will convert it to an ASHIFTRT for
10265 those machines (such as VAX) that don't have an LSHIFTRT. */
10282 {
10289 {
10292 }
10296 /* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
10297 minus the width of a smaller mode, we can do this with a
10298 SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
10305 {
10314 }
10318 /* If VAROP is a SUBREG, strip it as long as the inner operand has
10319 the same number of words as what we've seen so far. Then store
10320 the widest mode in MODE. */
10326 == mode_words
10329 {
10334 }
10338 /* Some machines use MULT instead of ASHIFT because MULT
10339 is cheaper. But it is still better on those machines to
10340 merge two shifts into one. */
10343 {
10344 varop
10350 }
10354 /* Similar, for when divides are cheaper. */
10357 {
10358 varop
10364 }
10368 /* If we are extracting just the sign bit of an arithmetic
10369 right shift, that shift is not needed. However, the sign
10370 bit of a wider mode may be different from what would be
10371 interpreted as the sign bit in a narrower mode, so, if
10372 the result is narrower, don't discard the shift. */
10377 {
10380 }
10382 /* ... fall through ... */
10387 /* Here we have two nested shifts. The result is usually the
10388 AND of a new shift with a mask. We compute the result below. */
10395 {
10399 rtx mask_rtx;
10401 /* We have one common special case. We can't do any merging if
10402 the inner code is an ASHIFTRT of a smaller mode. However, if
10403 we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
10404 with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
10405 we can convert it to
10406 (ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0) C3) C2) C1).
10407 This simplifies certain SIGN_EXTEND operations. */
10411 {
10412 /* C3 has the low-order C1 bits zero. */
10424 }
10426 /* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
10427 than C1 high-order bits equal to the sign bit, we can convert
10428 this to either an ASHIFT or an ASHIFTRT depending on the
10429 two counts.
10431 We cannot do this if VAROP's mode is not SHIFT_MODE. */
10437 {
10441 {
10444 }
10447 }
10449 /* There are some cases we can't do. If CODE is ASHIFTRT,
10450 we can only do this if FIRST_CODE is also ASHIFTRT.
10452 We can't do the case when CODE is ROTATE and FIRST_CODE is
10453 ASHIFTRT.
10455 If the mode of this shift is not the mode of the outer shift,
10456 we can't do this if either shift is a right shift or ROTATE.
10458 Finally, we can't do any of these if the mode is too wide
10459 unless the codes are the same.
10461 Handle the case where the shift codes are the same
10462 first. */
10465 {
10474 }
10485 /* To compute the mask to apply after the shift, shift the
10486 nonzero bits of the inner shift the same way the
10487 outer shift will. */
10490 result_mode);
10492 mask_rtx
10496 /* Give up if we can't compute an outer operation to use. */
10504 /* If the shifts are in the same direction, we add the
10505 counts. Otherwise, we subtract them. */
10509 else
10512 /* If COUNT is positive, the new shift is usually CODE,
10513 except for the two exceptions below, in which case it is
10514 FIRST_CODE. If the count is negative, FIRST_CODE should
10515 always be used */
10525 }
10527 /* If we have (A << B << C) for any shift, we can convert this to
10528 (A << C << B). This wins if A is a constant. Only try this if
10529 B is not a constant. */
10534 {
10535 /* For ((unsigned) (cstULL >> count)) >> cst2 we have to make
10536 sure the result will be masked. See PR70222. */
10544 /* For ((int) (cstLL >> count)) >> cst2 just give up. Queuing
10545 up outer sign extension (often left and right shift) is
10546 hardly more efficient than the original. See PR70429. */
10556 }
10563 /* Make this fit the case below. */
10570 /* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
10571 with C the size of VAROP - 1 and the shift is logical if
10572 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10573 we have an (le X 0) operation. If we have an arithmetic shift
10574 and STORE_FLAG_VALUE is 1 or we have a logical shift with
10575 STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
10583 {
10586 const0_rtx);
10592 }
10594 /* If we have (shift (logical)), move the logical to the outside
10595 to allow it to possibly combine with another logical and the
10596 shift to combine with another shift. This also canonicalizes to
10597 what a ZERO_EXTRACT looks like. Also, some machines have
10598 (and (shift)) insns. */
10601 /* We can't do this if we have (ashiftrt (xor)) and the
10602 constant has its sign bit set in shift_mode with shift_mode
10603 wider than result_mode. */
10607 shift_mode))
10615 {
10618 }
10620 /* If we can't do that, try to simplify the shift in each arm of the
10621 logical expression, make a new logical expression, and apply
10622 the inverse distributive law. This also can't be done for
10623 (ashiftrt (xor)) where we've widened the shift and the constant
10624 changes the sign bit. */
10629 shift_mode)))
10630 {
10642 }
10646 /* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
10647 says that the sign bit can be tested, FOO has mode MODE, C is
10648 GET_MODE_PRECISION (MODE) - 1, and FOO has only its low-order bit
10649 that may be nonzero. */
10659 {
10663 }
10667 /* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
10668 than the number of bits in the mode is equivalent to A. */
10672 {
10676 }
10678 /* NEG commutes with ASHIFT since it is multiplication. Move the
10679 NEG outside to allow shifts to combine. */
10683 {
10686 }
10690 /* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
10691 is one less than the number of bits in the mode is
10692 equivalent to (xor A 1). */
10699 {
10703 }
10705 /* If we have (xshiftrt (plus FOO BAR) C), and the only bits
10706 that might be nonzero in BAR are those being shifted out and those
10707 bits are known zero in FOO, we can replace the PLUS with FOO.
10708 Similarly in the other operand order. This code occurs when
10709 we are computing the size of a variable-size array. */
10716 {
10719 }
10727 result_mode)))
10728 {
10731 }
10733 /* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
10743 {
10746 }
10748 /* Check for 'PLUS signbit', which is the canonical form of 'XOR
10749 signbit', and attempt to change the PLUS to an XOR and move it to
10750 the outer operation as is done above in the AND/IOR/XOR case
10751 leg for shift(logical). See details in logical handling above
10752 for reasoning in doing so. */
10763 {
10766 }
10771 /* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
10772 with C the size of VAROP - 1 and the shift is logical if
10773 STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
10774 we have a (gt X 0) operation. If the shift is arithmetic with
10775 STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
10776 we have a (neg (gt X 0)) operation. */
10785 {
10788 const0_rtx);
10794 }
10798 /* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
10799 if the truncate does not affect the value. */
10806 {
10809 varop_inner
10812 GEN_INT
10817 }
10822 }
10825 }
10830 /* We have now finished analyzing the shift. The result should be
10831 a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
10832 OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
10833 to the result of the shift. OUTER_CONST is the relevant constant,
10834 but we must turn off all bits turned off in the shift. */
10842 /* Make a SUBREG if necessary. If we can't make it, fail. */
10847 /* If we have an outer operation and we just made a shift, it is
10848 possible that we could have simplified the shift were it not
10849 for the outer operation. So try to do the simplification
10850 recursively. */
10854 else
10860 /* If we were doing an LSHIFTRT in a wider mode than it was originally,
10861 turn off all the bits that the shift would have turned off. */
10866 /* Do the remainder of the processing in RESULT_MODE. */
10869 /* If COMPLEMENT_P is set, we have to complement X before doing the outer
10870 operation. */
10875 {
10883 {
10884 /* This means that we have determined that the result is
10885 equivalent to a constant. This should be rare. */
10888 }
10891 else
10894 }
10897 }
10899 /* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
10900 The result of the shift is RESULT_MODE. If we cannot simplify it,
10901 return X or, if it is NULL, synthesize the expression with
10902 simplify_gen_binary. Otherwise, return a simplified value.
10904 The shift is normally computed in the widest mode we find in VAROP, as
10905 long as it isn't a different number of words than RESULT_MODE. Exceptions
10906 are ASHIFTRT and ROTATE, which are always done in their original mode. */
10908 static rtx
10911 {
10921 }
10923 \f
10924 /* A subroutine of recog_for_combine. See there for arguments and
10925 return value. */
10927 static int
10929 {
10931 rtx pat_without_clobbers;
10939 /* If PAT is a PARALLEL, check to see if it contains the CLOBBER
10940 we use to indicate that something didn't match. If we find such a
10941 thing, force rejection. */
10955 {
10958 else
10961 }
10963 /* If it isn't, there is the possibility that we previously had an insn
10964 that clobbered some register as a side effect, but the combined
10965 insn doesn't need to do that. So try once more without the clobbers
10966 unless this represents an ASM insn. */
10970 {
10975 {
10978 pos++;
10979 }
10989 {
10992 else
10995 }
10996 }
11003 /* Recognize all noop sets, these will be killed by followup pass. */
11007 /* If we had any clobbers to add, make a new pattern than contains
11008 them. Then check to make sure that all of them are dead. */
11010 {
11020 else
11027 {
11032 {
11036 }
11037 }
11039 }
11043 {
11050 /* Allow targets to reject combined insn. */
11052 {
11055 dump_file);
11057 /* Callers expect recog_for_combine to strip
11058 clobbers from the pattern on failure. */
11063 }
11068 }
11074 }
11076 /* Change every ZERO_EXTRACT and ZERO_EXTEND of a SUBREG that can be
11077 expressed as an AND and maybe an LSHIFTRT, to that formulation.
11078 Return whether anything was so changed. */
11080 static bool
11082 {
11087 {
11096 {
11105 }
11110 {
11113 }
11114 else
11119 mask--;
11125 }
11128 }
11130 /* Like recog, but we receive the address of a pointer to a new pattern.
11131 We try to match the rtx that the pointer points to.
11132 If that fails, we may try to modify or replace the pattern,
11133 storing the replacement into the same pointer object.
11135 Modifications include deletion or addition of CLOBBERs. If the
11136 instruction will still not match, we change ZERO_EXTEND and ZERO_EXTRACT
11137 to the equivalent AND and perhaps LSHIFTRT patterns, and try with that
11138 (and undo if that fails).
11140 PNOTES is a pointer to a location where any REG_UNUSED notes added for
11141 the CLOBBERs are placed.
11143 The value is the final insn code from the pattern ultimately matched,
11144 or -1. */
11146 static int
11148 {
11160 {
11163 {
11167 }
11168 }
11171 {
11176 }
11179 }
11180 \f
11181 /* Like gen_lowpart_general but for use by combine. In combine it
11182 is not possible to create any new pseudoregs. However, it is
11183 safe to create invalid memory addresses, because combine will
11184 try to recognize them and all they will do is make the combine
11185 attempt fail.
11187 If for some reason this cannot do its job, an rtx
11188 (clobber (const_int 0)) is returned.
11189 An insn containing that will not be recognized. */
11191 static rtx
11193 {
11197 rtx result;
11202 /* We can only support MODE being wider than a word if X is a
11203 constant integer or has a mode the same size. */
11208 /* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
11209 won't know what to do. So we will strip off the SUBREG here and
11210 process normally. */
11212 {
11215 /* For use in case we fall down into the address adjustments
11216 further below, we need to adjust the known mode and size of
11217 x; imode and isize, since we just adjusted x. */
11224 }
11232 {
11235 /* Refuse to work on a volatile memory ref or one with a mode-dependent
11236 address. */
11241 /* If we want to refer to something bigger than the original memref,
11242 generate a paradoxical subreg instead. That will force a reload
11243 of the original memref X. */
11250 /* Adjust the address so that the address-after-the-data is
11251 unchanged. */
11256 }
11258 /* If X is a comparison operator, rewrite it in a new mode. This
11259 probably won't match, but may allow further simplifications. */
11263 /* If we couldn't simplify X any other way, just enclose it in a
11264 SUBREG. Normally, this SUBREG won't match, but some patterns may
11265 include an explicit SUBREG or we may simplify it further in combine. */
11266 else
11267 {
11268 rtx res;
11271 {
11276 }
11280 }
11282 fail:
11284 }
11285 \f
11286 /* Try to simplify a comparison between OP0 and a constant OP1,
11287 where CODE is the comparison code that will be tested, into a
11288 (CODE OP0 const0_rtx) form.
11290 The result is a possibly different comparison code to use.
11291 *POP1 may be updated. */
11293 static enum rtx_code
11296 {
11300 /* Get the constant we are comparing against and turn off all bits
11301 not on in our mode. */
11305 /* If we are comparing against a constant power of two and the value
11306 being compared can only have that single bit nonzero (e.g., it was
11307 `and'ed with that bit), we can replace this with a comparison
11308 with zero. */
11316 {
11319 }
11321 /* Similarly, if we are comparing a value known to be either -1 or
11322 0 with -1, change it to the opposite comparison against zero. */
11327 {
11330 }
11332 /* Do some canonicalizations based on the comparison code. We prefer
11333 comparisons against zero and then prefer equality comparisons.
11334 If we can reduce the size of a constant, we will do that too. */
11336 {
11338 /* < C is equivalent to <= (C - 1) */
11340 {
11343 /* ... fall through to LE case below. */
11344 }
11345 else
11349 /* <= C is equivalent to < (C + 1); we do this for C < 0 */
11351 {
11354 }
11356 /* If we are doing a <= 0 comparison on a value known to have
11357 a zero sign bit, we can replace this with == 0. */
11367 /* >= C is equivalent to > (C - 1). */
11369 {
11372 /* ... fall through to GT below. */
11373 }
11374 else
11378 /* > C is equivalent to >= (C + 1); we do this for C < 0. */
11380 {
11383 }
11385 /* If we are doing a > 0 comparison on a value known to have
11386 a zero sign bit, we can replace this with != 0. */
11396 /* < C is equivalent to <= (C - 1). */
11398 {
11401 /* ... fall through ... */
11402 }
11403 /* (unsigned) < 0x80000000 is equivalent to >= 0. */
11407 {
11411 }
11412 else
11416 /* unsigned <= 0 is equivalent to == 0 */
11419 /* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
11423 {
11426 }
11430 /* >= C is equivalent to > (C - 1). */
11432 {
11435 /* ... fall through ... */
11436 }
11438 /* (unsigned) >= 0x80000000 is equivalent to < 0. */
11442 {
11446 }
11447 else
11451 /* unsigned > 0 is equivalent to != 0 */
11454 /* (unsigned) > 0x7fffffff is equivalent to < 0. */
11458 {
11461 }
11466 }
11470 }
11471 \f
11472 /* Simplify a comparison between *POP0 and *POP1 where CODE is the
11473 comparison code that will be tested.
11475 The result is a possibly different comparison code to use. *POP0 and
11476 *POP1 may be updated.
11478 It is possible that we might detect that a comparison is either always
11479 true or always false. However, we do not perform general constant
11480 folding in combine, so this knowledge isn't useful. Such tautologies
11481 should have been detected earlier. Hence we ignore all such cases. */
11483 static enum rtx_code
11485 {
11492 /* Try a few ways of applying the same transformation to both operands. */
11494 {
11495 /* The test below this one won't handle SIGN_EXTENDs on these machines,
11496 so check specially. */
11512 - (GET_MODE_PRECISION
11514 {
11517 }
11519 /* If both operands are the same constant shift, see if we can ignore the
11520 shift. We can if the shift is a rotate or if the bits shifted out of
11521 this shift are known to be zero for both inputs and if the type of
11522 comparison is compatible with the shift. */
11535 {
11548 else
11550 }
11552 /* If both operands are AND's of a paradoxical SUBREG by constant, the
11553 SUBREGs are of the same mode, and, in both cases, the AND would
11554 be redundant if the comparison was done in the narrower mode,
11555 do the comparison in the narrower mode (e.g., we are AND'ing with 1
11556 and the operand's possibly nonzero bits are 0xffffff01; in that case
11557 if we only care about QImode, we don't need the AND). This case
11558 occurs if the output mode of an scc insn is not SImode and
11559 STORE_FLAG_VALUE == 1 (e.g., the 386).
11561 Similarly, check for a case where the AND's are ZERO_EXTEND
11562 operations from some narrower mode even though a SUBREG is not
11563 present. */
11568 {
11585 {
11589 /* The resulting comparison is always unsigned since we masked
11590 off the original sign bit. */
11594 }
11601 {
11607 }
11611 }
11613 /* If both operands are NOT, we can strip off the outer operation
11614 and adjust the comparison code for swapped operands; similarly for
11615 NEG, except that this must be an equality comparison. */
11621 else
11623 }
11625 /* If the first operand is a constant, swap the operands and adjust the
11626 comparison code appropriately, but don't do this if the second operand
11627 is already a constant integer. */
11629 {
11632 }
11634 /* We now enter a loop during which we will try to simplify the comparison.
11635 For the most part, we only are concerned with comparisons with zero,
11636 but some things may really be comparisons with zero but not start
11637 out looking that way. */
11640 {
11647 HOST_WIDE_INT const_op;
11649 /* We only want to handle integral modes. This catches VOIDmode,
11650 CCmode, and the floating-point modes. An exception is that we
11651 can handle VOIDmode if OP0 is a COMPARE or a comparison
11652 operation. */
11659 /* Try to simplify the compare to constant, possibly changing the
11660 comparison op, and/or changing op1 to zero. */
11664 /* Compute some predicates to simplify code below. */
11671 /* If this is a sign bit comparison and we can do arithmetic in
11672 MODE, say that we will only be needing the sign bit of OP0. */
11679 /* Now try cases based on the opcode of OP0. If none of the cases
11680 does a "continue", we exit this loop immediately after the
11681 switch. */
11684 {
11686 /* If we are extracting a single bit from a variable position in
11687 a constant that has only a single bit set and are comparing it
11688 with zero, we can convert this into an equality comparison
11689 between the position and the location of the single bit. */
11690 /* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
11691 have already reduced the shift count modulo the word size. */
11697 {
11705 /* Result is nonzero iff shift count is equal to I. */
11708 }
11710 /* ... fall through ... */
11715 {
11718 }
11722 /* If testing for equality, we can take the NOT of the constant. */
11725 {
11729 }
11731 /* If just looking at the sign bit, reverse the sense of the
11732 comparison. */
11734 {
11738 }
11742 /* If testing for equality, we can take the NEG of the constant. */
11745 {
11749 }
11751 /* The remaining cases only apply to comparisons with zero. */
11755 /* When X is ABS or is known positive,
11756 (neg X) is < 0 if and only if X != 0. */
11764 {
11768 }
11770 /* If we have NEG of something whose two high-order bits are the
11771 same, we know that "(-a) < 0" is equivalent to "a > 0". */
11773 {
11777 }
11781 /* If we are testing equality and our count is a constant, we
11782 can perform the inverse operation on our RHS. */
11786 {
11790 }
11792 /* If we are doing a < 0 or >= 0 comparison, it means we are testing
11793 a particular bit. Convert it to an AND of a constant of that
11794 bit. This will be converted into a ZERO_EXTRACT. */
11798 {
11805 }
11807 /* Fall through. */
11810 /* ABS is ignorable inside an equality comparison with zero. */
11812 {
11815 }
11819 /* Can simplify (compare (zero/sign_extend FOO) CONST) to
11820 (compare FOO CONST) if CONST fits in FOO's mode and we
11821 are either testing inequality or have an unsigned
11822 comparison with ZERO_EXTEND or a signed comparison with
11823 SIGN_EXTEND. But don't do it if we don't have a compare
11824 insn of the given mode, since we'd have to revert it
11825 later on, and then we wouldn't know whether to sign- or
11826 zero-extend. */
11829 && ! unsigned_comparison_p
11833 {
11836 }
11840 /* Check for the case where we are comparing A - C1 with C2, that is
11842 (subreg:MODE (plus (A) (-C1))) op (C2)
11844 with C1 a constant, and try to lift the SUBREG, i.e. to do the
11845 comparison in the wider mode. One of the following two conditions
11846 must be true in order for this to be valid:
11848 1. The mode extension results in the same bit pattern being added
11849 on both sides and the comparison is equality or unsigned. As
11850 C2 has been truncated to fit in MODE, the pattern can only be
11851 all 0s or all 1s.
11853 2. The mode extension results in the sign bit being copied on
11854 each side.
11856 The difficulty here is that we have predicates for A but not for
11857 (A - C1) so we need to check that C1 is within proper bounds so
11858 as to perturbate A as little as possible. */
11865 {
11874 /* (A - C1) zero-extends if it is positive and sign-extends
11875 if it is negative, C2 both zero- and sign-extends. */
11879 /* (A - C1) sign-extends if it is positive and 1-extends
11880 if it is negative, C2 both sign- and 1-extends. */
11887 /* (A - C1) always sign-extends, like C2. */
11891 {
11894 }
11895 }
11897 /* If the inner mode is narrower and we are extracting the low part,
11898 we can treat the SUBREG as if it were a ZERO_EXTEND. */
11902 else
11905 /* ... fall through ... */
11915 {
11918 }
11922 /* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
11923 this for equality comparisons due to pathological cases involving
11924 overflows. */
11928 {
11932 }
11934 /* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
11937 {
11941 }
11945 /* We used to optimize signed comparisons against zero, but that
11946 was incorrect. Unsigned comparisons against zero (GTU, LEU)
11947 arrive here as equality comparisons, or (GEU, LTU) are
11948 optimized away. No need to special-case them. */
11950 /* (eq (minus A B) C) -> (eq A (plus B C)) or
11951 (eq B (minus A C)), whichever simplifies. We can only do
11952 this for equality comparisons due to pathological cases involving
11953 overflows. */
11957 {
11961 }
11966 {
11970 }
11972 /* The sign bit of (minus (ashiftrt X C) X), where C is the number
11973 of bits in X minus 1, is one iff X > 0. */
11978 {
11982 }
11986 /* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
11987 if C is zero or B is a constant. */
11991 {
11995 }
12003 /* We can't do anything if OP0 is a condition code value, rather
12004 than an actual data value. */
12010 /* Get the two operands being compared. */
12013 else
12016 /* Check for the cases where we simply want the result of the
12017 earlier test or the opposite of that result. */
12021 {
12025 else
12029 {
12034 }
12035 }
12039 /* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
12040 iff X <= 0. */
12044 {
12048 }
12052 /* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
12053 will be converted to a ZERO_EXTRACT later. */
12057 {
12062 }
12064 /* If we are comparing (and (lshiftrt X C1) C2) for equality with
12065 zero and X is a comparison and C1 and C2 describe only bits set
12066 in STORE_FLAG_VALUE, we can compare with X. */
12074 {
12081 {
12084 }
12085 }
12087 /* If we are doing an equality comparison of an AND of a bit equal
12088 to the sign bit, replace this with a LT or GE comparison of
12089 the underlying value. */
12096 {
12100 }
12102 /* If this AND operation is really a ZERO_EXTEND from a narrower
12103 mode, the constant fits within that mode, and this is either an
12104 equality or unsigned comparison, try to do this comparison in
12105 the narrower mode.
12107 Note that in:
12109 (ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
12110 -> (ne:DI (reg:SI 4) (const_int 0))
12112 unless TRULY_NOOP_TRUNCATION allows it or the register is
12113 known to hold a value of the required mode the
12114 transformation is invalid. */
12122 {
12125 }
12127 /* If this is (and:M1 (subreg:M1 X:M2 0) (const_int C1)) where C1
12128 fits in both M1 and M2 and the SUBREG is either paradoxical
12129 or represents the low part, permute the SUBREG and the AND
12130 and try again. */
12133 {
12136 /* Require an integral mode, to avoid creating something like
12137 (AND:SF ...). */
12139 /* It is unsafe to commute the AND into the SUBREG if the
12140 SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
12141 not defined. As originally written the upper bits
12142 have a defined value due to the AND operation.
12143 However, if we commute the AND inside the SUBREG then
12144 they no longer have defined values and the meaning of
12145 the code has been changed.
12146 Also C1 should not change value in the smaller mode,
12147 see PR67028 (a positive C1 can become negative in the
12148 smaller mode, so that the AND does no longer mask the
12149 upper bits). */
12150 && ((WORD_REGISTER_OPERATIONS
12162 {
12168 }
12169 }
12171 /* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
12175 {
12180 }
12182 /* Convert (ne (and (lshiftrt (not X)) 1) 0) to
12183 (eq (and (lshiftrt X) 1) 0).
12184 Also handle the case where (not X) is expressed using xor. */
12188 {
12200 {
12201 op0
12206 }
12207 }
12211 /* If we have (compare (ashift FOO N) (const_int C)) and
12212 the high order N bits of FOO (N+1 if an inequality comparison)
12213 are known to be zero, we can do this by comparing FOO with C
12214 shifted right N bits so long as the low-order N bits of C are
12215 zero. */
12227 {
12228 /* We must perform a logical shift, not an arithmetic one,
12229 as we want the top N bits of C to be zero. */
12236 }
12238 /* If we are doing a sign bit comparison, it means we are testing
12239 a particular bit. Convert it to the appropriate AND. */
12242 {
12249 }
12251 /* If this an equality comparison with zero and we are shifting
12252 the low bit to the sign bit, we can convert this to an AND of the
12253 low-order bit. */
12257 {
12260 }
12264 /* If this is an equality comparison with zero, we can do this
12265 as a logical shift, which might be much simpler. */
12268 {
12273 }
12275 /* If OP0 is a sign extension and CODE is not an unsigned comparison,
12276 do the comparison in a narrower mode. */
12286 {
12289 }
12291 /* Likewise if OP0 is a PLUS of a sign extension with a
12292 constant, which is usually represented with the PLUS
12293 between the shifts. */
12305 {
12313 new_const);
12315 }
12317 /* ... fall through ... */
12319 /* If we have (compare (xshiftrt FOO N) (const_int C)) and
12320 the low order N bits of FOO are known to be zero, we can do this
12321 by comparing FOO with C shifted left N bits so long as no
12322 overflow occurs. Even if the low order N bits of FOO aren't known
12323 to be zero, if the comparison is >= or < we can use the same
12324 optimization and for > or <= by setting all the low
12325 order N bits in the comparison constant. */
12336 {
12337 unsigned HOST_WIDE_INT low_bits
12342 {
12343 /* If the shift was logical, then we must make the condition
12344 unsigned. */
12352 const_op
12357 }
12358 }
12360 /* If we are using this shift to extract just the sign bit, we
12361 can replace this with an LT or GE comparison. */
12366 {
12370 }
12375 }
12378 }
12380 /* Now make any compound operations involved in this comparison. Then,
12381 check for an outmost SUBREG on OP0 that is not doing anything or is
12382 paradoxical. The latter transformation must only be performed when
12383 it is known that the "extra" bits will be the same in op0 and op1 or
12384 that they don't matter. There are three cases to consider:
12386 1. SUBREG_REG (op0) is a register. In this case the bits are don't
12387 care bits and we can assume they have any convenient value. So
12388 making the transformation is safe.
12390 2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not defined.
12391 In this case the upper bits of op0 are undefined. We should not make
12392 the simplification in that case as we do not know the contents of
12393 those bits.
12395 3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is defined and not
12396 UNKNOWN. In that case we know those bits are zeros or ones. We must
12397 also be sure that they are the same as the upper bits of op1.
12399 We can never remove a SUBREG for a non-equality comparison because
12400 the sign bit is in a different place in the underlying object. */
12413 {
12415 {
12416 /* For paradoxical subregs, allow case 1 as above. Case 3 isn't
12417 implemented. */
12419 {
12422 }
12423 }
12429 {
12435 }
12436 }
12438 /* We now do the opposite procedure: Some machines don't have compare
12439 insns in all modes. If OP0's mode is an integer mode smaller than a
12440 word and we can't do a compare in that mode, see if there is a larger
12441 mode for which we can do the compare. There are a number of cases in
12442 which we can use the wider mode. */
12452 {
12455 /* If this is a test for negative, we can make an explicit
12456 test of the sign bit. Test this first so we can use
12457 a paradoxical subreg to extend OP0. */
12461 {
12462 unsigned HOST_WIDE_INT sign
12469 }
12471 /* If the only nonzero bits in OP0 and OP1 are those in the
12472 narrower mode and this is an equality or unsigned comparison,
12473 we can use the wider mode. Similarly for sign-extended
12474 values, in which case it is true for all comparisons. */
12491 {
12492 /* If OP0 is an AND and we don't have an AND in MODE either,
12493 make a new AND in the proper mode. */
12501 else
12502 {
12504 {
12507 }
12508 else
12509 {
12512 }
12514 }
12515 }
12516 }
12518 /* We may have changed the comparison operands. Re-canonicalize. */
12520 {
12523 }
12525 /* If this machine only supports a subset of valid comparisons, see if we
12526 can convert an unsupported one into a supported one. */
12533 }
12534 \f
12535 /* Utility function for record_value_for_reg. Count number of
12536 rtxs in X. */
12537 static int
12539 {
12546 {
12566 }
12577 }
12578 \f
12579 /* Utility function for following routine. Called when X is part of a value
12580 being stored into last_set_value. Sets last_set_table_tick
12581 for each register mentioned. Similar to mention_regs in cse.c */
12583 static void
12585 {
12591 {
12597 {
12600 }
12603 }
12607 {
12608 /* Check for identical subexpressions. If x contains
12609 identical subexpression we only have to traverse one of
12610 them. */
12612 {
12613 /* Note that at this point x1 has already been
12614 processed. */
12618 /* If x0 and x1 are identical then there is no need to
12619 process x0. */
12623 /* If x0 is identical to a subexpression of x1 then while
12624 processing x1, x0 has already been processed. Thus we
12625 are done with x. */
12630 /* If x1 is identical to a subexpression of x0 then we
12631 still have to process the rest of x0. */
12634 {
12637 }
12638 }
12641 }
12645 }
12647 /* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
12648 are saying that the register is clobbered and we no longer know its
12649 value. If INSN is zero, don't update reg_stat[].last_set; this is
12650 only permitted with VALUE also zero and is used to invalidate the
12651 register. */
12653 static void
12655 {
12661 /* If VALUE contains REG and we have a previous value for REG, substitute
12662 the previous value. */
12664 {
12665 rtx tem;
12667 /* Set things up so get_last_value is allowed to see anything set up to
12668 our insn. */
12672 /* If TEM is simply a binary operation with two CLOBBERs as operands,
12673 it isn't going to be useful and will take a lot of time to process,
12674 so just use the CLOBBER. */
12677 {
12683 {
12684 /* If there are two or more occurrences of REG in VALUE,
12685 prevent the value from growing too much. */
12688 }
12691 }
12692 }
12694 /* For each register modified, show we don't know its value, that
12695 we don't know about its bitwise content, that its value has been
12696 updated, and that we don't know the location of the death of the
12697 register. */
12699 {
12711 }
12713 /* Mark registers that are being referenced in this value. */
12717 /* Now update the status of each register being set.
12718 If someone is using this register in this block, set this register
12719 to invalid since we will get confused between the two lives in this
12720 basic block. This makes using this register always invalid. In cse, we
12721 scan the table to invalidate all entries using this register, but this
12722 is too much work for us. */
12725 {
12731 else
12733 }
12735 /* The value being assigned might refer to X (like in "x++;"). In that
12736 case, we must replace it with (clobber (const_int 0)) to prevent
12737 infinite loops. */
12740 {
12744 }
12746 /* For the main register being modified, update the value, the mode, the
12747 nonzero bits, and the number of sign bit copies. */
12752 {
12760 rsp->last_set_sign_bit_copies
12762 }
12763 }
12765 /* Called via note_stores from record_dead_and_set_regs to handle one
12766 SET or CLOBBER in an insn. DATA is the instruction in which the
12767 set is occurring. */
12769 static void
12771 {
12778 {
12782 }
12785 {
12786 /* If we are setting the whole register, we know its value. Otherwise
12787 show that we don't know the value. We can handle SUBREG in
12788 some cases. */
12799 else
12801 }
12803 /* Ignore pushes, they clobber nothing. */
12806 }
12808 /* Update the records of when each REG was most recently set or killed
12809 for the things done by INSN. This is the last thing done in processing
12810 INSN in the combiner loop.
12812 We update reg_stat[], in particular fields last_set, last_set_value,
12813 last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
12814 last_death, and also the similar information mem_last_set (which insn
12815 most recently modified memory) and last_call_luid (which insn was the
12816 most recent subroutine call). */
12818 static void
12820 {
12821 rtx link;
12825 {
12828 {
12833 {
12838 }
12839 }
12842 }
12845 {
12846 hard_reg_set_iterator hrsi;
12848 {
12860 }
12864 /* We can't combine into a call pattern. Remember, though, that
12865 the return value register is set at this LUID. We could
12866 still replace a register with the return value from the
12867 wrong subroutine call! */
12869 }
12870 else
12872 }
12874 /* If a SUBREG has the promoted bit set, it is in fact a property of the
12875 register present in the SUBREG, so for each such SUBREG go back and
12876 adjust nonzero and sign bit information of the registers that are
12877 known to have some zero/sign bits set.
12879 This is needed because when combine blows the SUBREGs away, the
12880 information on zero/sign bits is lost and further combines can be
12881 missed because of that. */
12883 static void
12885 {
12887 rtx set;
12895 {
12904 {
12907 }
12911 {
12914 }
12917 {
12920 }
12921 else
12923 }
12924 }
12926 /* Check if X, a register, is known to contain a value already
12927 truncated to MODE. In this case we can use a subreg to refer to
12928 the truncated value even though in the generic case we would need
12929 an explicit truncation. */
12931 static bool
12933 {
12945 }
12947 /* If X is a hard reg or a subreg record the mode that the register is
12948 accessed in. For non-TRULY_NOOP_TRUNCATION targets we might be able
12949 to turn a truncate into a subreg using this information. Return true
12950 if traversing X is complete. */
12952 static bool
12954 {
12955 machine_mode truncated_mode;
12959 {
12970 }
12971 /* ??? For hard-regs we now record everything. We might be able to
12972 optimize this using last_set_mode. */
12975 else
12983 {
12986 }
12989 }
12991 /* Callback for note_uses. Find hardregs and subregs of pseudos and
12992 the modes they are used in. This can help truning TRUNCATEs into
12993 SUBREGs. */
12995 static void
12997 {
13002 }
13004 /* Scan X for promoted SUBREGs. For each one found,
13005 note what it implies to the registers used in it. */
13007 static void
13009 {
13014 else
13015 {
13021 {
13031 }
13032 }
13033 }
13034 \f
13035 /* Verify that all the registers and memory references mentioned in *LOC are
13036 still valid. *LOC was part of a value set in INSN when label_tick was
13037 equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
13038 the invalid references with (clobber (const_int 0)) and return 1. This
13039 replacement is useful because we often can get useful information about
13040 the form of a value (e.g., if it was produced by a shift that always
13041 produces -1 or 0) even though we don't know exactly what registers it
13042 was produced from. */
13044 static int
13046 {
13053 {
13059 {
13062 /* If this is a pseudo-register that was only set once and not
13063 live at the beginning of the function, it is always valid. */
13067 && (!REGNO_REG_SET_P
13069 regno)))
13071 {
13075 }
13076 }
13079 }
13080 /* If this is a memory reference, make sure that there were no stores after
13081 it that might have clobbered the value. We don't have alias info, so we
13082 assume any store invalidates it. Moreover, we only have local UIDs, so
13083 we also assume that there were stores in the intervening basic blocks. */
13086 {
13090 }
13093 {
13095 {
13096 /* Check for identical subexpressions. If x contains
13097 identical subexpression we only have to traverse one of
13098 them. */
13100 {
13101 /* Note that at this point x0 has already been checked
13102 and found valid. */
13106 /* If x0 and x1 are identical then x is also valid. */
13110 /* If x1 is identical to a subexpression of x0 then
13111 while checking x0, x1 has already been checked. Thus
13112 it is valid and so as x. */
13117 /* If x0 is identical to a subexpression of x1 then x is
13118 valid iff the rest of x1 is valid. */
13121 return
13125 }
13130 }
13136 }
13138 /* If we haven't found a reason for it to be invalid, it is valid. */
13140 }
13142 /* Get the last value assigned to X, if known. Some registers
13143 in the value may be replaced with (clobber (const_int 0)) if their value
13144 is known longer known reliably. */
13146 static rtx
13148 {
13150 rtx value;
13153 /* If this is a non-paradoxical SUBREG, get the value of its operand and
13154 then convert it to the desired mode. If this is a paradoxical SUBREG,
13155 we cannot predict what values the "extra" bits might have. */
13169 /* If we don't have a value, or if it isn't for this basic block and
13170 it's either a hard register, set more than once, or it's a live
13171 at the beginning of the function, return 0.
13173 Because if it's not live at the beginning of the function then the reg
13174 is always set before being used (is never used without being set).
13175 And, if it's set only once, and it's always set before use, then all
13176 uses must have the same last value, even if it's not from this basic
13177 block. */
13184 || REGNO_REG_SET_P
13188 /* If the value was set in a later insn than the ones we are processing,
13189 we can't use it even if the register was only set once. */
13194 /* If fewer bits were set than what we are asked for now, we cannot use
13195 the value. */
13200 /* If the value has all its registers valid, return it. */
13204 /* Otherwise, make a copy and replace any invalid register with
13205 (clobber (const_int 0)). If that fails for some reason, return 0. */
13212 }
13213 \f
13214 /* Return nonzero if expression X refers to a REG or to memory
13215 that is set in an instruction more recent than FROM_LUID. */
13217 static int
13219 {
13225 {
13229 #ifdef PUSH_ROUNDING
13230 /* Don't allow uses of the stack pointer to be moved,
13231 because we don't know whether the move crosses a push insn. */
13234 #endif
13236 {
13242 }
13244 }
13252 {
13254 {
13259 }
13263 }
13265 }
13266 \f
13267 /* Define three variables used for communication between the following
13268 routines. */
13273 /* Function called via note_stores from reg_dead_at_p.
13275 If DEST is within [reg_dead_regno, reg_dead_endregno), set
13276 reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
13278 static void
13280 {
13290 }
13292 /* Return nonzero if REG is known to be dead at INSN.
13294 We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
13295 referencing REG, it is dead. If we hit a SET referencing REG, it is
13296 live. Otherwise, see if it is live or dead at the start of the basic
13297 block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
13298 must be assumed to be always live. */
13300 static int
13302 {
13303 basic_block block;
13306 /* Set variables for reg_dead_at_p_1. */
13312 /* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
13313 we allow the machine description to decide whether use-and-clobber
13314 patterns are OK. */
13316 {
13320 }
13322 /* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
13323 beginning of basic block. */
13326 {
13328 {
13338 }
13344 }
13346 /* Look at live-in sets for the basic block that we were in. */
13352 }
13353 \f
13354 /* Note hard registers in X that are used. */
13356 static void
13358 {
13364 {
13368 CASE_CONST_ANY:
13373 /* CC0 must die in the insn after it is set, so we don't need to take
13374 special note of it here. */
13379 /* If we are clobbering a MEM, mark any hard registers inside the
13380 address as used. */
13387 /* A hard reg in a wide mode may really be multiple registers.
13388 If so, mark all of them just like the first. */
13390 {
13391 /* None of this applies to the stack, frame or arg pointers. */
13393 || (!HARD_FRAME_POINTER_IS_FRAME_POINTER
13401 }
13405 {
13406 /* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
13407 the address. */
13419 }
13424 }
13426 /* Recursively scan the operands of this expression. */
13428 {
13432 {
13436 {
13441 }
13442 }
13443 }
13444 }
13445 \f
13446 /* Remove register number REGNO from the dead registers list of INSN.
13448 Return the note used to record the death, if there was one. */
13450 rtx
13452 {
13459 }
13461 /* For each register (hardware or pseudo) used within expression X, if its
13462 death is in an instruction with luid between FROM_LUID (inclusive) and
13463 TO_INSN (exclusive), put a REG_DEAD note for that register in the
13464 list headed by PNOTES.
13466 That said, don't move registers killed by maybe_kill_insn.
13468 This is done when X is being merged by combination into TO_INSN. These
13469 notes will then be distributed as needed. */
13471 static void
13474 {
13480 {
13484 /* Don't move the register if it gets killed in between from and to. */
13493 {
13496 /* It is possible for the call above to return 0. This can occur
13497 when last_death points to I2 or I1 that we combined with.
13498 In that case make a new note.
13500 We must also check for the case where X is a hard register
13501 and NOTE is a death note for a range of hard registers
13502 including X. In that case, we must put REG_DEAD notes for
13503 the remaining registers in place of NOTE. */
13508 {
13517 }
13519 /* If we didn't find any note, or if we found a REG_DEAD note that
13520 covers only part of the given reg, and we have a multi-reg hard
13521 register, then to be safe we must check for REG_DEAD notes
13522 for each register other than the first. They could have
13523 their own REG_DEAD notes lying around. */
13530 {
13537 else
13543 }
13546 {
13549 }
13550 else
13552 }
13555 }
13558 {
13563 /* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
13564 that accesses one word of a multi-word item, some
13565 piece of everything register in the expression is used by
13566 this insn, so remove any old death. */
13567 /* ??? So why do we test for equality of the sizes? */
13576 {
13579 }
13581 /* If this is some other SUBREG, we know it replaces the entire
13582 value, so use that as the destination. */
13586 /* If this is a MEM, adjust deaths of anything used in the address.
13587 For a REG (the only other possibility), the entire value is
13588 being replaced so the old value is not used in this insn. */
13594 }
13603 {
13605 {
13610 }
13613 }
13614 }
13615 \f
13616 /* Return 1 if X is the target of a bit-field assignment in BODY, the
13617 pattern of an insn. X must be a REG. */
13619 static int
13621 {
13625 {
13627 rtx target;
13634 else
13651 }
13659 }
13660 \f
13661 /* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
13662 as appropriate. I3 and I2 are the insns resulting from the combination
13663 insns including FROM (I2 may be zero).
13665 ELIM_I2 and ELIM_I1 are either zero or registers that we know will
13666 not need REG_DEAD notes because they are being substituted for. This
13667 saves searching in the most common cases.
13669 Each note in the list is either ignored or placed on some insns, depending
13670 on the type of note. */
13672 static void
13675 {
13677 rtx tem_note;
13681 {
13686 {
13689 /* Doesn't matter much where we put this, as long as it's somewhere.
13690 It is preferable to keep these notes on branches, which is most
13691 likely to be i3. */
13698 else
13699 {
13702 }
13706 /* These notes must remain with the call or trapping instruction. */
13711 else
13712 {
13718 /* ??? Otherwise assume we've combined things such that we
13719 can now prove that the instructions can't trap. Drop the
13720 note in this case. */
13721 }
13725 /* ??? How to distribute between i3-i1. Assume i3 contains the
13726 entire adjustment. Assert i3 contains at least some adjust. */
13728 {
13730 /* fixup_args_size_notes looks at REG_NORETURN note,
13731 so ensure the note is placed there first. */
13733 {
13737 {
13743 }
13744 }
13746 /* emit_call_1 adds for !ACCUMULATE_OUTGOING_ARGS
13747 REG_ARGS_SIZE note to all noreturn calls, allow that here. */
13750 && !ACCUMULATE_OUTGOING_ARGS
13752 }
13759 /* These notes must remain with the call. It should not be
13760 possible for both I2 and I3 to be a call. */
13763 else
13764 {
13767 }
13771 /* Any clobbers for i3 may still exist, and so we must process
13772 REG_UNUSED notes from that insn.
13774 Any clobbers from i2 or i1 can only exist if they were added by
13775 recog_for_combine. In that case, recog_for_combine created the
13776 necessary REG_UNUSED notes. Trying to keep any original
13777 REG_UNUSED notes from these insns can cause incorrect output
13778 if it is for the same register as the original i3 dest.
13779 In that case, we will notice that the register is set in i3,
13780 and then add a REG_UNUSED note for the destination of i3, which
13781 is wrong. However, it is possible to have REG_UNUSED notes from
13782 i2 or i1 for register which were both used and clobbered, so
13783 we keep notes from i2 or i1 if they will turn into REG_DEAD
13784 notes. */
13786 /* If this register is set or clobbered in I3, put the note there
13787 unless there is one already. */
13789 {
13797 }
13798 /* Otherwise, if this register is used by I3, then this register
13799 now dies here, so we must put a REG_DEAD note here unless there
13800 is one already. */
13806 {
13809 }
13815 /* These notes say something about results of an insn. We can
13816 only support them if they used to be on I3 in which case they
13817 remain on I3. Otherwise they are ignored.
13819 If the note refers to an expression that is not a constant, we
13820 must also ignore the note since we cannot tell whether the
13821 equivalence is still true. It might be possible to do
13822 slightly better than this (we only have a problem if I2DEST
13823 or I1DEST is present in the expression), but it doesn't
13824 seem worth the trouble. */
13832 /* These notes say something about how a register is used. They must
13833 be present on any use of the register in I2 or I3. */
13838 {
13841 else
13843 }
13848 /* This can show up in several ways -- either directly in the
13849 pattern, or hidden off in the constant pool with (or without?)
13850 a REG_EQUAL note. */
13851 /* ??? Ignore the without-reg_equal-note problem for now. */
13863 {
13866 else
13868 }
13870 /* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
13871 as a JUMP_LABEL or decrement LABEL_NUSES if it's already
13872 there. */
13877 {
13884 }
13890 {
13898 }
13902 /* This note says something about the value of a register prior
13903 to the execution of an insn. It is too much trouble to see
13904 if the note is still correct in all situations. It is better
13905 to simply delete it. */
13909 /* If we replaced the right hand side of FROM_INSN with a
13910 REG_EQUAL note, the original use of the dying register
13911 will not have been combined into I3 and I2. In such cases,
13912 FROM_INSN is guaranteed to be the first of the combined
13913 instructions, so we simply need to search back before
13914 FROM_INSN for the previous use or set of this register,
13915 then alter the notes there appropriately.
13917 If the register is used as an input in I3, it dies there.
13918 Similarly for I2, if it is nonzero and adjacent to I3.
13920 If the register is not used as an input in either I3 or I2
13921 and it is not one of the registers we were supposed to eliminate,
13922 there are two possibilities. We might have a non-adjacent I2
13923 or we might have somehow eliminated an additional register
13924 from a computation. For example, we might have had A & B where
13925 we discover that B will always be zero. In this case we will
13926 eliminate the reference to A.
13928 In both cases, we must search to see if we can find a previous
13929 use of A and put the death note there. */
13935 else
13936 {
13947 && !(i2mod
13949 i2mod_old_rhs)))
13954 /* If the new I2 sets the same register that is marked dead
13955 in the note, we do not know where to put the note.
13956 Give up. */
13959 }
13962 {
13966 {
13968 {
13972 }
13974 /* If the register is being set at TEM_INSN, see if that is all
13975 TEM_INSN is doing. If so, delete TEM_INSN. Otherwise, make this
13976 into a REG_UNUSED note instead. Don't delete sets to
13977 global register vars. */
13981 {
13992 ;
13994 /* Verify that it was the set, and not a clobber that
13995 modified the register.
13997 CC0 targets must be careful to maintain setter/user
13998 pairs. If we cannot delete the setter due to side
13999 effects, mark the user with an UNUSED note instead
14000 of deleting it. */
14004 && (!HAVE_cc0
14008 {
14009 /* Move the notes and links of TEM_INSN elsewhere.
14010 This might delete other dead insns recursively.
14011 First set the pattern to something that won't use
14012 any register. */
14026 /* Delete the setter too. */
14028 {
14041 }
14042 }
14043 else
14044 {
14047 /* If there isn't already a REG_UNUSED note, put one
14048 here. Do not place a REG_DEAD note, even if
14049 the register is also used here; that would not
14050 match the algorithm used in lifetime analysis
14051 and can cause the consistency check in the
14052 scheduler to fail. */
14057 }
14058 }
14062 {
14065 /* If we are doing a 3->2 combination, and we have a
14066 register which formerly died in i3 and was not used
14067 by i2, which now no longer dies in i3 and is used in
14068 i2 but does not die in i2, and place is between i2
14069 and i3, then we may need to move a link from place to
14070 i2. */
14072 && from_insn
14075 {
14079 }
14081 }
14085 }
14087 }
14089 /* If the register is set or already dead at PLACE, we needn't do
14090 anything with this note if it is still a REG_DEAD note.
14091 We check here if it is set at all, not if is it totally replaced,
14092 which is what `dead_or_set_p' checks, so also check for it being
14093 set partially. */
14096 {
14102 {
14103 /* Unless the register previously died in PLACE, clear
14104 last_death. [I no longer understand why this is
14105 being done.] */
14109 }
14110 else
14113 /* If this is a death note for a hard reg that is occupying
14114 multiple registers, ensure that we are still using all
14115 parts of the object. If we find a piece of the object
14116 that is unused, we must arrange for an appropriate REG_DEAD
14117 note to be added for it. However, we can't just emit a USE
14118 and tag the note to it, since the register might actually
14119 be dead; so we recourse, and the recursive call then finds
14120 the previous insn that used this register. */
14123 {
14132 {
14135 }
14138 {
14139 /* Put only REG_DEAD notes for pieces that are
14140 not already dead or set. */
14144 {
14151 {
14153 NULL_RTX);
14157 NULL_RTX);
14158 }
14163 {
14165 {
14169 }
14173 {
14176 }
14177 }
14178 }
14181 }
14182 }
14183 }
14187 /* Any other notes should not be present at this point in the
14188 compilation. */
14190 }
14193 {
14196 }
14200 }
14201 }
14202 \f
14203 /* Similarly to above, distribute the LOG_LINKS that used to be present on
14204 I3, I2, and I1 to new locations. This is also called to add a link
14205 pointing at I3 when I3's destination is changed. */
14207 static void
14209 {
14213 {
14220 /* If the insn that this link points to is a NOTE, ignore it. */
14229 {
14232 {
14248 }
14251 }
14252 else
14262 /* A LOG_LINK is defined as being placed on the first insn that uses
14263 a register and points to the insn that sets the register. Start
14264 searching at the next insn after the target of the link and stop
14265 when we reach a set of the register or the end of the basic block.
14267 Note that this correctly handles the link that used to point from
14268 I3 to I2. Also note that not much searching is typically done here
14269 since most links don't point very far away. */
14278 {
14282 }
14285 {
14288 }
14292 /* If we found a place to put the link, place it there unless there
14293 is already a link to the same insn as LINK at that point. */
14296 {
14304 {
14308 /* Set added_links_insn to the earliest insn we added a
14309 link to. */
14313 }
14314 }
14315 }
14316 }
14317 \f
14318 /* Check for any register or memory mentioned in EQUIV that is not
14319 mentioned in EXPR. This is used to restrict EQUIV to "specializations"
14320 of EXPR where some registers may have been replaced by constants. */
14322 static bool
14324 {
14327 {
14332 }
14334 }
14335 \f
14336 DEBUG_FUNCTION void
14338 {
14339 fprintf
14341 ";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
14343 }
14345 void
14347 {
14348 fprintf
14350 "\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
14352 }
14353 \f
14354 /* Try combining insns through substitution. */
14355 static unsigned int
14357 {
14367 rebuild_jump_labels_after_combine
14370 /* Combining insns may have turned an indirect jump into a
14371 direct jump. Rebuild the JUMP_LABEL fields of jumping
14372 instructions. */
14374 {
14379 }
14383 }
14388 {
14398 };
14401 {
14405 {}
14407 /* opt_pass methods: */
14410 {
14412 }
14418 rtl_opt_pass *
14420 {
14422 }