1 |
280 |
jeremybenn |
/* Optimize by combining instructions for GNU compiler.
|
2 |
|
|
Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
|
3 |
|
|
1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
|
4 |
|
|
Free Software Foundation, Inc.
|
5 |
|
|
|
6 |
|
|
This file is part of GCC.
|
7 |
|
|
|
8 |
|
|
GCC is free software; you can redistribute it and/or modify it under
|
9 |
|
|
the terms of the GNU General Public License as published by the Free
|
10 |
|
|
Software Foundation; either version 3, or (at your option) any later
|
11 |
|
|
version.
|
12 |
|
|
|
13 |
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
14 |
|
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
15 |
|
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
16 |
|
|
for more details.
|
17 |
|
|
|
18 |
|
|
You should have received a copy of the GNU General Public License
|
19 |
|
|
along with GCC; see the file COPYING3. If not see
|
20 |
|
|
<http://www.gnu.org/licenses/>. */
|
21 |
|
|
|
22 |
|
|
/* This module is essentially the "combiner" phase of the U. of Arizona
|
23 |
|
|
Portable Optimizer, but redone to work on our list-structured
|
24 |
|
|
representation for RTL instead of their string representation.
|
25 |
|
|
|
26 |
|
|
The LOG_LINKS of each insn identify the most recent assignment
|
27 |
|
|
to each REG used in the insn. It is a list of previous insns,
|
28 |
|
|
each of which contains a SET for a REG that is used in this insn
|
29 |
|
|
and not used or set in between. LOG_LINKs never cross basic blocks.
|
30 |
|
|
They were set up by the preceding pass (lifetime analysis).
|
31 |
|
|
|
32 |
|
|
We try to combine each pair of insns joined by a logical link.
|
33 |
|
|
We also try to combine triples of insns A, B and C when
|
34 |
|
|
C has a link back to B and B has a link back to A.
|
35 |
|
|
|
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.
|
41 |
|
|
|
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.
|
44 |
|
|
|
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.
|
51 |
|
|
|
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):
|
55 |
|
|
|
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
|
64 |
|
|
|
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.
|
68 |
|
|
|
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. */
|
77 |
|
|
|
78 |
|
|
#include "config.h"
|
79 |
|
|
#include "system.h"
|
80 |
|
|
#include "coretypes.h"
|
81 |
|
|
#include "tm.h"
|
82 |
|
|
#include "rtl.h"
|
83 |
|
|
#include "tree.h"
|
84 |
|
|
#include "tm_p.h"
|
85 |
|
|
#include "flags.h"
|
86 |
|
|
#include "regs.h"
|
87 |
|
|
#include "hard-reg-set.h"
|
88 |
|
|
#include "basic-block.h"
|
89 |
|
|
#include "insn-config.h"
|
90 |
|
|
#include "function.h"
|
91 |
|
|
/* Include expr.h after insn-config.h so we get HAVE_conditional_move. */
|
92 |
|
|
#include "expr.h"
|
93 |
|
|
#include "insn-attr.h"
|
94 |
|
|
#include "recog.h"
|
95 |
|
|
#include "real.h"
|
96 |
|
|
#include "toplev.h"
|
97 |
|
|
#include "target.h"
|
98 |
|
|
#include "optabs.h"
|
99 |
|
|
#include "insn-codes.h"
|
100 |
|
|
#include "rtlhooks-def.h"
|
101 |
|
|
/* Include output.h for dump_file. */
|
102 |
|
|
#include "output.h"
|
103 |
|
|
#include "params.h"
|
104 |
|
|
#include "timevar.h"
|
105 |
|
|
#include "tree-pass.h"
|
106 |
|
|
#include "df.h"
|
107 |
|
|
#include "cgraph.h"
|
108 |
|
|
|
109 |
|
|
/* Number of attempts to combine instructions in this function. */
|
110 |
|
|
|
111 |
|
|
static int combine_attempts;
|
112 |
|
|
|
113 |
|
|
/* Number of attempts that got as far as substitution in this function. */
|
114 |
|
|
|
115 |
|
|
static int combine_merges;
|
116 |
|
|
|
117 |
|
|
/* Number of instructions combined with added SETs in this function. */
|
118 |
|
|
|
119 |
|
|
static int combine_extras;
|
120 |
|
|
|
121 |
|
|
/* Number of instructions combined in this function. */
|
122 |
|
|
|
123 |
|
|
static int combine_successes;
|
124 |
|
|
|
125 |
|
|
/* Totals over entire compilation. */
|
126 |
|
|
|
127 |
|
|
static int total_attempts, total_merges, total_extras, total_successes;
|
128 |
|
|
|
129 |
|
|
/* combine_instructions may try to replace the right hand side of the
|
130 |
|
|
second instruction with the value of an associated REG_EQUAL note
|
131 |
|
|
before throwing it at try_combine. That is problematic when there
|
132 |
|
|
is a REG_DEAD note for a register used in the old right hand side
|
133 |
|
|
and can cause distribute_notes to do wrong things. This is the
|
134 |
|
|
second instruction if it has been so modified, null otherwise. */
|
135 |
|
|
|
136 |
|
|
static rtx i2mod;
|
137 |
|
|
|
138 |
|
|
/* When I2MOD is nonnull, this is a copy of the old right hand side. */
|
139 |
|
|
|
140 |
|
|
static rtx i2mod_old_rhs;
|
141 |
|
|
|
142 |
|
|
/* When I2MOD is nonnull, this is a copy of the new right hand side. */
|
143 |
|
|
|
144 |
|
|
static rtx i2mod_new_rhs;
|
145 |
|
|
|
146 |
|
|
typedef struct reg_stat_struct {
|
147 |
|
|
/* Record last point of death of (hard or pseudo) register n. */
|
148 |
|
|
rtx last_death;
|
149 |
|
|
|
150 |
|
|
/* Record last point of modification of (hard or pseudo) register n. */
|
151 |
|
|
rtx last_set;
|
152 |
|
|
|
153 |
|
|
/* The next group of fields allows the recording of the last value assigned
|
154 |
|
|
to (hard or pseudo) register n. We use this information to see if an
|
155 |
|
|
operation being processed is redundant given a prior operation performed
|
156 |
|
|
on the register. For example, an `and' with a constant is redundant if
|
157 |
|
|
all the zero bits are already known to be turned off.
|
158 |
|
|
|
159 |
|
|
We use an approach similar to that used by cse, but change it in the
|
160 |
|
|
following ways:
|
161 |
|
|
|
162 |
|
|
(1) We do not want to reinitialize at each label.
|
163 |
|
|
(2) It is useful, but not critical, to know the actual value assigned
|
164 |
|
|
to a register. Often just its form is helpful.
|
165 |
|
|
|
166 |
|
|
Therefore, we maintain the following fields:
|
167 |
|
|
|
168 |
|
|
last_set_value the last value assigned
|
169 |
|
|
last_set_label records the value of label_tick when the
|
170 |
|
|
register was assigned
|
171 |
|
|
last_set_table_tick records the value of label_tick when a
|
172 |
|
|
value using the register is assigned
|
173 |
|
|
last_set_invalid set to nonzero when it is not valid
|
174 |
|
|
to use the value of this register in some
|
175 |
|
|
register's value
|
176 |
|
|
|
177 |
|
|
To understand the usage of these tables, it is important to understand
|
178 |
|
|
the distinction between the value in last_set_value being valid and
|
179 |
|
|
the register being validly contained in some other expression in the
|
180 |
|
|
table.
|
181 |
|
|
|
182 |
|
|
(The next two parameters are out of date).
|
183 |
|
|
|
184 |
|
|
reg_stat[i].last_set_value is valid if it is nonzero, and either
|
185 |
|
|
reg_n_sets[i] is 1 or reg_stat[i].last_set_label == label_tick.
|
186 |
|
|
|
187 |
|
|
Register I may validly appear in any expression returned for the value
|
188 |
|
|
of another register if reg_n_sets[i] is 1. It may also appear in the
|
189 |
|
|
value for register J if reg_stat[j].last_set_invalid is zero, or
|
190 |
|
|
reg_stat[i].last_set_label < reg_stat[j].last_set_label.
|
191 |
|
|
|
192 |
|
|
If an expression is found in the table containing a register which may
|
193 |
|
|
not validly appear in an expression, the register is replaced by
|
194 |
|
|
something that won't match, (clobber (const_int 0)). */
|
195 |
|
|
|
196 |
|
|
/* Record last value assigned to (hard or pseudo) register n. */
|
197 |
|
|
|
198 |
|
|
rtx last_set_value;
|
199 |
|
|
|
200 |
|
|
/* Record the value of label_tick when an expression involving register n
|
201 |
|
|
is placed in last_set_value. */
|
202 |
|
|
|
203 |
|
|
int last_set_table_tick;
|
204 |
|
|
|
205 |
|
|
/* Record the value of label_tick when the value for register n is placed in
|
206 |
|
|
last_set_value. */
|
207 |
|
|
|
208 |
|
|
int last_set_label;
|
209 |
|
|
|
210 |
|
|
/* These fields are maintained in parallel with last_set_value and are
|
211 |
|
|
used to store the mode in which the register was last set, the bits
|
212 |
|
|
that were known to be zero when it was last set, and the number of
|
213 |
|
|
sign bits copies it was known to have when it was last set. */
|
214 |
|
|
|
215 |
|
|
unsigned HOST_WIDE_INT last_set_nonzero_bits;
|
216 |
|
|
char last_set_sign_bit_copies;
|
217 |
|
|
ENUM_BITFIELD(machine_mode) last_set_mode : 8;
|
218 |
|
|
|
219 |
|
|
/* Set nonzero if references to register n in expressions should not be
|
220 |
|
|
used. last_set_invalid is set nonzero when this register is being
|
221 |
|
|
assigned to and last_set_table_tick == label_tick. */
|
222 |
|
|
|
223 |
|
|
char last_set_invalid;
|
224 |
|
|
|
225 |
|
|
/* Some registers that are set more than once and used in more than one
|
226 |
|
|
basic block are nevertheless always set in similar ways. For example,
|
227 |
|
|
a QImode register may be loaded from memory in two places on a machine
|
228 |
|
|
where byte loads zero extend.
|
229 |
|
|
|
230 |
|
|
We record in the following fields if a register has some leading bits
|
231 |
|
|
that are always equal to the sign bit, and what we know about the
|
232 |
|
|
nonzero bits of a register, specifically which bits are known to be
|
233 |
|
|
zero.
|
234 |
|
|
|
235 |
|
|
If an entry is zero, it means that we don't know anything special. */
|
236 |
|
|
|
237 |
|
|
unsigned char sign_bit_copies;
|
238 |
|
|
|
239 |
|
|
unsigned HOST_WIDE_INT nonzero_bits;
|
240 |
|
|
|
241 |
|
|
/* Record the value of the label_tick when the last truncation
|
242 |
|
|
happened. The field truncated_to_mode is only valid if
|
243 |
|
|
truncation_label == label_tick. */
|
244 |
|
|
|
245 |
|
|
int truncation_label;
|
246 |
|
|
|
247 |
|
|
/* Record the last truncation seen for this register. If truncation
|
248 |
|
|
is not a nop to this mode we might be able to save an explicit
|
249 |
|
|
truncation if we know that value already contains a truncated
|
250 |
|
|
value. */
|
251 |
|
|
|
252 |
|
|
ENUM_BITFIELD(machine_mode) truncated_to_mode : 8;
|
253 |
|
|
} reg_stat_type;
|
254 |
|
|
|
255 |
|
|
DEF_VEC_O(reg_stat_type);
|
256 |
|
|
DEF_VEC_ALLOC_O(reg_stat_type,heap);
|
257 |
|
|
|
258 |
|
|
static VEC(reg_stat_type,heap) *reg_stat;
|
259 |
|
|
|
260 |
|
|
/* Record the luid of the last insn that invalidated memory
|
261 |
|
|
(anything that writes memory, and subroutine calls, but not pushes). */
|
262 |
|
|
|
263 |
|
|
static int mem_last_set;
|
264 |
|
|
|
265 |
|
|
/* Record the luid of the last CALL_INSN
|
266 |
|
|
so we can tell whether a potential combination crosses any calls. */
|
267 |
|
|
|
268 |
|
|
static int last_call_luid;
|
269 |
|
|
|
270 |
|
|
/* When `subst' is called, this is the insn that is being modified
|
271 |
|
|
(by combining in a previous insn). The PATTERN of this insn
|
272 |
|
|
is still the old pattern partially modified and it should not be
|
273 |
|
|
looked at, but this may be used to examine the successors of the insn
|
274 |
|
|
to judge whether a simplification is valid. */
|
275 |
|
|
|
276 |
|
|
static rtx subst_insn;
|
277 |
|
|
|
278 |
|
|
/* This is the lowest LUID that `subst' is currently dealing with.
|
279 |
|
|
get_last_value will not return a value if the register was set at or
|
280 |
|
|
after this LUID. If not for this mechanism, we could get confused if
|
281 |
|
|
I2 or I1 in try_combine were an insn that used the old value of a register
|
282 |
|
|
to obtain a new value. In that case, we might erroneously get the
|
283 |
|
|
new value of the register when we wanted the old one. */
|
284 |
|
|
|
285 |
|
|
static int subst_low_luid;
|
286 |
|
|
|
287 |
|
|
/* This contains any hard registers that are used in newpat; reg_dead_at_p
|
288 |
|
|
must consider all these registers to be always live. */
|
289 |
|
|
|
290 |
|
|
static HARD_REG_SET newpat_used_regs;
|
291 |
|
|
|
292 |
|
|
/* This is an insn to which a LOG_LINKS entry has been added. If this
|
293 |
|
|
insn is the earlier than I2 or I3, combine should rescan starting at
|
294 |
|
|
that location. */
|
295 |
|
|
|
296 |
|
|
static rtx added_links_insn;
|
297 |
|
|
|
298 |
|
|
/* Basic block in which we are performing combines. */
|
299 |
|
|
static basic_block this_basic_block;
|
300 |
|
|
static bool optimize_this_for_speed_p;
|
301 |
|
|
|
302 |
|
|
|
303 |
|
|
/* Length of the currently allocated uid_insn_cost array. */
|
304 |
|
|
|
305 |
|
|
static int max_uid_known;
|
306 |
|
|
|
307 |
|
|
/* The following array records the insn_rtx_cost for every insn
|
308 |
|
|
in the instruction stream. */
|
309 |
|
|
|
310 |
|
|
static int *uid_insn_cost;
|
311 |
|
|
|
312 |
|
|
/* The following array records the LOG_LINKS for every insn in the
|
313 |
|
|
instruction stream as an INSN_LIST rtx. */
|
314 |
|
|
|
315 |
|
|
static rtx *uid_log_links;
|
316 |
|
|
|
317 |
|
|
#define INSN_COST(INSN) (uid_insn_cost[INSN_UID (INSN)])
|
318 |
|
|
#define LOG_LINKS(INSN) (uid_log_links[INSN_UID (INSN)])
|
319 |
|
|
|
320 |
|
|
/* Incremented for each basic block. */
|
321 |
|
|
|
322 |
|
|
static int label_tick;
|
323 |
|
|
|
324 |
|
|
/* Reset to label_tick for each extended basic block in scanning order. */
|
325 |
|
|
|
326 |
|
|
static int label_tick_ebb_start;
|
327 |
|
|
|
328 |
|
|
/* Mode used to compute significance in reg_stat[].nonzero_bits. It is the
|
329 |
|
|
largest integer mode that can fit in HOST_BITS_PER_WIDE_INT. */
|
330 |
|
|
|
331 |
|
|
static enum machine_mode nonzero_bits_mode;
|
332 |
|
|
|
333 |
|
|
/* Nonzero when reg_stat[].nonzero_bits and reg_stat[].sign_bit_copies can
|
334 |
|
|
be safely used. It is zero while computing them and after combine has
|
335 |
|
|
completed. This former test prevents propagating values based on
|
336 |
|
|
previously set values, which can be incorrect if a variable is modified
|
337 |
|
|
in a loop. */
|
338 |
|
|
|
339 |
|
|
static int nonzero_sign_valid;
|
340 |
|
|
|
341 |
|
|
|
342 |
|
|
/* Record one modification to rtl structure
|
343 |
|
|
to be undone by storing old_contents into *where. */
|
344 |
|
|
|
345 |
|
|
enum undo_kind { UNDO_RTX, UNDO_INT, UNDO_MODE };
|
346 |
|
|
|
347 |
|
|
struct undo
|
348 |
|
|
{
|
349 |
|
|
struct undo *next;
|
350 |
|
|
enum undo_kind kind;
|
351 |
|
|
union { rtx r; int i; enum machine_mode m; } old_contents;
|
352 |
|
|
union { rtx *r; int *i; } where;
|
353 |
|
|
};
|
354 |
|
|
|
355 |
|
|
/* Record a bunch of changes to be undone, up to MAX_UNDO of them.
|
356 |
|
|
num_undo says how many are currently recorded.
|
357 |
|
|
|
358 |
|
|
other_insn is nonzero if we have modified some other insn in the process
|
359 |
|
|
of working on subst_insn. It must be verified too. */
|
360 |
|
|
|
361 |
|
|
struct undobuf
|
362 |
|
|
{
|
363 |
|
|
struct undo *undos;
|
364 |
|
|
struct undo *frees;
|
365 |
|
|
rtx other_insn;
|
366 |
|
|
};
|
367 |
|
|
|
368 |
|
|
static struct undobuf undobuf;
|
369 |
|
|
|
370 |
|
|
/* Number of times the pseudo being substituted for
|
371 |
|
|
was found and replaced. */
|
372 |
|
|
|
373 |
|
|
static int n_occurrences;
|
374 |
|
|
|
375 |
|
|
static rtx reg_nonzero_bits_for_combine (const_rtx, enum machine_mode, const_rtx,
|
376 |
|
|
enum machine_mode,
|
377 |
|
|
unsigned HOST_WIDE_INT,
|
378 |
|
|
unsigned HOST_WIDE_INT *);
|
379 |
|
|
static rtx reg_num_sign_bit_copies_for_combine (const_rtx, enum machine_mode, const_rtx,
|
380 |
|
|
enum machine_mode,
|
381 |
|
|
unsigned int, unsigned int *);
|
382 |
|
|
static void do_SUBST (rtx *, rtx);
|
383 |
|
|
static void do_SUBST_INT (int *, int);
|
384 |
|
|
static void init_reg_last (void);
|
385 |
|
|
static void setup_incoming_promotions (rtx);
|
386 |
|
|
static void set_nonzero_bits_and_sign_copies (rtx, const_rtx, void *);
|
387 |
|
|
static int cant_combine_insn_p (rtx);
|
388 |
|
|
static int can_combine_p (rtx, rtx, rtx, rtx, rtx *, rtx *);
|
389 |
|
|
static int combinable_i3pat (rtx, rtx *, rtx, rtx, int, rtx *);
|
390 |
|
|
static int contains_muldiv (rtx);
|
391 |
|
|
static rtx try_combine (rtx, rtx, rtx, int *);
|
392 |
|
|
static void undo_all (void);
|
393 |
|
|
static void undo_commit (void);
|
394 |
|
|
static rtx *find_split_point (rtx *, rtx);
|
395 |
|
|
static rtx subst (rtx, rtx, rtx, int, int);
|
396 |
|
|
static rtx combine_simplify_rtx (rtx, enum machine_mode, int);
|
397 |
|
|
static rtx simplify_if_then_else (rtx);
|
398 |
|
|
static rtx simplify_set (rtx);
|
399 |
|
|
static rtx simplify_logical (rtx);
|
400 |
|
|
static rtx expand_compound_operation (rtx);
|
401 |
|
|
static const_rtx expand_field_assignment (const_rtx);
|
402 |
|
|
static rtx make_extraction (enum machine_mode, rtx, HOST_WIDE_INT,
|
403 |
|
|
rtx, unsigned HOST_WIDE_INT, int, int, int);
|
404 |
|
|
static rtx extract_left_shift (rtx, int);
|
405 |
|
|
static rtx make_compound_operation (rtx, enum rtx_code);
|
406 |
|
|
static int get_pos_from_mask (unsigned HOST_WIDE_INT,
|
407 |
|
|
unsigned HOST_WIDE_INT *);
|
408 |
|
|
static rtx canon_reg_for_combine (rtx, rtx);
|
409 |
|
|
static rtx force_to_mode (rtx, enum machine_mode,
|
410 |
|
|
unsigned HOST_WIDE_INT, int);
|
411 |
|
|
static rtx if_then_else_cond (rtx, rtx *, rtx *);
|
412 |
|
|
static rtx known_cond (rtx, enum rtx_code, rtx, rtx);
|
413 |
|
|
static int rtx_equal_for_field_assignment_p (rtx, rtx);
|
414 |
|
|
static rtx make_field_assignment (rtx);
|
415 |
|
|
static rtx apply_distributive_law (rtx);
|
416 |
|
|
static rtx distribute_and_simplify_rtx (rtx, int);
|
417 |
|
|
static rtx simplify_and_const_int_1 (enum machine_mode, rtx,
|
418 |
|
|
unsigned HOST_WIDE_INT);
|
419 |
|
|
static rtx simplify_and_const_int (rtx, enum machine_mode, rtx,
|
420 |
|
|
unsigned HOST_WIDE_INT);
|
421 |
|
|
static int merge_outer_ops (enum rtx_code *, HOST_WIDE_INT *, enum rtx_code,
|
422 |
|
|
HOST_WIDE_INT, enum machine_mode, int *);
|
423 |
|
|
static rtx simplify_shift_const_1 (enum rtx_code, enum machine_mode, rtx, int);
|
424 |
|
|
static rtx simplify_shift_const (rtx, enum rtx_code, enum machine_mode, rtx,
|
425 |
|
|
int);
|
426 |
|
|
static int recog_for_combine (rtx *, rtx, rtx *);
|
427 |
|
|
static rtx gen_lowpart_for_combine (enum machine_mode, rtx);
|
428 |
|
|
static enum rtx_code simplify_comparison (enum rtx_code, rtx *, rtx *);
|
429 |
|
|
static void update_table_tick (rtx);
|
430 |
|
|
static void record_value_for_reg (rtx, rtx, rtx);
|
431 |
|
|
static void check_promoted_subreg (rtx, rtx);
|
432 |
|
|
static void record_dead_and_set_regs_1 (rtx, const_rtx, void *);
|
433 |
|
|
static void record_dead_and_set_regs (rtx);
|
434 |
|
|
static int get_last_value_validate (rtx *, rtx, int, int);
|
435 |
|
|
static rtx get_last_value (const_rtx);
|
436 |
|
|
static int use_crosses_set_p (const_rtx, int);
|
437 |
|
|
static void reg_dead_at_p_1 (rtx, const_rtx, void *);
|
438 |
|
|
static int reg_dead_at_p (rtx, rtx);
|
439 |
|
|
static void move_deaths (rtx, rtx, int, rtx, rtx *);
|
440 |
|
|
static int reg_bitfield_target_p (rtx, rtx);
|
441 |
|
|
static void distribute_notes (rtx, rtx, rtx, rtx, rtx, rtx);
|
442 |
|
|
static void distribute_links (rtx);
|
443 |
|
|
static void mark_used_regs_combine (rtx);
|
444 |
|
|
static void record_promoted_value (rtx, rtx);
|
445 |
|
|
static int unmentioned_reg_p_1 (rtx *, void *);
|
446 |
|
|
static bool unmentioned_reg_p (rtx, rtx);
|
447 |
|
|
static int record_truncated_value (rtx *, void *);
|
448 |
|
|
static void record_truncated_values (rtx *, void *);
|
449 |
|
|
static bool reg_truncated_to_mode (enum machine_mode, const_rtx);
|
450 |
|
|
static rtx gen_lowpart_or_truncate (enum machine_mode, rtx);
|
451 |
|
|
|
452 |
|
|
|
453 |
|
|
/* It is not safe to use ordinary gen_lowpart in combine.
|
454 |
|
|
See comments in gen_lowpart_for_combine. */
|
455 |
|
|
#undef RTL_HOOKS_GEN_LOWPART
|
456 |
|
|
#define RTL_HOOKS_GEN_LOWPART gen_lowpart_for_combine
|
457 |
|
|
|
458 |
|
|
/* Our implementation of gen_lowpart never emits a new pseudo. */
|
459 |
|
|
#undef RTL_HOOKS_GEN_LOWPART_NO_EMIT
|
460 |
|
|
#define RTL_HOOKS_GEN_LOWPART_NO_EMIT gen_lowpart_for_combine
|
461 |
|
|
|
462 |
|
|
#undef RTL_HOOKS_REG_NONZERO_REG_BITS
|
463 |
|
|
#define RTL_HOOKS_REG_NONZERO_REG_BITS reg_nonzero_bits_for_combine
|
464 |
|
|
|
465 |
|
|
#undef RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES
|
466 |
|
|
#define RTL_HOOKS_REG_NUM_SIGN_BIT_COPIES reg_num_sign_bit_copies_for_combine
|
467 |
|
|
|
468 |
|
|
#undef RTL_HOOKS_REG_TRUNCATED_TO_MODE
|
469 |
|
|
#define RTL_HOOKS_REG_TRUNCATED_TO_MODE reg_truncated_to_mode
|
470 |
|
|
|
471 |
|
|
static const struct rtl_hooks combine_rtl_hooks = RTL_HOOKS_INITIALIZER;
|
472 |
|
|
|
473 |
|
|
|
474 |
|
|
/* Try to split PATTERN found in INSN. This returns NULL_RTX if
|
475 |
|
|
PATTERN can not be split. Otherwise, it returns an insn sequence.
|
476 |
|
|
This is a wrapper around split_insns which ensures that the
|
477 |
|
|
reg_stat vector is made larger if the splitter creates a new
|
478 |
|
|
register. */
|
479 |
|
|
|
480 |
|
|
static rtx
|
481 |
|
|
combine_split_insns (rtx pattern, rtx insn)
|
482 |
|
|
{
|
483 |
|
|
rtx ret;
|
484 |
|
|
unsigned int nregs;
|
485 |
|
|
|
486 |
|
|
ret = split_insns (pattern, insn);
|
487 |
|
|
nregs = max_reg_num ();
|
488 |
|
|
if (nregs > VEC_length (reg_stat_type, reg_stat))
|
489 |
|
|
VEC_safe_grow_cleared (reg_stat_type, heap, reg_stat, nregs);
|
490 |
|
|
return ret;
|
491 |
|
|
}
|
492 |
|
|
|
493 |
|
|
/* This is used by find_single_use to locate an rtx in LOC that
|
494 |
|
|
contains exactly one use of DEST, which is typically either a REG
|
495 |
|
|
or CC0. It returns a pointer to the innermost rtx expression
|
496 |
|
|
containing DEST. Appearances of DEST that are being used to
|
497 |
|
|
totally replace it are not counted. */
|
498 |
|
|
|
499 |
|
|
static rtx *
|
500 |
|
|
find_single_use_1 (rtx dest, rtx *loc)
|
501 |
|
|
{
|
502 |
|
|
rtx x = *loc;
|
503 |
|
|
enum rtx_code code = GET_CODE (x);
|
504 |
|
|
rtx *result = NULL;
|
505 |
|
|
rtx *this_result;
|
506 |
|
|
int i;
|
507 |
|
|
const char *fmt;
|
508 |
|
|
|
509 |
|
|
switch (code)
|
510 |
|
|
{
|
511 |
|
|
case CONST_INT:
|
512 |
|
|
case CONST:
|
513 |
|
|
case LABEL_REF:
|
514 |
|
|
case SYMBOL_REF:
|
515 |
|
|
case CONST_DOUBLE:
|
516 |
|
|
case CONST_VECTOR:
|
517 |
|
|
case CLOBBER:
|
518 |
|
|
return 0;
|
519 |
|
|
|
520 |
|
|
case SET:
|
521 |
|
|
/* If the destination is anything other than CC0, PC, a REG or a SUBREG
|
522 |
|
|
of a REG that occupies all of the REG, the insn uses DEST if
|
523 |
|
|
it is mentioned in the destination or the source. Otherwise, we
|
524 |
|
|
need just check the source. */
|
525 |
|
|
if (GET_CODE (SET_DEST (x)) != CC0
|
526 |
|
|
&& GET_CODE (SET_DEST (x)) != PC
|
527 |
|
|
&& !REG_P (SET_DEST (x))
|
528 |
|
|
&& ! (GET_CODE (SET_DEST (x)) == SUBREG
|
529 |
|
|
&& REG_P (SUBREG_REG (SET_DEST (x)))
|
530 |
|
|
&& (((GET_MODE_SIZE (GET_MODE (SUBREG_REG (SET_DEST (x))))
|
531 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
|
532 |
|
|
== ((GET_MODE_SIZE (GET_MODE (SET_DEST (x)))
|
533 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD))))
|
534 |
|
|
break;
|
535 |
|
|
|
536 |
|
|
return find_single_use_1 (dest, &SET_SRC (x));
|
537 |
|
|
|
538 |
|
|
case MEM:
|
539 |
|
|
case SUBREG:
|
540 |
|
|
return find_single_use_1 (dest, &XEXP (x, 0));
|
541 |
|
|
|
542 |
|
|
default:
|
543 |
|
|
break;
|
544 |
|
|
}
|
545 |
|
|
|
546 |
|
|
/* If it wasn't one of the common cases above, check each expression and
|
547 |
|
|
vector of this code. Look for a unique usage of DEST. */
|
548 |
|
|
|
549 |
|
|
fmt = GET_RTX_FORMAT (code);
|
550 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
551 |
|
|
{
|
552 |
|
|
if (fmt[i] == 'e')
|
553 |
|
|
{
|
554 |
|
|
if (dest == XEXP (x, i)
|
555 |
|
|
|| (REG_P (dest) && REG_P (XEXP (x, i))
|
556 |
|
|
&& REGNO (dest) == REGNO (XEXP (x, i))))
|
557 |
|
|
this_result = loc;
|
558 |
|
|
else
|
559 |
|
|
this_result = find_single_use_1 (dest, &XEXP (x, i));
|
560 |
|
|
|
561 |
|
|
if (result == NULL)
|
562 |
|
|
result = this_result;
|
563 |
|
|
else if (this_result)
|
564 |
|
|
/* Duplicate usage. */
|
565 |
|
|
return NULL;
|
566 |
|
|
}
|
567 |
|
|
else if (fmt[i] == 'E')
|
568 |
|
|
{
|
569 |
|
|
int j;
|
570 |
|
|
|
571 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
572 |
|
|
{
|
573 |
|
|
if (XVECEXP (x, i, j) == dest
|
574 |
|
|
|| (REG_P (dest)
|
575 |
|
|
&& REG_P (XVECEXP (x, i, j))
|
576 |
|
|
&& REGNO (XVECEXP (x, i, j)) == REGNO (dest)))
|
577 |
|
|
this_result = loc;
|
578 |
|
|
else
|
579 |
|
|
this_result = find_single_use_1 (dest, &XVECEXP (x, i, j));
|
580 |
|
|
|
581 |
|
|
if (result == NULL)
|
582 |
|
|
result = this_result;
|
583 |
|
|
else if (this_result)
|
584 |
|
|
return NULL;
|
585 |
|
|
}
|
586 |
|
|
}
|
587 |
|
|
}
|
588 |
|
|
|
589 |
|
|
return result;
|
590 |
|
|
}
|
591 |
|
|
|
592 |
|
|
|
593 |
|
|
/* See if DEST, produced in INSN, is used only a single time in the
|
594 |
|
|
sequel. If so, return a pointer to the innermost rtx expression in which
|
595 |
|
|
it is used.
|
596 |
|
|
|
597 |
|
|
If PLOC is nonzero, *PLOC is set to the insn containing the single use.
|
598 |
|
|
|
599 |
|
|
If DEST is cc0_rtx, we look only at the next insn. In that case, we don't
|
600 |
|
|
care about REG_DEAD notes or LOG_LINKS.
|
601 |
|
|
|
602 |
|
|
Otherwise, we find the single use by finding an insn that has a
|
603 |
|
|
LOG_LINKS pointing at INSN and has a REG_DEAD note for DEST. If DEST is
|
604 |
|
|
only referenced once in that insn, we know that it must be the first
|
605 |
|
|
and last insn referencing DEST. */
|
606 |
|
|
|
607 |
|
|
static rtx *
|
608 |
|
|
find_single_use (rtx dest, rtx insn, rtx *ploc)
|
609 |
|
|
{
|
610 |
|
|
basic_block bb;
|
611 |
|
|
rtx next;
|
612 |
|
|
rtx *result;
|
613 |
|
|
rtx link;
|
614 |
|
|
|
615 |
|
|
#ifdef HAVE_cc0
|
616 |
|
|
if (dest == cc0_rtx)
|
617 |
|
|
{
|
618 |
|
|
next = NEXT_INSN (insn);
|
619 |
|
|
if (next == 0
|
620 |
|
|
|| (!NONJUMP_INSN_P (next) && !JUMP_P (next)))
|
621 |
|
|
return 0;
|
622 |
|
|
|
623 |
|
|
result = find_single_use_1 (dest, &PATTERN (next));
|
624 |
|
|
if (result && ploc)
|
625 |
|
|
*ploc = next;
|
626 |
|
|
return result;
|
627 |
|
|
}
|
628 |
|
|
#endif
|
629 |
|
|
|
630 |
|
|
if (!REG_P (dest))
|
631 |
|
|
return 0;
|
632 |
|
|
|
633 |
|
|
bb = BLOCK_FOR_INSN (insn);
|
634 |
|
|
for (next = NEXT_INSN (insn);
|
635 |
|
|
next && BLOCK_FOR_INSN (next) == bb;
|
636 |
|
|
next = NEXT_INSN (next))
|
637 |
|
|
if (INSN_P (next) && dead_or_set_p (next, dest))
|
638 |
|
|
{
|
639 |
|
|
for (link = LOG_LINKS (next); link; link = XEXP (link, 1))
|
640 |
|
|
if (XEXP (link, 0) == insn)
|
641 |
|
|
break;
|
642 |
|
|
|
643 |
|
|
if (link)
|
644 |
|
|
{
|
645 |
|
|
result = find_single_use_1 (dest, &PATTERN (next));
|
646 |
|
|
if (ploc)
|
647 |
|
|
*ploc = next;
|
648 |
|
|
return result;
|
649 |
|
|
}
|
650 |
|
|
}
|
651 |
|
|
|
652 |
|
|
return 0;
|
653 |
|
|
}
|
654 |
|
|
|
655 |
|
|
/* Substitute NEWVAL, an rtx expression, into INTO, a place in some
|
656 |
|
|
insn. The substitution can be undone by undo_all. If INTO is already
|
657 |
|
|
set to NEWVAL, do not record this change. Because computing NEWVAL might
|
658 |
|
|
also call SUBST, we have to compute it before we put anything into
|
659 |
|
|
the undo table. */
|
660 |
|
|
|
661 |
|
|
static void
|
662 |
|
|
do_SUBST (rtx *into, rtx newval)
|
663 |
|
|
{
|
664 |
|
|
struct undo *buf;
|
665 |
|
|
rtx oldval = *into;
|
666 |
|
|
|
667 |
|
|
if (oldval == newval)
|
668 |
|
|
return;
|
669 |
|
|
|
670 |
|
|
/* We'd like to catch as many invalid transformations here as
|
671 |
|
|
possible. Unfortunately, there are way too many mode changes
|
672 |
|
|
that are perfectly valid, so we'd waste too much effort for
|
673 |
|
|
little gain doing the checks here. Focus on catching invalid
|
674 |
|
|
transformations involving integer constants. */
|
675 |
|
|
if (GET_MODE_CLASS (GET_MODE (oldval)) == MODE_INT
|
676 |
|
|
&& CONST_INT_P (newval))
|
677 |
|
|
{
|
678 |
|
|
/* Sanity check that we're replacing oldval with a CONST_INT
|
679 |
|
|
that is a valid sign-extension for the original mode. */
|
680 |
|
|
gcc_assert (INTVAL (newval)
|
681 |
|
|
== trunc_int_for_mode (INTVAL (newval), GET_MODE (oldval)));
|
682 |
|
|
|
683 |
|
|
/* Replacing the operand of a SUBREG or a ZERO_EXTEND with a
|
684 |
|
|
CONST_INT is not valid, because after the replacement, the
|
685 |
|
|
original mode would be gone. Unfortunately, we can't tell
|
686 |
|
|
when do_SUBST is called to replace the operand thereof, so we
|
687 |
|
|
perform this test on oldval instead, checking whether an
|
688 |
|
|
invalid replacement took place before we got here. */
|
689 |
|
|
gcc_assert (!(GET_CODE (oldval) == SUBREG
|
690 |
|
|
&& CONST_INT_P (SUBREG_REG (oldval))));
|
691 |
|
|
gcc_assert (!(GET_CODE (oldval) == ZERO_EXTEND
|
692 |
|
|
&& CONST_INT_P (XEXP (oldval, 0))));
|
693 |
|
|
}
|
694 |
|
|
|
695 |
|
|
if (undobuf.frees)
|
696 |
|
|
buf = undobuf.frees, undobuf.frees = buf->next;
|
697 |
|
|
else
|
698 |
|
|
buf = XNEW (struct undo);
|
699 |
|
|
|
700 |
|
|
buf->kind = UNDO_RTX;
|
701 |
|
|
buf->where.r = into;
|
702 |
|
|
buf->old_contents.r = oldval;
|
703 |
|
|
*into = newval;
|
704 |
|
|
|
705 |
|
|
buf->next = undobuf.undos, undobuf.undos = buf;
|
706 |
|
|
}
|
707 |
|
|
|
708 |
|
|
#define SUBST(INTO, NEWVAL) do_SUBST(&(INTO), (NEWVAL))
|
709 |
|
|
|
710 |
|
|
/* Similar to SUBST, but NEWVAL is an int expression. Note that substitution
|
711 |
|
|
for the value of a HOST_WIDE_INT value (including CONST_INT) is
|
712 |
|
|
not safe. */
|
713 |
|
|
|
714 |
|
|
static void
|
715 |
|
|
do_SUBST_INT (int *into, int newval)
|
716 |
|
|
{
|
717 |
|
|
struct undo *buf;
|
718 |
|
|
int oldval = *into;
|
719 |
|
|
|
720 |
|
|
if (oldval == newval)
|
721 |
|
|
return;
|
722 |
|
|
|
723 |
|
|
if (undobuf.frees)
|
724 |
|
|
buf = undobuf.frees, undobuf.frees = buf->next;
|
725 |
|
|
else
|
726 |
|
|
buf = XNEW (struct undo);
|
727 |
|
|
|
728 |
|
|
buf->kind = UNDO_INT;
|
729 |
|
|
buf->where.i = into;
|
730 |
|
|
buf->old_contents.i = oldval;
|
731 |
|
|
*into = newval;
|
732 |
|
|
|
733 |
|
|
buf->next = undobuf.undos, undobuf.undos = buf;
|
734 |
|
|
}
|
735 |
|
|
|
736 |
|
|
#define SUBST_INT(INTO, NEWVAL) do_SUBST_INT(&(INTO), (NEWVAL))
|
737 |
|
|
|
738 |
|
|
/* Similar to SUBST, but just substitute the mode. This is used when
|
739 |
|
|
changing the mode of a pseudo-register, so that any other
|
740 |
|
|
references to the entry in the regno_reg_rtx array will change as
|
741 |
|
|
well. */
|
742 |
|
|
|
743 |
|
|
static void
|
744 |
|
|
do_SUBST_MODE (rtx *into, enum machine_mode newval)
|
745 |
|
|
{
|
746 |
|
|
struct undo *buf;
|
747 |
|
|
enum machine_mode oldval = GET_MODE (*into);
|
748 |
|
|
|
749 |
|
|
if (oldval == newval)
|
750 |
|
|
return;
|
751 |
|
|
|
752 |
|
|
if (undobuf.frees)
|
753 |
|
|
buf = undobuf.frees, undobuf.frees = buf->next;
|
754 |
|
|
else
|
755 |
|
|
buf = XNEW (struct undo);
|
756 |
|
|
|
757 |
|
|
buf->kind = UNDO_MODE;
|
758 |
|
|
buf->where.r = into;
|
759 |
|
|
buf->old_contents.m = oldval;
|
760 |
|
|
adjust_reg_mode (*into, newval);
|
761 |
|
|
|
762 |
|
|
buf->next = undobuf.undos, undobuf.undos = buf;
|
763 |
|
|
}
|
764 |
|
|
|
765 |
|
|
#define SUBST_MODE(INTO, NEWVAL) do_SUBST_MODE(&(INTO), (NEWVAL))
|
766 |
|
|
|
767 |
|
|
/* Subroutine of try_combine. Determine whether the combine replacement
|
768 |
|
|
patterns NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to
|
769 |
|
|
insn_rtx_cost that the original instruction sequence I1, I2, I3 and
|
770 |
|
|
undobuf.other_insn. Note that I1 and/or NEWI2PAT may be NULL_RTX.
|
771 |
|
|
NEWOTHERPAT and undobuf.other_insn may also both be NULL_RTX. This
|
772 |
|
|
function returns false, if the costs of all instructions can be
|
773 |
|
|
estimated, and the replacements are more expensive than the original
|
774 |
|
|
sequence. */
|
775 |
|
|
|
776 |
|
|
static bool
|
777 |
|
|
combine_validate_cost (rtx i1, rtx i2, rtx i3, rtx newpat, rtx newi2pat,
|
778 |
|
|
rtx newotherpat)
|
779 |
|
|
{
|
780 |
|
|
int i1_cost, i2_cost, i3_cost;
|
781 |
|
|
int new_i2_cost, new_i3_cost;
|
782 |
|
|
int old_cost, new_cost;
|
783 |
|
|
|
784 |
|
|
/* Lookup the original insn_rtx_costs. */
|
785 |
|
|
i2_cost = INSN_COST (i2);
|
786 |
|
|
i3_cost = INSN_COST (i3);
|
787 |
|
|
|
788 |
|
|
if (i1)
|
789 |
|
|
{
|
790 |
|
|
i1_cost = INSN_COST (i1);
|
791 |
|
|
old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0)
|
792 |
|
|
? i1_cost + i2_cost + i3_cost : 0;
|
793 |
|
|
}
|
794 |
|
|
else
|
795 |
|
|
{
|
796 |
|
|
old_cost = (i2_cost > 0 && i3_cost > 0) ? i2_cost + i3_cost : 0;
|
797 |
|
|
i1_cost = 0;
|
798 |
|
|
}
|
799 |
|
|
|
800 |
|
|
/* Calculate the replacement insn_rtx_costs. */
|
801 |
|
|
new_i3_cost = insn_rtx_cost (newpat, optimize_this_for_speed_p);
|
802 |
|
|
if (newi2pat)
|
803 |
|
|
{
|
804 |
|
|
new_i2_cost = insn_rtx_cost (newi2pat, optimize_this_for_speed_p);
|
805 |
|
|
new_cost = (new_i2_cost > 0 && new_i3_cost > 0)
|
806 |
|
|
? new_i2_cost + new_i3_cost : 0;
|
807 |
|
|
}
|
808 |
|
|
else
|
809 |
|
|
{
|
810 |
|
|
new_cost = new_i3_cost;
|
811 |
|
|
new_i2_cost = 0;
|
812 |
|
|
}
|
813 |
|
|
|
814 |
|
|
if (undobuf.other_insn)
|
815 |
|
|
{
|
816 |
|
|
int old_other_cost, new_other_cost;
|
817 |
|
|
|
818 |
|
|
old_other_cost = INSN_COST (undobuf.other_insn);
|
819 |
|
|
new_other_cost = insn_rtx_cost (newotherpat, optimize_this_for_speed_p);
|
820 |
|
|
if (old_other_cost > 0 && new_other_cost > 0)
|
821 |
|
|
{
|
822 |
|
|
old_cost += old_other_cost;
|
823 |
|
|
new_cost += new_other_cost;
|
824 |
|
|
}
|
825 |
|
|
else
|
826 |
|
|
old_cost = 0;
|
827 |
|
|
}
|
828 |
|
|
|
829 |
|
|
/* Disallow this recombination if both new_cost and old_cost are
|
830 |
|
|
greater than zero, and new_cost is greater than old cost. */
|
831 |
|
|
if (old_cost > 0
|
832 |
|
|
&& new_cost > old_cost)
|
833 |
|
|
{
|
834 |
|
|
if (dump_file)
|
835 |
|
|
{
|
836 |
|
|
if (i1)
|
837 |
|
|
{
|
838 |
|
|
fprintf (dump_file,
|
839 |
|
|
"rejecting combination of insns %d, %d and %d\n",
|
840 |
|
|
INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
|
841 |
|
|
fprintf (dump_file, "original costs %d + %d + %d = %d\n",
|
842 |
|
|
i1_cost, i2_cost, i3_cost, old_cost);
|
843 |
|
|
}
|
844 |
|
|
else
|
845 |
|
|
{
|
846 |
|
|
fprintf (dump_file,
|
847 |
|
|
"rejecting combination of insns %d and %d\n",
|
848 |
|
|
INSN_UID (i2), INSN_UID (i3));
|
849 |
|
|
fprintf (dump_file, "original costs %d + %d = %d\n",
|
850 |
|
|
i2_cost, i3_cost, old_cost);
|
851 |
|
|
}
|
852 |
|
|
|
853 |
|
|
if (newi2pat)
|
854 |
|
|
{
|
855 |
|
|
fprintf (dump_file, "replacement costs %d + %d = %d\n",
|
856 |
|
|
new_i2_cost, new_i3_cost, new_cost);
|
857 |
|
|
}
|
858 |
|
|
else
|
859 |
|
|
fprintf (dump_file, "replacement cost %d\n", new_cost);
|
860 |
|
|
}
|
861 |
|
|
|
862 |
|
|
return false;
|
863 |
|
|
}
|
864 |
|
|
|
865 |
|
|
/* Update the uid_insn_cost array with the replacement costs. */
|
866 |
|
|
INSN_COST (i2) = new_i2_cost;
|
867 |
|
|
INSN_COST (i3) = new_i3_cost;
|
868 |
|
|
if (i1)
|
869 |
|
|
INSN_COST (i1) = 0;
|
870 |
|
|
|
871 |
|
|
return true;
|
872 |
|
|
}
|
873 |
|
|
|
874 |
|
|
|
875 |
|
|
/* Delete any insns that copy a register to itself. */
|
876 |
|
|
|
877 |
|
|
static void
|
878 |
|
|
delete_noop_moves (void)
|
879 |
|
|
{
|
880 |
|
|
rtx insn, next;
|
881 |
|
|
basic_block bb;
|
882 |
|
|
|
883 |
|
|
FOR_EACH_BB (bb)
|
884 |
|
|
{
|
885 |
|
|
for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
|
886 |
|
|
{
|
887 |
|
|
next = NEXT_INSN (insn);
|
888 |
|
|
if (INSN_P (insn) && noop_move_p (insn))
|
889 |
|
|
{
|
890 |
|
|
if (dump_file)
|
891 |
|
|
fprintf (dump_file, "deleting noop move %d\n", INSN_UID (insn));
|
892 |
|
|
|
893 |
|
|
delete_insn_and_edges (insn);
|
894 |
|
|
}
|
895 |
|
|
}
|
896 |
|
|
}
|
897 |
|
|
}
|
898 |
|
|
|
899 |
|
|
|
900 |
|
|
/* Fill in log links field for all insns. */
|
901 |
|
|
|
902 |
|
|
static void
|
903 |
|
|
create_log_links (void)
|
904 |
|
|
{
|
905 |
|
|
basic_block bb;
|
906 |
|
|
rtx *next_use, insn;
|
907 |
|
|
df_ref *def_vec, *use_vec;
|
908 |
|
|
|
909 |
|
|
next_use = XCNEWVEC (rtx, max_reg_num ());
|
910 |
|
|
|
911 |
|
|
/* Pass through each block from the end, recording the uses of each
|
912 |
|
|
register and establishing log links when def is encountered.
|
913 |
|
|
Note that we do not clear next_use array in order to save time,
|
914 |
|
|
so we have to test whether the use is in the same basic block as def.
|
915 |
|
|
|
916 |
|
|
There are a few cases below when we do not consider the definition or
|
917 |
|
|
usage -- these are taken from original flow.c did. Don't ask me why it is
|
918 |
|
|
done this way; I don't know and if it works, I don't want to know. */
|
919 |
|
|
|
920 |
|
|
FOR_EACH_BB (bb)
|
921 |
|
|
{
|
922 |
|
|
FOR_BB_INSNS_REVERSE (bb, insn)
|
923 |
|
|
{
|
924 |
|
|
if (!NONDEBUG_INSN_P (insn))
|
925 |
|
|
continue;
|
926 |
|
|
|
927 |
|
|
/* Log links are created only once. */
|
928 |
|
|
gcc_assert (!LOG_LINKS (insn));
|
929 |
|
|
|
930 |
|
|
for (def_vec = DF_INSN_DEFS (insn); *def_vec; def_vec++)
|
931 |
|
|
{
|
932 |
|
|
df_ref def = *def_vec;
|
933 |
|
|
int regno = DF_REF_REGNO (def);
|
934 |
|
|
rtx use_insn;
|
935 |
|
|
|
936 |
|
|
if (!next_use[regno])
|
937 |
|
|
continue;
|
938 |
|
|
|
939 |
|
|
/* Do not consider if it is pre/post modification in MEM. */
|
940 |
|
|
if (DF_REF_FLAGS (def) & DF_REF_PRE_POST_MODIFY)
|
941 |
|
|
continue;
|
942 |
|
|
|
943 |
|
|
/* Do not make the log link for frame pointer. */
|
944 |
|
|
if ((regno == FRAME_POINTER_REGNUM
|
945 |
|
|
&& (! reload_completed || frame_pointer_needed))
|
946 |
|
|
#if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
|
947 |
|
|
|| (regno == HARD_FRAME_POINTER_REGNUM
|
948 |
|
|
&& (! reload_completed || frame_pointer_needed))
|
949 |
|
|
#endif
|
950 |
|
|
#if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
|
951 |
|
|
|| (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
|
952 |
|
|
#endif
|
953 |
|
|
)
|
954 |
|
|
continue;
|
955 |
|
|
|
956 |
|
|
use_insn = next_use[regno];
|
957 |
|
|
if (BLOCK_FOR_INSN (use_insn) == bb)
|
958 |
|
|
{
|
959 |
|
|
/* flow.c claimed:
|
960 |
|
|
|
961 |
|
|
We don't build a LOG_LINK for hard registers contained
|
962 |
|
|
in ASM_OPERANDs. If these registers get replaced,
|
963 |
|
|
we might wind up changing the semantics of the insn,
|
964 |
|
|
even if reload can make what appear to be valid
|
965 |
|
|
assignments later. */
|
966 |
|
|
if (regno >= FIRST_PSEUDO_REGISTER
|
967 |
|
|
|| asm_noperands (PATTERN (use_insn)) < 0)
|
968 |
|
|
{
|
969 |
|
|
/* Don't add duplicate links between instructions. */
|
970 |
|
|
rtx links;
|
971 |
|
|
for (links = LOG_LINKS (use_insn); links;
|
972 |
|
|
links = XEXP (links, 1))
|
973 |
|
|
if (insn == XEXP (links, 0))
|
974 |
|
|
break;
|
975 |
|
|
|
976 |
|
|
if (!links)
|
977 |
|
|
LOG_LINKS (use_insn) =
|
978 |
|
|
alloc_INSN_LIST (insn, LOG_LINKS (use_insn));
|
979 |
|
|
}
|
980 |
|
|
}
|
981 |
|
|
next_use[regno] = NULL_RTX;
|
982 |
|
|
}
|
983 |
|
|
|
984 |
|
|
for (use_vec = DF_INSN_USES (insn); *use_vec; use_vec++)
|
985 |
|
|
{
|
986 |
|
|
df_ref use = *use_vec;
|
987 |
|
|
int regno = DF_REF_REGNO (use);
|
988 |
|
|
|
989 |
|
|
/* Do not consider the usage of the stack pointer
|
990 |
|
|
by function call. */
|
991 |
|
|
if (DF_REF_FLAGS (use) & DF_REF_CALL_STACK_USAGE)
|
992 |
|
|
continue;
|
993 |
|
|
|
994 |
|
|
next_use[regno] = insn;
|
995 |
|
|
}
|
996 |
|
|
}
|
997 |
|
|
}
|
998 |
|
|
|
999 |
|
|
free (next_use);
|
1000 |
|
|
}
|
1001 |
|
|
|
1002 |
|
|
/* Clear LOG_LINKS fields of insns. */
|
1003 |
|
|
|
1004 |
|
|
static void
|
1005 |
|
|
clear_log_links (void)
|
1006 |
|
|
{
|
1007 |
|
|
rtx insn;
|
1008 |
|
|
|
1009 |
|
|
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
|
1010 |
|
|
if (INSN_P (insn))
|
1011 |
|
|
free_INSN_LIST_list (&LOG_LINKS (insn));
|
1012 |
|
|
}
|
1013 |
|
|
|
1014 |
|
|
/* Main entry point for combiner. F is the first insn of the function.
|
1015 |
|
|
NREGS is the first unused pseudo-reg number.
|
1016 |
|
|
|
1017 |
|
|
Return nonzero if the combiner has turned an indirect jump
|
1018 |
|
|
instruction into a direct jump. */
|
1019 |
|
|
static int
|
1020 |
|
|
combine_instructions (rtx f, unsigned int nregs)
|
1021 |
|
|
{
|
1022 |
|
|
rtx insn, next;
|
1023 |
|
|
#ifdef HAVE_cc0
|
1024 |
|
|
rtx prev;
|
1025 |
|
|
#endif
|
1026 |
|
|
rtx links, nextlinks;
|
1027 |
|
|
rtx first;
|
1028 |
|
|
basic_block last_bb;
|
1029 |
|
|
|
1030 |
|
|
int new_direct_jump_p = 0;
|
1031 |
|
|
|
1032 |
|
|
for (first = f; first && !INSN_P (first); )
|
1033 |
|
|
first = NEXT_INSN (first);
|
1034 |
|
|
if (!first)
|
1035 |
|
|
return 0;
|
1036 |
|
|
|
1037 |
|
|
combine_attempts = 0;
|
1038 |
|
|
combine_merges = 0;
|
1039 |
|
|
combine_extras = 0;
|
1040 |
|
|
combine_successes = 0;
|
1041 |
|
|
|
1042 |
|
|
rtl_hooks = combine_rtl_hooks;
|
1043 |
|
|
|
1044 |
|
|
VEC_safe_grow_cleared (reg_stat_type, heap, reg_stat, nregs);
|
1045 |
|
|
|
1046 |
|
|
init_recog_no_volatile ();
|
1047 |
|
|
|
1048 |
|
|
/* Allocate array for insn info. */
|
1049 |
|
|
max_uid_known = get_max_uid ();
|
1050 |
|
|
uid_log_links = XCNEWVEC (rtx, max_uid_known + 1);
|
1051 |
|
|
uid_insn_cost = XCNEWVEC (int, max_uid_known + 1);
|
1052 |
|
|
|
1053 |
|
|
nonzero_bits_mode = mode_for_size (HOST_BITS_PER_WIDE_INT, MODE_INT, 0);
|
1054 |
|
|
|
1055 |
|
|
/* Don't use reg_stat[].nonzero_bits when computing it. This can cause
|
1056 |
|
|
problems when, for example, we have j <<= 1 in a loop. */
|
1057 |
|
|
|
1058 |
|
|
nonzero_sign_valid = 0;
|
1059 |
|
|
label_tick = label_tick_ebb_start = 1;
|
1060 |
|
|
|
1061 |
|
|
/* Scan all SETs and see if we can deduce anything about what
|
1062 |
|
|
bits are known to be zero for some registers and how many copies
|
1063 |
|
|
of the sign bit are known to exist for those registers.
|
1064 |
|
|
|
1065 |
|
|
Also set any known values so that we can use it while searching
|
1066 |
|
|
for what bits are known to be set. */
|
1067 |
|
|
|
1068 |
|
|
setup_incoming_promotions (first);
|
1069 |
|
|
/* Allow the entry block and the first block to fall into the same EBB.
|
1070 |
|
|
Conceptually the incoming promotions are assigned to the entry block. */
|
1071 |
|
|
last_bb = ENTRY_BLOCK_PTR;
|
1072 |
|
|
|
1073 |
|
|
create_log_links ();
|
1074 |
|
|
FOR_EACH_BB (this_basic_block)
|
1075 |
|
|
{
|
1076 |
|
|
optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
|
1077 |
|
|
last_call_luid = 0;
|
1078 |
|
|
mem_last_set = -1;
|
1079 |
|
|
|
1080 |
|
|
label_tick++;
|
1081 |
|
|
if (!single_pred_p (this_basic_block)
|
1082 |
|
|
|| single_pred (this_basic_block) != last_bb)
|
1083 |
|
|
label_tick_ebb_start = label_tick;
|
1084 |
|
|
last_bb = this_basic_block;
|
1085 |
|
|
|
1086 |
|
|
FOR_BB_INSNS (this_basic_block, insn)
|
1087 |
|
|
if (INSN_P (insn) && BLOCK_FOR_INSN (insn))
|
1088 |
|
|
{
|
1089 |
|
|
subst_low_luid = DF_INSN_LUID (insn);
|
1090 |
|
|
subst_insn = insn;
|
1091 |
|
|
|
1092 |
|
|
note_stores (PATTERN (insn), set_nonzero_bits_and_sign_copies,
|
1093 |
|
|
insn);
|
1094 |
|
|
record_dead_and_set_regs (insn);
|
1095 |
|
|
|
1096 |
|
|
#ifdef AUTO_INC_DEC
|
1097 |
|
|
for (links = REG_NOTES (insn); links; links = XEXP (links, 1))
|
1098 |
|
|
if (REG_NOTE_KIND (links) == REG_INC)
|
1099 |
|
|
set_nonzero_bits_and_sign_copies (XEXP (links, 0), NULL_RTX,
|
1100 |
|
|
insn);
|
1101 |
|
|
#endif
|
1102 |
|
|
|
1103 |
|
|
/* Record the current insn_rtx_cost of this instruction. */
|
1104 |
|
|
if (NONJUMP_INSN_P (insn))
|
1105 |
|
|
INSN_COST (insn) = insn_rtx_cost (PATTERN (insn),
|
1106 |
|
|
optimize_this_for_speed_p);
|
1107 |
|
|
if (dump_file)
|
1108 |
|
|
fprintf(dump_file, "insn_cost %d: %d\n",
|
1109 |
|
|
INSN_UID (insn), INSN_COST (insn));
|
1110 |
|
|
}
|
1111 |
|
|
}
|
1112 |
|
|
|
1113 |
|
|
nonzero_sign_valid = 1;
|
1114 |
|
|
|
1115 |
|
|
/* Now scan all the insns in forward order. */
|
1116 |
|
|
label_tick = label_tick_ebb_start = 1;
|
1117 |
|
|
init_reg_last ();
|
1118 |
|
|
setup_incoming_promotions (first);
|
1119 |
|
|
last_bb = ENTRY_BLOCK_PTR;
|
1120 |
|
|
|
1121 |
|
|
FOR_EACH_BB (this_basic_block)
|
1122 |
|
|
{
|
1123 |
|
|
optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
|
1124 |
|
|
last_call_luid = 0;
|
1125 |
|
|
mem_last_set = -1;
|
1126 |
|
|
|
1127 |
|
|
label_tick++;
|
1128 |
|
|
if (!single_pred_p (this_basic_block)
|
1129 |
|
|
|| single_pred (this_basic_block) != last_bb)
|
1130 |
|
|
label_tick_ebb_start = label_tick;
|
1131 |
|
|
last_bb = this_basic_block;
|
1132 |
|
|
|
1133 |
|
|
rtl_profile_for_bb (this_basic_block);
|
1134 |
|
|
for (insn = BB_HEAD (this_basic_block);
|
1135 |
|
|
insn != NEXT_INSN (BB_END (this_basic_block));
|
1136 |
|
|
insn = next ? next : NEXT_INSN (insn))
|
1137 |
|
|
{
|
1138 |
|
|
next = 0;
|
1139 |
|
|
if (NONDEBUG_INSN_P (insn))
|
1140 |
|
|
{
|
1141 |
|
|
/* See if we know about function return values before this
|
1142 |
|
|
insn based upon SUBREG flags. */
|
1143 |
|
|
check_promoted_subreg (insn, PATTERN (insn));
|
1144 |
|
|
|
1145 |
|
|
/* See if we can find hardregs and subreg of pseudos in
|
1146 |
|
|
narrower modes. This could help turning TRUNCATEs
|
1147 |
|
|
into SUBREGs. */
|
1148 |
|
|
note_uses (&PATTERN (insn), record_truncated_values, NULL);
|
1149 |
|
|
|
1150 |
|
|
/* Try this insn with each insn it links back to. */
|
1151 |
|
|
|
1152 |
|
|
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
|
1153 |
|
|
if ((next = try_combine (insn, XEXP (links, 0),
|
1154 |
|
|
NULL_RTX, &new_direct_jump_p)) != 0)
|
1155 |
|
|
goto retry;
|
1156 |
|
|
|
1157 |
|
|
/* Try each sequence of three linked insns ending with this one. */
|
1158 |
|
|
|
1159 |
|
|
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
|
1160 |
|
|
{
|
1161 |
|
|
rtx link = XEXP (links, 0);
|
1162 |
|
|
|
1163 |
|
|
/* If the linked insn has been replaced by a note, then there
|
1164 |
|
|
is no point in pursuing this chain any further. */
|
1165 |
|
|
if (NOTE_P (link))
|
1166 |
|
|
continue;
|
1167 |
|
|
|
1168 |
|
|
for (nextlinks = LOG_LINKS (link);
|
1169 |
|
|
nextlinks;
|
1170 |
|
|
nextlinks = XEXP (nextlinks, 1))
|
1171 |
|
|
if ((next = try_combine (insn, link,
|
1172 |
|
|
XEXP (nextlinks, 0),
|
1173 |
|
|
&new_direct_jump_p)) != 0)
|
1174 |
|
|
goto retry;
|
1175 |
|
|
}
|
1176 |
|
|
|
1177 |
|
|
#ifdef HAVE_cc0
|
1178 |
|
|
/* Try to combine a jump insn that uses CC0
|
1179 |
|
|
with a preceding insn that sets CC0, and maybe with its
|
1180 |
|
|
logical predecessor as well.
|
1181 |
|
|
This is how we make decrement-and-branch insns.
|
1182 |
|
|
We need this special code because data flow connections
|
1183 |
|
|
via CC0 do not get entered in LOG_LINKS. */
|
1184 |
|
|
|
1185 |
|
|
if (JUMP_P (insn)
|
1186 |
|
|
&& (prev = prev_nonnote_insn (insn)) != 0
|
1187 |
|
|
&& NONJUMP_INSN_P (prev)
|
1188 |
|
|
&& sets_cc0_p (PATTERN (prev)))
|
1189 |
|
|
{
|
1190 |
|
|
if ((next = try_combine (insn, prev,
|
1191 |
|
|
NULL_RTX, &new_direct_jump_p)) != 0)
|
1192 |
|
|
goto retry;
|
1193 |
|
|
|
1194 |
|
|
for (nextlinks = LOG_LINKS (prev); nextlinks;
|
1195 |
|
|
nextlinks = XEXP (nextlinks, 1))
|
1196 |
|
|
if ((next = try_combine (insn, prev,
|
1197 |
|
|
XEXP (nextlinks, 0),
|
1198 |
|
|
&new_direct_jump_p)) != 0)
|
1199 |
|
|
goto retry;
|
1200 |
|
|
}
|
1201 |
|
|
|
1202 |
|
|
/* Do the same for an insn that explicitly references CC0. */
|
1203 |
|
|
if (NONJUMP_INSN_P (insn)
|
1204 |
|
|
&& (prev = prev_nonnote_insn (insn)) != 0
|
1205 |
|
|
&& NONJUMP_INSN_P (prev)
|
1206 |
|
|
&& sets_cc0_p (PATTERN (prev))
|
1207 |
|
|
&& GET_CODE (PATTERN (insn)) == SET
|
1208 |
|
|
&& reg_mentioned_p (cc0_rtx, SET_SRC (PATTERN (insn))))
|
1209 |
|
|
{
|
1210 |
|
|
if ((next = try_combine (insn, prev,
|
1211 |
|
|
NULL_RTX, &new_direct_jump_p)) != 0)
|
1212 |
|
|
goto retry;
|
1213 |
|
|
|
1214 |
|
|
for (nextlinks = LOG_LINKS (prev); nextlinks;
|
1215 |
|
|
nextlinks = XEXP (nextlinks, 1))
|
1216 |
|
|
if ((next = try_combine (insn, prev,
|
1217 |
|
|
XEXP (nextlinks, 0),
|
1218 |
|
|
&new_direct_jump_p)) != 0)
|
1219 |
|
|
goto retry;
|
1220 |
|
|
}
|
1221 |
|
|
|
1222 |
|
|
/* Finally, see if any of the insns that this insn links to
|
1223 |
|
|
explicitly references CC0. If so, try this insn, that insn,
|
1224 |
|
|
and its predecessor if it sets CC0. */
|
1225 |
|
|
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
|
1226 |
|
|
if (NONJUMP_INSN_P (XEXP (links, 0))
|
1227 |
|
|
&& GET_CODE (PATTERN (XEXP (links, 0))) == SET
|
1228 |
|
|
&& reg_mentioned_p (cc0_rtx, SET_SRC (PATTERN (XEXP (links, 0))))
|
1229 |
|
|
&& (prev = prev_nonnote_insn (XEXP (links, 0))) != 0
|
1230 |
|
|
&& NONJUMP_INSN_P (prev)
|
1231 |
|
|
&& sets_cc0_p (PATTERN (prev))
|
1232 |
|
|
&& (next = try_combine (insn, XEXP (links, 0),
|
1233 |
|
|
prev, &new_direct_jump_p)) != 0)
|
1234 |
|
|
goto retry;
|
1235 |
|
|
#endif
|
1236 |
|
|
|
1237 |
|
|
/* Try combining an insn with two different insns whose results it
|
1238 |
|
|
uses. */
|
1239 |
|
|
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
|
1240 |
|
|
for (nextlinks = XEXP (links, 1); nextlinks;
|
1241 |
|
|
nextlinks = XEXP (nextlinks, 1))
|
1242 |
|
|
if ((next = try_combine (insn, XEXP (links, 0),
|
1243 |
|
|
XEXP (nextlinks, 0),
|
1244 |
|
|
&new_direct_jump_p)) != 0)
|
1245 |
|
|
goto retry;
|
1246 |
|
|
|
1247 |
|
|
/* Try this insn with each REG_EQUAL note it links back to. */
|
1248 |
|
|
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
|
1249 |
|
|
{
|
1250 |
|
|
rtx set, note;
|
1251 |
|
|
rtx temp = XEXP (links, 0);
|
1252 |
|
|
if ((set = single_set (temp)) != 0
|
1253 |
|
|
&& (note = find_reg_equal_equiv_note (temp)) != 0
|
1254 |
|
|
&& (note = XEXP (note, 0), GET_CODE (note)) != EXPR_LIST
|
1255 |
|
|
/* Avoid using a register that may already been marked
|
1256 |
|
|
dead by an earlier instruction. */
|
1257 |
|
|
&& ! unmentioned_reg_p (note, SET_SRC (set))
|
1258 |
|
|
&& (GET_MODE (note) == VOIDmode
|
1259 |
|
|
? SCALAR_INT_MODE_P (GET_MODE (SET_DEST (set)))
|
1260 |
|
|
: GET_MODE (SET_DEST (set)) == GET_MODE (note)))
|
1261 |
|
|
{
|
1262 |
|
|
/* Temporarily replace the set's source with the
|
1263 |
|
|
contents of the REG_EQUAL note. The insn will
|
1264 |
|
|
be deleted or recognized by try_combine. */
|
1265 |
|
|
rtx orig = SET_SRC (set);
|
1266 |
|
|
SET_SRC (set) = note;
|
1267 |
|
|
i2mod = temp;
|
1268 |
|
|
i2mod_old_rhs = copy_rtx (orig);
|
1269 |
|
|
i2mod_new_rhs = copy_rtx (note);
|
1270 |
|
|
next = try_combine (insn, i2mod, NULL_RTX,
|
1271 |
|
|
&new_direct_jump_p);
|
1272 |
|
|
i2mod = NULL_RTX;
|
1273 |
|
|
if (next)
|
1274 |
|
|
goto retry;
|
1275 |
|
|
SET_SRC (set) = orig;
|
1276 |
|
|
}
|
1277 |
|
|
}
|
1278 |
|
|
|
1279 |
|
|
if (!NOTE_P (insn))
|
1280 |
|
|
record_dead_and_set_regs (insn);
|
1281 |
|
|
|
1282 |
|
|
retry:
|
1283 |
|
|
;
|
1284 |
|
|
}
|
1285 |
|
|
}
|
1286 |
|
|
}
|
1287 |
|
|
|
1288 |
|
|
default_rtl_profile ();
|
1289 |
|
|
clear_log_links ();
|
1290 |
|
|
clear_bb_flags ();
|
1291 |
|
|
new_direct_jump_p |= purge_all_dead_edges ();
|
1292 |
|
|
delete_noop_moves ();
|
1293 |
|
|
|
1294 |
|
|
/* Clean up. */
|
1295 |
|
|
free (uid_log_links);
|
1296 |
|
|
free (uid_insn_cost);
|
1297 |
|
|
VEC_free (reg_stat_type, heap, reg_stat);
|
1298 |
|
|
|
1299 |
|
|
{
|
1300 |
|
|
struct undo *undo, *next;
|
1301 |
|
|
for (undo = undobuf.frees; undo; undo = next)
|
1302 |
|
|
{
|
1303 |
|
|
next = undo->next;
|
1304 |
|
|
free (undo);
|
1305 |
|
|
}
|
1306 |
|
|
undobuf.frees = 0;
|
1307 |
|
|
}
|
1308 |
|
|
|
1309 |
|
|
total_attempts += combine_attempts;
|
1310 |
|
|
total_merges += combine_merges;
|
1311 |
|
|
total_extras += combine_extras;
|
1312 |
|
|
total_successes += combine_successes;
|
1313 |
|
|
|
1314 |
|
|
nonzero_sign_valid = 0;
|
1315 |
|
|
rtl_hooks = general_rtl_hooks;
|
1316 |
|
|
|
1317 |
|
|
/* Make recognizer allow volatile MEMs again. */
|
1318 |
|
|
init_recog ();
|
1319 |
|
|
|
1320 |
|
|
return new_direct_jump_p;
|
1321 |
|
|
}
|
1322 |
|
|
|
1323 |
|
|
/* Wipe the last_xxx fields of reg_stat in preparation for another pass. */
|
1324 |
|
|
|
1325 |
|
|
static void
|
1326 |
|
|
init_reg_last (void)
|
1327 |
|
|
{
|
1328 |
|
|
unsigned int i;
|
1329 |
|
|
reg_stat_type *p;
|
1330 |
|
|
|
1331 |
|
|
for (i = 0; VEC_iterate (reg_stat_type, reg_stat, i, p); ++i)
|
1332 |
|
|
memset (p, 0, offsetof (reg_stat_type, sign_bit_copies));
|
1333 |
|
|
}
|
1334 |
|
|
|
1335 |
|
|
/* Set up any promoted values for incoming argument registers. */
|
1336 |
|
|
|
1337 |
|
|
static void
|
1338 |
|
|
setup_incoming_promotions (rtx first)
|
1339 |
|
|
{
|
1340 |
|
|
tree arg;
|
1341 |
|
|
bool strictly_local = false;
|
1342 |
|
|
|
1343 |
|
|
for (arg = DECL_ARGUMENTS (current_function_decl); arg;
|
1344 |
|
|
arg = TREE_CHAIN (arg))
|
1345 |
|
|
{
|
1346 |
|
|
rtx x, reg = DECL_INCOMING_RTL (arg);
|
1347 |
|
|
int uns1, uns3;
|
1348 |
|
|
enum machine_mode mode1, mode2, mode3, mode4;
|
1349 |
|
|
|
1350 |
|
|
/* Only continue if the incoming argument is in a register. */
|
1351 |
|
|
if (!REG_P (reg))
|
1352 |
|
|
continue;
|
1353 |
|
|
|
1354 |
|
|
/* Determine, if possible, whether all call sites of the current
|
1355 |
|
|
function lie within the current compilation unit. (This does
|
1356 |
|
|
take into account the exporting of a function via taking its
|
1357 |
|
|
address, and so forth.) */
|
1358 |
|
|
strictly_local = cgraph_local_info (current_function_decl)->local;
|
1359 |
|
|
|
1360 |
|
|
/* The mode and signedness of the argument before any promotions happen
|
1361 |
|
|
(equal to the mode of the pseudo holding it at that stage). */
|
1362 |
|
|
mode1 = TYPE_MODE (TREE_TYPE (arg));
|
1363 |
|
|
uns1 = TYPE_UNSIGNED (TREE_TYPE (arg));
|
1364 |
|
|
|
1365 |
|
|
/* The mode and signedness of the argument after any source language and
|
1366 |
|
|
TARGET_PROMOTE_PROTOTYPES-driven promotions. */
|
1367 |
|
|
mode2 = TYPE_MODE (DECL_ARG_TYPE (arg));
|
1368 |
|
|
uns3 = TYPE_UNSIGNED (DECL_ARG_TYPE (arg));
|
1369 |
|
|
|
1370 |
|
|
/* The mode and signedness of the argument as it is actually passed,
|
1371 |
|
|
after any TARGET_PROMOTE_FUNCTION_ARGS-driven ABI promotions. */
|
1372 |
|
|
mode3 = promote_function_mode (DECL_ARG_TYPE (arg), mode2, &uns3,
|
1373 |
|
|
TREE_TYPE (cfun->decl), 0);
|
1374 |
|
|
|
1375 |
|
|
/* The mode of the register in which the argument is being passed. */
|
1376 |
|
|
mode4 = GET_MODE (reg);
|
1377 |
|
|
|
1378 |
|
|
/* Eliminate sign extensions in the callee when:
|
1379 |
|
|
(a) A mode promotion has occurred; */
|
1380 |
|
|
if (mode1 == mode3)
|
1381 |
|
|
continue;
|
1382 |
|
|
/* (b) The mode of the register is the same as the mode of
|
1383 |
|
|
the argument as it is passed; */
|
1384 |
|
|
if (mode3 != mode4)
|
1385 |
|
|
continue;
|
1386 |
|
|
/* (c) There's no language level extension; */
|
1387 |
|
|
if (mode1 == mode2)
|
1388 |
|
|
;
|
1389 |
|
|
/* (c.1) All callers are from the current compilation unit. If that's
|
1390 |
|
|
the case we don't have to rely on an ABI, we only have to know
|
1391 |
|
|
what we're generating right now, and we know that we will do the
|
1392 |
|
|
mode1 to mode2 promotion with the given sign. */
|
1393 |
|
|
else if (!strictly_local)
|
1394 |
|
|
continue;
|
1395 |
|
|
/* (c.2) The combination of the two promotions is useful. This is
|
1396 |
|
|
true when the signs match, or if the first promotion is unsigned.
|
1397 |
|
|
In the later case, (sign_extend (zero_extend x)) is the same as
|
1398 |
|
|
(zero_extend (zero_extend x)), so make sure to force UNS3 true. */
|
1399 |
|
|
else if (uns1)
|
1400 |
|
|
uns3 = true;
|
1401 |
|
|
else if (uns3)
|
1402 |
|
|
continue;
|
1403 |
|
|
|
1404 |
|
|
/* Record that the value was promoted from mode1 to mode3,
|
1405 |
|
|
so that any sign extension at the head of the current
|
1406 |
|
|
function may be eliminated. */
|
1407 |
|
|
x = gen_rtx_CLOBBER (mode1, const0_rtx);
|
1408 |
|
|
x = gen_rtx_fmt_e ((uns3 ? ZERO_EXTEND : SIGN_EXTEND), mode3, x);
|
1409 |
|
|
record_value_for_reg (reg, first, x);
|
1410 |
|
|
}
|
1411 |
|
|
}
|
1412 |
|
|
|
1413 |
|
|
/* Called via note_stores. If X is a pseudo that is narrower than
|
1414 |
|
|
HOST_BITS_PER_WIDE_INT and is being set, record what bits are known zero.
|
1415 |
|
|
|
1416 |
|
|
If we are setting only a portion of X and we can't figure out what
|
1417 |
|
|
portion, assume all bits will be used since we don't know what will
|
1418 |
|
|
be happening.
|
1419 |
|
|
|
1420 |
|
|
Similarly, set how many bits of X are known to be copies of the sign bit
|
1421 |
|
|
at all locations in the function. This is the smallest number implied
|
1422 |
|
|
by any set of X. */
|
1423 |
|
|
|
1424 |
|
|
static void
|
1425 |
|
|
set_nonzero_bits_and_sign_copies (rtx x, const_rtx set, void *data)
|
1426 |
|
|
{
|
1427 |
|
|
rtx insn = (rtx) data;
|
1428 |
|
|
unsigned int num;
|
1429 |
|
|
|
1430 |
|
|
if (REG_P (x)
|
1431 |
|
|
&& REGNO (x) >= FIRST_PSEUDO_REGISTER
|
1432 |
|
|
/* If this register is undefined at the start of the file, we can't
|
1433 |
|
|
say what its contents were. */
|
1434 |
|
|
&& ! REGNO_REG_SET_P
|
1435 |
|
|
(DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x))
|
1436 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT)
|
1437 |
|
|
{
|
1438 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
|
1439 |
|
|
|
1440 |
|
|
if (set == 0 || GET_CODE (set) == CLOBBER)
|
1441 |
|
|
{
|
1442 |
|
|
rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
|
1443 |
|
|
rsp->sign_bit_copies = 1;
|
1444 |
|
|
return;
|
1445 |
|
|
}
|
1446 |
|
|
|
1447 |
|
|
/* If this register is being initialized using itself, and the
|
1448 |
|
|
register is uninitialized in this basic block, and there are
|
1449 |
|
|
no LOG_LINKS which set the register, then part of the
|
1450 |
|
|
register is uninitialized. In that case we can't assume
|
1451 |
|
|
anything about the number of nonzero bits.
|
1452 |
|
|
|
1453 |
|
|
??? We could do better if we checked this in
|
1454 |
|
|
reg_{nonzero_bits,num_sign_bit_copies}_for_combine. Then we
|
1455 |
|
|
could avoid making assumptions about the insn which initially
|
1456 |
|
|
sets the register, while still using the information in other
|
1457 |
|
|
insns. We would have to be careful to check every insn
|
1458 |
|
|
involved in the combination. */
|
1459 |
|
|
|
1460 |
|
|
if (insn
|
1461 |
|
|
&& reg_referenced_p (x, PATTERN (insn))
|
1462 |
|
|
&& !REGNO_REG_SET_P (DF_LR_IN (BLOCK_FOR_INSN (insn)),
|
1463 |
|
|
REGNO (x)))
|
1464 |
|
|
{
|
1465 |
|
|
rtx link;
|
1466 |
|
|
|
1467 |
|
|
for (link = LOG_LINKS (insn); link; link = XEXP (link, 1))
|
1468 |
|
|
{
|
1469 |
|
|
if (dead_or_set_p (XEXP (link, 0), x))
|
1470 |
|
|
break;
|
1471 |
|
|
}
|
1472 |
|
|
if (!link)
|
1473 |
|
|
{
|
1474 |
|
|
rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
|
1475 |
|
|
rsp->sign_bit_copies = 1;
|
1476 |
|
|
return;
|
1477 |
|
|
}
|
1478 |
|
|
}
|
1479 |
|
|
|
1480 |
|
|
/* If this is a complex assignment, see if we can convert it into a
|
1481 |
|
|
simple assignment. */
|
1482 |
|
|
set = expand_field_assignment (set);
|
1483 |
|
|
|
1484 |
|
|
/* If this is a simple assignment, or we have a paradoxical SUBREG,
|
1485 |
|
|
set what we know about X. */
|
1486 |
|
|
|
1487 |
|
|
if (SET_DEST (set) == x
|
1488 |
|
|
|| (GET_CODE (SET_DEST (set)) == SUBREG
|
1489 |
|
|
&& (GET_MODE_SIZE (GET_MODE (SET_DEST (set)))
|
1490 |
|
|
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (SET_DEST (set)))))
|
1491 |
|
|
&& SUBREG_REG (SET_DEST (set)) == x))
|
1492 |
|
|
{
|
1493 |
|
|
rtx src = SET_SRC (set);
|
1494 |
|
|
|
1495 |
|
|
#ifdef SHORT_IMMEDIATES_SIGN_EXTEND
|
1496 |
|
|
/* If X is narrower than a word and SRC is a non-negative
|
1497 |
|
|
constant that would appear negative in the mode of X,
|
1498 |
|
|
sign-extend it for use in reg_stat[].nonzero_bits because some
|
1499 |
|
|
machines (maybe most) will actually do the sign-extension
|
1500 |
|
|
and this is the conservative approach.
|
1501 |
|
|
|
1502 |
|
|
??? For 2.5, try to tighten up the MD files in this regard
|
1503 |
|
|
instead of this kludge. */
|
1504 |
|
|
|
1505 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) < BITS_PER_WORD
|
1506 |
|
|
&& CONST_INT_P (src)
|
1507 |
|
|
&& INTVAL (src) > 0
|
1508 |
|
|
&& 0 != (INTVAL (src)
|
1509 |
|
|
& ((HOST_WIDE_INT) 1
|
1510 |
|
|
<< (GET_MODE_BITSIZE (GET_MODE (x)) - 1))))
|
1511 |
|
|
src = GEN_INT (INTVAL (src)
|
1512 |
|
|
| ((HOST_WIDE_INT) (-1)
|
1513 |
|
|
<< GET_MODE_BITSIZE (GET_MODE (x))));
|
1514 |
|
|
#endif
|
1515 |
|
|
|
1516 |
|
|
/* Don't call nonzero_bits if it cannot change anything. */
|
1517 |
|
|
if (rsp->nonzero_bits != ~(unsigned HOST_WIDE_INT) 0)
|
1518 |
|
|
rsp->nonzero_bits |= nonzero_bits (src, nonzero_bits_mode);
|
1519 |
|
|
num = num_sign_bit_copies (SET_SRC (set), GET_MODE (x));
|
1520 |
|
|
if (rsp->sign_bit_copies == 0
|
1521 |
|
|
|| rsp->sign_bit_copies > num)
|
1522 |
|
|
rsp->sign_bit_copies = num;
|
1523 |
|
|
}
|
1524 |
|
|
else
|
1525 |
|
|
{
|
1526 |
|
|
rsp->nonzero_bits = GET_MODE_MASK (GET_MODE (x));
|
1527 |
|
|
rsp->sign_bit_copies = 1;
|
1528 |
|
|
}
|
1529 |
|
|
}
|
1530 |
|
|
}
|
1531 |
|
|
|
1532 |
|
|
/* See if INSN can be combined into I3. PRED and SUCC are optionally
|
1533 |
|
|
insns that were previously combined into I3 or that will be combined
|
1534 |
|
|
into the merger of INSN and I3.
|
1535 |
|
|
|
1536 |
|
|
Return 0 if the combination is not allowed for any reason.
|
1537 |
|
|
|
1538 |
|
|
If the combination is allowed, *PDEST will be set to the single
|
1539 |
|
|
destination of INSN and *PSRC to the single source, and this function
|
1540 |
|
|
will return 1. */
|
1541 |
|
|
|
1542 |
|
|
static int
|
1543 |
|
|
can_combine_p (rtx insn, rtx i3, rtx pred ATTRIBUTE_UNUSED, rtx succ,
|
1544 |
|
|
rtx *pdest, rtx *psrc)
|
1545 |
|
|
{
|
1546 |
|
|
int i;
|
1547 |
|
|
const_rtx set = 0;
|
1548 |
|
|
rtx src, dest;
|
1549 |
|
|
rtx p;
|
1550 |
|
|
#ifdef AUTO_INC_DEC
|
1551 |
|
|
rtx link;
|
1552 |
|
|
#endif
|
1553 |
|
|
int all_adjacent = (succ ? (next_active_insn (insn) == succ
|
1554 |
|
|
&& next_active_insn (succ) == i3)
|
1555 |
|
|
: next_active_insn (insn) == i3);
|
1556 |
|
|
|
1557 |
|
|
/* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
|
1558 |
|
|
or a PARALLEL consisting of such a SET and CLOBBERs.
|
1559 |
|
|
|
1560 |
|
|
If INSN has CLOBBER parallel parts, ignore them for our processing.
|
1561 |
|
|
By definition, these happen during the execution of the insn. When it
|
1562 |
|
|
is merged with another insn, all bets are off. If they are, in fact,
|
1563 |
|
|
needed and aren't also supplied in I3, they may be added by
|
1564 |
|
|
recog_for_combine. Otherwise, it won't match.
|
1565 |
|
|
|
1566 |
|
|
We can also ignore a SET whose SET_DEST is mentioned in a REG_UNUSED
|
1567 |
|
|
note.
|
1568 |
|
|
|
1569 |
|
|
Get the source and destination of INSN. If more than one, can't
|
1570 |
|
|
combine. */
|
1571 |
|
|
|
1572 |
|
|
if (GET_CODE (PATTERN (insn)) == SET)
|
1573 |
|
|
set = PATTERN (insn);
|
1574 |
|
|
else if (GET_CODE (PATTERN (insn)) == PARALLEL
|
1575 |
|
|
&& GET_CODE (XVECEXP (PATTERN (insn), 0, 0)) == SET)
|
1576 |
|
|
{
|
1577 |
|
|
for (i = 0; i < XVECLEN (PATTERN (insn), 0); i++)
|
1578 |
|
|
{
|
1579 |
|
|
rtx elt = XVECEXP (PATTERN (insn), 0, i);
|
1580 |
|
|
|
1581 |
|
|
switch (GET_CODE (elt))
|
1582 |
|
|
{
|
1583 |
|
|
/* This is important to combine floating point insns
|
1584 |
|
|
for the SH4 port. */
|
1585 |
|
|
case USE:
|
1586 |
|
|
/* Combining an isolated USE doesn't make sense.
|
1587 |
|
|
We depend here on combinable_i3pat to reject them. */
|
1588 |
|
|
/* The code below this loop only verifies that the inputs of
|
1589 |
|
|
the SET in INSN do not change. We call reg_set_between_p
|
1590 |
|
|
to verify that the REG in the USE does not change between
|
1591 |
|
|
I3 and INSN.
|
1592 |
|
|
If the USE in INSN was for a pseudo register, the matching
|
1593 |
|
|
insn pattern will likely match any register; combining this
|
1594 |
|
|
with any other USE would only be safe if we knew that the
|
1595 |
|
|
used registers have identical values, or if there was
|
1596 |
|
|
something to tell them apart, e.g. different modes. For
|
1597 |
|
|
now, we forgo such complicated tests and simply disallow
|
1598 |
|
|
combining of USES of pseudo registers with any other USE. */
|
1599 |
|
|
if (REG_P (XEXP (elt, 0))
|
1600 |
|
|
&& GET_CODE (PATTERN (i3)) == PARALLEL)
|
1601 |
|
|
{
|
1602 |
|
|
rtx i3pat = PATTERN (i3);
|
1603 |
|
|
int i = XVECLEN (i3pat, 0) - 1;
|
1604 |
|
|
unsigned int regno = REGNO (XEXP (elt, 0));
|
1605 |
|
|
|
1606 |
|
|
do
|
1607 |
|
|
{
|
1608 |
|
|
rtx i3elt = XVECEXP (i3pat, 0, i);
|
1609 |
|
|
|
1610 |
|
|
if (GET_CODE (i3elt) == USE
|
1611 |
|
|
&& REG_P (XEXP (i3elt, 0))
|
1612 |
|
|
&& (REGNO (XEXP (i3elt, 0)) == regno
|
1613 |
|
|
? reg_set_between_p (XEXP (elt, 0),
|
1614 |
|
|
PREV_INSN (insn), i3)
|
1615 |
|
|
: regno >= FIRST_PSEUDO_REGISTER))
|
1616 |
|
|
return 0;
|
1617 |
|
|
}
|
1618 |
|
|
while (--i >= 0);
|
1619 |
|
|
}
|
1620 |
|
|
break;
|
1621 |
|
|
|
1622 |
|
|
/* We can ignore CLOBBERs. */
|
1623 |
|
|
case CLOBBER:
|
1624 |
|
|
break;
|
1625 |
|
|
|
1626 |
|
|
case SET:
|
1627 |
|
|
/* Ignore SETs whose result isn't used but not those that
|
1628 |
|
|
have side-effects. */
|
1629 |
|
|
if (find_reg_note (insn, REG_UNUSED, SET_DEST (elt))
|
1630 |
|
|
&& insn_nothrow_p (insn)
|
1631 |
|
|
&& !side_effects_p (elt))
|
1632 |
|
|
break;
|
1633 |
|
|
|
1634 |
|
|
/* If we have already found a SET, this is a second one and
|
1635 |
|
|
so we cannot combine with this insn. */
|
1636 |
|
|
if (set)
|
1637 |
|
|
return 0;
|
1638 |
|
|
|
1639 |
|
|
set = elt;
|
1640 |
|
|
break;
|
1641 |
|
|
|
1642 |
|
|
default:
|
1643 |
|
|
/* Anything else means we can't combine. */
|
1644 |
|
|
return 0;
|
1645 |
|
|
}
|
1646 |
|
|
}
|
1647 |
|
|
|
1648 |
|
|
if (set == 0
|
1649 |
|
|
/* If SET_SRC is an ASM_OPERANDS we can't throw away these CLOBBERs,
|
1650 |
|
|
so don't do anything with it. */
|
1651 |
|
|
|| GET_CODE (SET_SRC (set)) == ASM_OPERANDS)
|
1652 |
|
|
return 0;
|
1653 |
|
|
}
|
1654 |
|
|
else
|
1655 |
|
|
return 0;
|
1656 |
|
|
|
1657 |
|
|
if (set == 0)
|
1658 |
|
|
return 0;
|
1659 |
|
|
|
1660 |
|
|
set = expand_field_assignment (set);
|
1661 |
|
|
src = SET_SRC (set), dest = SET_DEST (set);
|
1662 |
|
|
|
1663 |
|
|
/* Don't eliminate a store in the stack pointer. */
|
1664 |
|
|
if (dest == stack_pointer_rtx
|
1665 |
|
|
/* Don't combine with an insn that sets a register to itself if it has
|
1666 |
|
|
a REG_EQUAL note. This may be part of a LIBCALL sequence. */
|
1667 |
|
|
|| (rtx_equal_p (src, dest) && find_reg_note (insn, REG_EQUAL, NULL_RTX))
|
1668 |
|
|
/* Can't merge an ASM_OPERANDS. */
|
1669 |
|
|
|| GET_CODE (src) == ASM_OPERANDS
|
1670 |
|
|
/* Can't merge a function call. */
|
1671 |
|
|
|| GET_CODE (src) == CALL
|
1672 |
|
|
/* Don't eliminate a function call argument. */
|
1673 |
|
|
|| (CALL_P (i3)
|
1674 |
|
|
&& (find_reg_fusage (i3, USE, dest)
|
1675 |
|
|
|| (REG_P (dest)
|
1676 |
|
|
&& REGNO (dest) < FIRST_PSEUDO_REGISTER
|
1677 |
|
|
&& global_regs[REGNO (dest)])))
|
1678 |
|
|
/* Don't substitute into an incremented register. */
|
1679 |
|
|
|| FIND_REG_INC_NOTE (i3, dest)
|
1680 |
|
|
|| (succ && FIND_REG_INC_NOTE (succ, dest))
|
1681 |
|
|
/* Don't substitute into a non-local goto, this confuses CFG. */
|
1682 |
|
|
|| (JUMP_P (i3) && find_reg_note (i3, REG_NON_LOCAL_GOTO, NULL_RTX))
|
1683 |
|
|
/* Make sure that DEST is not used after SUCC but before I3. */
|
1684 |
|
|
|| (succ && ! all_adjacent
|
1685 |
|
|
&& reg_used_between_p (dest, succ, i3))
|
1686 |
|
|
/* Make sure that the value that is to be substituted for the register
|
1687 |
|
|
does not use any registers whose values alter in between. However,
|
1688 |
|
|
If the insns are adjacent, a use can't cross a set even though we
|
1689 |
|
|
think it might (this can happen for a sequence of insns each setting
|
1690 |
|
|
the same destination; last_set of that register might point to
|
1691 |
|
|
a NOTE). If INSN has a REG_EQUIV note, the register is always
|
1692 |
|
|
equivalent to the memory so the substitution is valid even if there
|
1693 |
|
|
are intervening stores. Also, don't move a volatile asm or
|
1694 |
|
|
UNSPEC_VOLATILE across any other insns. */
|
1695 |
|
|
|| (! all_adjacent
|
1696 |
|
|
&& (((!MEM_P (src)
|
1697 |
|
|
|| ! find_reg_note (insn, REG_EQUIV, src))
|
1698 |
|
|
&& use_crosses_set_p (src, DF_INSN_LUID (insn)))
|
1699 |
|
|
|| (GET_CODE (src) == ASM_OPERANDS && MEM_VOLATILE_P (src))
|
1700 |
|
|
|| GET_CODE (src) == UNSPEC_VOLATILE))
|
1701 |
|
|
/* Don't combine across a CALL_INSN, because that would possibly
|
1702 |
|
|
change whether the life span of some REGs crosses calls or not,
|
1703 |
|
|
and it is a pain to update that information.
|
1704 |
|
|
Exception: if source is a constant, moving it later can't hurt.
|
1705 |
|
|
Accept that as a special case. */
|
1706 |
|
|
|| (DF_INSN_LUID (insn) < last_call_luid && ! CONSTANT_P (src)))
|
1707 |
|
|
return 0;
|
1708 |
|
|
|
1709 |
|
|
/* DEST must either be a REG or CC0. */
|
1710 |
|
|
if (REG_P (dest))
|
1711 |
|
|
{
|
1712 |
|
|
/* If register alignment is being enforced for multi-word items in all
|
1713 |
|
|
cases except for parameters, it is possible to have a register copy
|
1714 |
|
|
insn referencing a hard register that is not allowed to contain the
|
1715 |
|
|
mode being copied and which would not be valid as an operand of most
|
1716 |
|
|
insns. Eliminate this problem by not combining with such an insn.
|
1717 |
|
|
|
1718 |
|
|
Also, on some machines we don't want to extend the life of a hard
|
1719 |
|
|
register. */
|
1720 |
|
|
|
1721 |
|
|
if (REG_P (src)
|
1722 |
|
|
&& ((REGNO (dest) < FIRST_PSEUDO_REGISTER
|
1723 |
|
|
&& ! HARD_REGNO_MODE_OK (REGNO (dest), GET_MODE (dest)))
|
1724 |
|
|
/* Don't extend the life of a hard register unless it is
|
1725 |
|
|
user variable (if we have few registers) or it can't
|
1726 |
|
|
fit into the desired register (meaning something special
|
1727 |
|
|
is going on).
|
1728 |
|
|
Also avoid substituting a return register into I3, because
|
1729 |
|
|
reload can't handle a conflict with constraints of other
|
1730 |
|
|
inputs. */
|
1731 |
|
|
|| (REGNO (src) < FIRST_PSEUDO_REGISTER
|
1732 |
|
|
&& ! HARD_REGNO_MODE_OK (REGNO (src), GET_MODE (src)))))
|
1733 |
|
|
return 0;
|
1734 |
|
|
}
|
1735 |
|
|
else if (GET_CODE (dest) != CC0)
|
1736 |
|
|
return 0;
|
1737 |
|
|
|
1738 |
|
|
|
1739 |
|
|
if (GET_CODE (PATTERN (i3)) == PARALLEL)
|
1740 |
|
|
for (i = XVECLEN (PATTERN (i3), 0) - 1; i >= 0; i--)
|
1741 |
|
|
if (GET_CODE (XVECEXP (PATTERN (i3), 0, i)) == CLOBBER)
|
1742 |
|
|
{
|
1743 |
|
|
/* Don't substitute for a register intended as a clobberable
|
1744 |
|
|
operand. */
|
1745 |
|
|
rtx reg = XEXP (XVECEXP (PATTERN (i3), 0, i), 0);
|
1746 |
|
|
if (rtx_equal_p (reg, dest))
|
1747 |
|
|
return 0;
|
1748 |
|
|
|
1749 |
|
|
/* If the clobber represents an earlyclobber operand, we must not
|
1750 |
|
|
substitute an expression containing the clobbered register.
|
1751 |
|
|
As we do not analyze the constraint strings here, we have to
|
1752 |
|
|
make the conservative assumption. However, if the register is
|
1753 |
|
|
a fixed hard reg, the clobber cannot represent any operand;
|
1754 |
|
|
we leave it up to the machine description to either accept or
|
1755 |
|
|
reject use-and-clobber patterns. */
|
1756 |
|
|
if (!REG_P (reg)
|
1757 |
|
|
|| REGNO (reg) >= FIRST_PSEUDO_REGISTER
|
1758 |
|
|
|| !fixed_regs[REGNO (reg)])
|
1759 |
|
|
if (reg_overlap_mentioned_p (reg, src))
|
1760 |
|
|
return 0;
|
1761 |
|
|
}
|
1762 |
|
|
|
1763 |
|
|
/* If INSN contains anything volatile, or is an `asm' (whether volatile
|
1764 |
|
|
or not), reject, unless nothing volatile comes between it and I3 */
|
1765 |
|
|
|
1766 |
|
|
if (GET_CODE (src) == ASM_OPERANDS || volatile_refs_p (src))
|
1767 |
|
|
{
|
1768 |
|
|
/* Make sure succ doesn't contain a volatile reference. */
|
1769 |
|
|
if (succ != 0 && volatile_refs_p (PATTERN (succ)))
|
1770 |
|
|
return 0;
|
1771 |
|
|
|
1772 |
|
|
for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
|
1773 |
|
|
if (INSN_P (p) && p != succ && volatile_refs_p (PATTERN (p)))
|
1774 |
|
|
return 0;
|
1775 |
|
|
}
|
1776 |
|
|
|
1777 |
|
|
/* If INSN is an asm, and DEST is a hard register, reject, since it has
|
1778 |
|
|
to be an explicit register variable, and was chosen for a reason. */
|
1779 |
|
|
|
1780 |
|
|
if (GET_CODE (src) == ASM_OPERANDS
|
1781 |
|
|
&& REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER)
|
1782 |
|
|
return 0;
|
1783 |
|
|
|
1784 |
|
|
/* If there are any volatile insns between INSN and I3, reject, because
|
1785 |
|
|
they might affect machine state. */
|
1786 |
|
|
|
1787 |
|
|
for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
|
1788 |
|
|
if (INSN_P (p) && p != succ && volatile_insn_p (PATTERN (p)))
|
1789 |
|
|
return 0;
|
1790 |
|
|
|
1791 |
|
|
/* If INSN contains an autoincrement or autodecrement, make sure that
|
1792 |
|
|
register is not used between there and I3, and not already used in
|
1793 |
|
|
I3 either. Neither must it be used in PRED or SUCC, if they exist.
|
1794 |
|
|
Also insist that I3 not be a jump; if it were one
|
1795 |
|
|
and the incremented register were spilled, we would lose. */
|
1796 |
|
|
|
1797 |
|
|
#ifdef AUTO_INC_DEC
|
1798 |
|
|
for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
|
1799 |
|
|
if (REG_NOTE_KIND (link) == REG_INC
|
1800 |
|
|
&& (JUMP_P (i3)
|
1801 |
|
|
|| reg_used_between_p (XEXP (link, 0), insn, i3)
|
1802 |
|
|
|| (pred != NULL_RTX
|
1803 |
|
|
&& reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred)))
|
1804 |
|
|
|| (succ != NULL_RTX
|
1805 |
|
|
&& reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ)))
|
1806 |
|
|
|| reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i3))))
|
1807 |
|
|
return 0;
|
1808 |
|
|
#endif
|
1809 |
|
|
|
1810 |
|
|
#ifdef HAVE_cc0
|
1811 |
|
|
/* Don't combine an insn that follows a CC0-setting insn.
|
1812 |
|
|
An insn that uses CC0 must not be separated from the one that sets it.
|
1813 |
|
|
We do, however, allow I2 to follow a CC0-setting insn if that insn
|
1814 |
|
|
is passed as I1; in that case it will be deleted also.
|
1815 |
|
|
We also allow combining in this case if all the insns are adjacent
|
1816 |
|
|
because that would leave the two CC0 insns adjacent as well.
|
1817 |
|
|
It would be more logical to test whether CC0 occurs inside I1 or I2,
|
1818 |
|
|
but that would be much slower, and this ought to be equivalent. */
|
1819 |
|
|
|
1820 |
|
|
p = prev_nonnote_insn (insn);
|
1821 |
|
|
if (p && p != pred && NONJUMP_INSN_P (p) && sets_cc0_p (PATTERN (p))
|
1822 |
|
|
&& ! all_adjacent)
|
1823 |
|
|
return 0;
|
1824 |
|
|
#endif
|
1825 |
|
|
|
1826 |
|
|
/* If we get here, we have passed all the tests and the combination is
|
1827 |
|
|
to be allowed. */
|
1828 |
|
|
|
1829 |
|
|
*pdest = dest;
|
1830 |
|
|
*psrc = src;
|
1831 |
|
|
|
1832 |
|
|
return 1;
|
1833 |
|
|
}
|
1834 |
|
|
|
1835 |
|
|
/* LOC is the location within I3 that contains its pattern or the component
|
1836 |
|
|
of a PARALLEL of the pattern. We validate that it is valid for combining.
|
1837 |
|
|
|
1838 |
|
|
One problem is if I3 modifies its output, as opposed to replacing it
|
1839 |
|
|
entirely, we can't allow the output to contain I2DEST or I1DEST as doing
|
1840 |
|
|
so would produce an insn that is not equivalent to the original insns.
|
1841 |
|
|
|
1842 |
|
|
Consider:
|
1843 |
|
|
|
1844 |
|
|
(set (reg:DI 101) (reg:DI 100))
|
1845 |
|
|
(set (subreg:SI (reg:DI 101) 0) <foo>)
|
1846 |
|
|
|
1847 |
|
|
This is NOT equivalent to:
|
1848 |
|
|
|
1849 |
|
|
(parallel [(set (subreg:SI (reg:DI 100) 0) <foo>)
|
1850 |
|
|
(set (reg:DI 101) (reg:DI 100))])
|
1851 |
|
|
|
1852 |
|
|
Not only does this modify 100 (in which case it might still be valid
|
1853 |
|
|
if 100 were dead in I2), it sets 101 to the ORIGINAL value of 100.
|
1854 |
|
|
|
1855 |
|
|
We can also run into a problem if I2 sets a register that I1
|
1856 |
|
|
uses and I1 gets directly substituted into I3 (not via I2). In that
|
1857 |
|
|
case, we would be getting the wrong value of I2DEST into I3, so we
|
1858 |
|
|
must reject the combination. This case occurs when I2 and I1 both
|
1859 |
|
|
feed into I3, rather than when I1 feeds into I2, which feeds into I3.
|
1860 |
|
|
If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
|
1861 |
|
|
of a SET must prevent combination from occurring.
|
1862 |
|
|
|
1863 |
|
|
Before doing the above check, we first try to expand a field assignment
|
1864 |
|
|
into a set of logical operations.
|
1865 |
|
|
|
1866 |
|
|
If PI3_DEST_KILLED is nonzero, it is a pointer to a location in which
|
1867 |
|
|
we place a register that is both set and used within I3. If more than one
|
1868 |
|
|
such register is detected, we fail.
|
1869 |
|
|
|
1870 |
|
|
Return 1 if the combination is valid, zero otherwise. */
|
1871 |
|
|
|
1872 |
|
|
static int
|
1873 |
|
|
combinable_i3pat (rtx i3, rtx *loc, rtx i2dest, rtx i1dest,
|
1874 |
|
|
int i1_not_in_src, rtx *pi3dest_killed)
|
1875 |
|
|
{
|
1876 |
|
|
rtx x = *loc;
|
1877 |
|
|
|
1878 |
|
|
if (GET_CODE (x) == SET)
|
1879 |
|
|
{
|
1880 |
|
|
rtx set = x ;
|
1881 |
|
|
rtx dest = SET_DEST (set);
|
1882 |
|
|
rtx src = SET_SRC (set);
|
1883 |
|
|
rtx inner_dest = dest;
|
1884 |
|
|
rtx subdest;
|
1885 |
|
|
|
1886 |
|
|
while (GET_CODE (inner_dest) == STRICT_LOW_PART
|
1887 |
|
|
|| GET_CODE (inner_dest) == SUBREG
|
1888 |
|
|
|| GET_CODE (inner_dest) == ZERO_EXTRACT)
|
1889 |
|
|
inner_dest = XEXP (inner_dest, 0);
|
1890 |
|
|
|
1891 |
|
|
/* Check for the case where I3 modifies its output, as discussed
|
1892 |
|
|
above. We don't want to prevent pseudos from being combined
|
1893 |
|
|
into the address of a MEM, so only prevent the combination if
|
1894 |
|
|
i1 or i2 set the same MEM. */
|
1895 |
|
|
if ((inner_dest != dest &&
|
1896 |
|
|
(!MEM_P (inner_dest)
|
1897 |
|
|
|| rtx_equal_p (i2dest, inner_dest)
|
1898 |
|
|
|| (i1dest && rtx_equal_p (i1dest, inner_dest)))
|
1899 |
|
|
&& (reg_overlap_mentioned_p (i2dest, inner_dest)
|
1900 |
|
|
|| (i1dest && reg_overlap_mentioned_p (i1dest, inner_dest))))
|
1901 |
|
|
|
1902 |
|
|
/* This is the same test done in can_combine_p except we can't test
|
1903 |
|
|
all_adjacent; we don't have to, since this instruction will stay
|
1904 |
|
|
in place, thus we are not considering increasing the lifetime of
|
1905 |
|
|
INNER_DEST.
|
1906 |
|
|
|
1907 |
|
|
Also, if this insn sets a function argument, combining it with
|
1908 |
|
|
something that might need a spill could clobber a previous
|
1909 |
|
|
function argument; the all_adjacent test in can_combine_p also
|
1910 |
|
|
checks this; here, we do a more specific test for this case. */
|
1911 |
|
|
|
1912 |
|
|
|| (REG_P (inner_dest)
|
1913 |
|
|
&& REGNO (inner_dest) < FIRST_PSEUDO_REGISTER
|
1914 |
|
|
&& (! HARD_REGNO_MODE_OK (REGNO (inner_dest),
|
1915 |
|
|
GET_MODE (inner_dest))))
|
1916 |
|
|
|| (i1_not_in_src && reg_overlap_mentioned_p (i1dest, src)))
|
1917 |
|
|
return 0;
|
1918 |
|
|
|
1919 |
|
|
/* If DEST is used in I3, it is being killed in this insn, so
|
1920 |
|
|
record that for later. We have to consider paradoxical
|
1921 |
|
|
subregs here, since they kill the whole register, but we
|
1922 |
|
|
ignore partial subregs, STRICT_LOW_PART, etc.
|
1923 |
|
|
Never add REG_DEAD notes for the FRAME_POINTER_REGNUM or the
|
1924 |
|
|
STACK_POINTER_REGNUM, since these are always considered to be
|
1925 |
|
|
live. Similarly for ARG_POINTER_REGNUM if it is fixed. */
|
1926 |
|
|
subdest = dest;
|
1927 |
|
|
if (GET_CODE (subdest) == SUBREG
|
1928 |
|
|
&& (GET_MODE_SIZE (GET_MODE (subdest))
|
1929 |
|
|
>= GET_MODE_SIZE (GET_MODE (SUBREG_REG (subdest)))))
|
1930 |
|
|
subdest = SUBREG_REG (subdest);
|
1931 |
|
|
if (pi3dest_killed
|
1932 |
|
|
&& REG_P (subdest)
|
1933 |
|
|
&& reg_referenced_p (subdest, PATTERN (i3))
|
1934 |
|
|
&& REGNO (subdest) != FRAME_POINTER_REGNUM
|
1935 |
|
|
#if HARD_FRAME_POINTER_REGNUM != FRAME_POINTER_REGNUM
|
1936 |
|
|
&& REGNO (subdest) != HARD_FRAME_POINTER_REGNUM
|
1937 |
|
|
#endif
|
1938 |
|
|
#if ARG_POINTER_REGNUM != FRAME_POINTER_REGNUM
|
1939 |
|
|
&& (REGNO (subdest) != ARG_POINTER_REGNUM
|
1940 |
|
|
|| ! fixed_regs [REGNO (subdest)])
|
1941 |
|
|
#endif
|
1942 |
|
|
&& REGNO (subdest) != STACK_POINTER_REGNUM)
|
1943 |
|
|
{
|
1944 |
|
|
if (*pi3dest_killed)
|
1945 |
|
|
return 0;
|
1946 |
|
|
|
1947 |
|
|
*pi3dest_killed = subdest;
|
1948 |
|
|
}
|
1949 |
|
|
}
|
1950 |
|
|
|
1951 |
|
|
else if (GET_CODE (x) == PARALLEL)
|
1952 |
|
|
{
|
1953 |
|
|
int i;
|
1954 |
|
|
|
1955 |
|
|
for (i = 0; i < XVECLEN (x, 0); i++)
|
1956 |
|
|
if (! combinable_i3pat (i3, &XVECEXP (x, 0, i), i2dest, i1dest,
|
1957 |
|
|
i1_not_in_src, pi3dest_killed))
|
1958 |
|
|
return 0;
|
1959 |
|
|
}
|
1960 |
|
|
|
1961 |
|
|
return 1;
|
1962 |
|
|
}
|
1963 |
|
|
|
1964 |
|
|
/* Return 1 if X is an arithmetic expression that contains a multiplication
|
1965 |
|
|
and division. We don't count multiplications by powers of two here. */
|
1966 |
|
|
|
1967 |
|
|
static int
|
1968 |
|
|
contains_muldiv (rtx x)
|
1969 |
|
|
{
|
1970 |
|
|
switch (GET_CODE (x))
|
1971 |
|
|
{
|
1972 |
|
|
case MOD: case DIV: case UMOD: case UDIV:
|
1973 |
|
|
return 1;
|
1974 |
|
|
|
1975 |
|
|
case MULT:
|
1976 |
|
|
return ! (CONST_INT_P (XEXP (x, 1))
|
1977 |
|
|
&& exact_log2 (INTVAL (XEXP (x, 1))) >= 0);
|
1978 |
|
|
default:
|
1979 |
|
|
if (BINARY_P (x))
|
1980 |
|
|
return contains_muldiv (XEXP (x, 0))
|
1981 |
|
|
|| contains_muldiv (XEXP (x, 1));
|
1982 |
|
|
|
1983 |
|
|
if (UNARY_P (x))
|
1984 |
|
|
return contains_muldiv (XEXP (x, 0));
|
1985 |
|
|
|
1986 |
|
|
return 0;
|
1987 |
|
|
}
|
1988 |
|
|
}
|
1989 |
|
|
|
1990 |
|
|
/* Determine whether INSN can be used in a combination. Return nonzero if
|
1991 |
|
|
not. This is used in try_combine to detect early some cases where we
|
1992 |
|
|
can't perform combinations. */
|
1993 |
|
|
|
1994 |
|
|
static int
|
1995 |
|
|
cant_combine_insn_p (rtx insn)
|
1996 |
|
|
{
|
1997 |
|
|
rtx set;
|
1998 |
|
|
rtx src, dest;
|
1999 |
|
|
|
2000 |
|
|
/* If this isn't really an insn, we can't do anything.
|
2001 |
|
|
This can occur when flow deletes an insn that it has merged into an
|
2002 |
|
|
auto-increment address. */
|
2003 |
|
|
if (! INSN_P (insn))
|
2004 |
|
|
return 1;
|
2005 |
|
|
|
2006 |
|
|
/* Never combine loads and stores involving hard regs that are likely
|
2007 |
|
|
to be spilled. The register allocator can usually handle such
|
2008 |
|
|
reg-reg moves by tying. If we allow the combiner to make
|
2009 |
|
|
substitutions of likely-spilled regs, reload might die.
|
2010 |
|
|
As an exception, we allow combinations involving fixed regs; these are
|
2011 |
|
|
not available to the register allocator so there's no risk involved. */
|
2012 |
|
|
|
2013 |
|
|
set = single_set (insn);
|
2014 |
|
|
if (! set)
|
2015 |
|
|
return 0;
|
2016 |
|
|
src = SET_SRC (set);
|
2017 |
|
|
dest = SET_DEST (set);
|
2018 |
|
|
if (GET_CODE (src) == SUBREG)
|
2019 |
|
|
src = SUBREG_REG (src);
|
2020 |
|
|
if (GET_CODE (dest) == SUBREG)
|
2021 |
|
|
dest = SUBREG_REG (dest);
|
2022 |
|
|
if (REG_P (src) && REG_P (dest)
|
2023 |
|
|
&& ((REGNO (src) < FIRST_PSEUDO_REGISTER
|
2024 |
|
|
&& ! fixed_regs[REGNO (src)]
|
2025 |
|
|
&& CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (src))))
|
2026 |
|
|
|| (REGNO (dest) < FIRST_PSEUDO_REGISTER
|
2027 |
|
|
&& ! fixed_regs[REGNO (dest)]
|
2028 |
|
|
&& CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (dest))))))
|
2029 |
|
|
return 1;
|
2030 |
|
|
|
2031 |
|
|
return 0;
|
2032 |
|
|
}
|
2033 |
|
|
|
2034 |
|
|
struct likely_spilled_retval_info
|
2035 |
|
|
{
|
2036 |
|
|
unsigned regno, nregs;
|
2037 |
|
|
unsigned mask;
|
2038 |
|
|
};
|
2039 |
|
|
|
2040 |
|
|
/* Called via note_stores by likely_spilled_retval_p. Remove from info->mask
|
2041 |
|
|
hard registers that are known to be written to / clobbered in full. */
|
2042 |
|
|
static void
|
2043 |
|
|
likely_spilled_retval_1 (rtx x, const_rtx set, void *data)
|
2044 |
|
|
{
|
2045 |
|
|
struct likely_spilled_retval_info *const info =
|
2046 |
|
|
(struct likely_spilled_retval_info *) data;
|
2047 |
|
|
unsigned regno, nregs;
|
2048 |
|
|
unsigned new_mask;
|
2049 |
|
|
|
2050 |
|
|
if (!REG_P (XEXP (set, 0)))
|
2051 |
|
|
return;
|
2052 |
|
|
regno = REGNO (x);
|
2053 |
|
|
if (regno >= info->regno + info->nregs)
|
2054 |
|
|
return;
|
2055 |
|
|
nregs = hard_regno_nregs[regno][GET_MODE (x)];
|
2056 |
|
|
if (regno + nregs <= info->regno)
|
2057 |
|
|
return;
|
2058 |
|
|
new_mask = (2U << (nregs - 1)) - 1;
|
2059 |
|
|
if (regno < info->regno)
|
2060 |
|
|
new_mask >>= info->regno - regno;
|
2061 |
|
|
else
|
2062 |
|
|
new_mask <<= regno - info->regno;
|
2063 |
|
|
info->mask &= ~new_mask;
|
2064 |
|
|
}
|
2065 |
|
|
|
2066 |
|
|
/* Return nonzero iff part of the return value is live during INSN, and
|
2067 |
|
|
it is likely spilled. This can happen when more than one insn is needed
|
2068 |
|
|
to copy the return value, e.g. when we consider to combine into the
|
2069 |
|
|
second copy insn for a complex value. */
|
2070 |
|
|
|
2071 |
|
|
static int
|
2072 |
|
|
likely_spilled_retval_p (rtx insn)
|
2073 |
|
|
{
|
2074 |
|
|
rtx use = BB_END (this_basic_block);
|
2075 |
|
|
rtx reg, p;
|
2076 |
|
|
unsigned regno, nregs;
|
2077 |
|
|
/* We assume here that no machine mode needs more than
|
2078 |
|
|
32 hard registers when the value overlaps with a register
|
2079 |
|
|
for which FUNCTION_VALUE_REGNO_P is true. */
|
2080 |
|
|
unsigned mask;
|
2081 |
|
|
struct likely_spilled_retval_info info;
|
2082 |
|
|
|
2083 |
|
|
if (!NONJUMP_INSN_P (use) || GET_CODE (PATTERN (use)) != USE || insn == use)
|
2084 |
|
|
return 0;
|
2085 |
|
|
reg = XEXP (PATTERN (use), 0);
|
2086 |
|
|
if (!REG_P (reg) || !FUNCTION_VALUE_REGNO_P (REGNO (reg)))
|
2087 |
|
|
return 0;
|
2088 |
|
|
regno = REGNO (reg);
|
2089 |
|
|
nregs = hard_regno_nregs[regno][GET_MODE (reg)];
|
2090 |
|
|
if (nregs == 1)
|
2091 |
|
|
return 0;
|
2092 |
|
|
mask = (2U << (nregs - 1)) - 1;
|
2093 |
|
|
|
2094 |
|
|
/* Disregard parts of the return value that are set later. */
|
2095 |
|
|
info.regno = regno;
|
2096 |
|
|
info.nregs = nregs;
|
2097 |
|
|
info.mask = mask;
|
2098 |
|
|
for (p = PREV_INSN (use); info.mask && p != insn; p = PREV_INSN (p))
|
2099 |
|
|
if (INSN_P (p))
|
2100 |
|
|
note_stores (PATTERN (p), likely_spilled_retval_1, &info);
|
2101 |
|
|
mask = info.mask;
|
2102 |
|
|
|
2103 |
|
|
/* Check if any of the (probably) live return value registers is
|
2104 |
|
|
likely spilled. */
|
2105 |
|
|
nregs --;
|
2106 |
|
|
do
|
2107 |
|
|
{
|
2108 |
|
|
if ((mask & 1 << nregs)
|
2109 |
|
|
&& CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (regno + nregs)))
|
2110 |
|
|
return 1;
|
2111 |
|
|
} while (nregs--);
|
2112 |
|
|
return 0;
|
2113 |
|
|
}
|
2114 |
|
|
|
2115 |
|
|
/* Adjust INSN after we made a change to its destination.
|
2116 |
|
|
|
2117 |
|
|
Changing the destination can invalidate notes that say something about
|
2118 |
|
|
the results of the insn and a LOG_LINK pointing to the insn. */
|
2119 |
|
|
|
2120 |
|
|
static void
|
2121 |
|
|
adjust_for_new_dest (rtx insn)
|
2122 |
|
|
{
|
2123 |
|
|
/* For notes, be conservative and simply remove them. */
|
2124 |
|
|
remove_reg_equal_equiv_notes (insn);
|
2125 |
|
|
|
2126 |
|
|
/* The new insn will have a destination that was previously the destination
|
2127 |
|
|
of an insn just above it. Call distribute_links to make a LOG_LINK from
|
2128 |
|
|
the next use of that destination. */
|
2129 |
|
|
distribute_links (gen_rtx_INSN_LIST (VOIDmode, insn, NULL_RTX));
|
2130 |
|
|
|
2131 |
|
|
df_insn_rescan (insn);
|
2132 |
|
|
}
|
2133 |
|
|
|
2134 |
|
|
/* Return TRUE if combine can reuse reg X in mode MODE.
|
2135 |
|
|
ADDED_SETS is nonzero if the original set is still required. */
|
2136 |
|
|
static bool
|
2137 |
|
|
can_change_dest_mode (rtx x, int added_sets, enum machine_mode mode)
|
2138 |
|
|
{
|
2139 |
|
|
unsigned int regno;
|
2140 |
|
|
|
2141 |
|
|
if (!REG_P(x))
|
2142 |
|
|
return false;
|
2143 |
|
|
|
2144 |
|
|
regno = REGNO (x);
|
2145 |
|
|
/* Allow hard registers if the new mode is legal, and occupies no more
|
2146 |
|
|
registers than the old mode. */
|
2147 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
2148 |
|
|
return (HARD_REGNO_MODE_OK (regno, mode)
|
2149 |
|
|
&& (hard_regno_nregs[regno][GET_MODE (x)]
|
2150 |
|
|
>= hard_regno_nregs[regno][mode]));
|
2151 |
|
|
|
2152 |
|
|
/* Or a pseudo that is only used once. */
|
2153 |
|
|
return (REG_N_SETS (regno) == 1 && !added_sets
|
2154 |
|
|
&& !REG_USERVAR_P (x));
|
2155 |
|
|
}
|
2156 |
|
|
|
2157 |
|
|
|
2158 |
|
|
/* Check whether X, the destination of a set, refers to part of
|
2159 |
|
|
the register specified by REG. */
|
2160 |
|
|
|
2161 |
|
|
static bool
|
2162 |
|
|
reg_subword_p (rtx x, rtx reg)
|
2163 |
|
|
{
|
2164 |
|
|
/* Check that reg is an integer mode register. */
|
2165 |
|
|
if (!REG_P (reg) || GET_MODE_CLASS (GET_MODE (reg)) != MODE_INT)
|
2166 |
|
|
return false;
|
2167 |
|
|
|
2168 |
|
|
if (GET_CODE (x) == STRICT_LOW_PART
|
2169 |
|
|
|| GET_CODE (x) == ZERO_EXTRACT)
|
2170 |
|
|
x = XEXP (x, 0);
|
2171 |
|
|
|
2172 |
|
|
return GET_CODE (x) == SUBREG
|
2173 |
|
|
&& SUBREG_REG (x) == reg
|
2174 |
|
|
&& GET_MODE_CLASS (GET_MODE (x)) == MODE_INT;
|
2175 |
|
|
}
|
2176 |
|
|
|
2177 |
|
|
#ifdef AUTO_INC_DEC
|
2178 |
|
|
/* Replace auto-increment addressing modes with explicit operations to
|
2179 |
|
|
access the same addresses without modifying the corresponding
|
2180 |
|
|
registers. If AFTER holds, SRC is meant to be reused after the
|
2181 |
|
|
side effect, otherwise it is to be reused before that. */
|
2182 |
|
|
|
2183 |
|
|
static rtx
|
2184 |
|
|
cleanup_auto_inc_dec (rtx src, bool after, enum machine_mode mem_mode)
|
2185 |
|
|
{
|
2186 |
|
|
rtx x = src;
|
2187 |
|
|
const RTX_CODE code = GET_CODE (x);
|
2188 |
|
|
int i;
|
2189 |
|
|
const char *fmt;
|
2190 |
|
|
|
2191 |
|
|
switch (code)
|
2192 |
|
|
{
|
2193 |
|
|
case REG:
|
2194 |
|
|
case CONST_INT:
|
2195 |
|
|
case CONST_DOUBLE:
|
2196 |
|
|
case CONST_FIXED:
|
2197 |
|
|
case CONST_VECTOR:
|
2198 |
|
|
case SYMBOL_REF:
|
2199 |
|
|
case CODE_LABEL:
|
2200 |
|
|
case PC:
|
2201 |
|
|
case CC0:
|
2202 |
|
|
case SCRATCH:
|
2203 |
|
|
/* SCRATCH must be shared because they represent distinct values. */
|
2204 |
|
|
return x;
|
2205 |
|
|
case CLOBBER:
|
2206 |
|
|
if (REG_P (XEXP (x, 0)) && REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER)
|
2207 |
|
|
return x;
|
2208 |
|
|
break;
|
2209 |
|
|
|
2210 |
|
|
case CONST:
|
2211 |
|
|
if (shared_const_p (x))
|
2212 |
|
|
return x;
|
2213 |
|
|
break;
|
2214 |
|
|
|
2215 |
|
|
case MEM:
|
2216 |
|
|
mem_mode = GET_MODE (x);
|
2217 |
|
|
break;
|
2218 |
|
|
|
2219 |
|
|
case PRE_INC:
|
2220 |
|
|
case PRE_DEC:
|
2221 |
|
|
case POST_INC:
|
2222 |
|
|
case POST_DEC:
|
2223 |
|
|
gcc_assert (mem_mode != VOIDmode && mem_mode != BLKmode);
|
2224 |
|
|
if (after == (code == PRE_INC || code == PRE_DEC))
|
2225 |
|
|
x = cleanup_auto_inc_dec (XEXP (x, 0), after, mem_mode);
|
2226 |
|
|
else
|
2227 |
|
|
x = gen_rtx_PLUS (GET_MODE (x),
|
2228 |
|
|
cleanup_auto_inc_dec (XEXP (x, 0), after, mem_mode),
|
2229 |
|
|
GEN_INT ((code == PRE_INC || code == POST_INC)
|
2230 |
|
|
? GET_MODE_SIZE (mem_mode)
|
2231 |
|
|
: -GET_MODE_SIZE (mem_mode)));
|
2232 |
|
|
return x;
|
2233 |
|
|
|
2234 |
|
|
case PRE_MODIFY:
|
2235 |
|
|
case POST_MODIFY:
|
2236 |
|
|
if (after == (code == PRE_MODIFY))
|
2237 |
|
|
x = XEXP (x, 0);
|
2238 |
|
|
else
|
2239 |
|
|
x = XEXP (x, 1);
|
2240 |
|
|
return cleanup_auto_inc_dec (x, after, mem_mode);
|
2241 |
|
|
|
2242 |
|
|
default:
|
2243 |
|
|
break;
|
2244 |
|
|
}
|
2245 |
|
|
|
2246 |
|
|
/* Copy the various flags, fields, and other information. We assume
|
2247 |
|
|
that all fields need copying, and then clear the fields that should
|
2248 |
|
|
not be copied. That is the sensible default behavior, and forces
|
2249 |
|
|
us to explicitly document why we are *not* copying a flag. */
|
2250 |
|
|
x = shallow_copy_rtx (x);
|
2251 |
|
|
|
2252 |
|
|
/* We do not copy the USED flag, which is used as a mark bit during
|
2253 |
|
|
walks over the RTL. */
|
2254 |
|
|
RTX_FLAG (x, used) = 0;
|
2255 |
|
|
|
2256 |
|
|
/* We do not copy FRAME_RELATED for INSNs. */
|
2257 |
|
|
if (INSN_P (x))
|
2258 |
|
|
RTX_FLAG (x, frame_related) = 0;
|
2259 |
|
|
|
2260 |
|
|
fmt = GET_RTX_FORMAT (code);
|
2261 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
2262 |
|
|
if (fmt[i] == 'e')
|
2263 |
|
|
XEXP (x, i) = cleanup_auto_inc_dec (XEXP (x, i), after, mem_mode);
|
2264 |
|
|
else if (fmt[i] == 'E' || fmt[i] == 'V')
|
2265 |
|
|
{
|
2266 |
|
|
int j;
|
2267 |
|
|
XVEC (x, i) = rtvec_alloc (XVECLEN (x, i));
|
2268 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
2269 |
|
|
XVECEXP (x, i, j)
|
2270 |
|
|
= cleanup_auto_inc_dec (XVECEXP (src, i, j), after, mem_mode);
|
2271 |
|
|
}
|
2272 |
|
|
|
2273 |
|
|
return x;
|
2274 |
|
|
}
|
2275 |
|
|
|
2276 |
|
|
/* Auxiliary data structure for propagate_for_debug_stmt. */
|
2277 |
|
|
|
2278 |
|
|
struct rtx_subst_pair
|
2279 |
|
|
{
|
2280 |
|
|
rtx to;
|
2281 |
|
|
bool adjusted;
|
2282 |
|
|
bool after;
|
2283 |
|
|
};
|
2284 |
|
|
|
2285 |
|
|
/* DATA points to an rtx_subst_pair. Return the value that should be
|
2286 |
|
|
substituted. */
|
2287 |
|
|
|
2288 |
|
|
static rtx
|
2289 |
|
|
propagate_for_debug_subst (rtx from, const_rtx old_rtx, void *data)
|
2290 |
|
|
{
|
2291 |
|
|
struct rtx_subst_pair *pair = (struct rtx_subst_pair *)data;
|
2292 |
|
|
|
2293 |
|
|
if (!rtx_equal_p (from, old_rtx))
|
2294 |
|
|
return NULL_RTX;
|
2295 |
|
|
if (!pair->adjusted)
|
2296 |
|
|
{
|
2297 |
|
|
pair->adjusted = true;
|
2298 |
|
|
pair->to = cleanup_auto_inc_dec (pair->to, pair->after, VOIDmode);
|
2299 |
|
|
return pair->to;
|
2300 |
|
|
}
|
2301 |
|
|
return copy_rtx (pair->to);
|
2302 |
|
|
}
|
2303 |
|
|
#endif
|
2304 |
|
|
|
2305 |
|
|
/* Replace occurrences of DEST with SRC in DEBUG_INSNs between INSN
|
2306 |
|
|
and LAST. If MOVE holds, debug insns must also be moved past
|
2307 |
|
|
LAST. */
|
2308 |
|
|
|
2309 |
|
|
static void
|
2310 |
|
|
propagate_for_debug (rtx insn, rtx last, rtx dest, rtx src, bool move)
|
2311 |
|
|
{
|
2312 |
|
|
rtx next, move_pos = move ? last : NULL_RTX, loc;
|
2313 |
|
|
|
2314 |
|
|
#ifdef AUTO_INC_DEC
|
2315 |
|
|
struct rtx_subst_pair p;
|
2316 |
|
|
p.to = src;
|
2317 |
|
|
p.adjusted = false;
|
2318 |
|
|
p.after = move;
|
2319 |
|
|
#endif
|
2320 |
|
|
|
2321 |
|
|
next = NEXT_INSN (insn);
|
2322 |
|
|
while (next != last)
|
2323 |
|
|
{
|
2324 |
|
|
insn = next;
|
2325 |
|
|
next = NEXT_INSN (insn);
|
2326 |
|
|
if (DEBUG_INSN_P (insn))
|
2327 |
|
|
{
|
2328 |
|
|
#ifdef AUTO_INC_DEC
|
2329 |
|
|
loc = simplify_replace_fn_rtx (INSN_VAR_LOCATION_LOC (insn),
|
2330 |
|
|
dest, propagate_for_debug_subst, &p);
|
2331 |
|
|
#else
|
2332 |
|
|
loc = simplify_replace_rtx (INSN_VAR_LOCATION_LOC (insn), dest, src);
|
2333 |
|
|
#endif
|
2334 |
|
|
if (loc == INSN_VAR_LOCATION_LOC (insn))
|
2335 |
|
|
continue;
|
2336 |
|
|
INSN_VAR_LOCATION_LOC (insn) = loc;
|
2337 |
|
|
if (move_pos)
|
2338 |
|
|
{
|
2339 |
|
|
remove_insn (insn);
|
2340 |
|
|
PREV_INSN (insn) = NEXT_INSN (insn) = NULL_RTX;
|
2341 |
|
|
move_pos = emit_debug_insn_after (insn, move_pos);
|
2342 |
|
|
}
|
2343 |
|
|
else
|
2344 |
|
|
df_insn_rescan (insn);
|
2345 |
|
|
}
|
2346 |
|
|
}
|
2347 |
|
|
}
|
2348 |
|
|
|
2349 |
|
|
/* Delete the unconditional jump INSN and adjust the CFG correspondingly.
|
2350 |
|
|
Note that the INSN should be deleted *after* removing dead edges, so
|
2351 |
|
|
that the kept edge is the fallthrough edge for a (set (pc) (pc))
|
2352 |
|
|
but not for a (set (pc) (label_ref FOO)). */
|
2353 |
|
|
|
2354 |
|
|
static void
|
2355 |
|
|
update_cfg_for_uncondjump (rtx insn)
|
2356 |
|
|
{
|
2357 |
|
|
basic_block bb = BLOCK_FOR_INSN (insn);
|
2358 |
|
|
bool at_end = (BB_END (bb) == insn);
|
2359 |
|
|
|
2360 |
|
|
if (at_end)
|
2361 |
|
|
purge_dead_edges (bb);
|
2362 |
|
|
|
2363 |
|
|
delete_insn (insn);
|
2364 |
|
|
if (at_end && EDGE_COUNT (bb->succs) == 1)
|
2365 |
|
|
single_succ_edge (bb)->flags |= EDGE_FALLTHRU;
|
2366 |
|
|
}
|
2367 |
|
|
|
2368 |
|
|
|
2369 |
|
|
/* Try to combine the insns I1 and I2 into I3.
|
2370 |
|
|
Here I1 and I2 appear earlier than I3.
|
2371 |
|
|
I1 can be zero; then we combine just I2 into I3.
|
2372 |
|
|
|
2373 |
|
|
If we are combining three insns and the resulting insn is not recognized,
|
2374 |
|
|
try splitting it into two insns. If that happens, I2 and I3 are retained
|
2375 |
|
|
and I1 is pseudo-deleted by turning it into a NOTE. Otherwise, I1 and I2
|
2376 |
|
|
are pseudo-deleted.
|
2377 |
|
|
|
2378 |
|
|
Return 0 if the combination does not work. Then nothing is changed.
|
2379 |
|
|
If we did the combination, return the insn at which combine should
|
2380 |
|
|
resume scanning.
|
2381 |
|
|
|
2382 |
|
|
Set NEW_DIRECT_JUMP_P to a nonzero value if try_combine creates a
|
2383 |
|
|
new direct jump instruction. */
|
2384 |
|
|
|
2385 |
|
|
static rtx
|
2386 |
|
|
try_combine (rtx i3, rtx i2, rtx i1, int *new_direct_jump_p)
|
2387 |
|
|
{
|
2388 |
|
|
/* New patterns for I3 and I2, respectively. */
|
2389 |
|
|
rtx newpat, newi2pat = 0;
|
2390 |
|
|
rtvec newpat_vec_with_clobbers = 0;
|
2391 |
|
|
int substed_i2 = 0, substed_i1 = 0;
|
2392 |
|
|
/* Indicates need to preserve SET in I1 or I2 in I3 if it is not dead. */
|
2393 |
|
|
int added_sets_1, added_sets_2;
|
2394 |
|
|
/* Total number of SETs to put into I3. */
|
2395 |
|
|
int total_sets;
|
2396 |
|
|
/* Nonzero if I2's body now appears in I3. */
|
2397 |
|
|
int i2_is_used;
|
2398 |
|
|
/* INSN_CODEs for new I3, new I2, and user of condition code. */
|
2399 |
|
|
int insn_code_number, i2_code_number = 0, other_code_number = 0;
|
2400 |
|
|
/* Contains I3 if the destination of I3 is used in its source, which means
|
2401 |
|
|
that the old life of I3 is being killed. If that usage is placed into
|
2402 |
|
|
I2 and not in I3, a REG_DEAD note must be made. */
|
2403 |
|
|
rtx i3dest_killed = 0;
|
2404 |
|
|
/* SET_DEST and SET_SRC of I2 and I1. */
|
2405 |
|
|
rtx i2dest = 0, i2src = 0, i1dest = 0, i1src = 0;
|
2406 |
|
|
/* Set if I2DEST was reused as a scratch register. */
|
2407 |
|
|
bool i2scratch = false;
|
2408 |
|
|
/* PATTERN (I1) and PATTERN (I2), or a copy of it in certain cases. */
|
2409 |
|
|
rtx i1pat = 0, i2pat = 0;
|
2410 |
|
|
/* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
|
2411 |
|
|
int i2dest_in_i2src = 0, i1dest_in_i1src = 0, i2dest_in_i1src = 0;
|
2412 |
|
|
int i2dest_killed = 0, i1dest_killed = 0;
|
2413 |
|
|
int i1_feeds_i3 = 0;
|
2414 |
|
|
/* Notes that must be added to REG_NOTES in I3 and I2. */
|
2415 |
|
|
rtx new_i3_notes, new_i2_notes;
|
2416 |
|
|
/* Notes that we substituted I3 into I2 instead of the normal case. */
|
2417 |
|
|
int i3_subst_into_i2 = 0;
|
2418 |
|
|
/* Notes that I1, I2 or I3 is a MULT operation. */
|
2419 |
|
|
int have_mult = 0;
|
2420 |
|
|
int swap_i2i3 = 0;
|
2421 |
|
|
int changed_i3_dest = 0;
|
2422 |
|
|
|
2423 |
|
|
int maxreg;
|
2424 |
|
|
rtx temp;
|
2425 |
|
|
rtx link;
|
2426 |
|
|
rtx other_pat = 0;
|
2427 |
|
|
rtx new_other_notes;
|
2428 |
|
|
int i;
|
2429 |
|
|
|
2430 |
|
|
/* Exit early if one of the insns involved can't be used for
|
2431 |
|
|
combinations. */
|
2432 |
|
|
if (cant_combine_insn_p (i3)
|
2433 |
|
|
|| cant_combine_insn_p (i2)
|
2434 |
|
|
|| (i1 && cant_combine_insn_p (i1))
|
2435 |
|
|
|| likely_spilled_retval_p (i3))
|
2436 |
|
|
return 0;
|
2437 |
|
|
|
2438 |
|
|
combine_attempts++;
|
2439 |
|
|
undobuf.other_insn = 0;
|
2440 |
|
|
|
2441 |
|
|
/* Reset the hard register usage information. */
|
2442 |
|
|
CLEAR_HARD_REG_SET (newpat_used_regs);
|
2443 |
|
|
|
2444 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
2445 |
|
|
{
|
2446 |
|
|
if (i1)
|
2447 |
|
|
fprintf (dump_file, "\nTrying %d, %d -> %d:\n",
|
2448 |
|
|
INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
|
2449 |
|
|
else
|
2450 |
|
|
fprintf (dump_file, "\nTrying %d -> %d:\n",
|
2451 |
|
|
INSN_UID (i2), INSN_UID (i3));
|
2452 |
|
|
}
|
2453 |
|
|
|
2454 |
|
|
/* If I1 and I2 both feed I3, they can be in any order. To simplify the
|
2455 |
|
|
code below, set I1 to be the earlier of the two insns. */
|
2456 |
|
|
if (i1 && DF_INSN_LUID (i1) > DF_INSN_LUID (i2))
|
2457 |
|
|
temp = i1, i1 = i2, i2 = temp;
|
2458 |
|
|
|
2459 |
|
|
added_links_insn = 0;
|
2460 |
|
|
|
2461 |
|
|
/* First check for one important special-case that the code below will
|
2462 |
|
|
not handle. Namely, the case where I1 is zero, I2 is a PARALLEL
|
2463 |
|
|
and I3 is a SET whose SET_SRC is a SET_DEST in I2. In that case,
|
2464 |
|
|
we may be able to replace that destination with the destination of I3.
|
2465 |
|
|
This occurs in the common code where we compute both a quotient and
|
2466 |
|
|
remainder into a structure, in which case we want to do the computation
|
2467 |
|
|
directly into the structure to avoid register-register copies.
|
2468 |
|
|
|
2469 |
|
|
Note that this case handles both multiple sets in I2 and also
|
2470 |
|
|
cases where I2 has a number of CLOBBER or PARALLELs.
|
2471 |
|
|
|
2472 |
|
|
We make very conservative checks below and only try to handle the
|
2473 |
|
|
most common cases of this. For example, we only handle the case
|
2474 |
|
|
where I2 and I3 are adjacent to avoid making difficult register
|
2475 |
|
|
usage tests. */
|
2476 |
|
|
|
2477 |
|
|
if (i1 == 0 && NONJUMP_INSN_P (i3) && GET_CODE (PATTERN (i3)) == SET
|
2478 |
|
|
&& REG_P (SET_SRC (PATTERN (i3)))
|
2479 |
|
|
&& REGNO (SET_SRC (PATTERN (i3))) >= FIRST_PSEUDO_REGISTER
|
2480 |
|
|
&& find_reg_note (i3, REG_DEAD, SET_SRC (PATTERN (i3)))
|
2481 |
|
|
&& GET_CODE (PATTERN (i2)) == PARALLEL
|
2482 |
|
|
&& ! side_effects_p (SET_DEST (PATTERN (i3)))
|
2483 |
|
|
/* If the dest of I3 is a ZERO_EXTRACT or STRICT_LOW_PART, the code
|
2484 |
|
|
below would need to check what is inside (and reg_overlap_mentioned_p
|
2485 |
|
|
doesn't support those codes anyway). Don't allow those destinations;
|
2486 |
|
|
the resulting insn isn't likely to be recognized anyway. */
|
2487 |
|
|
&& GET_CODE (SET_DEST (PATTERN (i3))) != ZERO_EXTRACT
|
2488 |
|
|
&& GET_CODE (SET_DEST (PATTERN (i3))) != STRICT_LOW_PART
|
2489 |
|
|
&& ! reg_overlap_mentioned_p (SET_SRC (PATTERN (i3)),
|
2490 |
|
|
SET_DEST (PATTERN (i3)))
|
2491 |
|
|
&& next_active_insn (i2) == i3)
|
2492 |
|
|
{
|
2493 |
|
|
rtx p2 = PATTERN (i2);
|
2494 |
|
|
|
2495 |
|
|
/* Make sure that the destination of I3,
|
2496 |
|
|
which we are going to substitute into one output of I2,
|
2497 |
|
|
is not used within another output of I2. We must avoid making this:
|
2498 |
|
|
(parallel [(set (mem (reg 69)) ...)
|
2499 |
|
|
(set (reg 69) ...)])
|
2500 |
|
|
which is not well-defined as to order of actions.
|
2501 |
|
|
(Besides, reload can't handle output reloads for this.)
|
2502 |
|
|
|
2503 |
|
|
The problem can also happen if the dest of I3 is a memory ref,
|
2504 |
|
|
if another dest in I2 is an indirect memory ref. */
|
2505 |
|
|
for (i = 0; i < XVECLEN (p2, 0); i++)
|
2506 |
|
|
if ((GET_CODE (XVECEXP (p2, 0, i)) == SET
|
2507 |
|
|
|| GET_CODE (XVECEXP (p2, 0, i)) == CLOBBER)
|
2508 |
|
|
&& reg_overlap_mentioned_p (SET_DEST (PATTERN (i3)),
|
2509 |
|
|
SET_DEST (XVECEXP (p2, 0, i))))
|
2510 |
|
|
break;
|
2511 |
|
|
|
2512 |
|
|
if (i == XVECLEN (p2, 0))
|
2513 |
|
|
for (i = 0; i < XVECLEN (p2, 0); i++)
|
2514 |
|
|
if ((GET_CODE (XVECEXP (p2, 0, i)) == SET
|
2515 |
|
|
|| GET_CODE (XVECEXP (p2, 0, i)) == CLOBBER)
|
2516 |
|
|
&& SET_DEST (XVECEXP (p2, 0, i)) == SET_SRC (PATTERN (i3)))
|
2517 |
|
|
{
|
2518 |
|
|
combine_merges++;
|
2519 |
|
|
|
2520 |
|
|
subst_insn = i3;
|
2521 |
|
|
subst_low_luid = DF_INSN_LUID (i2);
|
2522 |
|
|
|
2523 |
|
|
added_sets_2 = added_sets_1 = 0;
|
2524 |
|
|
i2src = SET_DEST (PATTERN (i3));
|
2525 |
|
|
i2dest = SET_SRC (PATTERN (i3));
|
2526 |
|
|
i2dest_killed = dead_or_set_p (i2, i2dest);
|
2527 |
|
|
|
2528 |
|
|
/* Replace the dest in I2 with our dest and make the resulting
|
2529 |
|
|
insn the new pattern for I3. Then skip to where we
|
2530 |
|
|
validate the pattern. Everything was set up above. */
|
2531 |
|
|
SUBST (SET_DEST (XVECEXP (p2, 0, i)),
|
2532 |
|
|
SET_DEST (PATTERN (i3)));
|
2533 |
|
|
|
2534 |
|
|
newpat = p2;
|
2535 |
|
|
i3_subst_into_i2 = 1;
|
2536 |
|
|
goto validate_replacement;
|
2537 |
|
|
}
|
2538 |
|
|
}
|
2539 |
|
|
|
2540 |
|
|
/* If I2 is setting a pseudo to a constant and I3 is setting some
|
2541 |
|
|
sub-part of it to another constant, merge them by making a new
|
2542 |
|
|
constant. */
|
2543 |
|
|
if (i1 == 0
|
2544 |
|
|
&& (temp = single_set (i2)) != 0
|
2545 |
|
|
&& (CONST_INT_P (SET_SRC (temp))
|
2546 |
|
|
|| GET_CODE (SET_SRC (temp)) == CONST_DOUBLE)
|
2547 |
|
|
&& GET_CODE (PATTERN (i3)) == SET
|
2548 |
|
|
&& (CONST_INT_P (SET_SRC (PATTERN (i3)))
|
2549 |
|
|
|| GET_CODE (SET_SRC (PATTERN (i3))) == CONST_DOUBLE)
|
2550 |
|
|
&& reg_subword_p (SET_DEST (PATTERN (i3)), SET_DEST (temp)))
|
2551 |
|
|
{
|
2552 |
|
|
rtx dest = SET_DEST (PATTERN (i3));
|
2553 |
|
|
int offset = -1;
|
2554 |
|
|
int width = 0;
|
2555 |
|
|
|
2556 |
|
|
if (GET_CODE (dest) == ZERO_EXTRACT)
|
2557 |
|
|
{
|
2558 |
|
|
if (CONST_INT_P (XEXP (dest, 1))
|
2559 |
|
|
&& CONST_INT_P (XEXP (dest, 2)))
|
2560 |
|
|
{
|
2561 |
|
|
width = INTVAL (XEXP (dest, 1));
|
2562 |
|
|
offset = INTVAL (XEXP (dest, 2));
|
2563 |
|
|
dest = XEXP (dest, 0);
|
2564 |
|
|
if (BITS_BIG_ENDIAN)
|
2565 |
|
|
offset = GET_MODE_BITSIZE (GET_MODE (dest)) - width - offset;
|
2566 |
|
|
}
|
2567 |
|
|
}
|
2568 |
|
|
else
|
2569 |
|
|
{
|
2570 |
|
|
if (GET_CODE (dest) == STRICT_LOW_PART)
|
2571 |
|
|
dest = XEXP (dest, 0);
|
2572 |
|
|
width = GET_MODE_BITSIZE (GET_MODE (dest));
|
2573 |
|
|
offset = 0;
|
2574 |
|
|
}
|
2575 |
|
|
|
2576 |
|
|
if (offset >= 0)
|
2577 |
|
|
{
|
2578 |
|
|
/* If this is the low part, we're done. */
|
2579 |
|
|
if (subreg_lowpart_p (dest))
|
2580 |
|
|
;
|
2581 |
|
|
/* Handle the case where inner is twice the size of outer. */
|
2582 |
|
|
else if (GET_MODE_BITSIZE (GET_MODE (SET_DEST (temp)))
|
2583 |
|
|
== 2 * GET_MODE_BITSIZE (GET_MODE (dest)))
|
2584 |
|
|
offset += GET_MODE_BITSIZE (GET_MODE (dest));
|
2585 |
|
|
/* Otherwise give up for now. */
|
2586 |
|
|
else
|
2587 |
|
|
offset = -1;
|
2588 |
|
|
}
|
2589 |
|
|
|
2590 |
|
|
if (offset >= 0
|
2591 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (SET_DEST (temp)))
|
2592 |
|
|
<= HOST_BITS_PER_WIDE_INT * 2))
|
2593 |
|
|
{
|
2594 |
|
|
HOST_WIDE_INT mhi, ohi, ihi;
|
2595 |
|
|
HOST_WIDE_INT mlo, olo, ilo;
|
2596 |
|
|
rtx inner = SET_SRC (PATTERN (i3));
|
2597 |
|
|
rtx outer = SET_SRC (temp);
|
2598 |
|
|
|
2599 |
|
|
if (CONST_INT_P (outer))
|
2600 |
|
|
{
|
2601 |
|
|
olo = INTVAL (outer);
|
2602 |
|
|
ohi = olo < 0 ? -1 : 0;
|
2603 |
|
|
}
|
2604 |
|
|
else
|
2605 |
|
|
{
|
2606 |
|
|
olo = CONST_DOUBLE_LOW (outer);
|
2607 |
|
|
ohi = CONST_DOUBLE_HIGH (outer);
|
2608 |
|
|
}
|
2609 |
|
|
|
2610 |
|
|
if (CONST_INT_P (inner))
|
2611 |
|
|
{
|
2612 |
|
|
ilo = INTVAL (inner);
|
2613 |
|
|
ihi = ilo < 0 ? -1 : 0;
|
2614 |
|
|
}
|
2615 |
|
|
else
|
2616 |
|
|
{
|
2617 |
|
|
ilo = CONST_DOUBLE_LOW (inner);
|
2618 |
|
|
ihi = CONST_DOUBLE_HIGH (inner);
|
2619 |
|
|
}
|
2620 |
|
|
|
2621 |
|
|
if (width < HOST_BITS_PER_WIDE_INT)
|
2622 |
|
|
{
|
2623 |
|
|
mlo = ((unsigned HOST_WIDE_INT) 1 << width) - 1;
|
2624 |
|
|
mhi = 0;
|
2625 |
|
|
}
|
2626 |
|
|
else if (width < HOST_BITS_PER_WIDE_INT * 2)
|
2627 |
|
|
{
|
2628 |
|
|
mhi = ((unsigned HOST_WIDE_INT) 1
|
2629 |
|
|
<< (width - HOST_BITS_PER_WIDE_INT)) - 1;
|
2630 |
|
|
mlo = -1;
|
2631 |
|
|
}
|
2632 |
|
|
else
|
2633 |
|
|
{
|
2634 |
|
|
mlo = -1;
|
2635 |
|
|
mhi = -1;
|
2636 |
|
|
}
|
2637 |
|
|
|
2638 |
|
|
ilo &= mlo;
|
2639 |
|
|
ihi &= mhi;
|
2640 |
|
|
|
2641 |
|
|
if (offset >= HOST_BITS_PER_WIDE_INT)
|
2642 |
|
|
{
|
2643 |
|
|
mhi = mlo << (offset - HOST_BITS_PER_WIDE_INT);
|
2644 |
|
|
mlo = 0;
|
2645 |
|
|
ihi = ilo << (offset - HOST_BITS_PER_WIDE_INT);
|
2646 |
|
|
ilo = 0;
|
2647 |
|
|
}
|
2648 |
|
|
else if (offset > 0)
|
2649 |
|
|
{
|
2650 |
|
|
mhi = (mhi << offset) | ((unsigned HOST_WIDE_INT) mlo
|
2651 |
|
|
>> (HOST_BITS_PER_WIDE_INT - offset));
|
2652 |
|
|
mlo = mlo << offset;
|
2653 |
|
|
ihi = (ihi << offset) | ((unsigned HOST_WIDE_INT) ilo
|
2654 |
|
|
>> (HOST_BITS_PER_WIDE_INT - offset));
|
2655 |
|
|
ilo = ilo << offset;
|
2656 |
|
|
}
|
2657 |
|
|
|
2658 |
|
|
olo = (olo & ~mlo) | ilo;
|
2659 |
|
|
ohi = (ohi & ~mhi) | ihi;
|
2660 |
|
|
|
2661 |
|
|
combine_merges++;
|
2662 |
|
|
subst_insn = i3;
|
2663 |
|
|
subst_low_luid = DF_INSN_LUID (i2);
|
2664 |
|
|
added_sets_2 = added_sets_1 = 0;
|
2665 |
|
|
i2dest = SET_DEST (temp);
|
2666 |
|
|
i2dest_killed = dead_or_set_p (i2, i2dest);
|
2667 |
|
|
|
2668 |
|
|
/* Replace the source in I2 with the new constant and make the
|
2669 |
|
|
resulting insn the new pattern for I3. Then skip to where we
|
2670 |
|
|
validate the pattern. Everything was set up above. */
|
2671 |
|
|
SUBST (SET_SRC (temp),
|
2672 |
|
|
immed_double_const (olo, ohi, GET_MODE (SET_DEST (temp))));
|
2673 |
|
|
|
2674 |
|
|
newpat = PATTERN (i2);
|
2675 |
|
|
|
2676 |
|
|
/* The dest of I3 has been replaced with the dest of I2. */
|
2677 |
|
|
changed_i3_dest = 1;
|
2678 |
|
|
goto validate_replacement;
|
2679 |
|
|
}
|
2680 |
|
|
}
|
2681 |
|
|
|
2682 |
|
|
#ifndef HAVE_cc0
|
2683 |
|
|
/* If we have no I1 and I2 looks like:
|
2684 |
|
|
(parallel [(set (reg:CC X) (compare:CC OP (const_int 0)))
|
2685 |
|
|
(set Y OP)])
|
2686 |
|
|
make up a dummy I1 that is
|
2687 |
|
|
(set Y OP)
|
2688 |
|
|
and change I2 to be
|
2689 |
|
|
(set (reg:CC X) (compare:CC Y (const_int 0)))
|
2690 |
|
|
|
2691 |
|
|
(We can ignore any trailing CLOBBERs.)
|
2692 |
|
|
|
2693 |
|
|
This undoes a previous combination and allows us to match a branch-and-
|
2694 |
|
|
decrement insn. */
|
2695 |
|
|
|
2696 |
|
|
if (i1 == 0 && GET_CODE (PATTERN (i2)) == PARALLEL
|
2697 |
|
|
&& XVECLEN (PATTERN (i2), 0) >= 2
|
2698 |
|
|
&& GET_CODE (XVECEXP (PATTERN (i2), 0, 0)) == SET
|
2699 |
|
|
&& (GET_MODE_CLASS (GET_MODE (SET_DEST (XVECEXP (PATTERN (i2), 0, 0))))
|
2700 |
|
|
== MODE_CC)
|
2701 |
|
|
&& GET_CODE (SET_SRC (XVECEXP (PATTERN (i2), 0, 0))) == COMPARE
|
2702 |
|
|
&& XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 1) == const0_rtx
|
2703 |
|
|
&& GET_CODE (XVECEXP (PATTERN (i2), 0, 1)) == SET
|
2704 |
|
|
&& REG_P (SET_DEST (XVECEXP (PATTERN (i2), 0, 1)))
|
2705 |
|
|
&& rtx_equal_p (XEXP (SET_SRC (XVECEXP (PATTERN (i2), 0, 0)), 0),
|
2706 |
|
|
SET_SRC (XVECEXP (PATTERN (i2), 0, 1))))
|
2707 |
|
|
{
|
2708 |
|
|
for (i = XVECLEN (PATTERN (i2), 0) - 1; i >= 2; i--)
|
2709 |
|
|
if (GET_CODE (XVECEXP (PATTERN (i2), 0, i)) != CLOBBER)
|
2710 |
|
|
break;
|
2711 |
|
|
|
2712 |
|
|
if (i == 1)
|
2713 |
|
|
{
|
2714 |
|
|
/* We make I1 with the same INSN_UID as I2. This gives it
|
2715 |
|
|
the same DF_INSN_LUID for value tracking. Our fake I1 will
|
2716 |
|
|
never appear in the insn stream so giving it the same INSN_UID
|
2717 |
|
|
as I2 will not cause a problem. */
|
2718 |
|
|
|
2719 |
|
|
i1 = gen_rtx_INSN (VOIDmode, INSN_UID (i2), NULL_RTX, i2,
|
2720 |
|
|
BLOCK_FOR_INSN (i2), INSN_LOCATOR (i2),
|
2721 |
|
|
XVECEXP (PATTERN (i2), 0, 1), -1, NULL_RTX);
|
2722 |
|
|
|
2723 |
|
|
SUBST (PATTERN (i2), XVECEXP (PATTERN (i2), 0, 0));
|
2724 |
|
|
SUBST (XEXP (SET_SRC (PATTERN (i2)), 0),
|
2725 |
|
|
SET_DEST (PATTERN (i1)));
|
2726 |
|
|
}
|
2727 |
|
|
}
|
2728 |
|
|
#endif
|
2729 |
|
|
|
2730 |
|
|
/* Verify that I2 and I1 are valid for combining. */
|
2731 |
|
|
if (! can_combine_p (i2, i3, i1, NULL_RTX, &i2dest, &i2src)
|
2732 |
|
|
|| (i1 && ! can_combine_p (i1, i3, NULL_RTX, i2, &i1dest, &i1src)))
|
2733 |
|
|
{
|
2734 |
|
|
undo_all ();
|
2735 |
|
|
return 0;
|
2736 |
|
|
}
|
2737 |
|
|
|
2738 |
|
|
/* Record whether I2DEST is used in I2SRC and similarly for the other
|
2739 |
|
|
cases. Knowing this will help in register status updating below. */
|
2740 |
|
|
i2dest_in_i2src = reg_overlap_mentioned_p (i2dest, i2src);
|
2741 |
|
|
i1dest_in_i1src = i1 && reg_overlap_mentioned_p (i1dest, i1src);
|
2742 |
|
|
i2dest_in_i1src = i1 && reg_overlap_mentioned_p (i2dest, i1src);
|
2743 |
|
|
i2dest_killed = dead_or_set_p (i2, i2dest);
|
2744 |
|
|
i1dest_killed = i1 && dead_or_set_p (i1, i1dest);
|
2745 |
|
|
|
2746 |
|
|
/* See if I1 directly feeds into I3. It does if I1DEST is not used
|
2747 |
|
|
in I2SRC. */
|
2748 |
|
|
i1_feeds_i3 = i1 && ! reg_overlap_mentioned_p (i1dest, i2src);
|
2749 |
|
|
|
2750 |
|
|
/* Ensure that I3's pattern can be the destination of combines. */
|
2751 |
|
|
if (! combinable_i3pat (i3, &PATTERN (i3), i2dest, i1dest,
|
2752 |
|
|
i1 && i2dest_in_i1src && i1_feeds_i3,
|
2753 |
|
|
&i3dest_killed))
|
2754 |
|
|
{
|
2755 |
|
|
undo_all ();
|
2756 |
|
|
return 0;
|
2757 |
|
|
}
|
2758 |
|
|
|
2759 |
|
|
/* See if any of the insns is a MULT operation. Unless one is, we will
|
2760 |
|
|
reject a combination that is, since it must be slower. Be conservative
|
2761 |
|
|
here. */
|
2762 |
|
|
if (GET_CODE (i2src) == MULT
|
2763 |
|
|
|| (i1 != 0 && GET_CODE (i1src) == MULT)
|
2764 |
|
|
|| (GET_CODE (PATTERN (i3)) == SET
|
2765 |
|
|
&& GET_CODE (SET_SRC (PATTERN (i3))) == MULT))
|
2766 |
|
|
have_mult = 1;
|
2767 |
|
|
|
2768 |
|
|
/* If I3 has an inc, then give up if I1 or I2 uses the reg that is inc'd.
|
2769 |
|
|
We used to do this EXCEPT in one case: I3 has a post-inc in an
|
2770 |
|
|
output operand. However, that exception can give rise to insns like
|
2771 |
|
|
mov r3,(r3)+
|
2772 |
|
|
which is a famous insn on the PDP-11 where the value of r3 used as the
|
2773 |
|
|
source was model-dependent. Avoid this sort of thing. */
|
2774 |
|
|
|
2775 |
|
|
#if 0
|
2776 |
|
|
if (!(GET_CODE (PATTERN (i3)) == SET
|
2777 |
|
|
&& REG_P (SET_SRC (PATTERN (i3)))
|
2778 |
|
|
&& MEM_P (SET_DEST (PATTERN (i3)))
|
2779 |
|
|
&& (GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_INC
|
2780 |
|
|
|| GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_DEC)))
|
2781 |
|
|
/* It's not the exception. */
|
2782 |
|
|
#endif
|
2783 |
|
|
#ifdef AUTO_INC_DEC
|
2784 |
|
|
for (link = REG_NOTES (i3); link; link = XEXP (link, 1))
|
2785 |
|
|
if (REG_NOTE_KIND (link) == REG_INC
|
2786 |
|
|
&& (reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i2))
|
2787 |
|
|
|| (i1 != 0
|
2788 |
|
|
&& reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i1)))))
|
2789 |
|
|
{
|
2790 |
|
|
undo_all ();
|
2791 |
|
|
return 0;
|
2792 |
|
|
}
|
2793 |
|
|
#endif
|
2794 |
|
|
|
2795 |
|
|
/* See if the SETs in I1 or I2 need to be kept around in the merged
|
2796 |
|
|
instruction: whenever the value set there is still needed past I3.
|
2797 |
|
|
For the SETs in I2, this is easy: we see if I2DEST dies or is set in I3.
|
2798 |
|
|
|
2799 |
|
|
For the SET in I1, we have two cases: If I1 and I2 independently
|
2800 |
|
|
feed into I3, the set in I1 needs to be kept around if I1DEST dies
|
2801 |
|
|
or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
|
2802 |
|
|
in I1 needs to be kept around unless I1DEST dies or is set in either
|
2803 |
|
|
I2 or I3. We can distinguish these cases by seeing if I2SRC mentions
|
2804 |
|
|
I1DEST. If so, we know I1 feeds into I2. */
|
2805 |
|
|
|
2806 |
|
|
added_sets_2 = ! dead_or_set_p (i3, i2dest);
|
2807 |
|
|
|
2808 |
|
|
added_sets_1
|
2809 |
|
|
= i1 && ! (i1_feeds_i3 ? dead_or_set_p (i3, i1dest)
|
2810 |
|
|
: (dead_or_set_p (i3, i1dest) || dead_or_set_p (i2, i1dest)));
|
2811 |
|
|
|
2812 |
|
|
/* If the set in I2 needs to be kept around, we must make a copy of
|
2813 |
|
|
PATTERN (I2), so that when we substitute I1SRC for I1DEST in
|
2814 |
|
|
PATTERN (I2), we are only substituting for the original I1DEST, not into
|
2815 |
|
|
an already-substituted copy. This also prevents making self-referential
|
2816 |
|
|
rtx. If I2 is a PARALLEL, we just need the piece that assigns I2SRC to
|
2817 |
|
|
I2DEST. */
|
2818 |
|
|
|
2819 |
|
|
if (added_sets_2)
|
2820 |
|
|
{
|
2821 |
|
|
if (GET_CODE (PATTERN (i2)) == PARALLEL)
|
2822 |
|
|
i2pat = gen_rtx_SET (VOIDmode, i2dest, copy_rtx (i2src));
|
2823 |
|
|
else
|
2824 |
|
|
i2pat = copy_rtx (PATTERN (i2));
|
2825 |
|
|
}
|
2826 |
|
|
|
2827 |
|
|
if (added_sets_1)
|
2828 |
|
|
{
|
2829 |
|
|
if (GET_CODE (PATTERN (i1)) == PARALLEL)
|
2830 |
|
|
i1pat = gen_rtx_SET (VOIDmode, i1dest, copy_rtx (i1src));
|
2831 |
|
|
else
|
2832 |
|
|
i1pat = copy_rtx (PATTERN (i1));
|
2833 |
|
|
}
|
2834 |
|
|
|
2835 |
|
|
combine_merges++;
|
2836 |
|
|
|
2837 |
|
|
/* Substitute in the latest insn for the regs set by the earlier ones. */
|
2838 |
|
|
|
2839 |
|
|
maxreg = max_reg_num ();
|
2840 |
|
|
|
2841 |
|
|
subst_insn = i3;
|
2842 |
|
|
|
2843 |
|
|
#ifndef HAVE_cc0
|
2844 |
|
|
/* Many machines that don't use CC0 have insns that can both perform an
|
2845 |
|
|
arithmetic operation and set the condition code. These operations will
|
2846 |
|
|
be represented as a PARALLEL with the first element of the vector
|
2847 |
|
|
being a COMPARE of an arithmetic operation with the constant zero.
|
2848 |
|
|
The second element of the vector will set some pseudo to the result
|
2849 |
|
|
of the same arithmetic operation. If we simplify the COMPARE, we won't
|
2850 |
|
|
match such a pattern and so will generate an extra insn. Here we test
|
2851 |
|
|
for this case, where both the comparison and the operation result are
|
2852 |
|
|
needed, and make the PARALLEL by just replacing I2DEST in I3SRC with
|
2853 |
|
|
I2SRC. Later we will make the PARALLEL that contains I2. */
|
2854 |
|
|
|
2855 |
|
|
if (i1 == 0 && added_sets_2 && GET_CODE (PATTERN (i3)) == SET
|
2856 |
|
|
&& GET_CODE (SET_SRC (PATTERN (i3))) == COMPARE
|
2857 |
|
|
&& XEXP (SET_SRC (PATTERN (i3)), 1) == const0_rtx
|
2858 |
|
|
&& rtx_equal_p (XEXP (SET_SRC (PATTERN (i3)), 0), i2dest))
|
2859 |
|
|
{
|
2860 |
|
|
#ifdef SELECT_CC_MODE
|
2861 |
|
|
rtx *cc_use;
|
2862 |
|
|
enum machine_mode compare_mode;
|
2863 |
|
|
#endif
|
2864 |
|
|
|
2865 |
|
|
newpat = PATTERN (i3);
|
2866 |
|
|
SUBST (XEXP (SET_SRC (newpat), 0), i2src);
|
2867 |
|
|
|
2868 |
|
|
i2_is_used = 1;
|
2869 |
|
|
|
2870 |
|
|
#ifdef SELECT_CC_MODE
|
2871 |
|
|
/* See if a COMPARE with the operand we substituted in should be done
|
2872 |
|
|
with the mode that is currently being used. If not, do the same
|
2873 |
|
|
processing we do in `subst' for a SET; namely, if the destination
|
2874 |
|
|
is used only once, try to replace it with a register of the proper
|
2875 |
|
|
mode and also replace the COMPARE. */
|
2876 |
|
|
if (undobuf.other_insn == 0
|
2877 |
|
|
&& (cc_use = find_single_use (SET_DEST (newpat), i3,
|
2878 |
|
|
&undobuf.other_insn))
|
2879 |
|
|
&& ((compare_mode = SELECT_CC_MODE (GET_CODE (*cc_use),
|
2880 |
|
|
i2src, const0_rtx))
|
2881 |
|
|
!= GET_MODE (SET_DEST (newpat))))
|
2882 |
|
|
{
|
2883 |
|
|
if (can_change_dest_mode(SET_DEST (newpat), added_sets_2,
|
2884 |
|
|
compare_mode))
|
2885 |
|
|
{
|
2886 |
|
|
unsigned int regno = REGNO (SET_DEST (newpat));
|
2887 |
|
|
rtx new_dest;
|
2888 |
|
|
|
2889 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
2890 |
|
|
new_dest = gen_rtx_REG (compare_mode, regno);
|
2891 |
|
|
else
|
2892 |
|
|
{
|
2893 |
|
|
SUBST_MODE (regno_reg_rtx[regno], compare_mode);
|
2894 |
|
|
new_dest = regno_reg_rtx[regno];
|
2895 |
|
|
}
|
2896 |
|
|
|
2897 |
|
|
SUBST (SET_DEST (newpat), new_dest);
|
2898 |
|
|
SUBST (XEXP (*cc_use, 0), new_dest);
|
2899 |
|
|
SUBST (SET_SRC (newpat),
|
2900 |
|
|
gen_rtx_COMPARE (compare_mode, i2src, const0_rtx));
|
2901 |
|
|
}
|
2902 |
|
|
else
|
2903 |
|
|
undobuf.other_insn = 0;
|
2904 |
|
|
}
|
2905 |
|
|
#endif
|
2906 |
|
|
}
|
2907 |
|
|
else
|
2908 |
|
|
#endif
|
2909 |
|
|
{
|
2910 |
|
|
/* It is possible that the source of I2 or I1 may be performing
|
2911 |
|
|
an unneeded operation, such as a ZERO_EXTEND of something
|
2912 |
|
|
that is known to have the high part zero. Handle that case
|
2913 |
|
|
by letting subst look at the innermost one of them.
|
2914 |
|
|
|
2915 |
|
|
Another way to do this would be to have a function that tries
|
2916 |
|
|
to simplify a single insn instead of merging two or more
|
2917 |
|
|
insns. We don't do this because of the potential of infinite
|
2918 |
|
|
loops and because of the potential extra memory required.
|
2919 |
|
|
However, doing it the way we are is a bit of a kludge and
|
2920 |
|
|
doesn't catch all cases.
|
2921 |
|
|
|
2922 |
|
|
But only do this if -fexpensive-optimizations since it slows
|
2923 |
|
|
things down and doesn't usually win.
|
2924 |
|
|
|
2925 |
|
|
This is not done in the COMPARE case above because the
|
2926 |
|
|
unmodified I2PAT is used in the PARALLEL and so a pattern
|
2927 |
|
|
with a modified I2SRC would not match. */
|
2928 |
|
|
|
2929 |
|
|
if (flag_expensive_optimizations)
|
2930 |
|
|
{
|
2931 |
|
|
/* Pass pc_rtx so no substitutions are done, just
|
2932 |
|
|
simplifications. */
|
2933 |
|
|
if (i1)
|
2934 |
|
|
{
|
2935 |
|
|
subst_low_luid = DF_INSN_LUID (i1);
|
2936 |
|
|
i1src = subst (i1src, pc_rtx, pc_rtx, 0, 0);
|
2937 |
|
|
}
|
2938 |
|
|
else
|
2939 |
|
|
{
|
2940 |
|
|
subst_low_luid = DF_INSN_LUID (i2);
|
2941 |
|
|
i2src = subst (i2src, pc_rtx, pc_rtx, 0, 0);
|
2942 |
|
|
}
|
2943 |
|
|
}
|
2944 |
|
|
|
2945 |
|
|
n_occurrences = 0; /* `subst' counts here */
|
2946 |
|
|
|
2947 |
|
|
/* If I1 feeds into I2 (not into I3) and I1DEST is in I1SRC, we
|
2948 |
|
|
need to make a unique copy of I2SRC each time we substitute it
|
2949 |
|
|
to avoid self-referential rtl. */
|
2950 |
|
|
|
2951 |
|
|
subst_low_luid = DF_INSN_LUID (i2);
|
2952 |
|
|
newpat = subst (PATTERN (i3), i2dest, i2src, 0,
|
2953 |
|
|
! i1_feeds_i3 && i1dest_in_i1src);
|
2954 |
|
|
substed_i2 = 1;
|
2955 |
|
|
|
2956 |
|
|
/* Record whether i2's body now appears within i3's body. */
|
2957 |
|
|
i2_is_used = n_occurrences;
|
2958 |
|
|
}
|
2959 |
|
|
|
2960 |
|
|
/* If we already got a failure, don't try to do more. Otherwise,
|
2961 |
|
|
try to substitute in I1 if we have it. */
|
2962 |
|
|
|
2963 |
|
|
if (i1 && GET_CODE (newpat) != CLOBBER)
|
2964 |
|
|
{
|
2965 |
|
|
/* Check that an autoincrement side-effect on I1 has not been lost.
|
2966 |
|
|
This happens if I1DEST is mentioned in I2 and dies there, and
|
2967 |
|
|
has disappeared from the new pattern. */
|
2968 |
|
|
if ((FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
|
2969 |
|
|
&& !i1_feeds_i3
|
2970 |
|
|
&& dead_or_set_p (i2, i1dest)
|
2971 |
|
|
&& !reg_overlap_mentioned_p (i1dest, newpat))
|
2972 |
|
|
/* Before we can do this substitution, we must redo the test done
|
2973 |
|
|
above (see detailed comments there) that ensures that I1DEST
|
2974 |
|
|
isn't mentioned in any SETs in NEWPAT that are field assignments. */
|
2975 |
|
|
|| !combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX, 0, 0))
|
2976 |
|
|
{
|
2977 |
|
|
undo_all ();
|
2978 |
|
|
return 0;
|
2979 |
|
|
}
|
2980 |
|
|
|
2981 |
|
|
n_occurrences = 0;
|
2982 |
|
|
subst_low_luid = DF_INSN_LUID (i1);
|
2983 |
|
|
newpat = subst (newpat, i1dest, i1src, 0, 0);
|
2984 |
|
|
substed_i1 = 1;
|
2985 |
|
|
}
|
2986 |
|
|
|
2987 |
|
|
/* Fail if an autoincrement side-effect has been duplicated. Be careful
|
2988 |
|
|
to count all the ways that I2SRC and I1SRC can be used. */
|
2989 |
|
|
if ((FIND_REG_INC_NOTE (i2, NULL_RTX) != 0
|
2990 |
|
|
&& i2_is_used + added_sets_2 > 1)
|
2991 |
|
|
|| (i1 != 0 && FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
|
2992 |
|
|
&& (n_occurrences + added_sets_1 + (added_sets_2 && ! i1_feeds_i3)
|
2993 |
|
|
> 1))
|
2994 |
|
|
/* Fail if we tried to make a new register. */
|
2995 |
|
|
|| max_reg_num () != maxreg
|
2996 |
|
|
/* Fail if we couldn't do something and have a CLOBBER. */
|
2997 |
|
|
|| GET_CODE (newpat) == CLOBBER
|
2998 |
|
|
/* Fail if this new pattern is a MULT and we didn't have one before
|
2999 |
|
|
at the outer level. */
|
3000 |
|
|
|| (GET_CODE (newpat) == SET && GET_CODE (SET_SRC (newpat)) == MULT
|
3001 |
|
|
&& ! have_mult))
|
3002 |
|
|
{
|
3003 |
|
|
undo_all ();
|
3004 |
|
|
return 0;
|
3005 |
|
|
}
|
3006 |
|
|
|
3007 |
|
|
/* If the actions of the earlier insns must be kept
|
3008 |
|
|
in addition to substituting them into the latest one,
|
3009 |
|
|
we must make a new PARALLEL for the latest insn
|
3010 |
|
|
to hold additional the SETs. */
|
3011 |
|
|
|
3012 |
|
|
if (added_sets_1 || added_sets_2)
|
3013 |
|
|
{
|
3014 |
|
|
combine_extras++;
|
3015 |
|
|
|
3016 |
|
|
if (GET_CODE (newpat) == PARALLEL)
|
3017 |
|
|
{
|
3018 |
|
|
rtvec old = XVEC (newpat, 0);
|
3019 |
|
|
total_sets = XVECLEN (newpat, 0) + added_sets_1 + added_sets_2;
|
3020 |
|
|
newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
|
3021 |
|
|
memcpy (XVEC (newpat, 0)->elem, &old->elem[0],
|
3022 |
|
|
sizeof (old->elem[0]) * old->num_elem);
|
3023 |
|
|
}
|
3024 |
|
|
else
|
3025 |
|
|
{
|
3026 |
|
|
rtx old = newpat;
|
3027 |
|
|
total_sets = 1 + added_sets_1 + added_sets_2;
|
3028 |
|
|
newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
|
3029 |
|
|
XVECEXP (newpat, 0, 0) = old;
|
3030 |
|
|
}
|
3031 |
|
|
|
3032 |
|
|
if (added_sets_1)
|
3033 |
|
|
XVECEXP (newpat, 0, --total_sets) = i1pat;
|
3034 |
|
|
|
3035 |
|
|
if (added_sets_2)
|
3036 |
|
|
{
|
3037 |
|
|
/* If there is no I1, use I2's body as is. We used to also not do
|
3038 |
|
|
the subst call below if I2 was substituted into I3,
|
3039 |
|
|
but that could lose a simplification. */
|
3040 |
|
|
if (i1 == 0)
|
3041 |
|
|
XVECEXP (newpat, 0, --total_sets) = i2pat;
|
3042 |
|
|
else
|
3043 |
|
|
/* See comment where i2pat is assigned. */
|
3044 |
|
|
XVECEXP (newpat, 0, --total_sets)
|
3045 |
|
|
= subst (i2pat, i1dest, i1src, 0, 0);
|
3046 |
|
|
}
|
3047 |
|
|
}
|
3048 |
|
|
|
3049 |
|
|
validate_replacement:
|
3050 |
|
|
|
3051 |
|
|
/* Note which hard regs this insn has as inputs. */
|
3052 |
|
|
mark_used_regs_combine (newpat);
|
3053 |
|
|
|
3054 |
|
|
/* If recog_for_combine fails, it strips existing clobbers. If we'll
|
3055 |
|
|
consider splitting this pattern, we might need these clobbers. */
|
3056 |
|
|
if (i1 && GET_CODE (newpat) == PARALLEL
|
3057 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, XVECLEN (newpat, 0) - 1)) == CLOBBER)
|
3058 |
|
|
{
|
3059 |
|
|
int len = XVECLEN (newpat, 0);
|
3060 |
|
|
|
3061 |
|
|
newpat_vec_with_clobbers = rtvec_alloc (len);
|
3062 |
|
|
for (i = 0; i < len; i++)
|
3063 |
|
|
RTVEC_ELT (newpat_vec_with_clobbers, i) = XVECEXP (newpat, 0, i);
|
3064 |
|
|
}
|
3065 |
|
|
|
3066 |
|
|
/* Is the result of combination a valid instruction? */
|
3067 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3068 |
|
|
|
3069 |
|
|
/* If the result isn't valid, see if it is a PARALLEL of two SETs where
|
3070 |
|
|
the second SET's destination is a register that is unused and isn't
|
3071 |
|
|
marked as an instruction that might trap in an EH region. In that case,
|
3072 |
|
|
we just need the first SET. This can occur when simplifying a divmod
|
3073 |
|
|
insn. We *must* test for this case here because the code below that
|
3074 |
|
|
splits two independent SETs doesn't handle this case correctly when it
|
3075 |
|
|
updates the register status.
|
3076 |
|
|
|
3077 |
|
|
It's pointless doing this if we originally had two sets, one from
|
3078 |
|
|
i3, and one from i2. Combining then splitting the parallel results
|
3079 |
|
|
in the original i2 again plus an invalid insn (which we delete).
|
3080 |
|
|
The net effect is only to move instructions around, which makes
|
3081 |
|
|
debug info less accurate.
|
3082 |
|
|
|
3083 |
|
|
Also check the case where the first SET's destination is unused.
|
3084 |
|
|
That would not cause incorrect code, but does cause an unneeded
|
3085 |
|
|
insn to remain. */
|
3086 |
|
|
|
3087 |
|
|
if (insn_code_number < 0
|
3088 |
|
|
&& !(added_sets_2 && i1 == 0)
|
3089 |
|
|
&& GET_CODE (newpat) == PARALLEL
|
3090 |
|
|
&& XVECLEN (newpat, 0) == 2
|
3091 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 0)) == SET
|
3092 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 1)) == SET
|
3093 |
|
|
&& asm_noperands (newpat) < 0)
|
3094 |
|
|
{
|
3095 |
|
|
rtx set0 = XVECEXP (newpat, 0, 0);
|
3096 |
|
|
rtx set1 = XVECEXP (newpat, 0, 1);
|
3097 |
|
|
|
3098 |
|
|
if (((REG_P (SET_DEST (set1))
|
3099 |
|
|
&& find_reg_note (i3, REG_UNUSED, SET_DEST (set1)))
|
3100 |
|
|
|| (GET_CODE (SET_DEST (set1)) == SUBREG
|
3101 |
|
|
&& find_reg_note (i3, REG_UNUSED, SUBREG_REG (SET_DEST (set1)))))
|
3102 |
|
|
&& insn_nothrow_p (i3)
|
3103 |
|
|
&& !side_effects_p (SET_SRC (set1)))
|
3104 |
|
|
{
|
3105 |
|
|
newpat = set0;
|
3106 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3107 |
|
|
}
|
3108 |
|
|
|
3109 |
|
|
else if (((REG_P (SET_DEST (set0))
|
3110 |
|
|
&& find_reg_note (i3, REG_UNUSED, SET_DEST (set0)))
|
3111 |
|
|
|| (GET_CODE (SET_DEST (set0)) == SUBREG
|
3112 |
|
|
&& find_reg_note (i3, REG_UNUSED,
|
3113 |
|
|
SUBREG_REG (SET_DEST (set0)))))
|
3114 |
|
|
&& insn_nothrow_p (i3)
|
3115 |
|
|
&& !side_effects_p (SET_SRC (set0)))
|
3116 |
|
|
{
|
3117 |
|
|
newpat = set1;
|
3118 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3119 |
|
|
|
3120 |
|
|
if (insn_code_number >= 0)
|
3121 |
|
|
changed_i3_dest = 1;
|
3122 |
|
|
}
|
3123 |
|
|
}
|
3124 |
|
|
|
3125 |
|
|
/* If we were combining three insns and the result is a simple SET
|
3126 |
|
|
with no ASM_OPERANDS that wasn't recognized, try to split it into two
|
3127 |
|
|
insns. There are two ways to do this. It can be split using a
|
3128 |
|
|
machine-specific method (like when you have an addition of a large
|
3129 |
|
|
constant) or by combine in the function find_split_point. */
|
3130 |
|
|
|
3131 |
|
|
if (i1 && insn_code_number < 0 && GET_CODE (newpat) == SET
|
3132 |
|
|
&& asm_noperands (newpat) < 0)
|
3133 |
|
|
{
|
3134 |
|
|
rtx parallel, m_split, *split;
|
3135 |
|
|
|
3136 |
|
|
/* See if the MD file can split NEWPAT. If it can't, see if letting it
|
3137 |
|
|
use I2DEST as a scratch register will help. In the latter case,
|
3138 |
|
|
convert I2DEST to the mode of the source of NEWPAT if we can. */
|
3139 |
|
|
|
3140 |
|
|
m_split = combine_split_insns (newpat, i3);
|
3141 |
|
|
|
3142 |
|
|
/* We can only use I2DEST as a scratch reg if it doesn't overlap any
|
3143 |
|
|
inputs of NEWPAT. */
|
3144 |
|
|
|
3145 |
|
|
/* ??? If I2DEST is not safe, and I1DEST exists, then it would be
|
3146 |
|
|
possible to try that as a scratch reg. This would require adding
|
3147 |
|
|
more code to make it work though. */
|
3148 |
|
|
|
3149 |
|
|
if (m_split == 0 && ! reg_overlap_mentioned_p (i2dest, newpat))
|
3150 |
|
|
{
|
3151 |
|
|
enum machine_mode new_mode = GET_MODE (SET_DEST (newpat));
|
3152 |
|
|
|
3153 |
|
|
/* First try to split using the original register as a
|
3154 |
|
|
scratch register. */
|
3155 |
|
|
parallel = gen_rtx_PARALLEL (VOIDmode,
|
3156 |
|
|
gen_rtvec (2, newpat,
|
3157 |
|
|
gen_rtx_CLOBBER (VOIDmode,
|
3158 |
|
|
i2dest)));
|
3159 |
|
|
m_split = combine_split_insns (parallel, i3);
|
3160 |
|
|
|
3161 |
|
|
/* If that didn't work, try changing the mode of I2DEST if
|
3162 |
|
|
we can. */
|
3163 |
|
|
if (m_split == 0
|
3164 |
|
|
&& new_mode != GET_MODE (i2dest)
|
3165 |
|
|
&& new_mode != VOIDmode
|
3166 |
|
|
&& can_change_dest_mode (i2dest, added_sets_2, new_mode))
|
3167 |
|
|
{
|
3168 |
|
|
enum machine_mode old_mode = GET_MODE (i2dest);
|
3169 |
|
|
rtx ni2dest;
|
3170 |
|
|
|
3171 |
|
|
if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
|
3172 |
|
|
ni2dest = gen_rtx_REG (new_mode, REGNO (i2dest));
|
3173 |
|
|
else
|
3174 |
|
|
{
|
3175 |
|
|
SUBST_MODE (regno_reg_rtx[REGNO (i2dest)], new_mode);
|
3176 |
|
|
ni2dest = regno_reg_rtx[REGNO (i2dest)];
|
3177 |
|
|
}
|
3178 |
|
|
|
3179 |
|
|
parallel = (gen_rtx_PARALLEL
|
3180 |
|
|
(VOIDmode,
|
3181 |
|
|
gen_rtvec (2, newpat,
|
3182 |
|
|
gen_rtx_CLOBBER (VOIDmode,
|
3183 |
|
|
ni2dest))));
|
3184 |
|
|
m_split = combine_split_insns (parallel, i3);
|
3185 |
|
|
|
3186 |
|
|
if (m_split == 0
|
3187 |
|
|
&& REGNO (i2dest) >= FIRST_PSEUDO_REGISTER)
|
3188 |
|
|
{
|
3189 |
|
|
struct undo *buf;
|
3190 |
|
|
|
3191 |
|
|
adjust_reg_mode (regno_reg_rtx[REGNO (i2dest)], old_mode);
|
3192 |
|
|
buf = undobuf.undos;
|
3193 |
|
|
undobuf.undos = buf->next;
|
3194 |
|
|
buf->next = undobuf.frees;
|
3195 |
|
|
undobuf.frees = buf;
|
3196 |
|
|
}
|
3197 |
|
|
}
|
3198 |
|
|
|
3199 |
|
|
i2scratch = m_split != 0;
|
3200 |
|
|
}
|
3201 |
|
|
|
3202 |
|
|
/* If recog_for_combine has discarded clobbers, try to use them
|
3203 |
|
|
again for the split. */
|
3204 |
|
|
if (m_split == 0 && newpat_vec_with_clobbers)
|
3205 |
|
|
{
|
3206 |
|
|
parallel = gen_rtx_PARALLEL (VOIDmode, newpat_vec_with_clobbers);
|
3207 |
|
|
m_split = combine_split_insns (parallel, i3);
|
3208 |
|
|
}
|
3209 |
|
|
|
3210 |
|
|
if (m_split && NEXT_INSN (m_split) == NULL_RTX)
|
3211 |
|
|
{
|
3212 |
|
|
m_split = PATTERN (m_split);
|
3213 |
|
|
insn_code_number = recog_for_combine (&m_split, i3, &new_i3_notes);
|
3214 |
|
|
if (insn_code_number >= 0)
|
3215 |
|
|
newpat = m_split;
|
3216 |
|
|
}
|
3217 |
|
|
else if (m_split && NEXT_INSN (NEXT_INSN (m_split)) == NULL_RTX
|
3218 |
|
|
&& (next_real_insn (i2) == i3
|
3219 |
|
|
|| ! use_crosses_set_p (PATTERN (m_split), DF_INSN_LUID (i2))))
|
3220 |
|
|
{
|
3221 |
|
|
rtx i2set, i3set;
|
3222 |
|
|
rtx newi3pat = PATTERN (NEXT_INSN (m_split));
|
3223 |
|
|
newi2pat = PATTERN (m_split);
|
3224 |
|
|
|
3225 |
|
|
i3set = single_set (NEXT_INSN (m_split));
|
3226 |
|
|
i2set = single_set (m_split);
|
3227 |
|
|
|
3228 |
|
|
i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
|
3229 |
|
|
|
3230 |
|
|
/* If I2 or I3 has multiple SETs, we won't know how to track
|
3231 |
|
|
register status, so don't use these insns. If I2's destination
|
3232 |
|
|
is used between I2 and I3, we also can't use these insns. */
|
3233 |
|
|
|
3234 |
|
|
if (i2_code_number >= 0 && i2set && i3set
|
3235 |
|
|
&& (next_real_insn (i2) == i3
|
3236 |
|
|
|| ! reg_used_between_p (SET_DEST (i2set), i2, i3)))
|
3237 |
|
|
insn_code_number = recog_for_combine (&newi3pat, i3,
|
3238 |
|
|
&new_i3_notes);
|
3239 |
|
|
if (insn_code_number >= 0)
|
3240 |
|
|
newpat = newi3pat;
|
3241 |
|
|
|
3242 |
|
|
/* It is possible that both insns now set the destination of I3.
|
3243 |
|
|
If so, we must show an extra use of it. */
|
3244 |
|
|
|
3245 |
|
|
if (insn_code_number >= 0)
|
3246 |
|
|
{
|
3247 |
|
|
rtx new_i3_dest = SET_DEST (i3set);
|
3248 |
|
|
rtx new_i2_dest = SET_DEST (i2set);
|
3249 |
|
|
|
3250 |
|
|
while (GET_CODE (new_i3_dest) == ZERO_EXTRACT
|
3251 |
|
|
|| GET_CODE (new_i3_dest) == STRICT_LOW_PART
|
3252 |
|
|
|| GET_CODE (new_i3_dest) == SUBREG)
|
3253 |
|
|
new_i3_dest = XEXP (new_i3_dest, 0);
|
3254 |
|
|
|
3255 |
|
|
while (GET_CODE (new_i2_dest) == ZERO_EXTRACT
|
3256 |
|
|
|| GET_CODE (new_i2_dest) == STRICT_LOW_PART
|
3257 |
|
|
|| GET_CODE (new_i2_dest) == SUBREG)
|
3258 |
|
|
new_i2_dest = XEXP (new_i2_dest, 0);
|
3259 |
|
|
|
3260 |
|
|
if (REG_P (new_i3_dest)
|
3261 |
|
|
&& REG_P (new_i2_dest)
|
3262 |
|
|
&& REGNO (new_i3_dest) == REGNO (new_i2_dest))
|
3263 |
|
|
INC_REG_N_SETS (REGNO (new_i2_dest), 1);
|
3264 |
|
|
}
|
3265 |
|
|
}
|
3266 |
|
|
|
3267 |
|
|
/* If we can split it and use I2DEST, go ahead and see if that
|
3268 |
|
|
helps things be recognized. Verify that none of the registers
|
3269 |
|
|
are set between I2 and I3. */
|
3270 |
|
|
if (insn_code_number < 0 && (split = find_split_point (&newpat, i3)) != 0
|
3271 |
|
|
#ifdef HAVE_cc0
|
3272 |
|
|
&& REG_P (i2dest)
|
3273 |
|
|
#endif
|
3274 |
|
|
/* We need I2DEST in the proper mode. If it is a hard register
|
3275 |
|
|
or the only use of a pseudo, we can change its mode.
|
3276 |
|
|
Make sure we don't change a hard register to have a mode that
|
3277 |
|
|
isn't valid for it, or change the number of registers. */
|
3278 |
|
|
&& (GET_MODE (*split) == GET_MODE (i2dest)
|
3279 |
|
|
|| GET_MODE (*split) == VOIDmode
|
3280 |
|
|
|| can_change_dest_mode (i2dest, added_sets_2,
|
3281 |
|
|
GET_MODE (*split)))
|
3282 |
|
|
&& (next_real_insn (i2) == i3
|
3283 |
|
|
|| ! use_crosses_set_p (*split, DF_INSN_LUID (i2)))
|
3284 |
|
|
/* We can't overwrite I2DEST if its value is still used by
|
3285 |
|
|
NEWPAT. */
|
3286 |
|
|
&& ! reg_referenced_p (i2dest, newpat))
|
3287 |
|
|
{
|
3288 |
|
|
rtx newdest = i2dest;
|
3289 |
|
|
enum rtx_code split_code = GET_CODE (*split);
|
3290 |
|
|
enum machine_mode split_mode = GET_MODE (*split);
|
3291 |
|
|
bool subst_done = false;
|
3292 |
|
|
newi2pat = NULL_RTX;
|
3293 |
|
|
|
3294 |
|
|
i2scratch = true;
|
3295 |
|
|
|
3296 |
|
|
/* Get NEWDEST as a register in the proper mode. We have already
|
3297 |
|
|
validated that we can do this. */
|
3298 |
|
|
if (GET_MODE (i2dest) != split_mode && split_mode != VOIDmode)
|
3299 |
|
|
{
|
3300 |
|
|
if (REGNO (i2dest) < FIRST_PSEUDO_REGISTER)
|
3301 |
|
|
newdest = gen_rtx_REG (split_mode, REGNO (i2dest));
|
3302 |
|
|
else
|
3303 |
|
|
{
|
3304 |
|
|
SUBST_MODE (regno_reg_rtx[REGNO (i2dest)], split_mode);
|
3305 |
|
|
newdest = regno_reg_rtx[REGNO (i2dest)];
|
3306 |
|
|
}
|
3307 |
|
|
}
|
3308 |
|
|
|
3309 |
|
|
/* If *SPLIT is a (mult FOO (const_int pow2)), convert it to
|
3310 |
|
|
an ASHIFT. This can occur if it was inside a PLUS and hence
|
3311 |
|
|
appeared to be a memory address. This is a kludge. */
|
3312 |
|
|
if (split_code == MULT
|
3313 |
|
|
&& CONST_INT_P (XEXP (*split, 1))
|
3314 |
|
|
&& INTVAL (XEXP (*split, 1)) > 0
|
3315 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (*split, 1)))) >= 0)
|
3316 |
|
|
{
|
3317 |
|
|
SUBST (*split, gen_rtx_ASHIFT (split_mode,
|
3318 |
|
|
XEXP (*split, 0), GEN_INT (i)));
|
3319 |
|
|
/* Update split_code because we may not have a multiply
|
3320 |
|
|
anymore. */
|
3321 |
|
|
split_code = GET_CODE (*split);
|
3322 |
|
|
}
|
3323 |
|
|
|
3324 |
|
|
#ifdef INSN_SCHEDULING
|
3325 |
|
|
/* If *SPLIT is a paradoxical SUBREG, when we split it, it should
|
3326 |
|
|
be written as a ZERO_EXTEND. */
|
3327 |
|
|
if (split_code == SUBREG && MEM_P (SUBREG_REG (*split)))
|
3328 |
|
|
{
|
3329 |
|
|
#ifdef LOAD_EXTEND_OP
|
3330 |
|
|
/* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
|
3331 |
|
|
what it really is. */
|
3332 |
|
|
if (LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (*split)))
|
3333 |
|
|
== SIGN_EXTEND)
|
3334 |
|
|
SUBST (*split, gen_rtx_SIGN_EXTEND (split_mode,
|
3335 |
|
|
SUBREG_REG (*split)));
|
3336 |
|
|
else
|
3337 |
|
|
#endif
|
3338 |
|
|
SUBST (*split, gen_rtx_ZERO_EXTEND (split_mode,
|
3339 |
|
|
SUBREG_REG (*split)));
|
3340 |
|
|
}
|
3341 |
|
|
#endif
|
3342 |
|
|
|
3343 |
|
|
/* Attempt to split binary operators using arithmetic identities. */
|
3344 |
|
|
if (BINARY_P (SET_SRC (newpat))
|
3345 |
|
|
&& split_mode == GET_MODE (SET_SRC (newpat))
|
3346 |
|
|
&& ! side_effects_p (SET_SRC (newpat)))
|
3347 |
|
|
{
|
3348 |
|
|
rtx setsrc = SET_SRC (newpat);
|
3349 |
|
|
enum machine_mode mode = GET_MODE (setsrc);
|
3350 |
|
|
enum rtx_code code = GET_CODE (setsrc);
|
3351 |
|
|
rtx src_op0 = XEXP (setsrc, 0);
|
3352 |
|
|
rtx src_op1 = XEXP (setsrc, 1);
|
3353 |
|
|
|
3354 |
|
|
/* Split "X = Y op Y" as "Z = Y; X = Z op Z". */
|
3355 |
|
|
if (rtx_equal_p (src_op0, src_op1))
|
3356 |
|
|
{
|
3357 |
|
|
newi2pat = gen_rtx_SET (VOIDmode, newdest, src_op0);
|
3358 |
|
|
SUBST (XEXP (setsrc, 0), newdest);
|
3359 |
|
|
SUBST (XEXP (setsrc, 1), newdest);
|
3360 |
|
|
subst_done = true;
|
3361 |
|
|
}
|
3362 |
|
|
/* Split "((P op Q) op R) op S" where op is PLUS or MULT. */
|
3363 |
|
|
else if ((code == PLUS || code == MULT)
|
3364 |
|
|
&& GET_CODE (src_op0) == code
|
3365 |
|
|
&& GET_CODE (XEXP (src_op0, 0)) == code
|
3366 |
|
|
&& (INTEGRAL_MODE_P (mode)
|
3367 |
|
|
|| (FLOAT_MODE_P (mode)
|
3368 |
|
|
&& flag_unsafe_math_optimizations)))
|
3369 |
|
|
{
|
3370 |
|
|
rtx p = XEXP (XEXP (src_op0, 0), 0);
|
3371 |
|
|
rtx q = XEXP (XEXP (src_op0, 0), 1);
|
3372 |
|
|
rtx r = XEXP (src_op0, 1);
|
3373 |
|
|
rtx s = src_op1;
|
3374 |
|
|
|
3375 |
|
|
/* Split both "((X op Y) op X) op Y" and
|
3376 |
|
|
"((X op Y) op Y) op X" as "T op T" where T is
|
3377 |
|
|
"X op Y". */
|
3378 |
|
|
if ((rtx_equal_p (p,r) && rtx_equal_p (q,s))
|
3379 |
|
|
|| (rtx_equal_p (p,s) && rtx_equal_p (q,r)))
|
3380 |
|
|
{
|
3381 |
|
|
newi2pat = gen_rtx_SET (VOIDmode, newdest,
|
3382 |
|
|
XEXP (src_op0, 0));
|
3383 |
|
|
SUBST (XEXP (setsrc, 0), newdest);
|
3384 |
|
|
SUBST (XEXP (setsrc, 1), newdest);
|
3385 |
|
|
subst_done = true;
|
3386 |
|
|
}
|
3387 |
|
|
/* Split "((X op X) op Y) op Y)" as "T op T" where
|
3388 |
|
|
T is "X op Y". */
|
3389 |
|
|
else if (rtx_equal_p (p,q) && rtx_equal_p (r,s))
|
3390 |
|
|
{
|
3391 |
|
|
rtx tmp = simplify_gen_binary (code, mode, p, r);
|
3392 |
|
|
newi2pat = gen_rtx_SET (VOIDmode, newdest, tmp);
|
3393 |
|
|
SUBST (XEXP (setsrc, 0), newdest);
|
3394 |
|
|
SUBST (XEXP (setsrc, 1), newdest);
|
3395 |
|
|
subst_done = true;
|
3396 |
|
|
}
|
3397 |
|
|
}
|
3398 |
|
|
}
|
3399 |
|
|
|
3400 |
|
|
if (!subst_done)
|
3401 |
|
|
{
|
3402 |
|
|
newi2pat = gen_rtx_SET (VOIDmode, newdest, *split);
|
3403 |
|
|
SUBST (*split, newdest);
|
3404 |
|
|
}
|
3405 |
|
|
|
3406 |
|
|
i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
|
3407 |
|
|
|
3408 |
|
|
/* recog_for_combine might have added CLOBBERs to newi2pat.
|
3409 |
|
|
Make sure NEWPAT does not depend on the clobbered regs. */
|
3410 |
|
|
if (GET_CODE (newi2pat) == PARALLEL)
|
3411 |
|
|
for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
|
3412 |
|
|
if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
|
3413 |
|
|
{
|
3414 |
|
|
rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
|
3415 |
|
|
if (reg_overlap_mentioned_p (reg, newpat))
|
3416 |
|
|
{
|
3417 |
|
|
undo_all ();
|
3418 |
|
|
return 0;
|
3419 |
|
|
}
|
3420 |
|
|
}
|
3421 |
|
|
|
3422 |
|
|
/* If the split point was a MULT and we didn't have one before,
|
3423 |
|
|
don't use one now. */
|
3424 |
|
|
if (i2_code_number >= 0 && ! (split_code == MULT && ! have_mult))
|
3425 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3426 |
|
|
}
|
3427 |
|
|
}
|
3428 |
|
|
|
3429 |
|
|
/* Check for a case where we loaded from memory in a narrow mode and
|
3430 |
|
|
then sign extended it, but we need both registers. In that case,
|
3431 |
|
|
we have a PARALLEL with both loads from the same memory location.
|
3432 |
|
|
We can split this into a load from memory followed by a register-register
|
3433 |
|
|
copy. This saves at least one insn, more if register allocation can
|
3434 |
|
|
eliminate the copy.
|
3435 |
|
|
|
3436 |
|
|
We cannot do this if the destination of the first assignment is a
|
3437 |
|
|
condition code register or cc0. We eliminate this case by making sure
|
3438 |
|
|
the SET_DEST and SET_SRC have the same mode.
|
3439 |
|
|
|
3440 |
|
|
We cannot do this if the destination of the second assignment is
|
3441 |
|
|
a register that we have already assumed is zero-extended. Similarly
|
3442 |
|
|
for a SUBREG of such a register. */
|
3443 |
|
|
|
3444 |
|
|
else if (i1 && insn_code_number < 0 && asm_noperands (newpat) < 0
|
3445 |
|
|
&& GET_CODE (newpat) == PARALLEL
|
3446 |
|
|
&& XVECLEN (newpat, 0) == 2
|
3447 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 0)) == SET
|
3448 |
|
|
&& GET_CODE (SET_SRC (XVECEXP (newpat, 0, 0))) == SIGN_EXTEND
|
3449 |
|
|
&& (GET_MODE (SET_DEST (XVECEXP (newpat, 0, 0)))
|
3450 |
|
|
== GET_MODE (SET_SRC (XVECEXP (newpat, 0, 0))))
|
3451 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 1)) == SET
|
3452 |
|
|
&& rtx_equal_p (SET_SRC (XVECEXP (newpat, 0, 1)),
|
3453 |
|
|
XEXP (SET_SRC (XVECEXP (newpat, 0, 0)), 0))
|
3454 |
|
|
&& ! use_crosses_set_p (SET_SRC (XVECEXP (newpat, 0, 1)),
|
3455 |
|
|
DF_INSN_LUID (i2))
|
3456 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
|
3457 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
|
3458 |
|
|
&& ! (temp = SET_DEST (XVECEXP (newpat, 0, 1)),
|
3459 |
|
|
(REG_P (temp)
|
3460 |
|
|
&& VEC_index (reg_stat_type, reg_stat,
|
3461 |
|
|
REGNO (temp))->nonzero_bits != 0
|
3462 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (temp)) < BITS_PER_WORD
|
3463 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (temp)) < HOST_BITS_PER_INT
|
3464 |
|
|
&& (VEC_index (reg_stat_type, reg_stat,
|
3465 |
|
|
REGNO (temp))->nonzero_bits
|
3466 |
|
|
!= GET_MODE_MASK (word_mode))))
|
3467 |
|
|
&& ! (GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) == SUBREG
|
3468 |
|
|
&& (temp = SUBREG_REG (SET_DEST (XVECEXP (newpat, 0, 1))),
|
3469 |
|
|
(REG_P (temp)
|
3470 |
|
|
&& VEC_index (reg_stat_type, reg_stat,
|
3471 |
|
|
REGNO (temp))->nonzero_bits != 0
|
3472 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (temp)) < BITS_PER_WORD
|
3473 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (temp)) < HOST_BITS_PER_INT
|
3474 |
|
|
&& (VEC_index (reg_stat_type, reg_stat,
|
3475 |
|
|
REGNO (temp))->nonzero_bits
|
3476 |
|
|
!= GET_MODE_MASK (word_mode)))))
|
3477 |
|
|
&& ! reg_overlap_mentioned_p (SET_DEST (XVECEXP (newpat, 0, 1)),
|
3478 |
|
|
SET_SRC (XVECEXP (newpat, 0, 1)))
|
3479 |
|
|
&& ! find_reg_note (i3, REG_UNUSED,
|
3480 |
|
|
SET_DEST (XVECEXP (newpat, 0, 0))))
|
3481 |
|
|
{
|
3482 |
|
|
rtx ni2dest;
|
3483 |
|
|
|
3484 |
|
|
newi2pat = XVECEXP (newpat, 0, 0);
|
3485 |
|
|
ni2dest = SET_DEST (XVECEXP (newpat, 0, 0));
|
3486 |
|
|
newpat = XVECEXP (newpat, 0, 1);
|
3487 |
|
|
SUBST (SET_SRC (newpat),
|
3488 |
|
|
gen_lowpart (GET_MODE (SET_SRC (newpat)), ni2dest));
|
3489 |
|
|
i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
|
3490 |
|
|
|
3491 |
|
|
if (i2_code_number >= 0)
|
3492 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3493 |
|
|
|
3494 |
|
|
if (insn_code_number >= 0)
|
3495 |
|
|
swap_i2i3 = 1;
|
3496 |
|
|
}
|
3497 |
|
|
|
3498 |
|
|
/* Similarly, check for a case where we have a PARALLEL of two independent
|
3499 |
|
|
SETs but we started with three insns. In this case, we can do the sets
|
3500 |
|
|
as two separate insns. This case occurs when some SET allows two
|
3501 |
|
|
other insns to combine, but the destination of that SET is still live. */
|
3502 |
|
|
|
3503 |
|
|
else if (i1 && insn_code_number < 0 && asm_noperands (newpat) < 0
|
3504 |
|
|
&& GET_CODE (newpat) == PARALLEL
|
3505 |
|
|
&& XVECLEN (newpat, 0) == 2
|
3506 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 0)) == SET
|
3507 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != ZERO_EXTRACT
|
3508 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != STRICT_LOW_PART
|
3509 |
|
|
&& GET_CODE (XVECEXP (newpat, 0, 1)) == SET
|
3510 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
|
3511 |
|
|
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
|
3512 |
|
|
&& ! use_crosses_set_p (SET_SRC (XVECEXP (newpat, 0, 1)),
|
3513 |
|
|
DF_INSN_LUID (i2))
|
3514 |
|
|
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 1)),
|
3515 |
|
|
XVECEXP (newpat, 0, 0))
|
3516 |
|
|
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 0)),
|
3517 |
|
|
XVECEXP (newpat, 0, 1))
|
3518 |
|
|
&& ! (contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 0)))
|
3519 |
|
|
&& contains_muldiv (SET_SRC (XVECEXP (newpat, 0, 1))))
|
3520 |
|
|
#ifdef HAVE_cc0
|
3521 |
|
|
/* We cannot split the parallel into two sets if both sets
|
3522 |
|
|
reference cc0. */
|
3523 |
|
|
&& ! (reg_referenced_p (cc0_rtx, XVECEXP (newpat, 0, 0))
|
3524 |
|
|
&& reg_referenced_p (cc0_rtx, XVECEXP (newpat, 0, 1)))
|
3525 |
|
|
#endif
|
3526 |
|
|
)
|
3527 |
|
|
{
|
3528 |
|
|
/* Normally, it doesn't matter which of the two is done first,
|
3529 |
|
|
but it does if one references cc0. In that case, it has to
|
3530 |
|
|
be first. */
|
3531 |
|
|
#ifdef HAVE_cc0
|
3532 |
|
|
if (reg_referenced_p (cc0_rtx, XVECEXP (newpat, 0, 0)))
|
3533 |
|
|
{
|
3534 |
|
|
newi2pat = XVECEXP (newpat, 0, 0);
|
3535 |
|
|
newpat = XVECEXP (newpat, 0, 1);
|
3536 |
|
|
}
|
3537 |
|
|
else
|
3538 |
|
|
#endif
|
3539 |
|
|
{
|
3540 |
|
|
newi2pat = XVECEXP (newpat, 0, 1);
|
3541 |
|
|
newpat = XVECEXP (newpat, 0, 0);
|
3542 |
|
|
}
|
3543 |
|
|
|
3544 |
|
|
i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
|
3545 |
|
|
|
3546 |
|
|
if (i2_code_number >= 0)
|
3547 |
|
|
insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
|
3548 |
|
|
}
|
3549 |
|
|
|
3550 |
|
|
/* If it still isn't recognized, fail and change things back the way they
|
3551 |
|
|
were. */
|
3552 |
|
|
if ((insn_code_number < 0
|
3553 |
|
|
/* Is the result a reasonable ASM_OPERANDS? */
|
3554 |
|
|
&& (! check_asm_operands (newpat) || added_sets_1 || added_sets_2)))
|
3555 |
|
|
{
|
3556 |
|
|
undo_all ();
|
3557 |
|
|
return 0;
|
3558 |
|
|
}
|
3559 |
|
|
|
3560 |
|
|
/* If we had to change another insn, make sure it is valid also. */
|
3561 |
|
|
if (undobuf.other_insn)
|
3562 |
|
|
{
|
3563 |
|
|
CLEAR_HARD_REG_SET (newpat_used_regs);
|
3564 |
|
|
|
3565 |
|
|
other_pat = PATTERN (undobuf.other_insn);
|
3566 |
|
|
other_code_number = recog_for_combine (&other_pat, undobuf.other_insn,
|
3567 |
|
|
&new_other_notes);
|
3568 |
|
|
|
3569 |
|
|
if (other_code_number < 0 && ! check_asm_operands (other_pat))
|
3570 |
|
|
{
|
3571 |
|
|
undo_all ();
|
3572 |
|
|
return 0;
|
3573 |
|
|
}
|
3574 |
|
|
}
|
3575 |
|
|
|
3576 |
|
|
#ifdef HAVE_cc0
|
3577 |
|
|
/* If I2 is the CC0 setter and I3 is the CC0 user then check whether
|
3578 |
|
|
they are adjacent to each other or not. */
|
3579 |
|
|
{
|
3580 |
|
|
rtx p = prev_nonnote_insn (i3);
|
3581 |
|
|
if (p && p != i2 && NONJUMP_INSN_P (p) && newi2pat
|
3582 |
|
|
&& sets_cc0_p (newi2pat))
|
3583 |
|
|
{
|
3584 |
|
|
undo_all ();
|
3585 |
|
|
return 0;
|
3586 |
|
|
}
|
3587 |
|
|
}
|
3588 |
|
|
#endif
|
3589 |
|
|
|
3590 |
|
|
/* Only allow this combination if insn_rtx_costs reports that the
|
3591 |
|
|
replacement instructions are cheaper than the originals. */
|
3592 |
|
|
if (!combine_validate_cost (i1, i2, i3, newpat, newi2pat, other_pat))
|
3593 |
|
|
{
|
3594 |
|
|
undo_all ();
|
3595 |
|
|
return 0;
|
3596 |
|
|
}
|
3597 |
|
|
|
3598 |
|
|
if (MAY_HAVE_DEBUG_INSNS)
|
3599 |
|
|
{
|
3600 |
|
|
struct undo *undo;
|
3601 |
|
|
|
3602 |
|
|
for (undo = undobuf.undos; undo; undo = undo->next)
|
3603 |
|
|
if (undo->kind == UNDO_MODE)
|
3604 |
|
|
{
|
3605 |
|
|
rtx reg = *undo->where.r;
|
3606 |
|
|
enum machine_mode new_mode = GET_MODE (reg);
|
3607 |
|
|
enum machine_mode old_mode = undo->old_contents.m;
|
3608 |
|
|
|
3609 |
|
|
/* Temporarily revert mode back. */
|
3610 |
|
|
adjust_reg_mode (reg, old_mode);
|
3611 |
|
|
|
3612 |
|
|
if (reg == i2dest && i2scratch)
|
3613 |
|
|
{
|
3614 |
|
|
/* If we used i2dest as a scratch register with a
|
3615 |
|
|
different mode, substitute it for the original
|
3616 |
|
|
i2src while its original mode is temporarily
|
3617 |
|
|
restored, and then clear i2scratch so that we don't
|
3618 |
|
|
do it again later. */
|
3619 |
|
|
propagate_for_debug (i2, i3, reg, i2src, false);
|
3620 |
|
|
i2scratch = false;
|
3621 |
|
|
/* Put back the new mode. */
|
3622 |
|
|
adjust_reg_mode (reg, new_mode);
|
3623 |
|
|
}
|
3624 |
|
|
else
|
3625 |
|
|
{
|
3626 |
|
|
rtx tempreg = gen_raw_REG (old_mode, REGNO (reg));
|
3627 |
|
|
rtx first, last;
|
3628 |
|
|
|
3629 |
|
|
if (reg == i2dest)
|
3630 |
|
|
{
|
3631 |
|
|
first = i2;
|
3632 |
|
|
last = i3;
|
3633 |
|
|
}
|
3634 |
|
|
else
|
3635 |
|
|
{
|
3636 |
|
|
first = i3;
|
3637 |
|
|
last = undobuf.other_insn;
|
3638 |
|
|
gcc_assert (last);
|
3639 |
|
|
}
|
3640 |
|
|
|
3641 |
|
|
/* We're dealing with a reg that changed mode but not
|
3642 |
|
|
meaning, so we want to turn it into a subreg for
|
3643 |
|
|
the new mode. However, because of REG sharing and
|
3644 |
|
|
because its mode had already changed, we have to do
|
3645 |
|
|
it in two steps. First, replace any debug uses of
|
3646 |
|
|
reg, with its original mode temporarily restored,
|
3647 |
|
|
with this copy we have created; then, replace the
|
3648 |
|
|
copy with the SUBREG of the original shared reg,
|
3649 |
|
|
once again changed to the new mode. */
|
3650 |
|
|
propagate_for_debug (first, last, reg, tempreg, false);
|
3651 |
|
|
adjust_reg_mode (reg, new_mode);
|
3652 |
|
|
propagate_for_debug (first, last, tempreg,
|
3653 |
|
|
lowpart_subreg (old_mode, reg, new_mode),
|
3654 |
|
|
false);
|
3655 |
|
|
}
|
3656 |
|
|
}
|
3657 |
|
|
}
|
3658 |
|
|
|
3659 |
|
|
/* If we will be able to accept this, we have made a
|
3660 |
|
|
change to the destination of I3. This requires us to
|
3661 |
|
|
do a few adjustments. */
|
3662 |
|
|
|
3663 |
|
|
if (changed_i3_dest)
|
3664 |
|
|
{
|
3665 |
|
|
PATTERN (i3) = newpat;
|
3666 |
|
|
adjust_for_new_dest (i3);
|
3667 |
|
|
}
|
3668 |
|
|
|
3669 |
|
|
/* We now know that we can do this combination. Merge the insns and
|
3670 |
|
|
update the status of registers and LOG_LINKS. */
|
3671 |
|
|
|
3672 |
|
|
if (undobuf.other_insn)
|
3673 |
|
|
{
|
3674 |
|
|
rtx note, next;
|
3675 |
|
|
|
3676 |
|
|
PATTERN (undobuf.other_insn) = other_pat;
|
3677 |
|
|
|
3678 |
|
|
/* If any of the notes in OTHER_INSN were REG_UNUSED, ensure that they
|
3679 |
|
|
are still valid. Then add any non-duplicate notes added by
|
3680 |
|
|
recog_for_combine. */
|
3681 |
|
|
for (note = REG_NOTES (undobuf.other_insn); note; note = next)
|
3682 |
|
|
{
|
3683 |
|
|
next = XEXP (note, 1);
|
3684 |
|
|
|
3685 |
|
|
if (REG_NOTE_KIND (note) == REG_UNUSED
|
3686 |
|
|
&& ! reg_set_p (XEXP (note, 0), PATTERN (undobuf.other_insn)))
|
3687 |
|
|
remove_note (undobuf.other_insn, note);
|
3688 |
|
|
}
|
3689 |
|
|
|
3690 |
|
|
distribute_notes (new_other_notes, undobuf.other_insn,
|
3691 |
|
|
undobuf.other_insn, NULL_RTX, NULL_RTX, NULL_RTX);
|
3692 |
|
|
}
|
3693 |
|
|
|
3694 |
|
|
if (swap_i2i3)
|
3695 |
|
|
{
|
3696 |
|
|
rtx insn;
|
3697 |
|
|
rtx link;
|
3698 |
|
|
rtx ni2dest;
|
3699 |
|
|
|
3700 |
|
|
/* I3 now uses what used to be its destination and which is now
|
3701 |
|
|
I2's destination. This requires us to do a few adjustments. */
|
3702 |
|
|
PATTERN (i3) = newpat;
|
3703 |
|
|
adjust_for_new_dest (i3);
|
3704 |
|
|
|
3705 |
|
|
/* We need a LOG_LINK from I3 to I2. But we used to have one,
|
3706 |
|
|
so we still will.
|
3707 |
|
|
|
3708 |
|
|
However, some later insn might be using I2's dest and have
|
3709 |
|
|
a LOG_LINK pointing at I3. We must remove this link.
|
3710 |
|
|
The simplest way to remove the link is to point it at I1,
|
3711 |
|
|
which we know will be a NOTE. */
|
3712 |
|
|
|
3713 |
|
|
/* newi2pat is usually a SET here; however, recog_for_combine might
|
3714 |
|
|
have added some clobbers. */
|
3715 |
|
|
if (GET_CODE (newi2pat) == PARALLEL)
|
3716 |
|
|
ni2dest = SET_DEST (XVECEXP (newi2pat, 0, 0));
|
3717 |
|
|
else
|
3718 |
|
|
ni2dest = SET_DEST (newi2pat);
|
3719 |
|
|
|
3720 |
|
|
for (insn = NEXT_INSN (i3);
|
3721 |
|
|
insn && (this_basic_block->next_bb == EXIT_BLOCK_PTR
|
3722 |
|
|
|| insn != BB_HEAD (this_basic_block->next_bb));
|
3723 |
|
|
insn = NEXT_INSN (insn))
|
3724 |
|
|
{
|
3725 |
|
|
if (INSN_P (insn) && reg_referenced_p (ni2dest, PATTERN (insn)))
|
3726 |
|
|
{
|
3727 |
|
|
for (link = LOG_LINKS (insn); link;
|
3728 |
|
|
link = XEXP (link, 1))
|
3729 |
|
|
if (XEXP (link, 0) == i3)
|
3730 |
|
|
XEXP (link, 0) = i1;
|
3731 |
|
|
|
3732 |
|
|
break;
|
3733 |
|
|
}
|
3734 |
|
|
}
|
3735 |
|
|
}
|
3736 |
|
|
|
3737 |
|
|
{
|
3738 |
|
|
rtx i3notes, i2notes, i1notes = 0;
|
3739 |
|
|
rtx i3links, i2links, i1links = 0;
|
3740 |
|
|
rtx midnotes = 0;
|
3741 |
|
|
unsigned int regno;
|
3742 |
|
|
/* Compute which registers we expect to eliminate. newi2pat may be setting
|
3743 |
|
|
either i3dest or i2dest, so we must check it. Also, i1dest may be the
|
3744 |
|
|
same as i3dest, in which case newi2pat may be setting i1dest. */
|
3745 |
|
|
rtx elim_i2 = ((newi2pat && reg_set_p (i2dest, newi2pat))
|
3746 |
|
|
|| i2dest_in_i2src || i2dest_in_i1src
|
3747 |
|
|
|| !i2dest_killed
|
3748 |
|
|
? 0 : i2dest);
|
3749 |
|
|
rtx elim_i1 = (i1 == 0 || i1dest_in_i1src
|
3750 |
|
|
|| (newi2pat && reg_set_p (i1dest, newi2pat))
|
3751 |
|
|
|| !i1dest_killed
|
3752 |
|
|
? 0 : i1dest);
|
3753 |
|
|
|
3754 |
|
|
/* Get the old REG_NOTES and LOG_LINKS from all our insns and
|
3755 |
|
|
clear them. */
|
3756 |
|
|
i3notes = REG_NOTES (i3), i3links = LOG_LINKS (i3);
|
3757 |
|
|
i2notes = REG_NOTES (i2), i2links = LOG_LINKS (i2);
|
3758 |
|
|
if (i1)
|
3759 |
|
|
i1notes = REG_NOTES (i1), i1links = LOG_LINKS (i1);
|
3760 |
|
|
|
3761 |
|
|
/* Ensure that we do not have something that should not be shared but
|
3762 |
|
|
occurs multiple times in the new insns. Check this by first
|
3763 |
|
|
resetting all the `used' flags and then copying anything is shared. */
|
3764 |
|
|
|
3765 |
|
|
reset_used_flags (i3notes);
|
3766 |
|
|
reset_used_flags (i2notes);
|
3767 |
|
|
reset_used_flags (i1notes);
|
3768 |
|
|
reset_used_flags (newpat);
|
3769 |
|
|
reset_used_flags (newi2pat);
|
3770 |
|
|
if (undobuf.other_insn)
|
3771 |
|
|
reset_used_flags (PATTERN (undobuf.other_insn));
|
3772 |
|
|
|
3773 |
|
|
i3notes = copy_rtx_if_shared (i3notes);
|
3774 |
|
|
i2notes = copy_rtx_if_shared (i2notes);
|
3775 |
|
|
i1notes = copy_rtx_if_shared (i1notes);
|
3776 |
|
|
newpat = copy_rtx_if_shared (newpat);
|
3777 |
|
|
newi2pat = copy_rtx_if_shared (newi2pat);
|
3778 |
|
|
if (undobuf.other_insn)
|
3779 |
|
|
reset_used_flags (PATTERN (undobuf.other_insn));
|
3780 |
|
|
|
3781 |
|
|
INSN_CODE (i3) = insn_code_number;
|
3782 |
|
|
PATTERN (i3) = newpat;
|
3783 |
|
|
|
3784 |
|
|
if (CALL_P (i3) && CALL_INSN_FUNCTION_USAGE (i3))
|
3785 |
|
|
{
|
3786 |
|
|
rtx call_usage = CALL_INSN_FUNCTION_USAGE (i3);
|
3787 |
|
|
|
3788 |
|
|
reset_used_flags (call_usage);
|
3789 |
|
|
call_usage = copy_rtx (call_usage);
|
3790 |
|
|
|
3791 |
|
|
if (substed_i2)
|
3792 |
|
|
replace_rtx (call_usage, i2dest, i2src);
|
3793 |
|
|
|
3794 |
|
|
if (substed_i1)
|
3795 |
|
|
replace_rtx (call_usage, i1dest, i1src);
|
3796 |
|
|
|
3797 |
|
|
CALL_INSN_FUNCTION_USAGE (i3) = call_usage;
|
3798 |
|
|
}
|
3799 |
|
|
|
3800 |
|
|
if (undobuf.other_insn)
|
3801 |
|
|
INSN_CODE (undobuf.other_insn) = other_code_number;
|
3802 |
|
|
|
3803 |
|
|
/* We had one special case above where I2 had more than one set and
|
3804 |
|
|
we replaced a destination of one of those sets with the destination
|
3805 |
|
|
of I3. In that case, we have to update LOG_LINKS of insns later
|
3806 |
|
|
in this basic block. Note that this (expensive) case is rare.
|
3807 |
|
|
|
3808 |
|
|
Also, in this case, we must pretend that all REG_NOTEs for I2
|
3809 |
|
|
actually came from I3, so that REG_UNUSED notes from I2 will be
|
3810 |
|
|
properly handled. */
|
3811 |
|
|
|
3812 |
|
|
if (i3_subst_into_i2)
|
3813 |
|
|
{
|
3814 |
|
|
for (i = 0; i < XVECLEN (PATTERN (i2), 0); i++)
|
3815 |
|
|
if ((GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == SET
|
3816 |
|
|
|| GET_CODE (XVECEXP (PATTERN (i2), 0, i)) == CLOBBER)
|
3817 |
|
|
&& REG_P (SET_DEST (XVECEXP (PATTERN (i2), 0, i)))
|
3818 |
|
|
&& SET_DEST (XVECEXP (PATTERN (i2), 0, i)) != i2dest
|
3819 |
|
|
&& ! find_reg_note (i2, REG_UNUSED,
|
3820 |
|
|
SET_DEST (XVECEXP (PATTERN (i2), 0, i))))
|
3821 |
|
|
for (temp = NEXT_INSN (i2);
|
3822 |
|
|
temp && (this_basic_block->next_bb == EXIT_BLOCK_PTR
|
3823 |
|
|
|| BB_HEAD (this_basic_block) != temp);
|
3824 |
|
|
temp = NEXT_INSN (temp))
|
3825 |
|
|
if (temp != i3 && INSN_P (temp))
|
3826 |
|
|
for (link = LOG_LINKS (temp); link; link = XEXP (link, 1))
|
3827 |
|
|
if (XEXP (link, 0) == i2)
|
3828 |
|
|
XEXP (link, 0) = i3;
|
3829 |
|
|
|
3830 |
|
|
if (i3notes)
|
3831 |
|
|
{
|
3832 |
|
|
rtx link = i3notes;
|
3833 |
|
|
while (XEXP (link, 1))
|
3834 |
|
|
link = XEXP (link, 1);
|
3835 |
|
|
XEXP (link, 1) = i2notes;
|
3836 |
|
|
}
|
3837 |
|
|
else
|
3838 |
|
|
i3notes = i2notes;
|
3839 |
|
|
i2notes = 0;
|
3840 |
|
|
}
|
3841 |
|
|
|
3842 |
|
|
LOG_LINKS (i3) = 0;
|
3843 |
|
|
REG_NOTES (i3) = 0;
|
3844 |
|
|
LOG_LINKS (i2) = 0;
|
3845 |
|
|
REG_NOTES (i2) = 0;
|
3846 |
|
|
|
3847 |
|
|
if (newi2pat)
|
3848 |
|
|
{
|
3849 |
|
|
if (MAY_HAVE_DEBUG_INSNS && i2scratch)
|
3850 |
|
|
propagate_for_debug (i2, i3, i2dest, i2src, false);
|
3851 |
|
|
INSN_CODE (i2) = i2_code_number;
|
3852 |
|
|
PATTERN (i2) = newi2pat;
|
3853 |
|
|
}
|
3854 |
|
|
else
|
3855 |
|
|
{
|
3856 |
|
|
if (MAY_HAVE_DEBUG_INSNS && i2src)
|
3857 |
|
|
propagate_for_debug (i2, i3, i2dest, i2src, i3_subst_into_i2);
|
3858 |
|
|
SET_INSN_DELETED (i2);
|
3859 |
|
|
}
|
3860 |
|
|
|
3861 |
|
|
if (i1)
|
3862 |
|
|
{
|
3863 |
|
|
LOG_LINKS (i1) = 0;
|
3864 |
|
|
REG_NOTES (i1) = 0;
|
3865 |
|
|
if (MAY_HAVE_DEBUG_INSNS)
|
3866 |
|
|
propagate_for_debug (i1, i3, i1dest, i1src, false);
|
3867 |
|
|
SET_INSN_DELETED (i1);
|
3868 |
|
|
}
|
3869 |
|
|
|
3870 |
|
|
/* Get death notes for everything that is now used in either I3 or
|
3871 |
|
|
I2 and used to die in a previous insn. If we built two new
|
3872 |
|
|
patterns, move from I1 to I2 then I2 to I3 so that we get the
|
3873 |
|
|
proper movement on registers that I2 modifies. */
|
3874 |
|
|
|
3875 |
|
|
if (newi2pat)
|
3876 |
|
|
{
|
3877 |
|
|
move_deaths (newi2pat, NULL_RTX, DF_INSN_LUID (i1), i2, &midnotes);
|
3878 |
|
|
move_deaths (newpat, newi2pat, DF_INSN_LUID (i1), i3, &midnotes);
|
3879 |
|
|
}
|
3880 |
|
|
else
|
3881 |
|
|
move_deaths (newpat, NULL_RTX, i1 ? DF_INSN_LUID (i1) : DF_INSN_LUID (i2),
|
3882 |
|
|
i3, &midnotes);
|
3883 |
|
|
|
3884 |
|
|
/* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
|
3885 |
|
|
if (i3notes)
|
3886 |
|
|
distribute_notes (i3notes, i3, i3, newi2pat ? i2 : NULL_RTX,
|
3887 |
|
|
elim_i2, elim_i1);
|
3888 |
|
|
if (i2notes)
|
3889 |
|
|
distribute_notes (i2notes, i2, i3, newi2pat ? i2 : NULL_RTX,
|
3890 |
|
|
elim_i2, elim_i1);
|
3891 |
|
|
if (i1notes)
|
3892 |
|
|
distribute_notes (i1notes, i1, i3, newi2pat ? i2 : NULL_RTX,
|
3893 |
|
|
elim_i2, elim_i1);
|
3894 |
|
|
if (midnotes)
|
3895 |
|
|
distribute_notes (midnotes, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
|
3896 |
|
|
elim_i2, elim_i1);
|
3897 |
|
|
|
3898 |
|
|
/* Distribute any notes added to I2 or I3 by recog_for_combine. We
|
3899 |
|
|
know these are REG_UNUSED and want them to go to the desired insn,
|
3900 |
|
|
so we always pass it as i3. */
|
3901 |
|
|
|
3902 |
|
|
if (newi2pat && new_i2_notes)
|
3903 |
|
|
distribute_notes (new_i2_notes, i2, i2, NULL_RTX, NULL_RTX, NULL_RTX);
|
3904 |
|
|
|
3905 |
|
|
if (new_i3_notes)
|
3906 |
|
|
distribute_notes (new_i3_notes, i3, i3, NULL_RTX, NULL_RTX, NULL_RTX);
|
3907 |
|
|
|
3908 |
|
|
/* If I3DEST was used in I3SRC, it really died in I3. We may need to
|
3909 |
|
|
put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
|
3910 |
|
|
I3DEST, the death must be somewhere before I2, not I3. If we passed I3
|
3911 |
|
|
in that case, it might delete I2. Similarly for I2 and I1.
|
3912 |
|
|
Show an additional death due to the REG_DEAD note we make here. If
|
3913 |
|
|
we discard it in distribute_notes, we will decrement it again. */
|
3914 |
|
|
|
3915 |
|
|
if (i3dest_killed)
|
3916 |
|
|
{
|
3917 |
|
|
if (newi2pat && reg_set_p (i3dest_killed, newi2pat))
|
3918 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
|
3919 |
|
|
NULL_RTX),
|
3920 |
|
|
NULL_RTX, i2, NULL_RTX, elim_i2, elim_i1);
|
3921 |
|
|
else
|
3922 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
|
3923 |
|
|
NULL_RTX),
|
3924 |
|
|
NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
|
3925 |
|
|
elim_i2, elim_i1);
|
3926 |
|
|
}
|
3927 |
|
|
|
3928 |
|
|
if (i2dest_in_i2src)
|
3929 |
|
|
{
|
3930 |
|
|
if (newi2pat && reg_set_p (i2dest, newi2pat))
|
3931 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i2dest, NULL_RTX),
|
3932 |
|
|
NULL_RTX, i2, NULL_RTX, NULL_RTX, NULL_RTX);
|
3933 |
|
|
else
|
3934 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i2dest, NULL_RTX),
|
3935 |
|
|
NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
|
3936 |
|
|
NULL_RTX, NULL_RTX);
|
3937 |
|
|
}
|
3938 |
|
|
|
3939 |
|
|
if (i1dest_in_i1src)
|
3940 |
|
|
{
|
3941 |
|
|
if (newi2pat && reg_set_p (i1dest, newi2pat))
|
3942 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i1dest, NULL_RTX),
|
3943 |
|
|
NULL_RTX, i2, NULL_RTX, NULL_RTX, NULL_RTX);
|
3944 |
|
|
else
|
3945 |
|
|
distribute_notes (alloc_reg_note (REG_DEAD, i1dest, NULL_RTX),
|
3946 |
|
|
NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
|
3947 |
|
|
NULL_RTX, NULL_RTX);
|
3948 |
|
|
}
|
3949 |
|
|
|
3950 |
|
|
distribute_links (i3links);
|
3951 |
|
|
distribute_links (i2links);
|
3952 |
|
|
distribute_links (i1links);
|
3953 |
|
|
|
3954 |
|
|
if (REG_P (i2dest))
|
3955 |
|
|
{
|
3956 |
|
|
rtx link;
|
3957 |
|
|
rtx i2_insn = 0, i2_val = 0, set;
|
3958 |
|
|
|
3959 |
|
|
/* The insn that used to set this register doesn't exist, and
|
3960 |
|
|
this life of the register may not exist either. See if one of
|
3961 |
|
|
I3's links points to an insn that sets I2DEST. If it does,
|
3962 |
|
|
that is now the last known value for I2DEST. If we don't update
|
3963 |
|
|
this and I2 set the register to a value that depended on its old
|
3964 |
|
|
contents, we will get confused. If this insn is used, thing
|
3965 |
|
|
will be set correctly in combine_instructions. */
|
3966 |
|
|
|
3967 |
|
|
for (link = LOG_LINKS (i3); link; link = XEXP (link, 1))
|
3968 |
|
|
if ((set = single_set (XEXP (link, 0))) != 0
|
3969 |
|
|
&& rtx_equal_p (i2dest, SET_DEST (set)))
|
3970 |
|
|
i2_insn = XEXP (link, 0), i2_val = SET_SRC (set);
|
3971 |
|
|
|
3972 |
|
|
record_value_for_reg (i2dest, i2_insn, i2_val);
|
3973 |
|
|
|
3974 |
|
|
/* If the reg formerly set in I2 died only once and that was in I3,
|
3975 |
|
|
zero its use count so it won't make `reload' do any work. */
|
3976 |
|
|
if (! added_sets_2
|
3977 |
|
|
&& (newi2pat == 0 || ! reg_mentioned_p (i2dest, newi2pat))
|
3978 |
|
|
&& ! i2dest_in_i2src)
|
3979 |
|
|
{
|
3980 |
|
|
regno = REGNO (i2dest);
|
3981 |
|
|
INC_REG_N_SETS (regno, -1);
|
3982 |
|
|
}
|
3983 |
|
|
}
|
3984 |
|
|
|
3985 |
|
|
if (i1 && REG_P (i1dest))
|
3986 |
|
|
{
|
3987 |
|
|
rtx link;
|
3988 |
|
|
rtx i1_insn = 0, i1_val = 0, set;
|
3989 |
|
|
|
3990 |
|
|
for (link = LOG_LINKS (i3); link; link = XEXP (link, 1))
|
3991 |
|
|
if ((set = single_set (XEXP (link, 0))) != 0
|
3992 |
|
|
&& rtx_equal_p (i1dest, SET_DEST (set)))
|
3993 |
|
|
i1_insn = XEXP (link, 0), i1_val = SET_SRC (set);
|
3994 |
|
|
|
3995 |
|
|
record_value_for_reg (i1dest, i1_insn, i1_val);
|
3996 |
|
|
|
3997 |
|
|
regno = REGNO (i1dest);
|
3998 |
|
|
if (! added_sets_1 && ! i1dest_in_i1src)
|
3999 |
|
|
INC_REG_N_SETS (regno, -1);
|
4000 |
|
|
}
|
4001 |
|
|
|
4002 |
|
|
/* Update reg_stat[].nonzero_bits et al for any changes that may have
|
4003 |
|
|
been made to this insn. The order of
|
4004 |
|
|
set_nonzero_bits_and_sign_copies() is important. Because newi2pat
|
4005 |
|
|
can affect nonzero_bits of newpat */
|
4006 |
|
|
if (newi2pat)
|
4007 |
|
|
note_stores (newi2pat, set_nonzero_bits_and_sign_copies, NULL);
|
4008 |
|
|
note_stores (newpat, set_nonzero_bits_and_sign_copies, NULL);
|
4009 |
|
|
}
|
4010 |
|
|
|
4011 |
|
|
if (undobuf.other_insn != NULL_RTX)
|
4012 |
|
|
{
|
4013 |
|
|
if (dump_file)
|
4014 |
|
|
{
|
4015 |
|
|
fprintf (dump_file, "modifying other_insn ");
|
4016 |
|
|
dump_insn_slim (dump_file, undobuf.other_insn);
|
4017 |
|
|
}
|
4018 |
|
|
df_insn_rescan (undobuf.other_insn);
|
4019 |
|
|
}
|
4020 |
|
|
|
4021 |
|
|
if (i1 && !(NOTE_P(i1) && (NOTE_KIND (i1) == NOTE_INSN_DELETED)))
|
4022 |
|
|
{
|
4023 |
|
|
if (dump_file)
|
4024 |
|
|
{
|
4025 |
|
|
fprintf (dump_file, "modifying insn i1 ");
|
4026 |
|
|
dump_insn_slim (dump_file, i1);
|
4027 |
|
|
}
|
4028 |
|
|
df_insn_rescan (i1);
|
4029 |
|
|
}
|
4030 |
|
|
|
4031 |
|
|
if (i2 && !(NOTE_P(i2) && (NOTE_KIND (i2) == NOTE_INSN_DELETED)))
|
4032 |
|
|
{
|
4033 |
|
|
if (dump_file)
|
4034 |
|
|
{
|
4035 |
|
|
fprintf (dump_file, "modifying insn i2 ");
|
4036 |
|
|
dump_insn_slim (dump_file, i2);
|
4037 |
|
|
}
|
4038 |
|
|
df_insn_rescan (i2);
|
4039 |
|
|
}
|
4040 |
|
|
|
4041 |
|
|
if (i3 && !(NOTE_P(i3) && (NOTE_KIND (i3) == NOTE_INSN_DELETED)))
|
4042 |
|
|
{
|
4043 |
|
|
if (dump_file)
|
4044 |
|
|
{
|
4045 |
|
|
fprintf (dump_file, "modifying insn i3 ");
|
4046 |
|
|
dump_insn_slim (dump_file, i3);
|
4047 |
|
|
}
|
4048 |
|
|
df_insn_rescan (i3);
|
4049 |
|
|
}
|
4050 |
|
|
|
4051 |
|
|
/* Set new_direct_jump_p if a new return or simple jump instruction
|
4052 |
|
|
has been created. Adjust the CFG accordingly. */
|
4053 |
|
|
|
4054 |
|
|
if (returnjump_p (i3) || any_uncondjump_p (i3))
|
4055 |
|
|
{
|
4056 |
|
|
*new_direct_jump_p = 1;
|
4057 |
|
|
mark_jump_label (PATTERN (i3), i3, 0);
|
4058 |
|
|
update_cfg_for_uncondjump (i3);
|
4059 |
|
|
}
|
4060 |
|
|
|
4061 |
|
|
if (undobuf.other_insn != NULL_RTX
|
4062 |
|
|
&& (returnjump_p (undobuf.other_insn)
|
4063 |
|
|
|| any_uncondjump_p (undobuf.other_insn)))
|
4064 |
|
|
{
|
4065 |
|
|
*new_direct_jump_p = 1;
|
4066 |
|
|
update_cfg_for_uncondjump (undobuf.other_insn);
|
4067 |
|
|
}
|
4068 |
|
|
|
4069 |
|
|
/* A noop might also need cleaning up of CFG, if it comes from the
|
4070 |
|
|
simplification of a jump. */
|
4071 |
|
|
if (GET_CODE (newpat) == SET
|
4072 |
|
|
&& SET_SRC (newpat) == pc_rtx
|
4073 |
|
|
&& SET_DEST (newpat) == pc_rtx)
|
4074 |
|
|
{
|
4075 |
|
|
*new_direct_jump_p = 1;
|
4076 |
|
|
update_cfg_for_uncondjump (i3);
|
4077 |
|
|
}
|
4078 |
|
|
|
4079 |
|
|
combine_successes++;
|
4080 |
|
|
undo_commit ();
|
4081 |
|
|
|
4082 |
|
|
if (added_links_insn
|
4083 |
|
|
&& (newi2pat == 0 || DF_INSN_LUID (added_links_insn) < DF_INSN_LUID (i2))
|
4084 |
|
|
&& DF_INSN_LUID (added_links_insn) < DF_INSN_LUID (i3))
|
4085 |
|
|
return added_links_insn;
|
4086 |
|
|
else
|
4087 |
|
|
return newi2pat ? i2 : i3;
|
4088 |
|
|
}
|
4089 |
|
|
|
4090 |
|
|
/* Undo all the modifications recorded in undobuf. */
|
4091 |
|
|
|
4092 |
|
|
static void
|
4093 |
|
|
undo_all (void)
|
4094 |
|
|
{
|
4095 |
|
|
struct undo *undo, *next;
|
4096 |
|
|
|
4097 |
|
|
for (undo = undobuf.undos; undo; undo = next)
|
4098 |
|
|
{
|
4099 |
|
|
next = undo->next;
|
4100 |
|
|
switch (undo->kind)
|
4101 |
|
|
{
|
4102 |
|
|
case UNDO_RTX:
|
4103 |
|
|
*undo->where.r = undo->old_contents.r;
|
4104 |
|
|
break;
|
4105 |
|
|
case UNDO_INT:
|
4106 |
|
|
*undo->where.i = undo->old_contents.i;
|
4107 |
|
|
break;
|
4108 |
|
|
case UNDO_MODE:
|
4109 |
|
|
adjust_reg_mode (*undo->where.r, undo->old_contents.m);
|
4110 |
|
|
break;
|
4111 |
|
|
default:
|
4112 |
|
|
gcc_unreachable ();
|
4113 |
|
|
}
|
4114 |
|
|
|
4115 |
|
|
undo->next = undobuf.frees;
|
4116 |
|
|
undobuf.frees = undo;
|
4117 |
|
|
}
|
4118 |
|
|
|
4119 |
|
|
undobuf.undos = 0;
|
4120 |
|
|
}
|
4121 |
|
|
|
4122 |
|
|
/* We've committed to accepting the changes we made. Move all
|
4123 |
|
|
of the undos to the free list. */
|
4124 |
|
|
|
4125 |
|
|
static void
|
4126 |
|
|
undo_commit (void)
|
4127 |
|
|
{
|
4128 |
|
|
struct undo *undo, *next;
|
4129 |
|
|
|
4130 |
|
|
for (undo = undobuf.undos; undo; undo = next)
|
4131 |
|
|
{
|
4132 |
|
|
next = undo->next;
|
4133 |
|
|
undo->next = undobuf.frees;
|
4134 |
|
|
undobuf.frees = undo;
|
4135 |
|
|
}
|
4136 |
|
|
undobuf.undos = 0;
|
4137 |
|
|
}
|
4138 |
|
|
|
4139 |
|
|
/* Find the innermost point within the rtx at LOC, possibly LOC itself,
|
4140 |
|
|
where we have an arithmetic expression and return that point. LOC will
|
4141 |
|
|
be inside INSN.
|
4142 |
|
|
|
4143 |
|
|
try_combine will call this function to see if an insn can be split into
|
4144 |
|
|
two insns. */
|
4145 |
|
|
|
4146 |
|
|
static rtx *
|
4147 |
|
|
find_split_point (rtx *loc, rtx insn)
|
4148 |
|
|
{
|
4149 |
|
|
rtx x = *loc;
|
4150 |
|
|
enum rtx_code code = GET_CODE (x);
|
4151 |
|
|
rtx *split;
|
4152 |
|
|
unsigned HOST_WIDE_INT len = 0;
|
4153 |
|
|
HOST_WIDE_INT pos = 0;
|
4154 |
|
|
int unsignedp = 0;
|
4155 |
|
|
rtx inner = NULL_RTX;
|
4156 |
|
|
|
4157 |
|
|
/* First special-case some codes. */
|
4158 |
|
|
switch (code)
|
4159 |
|
|
{
|
4160 |
|
|
case SUBREG:
|
4161 |
|
|
#ifdef INSN_SCHEDULING
|
4162 |
|
|
/* If we are making a paradoxical SUBREG invalid, it becomes a split
|
4163 |
|
|
point. */
|
4164 |
|
|
if (MEM_P (SUBREG_REG (x)))
|
4165 |
|
|
return loc;
|
4166 |
|
|
#endif
|
4167 |
|
|
return find_split_point (&SUBREG_REG (x), insn);
|
4168 |
|
|
|
4169 |
|
|
case MEM:
|
4170 |
|
|
#ifdef HAVE_lo_sum
|
4171 |
|
|
/* If we have (mem (const ..)) or (mem (symbol_ref ...)), split it
|
4172 |
|
|
using LO_SUM and HIGH. */
|
4173 |
|
|
if (GET_CODE (XEXP (x, 0)) == CONST
|
4174 |
|
|
|| GET_CODE (XEXP (x, 0)) == SYMBOL_REF)
|
4175 |
|
|
{
|
4176 |
|
|
enum machine_mode address_mode
|
4177 |
|
|
= targetm.addr_space.address_mode (MEM_ADDR_SPACE (x));
|
4178 |
|
|
|
4179 |
|
|
SUBST (XEXP (x, 0),
|
4180 |
|
|
gen_rtx_LO_SUM (address_mode,
|
4181 |
|
|
gen_rtx_HIGH (address_mode, XEXP (x, 0)),
|
4182 |
|
|
XEXP (x, 0)));
|
4183 |
|
|
return &XEXP (XEXP (x, 0), 0);
|
4184 |
|
|
}
|
4185 |
|
|
#endif
|
4186 |
|
|
|
4187 |
|
|
/* If we have a PLUS whose second operand is a constant and the
|
4188 |
|
|
address is not valid, perhaps will can split it up using
|
4189 |
|
|
the machine-specific way to split large constants. We use
|
4190 |
|
|
the first pseudo-reg (one of the virtual regs) as a placeholder;
|
4191 |
|
|
it will not remain in the result. */
|
4192 |
|
|
if (GET_CODE (XEXP (x, 0)) == PLUS
|
4193 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
4194 |
|
|
&& ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
|
4195 |
|
|
MEM_ADDR_SPACE (x)))
|
4196 |
|
|
{
|
4197 |
|
|
rtx reg = regno_reg_rtx[FIRST_PSEUDO_REGISTER];
|
4198 |
|
|
rtx seq = combine_split_insns (gen_rtx_SET (VOIDmode, reg,
|
4199 |
|
|
XEXP (x, 0)),
|
4200 |
|
|
subst_insn);
|
4201 |
|
|
|
4202 |
|
|
/* This should have produced two insns, each of which sets our
|
4203 |
|
|
placeholder. If the source of the second is a valid address,
|
4204 |
|
|
we can make put both sources together and make a split point
|
4205 |
|
|
in the middle. */
|
4206 |
|
|
|
4207 |
|
|
if (seq
|
4208 |
|
|
&& NEXT_INSN (seq) != NULL_RTX
|
4209 |
|
|
&& NEXT_INSN (NEXT_INSN (seq)) == NULL_RTX
|
4210 |
|
|
&& NONJUMP_INSN_P (seq)
|
4211 |
|
|
&& GET_CODE (PATTERN (seq)) == SET
|
4212 |
|
|
&& SET_DEST (PATTERN (seq)) == reg
|
4213 |
|
|
&& ! reg_mentioned_p (reg,
|
4214 |
|
|
SET_SRC (PATTERN (seq)))
|
4215 |
|
|
&& NONJUMP_INSN_P (NEXT_INSN (seq))
|
4216 |
|
|
&& GET_CODE (PATTERN (NEXT_INSN (seq))) == SET
|
4217 |
|
|
&& SET_DEST (PATTERN (NEXT_INSN (seq))) == reg
|
4218 |
|
|
&& memory_address_addr_space_p
|
4219 |
|
|
(GET_MODE (x), SET_SRC (PATTERN (NEXT_INSN (seq))),
|
4220 |
|
|
MEM_ADDR_SPACE (x)))
|
4221 |
|
|
{
|
4222 |
|
|
rtx src1 = SET_SRC (PATTERN (seq));
|
4223 |
|
|
rtx src2 = SET_SRC (PATTERN (NEXT_INSN (seq)));
|
4224 |
|
|
|
4225 |
|
|
/* Replace the placeholder in SRC2 with SRC1. If we can
|
4226 |
|
|
find where in SRC2 it was placed, that can become our
|
4227 |
|
|
split point and we can replace this address with SRC2.
|
4228 |
|
|
Just try two obvious places. */
|
4229 |
|
|
|
4230 |
|
|
src2 = replace_rtx (src2, reg, src1);
|
4231 |
|
|
split = 0;
|
4232 |
|
|
if (XEXP (src2, 0) == src1)
|
4233 |
|
|
split = &XEXP (src2, 0);
|
4234 |
|
|
else if (GET_RTX_FORMAT (GET_CODE (XEXP (src2, 0)))[0] == 'e'
|
4235 |
|
|
&& XEXP (XEXP (src2, 0), 0) == src1)
|
4236 |
|
|
split = &XEXP (XEXP (src2, 0), 0);
|
4237 |
|
|
|
4238 |
|
|
if (split)
|
4239 |
|
|
{
|
4240 |
|
|
SUBST (XEXP (x, 0), src2);
|
4241 |
|
|
return split;
|
4242 |
|
|
}
|
4243 |
|
|
}
|
4244 |
|
|
|
4245 |
|
|
/* If that didn't work, perhaps the first operand is complex and
|
4246 |
|
|
needs to be computed separately, so make a split point there.
|
4247 |
|
|
This will occur on machines that just support REG + CONST
|
4248 |
|
|
and have a constant moved through some previous computation. */
|
4249 |
|
|
|
4250 |
|
|
else if (!OBJECT_P (XEXP (XEXP (x, 0), 0))
|
4251 |
|
|
&& ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
|
4252 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
|
4253 |
|
|
return &XEXP (XEXP (x, 0), 0);
|
4254 |
|
|
}
|
4255 |
|
|
|
4256 |
|
|
/* If we have a PLUS whose first operand is complex, try computing it
|
4257 |
|
|
separately by making a split there. */
|
4258 |
|
|
if (GET_CODE (XEXP (x, 0)) == PLUS
|
4259 |
|
|
&& ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
|
4260 |
|
|
MEM_ADDR_SPACE (x))
|
4261 |
|
|
&& ! OBJECT_P (XEXP (XEXP (x, 0), 0))
|
4262 |
|
|
&& ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
|
4263 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
|
4264 |
|
|
return &XEXP (XEXP (x, 0), 0);
|
4265 |
|
|
break;
|
4266 |
|
|
|
4267 |
|
|
case SET:
|
4268 |
|
|
#ifdef HAVE_cc0
|
4269 |
|
|
/* If SET_DEST is CC0 and SET_SRC is not an operand, a COMPARE, or a
|
4270 |
|
|
ZERO_EXTRACT, the most likely reason why this doesn't match is that
|
4271 |
|
|
we need to put the operand into a register. So split at that
|
4272 |
|
|
point. */
|
4273 |
|
|
|
4274 |
|
|
if (SET_DEST (x) == cc0_rtx
|
4275 |
|
|
&& GET_CODE (SET_SRC (x)) != COMPARE
|
4276 |
|
|
&& GET_CODE (SET_SRC (x)) != ZERO_EXTRACT
|
4277 |
|
|
&& !OBJECT_P (SET_SRC (x))
|
4278 |
|
|
&& ! (GET_CODE (SET_SRC (x)) == SUBREG
|
4279 |
|
|
&& OBJECT_P (SUBREG_REG (SET_SRC (x)))))
|
4280 |
|
|
return &SET_SRC (x);
|
4281 |
|
|
#endif
|
4282 |
|
|
|
4283 |
|
|
/* See if we can split SET_SRC as it stands. */
|
4284 |
|
|
split = find_split_point (&SET_SRC (x), insn);
|
4285 |
|
|
if (split && split != &SET_SRC (x))
|
4286 |
|
|
return split;
|
4287 |
|
|
|
4288 |
|
|
/* See if we can split SET_DEST as it stands. */
|
4289 |
|
|
split = find_split_point (&SET_DEST (x), insn);
|
4290 |
|
|
if (split && split != &SET_DEST (x))
|
4291 |
|
|
return split;
|
4292 |
|
|
|
4293 |
|
|
/* See if this is a bitfield assignment with everything constant. If
|
4294 |
|
|
so, this is an IOR of an AND, so split it into that. */
|
4295 |
|
|
if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
|
4296 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (XEXP (SET_DEST (x), 0)))
|
4297 |
|
|
<= HOST_BITS_PER_WIDE_INT)
|
4298 |
|
|
&& CONST_INT_P (XEXP (SET_DEST (x), 1))
|
4299 |
|
|
&& CONST_INT_P (XEXP (SET_DEST (x), 2))
|
4300 |
|
|
&& CONST_INT_P (SET_SRC (x))
|
4301 |
|
|
&& ((INTVAL (XEXP (SET_DEST (x), 1))
|
4302 |
|
|
+ INTVAL (XEXP (SET_DEST (x), 2)))
|
4303 |
|
|
<= GET_MODE_BITSIZE (GET_MODE (XEXP (SET_DEST (x), 0))))
|
4304 |
|
|
&& ! side_effects_p (XEXP (SET_DEST (x), 0)))
|
4305 |
|
|
{
|
4306 |
|
|
HOST_WIDE_INT pos = INTVAL (XEXP (SET_DEST (x), 2));
|
4307 |
|
|
unsigned HOST_WIDE_INT len = INTVAL (XEXP (SET_DEST (x), 1));
|
4308 |
|
|
unsigned HOST_WIDE_INT src = INTVAL (SET_SRC (x));
|
4309 |
|
|
rtx dest = XEXP (SET_DEST (x), 0);
|
4310 |
|
|
enum machine_mode mode = GET_MODE (dest);
|
4311 |
|
|
unsigned HOST_WIDE_INT mask = ((HOST_WIDE_INT) 1 << len) - 1;
|
4312 |
|
|
rtx or_mask;
|
4313 |
|
|
|
4314 |
|
|
if (BITS_BIG_ENDIAN)
|
4315 |
|
|
pos = GET_MODE_BITSIZE (mode) - len - pos;
|
4316 |
|
|
|
4317 |
|
|
or_mask = gen_int_mode (src << pos, mode);
|
4318 |
|
|
if (src == mask)
|
4319 |
|
|
SUBST (SET_SRC (x),
|
4320 |
|
|
simplify_gen_binary (IOR, mode, dest, or_mask));
|
4321 |
|
|
else
|
4322 |
|
|
{
|
4323 |
|
|
rtx negmask = gen_int_mode (~(mask << pos), mode);
|
4324 |
|
|
SUBST (SET_SRC (x),
|
4325 |
|
|
simplify_gen_binary (IOR, mode,
|
4326 |
|
|
simplify_gen_binary (AND, mode,
|
4327 |
|
|
dest, negmask),
|
4328 |
|
|
or_mask));
|
4329 |
|
|
}
|
4330 |
|
|
|
4331 |
|
|
SUBST (SET_DEST (x), dest);
|
4332 |
|
|
|
4333 |
|
|
split = find_split_point (&SET_SRC (x), insn);
|
4334 |
|
|
if (split && split != &SET_SRC (x))
|
4335 |
|
|
return split;
|
4336 |
|
|
}
|
4337 |
|
|
|
4338 |
|
|
/* Otherwise, see if this is an operation that we can split into two.
|
4339 |
|
|
If so, try to split that. */
|
4340 |
|
|
code = GET_CODE (SET_SRC (x));
|
4341 |
|
|
|
4342 |
|
|
switch (code)
|
4343 |
|
|
{
|
4344 |
|
|
case AND:
|
4345 |
|
|
/* If we are AND'ing with a large constant that is only a single
|
4346 |
|
|
bit and the result is only being used in a context where we
|
4347 |
|
|
need to know if it is zero or nonzero, replace it with a bit
|
4348 |
|
|
extraction. This will avoid the large constant, which might
|
4349 |
|
|
have taken more than one insn to make. If the constant were
|
4350 |
|
|
not a valid argument to the AND but took only one insn to make,
|
4351 |
|
|
this is no worse, but if it took more than one insn, it will
|
4352 |
|
|
be better. */
|
4353 |
|
|
|
4354 |
|
|
if (CONST_INT_P (XEXP (SET_SRC (x), 1))
|
4355 |
|
|
&& REG_P (XEXP (SET_SRC (x), 0))
|
4356 |
|
|
&& (pos = exact_log2 (INTVAL (XEXP (SET_SRC (x), 1)))) >= 7
|
4357 |
|
|
&& REG_P (SET_DEST (x))
|
4358 |
|
|
&& (split = find_single_use (SET_DEST (x), insn, (rtx*) 0)) != 0
|
4359 |
|
|
&& (GET_CODE (*split) == EQ || GET_CODE (*split) == NE)
|
4360 |
|
|
&& XEXP (*split, 0) == SET_DEST (x)
|
4361 |
|
|
&& XEXP (*split, 1) == const0_rtx)
|
4362 |
|
|
{
|
4363 |
|
|
rtx extraction = make_extraction (GET_MODE (SET_DEST (x)),
|
4364 |
|
|
XEXP (SET_SRC (x), 0),
|
4365 |
|
|
pos, NULL_RTX, 1, 1, 0, 0);
|
4366 |
|
|
if (extraction != 0)
|
4367 |
|
|
{
|
4368 |
|
|
SUBST (SET_SRC (x), extraction);
|
4369 |
|
|
return find_split_point (loc, insn);
|
4370 |
|
|
}
|
4371 |
|
|
}
|
4372 |
|
|
break;
|
4373 |
|
|
|
4374 |
|
|
case NE:
|
4375 |
|
|
/* If STORE_FLAG_VALUE is -1, this is (NE X 0) and only one bit of X
|
4376 |
|
|
is known to be on, this can be converted into a NEG of a shift. */
|
4377 |
|
|
if (STORE_FLAG_VALUE == -1 && XEXP (SET_SRC (x), 1) == const0_rtx
|
4378 |
|
|
&& GET_MODE (SET_SRC (x)) == GET_MODE (XEXP (SET_SRC (x), 0))
|
4379 |
|
|
&& 1 <= (pos = exact_log2
|
4380 |
|
|
(nonzero_bits (XEXP (SET_SRC (x), 0),
|
4381 |
|
|
GET_MODE (XEXP (SET_SRC (x), 0))))))
|
4382 |
|
|
{
|
4383 |
|
|
enum machine_mode mode = GET_MODE (XEXP (SET_SRC (x), 0));
|
4384 |
|
|
|
4385 |
|
|
SUBST (SET_SRC (x),
|
4386 |
|
|
gen_rtx_NEG (mode,
|
4387 |
|
|
gen_rtx_LSHIFTRT (mode,
|
4388 |
|
|
XEXP (SET_SRC (x), 0),
|
4389 |
|
|
GEN_INT (pos))));
|
4390 |
|
|
|
4391 |
|
|
split = find_split_point (&SET_SRC (x), insn);
|
4392 |
|
|
if (split && split != &SET_SRC (x))
|
4393 |
|
|
return split;
|
4394 |
|
|
}
|
4395 |
|
|
break;
|
4396 |
|
|
|
4397 |
|
|
case SIGN_EXTEND:
|
4398 |
|
|
inner = XEXP (SET_SRC (x), 0);
|
4399 |
|
|
|
4400 |
|
|
/* We can't optimize if either mode is a partial integer
|
4401 |
|
|
mode as we don't know how many bits are significant
|
4402 |
|
|
in those modes. */
|
4403 |
|
|
if (GET_MODE_CLASS (GET_MODE (inner)) == MODE_PARTIAL_INT
|
4404 |
|
|
|| GET_MODE_CLASS (GET_MODE (SET_SRC (x))) == MODE_PARTIAL_INT)
|
4405 |
|
|
break;
|
4406 |
|
|
|
4407 |
|
|
pos = 0;
|
4408 |
|
|
len = GET_MODE_BITSIZE (GET_MODE (inner));
|
4409 |
|
|
unsignedp = 0;
|
4410 |
|
|
break;
|
4411 |
|
|
|
4412 |
|
|
case SIGN_EXTRACT:
|
4413 |
|
|
case ZERO_EXTRACT:
|
4414 |
|
|
if (CONST_INT_P (XEXP (SET_SRC (x), 1))
|
4415 |
|
|
&& CONST_INT_P (XEXP (SET_SRC (x), 2)))
|
4416 |
|
|
{
|
4417 |
|
|
inner = XEXP (SET_SRC (x), 0);
|
4418 |
|
|
len = INTVAL (XEXP (SET_SRC (x), 1));
|
4419 |
|
|
pos = INTVAL (XEXP (SET_SRC (x), 2));
|
4420 |
|
|
|
4421 |
|
|
if (BITS_BIG_ENDIAN)
|
4422 |
|
|
pos = GET_MODE_BITSIZE (GET_MODE (inner)) - len - pos;
|
4423 |
|
|
unsignedp = (code == ZERO_EXTRACT);
|
4424 |
|
|
}
|
4425 |
|
|
break;
|
4426 |
|
|
|
4427 |
|
|
default:
|
4428 |
|
|
break;
|
4429 |
|
|
}
|
4430 |
|
|
|
4431 |
|
|
if (len && pos >= 0 && pos + len <= GET_MODE_BITSIZE (GET_MODE (inner)))
|
4432 |
|
|
{
|
4433 |
|
|
enum machine_mode mode = GET_MODE (SET_SRC (x));
|
4434 |
|
|
|
4435 |
|
|
/* For unsigned, we have a choice of a shift followed by an
|
4436 |
|
|
AND or two shifts. Use two shifts for field sizes where the
|
4437 |
|
|
constant might be too large. We assume here that we can
|
4438 |
|
|
always at least get 8-bit constants in an AND insn, which is
|
4439 |
|
|
true for every current RISC. */
|
4440 |
|
|
|
4441 |
|
|
if (unsignedp && len <= 8)
|
4442 |
|
|
{
|
4443 |
|
|
SUBST (SET_SRC (x),
|
4444 |
|
|
gen_rtx_AND (mode,
|
4445 |
|
|
gen_rtx_LSHIFTRT
|
4446 |
|
|
(mode, gen_lowpart (mode, inner),
|
4447 |
|
|
GEN_INT (pos)),
|
4448 |
|
|
GEN_INT (((HOST_WIDE_INT) 1 << len) - 1)));
|
4449 |
|
|
|
4450 |
|
|
split = find_split_point (&SET_SRC (x), insn);
|
4451 |
|
|
if (split && split != &SET_SRC (x))
|
4452 |
|
|
return split;
|
4453 |
|
|
}
|
4454 |
|
|
else
|
4455 |
|
|
{
|
4456 |
|
|
SUBST (SET_SRC (x),
|
4457 |
|
|
gen_rtx_fmt_ee
|
4458 |
|
|
(unsignedp ? LSHIFTRT : ASHIFTRT, mode,
|
4459 |
|
|
gen_rtx_ASHIFT (mode,
|
4460 |
|
|
gen_lowpart (mode, inner),
|
4461 |
|
|
GEN_INT (GET_MODE_BITSIZE (mode)
|
4462 |
|
|
- len - pos)),
|
4463 |
|
|
GEN_INT (GET_MODE_BITSIZE (mode) - len)));
|
4464 |
|
|
|
4465 |
|
|
split = find_split_point (&SET_SRC (x), insn);
|
4466 |
|
|
if (split && split != &SET_SRC (x))
|
4467 |
|
|
return split;
|
4468 |
|
|
}
|
4469 |
|
|
}
|
4470 |
|
|
|
4471 |
|
|
/* See if this is a simple operation with a constant as the second
|
4472 |
|
|
operand. It might be that this constant is out of range and hence
|
4473 |
|
|
could be used as a split point. */
|
4474 |
|
|
if (BINARY_P (SET_SRC (x))
|
4475 |
|
|
&& CONSTANT_P (XEXP (SET_SRC (x), 1))
|
4476 |
|
|
&& (OBJECT_P (XEXP (SET_SRC (x), 0))
|
4477 |
|
|
|| (GET_CODE (XEXP (SET_SRC (x), 0)) == SUBREG
|
4478 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (SET_SRC (x), 0))))))
|
4479 |
|
|
return &XEXP (SET_SRC (x), 1);
|
4480 |
|
|
|
4481 |
|
|
/* Finally, see if this is a simple operation with its first operand
|
4482 |
|
|
not in a register. The operation might require this operand in a
|
4483 |
|
|
register, so return it as a split point. We can always do this
|
4484 |
|
|
because if the first operand were another operation, we would have
|
4485 |
|
|
already found it as a split point. */
|
4486 |
|
|
if ((BINARY_P (SET_SRC (x)) || UNARY_P (SET_SRC (x)))
|
4487 |
|
|
&& ! register_operand (XEXP (SET_SRC (x), 0), VOIDmode))
|
4488 |
|
|
return &XEXP (SET_SRC (x), 0);
|
4489 |
|
|
|
4490 |
|
|
return 0;
|
4491 |
|
|
|
4492 |
|
|
case AND:
|
4493 |
|
|
case IOR:
|
4494 |
|
|
/* We write NOR as (and (not A) (not B)), but if we don't have a NOR,
|
4495 |
|
|
it is better to write this as (not (ior A B)) so we can split it.
|
4496 |
|
|
Similarly for IOR. */
|
4497 |
|
|
if (GET_CODE (XEXP (x, 0)) == NOT && GET_CODE (XEXP (x, 1)) == NOT)
|
4498 |
|
|
{
|
4499 |
|
|
SUBST (*loc,
|
4500 |
|
|
gen_rtx_NOT (GET_MODE (x),
|
4501 |
|
|
gen_rtx_fmt_ee (code == IOR ? AND : IOR,
|
4502 |
|
|
GET_MODE (x),
|
4503 |
|
|
XEXP (XEXP (x, 0), 0),
|
4504 |
|
|
XEXP (XEXP (x, 1), 0))));
|
4505 |
|
|
return find_split_point (loc, insn);
|
4506 |
|
|
}
|
4507 |
|
|
|
4508 |
|
|
/* Many RISC machines have a large set of logical insns. If the
|
4509 |
|
|
second operand is a NOT, put it first so we will try to split the
|
4510 |
|
|
other operand first. */
|
4511 |
|
|
if (GET_CODE (XEXP (x, 1)) == NOT)
|
4512 |
|
|
{
|
4513 |
|
|
rtx tem = XEXP (x, 0);
|
4514 |
|
|
SUBST (XEXP (x, 0), XEXP (x, 1));
|
4515 |
|
|
SUBST (XEXP (x, 1), tem);
|
4516 |
|
|
}
|
4517 |
|
|
break;
|
4518 |
|
|
|
4519 |
|
|
default:
|
4520 |
|
|
break;
|
4521 |
|
|
}
|
4522 |
|
|
|
4523 |
|
|
/* Otherwise, select our actions depending on our rtx class. */
|
4524 |
|
|
switch (GET_RTX_CLASS (code))
|
4525 |
|
|
{
|
4526 |
|
|
case RTX_BITFIELD_OPS: /* This is ZERO_EXTRACT and SIGN_EXTRACT. */
|
4527 |
|
|
case RTX_TERNARY:
|
4528 |
|
|
split = find_split_point (&XEXP (x, 2), insn);
|
4529 |
|
|
if (split)
|
4530 |
|
|
return split;
|
4531 |
|
|
/* ... fall through ... */
|
4532 |
|
|
case RTX_BIN_ARITH:
|
4533 |
|
|
case RTX_COMM_ARITH:
|
4534 |
|
|
case RTX_COMPARE:
|
4535 |
|
|
case RTX_COMM_COMPARE:
|
4536 |
|
|
split = find_split_point (&XEXP (x, 1), insn);
|
4537 |
|
|
if (split)
|
4538 |
|
|
return split;
|
4539 |
|
|
/* ... fall through ... */
|
4540 |
|
|
case RTX_UNARY:
|
4541 |
|
|
/* Some machines have (and (shift ...) ...) insns. If X is not
|
4542 |
|
|
an AND, but XEXP (X, 0) is, use it as our split point. */
|
4543 |
|
|
if (GET_CODE (x) != AND && GET_CODE (XEXP (x, 0)) == AND)
|
4544 |
|
|
return &XEXP (x, 0);
|
4545 |
|
|
|
4546 |
|
|
split = find_split_point (&XEXP (x, 0), insn);
|
4547 |
|
|
if (split)
|
4548 |
|
|
return split;
|
4549 |
|
|
return loc;
|
4550 |
|
|
|
4551 |
|
|
default:
|
4552 |
|
|
/* Otherwise, we don't have a split point. */
|
4553 |
|
|
return 0;
|
4554 |
|
|
}
|
4555 |
|
|
}
|
4556 |
|
|
|
4557 |
|
|
/* Throughout X, replace FROM with TO, and return the result.
|
4558 |
|
|
The result is TO if X is FROM;
|
4559 |
|
|
otherwise the result is X, but its contents may have been modified.
|
4560 |
|
|
If they were modified, a record was made in undobuf so that
|
4561 |
|
|
undo_all will (among other things) return X to its original state.
|
4562 |
|
|
|
4563 |
|
|
If the number of changes necessary is too much to record to undo,
|
4564 |
|
|
the excess changes are not made, so the result is invalid.
|
4565 |
|
|
The changes already made can still be undone.
|
4566 |
|
|
undobuf.num_undo is incremented for such changes, so by testing that
|
4567 |
|
|
the caller can tell whether the result is valid.
|
4568 |
|
|
|
4569 |
|
|
`n_occurrences' is incremented each time FROM is replaced.
|
4570 |
|
|
|
4571 |
|
|
IN_DEST is nonzero if we are processing the SET_DEST of a SET.
|
4572 |
|
|
|
4573 |
|
|
UNIQUE_COPY is nonzero if each substitution must be unique. We do this
|
4574 |
|
|
by copying if `n_occurrences' is nonzero. */
|
4575 |
|
|
|
4576 |
|
|
static rtx
|
4577 |
|
|
subst (rtx x, rtx from, rtx to, int in_dest, int unique_copy)
|
4578 |
|
|
{
|
4579 |
|
|
enum rtx_code code = GET_CODE (x);
|
4580 |
|
|
enum machine_mode op0_mode = VOIDmode;
|
4581 |
|
|
const char *fmt;
|
4582 |
|
|
int len, i;
|
4583 |
|
|
rtx new_rtx;
|
4584 |
|
|
|
4585 |
|
|
/* Two expressions are equal if they are identical copies of a shared
|
4586 |
|
|
RTX or if they are both registers with the same register number
|
4587 |
|
|
and mode. */
|
4588 |
|
|
|
4589 |
|
|
#define COMBINE_RTX_EQUAL_P(X,Y) \
|
4590 |
|
|
((X) == (Y) \
|
4591 |
|
|
|| (REG_P (X) && REG_P (Y) \
|
4592 |
|
|
&& REGNO (X) == REGNO (Y) && GET_MODE (X) == GET_MODE (Y)))
|
4593 |
|
|
|
4594 |
|
|
if (! in_dest && COMBINE_RTX_EQUAL_P (x, from))
|
4595 |
|
|
{
|
4596 |
|
|
n_occurrences++;
|
4597 |
|
|
return (unique_copy && n_occurrences > 1 ? copy_rtx (to) : to);
|
4598 |
|
|
}
|
4599 |
|
|
|
4600 |
|
|
/* If X and FROM are the same register but different modes, they
|
4601 |
|
|
will not have been seen as equal above. However, the log links code
|
4602 |
|
|
will make a LOG_LINKS entry for that case. If we do nothing, we
|
4603 |
|
|
will try to rerecognize our original insn and, when it succeeds,
|
4604 |
|
|
we will delete the feeding insn, which is incorrect.
|
4605 |
|
|
|
4606 |
|
|
So force this insn not to match in this (rare) case. */
|
4607 |
|
|
if (! in_dest && code == REG && REG_P (from)
|
4608 |
|
|
&& reg_overlap_mentioned_p (x, from))
|
4609 |
|
|
return gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
|
4610 |
|
|
|
4611 |
|
|
/* If this is an object, we are done unless it is a MEM or LO_SUM, both
|
4612 |
|
|
of which may contain things that can be combined. */
|
4613 |
|
|
if (code != MEM && code != LO_SUM && OBJECT_P (x))
|
4614 |
|
|
return x;
|
4615 |
|
|
|
4616 |
|
|
/* It is possible to have a subexpression appear twice in the insn.
|
4617 |
|
|
Suppose that FROM is a register that appears within TO.
|
4618 |
|
|
Then, after that subexpression has been scanned once by `subst',
|
4619 |
|
|
the second time it is scanned, TO may be found. If we were
|
4620 |
|
|
to scan TO here, we would find FROM within it and create a
|
4621 |
|
|
self-referent rtl structure which is completely wrong. */
|
4622 |
|
|
if (COMBINE_RTX_EQUAL_P (x, to))
|
4623 |
|
|
return to;
|
4624 |
|
|
|
4625 |
|
|
/* Parallel asm_operands need special attention because all of the
|
4626 |
|
|
inputs are shared across the arms. Furthermore, unsharing the
|
4627 |
|
|
rtl results in recognition failures. Failure to handle this case
|
4628 |
|
|
specially can result in circular rtl.
|
4629 |
|
|
|
4630 |
|
|
Solve this by doing a normal pass across the first entry of the
|
4631 |
|
|
parallel, and only processing the SET_DESTs of the subsequent
|
4632 |
|
|
entries. Ug. */
|
4633 |
|
|
|
4634 |
|
|
if (code == PARALLEL
|
4635 |
|
|
&& GET_CODE (XVECEXP (x, 0, 0)) == SET
|
4636 |
|
|
&& GET_CODE (SET_SRC (XVECEXP (x, 0, 0))) == ASM_OPERANDS)
|
4637 |
|
|
{
|
4638 |
|
|
new_rtx = subst (XVECEXP (x, 0, 0), from, to, 0, unique_copy);
|
4639 |
|
|
|
4640 |
|
|
/* If this substitution failed, this whole thing fails. */
|
4641 |
|
|
if (GET_CODE (new_rtx) == CLOBBER
|
4642 |
|
|
&& XEXP (new_rtx, 0) == const0_rtx)
|
4643 |
|
|
return new_rtx;
|
4644 |
|
|
|
4645 |
|
|
SUBST (XVECEXP (x, 0, 0), new_rtx);
|
4646 |
|
|
|
4647 |
|
|
for (i = XVECLEN (x, 0) - 1; i >= 1; i--)
|
4648 |
|
|
{
|
4649 |
|
|
rtx dest = SET_DEST (XVECEXP (x, 0, i));
|
4650 |
|
|
|
4651 |
|
|
if (!REG_P (dest)
|
4652 |
|
|
&& GET_CODE (dest) != CC0
|
4653 |
|
|
&& GET_CODE (dest) != PC)
|
4654 |
|
|
{
|
4655 |
|
|
new_rtx = subst (dest, from, to, 0, unique_copy);
|
4656 |
|
|
|
4657 |
|
|
/* If this substitution failed, this whole thing fails. */
|
4658 |
|
|
if (GET_CODE (new_rtx) == CLOBBER
|
4659 |
|
|
&& XEXP (new_rtx, 0) == const0_rtx)
|
4660 |
|
|
return new_rtx;
|
4661 |
|
|
|
4662 |
|
|
SUBST (SET_DEST (XVECEXP (x, 0, i)), new_rtx);
|
4663 |
|
|
}
|
4664 |
|
|
}
|
4665 |
|
|
}
|
4666 |
|
|
else
|
4667 |
|
|
{
|
4668 |
|
|
len = GET_RTX_LENGTH (code);
|
4669 |
|
|
fmt = GET_RTX_FORMAT (code);
|
4670 |
|
|
|
4671 |
|
|
/* We don't need to process a SET_DEST that is a register, CC0,
|
4672 |
|
|
or PC, so set up to skip this common case. All other cases
|
4673 |
|
|
where we want to suppress replacing something inside a
|
4674 |
|
|
SET_SRC are handled via the IN_DEST operand. */
|
4675 |
|
|
if (code == SET
|
4676 |
|
|
&& (REG_P (SET_DEST (x))
|
4677 |
|
|
|| GET_CODE (SET_DEST (x)) == CC0
|
4678 |
|
|
|| GET_CODE (SET_DEST (x)) == PC))
|
4679 |
|
|
fmt = "ie";
|
4680 |
|
|
|
4681 |
|
|
/* Get the mode of operand 0 in case X is now a SIGN_EXTEND of a
|
4682 |
|
|
constant. */
|
4683 |
|
|
if (fmt[0] == 'e')
|
4684 |
|
|
op0_mode = GET_MODE (XEXP (x, 0));
|
4685 |
|
|
|
4686 |
|
|
for (i = 0; i < len; i++)
|
4687 |
|
|
{
|
4688 |
|
|
if (fmt[i] == 'E')
|
4689 |
|
|
{
|
4690 |
|
|
int j;
|
4691 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
4692 |
|
|
{
|
4693 |
|
|
if (COMBINE_RTX_EQUAL_P (XVECEXP (x, i, j), from))
|
4694 |
|
|
{
|
4695 |
|
|
new_rtx = (unique_copy && n_occurrences
|
4696 |
|
|
? copy_rtx (to) : to);
|
4697 |
|
|
n_occurrences++;
|
4698 |
|
|
}
|
4699 |
|
|
else
|
4700 |
|
|
{
|
4701 |
|
|
new_rtx = subst (XVECEXP (x, i, j), from, to, 0,
|
4702 |
|
|
unique_copy);
|
4703 |
|
|
|
4704 |
|
|
/* If this substitution failed, this whole thing
|
4705 |
|
|
fails. */
|
4706 |
|
|
if (GET_CODE (new_rtx) == CLOBBER
|
4707 |
|
|
&& XEXP (new_rtx, 0) == const0_rtx)
|
4708 |
|
|
return new_rtx;
|
4709 |
|
|
}
|
4710 |
|
|
|
4711 |
|
|
SUBST (XVECEXP (x, i, j), new_rtx);
|
4712 |
|
|
}
|
4713 |
|
|
}
|
4714 |
|
|
else if (fmt[i] == 'e')
|
4715 |
|
|
{
|
4716 |
|
|
/* If this is a register being set, ignore it. */
|
4717 |
|
|
new_rtx = XEXP (x, i);
|
4718 |
|
|
if (in_dest
|
4719 |
|
|
&& i == 0
|
4720 |
|
|
&& (((code == SUBREG || code == ZERO_EXTRACT)
|
4721 |
|
|
&& REG_P (new_rtx))
|
4722 |
|
|
|| code == STRICT_LOW_PART))
|
4723 |
|
|
;
|
4724 |
|
|
|
4725 |
|
|
else if (COMBINE_RTX_EQUAL_P (XEXP (x, i), from))
|
4726 |
|
|
{
|
4727 |
|
|
/* In general, don't install a subreg involving two
|
4728 |
|
|
modes not tieable. It can worsen register
|
4729 |
|
|
allocation, and can even make invalid reload
|
4730 |
|
|
insns, since the reg inside may need to be copied
|
4731 |
|
|
from in the outside mode, and that may be invalid
|
4732 |
|
|
if it is an fp reg copied in integer mode.
|
4733 |
|
|
|
4734 |
|
|
We allow two exceptions to this: It is valid if
|
4735 |
|
|
it is inside another SUBREG and the mode of that
|
4736 |
|
|
SUBREG and the mode of the inside of TO is
|
4737 |
|
|
tieable and it is valid if X is a SET that copies
|
4738 |
|
|
FROM to CC0. */
|
4739 |
|
|
|
4740 |
|
|
if (GET_CODE (to) == SUBREG
|
4741 |
|
|
&& ! MODES_TIEABLE_P (GET_MODE (to),
|
4742 |
|
|
GET_MODE (SUBREG_REG (to)))
|
4743 |
|
|
&& ! (code == SUBREG
|
4744 |
|
|
&& MODES_TIEABLE_P (GET_MODE (x),
|
4745 |
|
|
GET_MODE (SUBREG_REG (to))))
|
4746 |
|
|
#ifdef HAVE_cc0
|
4747 |
|
|
&& ! (code == SET && i == 1 && XEXP (x, 0) == cc0_rtx)
|
4748 |
|
|
#endif
|
4749 |
|
|
)
|
4750 |
|
|
return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
|
4751 |
|
|
|
4752 |
|
|
#ifdef CANNOT_CHANGE_MODE_CLASS
|
4753 |
|
|
if (code == SUBREG
|
4754 |
|
|
&& REG_P (to)
|
4755 |
|
|
&& REGNO (to) < FIRST_PSEUDO_REGISTER
|
4756 |
|
|
&& REG_CANNOT_CHANGE_MODE_P (REGNO (to),
|
4757 |
|
|
GET_MODE (to),
|
4758 |
|
|
GET_MODE (x)))
|
4759 |
|
|
return gen_rtx_CLOBBER (VOIDmode, const0_rtx);
|
4760 |
|
|
#endif
|
4761 |
|
|
|
4762 |
|
|
new_rtx = (unique_copy && n_occurrences ? copy_rtx (to) : to);
|
4763 |
|
|
n_occurrences++;
|
4764 |
|
|
}
|
4765 |
|
|
else
|
4766 |
|
|
/* If we are in a SET_DEST, suppress most cases unless we
|
4767 |
|
|
have gone inside a MEM, in which case we want to
|
4768 |
|
|
simplify the address. We assume here that things that
|
4769 |
|
|
are actually part of the destination have their inner
|
4770 |
|
|
parts in the first expression. This is true for SUBREG,
|
4771 |
|
|
STRICT_LOW_PART, and ZERO_EXTRACT, which are the only
|
4772 |
|
|
things aside from REG and MEM that should appear in a
|
4773 |
|
|
SET_DEST. */
|
4774 |
|
|
new_rtx = subst (XEXP (x, i), from, to,
|
4775 |
|
|
(((in_dest
|
4776 |
|
|
&& (code == SUBREG || code == STRICT_LOW_PART
|
4777 |
|
|
|| code == ZERO_EXTRACT))
|
4778 |
|
|
|| code == SET)
|
4779 |
|
|
&& i == 0), unique_copy);
|
4780 |
|
|
|
4781 |
|
|
/* If we found that we will have to reject this combination,
|
4782 |
|
|
indicate that by returning the CLOBBER ourselves, rather than
|
4783 |
|
|
an expression containing it. This will speed things up as
|
4784 |
|
|
well as prevent accidents where two CLOBBERs are considered
|
4785 |
|
|
to be equal, thus producing an incorrect simplification. */
|
4786 |
|
|
|
4787 |
|
|
if (GET_CODE (new_rtx) == CLOBBER && XEXP (new_rtx, 0) == const0_rtx)
|
4788 |
|
|
return new_rtx;
|
4789 |
|
|
|
4790 |
|
|
if (GET_CODE (x) == SUBREG
|
4791 |
|
|
&& (CONST_INT_P (new_rtx)
|
4792 |
|
|
|| GET_CODE (new_rtx) == CONST_DOUBLE))
|
4793 |
|
|
{
|
4794 |
|
|
enum machine_mode mode = GET_MODE (x);
|
4795 |
|
|
|
4796 |
|
|
x = simplify_subreg (GET_MODE (x), new_rtx,
|
4797 |
|
|
GET_MODE (SUBREG_REG (x)),
|
4798 |
|
|
SUBREG_BYTE (x));
|
4799 |
|
|
if (! x)
|
4800 |
|
|
x = gen_rtx_CLOBBER (mode, const0_rtx);
|
4801 |
|
|
}
|
4802 |
|
|
else if (CONST_INT_P (new_rtx)
|
4803 |
|
|
&& GET_CODE (x) == ZERO_EXTEND)
|
4804 |
|
|
{
|
4805 |
|
|
x = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
|
4806 |
|
|
new_rtx, GET_MODE (XEXP (x, 0)));
|
4807 |
|
|
gcc_assert (x);
|
4808 |
|
|
}
|
4809 |
|
|
else
|
4810 |
|
|
SUBST (XEXP (x, i), new_rtx);
|
4811 |
|
|
}
|
4812 |
|
|
}
|
4813 |
|
|
}
|
4814 |
|
|
|
4815 |
|
|
/* Check if we are loading something from the constant pool via float
|
4816 |
|
|
extension; in this case we would undo compress_float_constant
|
4817 |
|
|
optimization and degenerate constant load to an immediate value. */
|
4818 |
|
|
if (GET_CODE (x) == FLOAT_EXTEND
|
4819 |
|
|
&& MEM_P (XEXP (x, 0))
|
4820 |
|
|
&& MEM_READONLY_P (XEXP (x, 0)))
|
4821 |
|
|
{
|
4822 |
|
|
rtx tmp = avoid_constant_pool_reference (x);
|
4823 |
|
|
if (x != tmp)
|
4824 |
|
|
return x;
|
4825 |
|
|
}
|
4826 |
|
|
|
4827 |
|
|
/* Try to simplify X. If the simplification changed the code, it is likely
|
4828 |
|
|
that further simplification will help, so loop, but limit the number
|
4829 |
|
|
of repetitions that will be performed. */
|
4830 |
|
|
|
4831 |
|
|
for (i = 0; i < 4; i++)
|
4832 |
|
|
{
|
4833 |
|
|
/* If X is sufficiently simple, don't bother trying to do anything
|
4834 |
|
|
with it. */
|
4835 |
|
|
if (code != CONST_INT && code != REG && code != CLOBBER)
|
4836 |
|
|
x = combine_simplify_rtx (x, op0_mode, in_dest);
|
4837 |
|
|
|
4838 |
|
|
if (GET_CODE (x) == code)
|
4839 |
|
|
break;
|
4840 |
|
|
|
4841 |
|
|
code = GET_CODE (x);
|
4842 |
|
|
|
4843 |
|
|
/* We no longer know the original mode of operand 0 since we
|
4844 |
|
|
have changed the form of X) */
|
4845 |
|
|
op0_mode = VOIDmode;
|
4846 |
|
|
}
|
4847 |
|
|
|
4848 |
|
|
return x;
|
4849 |
|
|
}
|
4850 |
|
|
|
4851 |
|
|
/* Simplify X, a piece of RTL. We just operate on the expression at the
|
4852 |
|
|
outer level; call `subst' to simplify recursively. Return the new
|
4853 |
|
|
expression.
|
4854 |
|
|
|
4855 |
|
|
OP0_MODE is the original mode of XEXP (x, 0). IN_DEST is nonzero
|
4856 |
|
|
if we are inside a SET_DEST. */
|
4857 |
|
|
|
4858 |
|
|
static rtx
|
4859 |
|
|
combine_simplify_rtx (rtx x, enum machine_mode op0_mode, int in_dest)
|
4860 |
|
|
{
|
4861 |
|
|
enum rtx_code code = GET_CODE (x);
|
4862 |
|
|
enum machine_mode mode = GET_MODE (x);
|
4863 |
|
|
rtx temp;
|
4864 |
|
|
int i;
|
4865 |
|
|
|
4866 |
|
|
/* If this is a commutative operation, put a constant last and a complex
|
4867 |
|
|
expression first. We don't need to do this for comparisons here. */
|
4868 |
|
|
if (COMMUTATIVE_ARITH_P (x)
|
4869 |
|
|
&& swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1)))
|
4870 |
|
|
{
|
4871 |
|
|
temp = XEXP (x, 0);
|
4872 |
|
|
SUBST (XEXP (x, 0), XEXP (x, 1));
|
4873 |
|
|
SUBST (XEXP (x, 1), temp);
|
4874 |
|
|
}
|
4875 |
|
|
|
4876 |
|
|
/* If this is a simple operation applied to an IF_THEN_ELSE, try
|
4877 |
|
|
applying it to the arms of the IF_THEN_ELSE. This often simplifies
|
4878 |
|
|
things. Check for cases where both arms are testing the same
|
4879 |
|
|
condition.
|
4880 |
|
|
|
4881 |
|
|
Don't do anything if all operands are very simple. */
|
4882 |
|
|
|
4883 |
|
|
if ((BINARY_P (x)
|
4884 |
|
|
&& ((!OBJECT_P (XEXP (x, 0))
|
4885 |
|
|
&& ! (GET_CODE (XEXP (x, 0)) == SUBREG
|
4886 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (x, 0)))))
|
4887 |
|
|
|| (!OBJECT_P (XEXP (x, 1))
|
4888 |
|
|
&& ! (GET_CODE (XEXP (x, 1)) == SUBREG
|
4889 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (x, 1)))))))
|
4890 |
|
|
|| (UNARY_P (x)
|
4891 |
|
|
&& (!OBJECT_P (XEXP (x, 0))
|
4892 |
|
|
&& ! (GET_CODE (XEXP (x, 0)) == SUBREG
|
4893 |
|
|
&& OBJECT_P (SUBREG_REG (XEXP (x, 0)))))))
|
4894 |
|
|
{
|
4895 |
|
|
rtx cond, true_rtx, false_rtx;
|
4896 |
|
|
|
4897 |
|
|
cond = if_then_else_cond (x, &true_rtx, &false_rtx);
|
4898 |
|
|
if (cond != 0
|
4899 |
|
|
/* If everything is a comparison, what we have is highly unlikely
|
4900 |
|
|
to be simpler, so don't use it. */
|
4901 |
|
|
&& ! (COMPARISON_P (x)
|
4902 |
|
|
&& (COMPARISON_P (true_rtx) || COMPARISON_P (false_rtx))))
|
4903 |
|
|
{
|
4904 |
|
|
rtx cop1 = const0_rtx;
|
4905 |
|
|
enum rtx_code cond_code = simplify_comparison (NE, &cond, &cop1);
|
4906 |
|
|
|
4907 |
|
|
if (cond_code == NE && COMPARISON_P (cond))
|
4908 |
|
|
return x;
|
4909 |
|
|
|
4910 |
|
|
/* Simplify the alternative arms; this may collapse the true and
|
4911 |
|
|
false arms to store-flag values. Be careful to use copy_rtx
|
4912 |
|
|
here since true_rtx or false_rtx might share RTL with x as a
|
4913 |
|
|
result of the if_then_else_cond call above. */
|
4914 |
|
|
true_rtx = subst (copy_rtx (true_rtx), pc_rtx, pc_rtx, 0, 0);
|
4915 |
|
|
false_rtx = subst (copy_rtx (false_rtx), pc_rtx, pc_rtx, 0, 0);
|
4916 |
|
|
|
4917 |
|
|
/* If true_rtx and false_rtx are not general_operands, an if_then_else
|
4918 |
|
|
is unlikely to be simpler. */
|
4919 |
|
|
if (general_operand (true_rtx, VOIDmode)
|
4920 |
|
|
&& general_operand (false_rtx, VOIDmode))
|
4921 |
|
|
{
|
4922 |
|
|
enum rtx_code reversed;
|
4923 |
|
|
|
4924 |
|
|
/* Restarting if we generate a store-flag expression will cause
|
4925 |
|
|
us to loop. Just drop through in this case. */
|
4926 |
|
|
|
4927 |
|
|
/* If the result values are STORE_FLAG_VALUE and zero, we can
|
4928 |
|
|
just make the comparison operation. */
|
4929 |
|
|
if (true_rtx == const_true_rtx && false_rtx == const0_rtx)
|
4930 |
|
|
x = simplify_gen_relational (cond_code, mode, VOIDmode,
|
4931 |
|
|
cond, cop1);
|
4932 |
|
|
else if (true_rtx == const0_rtx && false_rtx == const_true_rtx
|
4933 |
|
|
&& ((reversed = reversed_comparison_code_parts
|
4934 |
|
|
(cond_code, cond, cop1, NULL))
|
4935 |
|
|
!= UNKNOWN))
|
4936 |
|
|
x = simplify_gen_relational (reversed, mode, VOIDmode,
|
4937 |
|
|
cond, cop1);
|
4938 |
|
|
|
4939 |
|
|
/* Likewise, we can make the negate of a comparison operation
|
4940 |
|
|
if the result values are - STORE_FLAG_VALUE and zero. */
|
4941 |
|
|
else if (CONST_INT_P (true_rtx)
|
4942 |
|
|
&& INTVAL (true_rtx) == - STORE_FLAG_VALUE
|
4943 |
|
|
&& false_rtx == const0_rtx)
|
4944 |
|
|
x = simplify_gen_unary (NEG, mode,
|
4945 |
|
|
simplify_gen_relational (cond_code,
|
4946 |
|
|
mode, VOIDmode,
|
4947 |
|
|
cond, cop1),
|
4948 |
|
|
mode);
|
4949 |
|
|
else if (CONST_INT_P (false_rtx)
|
4950 |
|
|
&& INTVAL (false_rtx) == - STORE_FLAG_VALUE
|
4951 |
|
|
&& true_rtx == const0_rtx
|
4952 |
|
|
&& ((reversed = reversed_comparison_code_parts
|
4953 |
|
|
(cond_code, cond, cop1, NULL))
|
4954 |
|
|
!= UNKNOWN))
|
4955 |
|
|
x = simplify_gen_unary (NEG, mode,
|
4956 |
|
|
simplify_gen_relational (reversed,
|
4957 |
|
|
mode, VOIDmode,
|
4958 |
|
|
cond, cop1),
|
4959 |
|
|
mode);
|
4960 |
|
|
else
|
4961 |
|
|
return gen_rtx_IF_THEN_ELSE (mode,
|
4962 |
|
|
simplify_gen_relational (cond_code,
|
4963 |
|
|
mode,
|
4964 |
|
|
VOIDmode,
|
4965 |
|
|
cond,
|
4966 |
|
|
cop1),
|
4967 |
|
|
true_rtx, false_rtx);
|
4968 |
|
|
|
4969 |
|
|
code = GET_CODE (x);
|
4970 |
|
|
op0_mode = VOIDmode;
|
4971 |
|
|
}
|
4972 |
|
|
}
|
4973 |
|
|
}
|
4974 |
|
|
|
4975 |
|
|
/* Try to fold this expression in case we have constants that weren't
|
4976 |
|
|
present before. */
|
4977 |
|
|
temp = 0;
|
4978 |
|
|
switch (GET_RTX_CLASS (code))
|
4979 |
|
|
{
|
4980 |
|
|
case RTX_UNARY:
|
4981 |
|
|
if (op0_mode == VOIDmode)
|
4982 |
|
|
op0_mode = GET_MODE (XEXP (x, 0));
|
4983 |
|
|
temp = simplify_unary_operation (code, mode, XEXP (x, 0), op0_mode);
|
4984 |
|
|
break;
|
4985 |
|
|
case RTX_COMPARE:
|
4986 |
|
|
case RTX_COMM_COMPARE:
|
4987 |
|
|
{
|
4988 |
|
|
enum machine_mode cmp_mode = GET_MODE (XEXP (x, 0));
|
4989 |
|
|
if (cmp_mode == VOIDmode)
|
4990 |
|
|
{
|
4991 |
|
|
cmp_mode = GET_MODE (XEXP (x, 1));
|
4992 |
|
|
if (cmp_mode == VOIDmode)
|
4993 |
|
|
cmp_mode = op0_mode;
|
4994 |
|
|
}
|
4995 |
|
|
temp = simplify_relational_operation (code, mode, cmp_mode,
|
4996 |
|
|
XEXP (x, 0), XEXP (x, 1));
|
4997 |
|
|
}
|
4998 |
|
|
break;
|
4999 |
|
|
case RTX_COMM_ARITH:
|
5000 |
|
|
case RTX_BIN_ARITH:
|
5001 |
|
|
temp = simplify_binary_operation (code, mode, XEXP (x, 0), XEXP (x, 1));
|
5002 |
|
|
break;
|
5003 |
|
|
case RTX_BITFIELD_OPS:
|
5004 |
|
|
case RTX_TERNARY:
|
5005 |
|
|
temp = simplify_ternary_operation (code, mode, op0_mode, XEXP (x, 0),
|
5006 |
|
|
XEXP (x, 1), XEXP (x, 2));
|
5007 |
|
|
break;
|
5008 |
|
|
default:
|
5009 |
|
|
break;
|
5010 |
|
|
}
|
5011 |
|
|
|
5012 |
|
|
if (temp)
|
5013 |
|
|
{
|
5014 |
|
|
x = temp;
|
5015 |
|
|
code = GET_CODE (temp);
|
5016 |
|
|
op0_mode = VOIDmode;
|
5017 |
|
|
mode = GET_MODE (temp);
|
5018 |
|
|
}
|
5019 |
|
|
|
5020 |
|
|
/* First see if we can apply the inverse distributive law. */
|
5021 |
|
|
if (code == PLUS || code == MINUS
|
5022 |
|
|
|| code == AND || code == IOR || code == XOR)
|
5023 |
|
|
{
|
5024 |
|
|
x = apply_distributive_law (x);
|
5025 |
|
|
code = GET_CODE (x);
|
5026 |
|
|
op0_mode = VOIDmode;
|
5027 |
|
|
}
|
5028 |
|
|
|
5029 |
|
|
/* If CODE is an associative operation not otherwise handled, see if we
|
5030 |
|
|
can associate some operands. This can win if they are constants or
|
5031 |
|
|
if they are logically related (i.e. (a & b) & a). */
|
5032 |
|
|
if ((code == PLUS || code == MINUS || code == MULT || code == DIV
|
5033 |
|
|
|| code == AND || code == IOR || code == XOR
|
5034 |
|
|
|| code == SMAX || code == SMIN || code == UMAX || code == UMIN)
|
5035 |
|
|
&& ((INTEGRAL_MODE_P (mode) && code != DIV)
|
5036 |
|
|
|| (flag_associative_math && FLOAT_MODE_P (mode))))
|
5037 |
|
|
{
|
5038 |
|
|
if (GET_CODE (XEXP (x, 0)) == code)
|
5039 |
|
|
{
|
5040 |
|
|
rtx other = XEXP (XEXP (x, 0), 0);
|
5041 |
|
|
rtx inner_op0 = XEXP (XEXP (x, 0), 1);
|
5042 |
|
|
rtx inner_op1 = XEXP (x, 1);
|
5043 |
|
|
rtx inner;
|
5044 |
|
|
|
5045 |
|
|
/* Make sure we pass the constant operand if any as the second
|
5046 |
|
|
one if this is a commutative operation. */
|
5047 |
|
|
if (CONSTANT_P (inner_op0) && COMMUTATIVE_ARITH_P (x))
|
5048 |
|
|
{
|
5049 |
|
|
rtx tem = inner_op0;
|
5050 |
|
|
inner_op0 = inner_op1;
|
5051 |
|
|
inner_op1 = tem;
|
5052 |
|
|
}
|
5053 |
|
|
inner = simplify_binary_operation (code == MINUS ? PLUS
|
5054 |
|
|
: code == DIV ? MULT
|
5055 |
|
|
: code,
|
5056 |
|
|
mode, inner_op0, inner_op1);
|
5057 |
|
|
|
5058 |
|
|
/* For commutative operations, try the other pair if that one
|
5059 |
|
|
didn't simplify. */
|
5060 |
|
|
if (inner == 0 && COMMUTATIVE_ARITH_P (x))
|
5061 |
|
|
{
|
5062 |
|
|
other = XEXP (XEXP (x, 0), 1);
|
5063 |
|
|
inner = simplify_binary_operation (code, mode,
|
5064 |
|
|
XEXP (XEXP (x, 0), 0),
|
5065 |
|
|
XEXP (x, 1));
|
5066 |
|
|
}
|
5067 |
|
|
|
5068 |
|
|
if (inner)
|
5069 |
|
|
return simplify_gen_binary (code, mode, other, inner);
|
5070 |
|
|
}
|
5071 |
|
|
}
|
5072 |
|
|
|
5073 |
|
|
/* A little bit of algebraic simplification here. */
|
5074 |
|
|
switch (code)
|
5075 |
|
|
{
|
5076 |
|
|
case MEM:
|
5077 |
|
|
/* Ensure that our address has any ASHIFTs converted to MULT in case
|
5078 |
|
|
address-recognizing predicates are called later. */
|
5079 |
|
|
temp = make_compound_operation (XEXP (x, 0), MEM);
|
5080 |
|
|
SUBST (XEXP (x, 0), temp);
|
5081 |
|
|
break;
|
5082 |
|
|
|
5083 |
|
|
case SUBREG:
|
5084 |
|
|
if (op0_mode == VOIDmode)
|
5085 |
|
|
op0_mode = GET_MODE (SUBREG_REG (x));
|
5086 |
|
|
|
5087 |
|
|
/* See if this can be moved to simplify_subreg. */
|
5088 |
|
|
if (CONSTANT_P (SUBREG_REG (x))
|
5089 |
|
|
&& subreg_lowpart_offset (mode, op0_mode) == SUBREG_BYTE (x)
|
5090 |
|
|
/* Don't call gen_lowpart if the inner mode
|
5091 |
|
|
is VOIDmode and we cannot simplify it, as SUBREG without
|
5092 |
|
|
inner mode is invalid. */
|
5093 |
|
|
&& (GET_MODE (SUBREG_REG (x)) != VOIDmode
|
5094 |
|
|
|| gen_lowpart_common (mode, SUBREG_REG (x))))
|
5095 |
|
|
return gen_lowpart (mode, SUBREG_REG (x));
|
5096 |
|
|
|
5097 |
|
|
if (GET_MODE_CLASS (GET_MODE (SUBREG_REG (x))) == MODE_CC)
|
5098 |
|
|
break;
|
5099 |
|
|
{
|
5100 |
|
|
rtx temp;
|
5101 |
|
|
temp = simplify_subreg (mode, SUBREG_REG (x), op0_mode,
|
5102 |
|
|
SUBREG_BYTE (x));
|
5103 |
|
|
if (temp)
|
5104 |
|
|
return temp;
|
5105 |
|
|
}
|
5106 |
|
|
|
5107 |
|
|
/* Don't change the mode of the MEM if that would change the meaning
|
5108 |
|
|
of the address. */
|
5109 |
|
|
if (MEM_P (SUBREG_REG (x))
|
5110 |
|
|
&& (MEM_VOLATILE_P (SUBREG_REG (x))
|
5111 |
|
|
|| mode_dependent_address_p (XEXP (SUBREG_REG (x), 0))))
|
5112 |
|
|
return gen_rtx_CLOBBER (mode, const0_rtx);
|
5113 |
|
|
|
5114 |
|
|
/* Note that we cannot do any narrowing for non-constants since
|
5115 |
|
|
we might have been counting on using the fact that some bits were
|
5116 |
|
|
zero. We now do this in the SET. */
|
5117 |
|
|
|
5118 |
|
|
break;
|
5119 |
|
|
|
5120 |
|
|
case NEG:
|
5121 |
|
|
temp = expand_compound_operation (XEXP (x, 0));
|
5122 |
|
|
|
5123 |
|
|
/* For C equal to the width of MODE minus 1, (neg (ashiftrt X C)) can be
|
5124 |
|
|
replaced by (lshiftrt X C). This will convert
|
5125 |
|
|
(neg (sign_extract X 1 Y)) to (zero_extract X 1 Y). */
|
5126 |
|
|
|
5127 |
|
|
if (GET_CODE (temp) == ASHIFTRT
|
5128 |
|
|
&& CONST_INT_P (XEXP (temp, 1))
|
5129 |
|
|
&& INTVAL (XEXP (temp, 1)) == GET_MODE_BITSIZE (mode) - 1)
|
5130 |
|
|
return simplify_shift_const (NULL_RTX, LSHIFTRT, mode, XEXP (temp, 0),
|
5131 |
|
|
INTVAL (XEXP (temp, 1)));
|
5132 |
|
|
|
5133 |
|
|
/* If X has only a single bit that might be nonzero, say, bit I, convert
|
5134 |
|
|
(neg X) to (ashiftrt (ashift X C-I) C-I) where C is the bitsize of
|
5135 |
|
|
MODE minus 1. This will convert (neg (zero_extract X 1 Y)) to
|
5136 |
|
|
(sign_extract X 1 Y). But only do this if TEMP isn't a register
|
5137 |
|
|
or a SUBREG of one since we'd be making the expression more
|
5138 |
|
|
complex if it was just a register. */
|
5139 |
|
|
|
5140 |
|
|
if (!REG_P (temp)
|
5141 |
|
|
&& ! (GET_CODE (temp) == SUBREG
|
5142 |
|
|
&& REG_P (SUBREG_REG (temp)))
|
5143 |
|
|
&& (i = exact_log2 (nonzero_bits (temp, mode))) >= 0)
|
5144 |
|
|
{
|
5145 |
|
|
rtx temp1 = simplify_shift_const
|
5146 |
|
|
(NULL_RTX, ASHIFTRT, mode,
|
5147 |
|
|
simplify_shift_const (NULL_RTX, ASHIFT, mode, temp,
|
5148 |
|
|
GET_MODE_BITSIZE (mode) - 1 - i),
|
5149 |
|
|
GET_MODE_BITSIZE (mode) - 1 - i);
|
5150 |
|
|
|
5151 |
|
|
/* If all we did was surround TEMP with the two shifts, we
|
5152 |
|
|
haven't improved anything, so don't use it. Otherwise,
|
5153 |
|
|
we are better off with TEMP1. */
|
5154 |
|
|
if (GET_CODE (temp1) != ASHIFTRT
|
5155 |
|
|
|| GET_CODE (XEXP (temp1, 0)) != ASHIFT
|
5156 |
|
|
|| XEXP (XEXP (temp1, 0), 0) != temp)
|
5157 |
|
|
return temp1;
|
5158 |
|
|
}
|
5159 |
|
|
break;
|
5160 |
|
|
|
5161 |
|
|
case TRUNCATE:
|
5162 |
|
|
/* We can't handle truncation to a partial integer mode here
|
5163 |
|
|
because we don't know the real bitsize of the partial
|
5164 |
|
|
integer mode. */
|
5165 |
|
|
if (GET_MODE_CLASS (mode) == MODE_PARTIAL_INT)
|
5166 |
|
|
break;
|
5167 |
|
|
|
5168 |
|
|
if (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
5169 |
|
|
SUBST (XEXP (x, 0),
|
5170 |
|
|
force_to_mode (XEXP (x, 0), GET_MODE (XEXP (x, 0)),
|
5171 |
|
|
GET_MODE_MASK (mode), 0));
|
5172 |
|
|
|
5173 |
|
|
/* We can truncate a constant value and return it. */
|
5174 |
|
|
if (CONST_INT_P (XEXP (x, 0)))
|
5175 |
|
|
return gen_int_mode (INTVAL (XEXP (x, 0)), mode);
|
5176 |
|
|
|
5177 |
|
|
/* Similarly to what we do in simplify-rtx.c, a truncate of a register
|
5178 |
|
|
whose value is a comparison can be replaced with a subreg if
|
5179 |
|
|
STORE_FLAG_VALUE permits. */
|
5180 |
|
|
if (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5181 |
|
|
&& ((HOST_WIDE_INT) STORE_FLAG_VALUE & ~GET_MODE_MASK (mode)) == 0
|
5182 |
|
|
&& (temp = get_last_value (XEXP (x, 0)))
|
5183 |
|
|
&& COMPARISON_P (temp))
|
5184 |
|
|
return gen_lowpart (mode, XEXP (x, 0));
|
5185 |
|
|
break;
|
5186 |
|
|
|
5187 |
|
|
case CONST:
|
5188 |
|
|
/* (const (const X)) can become (const X). Do it this way rather than
|
5189 |
|
|
returning the inner CONST since CONST can be shared with a
|
5190 |
|
|
REG_EQUAL note. */
|
5191 |
|
|
if (GET_CODE (XEXP (x, 0)) == CONST)
|
5192 |
|
|
SUBST (XEXP (x, 0), XEXP (XEXP (x, 0), 0));
|
5193 |
|
|
break;
|
5194 |
|
|
|
5195 |
|
|
#ifdef HAVE_lo_sum
|
5196 |
|
|
case LO_SUM:
|
5197 |
|
|
/* Convert (lo_sum (high FOO) FOO) to FOO. This is necessary so we
|
5198 |
|
|
can add in an offset. find_split_point will split this address up
|
5199 |
|
|
again if it doesn't match. */
|
5200 |
|
|
if (GET_CODE (XEXP (x, 0)) == HIGH
|
5201 |
|
|
&& rtx_equal_p (XEXP (XEXP (x, 0), 0), XEXP (x, 1)))
|
5202 |
|
|
return XEXP (x, 1);
|
5203 |
|
|
break;
|
5204 |
|
|
#endif
|
5205 |
|
|
|
5206 |
|
|
case PLUS:
|
5207 |
|
|
/* (plus (xor (and <foo> (const_int pow2 - 1)) <c>) <-c>)
|
5208 |
|
|
when c is (const_int (pow2 + 1) / 2) is a sign extension of a
|
5209 |
|
|
bit-field and can be replaced by either a sign_extend or a
|
5210 |
|
|
sign_extract. The `and' may be a zero_extend and the two
|
5211 |
|
|
<c>, -<c> constants may be reversed. */
|
5212 |
|
|
if (GET_CODE (XEXP (x, 0)) == XOR
|
5213 |
|
|
&& CONST_INT_P (XEXP (x, 1))
|
5214 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
5215 |
|
|
&& INTVAL (XEXP (x, 1)) == -INTVAL (XEXP (XEXP (x, 0), 1))
|
5216 |
|
|
&& ((i = exact_log2 (INTVAL (XEXP (XEXP (x, 0), 1)))) >= 0
|
5217 |
|
|
|| (i = exact_log2 (INTVAL (XEXP (x, 1)))) >= 0)
|
5218 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5219 |
|
|
&& ((GET_CODE (XEXP (XEXP (x, 0), 0)) == AND
|
5220 |
|
|
&& CONST_INT_P (XEXP (XEXP (XEXP (x, 0), 0), 1))
|
5221 |
|
|
&& (INTVAL (XEXP (XEXP (XEXP (x, 0), 0), 1))
|
5222 |
|
|
== ((HOST_WIDE_INT) 1 << (i + 1)) - 1))
|
5223 |
|
|
|| (GET_CODE (XEXP (XEXP (x, 0), 0)) == ZERO_EXTEND
|
5224 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (XEXP (XEXP (XEXP (x, 0), 0), 0)))
|
5225 |
|
|
== (unsigned int) i + 1))))
|
5226 |
|
|
return simplify_shift_const
|
5227 |
|
|
(NULL_RTX, ASHIFTRT, mode,
|
5228 |
|
|
simplify_shift_const (NULL_RTX, ASHIFT, mode,
|
5229 |
|
|
XEXP (XEXP (XEXP (x, 0), 0), 0),
|
5230 |
|
|
GET_MODE_BITSIZE (mode) - (i + 1)),
|
5231 |
|
|
GET_MODE_BITSIZE (mode) - (i + 1));
|
5232 |
|
|
|
5233 |
|
|
/* If only the low-order bit of X is possibly nonzero, (plus x -1)
|
5234 |
|
|
can become (ashiftrt (ashift (xor x 1) C) C) where C is
|
5235 |
|
|
the bitsize of the mode - 1. This allows simplification of
|
5236 |
|
|
"a = (b & 8) == 0;" */
|
5237 |
|
|
if (XEXP (x, 1) == constm1_rtx
|
5238 |
|
|
&& !REG_P (XEXP (x, 0))
|
5239 |
|
|
&& ! (GET_CODE (XEXP (x, 0)) == SUBREG
|
5240 |
|
|
&& REG_P (SUBREG_REG (XEXP (x, 0))))
|
5241 |
|
|
&& nonzero_bits (XEXP (x, 0), mode) == 1)
|
5242 |
|
|
return simplify_shift_const (NULL_RTX, ASHIFTRT, mode,
|
5243 |
|
|
simplify_shift_const (NULL_RTX, ASHIFT, mode,
|
5244 |
|
|
gen_rtx_XOR (mode, XEXP (x, 0), const1_rtx),
|
5245 |
|
|
GET_MODE_BITSIZE (mode) - 1),
|
5246 |
|
|
GET_MODE_BITSIZE (mode) - 1);
|
5247 |
|
|
|
5248 |
|
|
/* If we are adding two things that have no bits in common, convert
|
5249 |
|
|
the addition into an IOR. This will often be further simplified,
|
5250 |
|
|
for example in cases like ((a & 1) + (a & 2)), which can
|
5251 |
|
|
become a & 3. */
|
5252 |
|
|
|
5253 |
|
|
if (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5254 |
|
|
&& (nonzero_bits (XEXP (x, 0), mode)
|
5255 |
|
|
& nonzero_bits (XEXP (x, 1), mode)) == 0)
|
5256 |
|
|
{
|
5257 |
|
|
/* Try to simplify the expression further. */
|
5258 |
|
|
rtx tor = simplify_gen_binary (IOR, mode, XEXP (x, 0), XEXP (x, 1));
|
5259 |
|
|
temp = combine_simplify_rtx (tor, mode, in_dest);
|
5260 |
|
|
|
5261 |
|
|
/* If we could, great. If not, do not go ahead with the IOR
|
5262 |
|
|
replacement, since PLUS appears in many special purpose
|
5263 |
|
|
address arithmetic instructions. */
|
5264 |
|
|
if (GET_CODE (temp) != CLOBBER && temp != tor)
|
5265 |
|
|
return temp;
|
5266 |
|
|
}
|
5267 |
|
|
break;
|
5268 |
|
|
|
5269 |
|
|
case MINUS:
|
5270 |
|
|
/* (minus <foo> (and <foo> (const_int -pow2))) becomes
|
5271 |
|
|
(and <foo> (const_int pow2-1)) */
|
5272 |
|
|
if (GET_CODE (XEXP (x, 1)) == AND
|
5273 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 1), 1))
|
5274 |
|
|
&& exact_log2 (-INTVAL (XEXP (XEXP (x, 1), 1))) >= 0
|
5275 |
|
|
&& rtx_equal_p (XEXP (XEXP (x, 1), 0), XEXP (x, 0)))
|
5276 |
|
|
return simplify_and_const_int (NULL_RTX, mode, XEXP (x, 0),
|
5277 |
|
|
-INTVAL (XEXP (XEXP (x, 1), 1)) - 1);
|
5278 |
|
|
break;
|
5279 |
|
|
|
5280 |
|
|
case MULT:
|
5281 |
|
|
/* If we have (mult (plus A B) C), apply the distributive law and then
|
5282 |
|
|
the inverse distributive law to see if things simplify. This
|
5283 |
|
|
occurs mostly in addresses, often when unrolling loops. */
|
5284 |
|
|
|
5285 |
|
|
if (GET_CODE (XEXP (x, 0)) == PLUS)
|
5286 |
|
|
{
|
5287 |
|
|
rtx result = distribute_and_simplify_rtx (x, 0);
|
5288 |
|
|
if (result)
|
5289 |
|
|
return result;
|
5290 |
|
|
}
|
5291 |
|
|
|
5292 |
|
|
/* Try simplify a*(b/c) as (a*b)/c. */
|
5293 |
|
|
if (FLOAT_MODE_P (mode) && flag_associative_math
|
5294 |
|
|
&& GET_CODE (XEXP (x, 0)) == DIV)
|
5295 |
|
|
{
|
5296 |
|
|
rtx tem = simplify_binary_operation (MULT, mode,
|
5297 |
|
|
XEXP (XEXP (x, 0), 0),
|
5298 |
|
|
XEXP (x, 1));
|
5299 |
|
|
if (tem)
|
5300 |
|
|
return simplify_gen_binary (DIV, mode, tem, XEXP (XEXP (x, 0), 1));
|
5301 |
|
|
}
|
5302 |
|
|
break;
|
5303 |
|
|
|
5304 |
|
|
case UDIV:
|
5305 |
|
|
/* If this is a divide by a power of two, treat it as a shift if
|
5306 |
|
|
its first operand is a shift. */
|
5307 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
5308 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)))) >= 0
|
5309 |
|
|
&& (GET_CODE (XEXP (x, 0)) == ASHIFT
|
5310 |
|
|
|| GET_CODE (XEXP (x, 0)) == LSHIFTRT
|
5311 |
|
|
|| GET_CODE (XEXP (x, 0)) == ASHIFTRT
|
5312 |
|
|
|| GET_CODE (XEXP (x, 0)) == ROTATE
|
5313 |
|
|
|| GET_CODE (XEXP (x, 0)) == ROTATERT))
|
5314 |
|
|
return simplify_shift_const (NULL_RTX, LSHIFTRT, mode, XEXP (x, 0), i);
|
5315 |
|
|
break;
|
5316 |
|
|
|
5317 |
|
|
case EQ: case NE:
|
5318 |
|
|
case GT: case GTU: case GE: case GEU:
|
5319 |
|
|
case LT: case LTU: case LE: case LEU:
|
5320 |
|
|
case UNEQ: case LTGT:
|
5321 |
|
|
case UNGT: case UNGE:
|
5322 |
|
|
case UNLT: case UNLE:
|
5323 |
|
|
case UNORDERED: case ORDERED:
|
5324 |
|
|
/* If the first operand is a condition code, we can't do anything
|
5325 |
|
|
with it. */
|
5326 |
|
|
if (GET_CODE (XEXP (x, 0)) == COMPARE
|
5327 |
|
|
|| (GET_MODE_CLASS (GET_MODE (XEXP (x, 0))) != MODE_CC
|
5328 |
|
|
&& ! CC0_P (XEXP (x, 0))))
|
5329 |
|
|
{
|
5330 |
|
|
rtx op0 = XEXP (x, 0);
|
5331 |
|
|
rtx op1 = XEXP (x, 1);
|
5332 |
|
|
enum rtx_code new_code;
|
5333 |
|
|
|
5334 |
|
|
if (GET_CODE (op0) == COMPARE)
|
5335 |
|
|
op1 = XEXP (op0, 1), op0 = XEXP (op0, 0);
|
5336 |
|
|
|
5337 |
|
|
/* Simplify our comparison, if possible. */
|
5338 |
|
|
new_code = simplify_comparison (code, &op0, &op1);
|
5339 |
|
|
|
5340 |
|
|
/* If STORE_FLAG_VALUE is 1, we can convert (ne x 0) to simply X
|
5341 |
|
|
if only the low-order bit is possibly nonzero in X (such as when
|
5342 |
|
|
X is a ZERO_EXTRACT of one bit). Similarly, we can convert EQ to
|
5343 |
|
|
(xor X 1) or (minus 1 X); we use the former. Finally, if X is
|
5344 |
|
|
known to be either 0 or -1, NE becomes a NEG and EQ becomes
|
5345 |
|
|
(plus X 1).
|
5346 |
|
|
|
5347 |
|
|
Remove any ZERO_EXTRACT we made when thinking this was a
|
5348 |
|
|
comparison. It may now be simpler to use, e.g., an AND. If a
|
5349 |
|
|
ZERO_EXTRACT is indeed appropriate, it will be placed back by
|
5350 |
|
|
the call to make_compound_operation in the SET case. */
|
5351 |
|
|
|
5352 |
|
|
if (STORE_FLAG_VALUE == 1
|
5353 |
|
|
&& new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
|
5354 |
|
|
&& op1 == const0_rtx
|
5355 |
|
|
&& mode == GET_MODE (op0)
|
5356 |
|
|
&& nonzero_bits (op0, mode) == 1)
|
5357 |
|
|
return gen_lowpart (mode,
|
5358 |
|
|
expand_compound_operation (op0));
|
5359 |
|
|
|
5360 |
|
|
else if (STORE_FLAG_VALUE == 1
|
5361 |
|
|
&& new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
|
5362 |
|
|
&& op1 == const0_rtx
|
5363 |
|
|
&& mode == GET_MODE (op0)
|
5364 |
|
|
&& (num_sign_bit_copies (op0, mode)
|
5365 |
|
|
== GET_MODE_BITSIZE (mode)))
|
5366 |
|
|
{
|
5367 |
|
|
op0 = expand_compound_operation (op0);
|
5368 |
|
|
return simplify_gen_unary (NEG, mode,
|
5369 |
|
|
gen_lowpart (mode, op0),
|
5370 |
|
|
mode);
|
5371 |
|
|
}
|
5372 |
|
|
|
5373 |
|
|
else if (STORE_FLAG_VALUE == 1
|
5374 |
|
|
&& new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
|
5375 |
|
|
&& op1 == const0_rtx
|
5376 |
|
|
&& mode == GET_MODE (op0)
|
5377 |
|
|
&& nonzero_bits (op0, mode) == 1)
|
5378 |
|
|
{
|
5379 |
|
|
op0 = expand_compound_operation (op0);
|
5380 |
|
|
return simplify_gen_binary (XOR, mode,
|
5381 |
|
|
gen_lowpart (mode, op0),
|
5382 |
|
|
const1_rtx);
|
5383 |
|
|
}
|
5384 |
|
|
|
5385 |
|
|
else if (STORE_FLAG_VALUE == 1
|
5386 |
|
|
&& new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
|
5387 |
|
|
&& op1 == const0_rtx
|
5388 |
|
|
&& mode == GET_MODE (op0)
|
5389 |
|
|
&& (num_sign_bit_copies (op0, mode)
|
5390 |
|
|
== GET_MODE_BITSIZE (mode)))
|
5391 |
|
|
{
|
5392 |
|
|
op0 = expand_compound_operation (op0);
|
5393 |
|
|
return plus_constant (gen_lowpart (mode, op0), 1);
|
5394 |
|
|
}
|
5395 |
|
|
|
5396 |
|
|
/* If STORE_FLAG_VALUE is -1, we have cases similar to
|
5397 |
|
|
those above. */
|
5398 |
|
|
if (STORE_FLAG_VALUE == -1
|
5399 |
|
|
&& new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
|
5400 |
|
|
&& op1 == const0_rtx
|
5401 |
|
|
&& (num_sign_bit_copies (op0, mode)
|
5402 |
|
|
== GET_MODE_BITSIZE (mode)))
|
5403 |
|
|
return gen_lowpart (mode,
|
5404 |
|
|
expand_compound_operation (op0));
|
5405 |
|
|
|
5406 |
|
|
else if (STORE_FLAG_VALUE == -1
|
5407 |
|
|
&& new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
|
5408 |
|
|
&& op1 == const0_rtx
|
5409 |
|
|
&& mode == GET_MODE (op0)
|
5410 |
|
|
&& nonzero_bits (op0, mode) == 1)
|
5411 |
|
|
{
|
5412 |
|
|
op0 = expand_compound_operation (op0);
|
5413 |
|
|
return simplify_gen_unary (NEG, mode,
|
5414 |
|
|
gen_lowpart (mode, op0),
|
5415 |
|
|
mode);
|
5416 |
|
|
}
|
5417 |
|
|
|
5418 |
|
|
else if (STORE_FLAG_VALUE == -1
|
5419 |
|
|
&& new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
|
5420 |
|
|
&& op1 == const0_rtx
|
5421 |
|
|
&& mode == GET_MODE (op0)
|
5422 |
|
|
&& (num_sign_bit_copies (op0, mode)
|
5423 |
|
|
== GET_MODE_BITSIZE (mode)))
|
5424 |
|
|
{
|
5425 |
|
|
op0 = expand_compound_operation (op0);
|
5426 |
|
|
return simplify_gen_unary (NOT, mode,
|
5427 |
|
|
gen_lowpart (mode, op0),
|
5428 |
|
|
mode);
|
5429 |
|
|
}
|
5430 |
|
|
|
5431 |
|
|
/* If X is 0/1, (eq X 0) is X-1. */
|
5432 |
|
|
else if (STORE_FLAG_VALUE == -1
|
5433 |
|
|
&& new_code == EQ && GET_MODE_CLASS (mode) == MODE_INT
|
5434 |
|
|
&& op1 == const0_rtx
|
5435 |
|
|
&& mode == GET_MODE (op0)
|
5436 |
|
|
&& nonzero_bits (op0, mode) == 1)
|
5437 |
|
|
{
|
5438 |
|
|
op0 = expand_compound_operation (op0);
|
5439 |
|
|
return plus_constant (gen_lowpart (mode, op0), -1);
|
5440 |
|
|
}
|
5441 |
|
|
|
5442 |
|
|
/* If STORE_FLAG_VALUE says to just test the sign bit and X has just
|
5443 |
|
|
one bit that might be nonzero, we can convert (ne x 0) to
|
5444 |
|
|
(ashift x c) where C puts the bit in the sign bit. Remove any
|
5445 |
|
|
AND with STORE_FLAG_VALUE when we are done, since we are only
|
5446 |
|
|
going to test the sign bit. */
|
5447 |
|
|
if (new_code == NE && GET_MODE_CLASS (mode) == MODE_INT
|
5448 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5449 |
|
|
&& ((STORE_FLAG_VALUE & GET_MODE_MASK (mode))
|
5450 |
|
|
== (unsigned HOST_WIDE_INT) 1 << (GET_MODE_BITSIZE (mode) - 1))
|
5451 |
|
|
&& op1 == const0_rtx
|
5452 |
|
|
&& mode == GET_MODE (op0)
|
5453 |
|
|
&& (i = exact_log2 (nonzero_bits (op0, mode))) >= 0)
|
5454 |
|
|
{
|
5455 |
|
|
x = simplify_shift_const (NULL_RTX, ASHIFT, mode,
|
5456 |
|
|
expand_compound_operation (op0),
|
5457 |
|
|
GET_MODE_BITSIZE (mode) - 1 - i);
|
5458 |
|
|
if (GET_CODE (x) == AND && XEXP (x, 1) == const_true_rtx)
|
5459 |
|
|
return XEXP (x, 0);
|
5460 |
|
|
else
|
5461 |
|
|
return x;
|
5462 |
|
|
}
|
5463 |
|
|
|
5464 |
|
|
/* If the code changed, return a whole new comparison. */
|
5465 |
|
|
if (new_code != code)
|
5466 |
|
|
return gen_rtx_fmt_ee (new_code, mode, op0, op1);
|
5467 |
|
|
|
5468 |
|
|
/* Otherwise, keep this operation, but maybe change its operands.
|
5469 |
|
|
This also converts (ne (compare FOO BAR) 0) to (ne FOO BAR). */
|
5470 |
|
|
SUBST (XEXP (x, 0), op0);
|
5471 |
|
|
SUBST (XEXP (x, 1), op1);
|
5472 |
|
|
}
|
5473 |
|
|
break;
|
5474 |
|
|
|
5475 |
|
|
case IF_THEN_ELSE:
|
5476 |
|
|
return simplify_if_then_else (x);
|
5477 |
|
|
|
5478 |
|
|
case ZERO_EXTRACT:
|
5479 |
|
|
case SIGN_EXTRACT:
|
5480 |
|
|
case ZERO_EXTEND:
|
5481 |
|
|
case SIGN_EXTEND:
|
5482 |
|
|
/* If we are processing SET_DEST, we are done. */
|
5483 |
|
|
if (in_dest)
|
5484 |
|
|
return x;
|
5485 |
|
|
|
5486 |
|
|
return expand_compound_operation (x);
|
5487 |
|
|
|
5488 |
|
|
case SET:
|
5489 |
|
|
return simplify_set (x);
|
5490 |
|
|
|
5491 |
|
|
case AND:
|
5492 |
|
|
case IOR:
|
5493 |
|
|
return simplify_logical (x);
|
5494 |
|
|
|
5495 |
|
|
case ASHIFT:
|
5496 |
|
|
case LSHIFTRT:
|
5497 |
|
|
case ASHIFTRT:
|
5498 |
|
|
case ROTATE:
|
5499 |
|
|
case ROTATERT:
|
5500 |
|
|
/* If this is a shift by a constant amount, simplify it. */
|
5501 |
|
|
if (CONST_INT_P (XEXP (x, 1)))
|
5502 |
|
|
return simplify_shift_const (x, code, mode, XEXP (x, 0),
|
5503 |
|
|
INTVAL (XEXP (x, 1)));
|
5504 |
|
|
|
5505 |
|
|
else if (SHIFT_COUNT_TRUNCATED && !REG_P (XEXP (x, 1)))
|
5506 |
|
|
SUBST (XEXP (x, 1),
|
5507 |
|
|
force_to_mode (XEXP (x, 1), GET_MODE (XEXP (x, 1)),
|
5508 |
|
|
((HOST_WIDE_INT) 1
|
5509 |
|
|
<< exact_log2 (GET_MODE_BITSIZE (GET_MODE (x))))
|
5510 |
|
|
- 1,
|
5511 |
|
|
0));
|
5512 |
|
|
break;
|
5513 |
|
|
|
5514 |
|
|
default:
|
5515 |
|
|
break;
|
5516 |
|
|
}
|
5517 |
|
|
|
5518 |
|
|
return x;
|
5519 |
|
|
}
|
5520 |
|
|
|
5521 |
|
|
/* Simplify X, an IF_THEN_ELSE expression. Return the new expression. */
|
5522 |
|
|
|
5523 |
|
|
static rtx
|
5524 |
|
|
simplify_if_then_else (rtx x)
|
5525 |
|
|
{
|
5526 |
|
|
enum machine_mode mode = GET_MODE (x);
|
5527 |
|
|
rtx cond = XEXP (x, 0);
|
5528 |
|
|
rtx true_rtx = XEXP (x, 1);
|
5529 |
|
|
rtx false_rtx = XEXP (x, 2);
|
5530 |
|
|
enum rtx_code true_code = GET_CODE (cond);
|
5531 |
|
|
int comparison_p = COMPARISON_P (cond);
|
5532 |
|
|
rtx temp;
|
5533 |
|
|
int i;
|
5534 |
|
|
enum rtx_code false_code;
|
5535 |
|
|
rtx reversed;
|
5536 |
|
|
|
5537 |
|
|
/* Simplify storing of the truth value. */
|
5538 |
|
|
if (comparison_p && true_rtx == const_true_rtx && false_rtx == const0_rtx)
|
5539 |
|
|
return simplify_gen_relational (true_code, mode, VOIDmode,
|
5540 |
|
|
XEXP (cond, 0), XEXP (cond, 1));
|
5541 |
|
|
|
5542 |
|
|
/* Also when the truth value has to be reversed. */
|
5543 |
|
|
if (comparison_p
|
5544 |
|
|
&& true_rtx == const0_rtx && false_rtx == const_true_rtx
|
5545 |
|
|
&& (reversed = reversed_comparison (cond, mode)))
|
5546 |
|
|
return reversed;
|
5547 |
|
|
|
5548 |
|
|
/* Sometimes we can simplify the arm of an IF_THEN_ELSE if a register used
|
5549 |
|
|
in it is being compared against certain values. Get the true and false
|
5550 |
|
|
comparisons and see if that says anything about the value of each arm. */
|
5551 |
|
|
|
5552 |
|
|
if (comparison_p
|
5553 |
|
|
&& ((false_code = reversed_comparison_code (cond, NULL))
|
5554 |
|
|
!= UNKNOWN)
|
5555 |
|
|
&& REG_P (XEXP (cond, 0)))
|
5556 |
|
|
{
|
5557 |
|
|
HOST_WIDE_INT nzb;
|
5558 |
|
|
rtx from = XEXP (cond, 0);
|
5559 |
|
|
rtx true_val = XEXP (cond, 1);
|
5560 |
|
|
rtx false_val = true_val;
|
5561 |
|
|
int swapped = 0;
|
5562 |
|
|
|
5563 |
|
|
/* If FALSE_CODE is EQ, swap the codes and arms. */
|
5564 |
|
|
|
5565 |
|
|
if (false_code == EQ)
|
5566 |
|
|
{
|
5567 |
|
|
swapped = 1, true_code = EQ, false_code = NE;
|
5568 |
|
|
temp = true_rtx, true_rtx = false_rtx, false_rtx = temp;
|
5569 |
|
|
}
|
5570 |
|
|
|
5571 |
|
|
/* If we are comparing against zero and the expression being tested has
|
5572 |
|
|
only a single bit that might be nonzero, that is its value when it is
|
5573 |
|
|
not equal to zero. Similarly if it is known to be -1 or 0. */
|
5574 |
|
|
|
5575 |
|
|
if (true_code == EQ && true_val == const0_rtx
|
5576 |
|
|
&& exact_log2 (nzb = nonzero_bits (from, GET_MODE (from))) >= 0)
|
5577 |
|
|
{
|
5578 |
|
|
false_code = EQ;
|
5579 |
|
|
false_val = GEN_INT (trunc_int_for_mode (nzb, GET_MODE (from)));
|
5580 |
|
|
}
|
5581 |
|
|
else if (true_code == EQ && true_val == const0_rtx
|
5582 |
|
|
&& (num_sign_bit_copies (from, GET_MODE (from))
|
5583 |
|
|
== GET_MODE_BITSIZE (GET_MODE (from))))
|
5584 |
|
|
{
|
5585 |
|
|
false_code = EQ;
|
5586 |
|
|
false_val = constm1_rtx;
|
5587 |
|
|
}
|
5588 |
|
|
|
5589 |
|
|
/* Now simplify an arm if we know the value of the register in the
|
5590 |
|
|
branch and it is used in the arm. Be careful due to the potential
|
5591 |
|
|
of locally-shared RTL. */
|
5592 |
|
|
|
5593 |
|
|
if (reg_mentioned_p (from, true_rtx))
|
5594 |
|
|
true_rtx = subst (known_cond (copy_rtx (true_rtx), true_code,
|
5595 |
|
|
from, true_val),
|
5596 |
|
|
pc_rtx, pc_rtx, 0, 0);
|
5597 |
|
|
if (reg_mentioned_p (from, false_rtx))
|
5598 |
|
|
false_rtx = subst (known_cond (copy_rtx (false_rtx), false_code,
|
5599 |
|
|
from, false_val),
|
5600 |
|
|
pc_rtx, pc_rtx, 0, 0);
|
5601 |
|
|
|
5602 |
|
|
SUBST (XEXP (x, 1), swapped ? false_rtx : true_rtx);
|
5603 |
|
|
SUBST (XEXP (x, 2), swapped ? true_rtx : false_rtx);
|
5604 |
|
|
|
5605 |
|
|
true_rtx = XEXP (x, 1);
|
5606 |
|
|
false_rtx = XEXP (x, 2);
|
5607 |
|
|
true_code = GET_CODE (cond);
|
5608 |
|
|
}
|
5609 |
|
|
|
5610 |
|
|
/* If we have (if_then_else FOO (pc) (label_ref BAR)) and FOO can be
|
5611 |
|
|
reversed, do so to avoid needing two sets of patterns for
|
5612 |
|
|
subtract-and-branch insns. Similarly if we have a constant in the true
|
5613 |
|
|
arm, the false arm is the same as the first operand of the comparison, or
|
5614 |
|
|
the false arm is more complicated than the true arm. */
|
5615 |
|
|
|
5616 |
|
|
if (comparison_p
|
5617 |
|
|
&& reversed_comparison_code (cond, NULL) != UNKNOWN
|
5618 |
|
|
&& (true_rtx == pc_rtx
|
5619 |
|
|
|| (CONSTANT_P (true_rtx)
|
5620 |
|
|
&& !CONST_INT_P (false_rtx) && false_rtx != pc_rtx)
|
5621 |
|
|
|| true_rtx == const0_rtx
|
5622 |
|
|
|| (OBJECT_P (true_rtx) && !OBJECT_P (false_rtx))
|
5623 |
|
|
|| (GET_CODE (true_rtx) == SUBREG && OBJECT_P (SUBREG_REG (true_rtx))
|
5624 |
|
|
&& !OBJECT_P (false_rtx))
|
5625 |
|
|
|| reg_mentioned_p (true_rtx, false_rtx)
|
5626 |
|
|
|| rtx_equal_p (false_rtx, XEXP (cond, 0))))
|
5627 |
|
|
{
|
5628 |
|
|
true_code = reversed_comparison_code (cond, NULL);
|
5629 |
|
|
SUBST (XEXP (x, 0), reversed_comparison (cond, GET_MODE (cond)));
|
5630 |
|
|
SUBST (XEXP (x, 1), false_rtx);
|
5631 |
|
|
SUBST (XEXP (x, 2), true_rtx);
|
5632 |
|
|
|
5633 |
|
|
temp = true_rtx, true_rtx = false_rtx, false_rtx = temp;
|
5634 |
|
|
cond = XEXP (x, 0);
|
5635 |
|
|
|
5636 |
|
|
/* It is possible that the conditional has been simplified out. */
|
5637 |
|
|
true_code = GET_CODE (cond);
|
5638 |
|
|
comparison_p = COMPARISON_P (cond);
|
5639 |
|
|
}
|
5640 |
|
|
|
5641 |
|
|
/* If the two arms are identical, we don't need the comparison. */
|
5642 |
|
|
|
5643 |
|
|
if (rtx_equal_p (true_rtx, false_rtx) && ! side_effects_p (cond))
|
5644 |
|
|
return true_rtx;
|
5645 |
|
|
|
5646 |
|
|
/* Convert a == b ? b : a to "a". */
|
5647 |
|
|
if (true_code == EQ && ! side_effects_p (cond)
|
5648 |
|
|
&& !HONOR_NANS (mode)
|
5649 |
|
|
&& rtx_equal_p (XEXP (cond, 0), false_rtx)
|
5650 |
|
|
&& rtx_equal_p (XEXP (cond, 1), true_rtx))
|
5651 |
|
|
return false_rtx;
|
5652 |
|
|
else if (true_code == NE && ! side_effects_p (cond)
|
5653 |
|
|
&& !HONOR_NANS (mode)
|
5654 |
|
|
&& rtx_equal_p (XEXP (cond, 0), true_rtx)
|
5655 |
|
|
&& rtx_equal_p (XEXP (cond, 1), false_rtx))
|
5656 |
|
|
return true_rtx;
|
5657 |
|
|
|
5658 |
|
|
/* Look for cases where we have (abs x) or (neg (abs X)). */
|
5659 |
|
|
|
5660 |
|
|
if (GET_MODE_CLASS (mode) == MODE_INT
|
5661 |
|
|
&& comparison_p
|
5662 |
|
|
&& XEXP (cond, 1) == const0_rtx
|
5663 |
|
|
&& GET_CODE (false_rtx) == NEG
|
5664 |
|
|
&& rtx_equal_p (true_rtx, XEXP (false_rtx, 0))
|
5665 |
|
|
&& rtx_equal_p (true_rtx, XEXP (cond, 0))
|
5666 |
|
|
&& ! side_effects_p (true_rtx))
|
5667 |
|
|
switch (true_code)
|
5668 |
|
|
{
|
5669 |
|
|
case GT:
|
5670 |
|
|
case GE:
|
5671 |
|
|
return simplify_gen_unary (ABS, mode, true_rtx, mode);
|
5672 |
|
|
case LT:
|
5673 |
|
|
case LE:
|
5674 |
|
|
return
|
5675 |
|
|
simplify_gen_unary (NEG, mode,
|
5676 |
|
|
simplify_gen_unary (ABS, mode, true_rtx, mode),
|
5677 |
|
|
mode);
|
5678 |
|
|
default:
|
5679 |
|
|
break;
|
5680 |
|
|
}
|
5681 |
|
|
|
5682 |
|
|
/* Look for MIN or MAX. */
|
5683 |
|
|
|
5684 |
|
|
if ((! FLOAT_MODE_P (mode) || flag_unsafe_math_optimizations)
|
5685 |
|
|
&& comparison_p
|
5686 |
|
|
&& rtx_equal_p (XEXP (cond, 0), true_rtx)
|
5687 |
|
|
&& rtx_equal_p (XEXP (cond, 1), false_rtx)
|
5688 |
|
|
&& ! side_effects_p (cond))
|
5689 |
|
|
switch (true_code)
|
5690 |
|
|
{
|
5691 |
|
|
case GE:
|
5692 |
|
|
case GT:
|
5693 |
|
|
return simplify_gen_binary (SMAX, mode, true_rtx, false_rtx);
|
5694 |
|
|
case LE:
|
5695 |
|
|
case LT:
|
5696 |
|
|
return simplify_gen_binary (SMIN, mode, true_rtx, false_rtx);
|
5697 |
|
|
case GEU:
|
5698 |
|
|
case GTU:
|
5699 |
|
|
return simplify_gen_binary (UMAX, mode, true_rtx, false_rtx);
|
5700 |
|
|
case LEU:
|
5701 |
|
|
case LTU:
|
5702 |
|
|
return simplify_gen_binary (UMIN, mode, true_rtx, false_rtx);
|
5703 |
|
|
default:
|
5704 |
|
|
break;
|
5705 |
|
|
}
|
5706 |
|
|
|
5707 |
|
|
/* If we have (if_then_else COND (OP Z C1) Z) and OP is an identity when its
|
5708 |
|
|
second operand is zero, this can be done as (OP Z (mult COND C2)) where
|
5709 |
|
|
C2 = C1 * STORE_FLAG_VALUE. Similarly if OP has an outer ZERO_EXTEND or
|
5710 |
|
|
SIGN_EXTEND as long as Z is already extended (so we don't destroy it).
|
5711 |
|
|
We can do this kind of thing in some cases when STORE_FLAG_VALUE is
|
5712 |
|
|
neither 1 or -1, but it isn't worth checking for. */
|
5713 |
|
|
|
5714 |
|
|
if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
|
5715 |
|
|
&& comparison_p
|
5716 |
|
|
&& GET_MODE_CLASS (mode) == MODE_INT
|
5717 |
|
|
&& ! side_effects_p (x))
|
5718 |
|
|
{
|
5719 |
|
|
rtx t = make_compound_operation (true_rtx, SET);
|
5720 |
|
|
rtx f = make_compound_operation (false_rtx, SET);
|
5721 |
|
|
rtx cond_op0 = XEXP (cond, 0);
|
5722 |
|
|
rtx cond_op1 = XEXP (cond, 1);
|
5723 |
|
|
enum rtx_code op = UNKNOWN, extend_op = UNKNOWN;
|
5724 |
|
|
enum machine_mode m = mode;
|
5725 |
|
|
rtx z = 0, c1 = NULL_RTX;
|
5726 |
|
|
|
5727 |
|
|
if ((GET_CODE (t) == PLUS || GET_CODE (t) == MINUS
|
5728 |
|
|
|| GET_CODE (t) == IOR || GET_CODE (t) == XOR
|
5729 |
|
|
|| GET_CODE (t) == ASHIFT
|
5730 |
|
|
|| GET_CODE (t) == LSHIFTRT || GET_CODE (t) == ASHIFTRT)
|
5731 |
|
|
&& rtx_equal_p (XEXP (t, 0), f))
|
5732 |
|
|
c1 = XEXP (t, 1), op = GET_CODE (t), z = f;
|
5733 |
|
|
|
5734 |
|
|
/* If an identity-zero op is commutative, check whether there
|
5735 |
|
|
would be a match if we swapped the operands. */
|
5736 |
|
|
else if ((GET_CODE (t) == PLUS || GET_CODE (t) == IOR
|
5737 |
|
|
|| GET_CODE (t) == XOR)
|
5738 |
|
|
&& rtx_equal_p (XEXP (t, 1), f))
|
5739 |
|
|
c1 = XEXP (t, 0), op = GET_CODE (t), z = f;
|
5740 |
|
|
else if (GET_CODE (t) == SIGN_EXTEND
|
5741 |
|
|
&& (GET_CODE (XEXP (t, 0)) == PLUS
|
5742 |
|
|
|| GET_CODE (XEXP (t, 0)) == MINUS
|
5743 |
|
|
|| GET_CODE (XEXP (t, 0)) == IOR
|
5744 |
|
|
|| GET_CODE (XEXP (t, 0)) == XOR
|
5745 |
|
|
|| GET_CODE (XEXP (t, 0)) == ASHIFT
|
5746 |
|
|
|| GET_CODE (XEXP (t, 0)) == LSHIFTRT
|
5747 |
|
|
|| GET_CODE (XEXP (t, 0)) == ASHIFTRT)
|
5748 |
|
|
&& GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
|
5749 |
|
|
&& subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
|
5750 |
|
|
&& rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
|
5751 |
|
|
&& (num_sign_bit_copies (f, GET_MODE (f))
|
5752 |
|
|
> (unsigned int)
|
5753 |
|
|
(GET_MODE_BITSIZE (mode)
|
5754 |
|
|
- GET_MODE_BITSIZE (GET_MODE (XEXP (XEXP (t, 0), 0))))))
|
5755 |
|
|
{
|
5756 |
|
|
c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
|
5757 |
|
|
extend_op = SIGN_EXTEND;
|
5758 |
|
|
m = GET_MODE (XEXP (t, 0));
|
5759 |
|
|
}
|
5760 |
|
|
else if (GET_CODE (t) == SIGN_EXTEND
|
5761 |
|
|
&& (GET_CODE (XEXP (t, 0)) == PLUS
|
5762 |
|
|
|| GET_CODE (XEXP (t, 0)) == IOR
|
5763 |
|
|
|| GET_CODE (XEXP (t, 0)) == XOR)
|
5764 |
|
|
&& GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
|
5765 |
|
|
&& subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
|
5766 |
|
|
&& rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
|
5767 |
|
|
&& (num_sign_bit_copies (f, GET_MODE (f))
|
5768 |
|
|
> (unsigned int)
|
5769 |
|
|
(GET_MODE_BITSIZE (mode)
|
5770 |
|
|
- GET_MODE_BITSIZE (GET_MODE (XEXP (XEXP (t, 0), 1))))))
|
5771 |
|
|
{
|
5772 |
|
|
c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
|
5773 |
|
|
extend_op = SIGN_EXTEND;
|
5774 |
|
|
m = GET_MODE (XEXP (t, 0));
|
5775 |
|
|
}
|
5776 |
|
|
else if (GET_CODE (t) == ZERO_EXTEND
|
5777 |
|
|
&& (GET_CODE (XEXP (t, 0)) == PLUS
|
5778 |
|
|
|| GET_CODE (XEXP (t, 0)) == MINUS
|
5779 |
|
|
|| GET_CODE (XEXP (t, 0)) == IOR
|
5780 |
|
|
|| GET_CODE (XEXP (t, 0)) == XOR
|
5781 |
|
|
|| GET_CODE (XEXP (t, 0)) == ASHIFT
|
5782 |
|
|
|| GET_CODE (XEXP (t, 0)) == LSHIFTRT
|
5783 |
|
|
|| GET_CODE (XEXP (t, 0)) == ASHIFTRT)
|
5784 |
|
|
&& GET_CODE (XEXP (XEXP (t, 0), 0)) == SUBREG
|
5785 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5786 |
|
|
&& subreg_lowpart_p (XEXP (XEXP (t, 0), 0))
|
5787 |
|
|
&& rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 0)), f)
|
5788 |
|
|
&& ((nonzero_bits (f, GET_MODE (f))
|
5789 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (XEXP (t, 0), 0))))
|
5790 |
|
|
== 0))
|
5791 |
|
|
{
|
5792 |
|
|
c1 = XEXP (XEXP (t, 0), 1); z = f; op = GET_CODE (XEXP (t, 0));
|
5793 |
|
|
extend_op = ZERO_EXTEND;
|
5794 |
|
|
m = GET_MODE (XEXP (t, 0));
|
5795 |
|
|
}
|
5796 |
|
|
else if (GET_CODE (t) == ZERO_EXTEND
|
5797 |
|
|
&& (GET_CODE (XEXP (t, 0)) == PLUS
|
5798 |
|
|
|| GET_CODE (XEXP (t, 0)) == IOR
|
5799 |
|
|
|| GET_CODE (XEXP (t, 0)) == XOR)
|
5800 |
|
|
&& GET_CODE (XEXP (XEXP (t, 0), 1)) == SUBREG
|
5801 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
5802 |
|
|
&& subreg_lowpart_p (XEXP (XEXP (t, 0), 1))
|
5803 |
|
|
&& rtx_equal_p (SUBREG_REG (XEXP (XEXP (t, 0), 1)), f)
|
5804 |
|
|
&& ((nonzero_bits (f, GET_MODE (f))
|
5805 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (XEXP (t, 0), 1))))
|
5806 |
|
|
== 0))
|
5807 |
|
|
{
|
5808 |
|
|
c1 = XEXP (XEXP (t, 0), 0); z = f; op = GET_CODE (XEXP (t, 0));
|
5809 |
|
|
extend_op = ZERO_EXTEND;
|
5810 |
|
|
m = GET_MODE (XEXP (t, 0));
|
5811 |
|
|
}
|
5812 |
|
|
|
5813 |
|
|
if (z)
|
5814 |
|
|
{
|
5815 |
|
|
temp = subst (simplify_gen_relational (true_code, m, VOIDmode,
|
5816 |
|
|
cond_op0, cond_op1),
|
5817 |
|
|
pc_rtx, pc_rtx, 0, 0);
|
5818 |
|
|
temp = simplify_gen_binary (MULT, m, temp,
|
5819 |
|
|
simplify_gen_binary (MULT, m, c1,
|
5820 |
|
|
const_true_rtx));
|
5821 |
|
|
temp = subst (temp, pc_rtx, pc_rtx, 0, 0);
|
5822 |
|
|
temp = simplify_gen_binary (op, m, gen_lowpart (m, z), temp);
|
5823 |
|
|
|
5824 |
|
|
if (extend_op != UNKNOWN)
|
5825 |
|
|
temp = simplify_gen_unary (extend_op, mode, temp, m);
|
5826 |
|
|
|
5827 |
|
|
return temp;
|
5828 |
|
|
}
|
5829 |
|
|
}
|
5830 |
|
|
|
5831 |
|
|
/* If we have (if_then_else (ne A 0) C1 0) and either A is known to be 0 or
|
5832 |
|
|
1 and C1 is a single bit or A is known to be 0 or -1 and C1 is the
|
5833 |
|
|
negation of a single bit, we can convert this operation to a shift. We
|
5834 |
|
|
can actually do this more generally, but it doesn't seem worth it. */
|
5835 |
|
|
|
5836 |
|
|
if (true_code == NE && XEXP (cond, 1) == const0_rtx
|
5837 |
|
|
&& false_rtx == const0_rtx && CONST_INT_P (true_rtx)
|
5838 |
|
|
&& ((1 == nonzero_bits (XEXP (cond, 0), mode)
|
5839 |
|
|
&& (i = exact_log2 (INTVAL (true_rtx))) >= 0)
|
5840 |
|
|
|| ((num_sign_bit_copies (XEXP (cond, 0), mode)
|
5841 |
|
|
== GET_MODE_BITSIZE (mode))
|
5842 |
|
|
&& (i = exact_log2 (-INTVAL (true_rtx))) >= 0)))
|
5843 |
|
|
return
|
5844 |
|
|
simplify_shift_const (NULL_RTX, ASHIFT, mode,
|
5845 |
|
|
gen_lowpart (mode, XEXP (cond, 0)), i);
|
5846 |
|
|
|
5847 |
|
|
/* (IF_THEN_ELSE (NE REG 0) (0) (8)) is REG for nonzero_bits (REG) == 8. */
|
5848 |
|
|
if (true_code == NE && XEXP (cond, 1) == const0_rtx
|
5849 |
|
|
&& false_rtx == const0_rtx && CONST_INT_P (true_rtx)
|
5850 |
|
|
&& GET_MODE (XEXP (cond, 0)) == mode
|
5851 |
|
|
&& (INTVAL (true_rtx) & GET_MODE_MASK (mode))
|
5852 |
|
|
== nonzero_bits (XEXP (cond, 0), mode)
|
5853 |
|
|
&& (i = exact_log2 (INTVAL (true_rtx) & GET_MODE_MASK (mode))) >= 0)
|
5854 |
|
|
return XEXP (cond, 0);
|
5855 |
|
|
|
5856 |
|
|
return x;
|
5857 |
|
|
}
|
5858 |
|
|
|
5859 |
|
|
/* Simplify X, a SET expression. Return the new expression. */
|
5860 |
|
|
|
5861 |
|
|
static rtx
|
5862 |
|
|
simplify_set (rtx x)
|
5863 |
|
|
{
|
5864 |
|
|
rtx src = SET_SRC (x);
|
5865 |
|
|
rtx dest = SET_DEST (x);
|
5866 |
|
|
enum machine_mode mode
|
5867 |
|
|
= GET_MODE (src) != VOIDmode ? GET_MODE (src) : GET_MODE (dest);
|
5868 |
|
|
rtx other_insn;
|
5869 |
|
|
rtx *cc_use;
|
5870 |
|
|
|
5871 |
|
|
/* (set (pc) (return)) gets written as (return). */
|
5872 |
|
|
if (GET_CODE (dest) == PC && GET_CODE (src) == RETURN)
|
5873 |
|
|
return src;
|
5874 |
|
|
|
5875 |
|
|
/* Now that we know for sure which bits of SRC we are using, see if we can
|
5876 |
|
|
simplify the expression for the object knowing that we only need the
|
5877 |
|
|
low-order bits. */
|
5878 |
|
|
|
5879 |
|
|
if (GET_MODE_CLASS (mode) == MODE_INT
|
5880 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
5881 |
|
|
{
|
5882 |
|
|
src = force_to_mode (src, mode, ~(HOST_WIDE_INT) 0, 0);
|
5883 |
|
|
SUBST (SET_SRC (x), src);
|
5884 |
|
|
}
|
5885 |
|
|
|
5886 |
|
|
/* If we are setting CC0 or if the source is a COMPARE, look for the use of
|
5887 |
|
|
the comparison result and try to simplify it unless we already have used
|
5888 |
|
|
undobuf.other_insn. */
|
5889 |
|
|
if ((GET_MODE_CLASS (mode) == MODE_CC
|
5890 |
|
|
|| GET_CODE (src) == COMPARE
|
5891 |
|
|
|| CC0_P (dest))
|
5892 |
|
|
&& (cc_use = find_single_use (dest, subst_insn, &other_insn)) != 0
|
5893 |
|
|
&& (undobuf.other_insn == 0 || other_insn == undobuf.other_insn)
|
5894 |
|
|
&& COMPARISON_P (*cc_use)
|
5895 |
|
|
&& rtx_equal_p (XEXP (*cc_use, 0), dest))
|
5896 |
|
|
{
|
5897 |
|
|
enum rtx_code old_code = GET_CODE (*cc_use);
|
5898 |
|
|
enum rtx_code new_code;
|
5899 |
|
|
rtx op0, op1, tmp;
|
5900 |
|
|
int other_changed = 0;
|
5901 |
|
|
enum machine_mode compare_mode = GET_MODE (dest);
|
5902 |
|
|
|
5903 |
|
|
if (GET_CODE (src) == COMPARE)
|
5904 |
|
|
op0 = XEXP (src, 0), op1 = XEXP (src, 1);
|
5905 |
|
|
else
|
5906 |
|
|
op0 = src, op1 = CONST0_RTX (GET_MODE (src));
|
5907 |
|
|
|
5908 |
|
|
tmp = simplify_relational_operation (old_code, compare_mode, VOIDmode,
|
5909 |
|
|
op0, op1);
|
5910 |
|
|
if (!tmp)
|
5911 |
|
|
new_code = old_code;
|
5912 |
|
|
else if (!CONSTANT_P (tmp))
|
5913 |
|
|
{
|
5914 |
|
|
new_code = GET_CODE (tmp);
|
5915 |
|
|
op0 = XEXP (tmp, 0);
|
5916 |
|
|
op1 = XEXP (tmp, 1);
|
5917 |
|
|
}
|
5918 |
|
|
else
|
5919 |
|
|
{
|
5920 |
|
|
rtx pat = PATTERN (other_insn);
|
5921 |
|
|
undobuf.other_insn = other_insn;
|
5922 |
|
|
SUBST (*cc_use, tmp);
|
5923 |
|
|
|
5924 |
|
|
/* Attempt to simplify CC user. */
|
5925 |
|
|
if (GET_CODE (pat) == SET)
|
5926 |
|
|
{
|
5927 |
|
|
rtx new_rtx = simplify_rtx (SET_SRC (pat));
|
5928 |
|
|
if (new_rtx != NULL_RTX)
|
5929 |
|
|
SUBST (SET_SRC (pat), new_rtx);
|
5930 |
|
|
}
|
5931 |
|
|
|
5932 |
|
|
/* Convert X into a no-op move. */
|
5933 |
|
|
SUBST (SET_DEST (x), pc_rtx);
|
5934 |
|
|
SUBST (SET_SRC (x), pc_rtx);
|
5935 |
|
|
return x;
|
5936 |
|
|
}
|
5937 |
|
|
|
5938 |
|
|
/* Simplify our comparison, if possible. */
|
5939 |
|
|
new_code = simplify_comparison (new_code, &op0, &op1);
|
5940 |
|
|
|
5941 |
|
|
#ifdef SELECT_CC_MODE
|
5942 |
|
|
/* If this machine has CC modes other than CCmode, check to see if we
|
5943 |
|
|
need to use a different CC mode here. */
|
5944 |
|
|
if (GET_MODE_CLASS (GET_MODE (op0)) == MODE_CC)
|
5945 |
|
|
compare_mode = GET_MODE (op0);
|
5946 |
|
|
else
|
5947 |
|
|
compare_mode = SELECT_CC_MODE (new_code, op0, op1);
|
5948 |
|
|
|
5949 |
|
|
#ifndef HAVE_cc0
|
5950 |
|
|
/* If the mode changed, we have to change SET_DEST, the mode in the
|
5951 |
|
|
compare, and the mode in the place SET_DEST is used. If SET_DEST is
|
5952 |
|
|
a hard register, just build new versions with the proper mode. If it
|
5953 |
|
|
is a pseudo, we lose unless it is only time we set the pseudo, in
|
5954 |
|
|
which case we can safely change its mode. */
|
5955 |
|
|
if (compare_mode != GET_MODE (dest))
|
5956 |
|
|
{
|
5957 |
|
|
if (can_change_dest_mode (dest, 0, compare_mode))
|
5958 |
|
|
{
|
5959 |
|
|
unsigned int regno = REGNO (dest);
|
5960 |
|
|
rtx new_dest;
|
5961 |
|
|
|
5962 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
5963 |
|
|
new_dest = gen_rtx_REG (compare_mode, regno);
|
5964 |
|
|
else
|
5965 |
|
|
{
|
5966 |
|
|
SUBST_MODE (regno_reg_rtx[regno], compare_mode);
|
5967 |
|
|
new_dest = regno_reg_rtx[regno];
|
5968 |
|
|
}
|
5969 |
|
|
|
5970 |
|
|
SUBST (SET_DEST (x), new_dest);
|
5971 |
|
|
SUBST (XEXP (*cc_use, 0), new_dest);
|
5972 |
|
|
other_changed = 1;
|
5973 |
|
|
|
5974 |
|
|
dest = new_dest;
|
5975 |
|
|
}
|
5976 |
|
|
}
|
5977 |
|
|
#endif /* cc0 */
|
5978 |
|
|
#endif /* SELECT_CC_MODE */
|
5979 |
|
|
|
5980 |
|
|
/* If the code changed, we have to build a new comparison in
|
5981 |
|
|
undobuf.other_insn. */
|
5982 |
|
|
if (new_code != old_code)
|
5983 |
|
|
{
|
5984 |
|
|
int other_changed_previously = other_changed;
|
5985 |
|
|
unsigned HOST_WIDE_INT mask;
|
5986 |
|
|
rtx old_cc_use = *cc_use;
|
5987 |
|
|
|
5988 |
|
|
SUBST (*cc_use, gen_rtx_fmt_ee (new_code, GET_MODE (*cc_use),
|
5989 |
|
|
dest, const0_rtx));
|
5990 |
|
|
other_changed = 1;
|
5991 |
|
|
|
5992 |
|
|
/* If the only change we made was to change an EQ into an NE or
|
5993 |
|
|
vice versa, OP0 has only one bit that might be nonzero, and OP1
|
5994 |
|
|
is zero, check if changing the user of the condition code will
|
5995 |
|
|
produce a valid insn. If it won't, we can keep the original code
|
5996 |
|
|
in that insn by surrounding our operation with an XOR. */
|
5997 |
|
|
|
5998 |
|
|
if (((old_code == NE && new_code == EQ)
|
5999 |
|
|
|| (old_code == EQ && new_code == NE))
|
6000 |
|
|
&& ! other_changed_previously && op1 == const0_rtx
|
6001 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (op0)) <= HOST_BITS_PER_WIDE_INT
|
6002 |
|
|
&& exact_log2 (mask = nonzero_bits (op0, GET_MODE (op0))) >= 0)
|
6003 |
|
|
{
|
6004 |
|
|
rtx pat = PATTERN (other_insn), note = 0;
|
6005 |
|
|
|
6006 |
|
|
if ((recog_for_combine (&pat, other_insn, ¬e) < 0
|
6007 |
|
|
&& ! check_asm_operands (pat)))
|
6008 |
|
|
{
|
6009 |
|
|
*cc_use = old_cc_use;
|
6010 |
|
|
other_changed = 0;
|
6011 |
|
|
|
6012 |
|
|
op0 = simplify_gen_binary (XOR, GET_MODE (op0),
|
6013 |
|
|
op0, GEN_INT (mask));
|
6014 |
|
|
}
|
6015 |
|
|
}
|
6016 |
|
|
}
|
6017 |
|
|
|
6018 |
|
|
if (other_changed)
|
6019 |
|
|
undobuf.other_insn = other_insn;
|
6020 |
|
|
|
6021 |
|
|
/* Otherwise, if we didn't previously have a COMPARE in the
|
6022 |
|
|
correct mode, we need one. */
|
6023 |
|
|
if (GET_CODE (src) != COMPARE || GET_MODE (src) != compare_mode)
|
6024 |
|
|
{
|
6025 |
|
|
SUBST (SET_SRC (x), gen_rtx_COMPARE (compare_mode, op0, op1));
|
6026 |
|
|
src = SET_SRC (x);
|
6027 |
|
|
}
|
6028 |
|
|
else if (GET_MODE (op0) == compare_mode && op1 == const0_rtx)
|
6029 |
|
|
{
|
6030 |
|
|
SUBST (SET_SRC (x), op0);
|
6031 |
|
|
src = SET_SRC (x);
|
6032 |
|
|
}
|
6033 |
|
|
/* Otherwise, update the COMPARE if needed. */
|
6034 |
|
|
else if (XEXP (src, 0) != op0 || XEXP (src, 1) != op1)
|
6035 |
|
|
{
|
6036 |
|
|
SUBST (SET_SRC (x), gen_rtx_COMPARE (compare_mode, op0, op1));
|
6037 |
|
|
src = SET_SRC (x);
|
6038 |
|
|
}
|
6039 |
|
|
}
|
6040 |
|
|
else
|
6041 |
|
|
{
|
6042 |
|
|
/* Get SET_SRC in a form where we have placed back any
|
6043 |
|
|
compound expressions. Then do the checks below. */
|
6044 |
|
|
src = make_compound_operation (src, SET);
|
6045 |
|
|
SUBST (SET_SRC (x), src);
|
6046 |
|
|
}
|
6047 |
|
|
|
6048 |
|
|
/* If we have (set x (subreg:m1 (op:m2 ...) 0)) with OP being some operation,
|
6049 |
|
|
and X being a REG or (subreg (reg)), we may be able to convert this to
|
6050 |
|
|
(set (subreg:m2 x) (op)).
|
6051 |
|
|
|
6052 |
|
|
We can always do this if M1 is narrower than M2 because that means that
|
6053 |
|
|
we only care about the low bits of the result.
|
6054 |
|
|
|
6055 |
|
|
However, on machines without WORD_REGISTER_OPERATIONS defined, we cannot
|
6056 |
|
|
perform a narrower operation than requested since the high-order bits will
|
6057 |
|
|
be undefined. On machine where it is defined, this transformation is safe
|
6058 |
|
|
as long as M1 and M2 have the same number of words. */
|
6059 |
|
|
|
6060 |
|
|
if (GET_CODE (src) == SUBREG && subreg_lowpart_p (src)
|
6061 |
|
|
&& !OBJECT_P (SUBREG_REG (src))
|
6062 |
|
|
&& (((GET_MODE_SIZE (GET_MODE (src)) + (UNITS_PER_WORD - 1))
|
6063 |
|
|
/ UNITS_PER_WORD)
|
6064 |
|
|
== ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))
|
6065 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD))
|
6066 |
|
|
#ifndef WORD_REGISTER_OPERATIONS
|
6067 |
|
|
&& (GET_MODE_SIZE (GET_MODE (src))
|
6068 |
|
|
< GET_MODE_SIZE (GET_MODE (SUBREG_REG (src))))
|
6069 |
|
|
#endif
|
6070 |
|
|
#ifdef CANNOT_CHANGE_MODE_CLASS
|
6071 |
|
|
&& ! (REG_P (dest) && REGNO (dest) < FIRST_PSEUDO_REGISTER
|
6072 |
|
|
&& REG_CANNOT_CHANGE_MODE_P (REGNO (dest),
|
6073 |
|
|
GET_MODE (SUBREG_REG (src)),
|
6074 |
|
|
GET_MODE (src)))
|
6075 |
|
|
#endif
|
6076 |
|
|
&& (REG_P (dest)
|
6077 |
|
|
|| (GET_CODE (dest) == SUBREG
|
6078 |
|
|
&& REG_P (SUBREG_REG (dest)))))
|
6079 |
|
|
{
|
6080 |
|
|
SUBST (SET_DEST (x),
|
6081 |
|
|
gen_lowpart (GET_MODE (SUBREG_REG (src)),
|
6082 |
|
|
dest));
|
6083 |
|
|
SUBST (SET_SRC (x), SUBREG_REG (src));
|
6084 |
|
|
|
6085 |
|
|
src = SET_SRC (x), dest = SET_DEST (x);
|
6086 |
|
|
}
|
6087 |
|
|
|
6088 |
|
|
#ifdef HAVE_cc0
|
6089 |
|
|
/* If we have (set (cc0) (subreg ...)), we try to remove the subreg
|
6090 |
|
|
in SRC. */
|
6091 |
|
|
if (dest == cc0_rtx
|
6092 |
|
|
&& GET_CODE (src) == SUBREG
|
6093 |
|
|
&& subreg_lowpart_p (src)
|
6094 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (src))
|
6095 |
|
|
< GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (src)))))
|
6096 |
|
|
{
|
6097 |
|
|
rtx inner = SUBREG_REG (src);
|
6098 |
|
|
enum machine_mode inner_mode = GET_MODE (inner);
|
6099 |
|
|
|
6100 |
|
|
/* Here we make sure that we don't have a sign bit on. */
|
6101 |
|
|
if (GET_MODE_BITSIZE (inner_mode) <= HOST_BITS_PER_WIDE_INT
|
6102 |
|
|
&& (nonzero_bits (inner, inner_mode)
|
6103 |
|
|
< ((unsigned HOST_WIDE_INT) 1
|
6104 |
|
|
<< (GET_MODE_BITSIZE (GET_MODE (src)) - 1))))
|
6105 |
|
|
{
|
6106 |
|
|
SUBST (SET_SRC (x), inner);
|
6107 |
|
|
src = SET_SRC (x);
|
6108 |
|
|
}
|
6109 |
|
|
}
|
6110 |
|
|
#endif
|
6111 |
|
|
|
6112 |
|
|
#ifdef LOAD_EXTEND_OP
|
6113 |
|
|
/* If we have (set FOO (subreg:M (mem:N BAR) 0)) with M wider than N, this
|
6114 |
|
|
would require a paradoxical subreg. Replace the subreg with a
|
6115 |
|
|
zero_extend to avoid the reload that would otherwise be required. */
|
6116 |
|
|
|
6117 |
|
|
if (GET_CODE (src) == SUBREG && subreg_lowpart_p (src)
|
6118 |
|
|
&& INTEGRAL_MODE_P (GET_MODE (SUBREG_REG (src)))
|
6119 |
|
|
&& LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (src))) != UNKNOWN
|
6120 |
|
|
&& SUBREG_BYTE (src) == 0
|
6121 |
|
|
&& (GET_MODE_SIZE (GET_MODE (src))
|
6122 |
|
|
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (src))))
|
6123 |
|
|
&& MEM_P (SUBREG_REG (src)))
|
6124 |
|
|
{
|
6125 |
|
|
SUBST (SET_SRC (x),
|
6126 |
|
|
gen_rtx_fmt_e (LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (src))),
|
6127 |
|
|
GET_MODE (src), SUBREG_REG (src)));
|
6128 |
|
|
|
6129 |
|
|
src = SET_SRC (x);
|
6130 |
|
|
}
|
6131 |
|
|
#endif
|
6132 |
|
|
|
6133 |
|
|
/* If we don't have a conditional move, SET_SRC is an IF_THEN_ELSE, and we
|
6134 |
|
|
are comparing an item known to be 0 or -1 against 0, use a logical
|
6135 |
|
|
operation instead. Check for one of the arms being an IOR of the other
|
6136 |
|
|
arm with some value. We compute three terms to be IOR'ed together. In
|
6137 |
|
|
practice, at most two will be nonzero. Then we do the IOR's. */
|
6138 |
|
|
|
6139 |
|
|
if (GET_CODE (dest) != PC
|
6140 |
|
|
&& GET_CODE (src) == IF_THEN_ELSE
|
6141 |
|
|
&& GET_MODE_CLASS (GET_MODE (src)) == MODE_INT
|
6142 |
|
|
&& (GET_CODE (XEXP (src, 0)) == EQ || GET_CODE (XEXP (src, 0)) == NE)
|
6143 |
|
|
&& XEXP (XEXP (src, 0), 1) == const0_rtx
|
6144 |
|
|
&& GET_MODE (src) == GET_MODE (XEXP (XEXP (src, 0), 0))
|
6145 |
|
|
#ifdef HAVE_conditional_move
|
6146 |
|
|
&& ! can_conditionally_move_p (GET_MODE (src))
|
6147 |
|
|
#endif
|
6148 |
|
|
&& (num_sign_bit_copies (XEXP (XEXP (src, 0), 0),
|
6149 |
|
|
GET_MODE (XEXP (XEXP (src, 0), 0)))
|
6150 |
|
|
== GET_MODE_BITSIZE (GET_MODE (XEXP (XEXP (src, 0), 0))))
|
6151 |
|
|
&& ! side_effects_p (src))
|
6152 |
|
|
{
|
6153 |
|
|
rtx true_rtx = (GET_CODE (XEXP (src, 0)) == NE
|
6154 |
|
|
? XEXP (src, 1) : XEXP (src, 2));
|
6155 |
|
|
rtx false_rtx = (GET_CODE (XEXP (src, 0)) == NE
|
6156 |
|
|
? XEXP (src, 2) : XEXP (src, 1));
|
6157 |
|
|
rtx term1 = const0_rtx, term2, term3;
|
6158 |
|
|
|
6159 |
|
|
if (GET_CODE (true_rtx) == IOR
|
6160 |
|
|
&& rtx_equal_p (XEXP (true_rtx, 0), false_rtx))
|
6161 |
|
|
term1 = false_rtx, true_rtx = XEXP (true_rtx, 1), false_rtx = const0_rtx;
|
6162 |
|
|
else if (GET_CODE (true_rtx) == IOR
|
6163 |
|
|
&& rtx_equal_p (XEXP (true_rtx, 1), false_rtx))
|
6164 |
|
|
term1 = false_rtx, true_rtx = XEXP (true_rtx, 0), false_rtx = const0_rtx;
|
6165 |
|
|
else if (GET_CODE (false_rtx) == IOR
|
6166 |
|
|
&& rtx_equal_p (XEXP (false_rtx, 0), true_rtx))
|
6167 |
|
|
term1 = true_rtx, false_rtx = XEXP (false_rtx, 1), true_rtx = const0_rtx;
|
6168 |
|
|
else if (GET_CODE (false_rtx) == IOR
|
6169 |
|
|
&& rtx_equal_p (XEXP (false_rtx, 1), true_rtx))
|
6170 |
|
|
term1 = true_rtx, false_rtx = XEXP (false_rtx, 0), true_rtx = const0_rtx;
|
6171 |
|
|
|
6172 |
|
|
term2 = simplify_gen_binary (AND, GET_MODE (src),
|
6173 |
|
|
XEXP (XEXP (src, 0), 0), true_rtx);
|
6174 |
|
|
term3 = simplify_gen_binary (AND, GET_MODE (src),
|
6175 |
|
|
simplify_gen_unary (NOT, GET_MODE (src),
|
6176 |
|
|
XEXP (XEXP (src, 0), 0),
|
6177 |
|
|
GET_MODE (src)),
|
6178 |
|
|
false_rtx);
|
6179 |
|
|
|
6180 |
|
|
SUBST (SET_SRC (x),
|
6181 |
|
|
simplify_gen_binary (IOR, GET_MODE (src),
|
6182 |
|
|
simplify_gen_binary (IOR, GET_MODE (src),
|
6183 |
|
|
term1, term2),
|
6184 |
|
|
term3));
|
6185 |
|
|
|
6186 |
|
|
src = SET_SRC (x);
|
6187 |
|
|
}
|
6188 |
|
|
|
6189 |
|
|
/* If either SRC or DEST is a CLOBBER of (const_int 0), make this
|
6190 |
|
|
whole thing fail. */
|
6191 |
|
|
if (GET_CODE (src) == CLOBBER && XEXP (src, 0) == const0_rtx)
|
6192 |
|
|
return src;
|
6193 |
|
|
else if (GET_CODE (dest) == CLOBBER && XEXP (dest, 0) == const0_rtx)
|
6194 |
|
|
return dest;
|
6195 |
|
|
else
|
6196 |
|
|
/* Convert this into a field assignment operation, if possible. */
|
6197 |
|
|
return make_field_assignment (x);
|
6198 |
|
|
}
|
6199 |
|
|
|
6200 |
|
|
/* Simplify, X, and AND, IOR, or XOR operation, and return the simplified
|
6201 |
|
|
result. */
|
6202 |
|
|
|
6203 |
|
|
static rtx
|
6204 |
|
|
simplify_logical (rtx x)
|
6205 |
|
|
{
|
6206 |
|
|
enum machine_mode mode = GET_MODE (x);
|
6207 |
|
|
rtx op0 = XEXP (x, 0);
|
6208 |
|
|
rtx op1 = XEXP (x, 1);
|
6209 |
|
|
|
6210 |
|
|
switch (GET_CODE (x))
|
6211 |
|
|
{
|
6212 |
|
|
case AND:
|
6213 |
|
|
/* We can call simplify_and_const_int only if we don't lose
|
6214 |
|
|
any (sign) bits when converting INTVAL (op1) to
|
6215 |
|
|
"unsigned HOST_WIDE_INT". */
|
6216 |
|
|
if (CONST_INT_P (op1)
|
6217 |
|
|
&& (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
6218 |
|
|
|| INTVAL (op1) > 0))
|
6219 |
|
|
{
|
6220 |
|
|
x = simplify_and_const_int (x, mode, op0, INTVAL (op1));
|
6221 |
|
|
if (GET_CODE (x) != AND)
|
6222 |
|
|
return x;
|
6223 |
|
|
|
6224 |
|
|
op0 = XEXP (x, 0);
|
6225 |
|
|
op1 = XEXP (x, 1);
|
6226 |
|
|
}
|
6227 |
|
|
|
6228 |
|
|
/* If we have any of (and (ior A B) C) or (and (xor A B) C),
|
6229 |
|
|
apply the distributive law and then the inverse distributive
|
6230 |
|
|
law to see if things simplify. */
|
6231 |
|
|
if (GET_CODE (op0) == IOR || GET_CODE (op0) == XOR)
|
6232 |
|
|
{
|
6233 |
|
|
rtx result = distribute_and_simplify_rtx (x, 0);
|
6234 |
|
|
if (result)
|
6235 |
|
|
return result;
|
6236 |
|
|
}
|
6237 |
|
|
if (GET_CODE (op1) == IOR || GET_CODE (op1) == XOR)
|
6238 |
|
|
{
|
6239 |
|
|
rtx result = distribute_and_simplify_rtx (x, 1);
|
6240 |
|
|
if (result)
|
6241 |
|
|
return result;
|
6242 |
|
|
}
|
6243 |
|
|
break;
|
6244 |
|
|
|
6245 |
|
|
case IOR:
|
6246 |
|
|
/* If we have (ior (and A B) C), apply the distributive law and then
|
6247 |
|
|
the inverse distributive law to see if things simplify. */
|
6248 |
|
|
|
6249 |
|
|
if (GET_CODE (op0) == AND)
|
6250 |
|
|
{
|
6251 |
|
|
rtx result = distribute_and_simplify_rtx (x, 0);
|
6252 |
|
|
if (result)
|
6253 |
|
|
return result;
|
6254 |
|
|
}
|
6255 |
|
|
|
6256 |
|
|
if (GET_CODE (op1) == AND)
|
6257 |
|
|
{
|
6258 |
|
|
rtx result = distribute_and_simplify_rtx (x, 1);
|
6259 |
|
|
if (result)
|
6260 |
|
|
return result;
|
6261 |
|
|
}
|
6262 |
|
|
break;
|
6263 |
|
|
|
6264 |
|
|
default:
|
6265 |
|
|
gcc_unreachable ();
|
6266 |
|
|
}
|
6267 |
|
|
|
6268 |
|
|
return x;
|
6269 |
|
|
}
|
6270 |
|
|
|
6271 |
|
|
/* We consider ZERO_EXTRACT, SIGN_EXTRACT, and SIGN_EXTEND as "compound
|
6272 |
|
|
operations" because they can be replaced with two more basic operations.
|
6273 |
|
|
ZERO_EXTEND is also considered "compound" because it can be replaced with
|
6274 |
|
|
an AND operation, which is simpler, though only one operation.
|
6275 |
|
|
|
6276 |
|
|
The function expand_compound_operation is called with an rtx expression
|
6277 |
|
|
and will convert it to the appropriate shifts and AND operations,
|
6278 |
|
|
simplifying at each stage.
|
6279 |
|
|
|
6280 |
|
|
The function make_compound_operation is called to convert an expression
|
6281 |
|
|
consisting of shifts and ANDs into the equivalent compound expression.
|
6282 |
|
|
It is the inverse of this function, loosely speaking. */
|
6283 |
|
|
|
6284 |
|
|
static rtx
|
6285 |
|
|
expand_compound_operation (rtx x)
|
6286 |
|
|
{
|
6287 |
|
|
unsigned HOST_WIDE_INT pos = 0, len;
|
6288 |
|
|
int unsignedp = 0;
|
6289 |
|
|
unsigned int modewidth;
|
6290 |
|
|
rtx tem;
|
6291 |
|
|
|
6292 |
|
|
switch (GET_CODE (x))
|
6293 |
|
|
{
|
6294 |
|
|
case ZERO_EXTEND:
|
6295 |
|
|
unsignedp = 1;
|
6296 |
|
|
case SIGN_EXTEND:
|
6297 |
|
|
/* We can't necessarily use a const_int for a multiword mode;
|
6298 |
|
|
it depends on implicitly extending the value.
|
6299 |
|
|
Since we don't know the right way to extend it,
|
6300 |
|
|
we can't tell whether the implicit way is right.
|
6301 |
|
|
|
6302 |
|
|
Even for a mode that is no wider than a const_int,
|
6303 |
|
|
we can't win, because we need to sign extend one of its bits through
|
6304 |
|
|
the rest of it, and we don't know which bit. */
|
6305 |
|
|
if (CONST_INT_P (XEXP (x, 0)))
|
6306 |
|
|
return x;
|
6307 |
|
|
|
6308 |
|
|
/* Return if (subreg:MODE FROM 0) is not a safe replacement for
|
6309 |
|
|
(zero_extend:MODE FROM) or (sign_extend:MODE FROM). It is for any MEM
|
6310 |
|
|
because (SUBREG (MEM...)) is guaranteed to cause the MEM to be
|
6311 |
|
|
reloaded. If not for that, MEM's would very rarely be safe.
|
6312 |
|
|
|
6313 |
|
|
Reject MODEs bigger than a word, because we might not be able
|
6314 |
|
|
to reference a two-register group starting with an arbitrary register
|
6315 |
|
|
(and currently gen_lowpart might crash for a SUBREG). */
|
6316 |
|
|
|
6317 |
|
|
if (GET_MODE_SIZE (GET_MODE (XEXP (x, 0))) > UNITS_PER_WORD)
|
6318 |
|
|
return x;
|
6319 |
|
|
|
6320 |
|
|
/* Reject MODEs that aren't scalar integers because turning vector
|
6321 |
|
|
or complex modes into shifts causes problems. */
|
6322 |
|
|
|
6323 |
|
|
if (! SCALAR_INT_MODE_P (GET_MODE (XEXP (x, 0))))
|
6324 |
|
|
return x;
|
6325 |
|
|
|
6326 |
|
|
len = GET_MODE_BITSIZE (GET_MODE (XEXP (x, 0)));
|
6327 |
|
|
/* If the inner object has VOIDmode (the only way this can happen
|
6328 |
|
|
is if it is an ASM_OPERANDS), we can't do anything since we don't
|
6329 |
|
|
know how much masking to do. */
|
6330 |
|
|
if (len == 0)
|
6331 |
|
|
return x;
|
6332 |
|
|
|
6333 |
|
|
break;
|
6334 |
|
|
|
6335 |
|
|
case ZERO_EXTRACT:
|
6336 |
|
|
unsignedp = 1;
|
6337 |
|
|
|
6338 |
|
|
/* ... fall through ... */
|
6339 |
|
|
|
6340 |
|
|
case SIGN_EXTRACT:
|
6341 |
|
|
/* If the operand is a CLOBBER, just return it. */
|
6342 |
|
|
if (GET_CODE (XEXP (x, 0)) == CLOBBER)
|
6343 |
|
|
return XEXP (x, 0);
|
6344 |
|
|
|
6345 |
|
|
if (!CONST_INT_P (XEXP (x, 1))
|
6346 |
|
|
|| !CONST_INT_P (XEXP (x, 2))
|
6347 |
|
|
|| GET_MODE (XEXP (x, 0)) == VOIDmode)
|
6348 |
|
|
return x;
|
6349 |
|
|
|
6350 |
|
|
/* Reject MODEs that aren't scalar integers because turning vector
|
6351 |
|
|
or complex modes into shifts causes problems. */
|
6352 |
|
|
|
6353 |
|
|
if (! SCALAR_INT_MODE_P (GET_MODE (XEXP (x, 0))))
|
6354 |
|
|
return x;
|
6355 |
|
|
|
6356 |
|
|
len = INTVAL (XEXP (x, 1));
|
6357 |
|
|
pos = INTVAL (XEXP (x, 2));
|
6358 |
|
|
|
6359 |
|
|
/* This should stay within the object being extracted, fail otherwise. */
|
6360 |
|
|
if (len + pos > GET_MODE_BITSIZE (GET_MODE (XEXP (x, 0))))
|
6361 |
|
|
return x;
|
6362 |
|
|
|
6363 |
|
|
if (BITS_BIG_ENDIAN)
|
6364 |
|
|
pos = GET_MODE_BITSIZE (GET_MODE (XEXP (x, 0))) - len - pos;
|
6365 |
|
|
|
6366 |
|
|
break;
|
6367 |
|
|
|
6368 |
|
|
default:
|
6369 |
|
|
return x;
|
6370 |
|
|
}
|
6371 |
|
|
/* Convert sign extension to zero extension, if we know that the high
|
6372 |
|
|
bit is not set, as this is easier to optimize. It will be converted
|
6373 |
|
|
back to cheaper alternative in make_extraction. */
|
6374 |
|
|
if (GET_CODE (x) == SIGN_EXTEND
|
6375 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT
|
6376 |
|
|
&& ((nonzero_bits (XEXP (x, 0), GET_MODE (XEXP (x, 0)))
|
6377 |
|
|
& ~(((unsigned HOST_WIDE_INT)
|
6378 |
|
|
GET_MODE_MASK (GET_MODE (XEXP (x, 0))))
|
6379 |
|
|
>> 1))
|
6380 |
|
|
== 0)))
|
6381 |
|
|
{
|
6382 |
|
|
rtx temp = gen_rtx_ZERO_EXTEND (GET_MODE (x), XEXP (x, 0));
|
6383 |
|
|
rtx temp2 = expand_compound_operation (temp);
|
6384 |
|
|
|
6385 |
|
|
/* Make sure this is a profitable operation. */
|
6386 |
|
|
if (rtx_cost (x, SET, optimize_this_for_speed_p)
|
6387 |
|
|
> rtx_cost (temp2, SET, optimize_this_for_speed_p))
|
6388 |
|
|
return temp2;
|
6389 |
|
|
else if (rtx_cost (x, SET, optimize_this_for_speed_p)
|
6390 |
|
|
> rtx_cost (temp, SET, optimize_this_for_speed_p))
|
6391 |
|
|
return temp;
|
6392 |
|
|
else
|
6393 |
|
|
return x;
|
6394 |
|
|
}
|
6395 |
|
|
|
6396 |
|
|
/* We can optimize some special cases of ZERO_EXTEND. */
|
6397 |
|
|
if (GET_CODE (x) == ZERO_EXTEND)
|
6398 |
|
|
{
|
6399 |
|
|
/* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI if we
|
6400 |
|
|
know that the last value didn't have any inappropriate bits
|
6401 |
|
|
set. */
|
6402 |
|
|
if (GET_CODE (XEXP (x, 0)) == TRUNCATE
|
6403 |
|
|
&& GET_MODE (XEXP (XEXP (x, 0), 0)) == GET_MODE (x)
|
6404 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT
|
6405 |
|
|
&& (nonzero_bits (XEXP (XEXP (x, 0), 0), GET_MODE (x))
|
6406 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
|
6407 |
|
|
return XEXP (XEXP (x, 0), 0);
|
6408 |
|
|
|
6409 |
|
|
/* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
|
6410 |
|
|
if (GET_CODE (XEXP (x, 0)) == SUBREG
|
6411 |
|
|
&& GET_MODE (SUBREG_REG (XEXP (x, 0))) == GET_MODE (x)
|
6412 |
|
|
&& subreg_lowpart_p (XEXP (x, 0))
|
6413 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT
|
6414 |
|
|
&& (nonzero_bits (SUBREG_REG (XEXP (x, 0)), GET_MODE (x))
|
6415 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
|
6416 |
|
|
return SUBREG_REG (XEXP (x, 0));
|
6417 |
|
|
|
6418 |
|
|
/* (zero_extend:DI (truncate:SI foo:DI)) is just foo:DI when foo
|
6419 |
|
|
is a comparison and STORE_FLAG_VALUE permits. This is like
|
6420 |
|
|
the first case, but it works even when GET_MODE (x) is larger
|
6421 |
|
|
than HOST_WIDE_INT. */
|
6422 |
|
|
if (GET_CODE (XEXP (x, 0)) == TRUNCATE
|
6423 |
|
|
&& GET_MODE (XEXP (XEXP (x, 0), 0)) == GET_MODE (x)
|
6424 |
|
|
&& COMPARISON_P (XEXP (XEXP (x, 0), 0))
|
6425 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (XEXP (x, 0)))
|
6426 |
|
|
<= HOST_BITS_PER_WIDE_INT)
|
6427 |
|
|
&& ((HOST_WIDE_INT) STORE_FLAG_VALUE
|
6428 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
|
6429 |
|
|
return XEXP (XEXP (x, 0), 0);
|
6430 |
|
|
|
6431 |
|
|
/* Likewise for (zero_extend:DI (subreg:SI foo:DI 0)). */
|
6432 |
|
|
if (GET_CODE (XEXP (x, 0)) == SUBREG
|
6433 |
|
|
&& GET_MODE (SUBREG_REG (XEXP (x, 0))) == GET_MODE (x)
|
6434 |
|
|
&& subreg_lowpart_p (XEXP (x, 0))
|
6435 |
|
|
&& COMPARISON_P (SUBREG_REG (XEXP (x, 0)))
|
6436 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (XEXP (x, 0)))
|
6437 |
|
|
<= HOST_BITS_PER_WIDE_INT)
|
6438 |
|
|
&& ((HOST_WIDE_INT) STORE_FLAG_VALUE
|
6439 |
|
|
& ~GET_MODE_MASK (GET_MODE (XEXP (x, 0)))) == 0)
|
6440 |
|
|
return SUBREG_REG (XEXP (x, 0));
|
6441 |
|
|
|
6442 |
|
|
}
|
6443 |
|
|
|
6444 |
|
|
/* If we reach here, we want to return a pair of shifts. The inner
|
6445 |
|
|
shift is a left shift of BITSIZE - POS - LEN bits. The outer
|
6446 |
|
|
shift is a right shift of BITSIZE - LEN bits. It is arithmetic or
|
6447 |
|
|
logical depending on the value of UNSIGNEDP.
|
6448 |
|
|
|
6449 |
|
|
If this was a ZERO_EXTEND or ZERO_EXTRACT, this pair of shifts will be
|
6450 |
|
|
converted into an AND of a shift.
|
6451 |
|
|
|
6452 |
|
|
We must check for the case where the left shift would have a negative
|
6453 |
|
|
count. This can happen in a case like (x >> 31) & 255 on machines
|
6454 |
|
|
that can't shift by a constant. On those machines, we would first
|
6455 |
|
|
combine the shift with the AND to produce a variable-position
|
6456 |
|
|
extraction. Then the constant of 31 would be substituted in to produce
|
6457 |
|
|
a such a position. */
|
6458 |
|
|
|
6459 |
|
|
modewidth = GET_MODE_BITSIZE (GET_MODE (x));
|
6460 |
|
|
if (modewidth + len >= pos)
|
6461 |
|
|
{
|
6462 |
|
|
enum machine_mode mode = GET_MODE (x);
|
6463 |
|
|
tem = gen_lowpart (mode, XEXP (x, 0));
|
6464 |
|
|
if (!tem || GET_CODE (tem) == CLOBBER)
|
6465 |
|
|
return x;
|
6466 |
|
|
tem = simplify_shift_const (NULL_RTX, ASHIFT, mode,
|
6467 |
|
|
tem, modewidth - pos - len);
|
6468 |
|
|
tem = simplify_shift_const (NULL_RTX, unsignedp ? LSHIFTRT : ASHIFTRT,
|
6469 |
|
|
mode, tem, modewidth - len);
|
6470 |
|
|
}
|
6471 |
|
|
else if (unsignedp && len < HOST_BITS_PER_WIDE_INT)
|
6472 |
|
|
tem = simplify_and_const_int (NULL_RTX, GET_MODE (x),
|
6473 |
|
|
simplify_shift_const (NULL_RTX, LSHIFTRT,
|
6474 |
|
|
GET_MODE (x),
|
6475 |
|
|
XEXP (x, 0), pos),
|
6476 |
|
|
((HOST_WIDE_INT) 1 << len) - 1);
|
6477 |
|
|
else
|
6478 |
|
|
/* Any other cases we can't handle. */
|
6479 |
|
|
return x;
|
6480 |
|
|
|
6481 |
|
|
/* If we couldn't do this for some reason, return the original
|
6482 |
|
|
expression. */
|
6483 |
|
|
if (GET_CODE (tem) == CLOBBER)
|
6484 |
|
|
return x;
|
6485 |
|
|
|
6486 |
|
|
return tem;
|
6487 |
|
|
}
|
6488 |
|
|
|
6489 |
|
|
/* X is a SET which contains an assignment of one object into
|
6490 |
|
|
a part of another (such as a bit-field assignment, STRICT_LOW_PART,
|
6491 |
|
|
or certain SUBREGS). If possible, convert it into a series of
|
6492 |
|
|
logical operations.
|
6493 |
|
|
|
6494 |
|
|
We half-heartedly support variable positions, but do not at all
|
6495 |
|
|
support variable lengths. */
|
6496 |
|
|
|
6497 |
|
|
static const_rtx
|
6498 |
|
|
expand_field_assignment (const_rtx x)
|
6499 |
|
|
{
|
6500 |
|
|
rtx inner;
|
6501 |
|
|
rtx pos; /* Always counts from low bit. */
|
6502 |
|
|
int len;
|
6503 |
|
|
rtx mask, cleared, masked;
|
6504 |
|
|
enum machine_mode compute_mode;
|
6505 |
|
|
|
6506 |
|
|
/* Loop until we find something we can't simplify. */
|
6507 |
|
|
while (1)
|
6508 |
|
|
{
|
6509 |
|
|
if (GET_CODE (SET_DEST (x)) == STRICT_LOW_PART
|
6510 |
|
|
&& GET_CODE (XEXP (SET_DEST (x), 0)) == SUBREG)
|
6511 |
|
|
{
|
6512 |
|
|
inner = SUBREG_REG (XEXP (SET_DEST (x), 0));
|
6513 |
|
|
len = GET_MODE_BITSIZE (GET_MODE (XEXP (SET_DEST (x), 0)));
|
6514 |
|
|
pos = GEN_INT (subreg_lsb (XEXP (SET_DEST (x), 0)));
|
6515 |
|
|
}
|
6516 |
|
|
else if (GET_CODE (SET_DEST (x)) == ZERO_EXTRACT
|
6517 |
|
|
&& CONST_INT_P (XEXP (SET_DEST (x), 1)))
|
6518 |
|
|
{
|
6519 |
|
|
inner = XEXP (SET_DEST (x), 0);
|
6520 |
|
|
len = INTVAL (XEXP (SET_DEST (x), 1));
|
6521 |
|
|
pos = XEXP (SET_DEST (x), 2);
|
6522 |
|
|
|
6523 |
|
|
/* A constant position should stay within the width of INNER. */
|
6524 |
|
|
if (CONST_INT_P (pos)
|
6525 |
|
|
&& INTVAL (pos) + len > GET_MODE_BITSIZE (GET_MODE (inner)))
|
6526 |
|
|
break;
|
6527 |
|
|
|
6528 |
|
|
if (BITS_BIG_ENDIAN)
|
6529 |
|
|
{
|
6530 |
|
|
if (CONST_INT_P (pos))
|
6531 |
|
|
pos = GEN_INT (GET_MODE_BITSIZE (GET_MODE (inner)) - len
|
6532 |
|
|
- INTVAL (pos));
|
6533 |
|
|
else if (GET_CODE (pos) == MINUS
|
6534 |
|
|
&& CONST_INT_P (XEXP (pos, 1))
|
6535 |
|
|
&& (INTVAL (XEXP (pos, 1))
|
6536 |
|
|
== GET_MODE_BITSIZE (GET_MODE (inner)) - len))
|
6537 |
|
|
/* If position is ADJUST - X, new position is X. */
|
6538 |
|
|
pos = XEXP (pos, 0);
|
6539 |
|
|
else
|
6540 |
|
|
pos = simplify_gen_binary (MINUS, GET_MODE (pos),
|
6541 |
|
|
GEN_INT (GET_MODE_BITSIZE (
|
6542 |
|
|
GET_MODE (inner))
|
6543 |
|
|
- len),
|
6544 |
|
|
pos);
|
6545 |
|
|
}
|
6546 |
|
|
}
|
6547 |
|
|
|
6548 |
|
|
/* A SUBREG between two modes that occupy the same numbers of words
|
6549 |
|
|
can be done by moving the SUBREG to the source. */
|
6550 |
|
|
else if (GET_CODE (SET_DEST (x)) == SUBREG
|
6551 |
|
|
/* We need SUBREGs to compute nonzero_bits properly. */
|
6552 |
|
|
&& nonzero_sign_valid
|
6553 |
|
|
&& (((GET_MODE_SIZE (GET_MODE (SET_DEST (x)))
|
6554 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
|
6555 |
|
|
== ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (SET_DEST (x))))
|
6556 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)))
|
6557 |
|
|
{
|
6558 |
|
|
x = gen_rtx_SET (VOIDmode, SUBREG_REG (SET_DEST (x)),
|
6559 |
|
|
gen_lowpart
|
6560 |
|
|
(GET_MODE (SUBREG_REG (SET_DEST (x))),
|
6561 |
|
|
SET_SRC (x)));
|
6562 |
|
|
continue;
|
6563 |
|
|
}
|
6564 |
|
|
else
|
6565 |
|
|
break;
|
6566 |
|
|
|
6567 |
|
|
while (GET_CODE (inner) == SUBREG && subreg_lowpart_p (inner))
|
6568 |
|
|
inner = SUBREG_REG (inner);
|
6569 |
|
|
|
6570 |
|
|
compute_mode = GET_MODE (inner);
|
6571 |
|
|
|
6572 |
|
|
/* Don't attempt bitwise arithmetic on non scalar integer modes. */
|
6573 |
|
|
if (! SCALAR_INT_MODE_P (compute_mode))
|
6574 |
|
|
{
|
6575 |
|
|
enum machine_mode imode;
|
6576 |
|
|
|
6577 |
|
|
/* Don't do anything for vector or complex integral types. */
|
6578 |
|
|
if (! FLOAT_MODE_P (compute_mode))
|
6579 |
|
|
break;
|
6580 |
|
|
|
6581 |
|
|
/* Try to find an integral mode to pun with. */
|
6582 |
|
|
imode = mode_for_size (GET_MODE_BITSIZE (compute_mode), MODE_INT, 0);
|
6583 |
|
|
if (imode == BLKmode)
|
6584 |
|
|
break;
|
6585 |
|
|
|
6586 |
|
|
compute_mode = imode;
|
6587 |
|
|
inner = gen_lowpart (imode, inner);
|
6588 |
|
|
}
|
6589 |
|
|
|
6590 |
|
|
/* Compute a mask of LEN bits, if we can do this on the host machine. */
|
6591 |
|
|
if (len >= HOST_BITS_PER_WIDE_INT)
|
6592 |
|
|
break;
|
6593 |
|
|
|
6594 |
|
|
/* Now compute the equivalent expression. Make a copy of INNER
|
6595 |
|
|
for the SET_DEST in case it is a MEM into which we will substitute;
|
6596 |
|
|
we don't want shared RTL in that case. */
|
6597 |
|
|
mask = GEN_INT (((HOST_WIDE_INT) 1 << len) - 1);
|
6598 |
|
|
cleared = simplify_gen_binary (AND, compute_mode,
|
6599 |
|
|
simplify_gen_unary (NOT, compute_mode,
|
6600 |
|
|
simplify_gen_binary (ASHIFT,
|
6601 |
|
|
compute_mode,
|
6602 |
|
|
mask, pos),
|
6603 |
|
|
compute_mode),
|
6604 |
|
|
inner);
|
6605 |
|
|
masked = simplify_gen_binary (ASHIFT, compute_mode,
|
6606 |
|
|
simplify_gen_binary (
|
6607 |
|
|
AND, compute_mode,
|
6608 |
|
|
gen_lowpart (compute_mode, SET_SRC (x)),
|
6609 |
|
|
mask),
|
6610 |
|
|
pos);
|
6611 |
|
|
|
6612 |
|
|
x = gen_rtx_SET (VOIDmode, copy_rtx (inner),
|
6613 |
|
|
simplify_gen_binary (IOR, compute_mode,
|
6614 |
|
|
cleared, masked));
|
6615 |
|
|
}
|
6616 |
|
|
|
6617 |
|
|
return x;
|
6618 |
|
|
}
|
6619 |
|
|
|
6620 |
|
|
/* Return an RTX for a reference to LEN bits of INNER. If POS_RTX is nonzero,
|
6621 |
|
|
it is an RTX that represents a variable starting position; otherwise,
|
6622 |
|
|
POS is the (constant) starting bit position (counted from the LSB).
|
6623 |
|
|
|
6624 |
|
|
UNSIGNEDP is nonzero for an unsigned reference and zero for a
|
6625 |
|
|
signed reference.
|
6626 |
|
|
|
6627 |
|
|
IN_DEST is nonzero if this is a reference in the destination of a
|
6628 |
|
|
SET. This is used when a ZERO_ or SIGN_EXTRACT isn't needed. If nonzero,
|
6629 |
|
|
a STRICT_LOW_PART will be used, if zero, ZERO_EXTEND or SIGN_EXTEND will
|
6630 |
|
|
be used.
|
6631 |
|
|
|
6632 |
|
|
IN_COMPARE is nonzero if we are in a COMPARE. This means that a
|
6633 |
|
|
ZERO_EXTRACT should be built even for bits starting at bit 0.
|
6634 |
|
|
|
6635 |
|
|
MODE is the desired mode of the result (if IN_DEST == 0).
|
6636 |
|
|
|
6637 |
|
|
The result is an RTX for the extraction or NULL_RTX if the target
|
6638 |
|
|
can't handle it. */
|
6639 |
|
|
|
6640 |
|
|
static rtx
|
6641 |
|
|
make_extraction (enum machine_mode mode, rtx inner, HOST_WIDE_INT pos,
|
6642 |
|
|
rtx pos_rtx, unsigned HOST_WIDE_INT len, int unsignedp,
|
6643 |
|
|
int in_dest, int in_compare)
|
6644 |
|
|
{
|
6645 |
|
|
/* This mode describes the size of the storage area
|
6646 |
|
|
to fetch the overall value from. Within that, we
|
6647 |
|
|
ignore the POS lowest bits, etc. */
|
6648 |
|
|
enum machine_mode is_mode = GET_MODE (inner);
|
6649 |
|
|
enum machine_mode inner_mode;
|
6650 |
|
|
enum machine_mode wanted_inner_mode;
|
6651 |
|
|
enum machine_mode wanted_inner_reg_mode = word_mode;
|
6652 |
|
|
enum machine_mode pos_mode = word_mode;
|
6653 |
|
|
enum machine_mode extraction_mode = word_mode;
|
6654 |
|
|
enum machine_mode tmode = mode_for_size (len, MODE_INT, 1);
|
6655 |
|
|
rtx new_rtx = 0;
|
6656 |
|
|
rtx orig_pos_rtx = pos_rtx;
|
6657 |
|
|
HOST_WIDE_INT orig_pos;
|
6658 |
|
|
|
6659 |
|
|
if (GET_CODE (inner) == SUBREG && subreg_lowpart_p (inner))
|
6660 |
|
|
{
|
6661 |
|
|
/* If going from (subreg:SI (mem:QI ...)) to (mem:QI ...),
|
6662 |
|
|
consider just the QI as the memory to extract from.
|
6663 |
|
|
The subreg adds or removes high bits; its mode is
|
6664 |
|
|
irrelevant to the meaning of this extraction,
|
6665 |
|
|
since POS and LEN count from the lsb. */
|
6666 |
|
|
if (MEM_P (SUBREG_REG (inner)))
|
6667 |
|
|
is_mode = GET_MODE (SUBREG_REG (inner));
|
6668 |
|
|
inner = SUBREG_REG (inner);
|
6669 |
|
|
}
|
6670 |
|
|
else if (GET_CODE (inner) == ASHIFT
|
6671 |
|
|
&& CONST_INT_P (XEXP (inner, 1))
|
6672 |
|
|
&& pos_rtx == 0 && pos == 0
|
6673 |
|
|
&& len > (unsigned HOST_WIDE_INT) INTVAL (XEXP (inner, 1)))
|
6674 |
|
|
{
|
6675 |
|
|
/* We're extracting the least significant bits of an rtx
|
6676 |
|
|
(ashift X (const_int C)), where LEN > C. Extract the
|
6677 |
|
|
least significant (LEN - C) bits of X, giving an rtx
|
6678 |
|
|
whose mode is MODE, then shift it left C times. */
|
6679 |
|
|
new_rtx = make_extraction (mode, XEXP (inner, 0),
|
6680 |
|
|
0, 0, len - INTVAL (XEXP (inner, 1)),
|
6681 |
|
|
unsignedp, in_dest, in_compare);
|
6682 |
|
|
if (new_rtx != 0)
|
6683 |
|
|
return gen_rtx_ASHIFT (mode, new_rtx, XEXP (inner, 1));
|
6684 |
|
|
}
|
6685 |
|
|
|
6686 |
|
|
inner_mode = GET_MODE (inner);
|
6687 |
|
|
|
6688 |
|
|
if (pos_rtx && CONST_INT_P (pos_rtx))
|
6689 |
|
|
pos = INTVAL (pos_rtx), pos_rtx = 0;
|
6690 |
|
|
|
6691 |
|
|
/* See if this can be done without an extraction. We never can if the
|
6692 |
|
|
width of the field is not the same as that of some integer mode. For
|
6693 |
|
|
registers, we can only avoid the extraction if the position is at the
|
6694 |
|
|
low-order bit and this is either not in the destination or we have the
|
6695 |
|
|
appropriate STRICT_LOW_PART operation available.
|
6696 |
|
|
|
6697 |
|
|
For MEM, we can avoid an extract if the field starts on an appropriate
|
6698 |
|
|
boundary and we can change the mode of the memory reference. */
|
6699 |
|
|
|
6700 |
|
|
if (tmode != BLKmode
|
6701 |
|
|
&& ((pos_rtx == 0 && (pos % BITS_PER_WORD) == 0
|
6702 |
|
|
&& !MEM_P (inner)
|
6703 |
|
|
&& (inner_mode == tmode
|
6704 |
|
|
|| !REG_P (inner)
|
6705 |
|
|
|| TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (tmode),
|
6706 |
|
|
GET_MODE_BITSIZE (inner_mode))
|
6707 |
|
|
|| reg_truncated_to_mode (tmode, inner))
|
6708 |
|
|
&& (! in_dest
|
6709 |
|
|
|| (REG_P (inner)
|
6710 |
|
|
&& have_insn_for (STRICT_LOW_PART, tmode))))
|
6711 |
|
|
|| (MEM_P (inner) && pos_rtx == 0
|
6712 |
|
|
&& (pos
|
6713 |
|
|
% (STRICT_ALIGNMENT ? GET_MODE_ALIGNMENT (tmode)
|
6714 |
|
|
: BITS_PER_UNIT)) == 0
|
6715 |
|
|
/* We can't do this if we are widening INNER_MODE (it
|
6716 |
|
|
may not be aligned, for one thing). */
|
6717 |
|
|
&& GET_MODE_BITSIZE (inner_mode) >= GET_MODE_BITSIZE (tmode)
|
6718 |
|
|
&& (inner_mode == tmode
|
6719 |
|
|
|| (! mode_dependent_address_p (XEXP (inner, 0))
|
6720 |
|
|
&& ! MEM_VOLATILE_P (inner))))))
|
6721 |
|
|
{
|
6722 |
|
|
/* If INNER is a MEM, make a new MEM that encompasses just the desired
|
6723 |
|
|
field. If the original and current mode are the same, we need not
|
6724 |
|
|
adjust the offset. Otherwise, we do if bytes big endian.
|
6725 |
|
|
|
6726 |
|
|
If INNER is not a MEM, get a piece consisting of just the field
|
6727 |
|
|
of interest (in this case POS % BITS_PER_WORD must be 0). */
|
6728 |
|
|
|
6729 |
|
|
if (MEM_P (inner))
|
6730 |
|
|
{
|
6731 |
|
|
HOST_WIDE_INT offset;
|
6732 |
|
|
|
6733 |
|
|
/* POS counts from lsb, but make OFFSET count in memory order. */
|
6734 |
|
|
if (BYTES_BIG_ENDIAN)
|
6735 |
|
|
offset = (GET_MODE_BITSIZE (is_mode) - len - pos) / BITS_PER_UNIT;
|
6736 |
|
|
else
|
6737 |
|
|
offset = pos / BITS_PER_UNIT;
|
6738 |
|
|
|
6739 |
|
|
new_rtx = adjust_address_nv (inner, tmode, offset);
|
6740 |
|
|
}
|
6741 |
|
|
else if (REG_P (inner))
|
6742 |
|
|
{
|
6743 |
|
|
if (tmode != inner_mode)
|
6744 |
|
|
{
|
6745 |
|
|
/* We can't call gen_lowpart in a DEST since we
|
6746 |
|
|
always want a SUBREG (see below) and it would sometimes
|
6747 |
|
|
return a new hard register. */
|
6748 |
|
|
if (pos || in_dest)
|
6749 |
|
|
{
|
6750 |
|
|
HOST_WIDE_INT final_word = pos / BITS_PER_WORD;
|
6751 |
|
|
|
6752 |
|
|
if (WORDS_BIG_ENDIAN
|
6753 |
|
|
&& GET_MODE_SIZE (inner_mode) > UNITS_PER_WORD)
|
6754 |
|
|
final_word = ((GET_MODE_SIZE (inner_mode)
|
6755 |
|
|
- GET_MODE_SIZE (tmode))
|
6756 |
|
|
/ UNITS_PER_WORD) - final_word;
|
6757 |
|
|
|
6758 |
|
|
final_word *= UNITS_PER_WORD;
|
6759 |
|
|
if (BYTES_BIG_ENDIAN &&
|
6760 |
|
|
GET_MODE_SIZE (inner_mode) > GET_MODE_SIZE (tmode))
|
6761 |
|
|
final_word += (GET_MODE_SIZE (inner_mode)
|
6762 |
|
|
- GET_MODE_SIZE (tmode)) % UNITS_PER_WORD;
|
6763 |
|
|
|
6764 |
|
|
/* Avoid creating invalid subregs, for example when
|
6765 |
|
|
simplifying (x>>32)&255. */
|
6766 |
|
|
if (!validate_subreg (tmode, inner_mode, inner, final_word))
|
6767 |
|
|
return NULL_RTX;
|
6768 |
|
|
|
6769 |
|
|
new_rtx = gen_rtx_SUBREG (tmode, inner, final_word);
|
6770 |
|
|
}
|
6771 |
|
|
else
|
6772 |
|
|
new_rtx = gen_lowpart (tmode, inner);
|
6773 |
|
|
}
|
6774 |
|
|
else
|
6775 |
|
|
new_rtx = inner;
|
6776 |
|
|
}
|
6777 |
|
|
else
|
6778 |
|
|
new_rtx = force_to_mode (inner, tmode,
|
6779 |
|
|
len >= HOST_BITS_PER_WIDE_INT
|
6780 |
|
|
? ~(unsigned HOST_WIDE_INT) 0
|
6781 |
|
|
: ((unsigned HOST_WIDE_INT) 1 << len) - 1,
|
6782 |
|
|
0);
|
6783 |
|
|
|
6784 |
|
|
/* If this extraction is going into the destination of a SET,
|
6785 |
|
|
make a STRICT_LOW_PART unless we made a MEM. */
|
6786 |
|
|
|
6787 |
|
|
if (in_dest)
|
6788 |
|
|
return (MEM_P (new_rtx) ? new_rtx
|
6789 |
|
|
: (GET_CODE (new_rtx) != SUBREG
|
6790 |
|
|
? gen_rtx_CLOBBER (tmode, const0_rtx)
|
6791 |
|
|
: gen_rtx_STRICT_LOW_PART (VOIDmode, new_rtx)));
|
6792 |
|
|
|
6793 |
|
|
if (mode == tmode)
|
6794 |
|
|
return new_rtx;
|
6795 |
|
|
|
6796 |
|
|
if (CONST_INT_P (new_rtx)
|
6797 |
|
|
|| GET_CODE (new_rtx) == CONST_DOUBLE)
|
6798 |
|
|
return simplify_unary_operation (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
|
6799 |
|
|
mode, new_rtx, tmode);
|
6800 |
|
|
|
6801 |
|
|
/* If we know that no extraneous bits are set, and that the high
|
6802 |
|
|
bit is not set, convert the extraction to the cheaper of
|
6803 |
|
|
sign and zero extension, that are equivalent in these cases. */
|
6804 |
|
|
if (flag_expensive_optimizations
|
6805 |
|
|
&& (GET_MODE_BITSIZE (tmode) <= HOST_BITS_PER_WIDE_INT
|
6806 |
|
|
&& ((nonzero_bits (new_rtx, tmode)
|
6807 |
|
|
& ~(((unsigned HOST_WIDE_INT)
|
6808 |
|
|
GET_MODE_MASK (tmode))
|
6809 |
|
|
>> 1))
|
6810 |
|
|
== 0)))
|
6811 |
|
|
{
|
6812 |
|
|
rtx temp = gen_rtx_ZERO_EXTEND (mode, new_rtx);
|
6813 |
|
|
rtx temp1 = gen_rtx_SIGN_EXTEND (mode, new_rtx);
|
6814 |
|
|
|
6815 |
|
|
/* Prefer ZERO_EXTENSION, since it gives more information to
|
6816 |
|
|
backends. */
|
6817 |
|
|
if (rtx_cost (temp, SET, optimize_this_for_speed_p)
|
6818 |
|
|
<= rtx_cost (temp1, SET, optimize_this_for_speed_p))
|
6819 |
|
|
return temp;
|
6820 |
|
|
return temp1;
|
6821 |
|
|
}
|
6822 |
|
|
|
6823 |
|
|
/* Otherwise, sign- or zero-extend unless we already are in the
|
6824 |
|
|
proper mode. */
|
6825 |
|
|
|
6826 |
|
|
return (gen_rtx_fmt_e (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
|
6827 |
|
|
mode, new_rtx));
|
6828 |
|
|
}
|
6829 |
|
|
|
6830 |
|
|
/* Unless this is a COMPARE or we have a funny memory reference,
|
6831 |
|
|
don't do anything with zero-extending field extracts starting at
|
6832 |
|
|
the low-order bit since they are simple AND operations. */
|
6833 |
|
|
if (pos_rtx == 0 && pos == 0 && ! in_dest
|
6834 |
|
|
&& ! in_compare && unsignedp)
|
6835 |
|
|
return 0;
|
6836 |
|
|
|
6837 |
|
|
/* Unless INNER is not MEM, reject this if we would be spanning bytes or
|
6838 |
|
|
if the position is not a constant and the length is not 1. In all
|
6839 |
|
|
other cases, we would only be going outside our object in cases when
|
6840 |
|
|
an original shift would have been undefined. */
|
6841 |
|
|
if (MEM_P (inner)
|
6842 |
|
|
&& ((pos_rtx == 0 && pos + len > GET_MODE_BITSIZE (is_mode))
|
6843 |
|
|
|| (pos_rtx != 0 && len != 1)))
|
6844 |
|
|
return 0;
|
6845 |
|
|
|
6846 |
|
|
/* Get the mode to use should INNER not be a MEM, the mode for the position,
|
6847 |
|
|
and the mode for the result. */
|
6848 |
|
|
if (in_dest && mode_for_extraction (EP_insv, -1) != MAX_MACHINE_MODE)
|
6849 |
|
|
{
|
6850 |
|
|
wanted_inner_reg_mode = mode_for_extraction (EP_insv, 0);
|
6851 |
|
|
pos_mode = mode_for_extraction (EP_insv, 2);
|
6852 |
|
|
extraction_mode = mode_for_extraction (EP_insv, 3);
|
6853 |
|
|
}
|
6854 |
|
|
|
6855 |
|
|
if (! in_dest && unsignedp
|
6856 |
|
|
&& mode_for_extraction (EP_extzv, -1) != MAX_MACHINE_MODE)
|
6857 |
|
|
{
|
6858 |
|
|
wanted_inner_reg_mode = mode_for_extraction (EP_extzv, 1);
|
6859 |
|
|
pos_mode = mode_for_extraction (EP_extzv, 3);
|
6860 |
|
|
extraction_mode = mode_for_extraction (EP_extzv, 0);
|
6861 |
|
|
}
|
6862 |
|
|
|
6863 |
|
|
if (! in_dest && ! unsignedp
|
6864 |
|
|
&& mode_for_extraction (EP_extv, -1) != MAX_MACHINE_MODE)
|
6865 |
|
|
{
|
6866 |
|
|
wanted_inner_reg_mode = mode_for_extraction (EP_extv, 1);
|
6867 |
|
|
pos_mode = mode_for_extraction (EP_extv, 3);
|
6868 |
|
|
extraction_mode = mode_for_extraction (EP_extv, 0);
|
6869 |
|
|
}
|
6870 |
|
|
|
6871 |
|
|
/* Never narrow an object, since that might not be safe. */
|
6872 |
|
|
|
6873 |
|
|
if (mode != VOIDmode
|
6874 |
|
|
&& GET_MODE_SIZE (extraction_mode) < GET_MODE_SIZE (mode))
|
6875 |
|
|
extraction_mode = mode;
|
6876 |
|
|
|
6877 |
|
|
if (pos_rtx && GET_MODE (pos_rtx) != VOIDmode
|
6878 |
|
|
&& GET_MODE_SIZE (pos_mode) < GET_MODE_SIZE (GET_MODE (pos_rtx)))
|
6879 |
|
|
pos_mode = GET_MODE (pos_rtx);
|
6880 |
|
|
|
6881 |
|
|
/* If this is not from memory, the desired mode is the preferred mode
|
6882 |
|
|
for an extraction pattern's first input operand, or word_mode if there
|
6883 |
|
|
is none. */
|
6884 |
|
|
if (!MEM_P (inner))
|
6885 |
|
|
wanted_inner_mode = wanted_inner_reg_mode;
|
6886 |
|
|
else
|
6887 |
|
|
{
|
6888 |
|
|
/* Be careful not to go beyond the extracted object and maintain the
|
6889 |
|
|
natural alignment of the memory. */
|
6890 |
|
|
wanted_inner_mode = smallest_mode_for_size (len, MODE_INT);
|
6891 |
|
|
while (pos % GET_MODE_BITSIZE (wanted_inner_mode) + len
|
6892 |
|
|
> GET_MODE_BITSIZE (wanted_inner_mode))
|
6893 |
|
|
{
|
6894 |
|
|
wanted_inner_mode = GET_MODE_WIDER_MODE (wanted_inner_mode);
|
6895 |
|
|
gcc_assert (wanted_inner_mode != VOIDmode);
|
6896 |
|
|
}
|
6897 |
|
|
|
6898 |
|
|
/* If we have to change the mode of memory and cannot, the desired mode
|
6899 |
|
|
is EXTRACTION_MODE. */
|
6900 |
|
|
if (inner_mode != wanted_inner_mode
|
6901 |
|
|
&& (mode_dependent_address_p (XEXP (inner, 0))
|
6902 |
|
|
|| MEM_VOLATILE_P (inner)
|
6903 |
|
|
|| pos_rtx))
|
6904 |
|
|
wanted_inner_mode = extraction_mode;
|
6905 |
|
|
}
|
6906 |
|
|
|
6907 |
|
|
orig_pos = pos;
|
6908 |
|
|
|
6909 |
|
|
if (BITS_BIG_ENDIAN)
|
6910 |
|
|
{
|
6911 |
|
|
/* POS is passed as if BITS_BIG_ENDIAN == 0, so we need to convert it to
|
6912 |
|
|
BITS_BIG_ENDIAN style. If position is constant, compute new
|
6913 |
|
|
position. Otherwise, build subtraction.
|
6914 |
|
|
Note that POS is relative to the mode of the original argument.
|
6915 |
|
|
If it's a MEM we need to recompute POS relative to that.
|
6916 |
|
|
However, if we're extracting from (or inserting into) a register,
|
6917 |
|
|
we want to recompute POS relative to wanted_inner_mode. */
|
6918 |
|
|
int width = (MEM_P (inner)
|
6919 |
|
|
? GET_MODE_BITSIZE (is_mode)
|
6920 |
|
|
: GET_MODE_BITSIZE (wanted_inner_mode));
|
6921 |
|
|
|
6922 |
|
|
if (pos_rtx == 0)
|
6923 |
|
|
pos = width - len - pos;
|
6924 |
|
|
else
|
6925 |
|
|
pos_rtx
|
6926 |
|
|
= gen_rtx_MINUS (GET_MODE (pos_rtx), GEN_INT (width - len), pos_rtx);
|
6927 |
|
|
/* POS may be less than 0 now, but we check for that below.
|
6928 |
|
|
Note that it can only be less than 0 if !MEM_P (inner). */
|
6929 |
|
|
}
|
6930 |
|
|
|
6931 |
|
|
/* If INNER has a wider mode, and this is a constant extraction, try to
|
6932 |
|
|
make it smaller and adjust the byte to point to the byte containing
|
6933 |
|
|
the value. */
|
6934 |
|
|
if (wanted_inner_mode != VOIDmode
|
6935 |
|
|
&& inner_mode != wanted_inner_mode
|
6936 |
|
|
&& ! pos_rtx
|
6937 |
|
|
&& GET_MODE_SIZE (wanted_inner_mode) < GET_MODE_SIZE (is_mode)
|
6938 |
|
|
&& MEM_P (inner)
|
6939 |
|
|
&& ! mode_dependent_address_p (XEXP (inner, 0))
|
6940 |
|
|
&& ! MEM_VOLATILE_P (inner))
|
6941 |
|
|
{
|
6942 |
|
|
int offset = 0;
|
6943 |
|
|
|
6944 |
|
|
/* The computations below will be correct if the machine is big
|
6945 |
|
|
endian in both bits and bytes or little endian in bits and bytes.
|
6946 |
|
|
If it is mixed, we must adjust. */
|
6947 |
|
|
|
6948 |
|
|
/* If bytes are big endian and we had a paradoxical SUBREG, we must
|
6949 |
|
|
adjust OFFSET to compensate. */
|
6950 |
|
|
if (BYTES_BIG_ENDIAN
|
6951 |
|
|
&& GET_MODE_SIZE (inner_mode) < GET_MODE_SIZE (is_mode))
|
6952 |
|
|
offset -= GET_MODE_SIZE (is_mode) - GET_MODE_SIZE (inner_mode);
|
6953 |
|
|
|
6954 |
|
|
/* We can now move to the desired byte. */
|
6955 |
|
|
offset += (pos / GET_MODE_BITSIZE (wanted_inner_mode))
|
6956 |
|
|
* GET_MODE_SIZE (wanted_inner_mode);
|
6957 |
|
|
pos %= GET_MODE_BITSIZE (wanted_inner_mode);
|
6958 |
|
|
|
6959 |
|
|
if (BYTES_BIG_ENDIAN != BITS_BIG_ENDIAN
|
6960 |
|
|
&& is_mode != wanted_inner_mode)
|
6961 |
|
|
offset = (GET_MODE_SIZE (is_mode)
|
6962 |
|
|
- GET_MODE_SIZE (wanted_inner_mode) - offset);
|
6963 |
|
|
|
6964 |
|
|
inner = adjust_address_nv (inner, wanted_inner_mode, offset);
|
6965 |
|
|
}
|
6966 |
|
|
|
6967 |
|
|
/* If INNER is not memory, get it into the proper mode. If we are changing
|
6968 |
|
|
its mode, POS must be a constant and smaller than the size of the new
|
6969 |
|
|
mode. */
|
6970 |
|
|
else if (!MEM_P (inner))
|
6971 |
|
|
{
|
6972 |
|
|
/* On the LHS, don't create paradoxical subregs implicitely truncating
|
6973 |
|
|
the register unless TRULY_NOOP_TRUNCATION. */
|
6974 |
|
|
if (in_dest
|
6975 |
|
|
&& !TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (GET_MODE (inner)),
|
6976 |
|
|
GET_MODE_BITSIZE (wanted_inner_mode)))
|
6977 |
|
|
return NULL_RTX;
|
6978 |
|
|
|
6979 |
|
|
if (GET_MODE (inner) != wanted_inner_mode
|
6980 |
|
|
&& (pos_rtx != 0
|
6981 |
|
|
|| orig_pos + len > GET_MODE_BITSIZE (wanted_inner_mode)))
|
6982 |
|
|
return NULL_RTX;
|
6983 |
|
|
|
6984 |
|
|
if (orig_pos < 0)
|
6985 |
|
|
return NULL_RTX;
|
6986 |
|
|
|
6987 |
|
|
inner = force_to_mode (inner, wanted_inner_mode,
|
6988 |
|
|
pos_rtx
|
6989 |
|
|
|| len + orig_pos >= HOST_BITS_PER_WIDE_INT
|
6990 |
|
|
? ~(unsigned HOST_WIDE_INT) 0
|
6991 |
|
|
: ((((unsigned HOST_WIDE_INT) 1 << len) - 1)
|
6992 |
|
|
<< orig_pos),
|
6993 |
|
|
0);
|
6994 |
|
|
}
|
6995 |
|
|
|
6996 |
|
|
/* Adjust mode of POS_RTX, if needed. If we want a wider mode, we
|
6997 |
|
|
have to zero extend. Otherwise, we can just use a SUBREG. */
|
6998 |
|
|
if (pos_rtx != 0
|
6999 |
|
|
&& GET_MODE_SIZE (pos_mode) > GET_MODE_SIZE (GET_MODE (pos_rtx)))
|
7000 |
|
|
{
|
7001 |
|
|
rtx temp = gen_rtx_ZERO_EXTEND (pos_mode, pos_rtx);
|
7002 |
|
|
|
7003 |
|
|
/* If we know that no extraneous bits are set, and that the high
|
7004 |
|
|
bit is not set, convert extraction to cheaper one - either
|
7005 |
|
|
SIGN_EXTENSION or ZERO_EXTENSION, that are equivalent in these
|
7006 |
|
|
cases. */
|
7007 |
|
|
if (flag_expensive_optimizations
|
7008 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (pos_rtx)) <= HOST_BITS_PER_WIDE_INT
|
7009 |
|
|
&& ((nonzero_bits (pos_rtx, GET_MODE (pos_rtx))
|
7010 |
|
|
& ~(((unsigned HOST_WIDE_INT)
|
7011 |
|
|
GET_MODE_MASK (GET_MODE (pos_rtx)))
|
7012 |
|
|
>> 1))
|
7013 |
|
|
== 0)))
|
7014 |
|
|
{
|
7015 |
|
|
rtx temp1 = gen_rtx_SIGN_EXTEND (pos_mode, pos_rtx);
|
7016 |
|
|
|
7017 |
|
|
/* Prefer ZERO_EXTENSION, since it gives more information to
|
7018 |
|
|
backends. */
|
7019 |
|
|
if (rtx_cost (temp1, SET, optimize_this_for_speed_p)
|
7020 |
|
|
< rtx_cost (temp, SET, optimize_this_for_speed_p))
|
7021 |
|
|
temp = temp1;
|
7022 |
|
|
}
|
7023 |
|
|
pos_rtx = temp;
|
7024 |
|
|
}
|
7025 |
|
|
else if (pos_rtx != 0
|
7026 |
|
|
&& GET_MODE_SIZE (pos_mode) < GET_MODE_SIZE (GET_MODE (pos_rtx)))
|
7027 |
|
|
pos_rtx = gen_lowpart (pos_mode, pos_rtx);
|
7028 |
|
|
|
7029 |
|
|
/* Make POS_RTX unless we already have it and it is correct. If we don't
|
7030 |
|
|
have a POS_RTX but we do have an ORIG_POS_RTX, the latter must
|
7031 |
|
|
be a CONST_INT. */
|
7032 |
|
|
if (pos_rtx == 0 && orig_pos_rtx != 0 && INTVAL (orig_pos_rtx) == pos)
|
7033 |
|
|
pos_rtx = orig_pos_rtx;
|
7034 |
|
|
|
7035 |
|
|
else if (pos_rtx == 0)
|
7036 |
|
|
pos_rtx = GEN_INT (pos);
|
7037 |
|
|
|
7038 |
|
|
/* Make the required operation. See if we can use existing rtx. */
|
7039 |
|
|
new_rtx = gen_rtx_fmt_eee (unsignedp ? ZERO_EXTRACT : SIGN_EXTRACT,
|
7040 |
|
|
extraction_mode, inner, GEN_INT (len), pos_rtx);
|
7041 |
|
|
if (! in_dest)
|
7042 |
|
|
new_rtx = gen_lowpart (mode, new_rtx);
|
7043 |
|
|
|
7044 |
|
|
return new_rtx;
|
7045 |
|
|
}
|
7046 |
|
|
|
7047 |
|
|
/* See if X contains an ASHIFT of COUNT or more bits that can be commuted
|
7048 |
|
|
with any other operations in X. Return X without that shift if so. */
|
7049 |
|
|
|
7050 |
|
|
static rtx
|
7051 |
|
|
extract_left_shift (rtx x, int count)
|
7052 |
|
|
{
|
7053 |
|
|
enum rtx_code code = GET_CODE (x);
|
7054 |
|
|
enum machine_mode mode = GET_MODE (x);
|
7055 |
|
|
rtx tem;
|
7056 |
|
|
|
7057 |
|
|
switch (code)
|
7058 |
|
|
{
|
7059 |
|
|
case ASHIFT:
|
7060 |
|
|
/* This is the shift itself. If it is wide enough, we will return
|
7061 |
|
|
either the value being shifted if the shift count is equal to
|
7062 |
|
|
COUNT or a shift for the difference. */
|
7063 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7064 |
|
|
&& INTVAL (XEXP (x, 1)) >= count)
|
7065 |
|
|
return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (x, 0),
|
7066 |
|
|
INTVAL (XEXP (x, 1)) - count);
|
7067 |
|
|
break;
|
7068 |
|
|
|
7069 |
|
|
case NEG: case NOT:
|
7070 |
|
|
if ((tem = extract_left_shift (XEXP (x, 0), count)) != 0)
|
7071 |
|
|
return simplify_gen_unary (code, mode, tem, mode);
|
7072 |
|
|
|
7073 |
|
|
break;
|
7074 |
|
|
|
7075 |
|
|
case PLUS: case IOR: case XOR: case AND:
|
7076 |
|
|
/* If we can safely shift this constant and we find the inner shift,
|
7077 |
|
|
make a new operation. */
|
7078 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7079 |
|
|
&& (INTVAL (XEXP (x, 1)) & ((((HOST_WIDE_INT) 1 << count)) - 1)) == 0
|
7080 |
|
|
&& (tem = extract_left_shift (XEXP (x, 0), count)) != 0)
|
7081 |
|
|
return simplify_gen_binary (code, mode, tem,
|
7082 |
|
|
GEN_INT (INTVAL (XEXP (x, 1)) >> count));
|
7083 |
|
|
|
7084 |
|
|
break;
|
7085 |
|
|
|
7086 |
|
|
default:
|
7087 |
|
|
break;
|
7088 |
|
|
}
|
7089 |
|
|
|
7090 |
|
|
return 0;
|
7091 |
|
|
}
|
7092 |
|
|
|
7093 |
|
|
/* Look at the expression rooted at X. Look for expressions
|
7094 |
|
|
equivalent to ZERO_EXTRACT, SIGN_EXTRACT, ZERO_EXTEND, SIGN_EXTEND.
|
7095 |
|
|
Form these expressions.
|
7096 |
|
|
|
7097 |
|
|
Return the new rtx, usually just X.
|
7098 |
|
|
|
7099 |
|
|
Also, for machines like the VAX that don't have logical shift insns,
|
7100 |
|
|
try to convert logical to arithmetic shift operations in cases where
|
7101 |
|
|
they are equivalent. This undoes the canonicalizations to logical
|
7102 |
|
|
shifts done elsewhere.
|
7103 |
|
|
|
7104 |
|
|
We try, as much as possible, to re-use rtl expressions to save memory.
|
7105 |
|
|
|
7106 |
|
|
IN_CODE says what kind of expression we are processing. Normally, it is
|
7107 |
|
|
SET. In a memory address (inside a MEM, PLUS or minus, the latter two
|
7108 |
|
|
being kludges), it is MEM. When processing the arguments of a comparison
|
7109 |
|
|
or a COMPARE against zero, it is COMPARE. */
|
7110 |
|
|
|
7111 |
|
|
static rtx
|
7112 |
|
|
make_compound_operation (rtx x, enum rtx_code in_code)
|
7113 |
|
|
{
|
7114 |
|
|
enum rtx_code code = GET_CODE (x);
|
7115 |
|
|
enum machine_mode mode = GET_MODE (x);
|
7116 |
|
|
int mode_width = GET_MODE_BITSIZE (mode);
|
7117 |
|
|
rtx rhs, lhs;
|
7118 |
|
|
enum rtx_code next_code;
|
7119 |
|
|
int i, j;
|
7120 |
|
|
rtx new_rtx = 0;
|
7121 |
|
|
rtx tem;
|
7122 |
|
|
const char *fmt;
|
7123 |
|
|
|
7124 |
|
|
/* Select the code to be used in recursive calls. Once we are inside an
|
7125 |
|
|
address, we stay there. If we have a comparison, set to COMPARE,
|
7126 |
|
|
but once inside, go back to our default of SET. */
|
7127 |
|
|
|
7128 |
|
|
next_code = (code == MEM || code == PLUS || code == MINUS ? MEM
|
7129 |
|
|
: ((code == COMPARE || COMPARISON_P (x))
|
7130 |
|
|
&& XEXP (x, 1) == const0_rtx) ? COMPARE
|
7131 |
|
|
: in_code == COMPARE ? SET : in_code);
|
7132 |
|
|
|
7133 |
|
|
/* Process depending on the code of this operation. If NEW is set
|
7134 |
|
|
nonzero, it will be returned. */
|
7135 |
|
|
|
7136 |
|
|
switch (code)
|
7137 |
|
|
{
|
7138 |
|
|
case ASHIFT:
|
7139 |
|
|
/* Convert shifts by constants into multiplications if inside
|
7140 |
|
|
an address. */
|
7141 |
|
|
if (in_code == MEM && CONST_INT_P (XEXP (x, 1))
|
7142 |
|
|
&& INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
|
7143 |
|
|
&& INTVAL (XEXP (x, 1)) >= 0)
|
7144 |
|
|
{
|
7145 |
|
|
new_rtx = make_compound_operation (XEXP (x, 0), next_code);
|
7146 |
|
|
new_rtx = gen_rtx_MULT (mode, new_rtx,
|
7147 |
|
|
GEN_INT ((HOST_WIDE_INT) 1
|
7148 |
|
|
<< INTVAL (XEXP (x, 1))));
|
7149 |
|
|
}
|
7150 |
|
|
break;
|
7151 |
|
|
|
7152 |
|
|
case AND:
|
7153 |
|
|
/* If the second operand is not a constant, we can't do anything
|
7154 |
|
|
with it. */
|
7155 |
|
|
if (!CONST_INT_P (XEXP (x, 1)))
|
7156 |
|
|
break;
|
7157 |
|
|
|
7158 |
|
|
/* If the constant is a power of two minus one and the first operand
|
7159 |
|
|
is a logical right shift, make an extraction. */
|
7160 |
|
|
if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
|
7161 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)) + 1)) >= 0)
|
7162 |
|
|
{
|
7163 |
|
|
new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
|
7164 |
|
|
new_rtx = make_extraction (mode, new_rtx, 0, XEXP (XEXP (x, 0), 1), i, 1,
|
7165 |
|
|
0, in_code == COMPARE);
|
7166 |
|
|
}
|
7167 |
|
|
|
7168 |
|
|
/* Same as previous, but for (subreg (lshiftrt ...)) in first op. */
|
7169 |
|
|
else if (GET_CODE (XEXP (x, 0)) == SUBREG
|
7170 |
|
|
&& subreg_lowpart_p (XEXP (x, 0))
|
7171 |
|
|
&& GET_CODE (SUBREG_REG (XEXP (x, 0))) == LSHIFTRT
|
7172 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)) + 1)) >= 0)
|
7173 |
|
|
{
|
7174 |
|
|
new_rtx = make_compound_operation (XEXP (SUBREG_REG (XEXP (x, 0)), 0),
|
7175 |
|
|
next_code);
|
7176 |
|
|
new_rtx = make_extraction (GET_MODE (SUBREG_REG (XEXP (x, 0))), new_rtx, 0,
|
7177 |
|
|
XEXP (SUBREG_REG (XEXP (x, 0)), 1), i, 1,
|
7178 |
|
|
0, in_code == COMPARE);
|
7179 |
|
|
}
|
7180 |
|
|
/* Same as previous, but for (xor/ior (lshiftrt...) (lshiftrt...)). */
|
7181 |
|
|
else if ((GET_CODE (XEXP (x, 0)) == XOR
|
7182 |
|
|
|| GET_CODE (XEXP (x, 0)) == IOR)
|
7183 |
|
|
&& GET_CODE (XEXP (XEXP (x, 0), 0)) == LSHIFTRT
|
7184 |
|
|
&& GET_CODE (XEXP (XEXP (x, 0), 1)) == LSHIFTRT
|
7185 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)) + 1)) >= 0)
|
7186 |
|
|
{
|
7187 |
|
|
/* Apply the distributive law, and then try to make extractions. */
|
7188 |
|
|
new_rtx = gen_rtx_fmt_ee (GET_CODE (XEXP (x, 0)), mode,
|
7189 |
|
|
gen_rtx_AND (mode, XEXP (XEXP (x, 0), 0),
|
7190 |
|
|
XEXP (x, 1)),
|
7191 |
|
|
gen_rtx_AND (mode, XEXP (XEXP (x, 0), 1),
|
7192 |
|
|
XEXP (x, 1)));
|
7193 |
|
|
new_rtx = make_compound_operation (new_rtx, in_code);
|
7194 |
|
|
}
|
7195 |
|
|
|
7196 |
|
|
/* If we are have (and (rotate X C) M) and C is larger than the number
|
7197 |
|
|
of bits in M, this is an extraction. */
|
7198 |
|
|
|
7199 |
|
|
else if (GET_CODE (XEXP (x, 0)) == ROTATE
|
7200 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
7201 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)) + 1)) >= 0
|
7202 |
|
|
&& i <= INTVAL (XEXP (XEXP (x, 0), 1)))
|
7203 |
|
|
{
|
7204 |
|
|
new_rtx = make_compound_operation (XEXP (XEXP (x, 0), 0), next_code);
|
7205 |
|
|
new_rtx = make_extraction (mode, new_rtx,
|
7206 |
|
|
(GET_MODE_BITSIZE (mode)
|
7207 |
|
|
- INTVAL (XEXP (XEXP (x, 0), 1))),
|
7208 |
|
|
NULL_RTX, i, 1, 0, in_code == COMPARE);
|
7209 |
|
|
}
|
7210 |
|
|
|
7211 |
|
|
/* On machines without logical shifts, if the operand of the AND is
|
7212 |
|
|
a logical shift and our mask turns off all the propagated sign
|
7213 |
|
|
bits, we can replace the logical shift with an arithmetic shift. */
|
7214 |
|
|
else if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
|
7215 |
|
|
&& !have_insn_for (LSHIFTRT, mode)
|
7216 |
|
|
&& have_insn_for (ASHIFTRT, mode)
|
7217 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
7218 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
|
7219 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
|
7220 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT)
|
7221 |
|
|
{
|
7222 |
|
|
unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
|
7223 |
|
|
|
7224 |
|
|
mask >>= INTVAL (XEXP (XEXP (x, 0), 1));
|
7225 |
|
|
if ((INTVAL (XEXP (x, 1)) & ~mask) == 0)
|
7226 |
|
|
SUBST (XEXP (x, 0),
|
7227 |
|
|
gen_rtx_ASHIFTRT (mode,
|
7228 |
|
|
make_compound_operation
|
7229 |
|
|
(XEXP (XEXP (x, 0), 0), next_code),
|
7230 |
|
|
XEXP (XEXP (x, 0), 1)));
|
7231 |
|
|
}
|
7232 |
|
|
|
7233 |
|
|
/* If the constant is one less than a power of two, this might be
|
7234 |
|
|
representable by an extraction even if no shift is present.
|
7235 |
|
|
If it doesn't end up being a ZERO_EXTEND, we will ignore it unless
|
7236 |
|
|
we are in a COMPARE. */
|
7237 |
|
|
else if ((i = exact_log2 (INTVAL (XEXP (x, 1)) + 1)) >= 0)
|
7238 |
|
|
new_rtx = make_extraction (mode,
|
7239 |
|
|
make_compound_operation (XEXP (x, 0),
|
7240 |
|
|
next_code),
|
7241 |
|
|
0, NULL_RTX, i, 1, 0, in_code == COMPARE);
|
7242 |
|
|
|
7243 |
|
|
/* If we are in a comparison and this is an AND with a power of two,
|
7244 |
|
|
convert this into the appropriate bit extract. */
|
7245 |
|
|
else if (in_code == COMPARE
|
7246 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (x, 1)))) >= 0)
|
7247 |
|
|
new_rtx = make_extraction (mode,
|
7248 |
|
|
make_compound_operation (XEXP (x, 0),
|
7249 |
|
|
next_code),
|
7250 |
|
|
i, NULL_RTX, 1, 1, 0, 1);
|
7251 |
|
|
|
7252 |
|
|
break;
|
7253 |
|
|
|
7254 |
|
|
case LSHIFTRT:
|
7255 |
|
|
/* If the sign bit is known to be zero, replace this with an
|
7256 |
|
|
arithmetic shift. */
|
7257 |
|
|
if (have_insn_for (ASHIFTRT, mode)
|
7258 |
|
|
&& ! have_insn_for (LSHIFTRT, mode)
|
7259 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
7260 |
|
|
&& (nonzero_bits (XEXP (x, 0), mode) & (1 << (mode_width - 1))) == 0)
|
7261 |
|
|
{
|
7262 |
|
|
new_rtx = gen_rtx_ASHIFTRT (mode,
|
7263 |
|
|
make_compound_operation (XEXP (x, 0),
|
7264 |
|
|
next_code),
|
7265 |
|
|
XEXP (x, 1));
|
7266 |
|
|
break;
|
7267 |
|
|
}
|
7268 |
|
|
|
7269 |
|
|
/* ... fall through ... */
|
7270 |
|
|
|
7271 |
|
|
case ASHIFTRT:
|
7272 |
|
|
lhs = XEXP (x, 0);
|
7273 |
|
|
rhs = XEXP (x, 1);
|
7274 |
|
|
|
7275 |
|
|
/* If we have (ashiftrt (ashift foo C1) C2) with C2 >= C1,
|
7276 |
|
|
this is a SIGN_EXTRACT. */
|
7277 |
|
|
if (CONST_INT_P (rhs)
|
7278 |
|
|
&& GET_CODE (lhs) == ASHIFT
|
7279 |
|
|
&& CONST_INT_P (XEXP (lhs, 1))
|
7280 |
|
|
&& INTVAL (rhs) >= INTVAL (XEXP (lhs, 1))
|
7281 |
|
|
&& INTVAL (rhs) < mode_width)
|
7282 |
|
|
{
|
7283 |
|
|
new_rtx = make_compound_operation (XEXP (lhs, 0), next_code);
|
7284 |
|
|
new_rtx = make_extraction (mode, new_rtx,
|
7285 |
|
|
INTVAL (rhs) - INTVAL (XEXP (lhs, 1)),
|
7286 |
|
|
NULL_RTX, mode_width - INTVAL (rhs),
|
7287 |
|
|
code == LSHIFTRT, 0, in_code == COMPARE);
|
7288 |
|
|
break;
|
7289 |
|
|
}
|
7290 |
|
|
|
7291 |
|
|
/* See if we have operations between an ASHIFTRT and an ASHIFT.
|
7292 |
|
|
If so, try to merge the shifts into a SIGN_EXTEND. We could
|
7293 |
|
|
also do this for some cases of SIGN_EXTRACT, but it doesn't
|
7294 |
|
|
seem worth the effort; the case checked for occurs on Alpha. */
|
7295 |
|
|
|
7296 |
|
|
if (!OBJECT_P (lhs)
|
7297 |
|
|
&& ! (GET_CODE (lhs) == SUBREG
|
7298 |
|
|
&& (OBJECT_P (SUBREG_REG (lhs))))
|
7299 |
|
|
&& CONST_INT_P (rhs)
|
7300 |
|
|
&& INTVAL (rhs) < HOST_BITS_PER_WIDE_INT
|
7301 |
|
|
&& INTVAL (rhs) < mode_width
|
7302 |
|
|
&& (new_rtx = extract_left_shift (lhs, INTVAL (rhs))) != 0)
|
7303 |
|
|
new_rtx = make_extraction (mode, make_compound_operation (new_rtx, next_code),
|
7304 |
|
|
0, NULL_RTX, mode_width - INTVAL (rhs),
|
7305 |
|
|
code == LSHIFTRT, 0, in_code == COMPARE);
|
7306 |
|
|
|
7307 |
|
|
break;
|
7308 |
|
|
|
7309 |
|
|
case SUBREG:
|
7310 |
|
|
/* Call ourselves recursively on the inner expression. If we are
|
7311 |
|
|
narrowing the object and it has a different RTL code from
|
7312 |
|
|
what it originally did, do this SUBREG as a force_to_mode. */
|
7313 |
|
|
|
7314 |
|
|
tem = make_compound_operation (SUBREG_REG (x), in_code);
|
7315 |
|
|
|
7316 |
|
|
{
|
7317 |
|
|
rtx simplified = simplify_subreg (mode, tem, GET_MODE (SUBREG_REG (x)),
|
7318 |
|
|
SUBREG_BYTE (x));
|
7319 |
|
|
|
7320 |
|
|
if (simplified)
|
7321 |
|
|
tem = simplified;
|
7322 |
|
|
|
7323 |
|
|
if (GET_CODE (tem) != GET_CODE (SUBREG_REG (x))
|
7324 |
|
|
&& GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (SUBREG_REG (x)))
|
7325 |
|
|
&& subreg_lowpart_p (x))
|
7326 |
|
|
{
|
7327 |
|
|
rtx newer = force_to_mode (tem, mode, ~(HOST_WIDE_INT) 0,
|
7328 |
|
|
0);
|
7329 |
|
|
|
7330 |
|
|
/* If we have something other than a SUBREG, we might have
|
7331 |
|
|
done an expansion, so rerun ourselves. */
|
7332 |
|
|
if (GET_CODE (newer) != SUBREG)
|
7333 |
|
|
newer = make_compound_operation (newer, in_code);
|
7334 |
|
|
|
7335 |
|
|
/* force_to_mode can expand compounds. If it just re-expanded the
|
7336 |
|
|
compound use gen_lowpart instead to convert to the desired
|
7337 |
|
|
mode. */
|
7338 |
|
|
if (rtx_equal_p (newer, x))
|
7339 |
|
|
return gen_lowpart (GET_MODE (x), tem);
|
7340 |
|
|
|
7341 |
|
|
return newer;
|
7342 |
|
|
}
|
7343 |
|
|
|
7344 |
|
|
if (simplified)
|
7345 |
|
|
return tem;
|
7346 |
|
|
}
|
7347 |
|
|
break;
|
7348 |
|
|
|
7349 |
|
|
default:
|
7350 |
|
|
break;
|
7351 |
|
|
}
|
7352 |
|
|
|
7353 |
|
|
if (new_rtx)
|
7354 |
|
|
{
|
7355 |
|
|
x = gen_lowpart (mode, new_rtx);
|
7356 |
|
|
code = GET_CODE (x);
|
7357 |
|
|
}
|
7358 |
|
|
|
7359 |
|
|
/* Now recursively process each operand of this operation. */
|
7360 |
|
|
fmt = GET_RTX_FORMAT (code);
|
7361 |
|
|
for (i = 0; i < GET_RTX_LENGTH (code); i++)
|
7362 |
|
|
if (fmt[i] == 'e')
|
7363 |
|
|
{
|
7364 |
|
|
new_rtx = make_compound_operation (XEXP (x, i), next_code);
|
7365 |
|
|
SUBST (XEXP (x, i), new_rtx);
|
7366 |
|
|
}
|
7367 |
|
|
else if (fmt[i] == 'E')
|
7368 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
7369 |
|
|
{
|
7370 |
|
|
new_rtx = make_compound_operation (XVECEXP (x, i, j), next_code);
|
7371 |
|
|
SUBST (XVECEXP (x, i, j), new_rtx);
|
7372 |
|
|
}
|
7373 |
|
|
|
7374 |
|
|
/* If this is a commutative operation, the changes to the operands
|
7375 |
|
|
may have made it noncanonical. */
|
7376 |
|
|
if (COMMUTATIVE_ARITH_P (x)
|
7377 |
|
|
&& swap_commutative_operands_p (XEXP (x, 0), XEXP (x, 1)))
|
7378 |
|
|
{
|
7379 |
|
|
tem = XEXP (x, 0);
|
7380 |
|
|
SUBST (XEXP (x, 0), XEXP (x, 1));
|
7381 |
|
|
SUBST (XEXP (x, 1), tem);
|
7382 |
|
|
}
|
7383 |
|
|
|
7384 |
|
|
return x;
|
7385 |
|
|
}
|
7386 |
|
|
|
7387 |
|
|
/* Given M see if it is a value that would select a field of bits
|
7388 |
|
|
within an item, but not the entire word. Return -1 if not.
|
7389 |
|
|
Otherwise, return the starting position of the field, where 0 is the
|
7390 |
|
|
low-order bit.
|
7391 |
|
|
|
7392 |
|
|
*PLEN is set to the length of the field. */
|
7393 |
|
|
|
7394 |
|
|
static int
|
7395 |
|
|
get_pos_from_mask (unsigned HOST_WIDE_INT m, unsigned HOST_WIDE_INT *plen)
|
7396 |
|
|
{
|
7397 |
|
|
/* Get the bit number of the first 1 bit from the right, -1 if none. */
|
7398 |
|
|
int pos = exact_log2 (m & -m);
|
7399 |
|
|
int len = 0;
|
7400 |
|
|
|
7401 |
|
|
if (pos >= 0)
|
7402 |
|
|
/* Now shift off the low-order zero bits and see if we have a
|
7403 |
|
|
power of two minus 1. */
|
7404 |
|
|
len = exact_log2 ((m >> pos) + 1);
|
7405 |
|
|
|
7406 |
|
|
if (len <= 0)
|
7407 |
|
|
pos = -1;
|
7408 |
|
|
|
7409 |
|
|
*plen = len;
|
7410 |
|
|
return pos;
|
7411 |
|
|
}
|
7412 |
|
|
|
7413 |
|
|
/* If X refers to a register that equals REG in value, replace these
|
7414 |
|
|
references with REG. */
|
7415 |
|
|
static rtx
|
7416 |
|
|
canon_reg_for_combine (rtx x, rtx reg)
|
7417 |
|
|
{
|
7418 |
|
|
rtx op0, op1, op2;
|
7419 |
|
|
const char *fmt;
|
7420 |
|
|
int i;
|
7421 |
|
|
bool copied;
|
7422 |
|
|
|
7423 |
|
|
enum rtx_code code = GET_CODE (x);
|
7424 |
|
|
switch (GET_RTX_CLASS (code))
|
7425 |
|
|
{
|
7426 |
|
|
case RTX_UNARY:
|
7427 |
|
|
op0 = canon_reg_for_combine (XEXP (x, 0), reg);
|
7428 |
|
|
if (op0 != XEXP (x, 0))
|
7429 |
|
|
return simplify_gen_unary (GET_CODE (x), GET_MODE (x), op0,
|
7430 |
|
|
GET_MODE (reg));
|
7431 |
|
|
break;
|
7432 |
|
|
|
7433 |
|
|
case RTX_BIN_ARITH:
|
7434 |
|
|
case RTX_COMM_ARITH:
|
7435 |
|
|
op0 = canon_reg_for_combine (XEXP (x, 0), reg);
|
7436 |
|
|
op1 = canon_reg_for_combine (XEXP (x, 1), reg);
|
7437 |
|
|
if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
|
7438 |
|
|
return simplify_gen_binary (GET_CODE (x), GET_MODE (x), op0, op1);
|
7439 |
|
|
break;
|
7440 |
|
|
|
7441 |
|
|
case RTX_COMPARE:
|
7442 |
|
|
case RTX_COMM_COMPARE:
|
7443 |
|
|
op0 = canon_reg_for_combine (XEXP (x, 0), reg);
|
7444 |
|
|
op1 = canon_reg_for_combine (XEXP (x, 1), reg);
|
7445 |
|
|
if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
|
7446 |
|
|
return simplify_gen_relational (GET_CODE (x), GET_MODE (x),
|
7447 |
|
|
GET_MODE (op0), op0, op1);
|
7448 |
|
|
break;
|
7449 |
|
|
|
7450 |
|
|
case RTX_TERNARY:
|
7451 |
|
|
case RTX_BITFIELD_OPS:
|
7452 |
|
|
op0 = canon_reg_for_combine (XEXP (x, 0), reg);
|
7453 |
|
|
op1 = canon_reg_for_combine (XEXP (x, 1), reg);
|
7454 |
|
|
op2 = canon_reg_for_combine (XEXP (x, 2), reg);
|
7455 |
|
|
if (op0 != XEXP (x, 0) || op1 != XEXP (x, 1) || op2 != XEXP (x, 2))
|
7456 |
|
|
return simplify_gen_ternary (GET_CODE (x), GET_MODE (x),
|
7457 |
|
|
GET_MODE (op0), op0, op1, op2);
|
7458 |
|
|
|
7459 |
|
|
case RTX_OBJ:
|
7460 |
|
|
if (REG_P (x))
|
7461 |
|
|
{
|
7462 |
|
|
if (rtx_equal_p (get_last_value (reg), x)
|
7463 |
|
|
|| rtx_equal_p (reg, get_last_value (x)))
|
7464 |
|
|
return reg;
|
7465 |
|
|
else
|
7466 |
|
|
break;
|
7467 |
|
|
}
|
7468 |
|
|
|
7469 |
|
|
/* fall through */
|
7470 |
|
|
|
7471 |
|
|
default:
|
7472 |
|
|
fmt = GET_RTX_FORMAT (code);
|
7473 |
|
|
copied = false;
|
7474 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
7475 |
|
|
if (fmt[i] == 'e')
|
7476 |
|
|
{
|
7477 |
|
|
rtx op = canon_reg_for_combine (XEXP (x, i), reg);
|
7478 |
|
|
if (op != XEXP (x, i))
|
7479 |
|
|
{
|
7480 |
|
|
if (!copied)
|
7481 |
|
|
{
|
7482 |
|
|
copied = true;
|
7483 |
|
|
x = copy_rtx (x);
|
7484 |
|
|
}
|
7485 |
|
|
XEXP (x, i) = op;
|
7486 |
|
|
}
|
7487 |
|
|
}
|
7488 |
|
|
else if (fmt[i] == 'E')
|
7489 |
|
|
{
|
7490 |
|
|
int j;
|
7491 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
7492 |
|
|
{
|
7493 |
|
|
rtx op = canon_reg_for_combine (XVECEXP (x, i, j), reg);
|
7494 |
|
|
if (op != XVECEXP (x, i, j))
|
7495 |
|
|
{
|
7496 |
|
|
if (!copied)
|
7497 |
|
|
{
|
7498 |
|
|
copied = true;
|
7499 |
|
|
x = copy_rtx (x);
|
7500 |
|
|
}
|
7501 |
|
|
XVECEXP (x, i, j) = op;
|
7502 |
|
|
}
|
7503 |
|
|
}
|
7504 |
|
|
}
|
7505 |
|
|
|
7506 |
|
|
break;
|
7507 |
|
|
}
|
7508 |
|
|
|
7509 |
|
|
return x;
|
7510 |
|
|
}
|
7511 |
|
|
|
7512 |
|
|
/* Return X converted to MODE. If the value is already truncated to
|
7513 |
|
|
MODE we can just return a subreg even though in the general case we
|
7514 |
|
|
would need an explicit truncation. */
|
7515 |
|
|
|
7516 |
|
|
static rtx
|
7517 |
|
|
gen_lowpart_or_truncate (enum machine_mode mode, rtx x)
|
7518 |
|
|
{
|
7519 |
|
|
if (!CONST_INT_P (x)
|
7520 |
|
|
&& GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (x))
|
7521 |
|
|
&& !TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (mode),
|
7522 |
|
|
GET_MODE_BITSIZE (GET_MODE (x)))
|
7523 |
|
|
&& !(REG_P (x) && reg_truncated_to_mode (mode, x)))
|
7524 |
|
|
{
|
7525 |
|
|
/* Bit-cast X into an integer mode. */
|
7526 |
|
|
if (!SCALAR_INT_MODE_P (GET_MODE (x)))
|
7527 |
|
|
x = gen_lowpart (int_mode_for_mode (GET_MODE (x)), x);
|
7528 |
|
|
x = simplify_gen_unary (TRUNCATE, int_mode_for_mode (mode),
|
7529 |
|
|
x, GET_MODE (x));
|
7530 |
|
|
}
|
7531 |
|
|
|
7532 |
|
|
return gen_lowpart (mode, x);
|
7533 |
|
|
}
|
7534 |
|
|
|
7535 |
|
|
/* See if X can be simplified knowing that we will only refer to it in
|
7536 |
|
|
MODE and will only refer to those bits that are nonzero in MASK.
|
7537 |
|
|
If other bits are being computed or if masking operations are done
|
7538 |
|
|
that select a superset of the bits in MASK, they can sometimes be
|
7539 |
|
|
ignored.
|
7540 |
|
|
|
7541 |
|
|
Return a possibly simplified expression, but always convert X to
|
7542 |
|
|
MODE. If X is a CONST_INT, AND the CONST_INT with MASK.
|
7543 |
|
|
|
7544 |
|
|
If JUST_SELECT is nonzero, don't optimize by noticing that bits in MASK
|
7545 |
|
|
are all off in X. This is used when X will be complemented, by either
|
7546 |
|
|
NOT, NEG, or XOR. */
|
7547 |
|
|
|
7548 |
|
|
static rtx
|
7549 |
|
|
force_to_mode (rtx x, enum machine_mode mode, unsigned HOST_WIDE_INT mask,
|
7550 |
|
|
int just_select)
|
7551 |
|
|
{
|
7552 |
|
|
enum rtx_code code = GET_CODE (x);
|
7553 |
|
|
int next_select = just_select || code == XOR || code == NOT || code == NEG;
|
7554 |
|
|
enum machine_mode op_mode;
|
7555 |
|
|
unsigned HOST_WIDE_INT fuller_mask, nonzero;
|
7556 |
|
|
rtx op0, op1, temp;
|
7557 |
|
|
|
7558 |
|
|
/* If this is a CALL or ASM_OPERANDS, don't do anything. Some of the
|
7559 |
|
|
code below will do the wrong thing since the mode of such an
|
7560 |
|
|
expression is VOIDmode.
|
7561 |
|
|
|
7562 |
|
|
Also do nothing if X is a CLOBBER; this can happen if X was
|
7563 |
|
|
the return value from a call to gen_lowpart. */
|
7564 |
|
|
if (code == CALL || code == ASM_OPERANDS || code == CLOBBER)
|
7565 |
|
|
return x;
|
7566 |
|
|
|
7567 |
|
|
/* We want to perform the operation is its present mode unless we know
|
7568 |
|
|
that the operation is valid in MODE, in which case we do the operation
|
7569 |
|
|
in MODE. */
|
7570 |
|
|
op_mode = ((GET_MODE_CLASS (mode) == GET_MODE_CLASS (GET_MODE (x))
|
7571 |
|
|
&& have_insn_for (code, mode))
|
7572 |
|
|
? mode : GET_MODE (x));
|
7573 |
|
|
|
7574 |
|
|
/* It is not valid to do a right-shift in a narrower mode
|
7575 |
|
|
than the one it came in with. */
|
7576 |
|
|
if ((code == LSHIFTRT || code == ASHIFTRT)
|
7577 |
|
|
&& GET_MODE_BITSIZE (mode) < GET_MODE_BITSIZE (GET_MODE (x)))
|
7578 |
|
|
op_mode = GET_MODE (x);
|
7579 |
|
|
|
7580 |
|
|
/* Truncate MASK to fit OP_MODE. */
|
7581 |
|
|
if (op_mode)
|
7582 |
|
|
mask &= GET_MODE_MASK (op_mode);
|
7583 |
|
|
|
7584 |
|
|
/* When we have an arithmetic operation, or a shift whose count we
|
7585 |
|
|
do not know, we need to assume that all bits up to the highest-order
|
7586 |
|
|
bit in MASK will be needed. This is how we form such a mask. */
|
7587 |
|
|
if (mask & ((unsigned HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT - 1)))
|
7588 |
|
|
fuller_mask = ~(unsigned HOST_WIDE_INT) 0;
|
7589 |
|
|
else
|
7590 |
|
|
fuller_mask = (((unsigned HOST_WIDE_INT) 1 << (floor_log2 (mask) + 1))
|
7591 |
|
|
- 1);
|
7592 |
|
|
|
7593 |
|
|
/* Determine what bits of X are guaranteed to be (non)zero. */
|
7594 |
|
|
nonzero = nonzero_bits (x, mode);
|
7595 |
|
|
|
7596 |
|
|
/* If none of the bits in X are needed, return a zero. */
|
7597 |
|
|
if (!just_select && (nonzero & mask) == 0 && !side_effects_p (x))
|
7598 |
|
|
x = const0_rtx;
|
7599 |
|
|
|
7600 |
|
|
/* If X is a CONST_INT, return a new one. Do this here since the
|
7601 |
|
|
test below will fail. */
|
7602 |
|
|
if (CONST_INT_P (x))
|
7603 |
|
|
{
|
7604 |
|
|
if (SCALAR_INT_MODE_P (mode))
|
7605 |
|
|
return gen_int_mode (INTVAL (x) & mask, mode);
|
7606 |
|
|
else
|
7607 |
|
|
{
|
7608 |
|
|
x = GEN_INT (INTVAL (x) & mask);
|
7609 |
|
|
return gen_lowpart_common (mode, x);
|
7610 |
|
|
}
|
7611 |
|
|
}
|
7612 |
|
|
|
7613 |
|
|
/* If X is narrower than MODE and we want all the bits in X's mode, just
|
7614 |
|
|
get X in the proper mode. */
|
7615 |
|
|
if (GET_MODE_SIZE (GET_MODE (x)) < GET_MODE_SIZE (mode)
|
7616 |
|
|
&& (GET_MODE_MASK (GET_MODE (x)) & ~mask) == 0)
|
7617 |
|
|
return gen_lowpart (mode, x);
|
7618 |
|
|
|
7619 |
|
|
/* We can ignore the effect of a SUBREG if it narrows the mode or
|
7620 |
|
|
if the constant masks to zero all the bits the mode doesn't have. */
|
7621 |
|
|
if (GET_CODE (x) == SUBREG
|
7622 |
|
|
&& subreg_lowpart_p (x)
|
7623 |
|
|
&& ((GET_MODE_SIZE (GET_MODE (x))
|
7624 |
|
|
< GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
|
7625 |
|
|
|| (0 == (mask
|
7626 |
|
|
& GET_MODE_MASK (GET_MODE (x))
|
7627 |
|
|
& ~GET_MODE_MASK (GET_MODE (SUBREG_REG (x)))))))
|
7628 |
|
|
return force_to_mode (SUBREG_REG (x), mode, mask, next_select);
|
7629 |
|
|
|
7630 |
|
|
/* The arithmetic simplifications here only work for scalar integer modes. */
|
7631 |
|
|
if (!SCALAR_INT_MODE_P (mode) || !SCALAR_INT_MODE_P (GET_MODE (x)))
|
7632 |
|
|
return gen_lowpart_or_truncate (mode, x);
|
7633 |
|
|
|
7634 |
|
|
switch (code)
|
7635 |
|
|
{
|
7636 |
|
|
case CLOBBER:
|
7637 |
|
|
/* If X is a (clobber (const_int)), return it since we know we are
|
7638 |
|
|
generating something that won't match. */
|
7639 |
|
|
return x;
|
7640 |
|
|
|
7641 |
|
|
case SIGN_EXTEND:
|
7642 |
|
|
case ZERO_EXTEND:
|
7643 |
|
|
case ZERO_EXTRACT:
|
7644 |
|
|
case SIGN_EXTRACT:
|
7645 |
|
|
x = expand_compound_operation (x);
|
7646 |
|
|
if (GET_CODE (x) != code)
|
7647 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7648 |
|
|
break;
|
7649 |
|
|
|
7650 |
|
|
case TRUNCATE:
|
7651 |
|
|
/* Similarly for a truncate. */
|
7652 |
|
|
return force_to_mode (XEXP (x, 0), mode, mask, next_select);
|
7653 |
|
|
|
7654 |
|
|
case AND:
|
7655 |
|
|
/* If this is an AND with a constant, convert it into an AND
|
7656 |
|
|
whose constant is the AND of that constant with MASK. If it
|
7657 |
|
|
remains an AND of MASK, delete it since it is redundant. */
|
7658 |
|
|
|
7659 |
|
|
if (CONST_INT_P (XEXP (x, 1)))
|
7660 |
|
|
{
|
7661 |
|
|
x = simplify_and_const_int (x, op_mode, XEXP (x, 0),
|
7662 |
|
|
mask & INTVAL (XEXP (x, 1)));
|
7663 |
|
|
|
7664 |
|
|
/* If X is still an AND, see if it is an AND with a mask that
|
7665 |
|
|
is just some low-order bits. If so, and it is MASK, we don't
|
7666 |
|
|
need it. */
|
7667 |
|
|
|
7668 |
|
|
if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
|
7669 |
|
|
&& ((INTVAL (XEXP (x, 1)) & GET_MODE_MASK (GET_MODE (x)))
|
7670 |
|
|
== mask))
|
7671 |
|
|
x = XEXP (x, 0);
|
7672 |
|
|
|
7673 |
|
|
/* If it remains an AND, try making another AND with the bits
|
7674 |
|
|
in the mode mask that aren't in MASK turned on. If the
|
7675 |
|
|
constant in the AND is wide enough, this might make a
|
7676 |
|
|
cheaper constant. */
|
7677 |
|
|
|
7678 |
|
|
if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1))
|
7679 |
|
|
&& GET_MODE_MASK (GET_MODE (x)) != mask
|
7680 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT)
|
7681 |
|
|
{
|
7682 |
|
|
HOST_WIDE_INT cval = (INTVAL (XEXP (x, 1))
|
7683 |
|
|
| (GET_MODE_MASK (GET_MODE (x)) & ~mask));
|
7684 |
|
|
int width = GET_MODE_BITSIZE (GET_MODE (x));
|
7685 |
|
|
rtx y;
|
7686 |
|
|
|
7687 |
|
|
/* If MODE is narrower than HOST_WIDE_INT and CVAL is a negative
|
7688 |
|
|
number, sign extend it. */
|
7689 |
|
|
if (width > 0 && width < HOST_BITS_PER_WIDE_INT
|
7690 |
|
|
&& (cval & ((HOST_WIDE_INT) 1 << (width - 1))) != 0)
|
7691 |
|
|
cval |= (HOST_WIDE_INT) -1 << width;
|
7692 |
|
|
|
7693 |
|
|
y = simplify_gen_binary (AND, GET_MODE (x),
|
7694 |
|
|
XEXP (x, 0), GEN_INT (cval));
|
7695 |
|
|
if (rtx_cost (y, SET, optimize_this_for_speed_p)
|
7696 |
|
|
< rtx_cost (x, SET, optimize_this_for_speed_p))
|
7697 |
|
|
x = y;
|
7698 |
|
|
}
|
7699 |
|
|
|
7700 |
|
|
break;
|
7701 |
|
|
}
|
7702 |
|
|
|
7703 |
|
|
goto binop;
|
7704 |
|
|
|
7705 |
|
|
case PLUS:
|
7706 |
|
|
/* In (and (plus FOO C1) M), if M is a mask that just turns off
|
7707 |
|
|
low-order bits (as in an alignment operation) and FOO is already
|
7708 |
|
|
aligned to that boundary, mask C1 to that boundary as well.
|
7709 |
|
|
This may eliminate that PLUS and, later, the AND. */
|
7710 |
|
|
|
7711 |
|
|
{
|
7712 |
|
|
unsigned int width = GET_MODE_BITSIZE (mode);
|
7713 |
|
|
unsigned HOST_WIDE_INT smask = mask;
|
7714 |
|
|
|
7715 |
|
|
/* If MODE is narrower than HOST_WIDE_INT and mask is a negative
|
7716 |
|
|
number, sign extend it. */
|
7717 |
|
|
|
7718 |
|
|
if (width < HOST_BITS_PER_WIDE_INT
|
7719 |
|
|
&& (smask & ((HOST_WIDE_INT) 1 << (width - 1))) != 0)
|
7720 |
|
|
smask |= (HOST_WIDE_INT) -1 << width;
|
7721 |
|
|
|
7722 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7723 |
|
|
&& exact_log2 (- smask) >= 0
|
7724 |
|
|
&& (nonzero_bits (XEXP (x, 0), mode) & ~smask) == 0
|
7725 |
|
|
&& (INTVAL (XEXP (x, 1)) & ~smask) != 0)
|
7726 |
|
|
return force_to_mode (plus_constant (XEXP (x, 0),
|
7727 |
|
|
(INTVAL (XEXP (x, 1)) & smask)),
|
7728 |
|
|
mode, smask, next_select);
|
7729 |
|
|
}
|
7730 |
|
|
|
7731 |
|
|
/* ... fall through ... */
|
7732 |
|
|
|
7733 |
|
|
case MULT:
|
7734 |
|
|
/* For PLUS, MINUS and MULT, we need any bits less significant than the
|
7735 |
|
|
most significant bit in MASK since carries from those bits will
|
7736 |
|
|
affect the bits we are interested in. */
|
7737 |
|
|
mask = fuller_mask;
|
7738 |
|
|
goto binop;
|
7739 |
|
|
|
7740 |
|
|
case MINUS:
|
7741 |
|
|
/* If X is (minus C Y) where C's least set bit is larger than any bit
|
7742 |
|
|
in the mask, then we may replace with (neg Y). */
|
7743 |
|
|
if (CONST_INT_P (XEXP (x, 0))
|
7744 |
|
|
&& (((unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 0))
|
7745 |
|
|
& -INTVAL (XEXP (x, 0))))
|
7746 |
|
|
> mask))
|
7747 |
|
|
{
|
7748 |
|
|
x = simplify_gen_unary (NEG, GET_MODE (x), XEXP (x, 1),
|
7749 |
|
|
GET_MODE (x));
|
7750 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7751 |
|
|
}
|
7752 |
|
|
|
7753 |
|
|
/* Similarly, if C contains every bit in the fuller_mask, then we may
|
7754 |
|
|
replace with (not Y). */
|
7755 |
|
|
if (CONST_INT_P (XEXP (x, 0))
|
7756 |
|
|
&& ((INTVAL (XEXP (x, 0)) | (HOST_WIDE_INT) fuller_mask)
|
7757 |
|
|
== INTVAL (XEXP (x, 0))))
|
7758 |
|
|
{
|
7759 |
|
|
x = simplify_gen_unary (NOT, GET_MODE (x),
|
7760 |
|
|
XEXP (x, 1), GET_MODE (x));
|
7761 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7762 |
|
|
}
|
7763 |
|
|
|
7764 |
|
|
mask = fuller_mask;
|
7765 |
|
|
goto binop;
|
7766 |
|
|
|
7767 |
|
|
case IOR:
|
7768 |
|
|
case XOR:
|
7769 |
|
|
/* If X is (ior (lshiftrt FOO C1) C2), try to commute the IOR and
|
7770 |
|
|
LSHIFTRT so we end up with an (and (lshiftrt (ior ...) ...) ...)
|
7771 |
|
|
operation which may be a bitfield extraction. Ensure that the
|
7772 |
|
|
constant we form is not wider than the mode of X. */
|
7773 |
|
|
|
7774 |
|
|
if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
|
7775 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
7776 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
|
7777 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT
|
7778 |
|
|
&& CONST_INT_P (XEXP (x, 1))
|
7779 |
|
|
&& ((INTVAL (XEXP (XEXP (x, 0), 1))
|
7780 |
|
|
+ floor_log2 (INTVAL (XEXP (x, 1))))
|
7781 |
|
|
< GET_MODE_BITSIZE (GET_MODE (x)))
|
7782 |
|
|
&& (INTVAL (XEXP (x, 1))
|
7783 |
|
|
& ~nonzero_bits (XEXP (x, 0), GET_MODE (x))) == 0)
|
7784 |
|
|
{
|
7785 |
|
|
temp = GEN_INT ((INTVAL (XEXP (x, 1)) & mask)
|
7786 |
|
|
<< INTVAL (XEXP (XEXP (x, 0), 1)));
|
7787 |
|
|
temp = simplify_gen_binary (GET_CODE (x), GET_MODE (x),
|
7788 |
|
|
XEXP (XEXP (x, 0), 0), temp);
|
7789 |
|
|
x = simplify_gen_binary (LSHIFTRT, GET_MODE (x), temp,
|
7790 |
|
|
XEXP (XEXP (x, 0), 1));
|
7791 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7792 |
|
|
}
|
7793 |
|
|
|
7794 |
|
|
binop:
|
7795 |
|
|
/* For most binary operations, just propagate into the operation and
|
7796 |
|
|
change the mode if we have an operation of that mode. */
|
7797 |
|
|
|
7798 |
|
|
op0 = force_to_mode (XEXP (x, 0), mode, mask, next_select);
|
7799 |
|
|
op1 = force_to_mode (XEXP (x, 1), mode, mask, next_select);
|
7800 |
|
|
|
7801 |
|
|
/* If we ended up truncating both operands, truncate the result of the
|
7802 |
|
|
operation instead. */
|
7803 |
|
|
if (GET_CODE (op0) == TRUNCATE
|
7804 |
|
|
&& GET_CODE (op1) == TRUNCATE)
|
7805 |
|
|
{
|
7806 |
|
|
op0 = XEXP (op0, 0);
|
7807 |
|
|
op1 = XEXP (op1, 0);
|
7808 |
|
|
}
|
7809 |
|
|
|
7810 |
|
|
op0 = gen_lowpart_or_truncate (op_mode, op0);
|
7811 |
|
|
op1 = gen_lowpart_or_truncate (op_mode, op1);
|
7812 |
|
|
|
7813 |
|
|
if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0) || op1 != XEXP (x, 1))
|
7814 |
|
|
x = simplify_gen_binary (code, op_mode, op0, op1);
|
7815 |
|
|
break;
|
7816 |
|
|
|
7817 |
|
|
case ASHIFT:
|
7818 |
|
|
/* For left shifts, do the same, but just for the first operand.
|
7819 |
|
|
However, we cannot do anything with shifts where we cannot
|
7820 |
|
|
guarantee that the counts are smaller than the size of the mode
|
7821 |
|
|
because such a count will have a different meaning in a
|
7822 |
|
|
wider mode. */
|
7823 |
|
|
|
7824 |
|
|
if (! (CONST_INT_P (XEXP (x, 1))
|
7825 |
|
|
&& INTVAL (XEXP (x, 1)) >= 0
|
7826 |
|
|
&& INTVAL (XEXP (x, 1)) < GET_MODE_BITSIZE (mode))
|
7827 |
|
|
&& ! (GET_MODE (XEXP (x, 1)) != VOIDmode
|
7828 |
|
|
&& (nonzero_bits (XEXP (x, 1), GET_MODE (XEXP (x, 1)))
|
7829 |
|
|
< (unsigned HOST_WIDE_INT) GET_MODE_BITSIZE (mode))))
|
7830 |
|
|
break;
|
7831 |
|
|
|
7832 |
|
|
/* If the shift count is a constant and we can do arithmetic in
|
7833 |
|
|
the mode of the shift, refine which bits we need. Otherwise, use the
|
7834 |
|
|
conservative form of the mask. */
|
7835 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7836 |
|
|
&& INTVAL (XEXP (x, 1)) >= 0
|
7837 |
|
|
&& INTVAL (XEXP (x, 1)) < GET_MODE_BITSIZE (op_mode)
|
7838 |
|
|
&& GET_MODE_BITSIZE (op_mode) <= HOST_BITS_PER_WIDE_INT)
|
7839 |
|
|
mask >>= INTVAL (XEXP (x, 1));
|
7840 |
|
|
else
|
7841 |
|
|
mask = fuller_mask;
|
7842 |
|
|
|
7843 |
|
|
op0 = gen_lowpart_or_truncate (op_mode,
|
7844 |
|
|
force_to_mode (XEXP (x, 0), op_mode,
|
7845 |
|
|
mask, next_select));
|
7846 |
|
|
|
7847 |
|
|
if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0))
|
7848 |
|
|
x = simplify_gen_binary (code, op_mode, op0, XEXP (x, 1));
|
7849 |
|
|
break;
|
7850 |
|
|
|
7851 |
|
|
case LSHIFTRT:
|
7852 |
|
|
/* Here we can only do something if the shift count is a constant,
|
7853 |
|
|
this shift constant is valid for the host, and we can do arithmetic
|
7854 |
|
|
in OP_MODE. */
|
7855 |
|
|
|
7856 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7857 |
|
|
&& INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
|
7858 |
|
|
&& GET_MODE_BITSIZE (op_mode) <= HOST_BITS_PER_WIDE_INT)
|
7859 |
|
|
{
|
7860 |
|
|
rtx inner = XEXP (x, 0);
|
7861 |
|
|
unsigned HOST_WIDE_INT inner_mask;
|
7862 |
|
|
|
7863 |
|
|
/* Select the mask of the bits we need for the shift operand. */
|
7864 |
|
|
inner_mask = mask << INTVAL (XEXP (x, 1));
|
7865 |
|
|
|
7866 |
|
|
/* We can only change the mode of the shift if we can do arithmetic
|
7867 |
|
|
in the mode of the shift and INNER_MASK is no wider than the
|
7868 |
|
|
width of X's mode. */
|
7869 |
|
|
if ((inner_mask & ~GET_MODE_MASK (GET_MODE (x))) != 0)
|
7870 |
|
|
op_mode = GET_MODE (x);
|
7871 |
|
|
|
7872 |
|
|
inner = force_to_mode (inner, op_mode, inner_mask, next_select);
|
7873 |
|
|
|
7874 |
|
|
if (GET_MODE (x) != op_mode || inner != XEXP (x, 0))
|
7875 |
|
|
x = simplify_gen_binary (LSHIFTRT, op_mode, inner, XEXP (x, 1));
|
7876 |
|
|
}
|
7877 |
|
|
|
7878 |
|
|
/* If we have (and (lshiftrt FOO C1) C2) where the combination of the
|
7879 |
|
|
shift and AND produces only copies of the sign bit (C2 is one less
|
7880 |
|
|
than a power of two), we can do this with just a shift. */
|
7881 |
|
|
|
7882 |
|
|
if (GET_CODE (x) == LSHIFTRT
|
7883 |
|
|
&& CONST_INT_P (XEXP (x, 1))
|
7884 |
|
|
/* The shift puts one of the sign bit copies in the least significant
|
7885 |
|
|
bit. */
|
7886 |
|
|
&& ((INTVAL (XEXP (x, 1))
|
7887 |
|
|
+ num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0))))
|
7888 |
|
|
>= GET_MODE_BITSIZE (GET_MODE (x)))
|
7889 |
|
|
&& exact_log2 (mask + 1) >= 0
|
7890 |
|
|
/* Number of bits left after the shift must be more than the mask
|
7891 |
|
|
needs. */
|
7892 |
|
|
&& ((INTVAL (XEXP (x, 1)) + exact_log2 (mask + 1))
|
7893 |
|
|
<= GET_MODE_BITSIZE (GET_MODE (x)))
|
7894 |
|
|
/* Must be more sign bit copies than the mask needs. */
|
7895 |
|
|
&& ((int) num_sign_bit_copies (XEXP (x, 0), GET_MODE (XEXP (x, 0)))
|
7896 |
|
|
>= exact_log2 (mask + 1)))
|
7897 |
|
|
x = simplify_gen_binary (LSHIFTRT, GET_MODE (x), XEXP (x, 0),
|
7898 |
|
|
GEN_INT (GET_MODE_BITSIZE (GET_MODE (x))
|
7899 |
|
|
- exact_log2 (mask + 1)));
|
7900 |
|
|
|
7901 |
|
|
goto shiftrt;
|
7902 |
|
|
|
7903 |
|
|
case ASHIFTRT:
|
7904 |
|
|
/* If we are just looking for the sign bit, we don't need this shift at
|
7905 |
|
|
all, even if it has a variable count. */
|
7906 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) <= HOST_BITS_PER_WIDE_INT
|
7907 |
|
|
&& (mask == ((unsigned HOST_WIDE_INT) 1
|
7908 |
|
|
<< (GET_MODE_BITSIZE (GET_MODE (x)) - 1))))
|
7909 |
|
|
return force_to_mode (XEXP (x, 0), mode, mask, next_select);
|
7910 |
|
|
|
7911 |
|
|
/* If this is a shift by a constant, get a mask that contains those bits
|
7912 |
|
|
that are not copies of the sign bit. We then have two cases: If
|
7913 |
|
|
MASK only includes those bits, this can be a logical shift, which may
|
7914 |
|
|
allow simplifications. If MASK is a single-bit field not within
|
7915 |
|
|
those bits, we are requesting a copy of the sign bit and hence can
|
7916 |
|
|
shift the sign bit to the appropriate location. */
|
7917 |
|
|
|
7918 |
|
|
if (CONST_INT_P (XEXP (x, 1)) && INTVAL (XEXP (x, 1)) >= 0
|
7919 |
|
|
&& INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT)
|
7920 |
|
|
{
|
7921 |
|
|
int i;
|
7922 |
|
|
|
7923 |
|
|
/* If the considered data is wider than HOST_WIDE_INT, we can't
|
7924 |
|
|
represent a mask for all its bits in a single scalar.
|
7925 |
|
|
But we only care about the lower bits, so calculate these. */
|
7926 |
|
|
|
7927 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) > HOST_BITS_PER_WIDE_INT)
|
7928 |
|
|
{
|
7929 |
|
|
nonzero = ~(HOST_WIDE_INT) 0;
|
7930 |
|
|
|
7931 |
|
|
/* GET_MODE_BITSIZE (GET_MODE (x)) - INTVAL (XEXP (x, 1))
|
7932 |
|
|
is the number of bits a full-width mask would have set.
|
7933 |
|
|
We need only shift if these are fewer than nonzero can
|
7934 |
|
|
hold. If not, we must keep all bits set in nonzero. */
|
7935 |
|
|
|
7936 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) - INTVAL (XEXP (x, 1))
|
7937 |
|
|
< HOST_BITS_PER_WIDE_INT)
|
7938 |
|
|
nonzero >>= INTVAL (XEXP (x, 1))
|
7939 |
|
|
+ HOST_BITS_PER_WIDE_INT
|
7940 |
|
|
- GET_MODE_BITSIZE (GET_MODE (x)) ;
|
7941 |
|
|
}
|
7942 |
|
|
else
|
7943 |
|
|
{
|
7944 |
|
|
nonzero = GET_MODE_MASK (GET_MODE (x));
|
7945 |
|
|
nonzero >>= INTVAL (XEXP (x, 1));
|
7946 |
|
|
}
|
7947 |
|
|
|
7948 |
|
|
if ((mask & ~nonzero) == 0)
|
7949 |
|
|
{
|
7950 |
|
|
x = simplify_shift_const (NULL_RTX, LSHIFTRT, GET_MODE (x),
|
7951 |
|
|
XEXP (x, 0), INTVAL (XEXP (x, 1)));
|
7952 |
|
|
if (GET_CODE (x) != ASHIFTRT)
|
7953 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7954 |
|
|
}
|
7955 |
|
|
|
7956 |
|
|
else if ((i = exact_log2 (mask)) >= 0)
|
7957 |
|
|
{
|
7958 |
|
|
x = simplify_shift_const
|
7959 |
|
|
(NULL_RTX, LSHIFTRT, GET_MODE (x), XEXP (x, 0),
|
7960 |
|
|
GET_MODE_BITSIZE (GET_MODE (x)) - 1 - i);
|
7961 |
|
|
|
7962 |
|
|
if (GET_CODE (x) != ASHIFTRT)
|
7963 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
7964 |
|
|
}
|
7965 |
|
|
}
|
7966 |
|
|
|
7967 |
|
|
/* If MASK is 1, convert this to an LSHIFTRT. This can be done
|
7968 |
|
|
even if the shift count isn't a constant. */
|
7969 |
|
|
if (mask == 1)
|
7970 |
|
|
x = simplify_gen_binary (LSHIFTRT, GET_MODE (x),
|
7971 |
|
|
XEXP (x, 0), XEXP (x, 1));
|
7972 |
|
|
|
7973 |
|
|
shiftrt:
|
7974 |
|
|
|
7975 |
|
|
/* If this is a zero- or sign-extension operation that just affects bits
|
7976 |
|
|
we don't care about, remove it. Be sure the call above returned
|
7977 |
|
|
something that is still a shift. */
|
7978 |
|
|
|
7979 |
|
|
if ((GET_CODE (x) == LSHIFTRT || GET_CODE (x) == ASHIFTRT)
|
7980 |
|
|
&& CONST_INT_P (XEXP (x, 1))
|
7981 |
|
|
&& INTVAL (XEXP (x, 1)) >= 0
|
7982 |
|
|
&& (INTVAL (XEXP (x, 1))
|
7983 |
|
|
<= GET_MODE_BITSIZE (GET_MODE (x)) - (floor_log2 (mask) + 1))
|
7984 |
|
|
&& GET_CODE (XEXP (x, 0)) == ASHIFT
|
7985 |
|
|
&& XEXP (XEXP (x, 0), 1) == XEXP (x, 1))
|
7986 |
|
|
return force_to_mode (XEXP (XEXP (x, 0), 0), mode, mask,
|
7987 |
|
|
next_select);
|
7988 |
|
|
|
7989 |
|
|
break;
|
7990 |
|
|
|
7991 |
|
|
case ROTATE:
|
7992 |
|
|
case ROTATERT:
|
7993 |
|
|
/* If the shift count is constant and we can do computations
|
7994 |
|
|
in the mode of X, compute where the bits we care about are.
|
7995 |
|
|
Otherwise, we can't do anything. Don't change the mode of
|
7996 |
|
|
the shift or propagate MODE into the shift, though. */
|
7997 |
|
|
if (CONST_INT_P (XEXP (x, 1))
|
7998 |
|
|
&& INTVAL (XEXP (x, 1)) >= 0)
|
7999 |
|
|
{
|
8000 |
|
|
temp = simplify_binary_operation (code == ROTATE ? ROTATERT : ROTATE,
|
8001 |
|
|
GET_MODE (x), GEN_INT (mask),
|
8002 |
|
|
XEXP (x, 1));
|
8003 |
|
|
if (temp && CONST_INT_P (temp))
|
8004 |
|
|
SUBST (XEXP (x, 0),
|
8005 |
|
|
force_to_mode (XEXP (x, 0), GET_MODE (x),
|
8006 |
|
|
INTVAL (temp), next_select));
|
8007 |
|
|
}
|
8008 |
|
|
break;
|
8009 |
|
|
|
8010 |
|
|
case NEG:
|
8011 |
|
|
/* If we just want the low-order bit, the NEG isn't needed since it
|
8012 |
|
|
won't change the low-order bit. */
|
8013 |
|
|
if (mask == 1)
|
8014 |
|
|
return force_to_mode (XEXP (x, 0), mode, mask, just_select);
|
8015 |
|
|
|
8016 |
|
|
/* We need any bits less significant than the most significant bit in
|
8017 |
|
|
MASK since carries from those bits will affect the bits we are
|
8018 |
|
|
interested in. */
|
8019 |
|
|
mask = fuller_mask;
|
8020 |
|
|
goto unop;
|
8021 |
|
|
|
8022 |
|
|
case NOT:
|
8023 |
|
|
/* (not FOO) is (xor FOO CONST), so if FOO is an LSHIFTRT, we can do the
|
8024 |
|
|
same as the XOR case above. Ensure that the constant we form is not
|
8025 |
|
|
wider than the mode of X. */
|
8026 |
|
|
|
8027 |
|
|
if (GET_CODE (XEXP (x, 0)) == LSHIFTRT
|
8028 |
|
|
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
|
8029 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) >= 0
|
8030 |
|
|
&& (INTVAL (XEXP (XEXP (x, 0), 1)) + floor_log2 (mask)
|
8031 |
|
|
< GET_MODE_BITSIZE (GET_MODE (x)))
|
8032 |
|
|
&& INTVAL (XEXP (XEXP (x, 0), 1)) < HOST_BITS_PER_WIDE_INT)
|
8033 |
|
|
{
|
8034 |
|
|
temp = gen_int_mode (mask << INTVAL (XEXP (XEXP (x, 0), 1)),
|
8035 |
|
|
GET_MODE (x));
|
8036 |
|
|
temp = simplify_gen_binary (XOR, GET_MODE (x),
|
8037 |
|
|
XEXP (XEXP (x, 0), 0), temp);
|
8038 |
|
|
x = simplify_gen_binary (LSHIFTRT, GET_MODE (x),
|
8039 |
|
|
temp, XEXP (XEXP (x, 0), 1));
|
8040 |
|
|
|
8041 |
|
|
return force_to_mode (x, mode, mask, next_select);
|
8042 |
|
|
}
|
8043 |
|
|
|
8044 |
|
|
/* (and (not FOO) CONST) is (not (or FOO (not CONST))), so we must
|
8045 |
|
|
use the full mask inside the NOT. */
|
8046 |
|
|
mask = fuller_mask;
|
8047 |
|
|
|
8048 |
|
|
unop:
|
8049 |
|
|
op0 = gen_lowpart_or_truncate (op_mode,
|
8050 |
|
|
force_to_mode (XEXP (x, 0), mode, mask,
|
8051 |
|
|
next_select));
|
8052 |
|
|
if (op_mode != GET_MODE (x) || op0 != XEXP (x, 0))
|
8053 |
|
|
x = simplify_gen_unary (code, op_mode, op0, op_mode);
|
8054 |
|
|
break;
|
8055 |
|
|
|
8056 |
|
|
case NE:
|
8057 |
|
|
/* (and (ne FOO 0) CONST) can be (and FOO CONST) if CONST is included
|
8058 |
|
|
in STORE_FLAG_VALUE and FOO has a single bit that might be nonzero,
|
8059 |
|
|
which is equal to STORE_FLAG_VALUE. */
|
8060 |
|
|
if ((mask & ~STORE_FLAG_VALUE) == 0 && XEXP (x, 1) == const0_rtx
|
8061 |
|
|
&& GET_MODE (XEXP (x, 0)) == mode
|
8062 |
|
|
&& exact_log2 (nonzero_bits (XEXP (x, 0), mode)) >= 0
|
8063 |
|
|
&& (nonzero_bits (XEXP (x, 0), mode)
|
8064 |
|
|
== (unsigned HOST_WIDE_INT) STORE_FLAG_VALUE))
|
8065 |
|
|
return force_to_mode (XEXP (x, 0), mode, mask, next_select);
|
8066 |
|
|
|
8067 |
|
|
break;
|
8068 |
|
|
|
8069 |
|
|
case IF_THEN_ELSE:
|
8070 |
|
|
/* We have no way of knowing if the IF_THEN_ELSE can itself be
|
8071 |
|
|
written in a narrower mode. We play it safe and do not do so. */
|
8072 |
|
|
|
8073 |
|
|
SUBST (XEXP (x, 1),
|
8074 |
|
|
gen_lowpart_or_truncate (GET_MODE (x),
|
8075 |
|
|
force_to_mode (XEXP (x, 1), mode,
|
8076 |
|
|
mask, next_select)));
|
8077 |
|
|
SUBST (XEXP (x, 2),
|
8078 |
|
|
gen_lowpart_or_truncate (GET_MODE (x),
|
8079 |
|
|
force_to_mode (XEXP (x, 2), mode,
|
8080 |
|
|
mask, next_select)));
|
8081 |
|
|
break;
|
8082 |
|
|
|
8083 |
|
|
default:
|
8084 |
|
|
break;
|
8085 |
|
|
}
|
8086 |
|
|
|
8087 |
|
|
/* Ensure we return a value of the proper mode. */
|
8088 |
|
|
return gen_lowpart_or_truncate (mode, x);
|
8089 |
|
|
}
|
8090 |
|
|
|
8091 |
|
|
/* Return nonzero if X is an expression that has one of two values depending on
|
8092 |
|
|
whether some other value is zero or nonzero. In that case, we return the
|
8093 |
|
|
value that is being tested, *PTRUE is set to the value if the rtx being
|
8094 |
|
|
returned has a nonzero value, and *PFALSE is set to the other alternative.
|
8095 |
|
|
|
8096 |
|
|
If we return zero, we set *PTRUE and *PFALSE to X. */
|
8097 |
|
|
|
8098 |
|
|
static rtx
|
8099 |
|
|
if_then_else_cond (rtx x, rtx *ptrue, rtx *pfalse)
|
8100 |
|
|
{
|
8101 |
|
|
enum machine_mode mode = GET_MODE (x);
|
8102 |
|
|
enum rtx_code code = GET_CODE (x);
|
8103 |
|
|
rtx cond0, cond1, true0, true1, false0, false1;
|
8104 |
|
|
unsigned HOST_WIDE_INT nz;
|
8105 |
|
|
|
8106 |
|
|
/* If we are comparing a value against zero, we are done. */
|
8107 |
|
|
if ((code == NE || code == EQ)
|
8108 |
|
|
&& XEXP (x, 1) == const0_rtx)
|
8109 |
|
|
{
|
8110 |
|
|
*ptrue = (code == NE) ? const_true_rtx : const0_rtx;
|
8111 |
|
|
*pfalse = (code == NE) ? const0_rtx : const_true_rtx;
|
8112 |
|
|
return XEXP (x, 0);
|
8113 |
|
|
}
|
8114 |
|
|
|
8115 |
|
|
/* If this is a unary operation whose operand has one of two values, apply
|
8116 |
|
|
our opcode to compute those values. */
|
8117 |
|
|
else if (UNARY_P (x)
|
8118 |
|
|
&& (cond0 = if_then_else_cond (XEXP (x, 0), &true0, &false0)) != 0)
|
8119 |
|
|
{
|
8120 |
|
|
*ptrue = simplify_gen_unary (code, mode, true0, GET_MODE (XEXP (x, 0)));
|
8121 |
|
|
*pfalse = simplify_gen_unary (code, mode, false0,
|
8122 |
|
|
GET_MODE (XEXP (x, 0)));
|
8123 |
|
|
return cond0;
|
8124 |
|
|
}
|
8125 |
|
|
|
8126 |
|
|
/* If this is a COMPARE, do nothing, since the IF_THEN_ELSE we would
|
8127 |
|
|
make can't possibly match and would suppress other optimizations. */
|
8128 |
|
|
else if (code == COMPARE)
|
8129 |
|
|
;
|
8130 |
|
|
|
8131 |
|
|
/* If this is a binary operation, see if either side has only one of two
|
8132 |
|
|
values. If either one does or if both do and they are conditional on
|
8133 |
|
|
the same value, compute the new true and false values. */
|
8134 |
|
|
else if (BINARY_P (x))
|
8135 |
|
|
{
|
8136 |
|
|
cond0 = if_then_else_cond (XEXP (x, 0), &true0, &false0);
|
8137 |
|
|
cond1 = if_then_else_cond (XEXP (x, 1), &true1, &false1);
|
8138 |
|
|
|
8139 |
|
|
if ((cond0 != 0 || cond1 != 0)
|
8140 |
|
|
&& ! (cond0 != 0 && cond1 != 0 && ! rtx_equal_p (cond0, cond1)))
|
8141 |
|
|
{
|
8142 |
|
|
/* If if_then_else_cond returned zero, then true/false are the
|
8143 |
|
|
same rtl. We must copy one of them to prevent invalid rtl
|
8144 |
|
|
sharing. */
|
8145 |
|
|
if (cond0 == 0)
|
8146 |
|
|
true0 = copy_rtx (true0);
|
8147 |
|
|
else if (cond1 == 0)
|
8148 |
|
|
true1 = copy_rtx (true1);
|
8149 |
|
|
|
8150 |
|
|
if (COMPARISON_P (x))
|
8151 |
|
|
{
|
8152 |
|
|
*ptrue = simplify_gen_relational (code, mode, VOIDmode,
|
8153 |
|
|
true0, true1);
|
8154 |
|
|
*pfalse = simplify_gen_relational (code, mode, VOIDmode,
|
8155 |
|
|
false0, false1);
|
8156 |
|
|
}
|
8157 |
|
|
else
|
8158 |
|
|
{
|
8159 |
|
|
*ptrue = simplify_gen_binary (code, mode, true0, true1);
|
8160 |
|
|
*pfalse = simplify_gen_binary (code, mode, false0, false1);
|
8161 |
|
|
}
|
8162 |
|
|
|
8163 |
|
|
return cond0 ? cond0 : cond1;
|
8164 |
|
|
}
|
8165 |
|
|
|
8166 |
|
|
/* See if we have PLUS, IOR, XOR, MINUS or UMAX, where one of the
|
8167 |
|
|
operands is zero when the other is nonzero, and vice-versa,
|
8168 |
|
|
and STORE_FLAG_VALUE is 1 or -1. */
|
8169 |
|
|
|
8170 |
|
|
if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
|
8171 |
|
|
&& (code == PLUS || code == IOR || code == XOR || code == MINUS
|
8172 |
|
|
|| code == UMAX)
|
8173 |
|
|
&& GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
|
8174 |
|
|
{
|
8175 |
|
|
rtx op0 = XEXP (XEXP (x, 0), 1);
|
8176 |
|
|
rtx op1 = XEXP (XEXP (x, 1), 1);
|
8177 |
|
|
|
8178 |
|
|
cond0 = XEXP (XEXP (x, 0), 0);
|
8179 |
|
|
cond1 = XEXP (XEXP (x, 1), 0);
|
8180 |
|
|
|
8181 |
|
|
if (COMPARISON_P (cond0)
|
8182 |
|
|
&& COMPARISON_P (cond1)
|
8183 |
|
|
&& ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
|
8184 |
|
|
&& rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
|
8185 |
|
|
&& rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
|
8186 |
|
|
|| ((swap_condition (GET_CODE (cond0))
|
8187 |
|
|
== reversed_comparison_code (cond1, NULL))
|
8188 |
|
|
&& rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
|
8189 |
|
|
&& rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
|
8190 |
|
|
&& ! side_effects_p (x))
|
8191 |
|
|
{
|
8192 |
|
|
*ptrue = simplify_gen_binary (MULT, mode, op0, const_true_rtx);
|
8193 |
|
|
*pfalse = simplify_gen_binary (MULT, mode,
|
8194 |
|
|
(code == MINUS
|
8195 |
|
|
? simplify_gen_unary (NEG, mode,
|
8196 |
|
|
op1, mode)
|
8197 |
|
|
: op1),
|
8198 |
|
|
const_true_rtx);
|
8199 |
|
|
return cond0;
|
8200 |
|
|
}
|
8201 |
|
|
}
|
8202 |
|
|
|
8203 |
|
|
/* Similarly for MULT, AND and UMIN, except that for these the result
|
8204 |
|
|
is always zero. */
|
8205 |
|
|
if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
|
8206 |
|
|
&& (code == MULT || code == AND || code == UMIN)
|
8207 |
|
|
&& GET_CODE (XEXP (x, 0)) == MULT && GET_CODE (XEXP (x, 1)) == MULT)
|
8208 |
|
|
{
|
8209 |
|
|
cond0 = XEXP (XEXP (x, 0), 0);
|
8210 |
|
|
cond1 = XEXP (XEXP (x, 1), 0);
|
8211 |
|
|
|
8212 |
|
|
if (COMPARISON_P (cond0)
|
8213 |
|
|
&& COMPARISON_P (cond1)
|
8214 |
|
|
&& ((GET_CODE (cond0) == reversed_comparison_code (cond1, NULL)
|
8215 |
|
|
&& rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 0))
|
8216 |
|
|
&& rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 1)))
|
8217 |
|
|
|| ((swap_condition (GET_CODE (cond0))
|
8218 |
|
|
== reversed_comparison_code (cond1, NULL))
|
8219 |
|
|
&& rtx_equal_p (XEXP (cond0, 0), XEXP (cond1, 1))
|
8220 |
|
|
&& rtx_equal_p (XEXP (cond0, 1), XEXP (cond1, 0))))
|
8221 |
|
|
&& ! side_effects_p (x))
|
8222 |
|
|
{
|
8223 |
|
|
*ptrue = *pfalse = const0_rtx;
|
8224 |
|
|
return cond0;
|
8225 |
|
|
}
|
8226 |
|
|
}
|
8227 |
|
|
}
|
8228 |
|
|
|
8229 |
|
|
else if (code == IF_THEN_ELSE)
|
8230 |
|
|
{
|
8231 |
|
|
/* If we have IF_THEN_ELSE already, extract the condition and
|
8232 |
|
|
canonicalize it if it is NE or EQ. */
|
8233 |
|
|
cond0 = XEXP (x, 0);
|
8234 |
|
|
*ptrue = XEXP (x, 1), *pfalse = XEXP (x, 2);
|
8235 |
|
|
if (GET_CODE (cond0) == NE && XEXP (cond0, 1) == const0_rtx)
|
8236 |
|
|
return XEXP (cond0, 0);
|
8237 |
|
|
else if (GET_CODE (cond0) == EQ && XEXP (cond0, 1) == const0_rtx)
|
8238 |
|
|
{
|
8239 |
|
|
*ptrue = XEXP (x, 2), *pfalse = XEXP (x, 1);
|
8240 |
|
|
return XEXP (cond0, 0);
|
8241 |
|
|
}
|
8242 |
|
|
else
|
8243 |
|
|
return cond0;
|
8244 |
|
|
}
|
8245 |
|
|
|
8246 |
|
|
/* If X is a SUBREG, we can narrow both the true and false values
|
8247 |
|
|
if the inner expression, if there is a condition. */
|
8248 |
|
|
else if (code == SUBREG
|
8249 |
|
|
&& 0 != (cond0 = if_then_else_cond (SUBREG_REG (x),
|
8250 |
|
|
&true0, &false0)))
|
8251 |
|
|
{
|
8252 |
|
|
true0 = simplify_gen_subreg (mode, true0,
|
8253 |
|
|
GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
|
8254 |
|
|
false0 = simplify_gen_subreg (mode, false0,
|
8255 |
|
|
GET_MODE (SUBREG_REG (x)), SUBREG_BYTE (x));
|
8256 |
|
|
if (true0 && false0)
|
8257 |
|
|
{
|
8258 |
|
|
*ptrue = true0;
|
8259 |
|
|
*pfalse = false0;
|
8260 |
|
|
return cond0;
|
8261 |
|
|
}
|
8262 |
|
|
}
|
8263 |
|
|
|
8264 |
|
|
/* If X is a constant, this isn't special and will cause confusions
|
8265 |
|
|
if we treat it as such. Likewise if it is equivalent to a constant. */
|
8266 |
|
|
else if (CONSTANT_P (x)
|
8267 |
|
|
|| ((cond0 = get_last_value (x)) != 0 && CONSTANT_P (cond0)))
|
8268 |
|
|
;
|
8269 |
|
|
|
8270 |
|
|
/* If we're in BImode, canonicalize on 0 and STORE_FLAG_VALUE, as that
|
8271 |
|
|
will be least confusing to the rest of the compiler. */
|
8272 |
|
|
else if (mode == BImode)
|
8273 |
|
|
{
|
8274 |
|
|
*ptrue = GEN_INT (STORE_FLAG_VALUE), *pfalse = const0_rtx;
|
8275 |
|
|
return x;
|
8276 |
|
|
}
|
8277 |
|
|
|
8278 |
|
|
/* If X is known to be either 0 or -1, those are the true and
|
8279 |
|
|
false values when testing X. */
|
8280 |
|
|
else if (x == constm1_rtx || x == const0_rtx
|
8281 |
|
|
|| (mode != VOIDmode
|
8282 |
|
|
&& num_sign_bit_copies (x, mode) == GET_MODE_BITSIZE (mode)))
|
8283 |
|
|
{
|
8284 |
|
|
*ptrue = constm1_rtx, *pfalse = const0_rtx;
|
8285 |
|
|
return x;
|
8286 |
|
|
}
|
8287 |
|
|
|
8288 |
|
|
/* Likewise for 0 or a single bit. */
|
8289 |
|
|
else if (SCALAR_INT_MODE_P (mode)
|
8290 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
8291 |
|
|
&& exact_log2 (nz = nonzero_bits (x, mode)) >= 0)
|
8292 |
|
|
{
|
8293 |
|
|
*ptrue = gen_int_mode (nz, mode), *pfalse = const0_rtx;
|
8294 |
|
|
return x;
|
8295 |
|
|
}
|
8296 |
|
|
|
8297 |
|
|
/* Otherwise fail; show no condition with true and false values the same. */
|
8298 |
|
|
*ptrue = *pfalse = x;
|
8299 |
|
|
return 0;
|
8300 |
|
|
}
|
8301 |
|
|
|
8302 |
|
|
/* Return the value of expression X given the fact that condition COND
|
8303 |
|
|
is known to be true when applied to REG as its first operand and VAL
|
8304 |
|
|
as its second. X is known to not be shared and so can be modified in
|
8305 |
|
|
place.
|
8306 |
|
|
|
8307 |
|
|
We only handle the simplest cases, and specifically those cases that
|
8308 |
|
|
arise with IF_THEN_ELSE expressions. */
|
8309 |
|
|
|
8310 |
|
|
static rtx
|
8311 |
|
|
known_cond (rtx x, enum rtx_code cond, rtx reg, rtx val)
|
8312 |
|
|
{
|
8313 |
|
|
enum rtx_code code = GET_CODE (x);
|
8314 |
|
|
rtx temp;
|
8315 |
|
|
const char *fmt;
|
8316 |
|
|
int i, j;
|
8317 |
|
|
|
8318 |
|
|
if (side_effects_p (x))
|
8319 |
|
|
return x;
|
8320 |
|
|
|
8321 |
|
|
/* If either operand of the condition is a floating point value,
|
8322 |
|
|
then we have to avoid collapsing an EQ comparison. */
|
8323 |
|
|
if (cond == EQ
|
8324 |
|
|
&& rtx_equal_p (x, reg)
|
8325 |
|
|
&& ! FLOAT_MODE_P (GET_MODE (x))
|
8326 |
|
|
&& ! FLOAT_MODE_P (GET_MODE (val)))
|
8327 |
|
|
return val;
|
8328 |
|
|
|
8329 |
|
|
if (cond == UNEQ && rtx_equal_p (x, reg))
|
8330 |
|
|
return val;
|
8331 |
|
|
|
8332 |
|
|
/* If X is (abs REG) and we know something about REG's relationship
|
8333 |
|
|
with zero, we may be able to simplify this. */
|
8334 |
|
|
|
8335 |
|
|
if (code == ABS && rtx_equal_p (XEXP (x, 0), reg) && val == const0_rtx)
|
8336 |
|
|
switch (cond)
|
8337 |
|
|
{
|
8338 |
|
|
case GE: case GT: case EQ:
|
8339 |
|
|
return XEXP (x, 0);
|
8340 |
|
|
case LT: case LE:
|
8341 |
|
|
return simplify_gen_unary (NEG, GET_MODE (XEXP (x, 0)),
|
8342 |
|
|
XEXP (x, 0),
|
8343 |
|
|
GET_MODE (XEXP (x, 0)));
|
8344 |
|
|
default:
|
8345 |
|
|
break;
|
8346 |
|
|
}
|
8347 |
|
|
|
8348 |
|
|
/* The only other cases we handle are MIN, MAX, and comparisons if the
|
8349 |
|
|
operands are the same as REG and VAL. */
|
8350 |
|
|
|
8351 |
|
|
else if (COMPARISON_P (x) || COMMUTATIVE_ARITH_P (x))
|
8352 |
|
|
{
|
8353 |
|
|
if (rtx_equal_p (XEXP (x, 0), val))
|
8354 |
|
|
cond = swap_condition (cond), temp = val, val = reg, reg = temp;
|
8355 |
|
|
|
8356 |
|
|
if (rtx_equal_p (XEXP (x, 0), reg) && rtx_equal_p (XEXP (x, 1), val))
|
8357 |
|
|
{
|
8358 |
|
|
if (COMPARISON_P (x))
|
8359 |
|
|
{
|
8360 |
|
|
if (comparison_dominates_p (cond, code))
|
8361 |
|
|
return const_true_rtx;
|
8362 |
|
|
|
8363 |
|
|
code = reversed_comparison_code (x, NULL);
|
8364 |
|
|
if (code != UNKNOWN
|
8365 |
|
|
&& comparison_dominates_p (cond, code))
|
8366 |
|
|
return const0_rtx;
|
8367 |
|
|
else
|
8368 |
|
|
return x;
|
8369 |
|
|
}
|
8370 |
|
|
else if (code == SMAX || code == SMIN
|
8371 |
|
|
|| code == UMIN || code == UMAX)
|
8372 |
|
|
{
|
8373 |
|
|
int unsignedp = (code == UMIN || code == UMAX);
|
8374 |
|
|
|
8375 |
|
|
/* Do not reverse the condition when it is NE or EQ.
|
8376 |
|
|
This is because we cannot conclude anything about
|
8377 |
|
|
the value of 'SMAX (x, y)' when x is not equal to y,
|
8378 |
|
|
but we can when x equals y. */
|
8379 |
|
|
if ((code == SMAX || code == UMAX)
|
8380 |
|
|
&& ! (cond == EQ || cond == NE))
|
8381 |
|
|
cond = reverse_condition (cond);
|
8382 |
|
|
|
8383 |
|
|
switch (cond)
|
8384 |
|
|
{
|
8385 |
|
|
case GE: case GT:
|
8386 |
|
|
return unsignedp ? x : XEXP (x, 1);
|
8387 |
|
|
case LE: case LT:
|
8388 |
|
|
return unsignedp ? x : XEXP (x, 0);
|
8389 |
|
|
case GEU: case GTU:
|
8390 |
|
|
return unsignedp ? XEXP (x, 1) : x;
|
8391 |
|
|
case LEU: case LTU:
|
8392 |
|
|
return unsignedp ? XEXP (x, 0) : x;
|
8393 |
|
|
default:
|
8394 |
|
|
break;
|
8395 |
|
|
}
|
8396 |
|
|
}
|
8397 |
|
|
}
|
8398 |
|
|
}
|
8399 |
|
|
else if (code == SUBREG)
|
8400 |
|
|
{
|
8401 |
|
|
enum machine_mode inner_mode = GET_MODE (SUBREG_REG (x));
|
8402 |
|
|
rtx new_rtx, r = known_cond (SUBREG_REG (x), cond, reg, val);
|
8403 |
|
|
|
8404 |
|
|
if (SUBREG_REG (x) != r)
|
8405 |
|
|
{
|
8406 |
|
|
/* We must simplify subreg here, before we lose track of the
|
8407 |
|
|
original inner_mode. */
|
8408 |
|
|
new_rtx = simplify_subreg (GET_MODE (x), r,
|
8409 |
|
|
inner_mode, SUBREG_BYTE (x));
|
8410 |
|
|
if (new_rtx)
|
8411 |
|
|
return new_rtx;
|
8412 |
|
|
else
|
8413 |
|
|
SUBST (SUBREG_REG (x), r);
|
8414 |
|
|
}
|
8415 |
|
|
|
8416 |
|
|
return x;
|
8417 |
|
|
}
|
8418 |
|
|
/* We don't have to handle SIGN_EXTEND here, because even in the
|
8419 |
|
|
case of replacing something with a modeless CONST_INT, a
|
8420 |
|
|
CONST_INT is already (supposed to be) a valid sign extension for
|
8421 |
|
|
its narrower mode, which implies it's already properly
|
8422 |
|
|
sign-extended for the wider mode. Now, for ZERO_EXTEND, the
|
8423 |
|
|
story is different. */
|
8424 |
|
|
else if (code == ZERO_EXTEND)
|
8425 |
|
|
{
|
8426 |
|
|
enum machine_mode inner_mode = GET_MODE (XEXP (x, 0));
|
8427 |
|
|
rtx new_rtx, r = known_cond (XEXP (x, 0), cond, reg, val);
|
8428 |
|
|
|
8429 |
|
|
if (XEXP (x, 0) != r)
|
8430 |
|
|
{
|
8431 |
|
|
/* We must simplify the zero_extend here, before we lose
|
8432 |
|
|
track of the original inner_mode. */
|
8433 |
|
|
new_rtx = simplify_unary_operation (ZERO_EXTEND, GET_MODE (x),
|
8434 |
|
|
r, inner_mode);
|
8435 |
|
|
if (new_rtx)
|
8436 |
|
|
return new_rtx;
|
8437 |
|
|
else
|
8438 |
|
|
SUBST (XEXP (x, 0), r);
|
8439 |
|
|
}
|
8440 |
|
|
|
8441 |
|
|
return x;
|
8442 |
|
|
}
|
8443 |
|
|
|
8444 |
|
|
fmt = GET_RTX_FORMAT (code);
|
8445 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
8446 |
|
|
{
|
8447 |
|
|
if (fmt[i] == 'e')
|
8448 |
|
|
SUBST (XEXP (x, i), known_cond (XEXP (x, i), cond, reg, val));
|
8449 |
|
|
else if (fmt[i] == 'E')
|
8450 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
8451 |
|
|
SUBST (XVECEXP (x, i, j), known_cond (XVECEXP (x, i, j),
|
8452 |
|
|
cond, reg, val));
|
8453 |
|
|
}
|
8454 |
|
|
|
8455 |
|
|
return x;
|
8456 |
|
|
}
|
8457 |
|
|
|
8458 |
|
|
/* See if X and Y are equal for the purposes of seeing if we can rewrite an
|
8459 |
|
|
assignment as a field assignment. */
|
8460 |
|
|
|
8461 |
|
|
static int
|
8462 |
|
|
rtx_equal_for_field_assignment_p (rtx x, rtx y)
|
8463 |
|
|
{
|
8464 |
|
|
if (x == y || rtx_equal_p (x, y))
|
8465 |
|
|
return 1;
|
8466 |
|
|
|
8467 |
|
|
if (x == 0 || y == 0 || GET_MODE (x) != GET_MODE (y))
|
8468 |
|
|
return 0;
|
8469 |
|
|
|
8470 |
|
|
/* Check for a paradoxical SUBREG of a MEM compared with the MEM.
|
8471 |
|
|
Note that all SUBREGs of MEM are paradoxical; otherwise they
|
8472 |
|
|
would have been rewritten. */
|
8473 |
|
|
if (MEM_P (x) && GET_CODE (y) == SUBREG
|
8474 |
|
|
&& MEM_P (SUBREG_REG (y))
|
8475 |
|
|
&& rtx_equal_p (SUBREG_REG (y),
|
8476 |
|
|
gen_lowpart (GET_MODE (SUBREG_REG (y)), x)))
|
8477 |
|
|
return 1;
|
8478 |
|
|
|
8479 |
|
|
if (MEM_P (y) && GET_CODE (x) == SUBREG
|
8480 |
|
|
&& MEM_P (SUBREG_REG (x))
|
8481 |
|
|
&& rtx_equal_p (SUBREG_REG (x),
|
8482 |
|
|
gen_lowpart (GET_MODE (SUBREG_REG (x)), y)))
|
8483 |
|
|
return 1;
|
8484 |
|
|
|
8485 |
|
|
/* We used to see if get_last_value of X and Y were the same but that's
|
8486 |
|
|
not correct. In one direction, we'll cause the assignment to have
|
8487 |
|
|
the wrong destination and in the case, we'll import a register into this
|
8488 |
|
|
insn that might have already have been dead. So fail if none of the
|
8489 |
|
|
above cases are true. */
|
8490 |
|
|
return 0;
|
8491 |
|
|
}
|
8492 |
|
|
|
8493 |
|
|
/* See if X, a SET operation, can be rewritten as a bit-field assignment.
|
8494 |
|
|
Return that assignment if so.
|
8495 |
|
|
|
8496 |
|
|
We only handle the most common cases. */
|
8497 |
|
|
|
8498 |
|
|
static rtx
|
8499 |
|
|
make_field_assignment (rtx x)
|
8500 |
|
|
{
|
8501 |
|
|
rtx dest = SET_DEST (x);
|
8502 |
|
|
rtx src = SET_SRC (x);
|
8503 |
|
|
rtx assign;
|
8504 |
|
|
rtx rhs, lhs;
|
8505 |
|
|
HOST_WIDE_INT c1;
|
8506 |
|
|
HOST_WIDE_INT pos;
|
8507 |
|
|
unsigned HOST_WIDE_INT len;
|
8508 |
|
|
rtx other;
|
8509 |
|
|
enum machine_mode mode;
|
8510 |
|
|
|
8511 |
|
|
/* If SRC was (and (not (ashift (const_int 1) POS)) DEST), this is
|
8512 |
|
|
a clear of a one-bit field. We will have changed it to
|
8513 |
|
|
(and (rotate (const_int -2) POS) DEST), so check for that. Also check
|
8514 |
|
|
for a SUBREG. */
|
8515 |
|
|
|
8516 |
|
|
if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == ROTATE
|
8517 |
|
|
&& CONST_INT_P (XEXP (XEXP (src, 0), 0))
|
8518 |
|
|
&& INTVAL (XEXP (XEXP (src, 0), 0)) == -2
|
8519 |
|
|
&& rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
|
8520 |
|
|
{
|
8521 |
|
|
assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
|
8522 |
|
|
1, 1, 1, 0);
|
8523 |
|
|
if (assign != 0)
|
8524 |
|
|
return gen_rtx_SET (VOIDmode, assign, const0_rtx);
|
8525 |
|
|
return x;
|
8526 |
|
|
}
|
8527 |
|
|
|
8528 |
|
|
if (GET_CODE (src) == AND && GET_CODE (XEXP (src, 0)) == SUBREG
|
8529 |
|
|
&& subreg_lowpart_p (XEXP (src, 0))
|
8530 |
|
|
&& (GET_MODE_SIZE (GET_MODE (XEXP (src, 0)))
|
8531 |
|
|
< GET_MODE_SIZE (GET_MODE (SUBREG_REG (XEXP (src, 0)))))
|
8532 |
|
|
&& GET_CODE (SUBREG_REG (XEXP (src, 0))) == ROTATE
|
8533 |
|
|
&& CONST_INT_P (XEXP (SUBREG_REG (XEXP (src, 0)), 0))
|
8534 |
|
|
&& INTVAL (XEXP (SUBREG_REG (XEXP (src, 0)), 0)) == -2
|
8535 |
|
|
&& rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
|
8536 |
|
|
{
|
8537 |
|
|
assign = make_extraction (VOIDmode, dest, 0,
|
8538 |
|
|
XEXP (SUBREG_REG (XEXP (src, 0)), 1),
|
8539 |
|
|
1, 1, 1, 0);
|
8540 |
|
|
if (assign != 0)
|
8541 |
|
|
return gen_rtx_SET (VOIDmode, assign, const0_rtx);
|
8542 |
|
|
return x;
|
8543 |
|
|
}
|
8544 |
|
|
|
8545 |
|
|
/* If SRC is (ior (ashift (const_int 1) POS) DEST), this is a set of a
|
8546 |
|
|
one-bit field. */
|
8547 |
|
|
if (GET_CODE (src) == IOR && GET_CODE (XEXP (src, 0)) == ASHIFT
|
8548 |
|
|
&& XEXP (XEXP (src, 0), 0) == const1_rtx
|
8549 |
|
|
&& rtx_equal_for_field_assignment_p (dest, XEXP (src, 1)))
|
8550 |
|
|
{
|
8551 |
|
|
assign = make_extraction (VOIDmode, dest, 0, XEXP (XEXP (src, 0), 1),
|
8552 |
|
|
1, 1, 1, 0);
|
8553 |
|
|
if (assign != 0)
|
8554 |
|
|
return gen_rtx_SET (VOIDmode, assign, const1_rtx);
|
8555 |
|
|
return x;
|
8556 |
|
|
}
|
8557 |
|
|
|
8558 |
|
|
/* If DEST is already a field assignment, i.e. ZERO_EXTRACT, and the
|
8559 |
|
|
SRC is an AND with all bits of that field set, then we can discard
|
8560 |
|
|
the AND. */
|
8561 |
|
|
if (GET_CODE (dest) == ZERO_EXTRACT
|
8562 |
|
|
&& CONST_INT_P (XEXP (dest, 1))
|
8563 |
|
|
&& GET_CODE (src) == AND
|
8564 |
|
|
&& CONST_INT_P (XEXP (src, 1)))
|
8565 |
|
|
{
|
8566 |
|
|
HOST_WIDE_INT width = INTVAL (XEXP (dest, 1));
|
8567 |
|
|
unsigned HOST_WIDE_INT and_mask = INTVAL (XEXP (src, 1));
|
8568 |
|
|
unsigned HOST_WIDE_INT ze_mask;
|
8569 |
|
|
|
8570 |
|
|
if (width >= HOST_BITS_PER_WIDE_INT)
|
8571 |
|
|
ze_mask = -1;
|
8572 |
|
|
else
|
8573 |
|
|
ze_mask = ((unsigned HOST_WIDE_INT)1 << width) - 1;
|
8574 |
|
|
|
8575 |
|
|
/* Complete overlap. We can remove the source AND. */
|
8576 |
|
|
if ((and_mask & ze_mask) == ze_mask)
|
8577 |
|
|
return gen_rtx_SET (VOIDmode, dest, XEXP (src, 0));
|
8578 |
|
|
|
8579 |
|
|
/* Partial overlap. We can reduce the source AND. */
|
8580 |
|
|
if ((and_mask & ze_mask) != and_mask)
|
8581 |
|
|
{
|
8582 |
|
|
mode = GET_MODE (src);
|
8583 |
|
|
src = gen_rtx_AND (mode, XEXP (src, 0),
|
8584 |
|
|
gen_int_mode (and_mask & ze_mask, mode));
|
8585 |
|
|
return gen_rtx_SET (VOIDmode, dest, src);
|
8586 |
|
|
}
|
8587 |
|
|
}
|
8588 |
|
|
|
8589 |
|
|
/* The other case we handle is assignments into a constant-position
|
8590 |
|
|
field. They look like (ior/xor (and DEST C1) OTHER). If C1 represents
|
8591 |
|
|
a mask that has all one bits except for a group of zero bits and
|
8592 |
|
|
OTHER is known to have zeros where C1 has ones, this is such an
|
8593 |
|
|
assignment. Compute the position and length from C1. Shift OTHER
|
8594 |
|
|
to the appropriate position, force it to the required mode, and
|
8595 |
|
|
make the extraction. Check for the AND in both operands. */
|
8596 |
|
|
|
8597 |
|
|
if (GET_CODE (src) != IOR && GET_CODE (src) != XOR)
|
8598 |
|
|
return x;
|
8599 |
|
|
|
8600 |
|
|
rhs = expand_compound_operation (XEXP (src, 0));
|
8601 |
|
|
lhs = expand_compound_operation (XEXP (src, 1));
|
8602 |
|
|
|
8603 |
|
|
if (GET_CODE (rhs) == AND
|
8604 |
|
|
&& CONST_INT_P (XEXP (rhs, 1))
|
8605 |
|
|
&& rtx_equal_for_field_assignment_p (XEXP (rhs, 0), dest))
|
8606 |
|
|
c1 = INTVAL (XEXP (rhs, 1)), other = lhs;
|
8607 |
|
|
else if (GET_CODE (lhs) == AND
|
8608 |
|
|
&& CONST_INT_P (XEXP (lhs, 1))
|
8609 |
|
|
&& rtx_equal_for_field_assignment_p (XEXP (lhs, 0), dest))
|
8610 |
|
|
c1 = INTVAL (XEXP (lhs, 1)), other = rhs;
|
8611 |
|
|
else
|
8612 |
|
|
return x;
|
8613 |
|
|
|
8614 |
|
|
pos = get_pos_from_mask ((~c1) & GET_MODE_MASK (GET_MODE (dest)), &len);
|
8615 |
|
|
if (pos < 0 || pos + len > GET_MODE_BITSIZE (GET_MODE (dest))
|
8616 |
|
|
|| GET_MODE_BITSIZE (GET_MODE (dest)) > HOST_BITS_PER_WIDE_INT
|
8617 |
|
|
|| (c1 & nonzero_bits (other, GET_MODE (dest))) != 0)
|
8618 |
|
|
return x;
|
8619 |
|
|
|
8620 |
|
|
assign = make_extraction (VOIDmode, dest, pos, NULL_RTX, len, 1, 1, 0);
|
8621 |
|
|
if (assign == 0)
|
8622 |
|
|
return x;
|
8623 |
|
|
|
8624 |
|
|
/* The mode to use for the source is the mode of the assignment, or of
|
8625 |
|
|
what is inside a possible STRICT_LOW_PART. */
|
8626 |
|
|
mode = (GET_CODE (assign) == STRICT_LOW_PART
|
8627 |
|
|
? GET_MODE (XEXP (assign, 0)) : GET_MODE (assign));
|
8628 |
|
|
|
8629 |
|
|
/* Shift OTHER right POS places and make it the source, restricting it
|
8630 |
|
|
to the proper length and mode. */
|
8631 |
|
|
|
8632 |
|
|
src = canon_reg_for_combine (simplify_shift_const (NULL_RTX, LSHIFTRT,
|
8633 |
|
|
GET_MODE (src),
|
8634 |
|
|
other, pos),
|
8635 |
|
|
dest);
|
8636 |
|
|
src = force_to_mode (src, mode,
|
8637 |
|
|
GET_MODE_BITSIZE (mode) >= HOST_BITS_PER_WIDE_INT
|
8638 |
|
|
? ~(unsigned HOST_WIDE_INT) 0
|
8639 |
|
|
: ((unsigned HOST_WIDE_INT) 1 << len) - 1,
|
8640 |
|
|
0);
|
8641 |
|
|
|
8642 |
|
|
/* If SRC is masked by an AND that does not make a difference in
|
8643 |
|
|
the value being stored, strip it. */
|
8644 |
|
|
if (GET_CODE (assign) == ZERO_EXTRACT
|
8645 |
|
|
&& CONST_INT_P (XEXP (assign, 1))
|
8646 |
|
|
&& INTVAL (XEXP (assign, 1)) < HOST_BITS_PER_WIDE_INT
|
8647 |
|
|
&& GET_CODE (src) == AND
|
8648 |
|
|
&& CONST_INT_P (XEXP (src, 1))
|
8649 |
|
|
&& ((unsigned HOST_WIDE_INT) INTVAL (XEXP (src, 1))
|
8650 |
|
|
== ((unsigned HOST_WIDE_INT) 1 << INTVAL (XEXP (assign, 1))) - 1))
|
8651 |
|
|
src = XEXP (src, 0);
|
8652 |
|
|
|
8653 |
|
|
return gen_rtx_SET (VOIDmode, assign, src);
|
8654 |
|
|
}
|
8655 |
|
|
|
8656 |
|
|
/* See if X is of the form (+ (* a c) (* b c)) and convert to (* (+ a b) c)
|
8657 |
|
|
if so. */
|
8658 |
|
|
|
8659 |
|
|
static rtx
|
8660 |
|
|
apply_distributive_law (rtx x)
|
8661 |
|
|
{
|
8662 |
|
|
enum rtx_code code = GET_CODE (x);
|
8663 |
|
|
enum rtx_code inner_code;
|
8664 |
|
|
rtx lhs, rhs, other;
|
8665 |
|
|
rtx tem;
|
8666 |
|
|
|
8667 |
|
|
/* Distributivity is not true for floating point as it can change the
|
8668 |
|
|
value. So we don't do it unless -funsafe-math-optimizations. */
|
8669 |
|
|
if (FLOAT_MODE_P (GET_MODE (x))
|
8670 |
|
|
&& ! flag_unsafe_math_optimizations)
|
8671 |
|
|
return x;
|
8672 |
|
|
|
8673 |
|
|
/* The outer operation can only be one of the following: */
|
8674 |
|
|
if (code != IOR && code != AND && code != XOR
|
8675 |
|
|
&& code != PLUS && code != MINUS)
|
8676 |
|
|
return x;
|
8677 |
|
|
|
8678 |
|
|
lhs = XEXP (x, 0);
|
8679 |
|
|
rhs = XEXP (x, 1);
|
8680 |
|
|
|
8681 |
|
|
/* If either operand is a primitive we can't do anything, so get out
|
8682 |
|
|
fast. */
|
8683 |
|
|
if (OBJECT_P (lhs) || OBJECT_P (rhs))
|
8684 |
|
|
return x;
|
8685 |
|
|
|
8686 |
|
|
lhs = expand_compound_operation (lhs);
|
8687 |
|
|
rhs = expand_compound_operation (rhs);
|
8688 |
|
|
inner_code = GET_CODE (lhs);
|
8689 |
|
|
if (inner_code != GET_CODE (rhs))
|
8690 |
|
|
return x;
|
8691 |
|
|
|
8692 |
|
|
/* See if the inner and outer operations distribute. */
|
8693 |
|
|
switch (inner_code)
|
8694 |
|
|
{
|
8695 |
|
|
case LSHIFTRT:
|
8696 |
|
|
case ASHIFTRT:
|
8697 |
|
|
case AND:
|
8698 |
|
|
case IOR:
|
8699 |
|
|
/* These all distribute except over PLUS. */
|
8700 |
|
|
if (code == PLUS || code == MINUS)
|
8701 |
|
|
return x;
|
8702 |
|
|
break;
|
8703 |
|
|
|
8704 |
|
|
case MULT:
|
8705 |
|
|
if (code != PLUS && code != MINUS)
|
8706 |
|
|
return x;
|
8707 |
|
|
break;
|
8708 |
|
|
|
8709 |
|
|
case ASHIFT:
|
8710 |
|
|
/* This is also a multiply, so it distributes over everything. */
|
8711 |
|
|
break;
|
8712 |
|
|
|
8713 |
|
|
case SUBREG:
|
8714 |
|
|
/* Non-paradoxical SUBREGs distributes over all operations,
|
8715 |
|
|
provided the inner modes and byte offsets are the same, this
|
8716 |
|
|
is an extraction of a low-order part, we don't convert an fp
|
8717 |
|
|
operation to int or vice versa, this is not a vector mode,
|
8718 |
|
|
and we would not be converting a single-word operation into a
|
8719 |
|
|
multi-word operation. The latter test is not required, but
|
8720 |
|
|
it prevents generating unneeded multi-word operations. Some
|
8721 |
|
|
of the previous tests are redundant given the latter test,
|
8722 |
|
|
but are retained because they are required for correctness.
|
8723 |
|
|
|
8724 |
|
|
We produce the result slightly differently in this case. */
|
8725 |
|
|
|
8726 |
|
|
if (GET_MODE (SUBREG_REG (lhs)) != GET_MODE (SUBREG_REG (rhs))
|
8727 |
|
|
|| SUBREG_BYTE (lhs) != SUBREG_BYTE (rhs)
|
8728 |
|
|
|| ! subreg_lowpart_p (lhs)
|
8729 |
|
|
|| (GET_MODE_CLASS (GET_MODE (lhs))
|
8730 |
|
|
!= GET_MODE_CLASS (GET_MODE (SUBREG_REG (lhs))))
|
8731 |
|
|
|| (GET_MODE_SIZE (GET_MODE (lhs))
|
8732 |
|
|
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (lhs))))
|
8733 |
|
|
|| VECTOR_MODE_P (GET_MODE (lhs))
|
8734 |
|
|
|| GET_MODE_SIZE (GET_MODE (SUBREG_REG (lhs))) > UNITS_PER_WORD
|
8735 |
|
|
/* Result might need to be truncated. Don't change mode if
|
8736 |
|
|
explicit truncation is needed. */
|
8737 |
|
|
|| !TRULY_NOOP_TRUNCATION
|
8738 |
|
|
(GET_MODE_BITSIZE (GET_MODE (x)),
|
8739 |
|
|
GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (lhs)))))
|
8740 |
|
|
return x;
|
8741 |
|
|
|
8742 |
|
|
tem = simplify_gen_binary (code, GET_MODE (SUBREG_REG (lhs)),
|
8743 |
|
|
SUBREG_REG (lhs), SUBREG_REG (rhs));
|
8744 |
|
|
return gen_lowpart (GET_MODE (x), tem);
|
8745 |
|
|
|
8746 |
|
|
default:
|
8747 |
|
|
return x;
|
8748 |
|
|
}
|
8749 |
|
|
|
8750 |
|
|
/* Set LHS and RHS to the inner operands (A and B in the example
|
8751 |
|
|
above) and set OTHER to the common operand (C in the example).
|
8752 |
|
|
There is only one way to do this unless the inner operation is
|
8753 |
|
|
commutative. */
|
8754 |
|
|
if (COMMUTATIVE_ARITH_P (lhs)
|
8755 |
|
|
&& rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 0)))
|
8756 |
|
|
other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 1);
|
8757 |
|
|
else if (COMMUTATIVE_ARITH_P (lhs)
|
8758 |
|
|
&& rtx_equal_p (XEXP (lhs, 0), XEXP (rhs, 1)))
|
8759 |
|
|
other = XEXP (lhs, 0), lhs = XEXP (lhs, 1), rhs = XEXP (rhs, 0);
|
8760 |
|
|
else if (COMMUTATIVE_ARITH_P (lhs)
|
8761 |
|
|
&& rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 0)))
|
8762 |
|
|
other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 1);
|
8763 |
|
|
else if (rtx_equal_p (XEXP (lhs, 1), XEXP (rhs, 1)))
|
8764 |
|
|
other = XEXP (lhs, 1), lhs = XEXP (lhs, 0), rhs = XEXP (rhs, 0);
|
8765 |
|
|
else
|
8766 |
|
|
return x;
|
8767 |
|
|
|
8768 |
|
|
/* Form the new inner operation, seeing if it simplifies first. */
|
8769 |
|
|
tem = simplify_gen_binary (code, GET_MODE (x), lhs, rhs);
|
8770 |
|
|
|
8771 |
|
|
/* There is one exception to the general way of distributing:
|
8772 |
|
|
(a | c) ^ (b | c) -> (a ^ b) & ~c */
|
8773 |
|
|
if (code == XOR && inner_code == IOR)
|
8774 |
|
|
{
|
8775 |
|
|
inner_code = AND;
|
8776 |
|
|
other = simplify_gen_unary (NOT, GET_MODE (x), other, GET_MODE (x));
|
8777 |
|
|
}
|
8778 |
|
|
|
8779 |
|
|
/* We may be able to continuing distributing the result, so call
|
8780 |
|
|
ourselves recursively on the inner operation before forming the
|
8781 |
|
|
outer operation, which we return. */
|
8782 |
|
|
return simplify_gen_binary (inner_code, GET_MODE (x),
|
8783 |
|
|
apply_distributive_law (tem), other);
|
8784 |
|
|
}
|
8785 |
|
|
|
8786 |
|
|
/* See if X is of the form (* (+ A B) C), and if so convert to
|
8787 |
|
|
(+ (* A C) (* B C)) and try to simplify.
|
8788 |
|
|
|
8789 |
|
|
Most of the time, this results in no change. However, if some of
|
8790 |
|
|
the operands are the same or inverses of each other, simplifications
|
8791 |
|
|
will result.
|
8792 |
|
|
|
8793 |
|
|
For example, (and (ior A B) (not B)) can occur as the result of
|
8794 |
|
|
expanding a bit field assignment. When we apply the distributive
|
8795 |
|
|
law to this, we get (ior (and (A (not B))) (and (B (not B)))),
|
8796 |
|
|
which then simplifies to (and (A (not B))).
|
8797 |
|
|
|
8798 |
|
|
Note that no checks happen on the validity of applying the inverse
|
8799 |
|
|
distributive law. This is pointless since we can do it in the
|
8800 |
|
|
few places where this routine is called.
|
8801 |
|
|
|
8802 |
|
|
N is the index of the term that is decomposed (the arithmetic operation,
|
8803 |
|
|
i.e. (+ A B) in the first example above). !N is the index of the term that
|
8804 |
|
|
is distributed, i.e. of C in the first example above. */
|
8805 |
|
|
static rtx
|
8806 |
|
|
distribute_and_simplify_rtx (rtx x, int n)
|
8807 |
|
|
{
|
8808 |
|
|
enum machine_mode mode;
|
8809 |
|
|
enum rtx_code outer_code, inner_code;
|
8810 |
|
|
rtx decomposed, distributed, inner_op0, inner_op1, new_op0, new_op1, tmp;
|
8811 |
|
|
|
8812 |
|
|
/* Distributivity is not true for floating point as it can change the
|
8813 |
|
|
value. So we don't do it unless -funsafe-math-optimizations. */
|
8814 |
|
|
if (FLOAT_MODE_P (GET_MODE (x))
|
8815 |
|
|
&& ! flag_unsafe_math_optimizations)
|
8816 |
|
|
return NULL_RTX;
|
8817 |
|
|
|
8818 |
|
|
decomposed = XEXP (x, n);
|
8819 |
|
|
if (!ARITHMETIC_P (decomposed))
|
8820 |
|
|
return NULL_RTX;
|
8821 |
|
|
|
8822 |
|
|
mode = GET_MODE (x);
|
8823 |
|
|
outer_code = GET_CODE (x);
|
8824 |
|
|
distributed = XEXP (x, !n);
|
8825 |
|
|
|
8826 |
|
|
inner_code = GET_CODE (decomposed);
|
8827 |
|
|
inner_op0 = XEXP (decomposed, 0);
|
8828 |
|
|
inner_op1 = XEXP (decomposed, 1);
|
8829 |
|
|
|
8830 |
|
|
/* Special case (and (xor B C) (not A)), which is equivalent to
|
8831 |
|
|
(xor (ior A B) (ior A C)) */
|
8832 |
|
|
if (outer_code == AND && inner_code == XOR && GET_CODE (distributed) == NOT)
|
8833 |
|
|
{
|
8834 |
|
|
distributed = XEXP (distributed, 0);
|
8835 |
|
|
outer_code = IOR;
|
8836 |
|
|
}
|
8837 |
|
|
|
8838 |
|
|
if (n == 0)
|
8839 |
|
|
{
|
8840 |
|
|
/* Distribute the second term. */
|
8841 |
|
|
new_op0 = simplify_gen_binary (outer_code, mode, inner_op0, distributed);
|
8842 |
|
|
new_op1 = simplify_gen_binary (outer_code, mode, inner_op1, distributed);
|
8843 |
|
|
}
|
8844 |
|
|
else
|
8845 |
|
|
{
|
8846 |
|
|
/* Distribute the first term. */
|
8847 |
|
|
new_op0 = simplify_gen_binary (outer_code, mode, distributed, inner_op0);
|
8848 |
|
|
new_op1 = simplify_gen_binary (outer_code, mode, distributed, inner_op1);
|
8849 |
|
|
}
|
8850 |
|
|
|
8851 |
|
|
tmp = apply_distributive_law (simplify_gen_binary (inner_code, mode,
|
8852 |
|
|
new_op0, new_op1));
|
8853 |
|
|
if (GET_CODE (tmp) != outer_code
|
8854 |
|
|
&& rtx_cost (tmp, SET, optimize_this_for_speed_p)
|
8855 |
|
|
< rtx_cost (x, SET, optimize_this_for_speed_p))
|
8856 |
|
|
return tmp;
|
8857 |
|
|
|
8858 |
|
|
return NULL_RTX;
|
8859 |
|
|
}
|
8860 |
|
|
|
8861 |
|
|
/* Simplify a logical `and' of VAROP with the constant CONSTOP, to be done
|
8862 |
|
|
in MODE. Return an equivalent form, if different from (and VAROP
|
8863 |
|
|
(const_int CONSTOP)). Otherwise, return NULL_RTX. */
|
8864 |
|
|
|
8865 |
|
|
static rtx
|
8866 |
|
|
simplify_and_const_int_1 (enum machine_mode mode, rtx varop,
|
8867 |
|
|
unsigned HOST_WIDE_INT constop)
|
8868 |
|
|
{
|
8869 |
|
|
unsigned HOST_WIDE_INT nonzero;
|
8870 |
|
|
unsigned HOST_WIDE_INT orig_constop;
|
8871 |
|
|
rtx orig_varop;
|
8872 |
|
|
int i;
|
8873 |
|
|
|
8874 |
|
|
orig_varop = varop;
|
8875 |
|
|
orig_constop = constop;
|
8876 |
|
|
if (GET_CODE (varop) == CLOBBER)
|
8877 |
|
|
return NULL_RTX;
|
8878 |
|
|
|
8879 |
|
|
/* Simplify VAROP knowing that we will be only looking at some of the
|
8880 |
|
|
bits in it.
|
8881 |
|
|
|
8882 |
|
|
Note by passing in CONSTOP, we guarantee that the bits not set in
|
8883 |
|
|
CONSTOP are not significant and will never be examined. We must
|
8884 |
|
|
ensure that is the case by explicitly masking out those bits
|
8885 |
|
|
before returning. */
|
8886 |
|
|
varop = force_to_mode (varop, mode, constop, 0);
|
8887 |
|
|
|
8888 |
|
|
/* If VAROP is a CLOBBER, we will fail so return it. */
|
8889 |
|
|
if (GET_CODE (varop) == CLOBBER)
|
8890 |
|
|
return varop;
|
8891 |
|
|
|
8892 |
|
|
/* If VAROP is a CONST_INT, then we need to apply the mask in CONSTOP
|
8893 |
|
|
to VAROP and return the new constant. */
|
8894 |
|
|
if (CONST_INT_P (varop))
|
8895 |
|
|
return gen_int_mode (INTVAL (varop) & constop, mode);
|
8896 |
|
|
|
8897 |
|
|
/* See what bits may be nonzero in VAROP. Unlike the general case of
|
8898 |
|
|
a call to nonzero_bits, here we don't care about bits outside
|
8899 |
|
|
MODE. */
|
8900 |
|
|
|
8901 |
|
|
nonzero = nonzero_bits (varop, mode) & GET_MODE_MASK (mode);
|
8902 |
|
|
|
8903 |
|
|
/* Turn off all bits in the constant that are known to already be zero.
|
8904 |
|
|
Thus, if the AND isn't needed at all, we will have CONSTOP == NONZERO_BITS
|
8905 |
|
|
which is tested below. */
|
8906 |
|
|
|
8907 |
|
|
constop &= nonzero;
|
8908 |
|
|
|
8909 |
|
|
/* If we don't have any bits left, return zero. */
|
8910 |
|
|
if (constop == 0)
|
8911 |
|
|
return const0_rtx;
|
8912 |
|
|
|
8913 |
|
|
/* If VAROP is a NEG of something known to be zero or 1 and CONSTOP is
|
8914 |
|
|
a power of two, we can replace this with an ASHIFT. */
|
8915 |
|
|
if (GET_CODE (varop) == NEG && nonzero_bits (XEXP (varop, 0), mode) == 1
|
8916 |
|
|
&& (i = exact_log2 (constop)) >= 0)
|
8917 |
|
|
return simplify_shift_const (NULL_RTX, ASHIFT, mode, XEXP (varop, 0), i);
|
8918 |
|
|
|
8919 |
|
|
/* If VAROP is an IOR or XOR, apply the AND to both branches of the IOR
|
8920 |
|
|
or XOR, then try to apply the distributive law. This may eliminate
|
8921 |
|
|
operations if either branch can be simplified because of the AND.
|
8922 |
|
|
It may also make some cases more complex, but those cases probably
|
8923 |
|
|
won't match a pattern either with or without this. */
|
8924 |
|
|
|
8925 |
|
|
if (GET_CODE (varop) == IOR || GET_CODE (varop) == XOR)
|
8926 |
|
|
return
|
8927 |
|
|
gen_lowpart
|
8928 |
|
|
(mode,
|
8929 |
|
|
apply_distributive_law
|
8930 |
|
|
(simplify_gen_binary (GET_CODE (varop), GET_MODE (varop),
|
8931 |
|
|
simplify_and_const_int (NULL_RTX,
|
8932 |
|
|
GET_MODE (varop),
|
8933 |
|
|
XEXP (varop, 0),
|
8934 |
|
|
constop),
|
8935 |
|
|
simplify_and_const_int (NULL_RTX,
|
8936 |
|
|
GET_MODE (varop),
|
8937 |
|
|
XEXP (varop, 1),
|
8938 |
|
|
constop))));
|
8939 |
|
|
|
8940 |
|
|
/* If VAROP is PLUS, and the constant is a mask of low bits, distribute
|
8941 |
|
|
the AND and see if one of the operands simplifies to zero. If so, we
|
8942 |
|
|
may eliminate it. */
|
8943 |
|
|
|
8944 |
|
|
if (GET_CODE (varop) == PLUS
|
8945 |
|
|
&& exact_log2 (constop + 1) >= 0)
|
8946 |
|
|
{
|
8947 |
|
|
rtx o0, o1;
|
8948 |
|
|
|
8949 |
|
|
o0 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 0), constop);
|
8950 |
|
|
o1 = simplify_and_const_int (NULL_RTX, mode, XEXP (varop, 1), constop);
|
8951 |
|
|
if (o0 == const0_rtx)
|
8952 |
|
|
return o1;
|
8953 |
|
|
if (o1 == const0_rtx)
|
8954 |
|
|
return o0;
|
8955 |
|
|
}
|
8956 |
|
|
|
8957 |
|
|
/* Make a SUBREG if necessary. If we can't make it, fail. */
|
8958 |
|
|
varop = gen_lowpart (mode, varop);
|
8959 |
|
|
if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
|
8960 |
|
|
return NULL_RTX;
|
8961 |
|
|
|
8962 |
|
|
/* If we are only masking insignificant bits, return VAROP. */
|
8963 |
|
|
if (constop == nonzero)
|
8964 |
|
|
return varop;
|
8965 |
|
|
|
8966 |
|
|
if (varop == orig_varop && constop == orig_constop)
|
8967 |
|
|
return NULL_RTX;
|
8968 |
|
|
|
8969 |
|
|
/* Otherwise, return an AND. */
|
8970 |
|
|
return simplify_gen_binary (AND, mode, varop, gen_int_mode (constop, mode));
|
8971 |
|
|
}
|
8972 |
|
|
|
8973 |
|
|
|
8974 |
|
|
/* We have X, a logical `and' of VAROP with the constant CONSTOP, to be done
|
8975 |
|
|
in MODE.
|
8976 |
|
|
|
8977 |
|
|
Return an equivalent form, if different from X. Otherwise, return X. If
|
8978 |
|
|
X is zero, we are to always construct the equivalent form. */
|
8979 |
|
|
|
8980 |
|
|
static rtx
|
8981 |
|
|
simplify_and_const_int (rtx x, enum machine_mode mode, rtx varop,
|
8982 |
|
|
unsigned HOST_WIDE_INT constop)
|
8983 |
|
|
{
|
8984 |
|
|
rtx tem = simplify_and_const_int_1 (mode, varop, constop);
|
8985 |
|
|
if (tem)
|
8986 |
|
|
return tem;
|
8987 |
|
|
|
8988 |
|
|
if (!x)
|
8989 |
|
|
x = simplify_gen_binary (AND, GET_MODE (varop), varop,
|
8990 |
|
|
gen_int_mode (constop, mode));
|
8991 |
|
|
if (GET_MODE (x) != mode)
|
8992 |
|
|
x = gen_lowpart (mode, x);
|
8993 |
|
|
return x;
|
8994 |
|
|
}
|
8995 |
|
|
|
8996 |
|
|
/* Given a REG, X, compute which bits in X can be nonzero.
|
8997 |
|
|
We don't care about bits outside of those defined in MODE.
|
8998 |
|
|
|
8999 |
|
|
For most X this is simply GET_MODE_MASK (GET_MODE (MODE)), but if X is
|
9000 |
|
|
a shift, AND, or zero_extract, we can do better. */
|
9001 |
|
|
|
9002 |
|
|
static rtx
|
9003 |
|
|
reg_nonzero_bits_for_combine (const_rtx x, enum machine_mode mode,
|
9004 |
|
|
const_rtx known_x ATTRIBUTE_UNUSED,
|
9005 |
|
|
enum machine_mode known_mode ATTRIBUTE_UNUSED,
|
9006 |
|
|
unsigned HOST_WIDE_INT known_ret ATTRIBUTE_UNUSED,
|
9007 |
|
|
unsigned HOST_WIDE_INT *nonzero)
|
9008 |
|
|
{
|
9009 |
|
|
rtx tem;
|
9010 |
|
|
reg_stat_type *rsp;
|
9011 |
|
|
|
9012 |
|
|
/* If X is a register whose nonzero bits value is current, use it.
|
9013 |
|
|
Otherwise, if X is a register whose value we can find, use that
|
9014 |
|
|
value. Otherwise, use the previously-computed global nonzero bits
|
9015 |
|
|
for this register. */
|
9016 |
|
|
|
9017 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
|
9018 |
|
|
if (rsp->last_set_value != 0
|
9019 |
|
|
&& (rsp->last_set_mode == mode
|
9020 |
|
|
|| (GET_MODE_CLASS (rsp->last_set_mode) == MODE_INT
|
9021 |
|
|
&& GET_MODE_CLASS (mode) == MODE_INT))
|
9022 |
|
|
&& ((rsp->last_set_label >= label_tick_ebb_start
|
9023 |
|
|
&& rsp->last_set_label < label_tick)
|
9024 |
|
|
|| (rsp->last_set_label == label_tick
|
9025 |
|
|
&& DF_INSN_LUID (rsp->last_set) < subst_low_luid)
|
9026 |
|
|
|| (REGNO (x) >= FIRST_PSEUDO_REGISTER
|
9027 |
|
|
&& REG_N_SETS (REGNO (x)) == 1
|
9028 |
|
|
&& !REGNO_REG_SET_P
|
9029 |
|
|
(DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x)))))
|
9030 |
|
|
{
|
9031 |
|
|
*nonzero &= rsp->last_set_nonzero_bits;
|
9032 |
|
|
return NULL;
|
9033 |
|
|
}
|
9034 |
|
|
|
9035 |
|
|
tem = get_last_value (x);
|
9036 |
|
|
|
9037 |
|
|
if (tem)
|
9038 |
|
|
{
|
9039 |
|
|
#ifdef SHORT_IMMEDIATES_SIGN_EXTEND
|
9040 |
|
|
/* If X is narrower than MODE and TEM is a non-negative
|
9041 |
|
|
constant that would appear negative in the mode of X,
|
9042 |
|
|
sign-extend it for use in reg_nonzero_bits because some
|
9043 |
|
|
machines (maybe most) will actually do the sign-extension
|
9044 |
|
|
and this is the conservative approach.
|
9045 |
|
|
|
9046 |
|
|
??? For 2.5, try to tighten up the MD files in this regard
|
9047 |
|
|
instead of this kludge. */
|
9048 |
|
|
|
9049 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) < GET_MODE_BITSIZE (mode)
|
9050 |
|
|
&& CONST_INT_P (tem)
|
9051 |
|
|
&& INTVAL (tem) > 0
|
9052 |
|
|
&& 0 != (INTVAL (tem)
|
9053 |
|
|
& ((HOST_WIDE_INT) 1
|
9054 |
|
|
<< (GET_MODE_BITSIZE (GET_MODE (x)) - 1))))
|
9055 |
|
|
tem = GEN_INT (INTVAL (tem)
|
9056 |
|
|
| ((HOST_WIDE_INT) (-1)
|
9057 |
|
|
<< GET_MODE_BITSIZE (GET_MODE (x))));
|
9058 |
|
|
#endif
|
9059 |
|
|
return tem;
|
9060 |
|
|
}
|
9061 |
|
|
else if (nonzero_sign_valid && rsp->nonzero_bits)
|
9062 |
|
|
{
|
9063 |
|
|
unsigned HOST_WIDE_INT mask = rsp->nonzero_bits;
|
9064 |
|
|
|
9065 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (x)) < GET_MODE_BITSIZE (mode))
|
9066 |
|
|
/* We don't know anything about the upper bits. */
|
9067 |
|
|
mask |= GET_MODE_MASK (mode) ^ GET_MODE_MASK (GET_MODE (x));
|
9068 |
|
|
*nonzero &= mask;
|
9069 |
|
|
}
|
9070 |
|
|
|
9071 |
|
|
return NULL;
|
9072 |
|
|
}
|
9073 |
|
|
|
9074 |
|
|
/* Return the number of bits at the high-order end of X that are known to
|
9075 |
|
|
be equal to the sign bit. X will be used in mode MODE; if MODE is
|
9076 |
|
|
VOIDmode, X will be used in its own mode. The returned value will always
|
9077 |
|
|
be between 1 and the number of bits in MODE. */
|
9078 |
|
|
|
9079 |
|
|
static rtx
|
9080 |
|
|
reg_num_sign_bit_copies_for_combine (const_rtx x, enum machine_mode mode,
|
9081 |
|
|
const_rtx known_x ATTRIBUTE_UNUSED,
|
9082 |
|
|
enum machine_mode known_mode
|
9083 |
|
|
ATTRIBUTE_UNUSED,
|
9084 |
|
|
unsigned int known_ret ATTRIBUTE_UNUSED,
|
9085 |
|
|
unsigned int *result)
|
9086 |
|
|
{
|
9087 |
|
|
rtx tem;
|
9088 |
|
|
reg_stat_type *rsp;
|
9089 |
|
|
|
9090 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
|
9091 |
|
|
if (rsp->last_set_value != 0
|
9092 |
|
|
&& rsp->last_set_mode == mode
|
9093 |
|
|
&& ((rsp->last_set_label >= label_tick_ebb_start
|
9094 |
|
|
&& rsp->last_set_label < label_tick)
|
9095 |
|
|
|| (rsp->last_set_label == label_tick
|
9096 |
|
|
&& DF_INSN_LUID (rsp->last_set) < subst_low_luid)
|
9097 |
|
|
|| (REGNO (x) >= FIRST_PSEUDO_REGISTER
|
9098 |
|
|
&& REG_N_SETS (REGNO (x)) == 1
|
9099 |
|
|
&& !REGNO_REG_SET_P
|
9100 |
|
|
(DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), REGNO (x)))))
|
9101 |
|
|
{
|
9102 |
|
|
*result = rsp->last_set_sign_bit_copies;
|
9103 |
|
|
return NULL;
|
9104 |
|
|
}
|
9105 |
|
|
|
9106 |
|
|
tem = get_last_value (x);
|
9107 |
|
|
if (tem != 0)
|
9108 |
|
|
return tem;
|
9109 |
|
|
|
9110 |
|
|
if (nonzero_sign_valid && rsp->sign_bit_copies != 0
|
9111 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (x)) == GET_MODE_BITSIZE (mode))
|
9112 |
|
|
*result = rsp->sign_bit_copies;
|
9113 |
|
|
|
9114 |
|
|
return NULL;
|
9115 |
|
|
}
|
9116 |
|
|
|
9117 |
|
|
/* Return the number of "extended" bits there are in X, when interpreted
|
9118 |
|
|
as a quantity in MODE whose signedness is indicated by UNSIGNEDP. For
|
9119 |
|
|
unsigned quantities, this is the number of high-order zero bits.
|
9120 |
|
|
For signed quantities, this is the number of copies of the sign bit
|
9121 |
|
|
minus 1. In both case, this function returns the number of "spare"
|
9122 |
|
|
bits. For example, if two quantities for which this function returns
|
9123 |
|
|
at least 1 are added, the addition is known not to overflow.
|
9124 |
|
|
|
9125 |
|
|
This function will always return 0 unless called during combine, which
|
9126 |
|
|
implies that it must be called from a define_split. */
|
9127 |
|
|
|
9128 |
|
|
unsigned int
|
9129 |
|
|
extended_count (const_rtx x, enum machine_mode mode, int unsignedp)
|
9130 |
|
|
{
|
9131 |
|
|
if (nonzero_sign_valid == 0)
|
9132 |
|
|
return 0;
|
9133 |
|
|
|
9134 |
|
|
return (unsignedp
|
9135 |
|
|
? (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
9136 |
|
|
? (unsigned int) (GET_MODE_BITSIZE (mode) - 1
|
9137 |
|
|
- floor_log2 (nonzero_bits (x, mode)))
|
9138 |
|
|
: 0)
|
9139 |
|
|
: num_sign_bit_copies (x, mode) - 1);
|
9140 |
|
|
}
|
9141 |
|
|
|
9142 |
|
|
/* This function is called from `simplify_shift_const' to merge two
|
9143 |
|
|
outer operations. Specifically, we have already found that we need
|
9144 |
|
|
to perform operation *POP0 with constant *PCONST0 at the outermost
|
9145 |
|
|
position. We would now like to also perform OP1 with constant CONST1
|
9146 |
|
|
(with *POP0 being done last).
|
9147 |
|
|
|
9148 |
|
|
Return 1 if we can do the operation and update *POP0 and *PCONST0 with
|
9149 |
|
|
the resulting operation. *PCOMP_P is set to 1 if we would need to
|
9150 |
|
|
complement the innermost operand, otherwise it is unchanged.
|
9151 |
|
|
|
9152 |
|
|
MODE is the mode in which the operation will be done. No bits outside
|
9153 |
|
|
the width of this mode matter. It is assumed that the width of this mode
|
9154 |
|
|
is smaller than or equal to HOST_BITS_PER_WIDE_INT.
|
9155 |
|
|
|
9156 |
|
|
If *POP0 or OP1 are UNKNOWN, it means no operation is required. Only NEG, PLUS,
|
9157 |
|
|
IOR, XOR, and AND are supported. We may set *POP0 to SET if the proper
|
9158 |
|
|
result is simply *PCONST0.
|
9159 |
|
|
|
9160 |
|
|
If the resulting operation cannot be expressed as one operation, we
|
9161 |
|
|
return 0 and do not change *POP0, *PCONST0, and *PCOMP_P. */
|
9162 |
|
|
|
9163 |
|
|
static int
|
9164 |
|
|
merge_outer_ops (enum rtx_code *pop0, HOST_WIDE_INT *pconst0, enum rtx_code op1, HOST_WIDE_INT const1, enum machine_mode mode, int *pcomp_p)
|
9165 |
|
|
{
|
9166 |
|
|
enum rtx_code op0 = *pop0;
|
9167 |
|
|
HOST_WIDE_INT const0 = *pconst0;
|
9168 |
|
|
|
9169 |
|
|
const0 &= GET_MODE_MASK (mode);
|
9170 |
|
|
const1 &= GET_MODE_MASK (mode);
|
9171 |
|
|
|
9172 |
|
|
/* If OP0 is an AND, clear unimportant bits in CONST1. */
|
9173 |
|
|
if (op0 == AND)
|
9174 |
|
|
const1 &= const0;
|
9175 |
|
|
|
9176 |
|
|
/* If OP0 or OP1 is UNKNOWN, this is easy. Similarly if they are the same or
|
9177 |
|
|
if OP0 is SET. */
|
9178 |
|
|
|
9179 |
|
|
if (op1 == UNKNOWN || op0 == SET)
|
9180 |
|
|
return 1;
|
9181 |
|
|
|
9182 |
|
|
else if (op0 == UNKNOWN)
|
9183 |
|
|
op0 = op1, const0 = const1;
|
9184 |
|
|
|
9185 |
|
|
else if (op0 == op1)
|
9186 |
|
|
{
|
9187 |
|
|
switch (op0)
|
9188 |
|
|
{
|
9189 |
|
|
case AND:
|
9190 |
|
|
const0 &= const1;
|
9191 |
|
|
break;
|
9192 |
|
|
case IOR:
|
9193 |
|
|
const0 |= const1;
|
9194 |
|
|
break;
|
9195 |
|
|
case XOR:
|
9196 |
|
|
const0 ^= const1;
|
9197 |
|
|
break;
|
9198 |
|
|
case PLUS:
|
9199 |
|
|
const0 += const1;
|
9200 |
|
|
break;
|
9201 |
|
|
case NEG:
|
9202 |
|
|
op0 = UNKNOWN;
|
9203 |
|
|
break;
|
9204 |
|
|
default:
|
9205 |
|
|
break;
|
9206 |
|
|
}
|
9207 |
|
|
}
|
9208 |
|
|
|
9209 |
|
|
/* Otherwise, if either is a PLUS or NEG, we can't do anything. */
|
9210 |
|
|
else if (op0 == PLUS || op1 == PLUS || op0 == NEG || op1 == NEG)
|
9211 |
|
|
return 0;
|
9212 |
|
|
|
9213 |
|
|
/* If the two constants aren't the same, we can't do anything. The
|
9214 |
|
|
remaining six cases can all be done. */
|
9215 |
|
|
else if (const0 != const1)
|
9216 |
|
|
return 0;
|
9217 |
|
|
|
9218 |
|
|
else
|
9219 |
|
|
switch (op0)
|
9220 |
|
|
{
|
9221 |
|
|
case IOR:
|
9222 |
|
|
if (op1 == AND)
|
9223 |
|
|
/* (a & b) | b == b */
|
9224 |
|
|
op0 = SET;
|
9225 |
|
|
else /* op1 == XOR */
|
9226 |
|
|
/* (a ^ b) | b == a | b */
|
9227 |
|
|
{;}
|
9228 |
|
|
break;
|
9229 |
|
|
|
9230 |
|
|
case XOR:
|
9231 |
|
|
if (op1 == AND)
|
9232 |
|
|
/* (a & b) ^ b == (~a) & b */
|
9233 |
|
|
op0 = AND, *pcomp_p = 1;
|
9234 |
|
|
else /* op1 == IOR */
|
9235 |
|
|
/* (a | b) ^ b == a & ~b */
|
9236 |
|
|
op0 = AND, const0 = ~const0;
|
9237 |
|
|
break;
|
9238 |
|
|
|
9239 |
|
|
case AND:
|
9240 |
|
|
if (op1 == IOR)
|
9241 |
|
|
/* (a | b) & b == b */
|
9242 |
|
|
op0 = SET;
|
9243 |
|
|
else /* op1 == XOR */
|
9244 |
|
|
/* (a ^ b) & b) == (~a) & b */
|
9245 |
|
|
*pcomp_p = 1;
|
9246 |
|
|
break;
|
9247 |
|
|
default:
|
9248 |
|
|
break;
|
9249 |
|
|
}
|
9250 |
|
|
|
9251 |
|
|
/* Check for NO-OP cases. */
|
9252 |
|
|
const0 &= GET_MODE_MASK (mode);
|
9253 |
|
|
if (const0 == 0
|
9254 |
|
|
&& (op0 == IOR || op0 == XOR || op0 == PLUS))
|
9255 |
|
|
op0 = UNKNOWN;
|
9256 |
|
|
else if (const0 == 0 && op0 == AND)
|
9257 |
|
|
op0 = SET;
|
9258 |
|
|
else if ((unsigned HOST_WIDE_INT) const0 == GET_MODE_MASK (mode)
|
9259 |
|
|
&& op0 == AND)
|
9260 |
|
|
op0 = UNKNOWN;
|
9261 |
|
|
|
9262 |
|
|
*pop0 = op0;
|
9263 |
|
|
|
9264 |
|
|
/* ??? Slightly redundant with the above mask, but not entirely.
|
9265 |
|
|
Moving this above means we'd have to sign-extend the mode mask
|
9266 |
|
|
for the final test. */
|
9267 |
|
|
if (op0 != UNKNOWN && op0 != NEG)
|
9268 |
|
|
*pconst0 = trunc_int_for_mode (const0, mode);
|
9269 |
|
|
|
9270 |
|
|
return 1;
|
9271 |
|
|
}
|
9272 |
|
|
|
9273 |
|
|
/* A helper to simplify_shift_const_1 to determine the mode we can perform
|
9274 |
|
|
the shift in. The original shift operation CODE is performed on OP in
|
9275 |
|
|
ORIG_MODE. Return the wider mode MODE if we can perform the operation
|
9276 |
|
|
in that mode. Return ORIG_MODE otherwise. We can also assume that the
|
9277 |
|
|
result of the shift is subject to operation OUTER_CODE with operand
|
9278 |
|
|
OUTER_CONST. */
|
9279 |
|
|
|
9280 |
|
|
static enum machine_mode
|
9281 |
|
|
try_widen_shift_mode (enum rtx_code code, rtx op, int count,
|
9282 |
|
|
enum machine_mode orig_mode, enum machine_mode mode,
|
9283 |
|
|
enum rtx_code outer_code, HOST_WIDE_INT outer_const)
|
9284 |
|
|
{
|
9285 |
|
|
if (orig_mode == mode)
|
9286 |
|
|
return mode;
|
9287 |
|
|
gcc_assert (GET_MODE_BITSIZE (mode) > GET_MODE_BITSIZE (orig_mode));
|
9288 |
|
|
|
9289 |
|
|
/* In general we can't perform in wider mode for right shift and rotate. */
|
9290 |
|
|
switch (code)
|
9291 |
|
|
{
|
9292 |
|
|
case ASHIFTRT:
|
9293 |
|
|
/* We can still widen if the bits brought in from the left are identical
|
9294 |
|
|
to the sign bit of ORIG_MODE. */
|
9295 |
|
|
if (num_sign_bit_copies (op, mode)
|
9296 |
|
|
> (unsigned) (GET_MODE_BITSIZE (mode)
|
9297 |
|
|
- GET_MODE_BITSIZE (orig_mode)))
|
9298 |
|
|
return mode;
|
9299 |
|
|
return orig_mode;
|
9300 |
|
|
|
9301 |
|
|
case LSHIFTRT:
|
9302 |
|
|
/* Similarly here but with zero bits. */
|
9303 |
|
|
if (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
9304 |
|
|
&& (nonzero_bits (op, mode) & ~GET_MODE_MASK (orig_mode)) == 0)
|
9305 |
|
|
return mode;
|
9306 |
|
|
|
9307 |
|
|
/* We can also widen if the bits brought in will be masked off. This
|
9308 |
|
|
operation is performed in ORIG_MODE. */
|
9309 |
|
|
if (outer_code == AND)
|
9310 |
|
|
{
|
9311 |
|
|
int care_bits = low_bitmask_len (orig_mode, outer_const);
|
9312 |
|
|
|
9313 |
|
|
if (care_bits >= 0
|
9314 |
|
|
&& GET_MODE_BITSIZE (orig_mode) - care_bits >= count)
|
9315 |
|
|
return mode;
|
9316 |
|
|
}
|
9317 |
|
|
/* fall through */
|
9318 |
|
|
|
9319 |
|
|
case ROTATE:
|
9320 |
|
|
return orig_mode;
|
9321 |
|
|
|
9322 |
|
|
case ROTATERT:
|
9323 |
|
|
gcc_unreachable ();
|
9324 |
|
|
|
9325 |
|
|
default:
|
9326 |
|
|
return mode;
|
9327 |
|
|
}
|
9328 |
|
|
}
|
9329 |
|
|
|
9330 |
|
|
/* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
|
9331 |
|
|
The result of the shift is RESULT_MODE. Return NULL_RTX if we cannot
|
9332 |
|
|
simplify it. Otherwise, return a simplified value.
|
9333 |
|
|
|
9334 |
|
|
The shift is normally computed in the widest mode we find in VAROP, as
|
9335 |
|
|
long as it isn't a different number of words than RESULT_MODE. Exceptions
|
9336 |
|
|
are ASHIFTRT and ROTATE, which are always done in their original mode. */
|
9337 |
|
|
|
9338 |
|
|
static rtx
|
9339 |
|
|
simplify_shift_const_1 (enum rtx_code code, enum machine_mode result_mode,
|
9340 |
|
|
rtx varop, int orig_count)
|
9341 |
|
|
{
|
9342 |
|
|
enum rtx_code orig_code = code;
|
9343 |
|
|
rtx orig_varop = varop;
|
9344 |
|
|
int count;
|
9345 |
|
|
enum machine_mode mode = result_mode;
|
9346 |
|
|
enum machine_mode shift_mode, tmode;
|
9347 |
|
|
unsigned int mode_words
|
9348 |
|
|
= (GET_MODE_SIZE (mode) + (UNITS_PER_WORD - 1)) / UNITS_PER_WORD;
|
9349 |
|
|
/* We form (outer_op (code varop count) (outer_const)). */
|
9350 |
|
|
enum rtx_code outer_op = UNKNOWN;
|
9351 |
|
|
HOST_WIDE_INT outer_const = 0;
|
9352 |
|
|
int complement_p = 0;
|
9353 |
|
|
rtx new_rtx, x;
|
9354 |
|
|
|
9355 |
|
|
/* Make sure and truncate the "natural" shift on the way in. We don't
|
9356 |
|
|
want to do this inside the loop as it makes it more difficult to
|
9357 |
|
|
combine shifts. */
|
9358 |
|
|
if (SHIFT_COUNT_TRUNCATED)
|
9359 |
|
|
orig_count &= GET_MODE_BITSIZE (mode) - 1;
|
9360 |
|
|
|
9361 |
|
|
/* If we were given an invalid count, don't do anything except exactly
|
9362 |
|
|
what was requested. */
|
9363 |
|
|
|
9364 |
|
|
if (orig_count < 0 || orig_count >= (int) GET_MODE_BITSIZE (mode))
|
9365 |
|
|
return NULL_RTX;
|
9366 |
|
|
|
9367 |
|
|
count = orig_count;
|
9368 |
|
|
|
9369 |
|
|
/* Unless one of the branches of the `if' in this loop does a `continue',
|
9370 |
|
|
we will `break' the loop after the `if'. */
|
9371 |
|
|
|
9372 |
|
|
while (count != 0)
|
9373 |
|
|
{
|
9374 |
|
|
/* If we have an operand of (clobber (const_int 0)), fail. */
|
9375 |
|
|
if (GET_CODE (varop) == CLOBBER)
|
9376 |
|
|
return NULL_RTX;
|
9377 |
|
|
|
9378 |
|
|
/* Convert ROTATERT to ROTATE. */
|
9379 |
|
|
if (code == ROTATERT)
|
9380 |
|
|
{
|
9381 |
|
|
unsigned int bitsize = GET_MODE_BITSIZE (result_mode);;
|
9382 |
|
|
code = ROTATE;
|
9383 |
|
|
if (VECTOR_MODE_P (result_mode))
|
9384 |
|
|
count = bitsize / GET_MODE_NUNITS (result_mode) - count;
|
9385 |
|
|
else
|
9386 |
|
|
count = bitsize - count;
|
9387 |
|
|
}
|
9388 |
|
|
|
9389 |
|
|
shift_mode = try_widen_shift_mode (code, varop, count, result_mode,
|
9390 |
|
|
mode, outer_op, outer_const);
|
9391 |
|
|
|
9392 |
|
|
/* Handle cases where the count is greater than the size of the mode
|
9393 |
|
|
minus 1. For ASHIFT, use the size minus one as the count (this can
|
9394 |
|
|
occur when simplifying (lshiftrt (ashiftrt ..))). For rotates,
|
9395 |
|
|
take the count modulo the size. For other shifts, the result is
|
9396 |
|
|
zero.
|
9397 |
|
|
|
9398 |
|
|
Since these shifts are being produced by the compiler by combining
|
9399 |
|
|
multiple operations, each of which are defined, we know what the
|
9400 |
|
|
result is supposed to be. */
|
9401 |
|
|
|
9402 |
|
|
if (count > (GET_MODE_BITSIZE (shift_mode) - 1))
|
9403 |
|
|
{
|
9404 |
|
|
if (code == ASHIFTRT)
|
9405 |
|
|
count = GET_MODE_BITSIZE (shift_mode) - 1;
|
9406 |
|
|
else if (code == ROTATE || code == ROTATERT)
|
9407 |
|
|
count %= GET_MODE_BITSIZE (shift_mode);
|
9408 |
|
|
else
|
9409 |
|
|
{
|
9410 |
|
|
/* We can't simply return zero because there may be an
|
9411 |
|
|
outer op. */
|
9412 |
|
|
varop = const0_rtx;
|
9413 |
|
|
count = 0;
|
9414 |
|
|
break;
|
9415 |
|
|
}
|
9416 |
|
|
}
|
9417 |
|
|
|
9418 |
|
|
/* If we discovered we had to complement VAROP, leave. Making a NOT
|
9419 |
|
|
here would cause an infinite loop. */
|
9420 |
|
|
if (complement_p)
|
9421 |
|
|
break;
|
9422 |
|
|
|
9423 |
|
|
/* An arithmetic right shift of a quantity known to be -1 or 0
|
9424 |
|
|
is a no-op. */
|
9425 |
|
|
if (code == ASHIFTRT
|
9426 |
|
|
&& (num_sign_bit_copies (varop, shift_mode)
|
9427 |
|
|
== GET_MODE_BITSIZE (shift_mode)))
|
9428 |
|
|
{
|
9429 |
|
|
count = 0;
|
9430 |
|
|
break;
|
9431 |
|
|
}
|
9432 |
|
|
|
9433 |
|
|
/* If we are doing an arithmetic right shift and discarding all but
|
9434 |
|
|
the sign bit copies, this is equivalent to doing a shift by the
|
9435 |
|
|
bitsize minus one. Convert it into that shift because it will often
|
9436 |
|
|
allow other simplifications. */
|
9437 |
|
|
|
9438 |
|
|
if (code == ASHIFTRT
|
9439 |
|
|
&& (count + num_sign_bit_copies (varop, shift_mode)
|
9440 |
|
|
>= GET_MODE_BITSIZE (shift_mode)))
|
9441 |
|
|
count = GET_MODE_BITSIZE (shift_mode) - 1;
|
9442 |
|
|
|
9443 |
|
|
/* We simplify the tests below and elsewhere by converting
|
9444 |
|
|
ASHIFTRT to LSHIFTRT if we know the sign bit is clear.
|
9445 |
|
|
`make_compound_operation' will convert it to an ASHIFTRT for
|
9446 |
|
|
those machines (such as VAX) that don't have an LSHIFTRT. */
|
9447 |
|
|
if (GET_MODE_BITSIZE (shift_mode) <= HOST_BITS_PER_WIDE_INT
|
9448 |
|
|
&& code == ASHIFTRT
|
9449 |
|
|
&& ((nonzero_bits (varop, shift_mode)
|
9450 |
|
|
& ((HOST_WIDE_INT) 1 << (GET_MODE_BITSIZE (shift_mode) - 1)))
|
9451 |
|
|
== 0))
|
9452 |
|
|
code = LSHIFTRT;
|
9453 |
|
|
|
9454 |
|
|
if (((code == LSHIFTRT
|
9455 |
|
|
&& GET_MODE_BITSIZE (shift_mode) <= HOST_BITS_PER_WIDE_INT
|
9456 |
|
|
&& !(nonzero_bits (varop, shift_mode) >> count))
|
9457 |
|
|
|| (code == ASHIFT
|
9458 |
|
|
&& GET_MODE_BITSIZE (shift_mode) <= HOST_BITS_PER_WIDE_INT
|
9459 |
|
|
&& !((nonzero_bits (varop, shift_mode) << count)
|
9460 |
|
|
& GET_MODE_MASK (shift_mode))))
|
9461 |
|
|
&& !side_effects_p (varop))
|
9462 |
|
|
varop = const0_rtx;
|
9463 |
|
|
|
9464 |
|
|
switch (GET_CODE (varop))
|
9465 |
|
|
{
|
9466 |
|
|
case SIGN_EXTEND:
|
9467 |
|
|
case ZERO_EXTEND:
|
9468 |
|
|
case SIGN_EXTRACT:
|
9469 |
|
|
case ZERO_EXTRACT:
|
9470 |
|
|
new_rtx = expand_compound_operation (varop);
|
9471 |
|
|
if (new_rtx != varop)
|
9472 |
|
|
{
|
9473 |
|
|
varop = new_rtx;
|
9474 |
|
|
continue;
|
9475 |
|
|
}
|
9476 |
|
|
break;
|
9477 |
|
|
|
9478 |
|
|
case MEM:
|
9479 |
|
|
/* If we have (xshiftrt (mem ...) C) and C is MODE_WIDTH
|
9480 |
|
|
minus the width of a smaller mode, we can do this with a
|
9481 |
|
|
SIGN_EXTEND or ZERO_EXTEND from the narrower memory location. */
|
9482 |
|
|
if ((code == ASHIFTRT || code == LSHIFTRT)
|
9483 |
|
|
&& ! mode_dependent_address_p (XEXP (varop, 0))
|
9484 |
|
|
&& ! MEM_VOLATILE_P (varop)
|
9485 |
|
|
&& (tmode = mode_for_size (GET_MODE_BITSIZE (mode) - count,
|
9486 |
|
|
MODE_INT, 1)) != BLKmode)
|
9487 |
|
|
{
|
9488 |
|
|
new_rtx = adjust_address_nv (varop, tmode,
|
9489 |
|
|
BYTES_BIG_ENDIAN ? 0
|
9490 |
|
|
: count / BITS_PER_UNIT);
|
9491 |
|
|
|
9492 |
|
|
varop = gen_rtx_fmt_e (code == ASHIFTRT ? SIGN_EXTEND
|
9493 |
|
|
: ZERO_EXTEND, mode, new_rtx);
|
9494 |
|
|
count = 0;
|
9495 |
|
|
continue;
|
9496 |
|
|
}
|
9497 |
|
|
break;
|
9498 |
|
|
|
9499 |
|
|
case SUBREG:
|
9500 |
|
|
/* If VAROP is a SUBREG, strip it as long as the inner operand has
|
9501 |
|
|
the same number of words as what we've seen so far. Then store
|
9502 |
|
|
the widest mode in MODE. */
|
9503 |
|
|
if (subreg_lowpart_p (varop)
|
9504 |
|
|
&& (GET_MODE_SIZE (GET_MODE (SUBREG_REG (varop)))
|
9505 |
|
|
> GET_MODE_SIZE (GET_MODE (varop)))
|
9506 |
|
|
&& (unsigned int) ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (varop)))
|
9507 |
|
|
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
|
9508 |
|
|
== mode_words)
|
9509 |
|
|
{
|
9510 |
|
|
varop = SUBREG_REG (varop);
|
9511 |
|
|
if (GET_MODE_SIZE (GET_MODE (varop)) > GET_MODE_SIZE (mode))
|
9512 |
|
|
mode = GET_MODE (varop);
|
9513 |
|
|
continue;
|
9514 |
|
|
}
|
9515 |
|
|
break;
|
9516 |
|
|
|
9517 |
|
|
case MULT:
|
9518 |
|
|
/* Some machines use MULT instead of ASHIFT because MULT
|
9519 |
|
|
is cheaper. But it is still better on those machines to
|
9520 |
|
|
merge two shifts into one. */
|
9521 |
|
|
if (CONST_INT_P (XEXP (varop, 1))
|
9522 |
|
|
&& exact_log2 (INTVAL (XEXP (varop, 1))) >= 0)
|
9523 |
|
|
{
|
9524 |
|
|
varop
|
9525 |
|
|
= simplify_gen_binary (ASHIFT, GET_MODE (varop),
|
9526 |
|
|
XEXP (varop, 0),
|
9527 |
|
|
GEN_INT (exact_log2 (
|
9528 |
|
|
INTVAL (XEXP (varop, 1)))));
|
9529 |
|
|
continue;
|
9530 |
|
|
}
|
9531 |
|
|
break;
|
9532 |
|
|
|
9533 |
|
|
case UDIV:
|
9534 |
|
|
/* Similar, for when divides are cheaper. */
|
9535 |
|
|
if (CONST_INT_P (XEXP (varop, 1))
|
9536 |
|
|
&& exact_log2 (INTVAL (XEXP (varop, 1))) >= 0)
|
9537 |
|
|
{
|
9538 |
|
|
varop
|
9539 |
|
|
= simplify_gen_binary (LSHIFTRT, GET_MODE (varop),
|
9540 |
|
|
XEXP (varop, 0),
|
9541 |
|
|
GEN_INT (exact_log2 (
|
9542 |
|
|
INTVAL (XEXP (varop, 1)))));
|
9543 |
|
|
continue;
|
9544 |
|
|
}
|
9545 |
|
|
break;
|
9546 |
|
|
|
9547 |
|
|
case ASHIFTRT:
|
9548 |
|
|
/* If we are extracting just the sign bit of an arithmetic
|
9549 |
|
|
right shift, that shift is not needed. However, the sign
|
9550 |
|
|
bit of a wider mode may be different from what would be
|
9551 |
|
|
interpreted as the sign bit in a narrower mode, so, if
|
9552 |
|
|
the result is narrower, don't discard the shift. */
|
9553 |
|
|
if (code == LSHIFTRT
|
9554 |
|
|
&& count == (GET_MODE_BITSIZE (result_mode) - 1)
|
9555 |
|
|
&& (GET_MODE_BITSIZE (result_mode)
|
9556 |
|
|
>= GET_MODE_BITSIZE (GET_MODE (varop))))
|
9557 |
|
|
{
|
9558 |
|
|
varop = XEXP (varop, 0);
|
9559 |
|
|
continue;
|
9560 |
|
|
}
|
9561 |
|
|
|
9562 |
|
|
/* ... fall through ... */
|
9563 |
|
|
|
9564 |
|
|
case LSHIFTRT:
|
9565 |
|
|
case ASHIFT:
|
9566 |
|
|
case ROTATE:
|
9567 |
|
|
/* Here we have two nested shifts. The result is usually the
|
9568 |
|
|
AND of a new shift with a mask. We compute the result below. */
|
9569 |
|
|
if (CONST_INT_P (XEXP (varop, 1))
|
9570 |
|
|
&& INTVAL (XEXP (varop, 1)) >= 0
|
9571 |
|
|
&& INTVAL (XEXP (varop, 1)) < GET_MODE_BITSIZE (GET_MODE (varop))
|
9572 |
|
|
&& GET_MODE_BITSIZE (result_mode) <= HOST_BITS_PER_WIDE_INT
|
9573 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
9574 |
|
|
&& !VECTOR_MODE_P (result_mode))
|
9575 |
|
|
{
|
9576 |
|
|
enum rtx_code first_code = GET_CODE (varop);
|
9577 |
|
|
unsigned int first_count = INTVAL (XEXP (varop, 1));
|
9578 |
|
|
unsigned HOST_WIDE_INT mask;
|
9579 |
|
|
rtx mask_rtx;
|
9580 |
|
|
|
9581 |
|
|
/* We have one common special case. We can't do any merging if
|
9582 |
|
|
the inner code is an ASHIFTRT of a smaller mode. However, if
|
9583 |
|
|
we have (ashift:M1 (subreg:M1 (ashiftrt:M2 FOO C1) 0) C2)
|
9584 |
|
|
with C2 == GET_MODE_BITSIZE (M1) - GET_MODE_BITSIZE (M2),
|
9585 |
|
|
we can convert it to
|
9586 |
|
|
(ashiftrt:M1 (ashift:M1 (and:M1 (subreg:M1 FOO 0 C2) C3) C1).
|
9587 |
|
|
This simplifies certain SIGN_EXTEND operations. */
|
9588 |
|
|
if (code == ASHIFT && first_code == ASHIFTRT
|
9589 |
|
|
&& count == (GET_MODE_BITSIZE (result_mode)
|
9590 |
|
|
- GET_MODE_BITSIZE (GET_MODE (varop))))
|
9591 |
|
|
{
|
9592 |
|
|
/* C3 has the low-order C1 bits zero. */
|
9593 |
|
|
|
9594 |
|
|
mask = (GET_MODE_MASK (mode)
|
9595 |
|
|
& ~(((HOST_WIDE_INT) 1 << first_count) - 1));
|
9596 |
|
|
|
9597 |
|
|
varop = simplify_and_const_int (NULL_RTX, result_mode,
|
9598 |
|
|
XEXP (varop, 0), mask);
|
9599 |
|
|
varop = simplify_shift_const (NULL_RTX, ASHIFT, result_mode,
|
9600 |
|
|
varop, count);
|
9601 |
|
|
count = first_count;
|
9602 |
|
|
code = ASHIFTRT;
|
9603 |
|
|
continue;
|
9604 |
|
|
}
|
9605 |
|
|
|
9606 |
|
|
/* If this was (ashiftrt (ashift foo C1) C2) and FOO has more
|
9607 |
|
|
than C1 high-order bits equal to the sign bit, we can convert
|
9608 |
|
|
this to either an ASHIFT or an ASHIFTRT depending on the
|
9609 |
|
|
two counts.
|
9610 |
|
|
|
9611 |
|
|
We cannot do this if VAROP's mode is not SHIFT_MODE. */
|
9612 |
|
|
|
9613 |
|
|
if (code == ASHIFTRT && first_code == ASHIFT
|
9614 |
|
|
&& GET_MODE (varop) == shift_mode
|
9615 |
|
|
&& (num_sign_bit_copies (XEXP (varop, 0), shift_mode)
|
9616 |
|
|
> first_count))
|
9617 |
|
|
{
|
9618 |
|
|
varop = XEXP (varop, 0);
|
9619 |
|
|
count -= first_count;
|
9620 |
|
|
if (count < 0)
|
9621 |
|
|
{
|
9622 |
|
|
count = -count;
|
9623 |
|
|
code = ASHIFT;
|
9624 |
|
|
}
|
9625 |
|
|
|
9626 |
|
|
continue;
|
9627 |
|
|
}
|
9628 |
|
|
|
9629 |
|
|
/* There are some cases we can't do. If CODE is ASHIFTRT,
|
9630 |
|
|
we can only do this if FIRST_CODE is also ASHIFTRT.
|
9631 |
|
|
|
9632 |
|
|
We can't do the case when CODE is ROTATE and FIRST_CODE is
|
9633 |
|
|
ASHIFTRT.
|
9634 |
|
|
|
9635 |
|
|
If the mode of this shift is not the mode of the outer shift,
|
9636 |
|
|
we can't do this if either shift is a right shift or ROTATE.
|
9637 |
|
|
|
9638 |
|
|
Finally, we can't do any of these if the mode is too wide
|
9639 |
|
|
unless the codes are the same.
|
9640 |
|
|
|
9641 |
|
|
Handle the case where the shift codes are the same
|
9642 |
|
|
first. */
|
9643 |
|
|
|
9644 |
|
|
if (code == first_code)
|
9645 |
|
|
{
|
9646 |
|
|
if (GET_MODE (varop) != result_mode
|
9647 |
|
|
&& (code == ASHIFTRT || code == LSHIFTRT
|
9648 |
|
|
|| code == ROTATE))
|
9649 |
|
|
break;
|
9650 |
|
|
|
9651 |
|
|
count += first_count;
|
9652 |
|
|
varop = XEXP (varop, 0);
|
9653 |
|
|
continue;
|
9654 |
|
|
}
|
9655 |
|
|
|
9656 |
|
|
if (code == ASHIFTRT
|
9657 |
|
|
|| (code == ROTATE && first_code == ASHIFTRT)
|
9658 |
|
|
|| GET_MODE_BITSIZE (mode) > HOST_BITS_PER_WIDE_INT
|
9659 |
|
|
|| (GET_MODE (varop) != result_mode
|
9660 |
|
|
&& (first_code == ASHIFTRT || first_code == LSHIFTRT
|
9661 |
|
|
|| first_code == ROTATE
|
9662 |
|
|
|| code == ROTATE)))
|
9663 |
|
|
break;
|
9664 |
|
|
|
9665 |
|
|
/* To compute the mask to apply after the shift, shift the
|
9666 |
|
|
nonzero bits of the inner shift the same way the
|
9667 |
|
|
outer shift will. */
|
9668 |
|
|
|
9669 |
|
|
mask_rtx = GEN_INT (nonzero_bits (varop, GET_MODE (varop)));
|
9670 |
|
|
|
9671 |
|
|
mask_rtx
|
9672 |
|
|
= simplify_const_binary_operation (code, result_mode, mask_rtx,
|
9673 |
|
|
GEN_INT (count));
|
9674 |
|
|
|
9675 |
|
|
/* Give up if we can't compute an outer operation to use. */
|
9676 |
|
|
if (mask_rtx == 0
|
9677 |
|
|
|| !CONST_INT_P (mask_rtx)
|
9678 |
|
|
|| ! merge_outer_ops (&outer_op, &outer_const, AND,
|
9679 |
|
|
INTVAL (mask_rtx),
|
9680 |
|
|
result_mode, &complement_p))
|
9681 |
|
|
break;
|
9682 |
|
|
|
9683 |
|
|
/* If the shifts are in the same direction, we add the
|
9684 |
|
|
counts. Otherwise, we subtract them. */
|
9685 |
|
|
if ((code == ASHIFTRT || code == LSHIFTRT)
|
9686 |
|
|
== (first_code == ASHIFTRT || first_code == LSHIFTRT))
|
9687 |
|
|
count += first_count;
|
9688 |
|
|
else
|
9689 |
|
|
count -= first_count;
|
9690 |
|
|
|
9691 |
|
|
/* If COUNT is positive, the new shift is usually CODE,
|
9692 |
|
|
except for the two exceptions below, in which case it is
|
9693 |
|
|
FIRST_CODE. If the count is negative, FIRST_CODE should
|
9694 |
|
|
always be used */
|
9695 |
|
|
if (count > 0
|
9696 |
|
|
&& ((first_code == ROTATE && code == ASHIFT)
|
9697 |
|
|
|| (first_code == ASHIFTRT && code == LSHIFTRT)))
|
9698 |
|
|
code = first_code;
|
9699 |
|
|
else if (count < 0)
|
9700 |
|
|
code = first_code, count = -count;
|
9701 |
|
|
|
9702 |
|
|
varop = XEXP (varop, 0);
|
9703 |
|
|
continue;
|
9704 |
|
|
}
|
9705 |
|
|
|
9706 |
|
|
/* If we have (A << B << C) for any shift, we can convert this to
|
9707 |
|
|
(A << C << B). This wins if A is a constant. Only try this if
|
9708 |
|
|
B is not a constant. */
|
9709 |
|
|
|
9710 |
|
|
else if (GET_CODE (varop) == code
|
9711 |
|
|
&& CONST_INT_P (XEXP (varop, 0))
|
9712 |
|
|
&& !CONST_INT_P (XEXP (varop, 1)))
|
9713 |
|
|
{
|
9714 |
|
|
rtx new_rtx = simplify_const_binary_operation (code, mode,
|
9715 |
|
|
XEXP (varop, 0),
|
9716 |
|
|
GEN_INT (count));
|
9717 |
|
|
varop = gen_rtx_fmt_ee (code, mode, new_rtx, XEXP (varop, 1));
|
9718 |
|
|
count = 0;
|
9719 |
|
|
continue;
|
9720 |
|
|
}
|
9721 |
|
|
break;
|
9722 |
|
|
|
9723 |
|
|
case NOT:
|
9724 |
|
|
if (VECTOR_MODE_P (mode))
|
9725 |
|
|
break;
|
9726 |
|
|
|
9727 |
|
|
/* Make this fit the case below. */
|
9728 |
|
|
varop = gen_rtx_XOR (mode, XEXP (varop, 0),
|
9729 |
|
|
GEN_INT (GET_MODE_MASK (mode)));
|
9730 |
|
|
continue;
|
9731 |
|
|
|
9732 |
|
|
case IOR:
|
9733 |
|
|
case AND:
|
9734 |
|
|
case XOR:
|
9735 |
|
|
/* If we have (xshiftrt (ior (plus X (const_int -1)) X) C)
|
9736 |
|
|
with C the size of VAROP - 1 and the shift is logical if
|
9737 |
|
|
STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
|
9738 |
|
|
we have an (le X 0) operation. If we have an arithmetic shift
|
9739 |
|
|
and STORE_FLAG_VALUE is 1 or we have a logical shift with
|
9740 |
|
|
STORE_FLAG_VALUE of -1, we have a (neg (le X 0)) operation. */
|
9741 |
|
|
|
9742 |
|
|
if (GET_CODE (varop) == IOR && GET_CODE (XEXP (varop, 0)) == PLUS
|
9743 |
|
|
&& XEXP (XEXP (varop, 0), 1) == constm1_rtx
|
9744 |
|
|
&& (STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
|
9745 |
|
|
&& (code == LSHIFTRT || code == ASHIFTRT)
|
9746 |
|
|
&& count == (GET_MODE_BITSIZE (GET_MODE (varop)) - 1)
|
9747 |
|
|
&& rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
|
9748 |
|
|
{
|
9749 |
|
|
count = 0;
|
9750 |
|
|
varop = gen_rtx_LE (GET_MODE (varop), XEXP (varop, 1),
|
9751 |
|
|
const0_rtx);
|
9752 |
|
|
|
9753 |
|
|
if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
|
9754 |
|
|
varop = gen_rtx_NEG (GET_MODE (varop), varop);
|
9755 |
|
|
|
9756 |
|
|
continue;
|
9757 |
|
|
}
|
9758 |
|
|
|
9759 |
|
|
/* If we have (shift (logical)), move the logical to the outside
|
9760 |
|
|
to allow it to possibly combine with another logical and the
|
9761 |
|
|
shift to combine with another shift. This also canonicalizes to
|
9762 |
|
|
what a ZERO_EXTRACT looks like. Also, some machines have
|
9763 |
|
|
(and (shift)) insns. */
|
9764 |
|
|
|
9765 |
|
|
if (CONST_INT_P (XEXP (varop, 1))
|
9766 |
|
|
/* We can't do this if we have (ashiftrt (xor)) and the
|
9767 |
|
|
constant has its sign bit set in shift_mode. */
|
9768 |
|
|
&& !(code == ASHIFTRT && GET_CODE (varop) == XOR
|
9769 |
|
|
&& 0 > trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
|
9770 |
|
|
shift_mode))
|
9771 |
|
|
&& (new_rtx = simplify_const_binary_operation (code, result_mode,
|
9772 |
|
|
XEXP (varop, 1),
|
9773 |
|
|
GEN_INT (count))) != 0
|
9774 |
|
|
&& CONST_INT_P (new_rtx)
|
9775 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, GET_CODE (varop),
|
9776 |
|
|
INTVAL (new_rtx), result_mode, &complement_p))
|
9777 |
|
|
{
|
9778 |
|
|
varop = XEXP (varop, 0);
|
9779 |
|
|
continue;
|
9780 |
|
|
}
|
9781 |
|
|
|
9782 |
|
|
/* If we can't do that, try to simplify the shift in each arm of the
|
9783 |
|
|
logical expression, make a new logical expression, and apply
|
9784 |
|
|
the inverse distributive law. This also can't be done
|
9785 |
|
|
for some (ashiftrt (xor)). */
|
9786 |
|
|
if (CONST_INT_P (XEXP (varop, 1))
|
9787 |
|
|
&& !(code == ASHIFTRT && GET_CODE (varop) == XOR
|
9788 |
|
|
&& 0 > trunc_int_for_mode (INTVAL (XEXP (varop, 1)),
|
9789 |
|
|
shift_mode)))
|
9790 |
|
|
{
|
9791 |
|
|
rtx lhs = simplify_shift_const (NULL_RTX, code, shift_mode,
|
9792 |
|
|
XEXP (varop, 0), count);
|
9793 |
|
|
rtx rhs = simplify_shift_const (NULL_RTX, code, shift_mode,
|
9794 |
|
|
XEXP (varop, 1), count);
|
9795 |
|
|
|
9796 |
|
|
varop = simplify_gen_binary (GET_CODE (varop), shift_mode,
|
9797 |
|
|
lhs, rhs);
|
9798 |
|
|
varop = apply_distributive_law (varop);
|
9799 |
|
|
|
9800 |
|
|
count = 0;
|
9801 |
|
|
continue;
|
9802 |
|
|
}
|
9803 |
|
|
break;
|
9804 |
|
|
|
9805 |
|
|
case EQ:
|
9806 |
|
|
/* Convert (lshiftrt (eq FOO 0) C) to (xor FOO 1) if STORE_FLAG_VALUE
|
9807 |
|
|
says that the sign bit can be tested, FOO has mode MODE, C is
|
9808 |
|
|
GET_MODE_BITSIZE (MODE) - 1, and FOO has only its low-order bit
|
9809 |
|
|
that may be nonzero. */
|
9810 |
|
|
if (code == LSHIFTRT
|
9811 |
|
|
&& XEXP (varop, 1) == const0_rtx
|
9812 |
|
|
&& GET_MODE (XEXP (varop, 0)) == result_mode
|
9813 |
|
|
&& count == (GET_MODE_BITSIZE (result_mode) - 1)
|
9814 |
|
|
&& GET_MODE_BITSIZE (result_mode) <= HOST_BITS_PER_WIDE_INT
|
9815 |
|
|
&& STORE_FLAG_VALUE == -1
|
9816 |
|
|
&& nonzero_bits (XEXP (varop, 0), result_mode) == 1
|
9817 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, XOR,
|
9818 |
|
|
(HOST_WIDE_INT) 1, result_mode,
|
9819 |
|
|
&complement_p))
|
9820 |
|
|
{
|
9821 |
|
|
varop = XEXP (varop, 0);
|
9822 |
|
|
count = 0;
|
9823 |
|
|
continue;
|
9824 |
|
|
}
|
9825 |
|
|
break;
|
9826 |
|
|
|
9827 |
|
|
case NEG:
|
9828 |
|
|
/* (lshiftrt (neg A) C) where A is either 0 or 1 and C is one less
|
9829 |
|
|
than the number of bits in the mode is equivalent to A. */
|
9830 |
|
|
if (code == LSHIFTRT
|
9831 |
|
|
&& count == (GET_MODE_BITSIZE (result_mode) - 1)
|
9832 |
|
|
&& nonzero_bits (XEXP (varop, 0), result_mode) == 1)
|
9833 |
|
|
{
|
9834 |
|
|
varop = XEXP (varop, 0);
|
9835 |
|
|
count = 0;
|
9836 |
|
|
continue;
|
9837 |
|
|
}
|
9838 |
|
|
|
9839 |
|
|
/* NEG commutes with ASHIFT since it is multiplication. Move the
|
9840 |
|
|
NEG outside to allow shifts to combine. */
|
9841 |
|
|
if (code == ASHIFT
|
9842 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, NEG,
|
9843 |
|
|
(HOST_WIDE_INT) 0, result_mode,
|
9844 |
|
|
&complement_p))
|
9845 |
|
|
{
|
9846 |
|
|
varop = XEXP (varop, 0);
|
9847 |
|
|
continue;
|
9848 |
|
|
}
|
9849 |
|
|
break;
|
9850 |
|
|
|
9851 |
|
|
case PLUS:
|
9852 |
|
|
/* (lshiftrt (plus A -1) C) where A is either 0 or 1 and C
|
9853 |
|
|
is one less than the number of bits in the mode is
|
9854 |
|
|
equivalent to (xor A 1). */
|
9855 |
|
|
if (code == LSHIFTRT
|
9856 |
|
|
&& count == (GET_MODE_BITSIZE (result_mode) - 1)
|
9857 |
|
|
&& XEXP (varop, 1) == constm1_rtx
|
9858 |
|
|
&& nonzero_bits (XEXP (varop, 0), result_mode) == 1
|
9859 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, XOR,
|
9860 |
|
|
(HOST_WIDE_INT) 1, result_mode,
|
9861 |
|
|
&complement_p))
|
9862 |
|
|
{
|
9863 |
|
|
count = 0;
|
9864 |
|
|
varop = XEXP (varop, 0);
|
9865 |
|
|
continue;
|
9866 |
|
|
}
|
9867 |
|
|
|
9868 |
|
|
/* If we have (xshiftrt (plus FOO BAR) C), and the only bits
|
9869 |
|
|
that might be nonzero in BAR are those being shifted out and those
|
9870 |
|
|
bits are known zero in FOO, we can replace the PLUS with FOO.
|
9871 |
|
|
Similarly in the other operand order. This code occurs when
|
9872 |
|
|
we are computing the size of a variable-size array. */
|
9873 |
|
|
|
9874 |
|
|
if ((code == ASHIFTRT || code == LSHIFTRT)
|
9875 |
|
|
&& count < HOST_BITS_PER_WIDE_INT
|
9876 |
|
|
&& nonzero_bits (XEXP (varop, 1), result_mode) >> count == 0
|
9877 |
|
|
&& (nonzero_bits (XEXP (varop, 1), result_mode)
|
9878 |
|
|
& nonzero_bits (XEXP (varop, 0), result_mode)) == 0)
|
9879 |
|
|
{
|
9880 |
|
|
varop = XEXP (varop, 0);
|
9881 |
|
|
continue;
|
9882 |
|
|
}
|
9883 |
|
|
else if ((code == ASHIFTRT || code == LSHIFTRT)
|
9884 |
|
|
&& count < HOST_BITS_PER_WIDE_INT
|
9885 |
|
|
&& GET_MODE_BITSIZE (result_mode) <= HOST_BITS_PER_WIDE_INT
|
9886 |
|
|
&& 0 == (nonzero_bits (XEXP (varop, 0), result_mode)
|
9887 |
|
|
>> count)
|
9888 |
|
|
&& 0 == (nonzero_bits (XEXP (varop, 0), result_mode)
|
9889 |
|
|
& nonzero_bits (XEXP (varop, 1),
|
9890 |
|
|
result_mode)))
|
9891 |
|
|
{
|
9892 |
|
|
varop = XEXP (varop, 1);
|
9893 |
|
|
continue;
|
9894 |
|
|
}
|
9895 |
|
|
|
9896 |
|
|
/* (ashift (plus foo C) N) is (plus (ashift foo N) C'). */
|
9897 |
|
|
if (code == ASHIFT
|
9898 |
|
|
&& CONST_INT_P (XEXP (varop, 1))
|
9899 |
|
|
&& (new_rtx = simplify_const_binary_operation (ASHIFT, result_mode,
|
9900 |
|
|
XEXP (varop, 1),
|
9901 |
|
|
GEN_INT (count))) != 0
|
9902 |
|
|
&& CONST_INT_P (new_rtx)
|
9903 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, PLUS,
|
9904 |
|
|
INTVAL (new_rtx), result_mode, &complement_p))
|
9905 |
|
|
{
|
9906 |
|
|
varop = XEXP (varop, 0);
|
9907 |
|
|
continue;
|
9908 |
|
|
}
|
9909 |
|
|
|
9910 |
|
|
/* Check for 'PLUS signbit', which is the canonical form of 'XOR
|
9911 |
|
|
signbit', and attempt to change the PLUS to an XOR and move it to
|
9912 |
|
|
the outer operation as is done above in the AND/IOR/XOR case
|
9913 |
|
|
leg for shift(logical). See details in logical handling above
|
9914 |
|
|
for reasoning in doing so. */
|
9915 |
|
|
if (code == LSHIFTRT
|
9916 |
|
|
&& CONST_INT_P (XEXP (varop, 1))
|
9917 |
|
|
&& mode_signbit_p (result_mode, XEXP (varop, 1))
|
9918 |
|
|
&& (new_rtx = simplify_const_binary_operation (code, result_mode,
|
9919 |
|
|
XEXP (varop, 1),
|
9920 |
|
|
GEN_INT (count))) != 0
|
9921 |
|
|
&& CONST_INT_P (new_rtx)
|
9922 |
|
|
&& merge_outer_ops (&outer_op, &outer_const, XOR,
|
9923 |
|
|
INTVAL (new_rtx), result_mode, &complement_p))
|
9924 |
|
|
{
|
9925 |
|
|
varop = XEXP (varop, 0);
|
9926 |
|
|
continue;
|
9927 |
|
|
}
|
9928 |
|
|
|
9929 |
|
|
break;
|
9930 |
|
|
|
9931 |
|
|
case MINUS:
|
9932 |
|
|
/* If we have (xshiftrt (minus (ashiftrt X C)) X) C)
|
9933 |
|
|
with C the size of VAROP - 1 and the shift is logical if
|
9934 |
|
|
STORE_FLAG_VALUE is 1 and arithmetic if STORE_FLAG_VALUE is -1,
|
9935 |
|
|
we have a (gt X 0) operation. If the shift is arithmetic with
|
9936 |
|
|
STORE_FLAG_VALUE of 1 or logical with STORE_FLAG_VALUE == -1,
|
9937 |
|
|
we have a (neg (gt X 0)) operation. */
|
9938 |
|
|
|
9939 |
|
|
if ((STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
|
9940 |
|
|
&& GET_CODE (XEXP (varop, 0)) == ASHIFTRT
|
9941 |
|
|
&& count == (GET_MODE_BITSIZE (GET_MODE (varop)) - 1)
|
9942 |
|
|
&& (code == LSHIFTRT || code == ASHIFTRT)
|
9943 |
|
|
&& CONST_INT_P (XEXP (XEXP (varop, 0), 1))
|
9944 |
|
|
&& INTVAL (XEXP (XEXP (varop, 0), 1)) == count
|
9945 |
|
|
&& rtx_equal_p (XEXP (XEXP (varop, 0), 0), XEXP (varop, 1)))
|
9946 |
|
|
{
|
9947 |
|
|
count = 0;
|
9948 |
|
|
varop = gen_rtx_GT (GET_MODE (varop), XEXP (varop, 1),
|
9949 |
|
|
const0_rtx);
|
9950 |
|
|
|
9951 |
|
|
if (STORE_FLAG_VALUE == 1 ? code == ASHIFTRT : code == LSHIFTRT)
|
9952 |
|
|
varop = gen_rtx_NEG (GET_MODE (varop), varop);
|
9953 |
|
|
|
9954 |
|
|
continue;
|
9955 |
|
|
}
|
9956 |
|
|
break;
|
9957 |
|
|
|
9958 |
|
|
case TRUNCATE:
|
9959 |
|
|
/* Change (lshiftrt (truncate (lshiftrt))) to (truncate (lshiftrt))
|
9960 |
|
|
if the truncate does not affect the value. */
|
9961 |
|
|
if (code == LSHIFTRT
|
9962 |
|
|
&& GET_CODE (XEXP (varop, 0)) == LSHIFTRT
|
9963 |
|
|
&& CONST_INT_P (XEXP (XEXP (varop, 0), 1))
|
9964 |
|
|
&& (INTVAL (XEXP (XEXP (varop, 0), 1))
|
9965 |
|
|
>= (GET_MODE_BITSIZE (GET_MODE (XEXP (varop, 0)))
|
9966 |
|
|
- GET_MODE_BITSIZE (GET_MODE (varop)))))
|
9967 |
|
|
{
|
9968 |
|
|
rtx varop_inner = XEXP (varop, 0);
|
9969 |
|
|
|
9970 |
|
|
varop_inner
|
9971 |
|
|
= gen_rtx_LSHIFTRT (GET_MODE (varop_inner),
|
9972 |
|
|
XEXP (varop_inner, 0),
|
9973 |
|
|
GEN_INT
|
9974 |
|
|
(count + INTVAL (XEXP (varop_inner, 1))));
|
9975 |
|
|
varop = gen_rtx_TRUNCATE (GET_MODE (varop), varop_inner);
|
9976 |
|
|
count = 0;
|
9977 |
|
|
continue;
|
9978 |
|
|
}
|
9979 |
|
|
break;
|
9980 |
|
|
|
9981 |
|
|
default:
|
9982 |
|
|
break;
|
9983 |
|
|
}
|
9984 |
|
|
|
9985 |
|
|
break;
|
9986 |
|
|
}
|
9987 |
|
|
|
9988 |
|
|
shift_mode = try_widen_shift_mode (code, varop, count, result_mode, mode,
|
9989 |
|
|
outer_op, outer_const);
|
9990 |
|
|
|
9991 |
|
|
/* We have now finished analyzing the shift. The result should be
|
9992 |
|
|
a shift of type CODE with SHIFT_MODE shifting VAROP COUNT places. If
|
9993 |
|
|
OUTER_OP is non-UNKNOWN, it is an operation that needs to be applied
|
9994 |
|
|
to the result of the shift. OUTER_CONST is the relevant constant,
|
9995 |
|
|
but we must turn off all bits turned off in the shift. */
|
9996 |
|
|
|
9997 |
|
|
if (outer_op == UNKNOWN
|
9998 |
|
|
&& orig_code == code && orig_count == count
|
9999 |
|
|
&& varop == orig_varop
|
10000 |
|
|
&& shift_mode == GET_MODE (varop))
|
10001 |
|
|
return NULL_RTX;
|
10002 |
|
|
|
10003 |
|
|
/* Make a SUBREG if necessary. If we can't make it, fail. */
|
10004 |
|
|
varop = gen_lowpart (shift_mode, varop);
|
10005 |
|
|
if (varop == NULL_RTX || GET_CODE (varop) == CLOBBER)
|
10006 |
|
|
return NULL_RTX;
|
10007 |
|
|
|
10008 |
|
|
/* If we have an outer operation and we just made a shift, it is
|
10009 |
|
|
possible that we could have simplified the shift were it not
|
10010 |
|
|
for the outer operation. So try to do the simplification
|
10011 |
|
|
recursively. */
|
10012 |
|
|
|
10013 |
|
|
if (outer_op != UNKNOWN)
|
10014 |
|
|
x = simplify_shift_const_1 (code, shift_mode, varop, count);
|
10015 |
|
|
else
|
10016 |
|
|
x = NULL_RTX;
|
10017 |
|
|
|
10018 |
|
|
if (x == NULL_RTX)
|
10019 |
|
|
x = simplify_gen_binary (code, shift_mode, varop, GEN_INT (count));
|
10020 |
|
|
|
10021 |
|
|
/* If we were doing an LSHIFTRT in a wider mode than it was originally,
|
10022 |
|
|
turn off all the bits that the shift would have turned off. */
|
10023 |
|
|
if (orig_code == LSHIFTRT && result_mode != shift_mode)
|
10024 |
|
|
x = simplify_and_const_int (NULL_RTX, shift_mode, x,
|
10025 |
|
|
GET_MODE_MASK (result_mode) >> orig_count);
|
10026 |
|
|
|
10027 |
|
|
/* Do the remainder of the processing in RESULT_MODE. */
|
10028 |
|
|
x = gen_lowpart_or_truncate (result_mode, x);
|
10029 |
|
|
|
10030 |
|
|
/* If COMPLEMENT_P is set, we have to complement X before doing the outer
|
10031 |
|
|
operation. */
|
10032 |
|
|
if (complement_p)
|
10033 |
|
|
x = simplify_gen_unary (NOT, result_mode, x, result_mode);
|
10034 |
|
|
|
10035 |
|
|
if (outer_op != UNKNOWN)
|
10036 |
|
|
{
|
10037 |
|
|
if (GET_RTX_CLASS (outer_op) != RTX_UNARY
|
10038 |
|
|
&& GET_MODE_BITSIZE (result_mode) < HOST_BITS_PER_WIDE_INT)
|
10039 |
|
|
outer_const = trunc_int_for_mode (outer_const, result_mode);
|
10040 |
|
|
|
10041 |
|
|
if (outer_op == AND)
|
10042 |
|
|
x = simplify_and_const_int (NULL_RTX, result_mode, x, outer_const);
|
10043 |
|
|
else if (outer_op == SET)
|
10044 |
|
|
{
|
10045 |
|
|
/* This means that we have determined that the result is
|
10046 |
|
|
equivalent to a constant. This should be rare. */
|
10047 |
|
|
if (!side_effects_p (x))
|
10048 |
|
|
x = GEN_INT (outer_const);
|
10049 |
|
|
}
|
10050 |
|
|
else if (GET_RTX_CLASS (outer_op) == RTX_UNARY)
|
10051 |
|
|
x = simplify_gen_unary (outer_op, result_mode, x, result_mode);
|
10052 |
|
|
else
|
10053 |
|
|
x = simplify_gen_binary (outer_op, result_mode, x,
|
10054 |
|
|
GEN_INT (outer_const));
|
10055 |
|
|
}
|
10056 |
|
|
|
10057 |
|
|
return x;
|
10058 |
|
|
}
|
10059 |
|
|
|
10060 |
|
|
/* Simplify a shift of VAROP by COUNT bits. CODE says what kind of shift.
|
10061 |
|
|
The result of the shift is RESULT_MODE. If we cannot simplify it,
|
10062 |
|
|
return X or, if it is NULL, synthesize the expression with
|
10063 |
|
|
simplify_gen_binary. Otherwise, return a simplified value.
|
10064 |
|
|
|
10065 |
|
|
The shift is normally computed in the widest mode we find in VAROP, as
|
10066 |
|
|
long as it isn't a different number of words than RESULT_MODE. Exceptions
|
10067 |
|
|
are ASHIFTRT and ROTATE, which are always done in their original mode. */
|
10068 |
|
|
|
10069 |
|
|
static rtx
|
10070 |
|
|
simplify_shift_const (rtx x, enum rtx_code code, enum machine_mode result_mode,
|
10071 |
|
|
rtx varop, int count)
|
10072 |
|
|
{
|
10073 |
|
|
rtx tem = simplify_shift_const_1 (code, result_mode, varop, count);
|
10074 |
|
|
if (tem)
|
10075 |
|
|
return tem;
|
10076 |
|
|
|
10077 |
|
|
if (!x)
|
10078 |
|
|
x = simplify_gen_binary (code, GET_MODE (varop), varop, GEN_INT (count));
|
10079 |
|
|
if (GET_MODE (x) != result_mode)
|
10080 |
|
|
x = gen_lowpart (result_mode, x);
|
10081 |
|
|
return x;
|
10082 |
|
|
}
|
10083 |
|
|
|
10084 |
|
|
|
10085 |
|
|
/* Like recog, but we receive the address of a pointer to a new pattern.
|
10086 |
|
|
We try to match the rtx that the pointer points to.
|
10087 |
|
|
If that fails, we may try to modify or replace the pattern,
|
10088 |
|
|
storing the replacement into the same pointer object.
|
10089 |
|
|
|
10090 |
|
|
Modifications include deletion or addition of CLOBBERs.
|
10091 |
|
|
|
10092 |
|
|
PNOTES is a pointer to a location where any REG_UNUSED notes added for
|
10093 |
|
|
the CLOBBERs are placed.
|
10094 |
|
|
|
10095 |
|
|
The value is the final insn code from the pattern ultimately matched,
|
10096 |
|
|
or -1. */
|
10097 |
|
|
|
10098 |
|
|
static int
|
10099 |
|
|
recog_for_combine (rtx *pnewpat, rtx insn, rtx *pnotes)
|
10100 |
|
|
{
|
10101 |
|
|
rtx pat = *pnewpat;
|
10102 |
|
|
int insn_code_number;
|
10103 |
|
|
int num_clobbers_to_add = 0;
|
10104 |
|
|
int i;
|
10105 |
|
|
rtx notes = 0;
|
10106 |
|
|
rtx old_notes, old_pat;
|
10107 |
|
|
|
10108 |
|
|
/* If PAT is a PARALLEL, check to see if it contains the CLOBBER
|
10109 |
|
|
we use to indicate that something didn't match. If we find such a
|
10110 |
|
|
thing, force rejection. */
|
10111 |
|
|
if (GET_CODE (pat) == PARALLEL)
|
10112 |
|
|
for (i = XVECLEN (pat, 0) - 1; i >= 0; i--)
|
10113 |
|
|
if (GET_CODE (XVECEXP (pat, 0, i)) == CLOBBER
|
10114 |
|
|
&& XEXP (XVECEXP (pat, 0, i), 0) == const0_rtx)
|
10115 |
|
|
return -1;
|
10116 |
|
|
|
10117 |
|
|
old_pat = PATTERN (insn);
|
10118 |
|
|
old_notes = REG_NOTES (insn);
|
10119 |
|
|
PATTERN (insn) = pat;
|
10120 |
|
|
REG_NOTES (insn) = 0;
|
10121 |
|
|
|
10122 |
|
|
insn_code_number = recog (pat, insn, &num_clobbers_to_add);
|
10123 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
10124 |
|
|
{
|
10125 |
|
|
if (insn_code_number < 0)
|
10126 |
|
|
fputs ("Failed to match this instruction:\n", dump_file);
|
10127 |
|
|
else
|
10128 |
|
|
fputs ("Successfully matched this instruction:\n", dump_file);
|
10129 |
|
|
print_rtl_single (dump_file, pat);
|
10130 |
|
|
}
|
10131 |
|
|
|
10132 |
|
|
/* If it isn't, there is the possibility that we previously had an insn
|
10133 |
|
|
that clobbered some register as a side effect, but the combined
|
10134 |
|
|
insn doesn't need to do that. So try once more without the clobbers
|
10135 |
|
|
unless this represents an ASM insn. */
|
10136 |
|
|
|
10137 |
|
|
if (insn_code_number < 0 && ! check_asm_operands (pat)
|
10138 |
|
|
&& GET_CODE (pat) == PARALLEL)
|
10139 |
|
|
{
|
10140 |
|
|
int pos;
|
10141 |
|
|
|
10142 |
|
|
for (pos = 0, i = 0; i < XVECLEN (pat, 0); i++)
|
10143 |
|
|
if (GET_CODE (XVECEXP (pat, 0, i)) != CLOBBER)
|
10144 |
|
|
{
|
10145 |
|
|
if (i != pos)
|
10146 |
|
|
SUBST (XVECEXP (pat, 0, pos), XVECEXP (pat, 0, i));
|
10147 |
|
|
pos++;
|
10148 |
|
|
}
|
10149 |
|
|
|
10150 |
|
|
SUBST_INT (XVECLEN (pat, 0), pos);
|
10151 |
|
|
|
10152 |
|
|
if (pos == 1)
|
10153 |
|
|
pat = XVECEXP (pat, 0, 0);
|
10154 |
|
|
|
10155 |
|
|
PATTERN (insn) = pat;
|
10156 |
|
|
insn_code_number = recog (pat, insn, &num_clobbers_to_add);
|
10157 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
10158 |
|
|
{
|
10159 |
|
|
if (insn_code_number < 0)
|
10160 |
|
|
fputs ("Failed to match this instruction:\n", dump_file);
|
10161 |
|
|
else
|
10162 |
|
|
fputs ("Successfully matched this instruction:\n", dump_file);
|
10163 |
|
|
print_rtl_single (dump_file, pat);
|
10164 |
|
|
}
|
10165 |
|
|
}
|
10166 |
|
|
PATTERN (insn) = old_pat;
|
10167 |
|
|
REG_NOTES (insn) = old_notes;
|
10168 |
|
|
|
10169 |
|
|
/* Recognize all noop sets, these will be killed by followup pass. */
|
10170 |
|
|
if (insn_code_number < 0 && GET_CODE (pat) == SET && set_noop_p (pat))
|
10171 |
|
|
insn_code_number = NOOP_MOVE_INSN_CODE, num_clobbers_to_add = 0;
|
10172 |
|
|
|
10173 |
|
|
/* If we had any clobbers to add, make a new pattern than contains
|
10174 |
|
|
them. Then check to make sure that all of them are dead. */
|
10175 |
|
|
if (num_clobbers_to_add)
|
10176 |
|
|
{
|
10177 |
|
|
rtx newpat = gen_rtx_PARALLEL (VOIDmode,
|
10178 |
|
|
rtvec_alloc (GET_CODE (pat) == PARALLEL
|
10179 |
|
|
? (XVECLEN (pat, 0)
|
10180 |
|
|
+ num_clobbers_to_add)
|
10181 |
|
|
: num_clobbers_to_add + 1));
|
10182 |
|
|
|
10183 |
|
|
if (GET_CODE (pat) == PARALLEL)
|
10184 |
|
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
10185 |
|
|
XVECEXP (newpat, 0, i) = XVECEXP (pat, 0, i);
|
10186 |
|
|
else
|
10187 |
|
|
XVECEXP (newpat, 0, 0) = pat;
|
10188 |
|
|
|
10189 |
|
|
add_clobbers (newpat, insn_code_number);
|
10190 |
|
|
|
10191 |
|
|
for (i = XVECLEN (newpat, 0) - num_clobbers_to_add;
|
10192 |
|
|
i < XVECLEN (newpat, 0); i++)
|
10193 |
|
|
{
|
10194 |
|
|
if (REG_P (XEXP (XVECEXP (newpat, 0, i), 0))
|
10195 |
|
|
&& ! reg_dead_at_p (XEXP (XVECEXP (newpat, 0, i), 0), insn))
|
10196 |
|
|
return -1;
|
10197 |
|
|
if (GET_CODE (XEXP (XVECEXP (newpat, 0, i), 0)) != SCRATCH)
|
10198 |
|
|
{
|
10199 |
|
|
gcc_assert (REG_P (XEXP (XVECEXP (newpat, 0, i), 0)));
|
10200 |
|
|
notes = alloc_reg_note (REG_UNUSED,
|
10201 |
|
|
XEXP (XVECEXP (newpat, 0, i), 0), notes);
|
10202 |
|
|
}
|
10203 |
|
|
}
|
10204 |
|
|
pat = newpat;
|
10205 |
|
|
}
|
10206 |
|
|
|
10207 |
|
|
*pnewpat = pat;
|
10208 |
|
|
*pnotes = notes;
|
10209 |
|
|
|
10210 |
|
|
return insn_code_number;
|
10211 |
|
|
}
|
10212 |
|
|
|
10213 |
|
|
/* Like gen_lowpart_general but for use by combine. In combine it
|
10214 |
|
|
is not possible to create any new pseudoregs. However, it is
|
10215 |
|
|
safe to create invalid memory addresses, because combine will
|
10216 |
|
|
try to recognize them and all they will do is make the combine
|
10217 |
|
|
attempt fail.
|
10218 |
|
|
|
10219 |
|
|
If for some reason this cannot do its job, an rtx
|
10220 |
|
|
(clobber (const_int 0)) is returned.
|
10221 |
|
|
An insn containing that will not be recognized. */
|
10222 |
|
|
|
10223 |
|
|
static rtx
|
10224 |
|
|
gen_lowpart_for_combine (enum machine_mode omode, rtx x)
|
10225 |
|
|
{
|
10226 |
|
|
enum machine_mode imode = GET_MODE (x);
|
10227 |
|
|
unsigned int osize = GET_MODE_SIZE (omode);
|
10228 |
|
|
unsigned int isize = GET_MODE_SIZE (imode);
|
10229 |
|
|
rtx result;
|
10230 |
|
|
|
10231 |
|
|
if (omode == imode)
|
10232 |
|
|
return x;
|
10233 |
|
|
|
10234 |
|
|
/* Return identity if this is a CONST or symbolic reference. */
|
10235 |
|
|
if (omode == Pmode
|
10236 |
|
|
&& (GET_CODE (x) == CONST
|
10237 |
|
|
|| GET_CODE (x) == SYMBOL_REF
|
10238 |
|
|
|| GET_CODE (x) == LABEL_REF))
|
10239 |
|
|
return x;
|
10240 |
|
|
|
10241 |
|
|
/* We can only support MODE being wider than a word if X is a
|
10242 |
|
|
constant integer or has a mode the same size. */
|
10243 |
|
|
if (GET_MODE_SIZE (omode) > UNITS_PER_WORD
|
10244 |
|
|
&& ! ((imode == VOIDmode
|
10245 |
|
|
&& (CONST_INT_P (x)
|
10246 |
|
|
|| GET_CODE (x) == CONST_DOUBLE))
|
10247 |
|
|
|| isize == osize))
|
10248 |
|
|
goto fail;
|
10249 |
|
|
|
10250 |
|
|
/* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
|
10251 |
|
|
won't know what to do. So we will strip off the SUBREG here and
|
10252 |
|
|
process normally. */
|
10253 |
|
|
if (GET_CODE (x) == SUBREG && MEM_P (SUBREG_REG (x)))
|
10254 |
|
|
{
|
10255 |
|
|
x = SUBREG_REG (x);
|
10256 |
|
|
|
10257 |
|
|
/* For use in case we fall down into the address adjustments
|
10258 |
|
|
further below, we need to adjust the known mode and size of
|
10259 |
|
|
x; imode and isize, since we just adjusted x. */
|
10260 |
|
|
imode = GET_MODE (x);
|
10261 |
|
|
|
10262 |
|
|
if (imode == omode)
|
10263 |
|
|
return x;
|
10264 |
|
|
|
10265 |
|
|
isize = GET_MODE_SIZE (imode);
|
10266 |
|
|
}
|
10267 |
|
|
|
10268 |
|
|
result = gen_lowpart_common (omode, x);
|
10269 |
|
|
|
10270 |
|
|
if (result)
|
10271 |
|
|
return result;
|
10272 |
|
|
|
10273 |
|
|
if (MEM_P (x))
|
10274 |
|
|
{
|
10275 |
|
|
int offset = 0;
|
10276 |
|
|
|
10277 |
|
|
/* Refuse to work on a volatile memory ref or one with a mode-dependent
|
10278 |
|
|
address. */
|
10279 |
|
|
if (MEM_VOLATILE_P (x) || mode_dependent_address_p (XEXP (x, 0)))
|
10280 |
|
|
goto fail;
|
10281 |
|
|
|
10282 |
|
|
/* If we want to refer to something bigger than the original memref,
|
10283 |
|
|
generate a paradoxical subreg instead. That will force a reload
|
10284 |
|
|
of the original memref X. */
|
10285 |
|
|
if (isize < osize)
|
10286 |
|
|
return gen_rtx_SUBREG (omode, x, 0);
|
10287 |
|
|
|
10288 |
|
|
if (WORDS_BIG_ENDIAN)
|
10289 |
|
|
offset = MAX (isize, UNITS_PER_WORD) - MAX (osize, UNITS_PER_WORD);
|
10290 |
|
|
|
10291 |
|
|
/* Adjust the address so that the address-after-the-data is
|
10292 |
|
|
unchanged. */
|
10293 |
|
|
if (BYTES_BIG_ENDIAN)
|
10294 |
|
|
offset -= MIN (UNITS_PER_WORD, osize) - MIN (UNITS_PER_WORD, isize);
|
10295 |
|
|
|
10296 |
|
|
return adjust_address_nv (x, omode, offset);
|
10297 |
|
|
}
|
10298 |
|
|
|
10299 |
|
|
/* If X is a comparison operator, rewrite it in a new mode. This
|
10300 |
|
|
probably won't match, but may allow further simplifications. */
|
10301 |
|
|
else if (COMPARISON_P (x))
|
10302 |
|
|
return gen_rtx_fmt_ee (GET_CODE (x), omode, XEXP (x, 0), XEXP (x, 1));
|
10303 |
|
|
|
10304 |
|
|
/* If we couldn't simplify X any other way, just enclose it in a
|
10305 |
|
|
SUBREG. Normally, this SUBREG won't match, but some patterns may
|
10306 |
|
|
include an explicit SUBREG or we may simplify it further in combine. */
|
10307 |
|
|
else
|
10308 |
|
|
{
|
10309 |
|
|
int offset = 0;
|
10310 |
|
|
rtx res;
|
10311 |
|
|
|
10312 |
|
|
offset = subreg_lowpart_offset (omode, imode);
|
10313 |
|
|
if (imode == VOIDmode)
|
10314 |
|
|
{
|
10315 |
|
|
imode = int_mode_for_mode (omode);
|
10316 |
|
|
x = gen_lowpart_common (imode, x);
|
10317 |
|
|
if (x == NULL)
|
10318 |
|
|
goto fail;
|
10319 |
|
|
}
|
10320 |
|
|
res = simplify_gen_subreg (omode, x, imode, offset);
|
10321 |
|
|
if (res)
|
10322 |
|
|
return res;
|
10323 |
|
|
}
|
10324 |
|
|
|
10325 |
|
|
fail:
|
10326 |
|
|
return gen_rtx_CLOBBER (omode, const0_rtx);
|
10327 |
|
|
}
|
10328 |
|
|
|
10329 |
|
|
/* Simplify a comparison between *POP0 and *POP1 where CODE is the
|
10330 |
|
|
comparison code that will be tested.
|
10331 |
|
|
|
10332 |
|
|
The result is a possibly different comparison code to use. *POP0 and
|
10333 |
|
|
*POP1 may be updated.
|
10334 |
|
|
|
10335 |
|
|
It is possible that we might detect that a comparison is either always
|
10336 |
|
|
true or always false. However, we do not perform general constant
|
10337 |
|
|
folding in combine, so this knowledge isn't useful. Such tautologies
|
10338 |
|
|
should have been detected earlier. Hence we ignore all such cases. */
|
10339 |
|
|
|
10340 |
|
|
static enum rtx_code
|
10341 |
|
|
simplify_comparison (enum rtx_code code, rtx *pop0, rtx *pop1)
|
10342 |
|
|
{
|
10343 |
|
|
rtx op0 = *pop0;
|
10344 |
|
|
rtx op1 = *pop1;
|
10345 |
|
|
rtx tem, tem1;
|
10346 |
|
|
int i;
|
10347 |
|
|
enum machine_mode mode, tmode;
|
10348 |
|
|
|
10349 |
|
|
/* Try a few ways of applying the same transformation to both operands. */
|
10350 |
|
|
while (1)
|
10351 |
|
|
{
|
10352 |
|
|
#ifndef WORD_REGISTER_OPERATIONS
|
10353 |
|
|
/* The test below this one won't handle SIGN_EXTENDs on these machines,
|
10354 |
|
|
so check specially. */
|
10355 |
|
|
if (code != GTU && code != GEU && code != LTU && code != LEU
|
10356 |
|
|
&& GET_CODE (op0) == ASHIFTRT && GET_CODE (op1) == ASHIFTRT
|
10357 |
|
|
&& GET_CODE (XEXP (op0, 0)) == ASHIFT
|
10358 |
|
|
&& GET_CODE (XEXP (op1, 0)) == ASHIFT
|
10359 |
|
|
&& GET_CODE (XEXP (XEXP (op0, 0), 0)) == SUBREG
|
10360 |
|
|
&& GET_CODE (XEXP (XEXP (op1, 0), 0)) == SUBREG
|
10361 |
|
|
&& (GET_MODE (SUBREG_REG (XEXP (XEXP (op0, 0), 0)))
|
10362 |
|
|
== GET_MODE (SUBREG_REG (XEXP (XEXP (op1, 0), 0))))
|
10363 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
10364 |
|
|
&& XEXP (op0, 1) == XEXP (op1, 1)
|
10365 |
|
|
&& XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
|
10366 |
|
|
&& XEXP (op0, 1) == XEXP (XEXP (op1, 0), 1)
|
10367 |
|
|
&& (INTVAL (XEXP (op0, 1))
|
10368 |
|
|
== (GET_MODE_BITSIZE (GET_MODE (op0))
|
10369 |
|
|
- (GET_MODE_BITSIZE
|
10370 |
|
|
(GET_MODE (SUBREG_REG (XEXP (XEXP (op0, 0), 0))))))))
|
10371 |
|
|
{
|
10372 |
|
|
op0 = SUBREG_REG (XEXP (XEXP (op0, 0), 0));
|
10373 |
|
|
op1 = SUBREG_REG (XEXP (XEXP (op1, 0), 0));
|
10374 |
|
|
}
|
10375 |
|
|
#endif
|
10376 |
|
|
|
10377 |
|
|
/* If both operands are the same constant shift, see if we can ignore the
|
10378 |
|
|
shift. We can if the shift is a rotate or if the bits shifted out of
|
10379 |
|
|
this shift are known to be zero for both inputs and if the type of
|
10380 |
|
|
comparison is compatible with the shift. */
|
10381 |
|
|
if (GET_CODE (op0) == GET_CODE (op1)
|
10382 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (op0)) <= HOST_BITS_PER_WIDE_INT
|
10383 |
|
|
&& ((GET_CODE (op0) == ROTATE && (code == NE || code == EQ))
|
10384 |
|
|
|| ((GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFT)
|
10385 |
|
|
&& (code != GT && code != LT && code != GE && code != LE))
|
10386 |
|
|
|| (GET_CODE (op0) == ASHIFTRT
|
10387 |
|
|
&& (code != GTU && code != LTU
|
10388 |
|
|
&& code != GEU && code != LEU)))
|
10389 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
10390 |
|
|
&& INTVAL (XEXP (op0, 1)) >= 0
|
10391 |
|
|
&& INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
|
10392 |
|
|
&& XEXP (op0, 1) == XEXP (op1, 1))
|
10393 |
|
|
{
|
10394 |
|
|
enum machine_mode mode = GET_MODE (op0);
|
10395 |
|
|
unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
|
10396 |
|
|
int shift_count = INTVAL (XEXP (op0, 1));
|
10397 |
|
|
|
10398 |
|
|
if (GET_CODE (op0) == LSHIFTRT || GET_CODE (op0) == ASHIFTRT)
|
10399 |
|
|
mask &= (mask >> shift_count) << shift_count;
|
10400 |
|
|
else if (GET_CODE (op0) == ASHIFT)
|
10401 |
|
|
mask = (mask & (mask << shift_count)) >> shift_count;
|
10402 |
|
|
|
10403 |
|
|
if ((nonzero_bits (XEXP (op0, 0), mode) & ~mask) == 0
|
10404 |
|
|
&& (nonzero_bits (XEXP (op1, 0), mode) & ~mask) == 0)
|
10405 |
|
|
op0 = XEXP (op0, 0), op1 = XEXP (op1, 0);
|
10406 |
|
|
else
|
10407 |
|
|
break;
|
10408 |
|
|
}
|
10409 |
|
|
|
10410 |
|
|
/* If both operands are AND's of a paradoxical SUBREG by constant, the
|
10411 |
|
|
SUBREGs are of the same mode, and, in both cases, the AND would
|
10412 |
|
|
be redundant if the comparison was done in the narrower mode,
|
10413 |
|
|
do the comparison in the narrower mode (e.g., we are AND'ing with 1
|
10414 |
|
|
and the operand's possibly nonzero bits are 0xffffff01; in that case
|
10415 |
|
|
if we only care about QImode, we don't need the AND). This case
|
10416 |
|
|
occurs if the output mode of an scc insn is not SImode and
|
10417 |
|
|
STORE_FLAG_VALUE == 1 (e.g., the 386).
|
10418 |
|
|
|
10419 |
|
|
Similarly, check for a case where the AND's are ZERO_EXTEND
|
10420 |
|
|
operations from some narrower mode even though a SUBREG is not
|
10421 |
|
|
present. */
|
10422 |
|
|
|
10423 |
|
|
else if (GET_CODE (op0) == AND && GET_CODE (op1) == AND
|
10424 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
10425 |
|
|
&& CONST_INT_P (XEXP (op1, 1)))
|
10426 |
|
|
{
|
10427 |
|
|
rtx inner_op0 = XEXP (op0, 0);
|
10428 |
|
|
rtx inner_op1 = XEXP (op1, 0);
|
10429 |
|
|
HOST_WIDE_INT c0 = INTVAL (XEXP (op0, 1));
|
10430 |
|
|
HOST_WIDE_INT c1 = INTVAL (XEXP (op1, 1));
|
10431 |
|
|
int changed = 0;
|
10432 |
|
|
|
10433 |
|
|
if (GET_CODE (inner_op0) == SUBREG && GET_CODE (inner_op1) == SUBREG
|
10434 |
|
|
&& (GET_MODE_SIZE (GET_MODE (inner_op0))
|
10435 |
|
|
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (inner_op0))))
|
10436 |
|
|
&& (GET_MODE (SUBREG_REG (inner_op0))
|
10437 |
|
|
== GET_MODE (SUBREG_REG (inner_op1)))
|
10438 |
|
|
&& (GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (inner_op0)))
|
10439 |
|
|
<= HOST_BITS_PER_WIDE_INT)
|
10440 |
|
|
&& (0 == ((~c0) & nonzero_bits (SUBREG_REG (inner_op0),
|
10441 |
|
|
GET_MODE (SUBREG_REG (inner_op0)))))
|
10442 |
|
|
&& (0 == ((~c1) & nonzero_bits (SUBREG_REG (inner_op1),
|
10443 |
|
|
GET_MODE (SUBREG_REG (inner_op1))))))
|
10444 |
|
|
{
|
10445 |
|
|
op0 = SUBREG_REG (inner_op0);
|
10446 |
|
|
op1 = SUBREG_REG (inner_op1);
|
10447 |
|
|
|
10448 |
|
|
/* The resulting comparison is always unsigned since we masked
|
10449 |
|
|
off the original sign bit. */
|
10450 |
|
|
code = unsigned_condition (code);
|
10451 |
|
|
|
10452 |
|
|
changed = 1;
|
10453 |
|
|
}
|
10454 |
|
|
|
10455 |
|
|
else if (c0 == c1)
|
10456 |
|
|
for (tmode = GET_CLASS_NARROWEST_MODE
|
10457 |
|
|
(GET_MODE_CLASS (GET_MODE (op0)));
|
10458 |
|
|
tmode != GET_MODE (op0); tmode = GET_MODE_WIDER_MODE (tmode))
|
10459 |
|
|
if ((unsigned HOST_WIDE_INT) c0 == GET_MODE_MASK (tmode))
|
10460 |
|
|
{
|
10461 |
|
|
op0 = gen_lowpart (tmode, inner_op0);
|
10462 |
|
|
op1 = gen_lowpart (tmode, inner_op1);
|
10463 |
|
|
code = unsigned_condition (code);
|
10464 |
|
|
changed = 1;
|
10465 |
|
|
break;
|
10466 |
|
|
}
|
10467 |
|
|
|
10468 |
|
|
if (! changed)
|
10469 |
|
|
break;
|
10470 |
|
|
}
|
10471 |
|
|
|
10472 |
|
|
/* If both operands are NOT, we can strip off the outer operation
|
10473 |
|
|
and adjust the comparison code for swapped operands; similarly for
|
10474 |
|
|
NEG, except that this must be an equality comparison. */
|
10475 |
|
|
else if ((GET_CODE (op0) == NOT && GET_CODE (op1) == NOT)
|
10476 |
|
|
|| (GET_CODE (op0) == NEG && GET_CODE (op1) == NEG
|
10477 |
|
|
&& (code == EQ || code == NE)))
|
10478 |
|
|
op0 = XEXP (op0, 0), op1 = XEXP (op1, 0), code = swap_condition (code);
|
10479 |
|
|
|
10480 |
|
|
else
|
10481 |
|
|
break;
|
10482 |
|
|
}
|
10483 |
|
|
|
10484 |
|
|
/* If the first operand is a constant, swap the operands and adjust the
|
10485 |
|
|
comparison code appropriately, but don't do this if the second operand
|
10486 |
|
|
is already a constant integer. */
|
10487 |
|
|
if (swap_commutative_operands_p (op0, op1))
|
10488 |
|
|
{
|
10489 |
|
|
tem = op0, op0 = op1, op1 = tem;
|
10490 |
|
|
code = swap_condition (code);
|
10491 |
|
|
}
|
10492 |
|
|
|
10493 |
|
|
/* We now enter a loop during which we will try to simplify the comparison.
|
10494 |
|
|
For the most part, we only are concerned with comparisons with zero,
|
10495 |
|
|
but some things may really be comparisons with zero but not start
|
10496 |
|
|
out looking that way. */
|
10497 |
|
|
|
10498 |
|
|
while (CONST_INT_P (op1))
|
10499 |
|
|
{
|
10500 |
|
|
enum machine_mode mode = GET_MODE (op0);
|
10501 |
|
|
unsigned int mode_width = GET_MODE_BITSIZE (mode);
|
10502 |
|
|
unsigned HOST_WIDE_INT mask = GET_MODE_MASK (mode);
|
10503 |
|
|
int equality_comparison_p;
|
10504 |
|
|
int sign_bit_comparison_p;
|
10505 |
|
|
int unsigned_comparison_p;
|
10506 |
|
|
HOST_WIDE_INT const_op;
|
10507 |
|
|
|
10508 |
|
|
/* We only want to handle integral modes. This catches VOIDmode,
|
10509 |
|
|
CCmode, and the floating-point modes. An exception is that we
|
10510 |
|
|
can handle VOIDmode if OP0 is a COMPARE or a comparison
|
10511 |
|
|
operation. */
|
10512 |
|
|
|
10513 |
|
|
if (GET_MODE_CLASS (mode) != MODE_INT
|
10514 |
|
|
&& ! (mode == VOIDmode
|
10515 |
|
|
&& (GET_CODE (op0) == COMPARE || COMPARISON_P (op0))))
|
10516 |
|
|
break;
|
10517 |
|
|
|
10518 |
|
|
/* Get the constant we are comparing against and turn off all bits
|
10519 |
|
|
not on in our mode. */
|
10520 |
|
|
const_op = INTVAL (op1);
|
10521 |
|
|
if (mode != VOIDmode)
|
10522 |
|
|
const_op = trunc_int_for_mode (const_op, mode);
|
10523 |
|
|
op1 = GEN_INT (const_op);
|
10524 |
|
|
|
10525 |
|
|
/* If we are comparing against a constant power of two and the value
|
10526 |
|
|
being compared can only have that single bit nonzero (e.g., it was
|
10527 |
|
|
`and'ed with that bit), we can replace this with a comparison
|
10528 |
|
|
with zero. */
|
10529 |
|
|
if (const_op
|
10530 |
|
|
&& (code == EQ || code == NE || code == GE || code == GEU
|
10531 |
|
|
|| code == LT || code == LTU)
|
10532 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
10533 |
|
|
&& exact_log2 (const_op) >= 0
|
10534 |
|
|
&& nonzero_bits (op0, mode) == (unsigned HOST_WIDE_INT) const_op)
|
10535 |
|
|
{
|
10536 |
|
|
code = (code == EQ || code == GE || code == GEU ? NE : EQ);
|
10537 |
|
|
op1 = const0_rtx, const_op = 0;
|
10538 |
|
|
}
|
10539 |
|
|
|
10540 |
|
|
/* Similarly, if we are comparing a value known to be either -1 or
|
10541 |
|
|
|
10542 |
|
|
|
10543 |
|
|
if (const_op == -1
|
10544 |
|
|
&& (code == EQ || code == NE || code == GT || code == LE
|
10545 |
|
|
|| code == GEU || code == LTU)
|
10546 |
|
|
&& num_sign_bit_copies (op0, mode) == mode_width)
|
10547 |
|
|
{
|
10548 |
|
|
code = (code == EQ || code == LE || code == GEU ? NE : EQ);
|
10549 |
|
|
op1 = const0_rtx, const_op = 0;
|
10550 |
|
|
}
|
10551 |
|
|
|
10552 |
|
|
/* Do some canonicalizations based on the comparison code. We prefer
|
10553 |
|
|
comparisons against zero and then prefer equality comparisons.
|
10554 |
|
|
If we can reduce the size of a constant, we will do that too. */
|
10555 |
|
|
|
10556 |
|
|
switch (code)
|
10557 |
|
|
{
|
10558 |
|
|
case LT:
|
10559 |
|
|
/* < C is equivalent to <= (C - 1) */
|
10560 |
|
|
if (const_op > 0)
|
10561 |
|
|
{
|
10562 |
|
|
const_op -= 1;
|
10563 |
|
|
op1 = GEN_INT (const_op);
|
10564 |
|
|
code = LE;
|
10565 |
|
|
/* ... fall through to LE case below. */
|
10566 |
|
|
}
|
10567 |
|
|
else
|
10568 |
|
|
break;
|
10569 |
|
|
|
10570 |
|
|
case LE:
|
10571 |
|
|
/* <= C is equivalent to < (C + 1); we do this for C < 0 */
|
10572 |
|
|
if (const_op < 0)
|
10573 |
|
|
{
|
10574 |
|
|
const_op += 1;
|
10575 |
|
|
op1 = GEN_INT (const_op);
|
10576 |
|
|
code = LT;
|
10577 |
|
|
}
|
10578 |
|
|
|
10579 |
|
|
/* If we are doing a <= 0 comparison on a value known to have
|
10580 |
|
|
a zero sign bit, we can replace this with == 0. */
|
10581 |
|
|
else if (const_op == 0
|
10582 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
10583 |
|
|
&& (nonzero_bits (op0, mode)
|
10584 |
|
|
& ((HOST_WIDE_INT) 1 << (mode_width - 1))) == 0)
|
10585 |
|
|
code = EQ;
|
10586 |
|
|
break;
|
10587 |
|
|
|
10588 |
|
|
case GE:
|
10589 |
|
|
/* >= C is equivalent to > (C - 1). */
|
10590 |
|
|
if (const_op > 0)
|
10591 |
|
|
{
|
10592 |
|
|
const_op -= 1;
|
10593 |
|
|
op1 = GEN_INT (const_op);
|
10594 |
|
|
code = GT;
|
10595 |
|
|
/* ... fall through to GT below. */
|
10596 |
|
|
}
|
10597 |
|
|
else
|
10598 |
|
|
break;
|
10599 |
|
|
|
10600 |
|
|
case GT:
|
10601 |
|
|
/* > C is equivalent to >= (C + 1); we do this for C < 0. */
|
10602 |
|
|
if (const_op < 0)
|
10603 |
|
|
{
|
10604 |
|
|
const_op += 1;
|
10605 |
|
|
op1 = GEN_INT (const_op);
|
10606 |
|
|
code = GE;
|
10607 |
|
|
}
|
10608 |
|
|
|
10609 |
|
|
/* If we are doing a > 0 comparison on a value known to have
|
10610 |
|
|
a zero sign bit, we can replace this with != 0. */
|
10611 |
|
|
else if (const_op == 0
|
10612 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
10613 |
|
|
&& (nonzero_bits (op0, mode)
|
10614 |
|
|
& ((HOST_WIDE_INT) 1 << (mode_width - 1))) == 0)
|
10615 |
|
|
code = NE;
|
10616 |
|
|
break;
|
10617 |
|
|
|
10618 |
|
|
case LTU:
|
10619 |
|
|
/* < C is equivalent to <= (C - 1). */
|
10620 |
|
|
if (const_op > 0)
|
10621 |
|
|
{
|
10622 |
|
|
const_op -= 1;
|
10623 |
|
|
op1 = GEN_INT (const_op);
|
10624 |
|
|
code = LEU;
|
10625 |
|
|
/* ... fall through ... */
|
10626 |
|
|
}
|
10627 |
|
|
|
10628 |
|
|
/* (unsigned) < 0x80000000 is equivalent to >= 0. */
|
10629 |
|
|
else if ((mode_width <= HOST_BITS_PER_WIDE_INT)
|
10630 |
|
|
&& (const_op == (HOST_WIDE_INT) 1 << (mode_width - 1)))
|
10631 |
|
|
{
|
10632 |
|
|
const_op = 0, op1 = const0_rtx;
|
10633 |
|
|
code = GE;
|
10634 |
|
|
break;
|
10635 |
|
|
}
|
10636 |
|
|
else
|
10637 |
|
|
break;
|
10638 |
|
|
|
10639 |
|
|
case LEU:
|
10640 |
|
|
/* unsigned <= 0 is equivalent to == 0 */
|
10641 |
|
|
if (const_op == 0)
|
10642 |
|
|
code = EQ;
|
10643 |
|
|
|
10644 |
|
|
/* (unsigned) <= 0x7fffffff is equivalent to >= 0. */
|
10645 |
|
|
else if ((mode_width <= HOST_BITS_PER_WIDE_INT)
|
10646 |
|
|
&& (const_op == ((HOST_WIDE_INT) 1 << (mode_width - 1)) - 1))
|
10647 |
|
|
{
|
10648 |
|
|
const_op = 0, op1 = const0_rtx;
|
10649 |
|
|
code = GE;
|
10650 |
|
|
}
|
10651 |
|
|
break;
|
10652 |
|
|
|
10653 |
|
|
case GEU:
|
10654 |
|
|
/* >= C is equivalent to > (C - 1). */
|
10655 |
|
|
if (const_op > 1)
|
10656 |
|
|
{
|
10657 |
|
|
const_op -= 1;
|
10658 |
|
|
op1 = GEN_INT (const_op);
|
10659 |
|
|
code = GTU;
|
10660 |
|
|
/* ... fall through ... */
|
10661 |
|
|
}
|
10662 |
|
|
|
10663 |
|
|
/* (unsigned) >= 0x80000000 is equivalent to < 0. */
|
10664 |
|
|
else if ((mode_width <= HOST_BITS_PER_WIDE_INT)
|
10665 |
|
|
&& (const_op == (HOST_WIDE_INT) 1 << (mode_width - 1)))
|
10666 |
|
|
{
|
10667 |
|
|
const_op = 0, op1 = const0_rtx;
|
10668 |
|
|
code = LT;
|
10669 |
|
|
break;
|
10670 |
|
|
}
|
10671 |
|
|
else
|
10672 |
|
|
break;
|
10673 |
|
|
|
10674 |
|
|
case GTU:
|
10675 |
|
|
/* unsigned > 0 is equivalent to != 0 */
|
10676 |
|
|
if (const_op == 0)
|
10677 |
|
|
code = NE;
|
10678 |
|
|
|
10679 |
|
|
/* (unsigned) > 0x7fffffff is equivalent to < 0. */
|
10680 |
|
|
else if ((mode_width <= HOST_BITS_PER_WIDE_INT)
|
10681 |
|
|
&& (const_op == ((HOST_WIDE_INT) 1 << (mode_width - 1)) - 1))
|
10682 |
|
|
{
|
10683 |
|
|
const_op = 0, op1 = const0_rtx;
|
10684 |
|
|
code = LT;
|
10685 |
|
|
}
|
10686 |
|
|
break;
|
10687 |
|
|
|
10688 |
|
|
default:
|
10689 |
|
|
break;
|
10690 |
|
|
}
|
10691 |
|
|
|
10692 |
|
|
/* Compute some predicates to simplify code below. */
|
10693 |
|
|
|
10694 |
|
|
equality_comparison_p = (code == EQ || code == NE);
|
10695 |
|
|
sign_bit_comparison_p = ((code == LT || code == GE) && const_op == 0);
|
10696 |
|
|
unsigned_comparison_p = (code == LTU || code == LEU || code == GTU
|
10697 |
|
|
|| code == GEU);
|
10698 |
|
|
|
10699 |
|
|
/* If this is a sign bit comparison and we can do arithmetic in
|
10700 |
|
|
MODE, say that we will only be needing the sign bit of OP0. */
|
10701 |
|
|
if (sign_bit_comparison_p
|
10702 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
10703 |
|
|
op0 = force_to_mode (op0, mode,
|
10704 |
|
|
((HOST_WIDE_INT) 1
|
10705 |
|
|
<< (GET_MODE_BITSIZE (mode) - 1)),
|
10706 |
|
|
0);
|
10707 |
|
|
|
10708 |
|
|
/* Now try cases based on the opcode of OP0. If none of the cases
|
10709 |
|
|
does a "continue", we exit this loop immediately after the
|
10710 |
|
|
switch. */
|
10711 |
|
|
|
10712 |
|
|
switch (GET_CODE (op0))
|
10713 |
|
|
{
|
10714 |
|
|
case ZERO_EXTRACT:
|
10715 |
|
|
/* If we are extracting a single bit from a variable position in
|
10716 |
|
|
a constant that has only a single bit set and are comparing it
|
10717 |
|
|
with zero, we can convert this into an equality comparison
|
10718 |
|
|
between the position and the location of the single bit. */
|
10719 |
|
|
/* Except we can't if SHIFT_COUNT_TRUNCATED is set, since we might
|
10720 |
|
|
have already reduced the shift count modulo the word size. */
|
10721 |
|
|
if (!SHIFT_COUNT_TRUNCATED
|
10722 |
|
|
&& CONST_INT_P (XEXP (op0, 0))
|
10723 |
|
|
&& XEXP (op0, 1) == const1_rtx
|
10724 |
|
|
&& equality_comparison_p && const_op == 0
|
10725 |
|
|
&& (i = exact_log2 (INTVAL (XEXP (op0, 0)))) >= 0)
|
10726 |
|
|
{
|
10727 |
|
|
if (BITS_BIG_ENDIAN)
|
10728 |
|
|
{
|
10729 |
|
|
enum machine_mode new_mode
|
10730 |
|
|
= mode_for_extraction (EP_extzv, 1);
|
10731 |
|
|
if (new_mode == MAX_MACHINE_MODE)
|
10732 |
|
|
i = BITS_PER_WORD - 1 - i;
|
10733 |
|
|
else
|
10734 |
|
|
{
|
10735 |
|
|
mode = new_mode;
|
10736 |
|
|
i = (GET_MODE_BITSIZE (mode) - 1 - i);
|
10737 |
|
|
}
|
10738 |
|
|
}
|
10739 |
|
|
|
10740 |
|
|
op0 = XEXP (op0, 2);
|
10741 |
|
|
op1 = GEN_INT (i);
|
10742 |
|
|
const_op = i;
|
10743 |
|
|
|
10744 |
|
|
/* Result is nonzero iff shift count is equal to I. */
|
10745 |
|
|
code = reverse_condition (code);
|
10746 |
|
|
continue;
|
10747 |
|
|
}
|
10748 |
|
|
|
10749 |
|
|
/* ... fall through ... */
|
10750 |
|
|
|
10751 |
|
|
case SIGN_EXTRACT:
|
10752 |
|
|
tem = expand_compound_operation (op0);
|
10753 |
|
|
if (tem != op0)
|
10754 |
|
|
{
|
10755 |
|
|
op0 = tem;
|
10756 |
|
|
continue;
|
10757 |
|
|
}
|
10758 |
|
|
break;
|
10759 |
|
|
|
10760 |
|
|
case NOT:
|
10761 |
|
|
/* If testing for equality, we can take the NOT of the constant. */
|
10762 |
|
|
if (equality_comparison_p
|
10763 |
|
|
&& (tem = simplify_unary_operation (NOT, mode, op1, mode)) != 0)
|
10764 |
|
|
{
|
10765 |
|
|
op0 = XEXP (op0, 0);
|
10766 |
|
|
op1 = tem;
|
10767 |
|
|
continue;
|
10768 |
|
|
}
|
10769 |
|
|
|
10770 |
|
|
/* If just looking at the sign bit, reverse the sense of the
|
10771 |
|
|
comparison. */
|
10772 |
|
|
if (sign_bit_comparison_p)
|
10773 |
|
|
{
|
10774 |
|
|
op0 = XEXP (op0, 0);
|
10775 |
|
|
code = (code == GE ? LT : GE);
|
10776 |
|
|
continue;
|
10777 |
|
|
}
|
10778 |
|
|
break;
|
10779 |
|
|
|
10780 |
|
|
case NEG:
|
10781 |
|
|
/* If testing for equality, we can take the NEG of the constant. */
|
10782 |
|
|
if (equality_comparison_p
|
10783 |
|
|
&& (tem = simplify_unary_operation (NEG, mode, op1, mode)) != 0)
|
10784 |
|
|
{
|
10785 |
|
|
op0 = XEXP (op0, 0);
|
10786 |
|
|
op1 = tem;
|
10787 |
|
|
continue;
|
10788 |
|
|
}
|
10789 |
|
|
|
10790 |
|
|
/* The remaining cases only apply to comparisons with zero. */
|
10791 |
|
|
if (const_op != 0)
|
10792 |
|
|
break;
|
10793 |
|
|
|
10794 |
|
|
/* When X is ABS or is known positive,
|
10795 |
|
|
(neg X) is < 0 if and only if X != 0. */
|
10796 |
|
|
|
10797 |
|
|
if (sign_bit_comparison_p
|
10798 |
|
|
&& (GET_CODE (XEXP (op0, 0)) == ABS
|
10799 |
|
|
|| (mode_width <= HOST_BITS_PER_WIDE_INT
|
10800 |
|
|
&& (nonzero_bits (XEXP (op0, 0), mode)
|
10801 |
|
|
& ((HOST_WIDE_INT) 1 << (mode_width - 1))) == 0)))
|
10802 |
|
|
{
|
10803 |
|
|
op0 = XEXP (op0, 0);
|
10804 |
|
|
code = (code == LT ? NE : EQ);
|
10805 |
|
|
continue;
|
10806 |
|
|
}
|
10807 |
|
|
|
10808 |
|
|
/* If we have NEG of something whose two high-order bits are the
|
10809 |
|
|
same, we know that "(-a) < 0" is equivalent to "a > 0". */
|
10810 |
|
|
if (num_sign_bit_copies (op0, mode) >= 2)
|
10811 |
|
|
{
|
10812 |
|
|
op0 = XEXP (op0, 0);
|
10813 |
|
|
code = swap_condition (code);
|
10814 |
|
|
continue;
|
10815 |
|
|
}
|
10816 |
|
|
break;
|
10817 |
|
|
|
10818 |
|
|
case ROTATE:
|
10819 |
|
|
/* If we are testing equality and our count is a constant, we
|
10820 |
|
|
can perform the inverse operation on our RHS. */
|
10821 |
|
|
if (equality_comparison_p && CONST_INT_P (XEXP (op0, 1))
|
10822 |
|
|
&& (tem = simplify_binary_operation (ROTATERT, mode,
|
10823 |
|
|
op1, XEXP (op0, 1))) != 0)
|
10824 |
|
|
{
|
10825 |
|
|
op0 = XEXP (op0, 0);
|
10826 |
|
|
op1 = tem;
|
10827 |
|
|
continue;
|
10828 |
|
|
}
|
10829 |
|
|
|
10830 |
|
|
/* If we are doing a < 0 or >= 0 comparison, it means we are testing
|
10831 |
|
|
a particular bit. Convert it to an AND of a constant of that
|
10832 |
|
|
bit. This will be converted into a ZERO_EXTRACT. */
|
10833 |
|
|
if (const_op == 0 && sign_bit_comparison_p
|
10834 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
10835 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT)
|
10836 |
|
|
{
|
10837 |
|
|
op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
|
10838 |
|
|
((HOST_WIDE_INT) 1
|
10839 |
|
|
<< (mode_width - 1
|
10840 |
|
|
- INTVAL (XEXP (op0, 1)))));
|
10841 |
|
|
code = (code == LT ? NE : EQ);
|
10842 |
|
|
continue;
|
10843 |
|
|
}
|
10844 |
|
|
|
10845 |
|
|
/* Fall through. */
|
10846 |
|
|
|
10847 |
|
|
case ABS:
|
10848 |
|
|
/* ABS is ignorable inside an equality comparison with zero. */
|
10849 |
|
|
if (const_op == 0 && equality_comparison_p)
|
10850 |
|
|
{
|
10851 |
|
|
op0 = XEXP (op0, 0);
|
10852 |
|
|
continue;
|
10853 |
|
|
}
|
10854 |
|
|
break;
|
10855 |
|
|
|
10856 |
|
|
case SIGN_EXTEND:
|
10857 |
|
|
/* Can simplify (compare (zero/sign_extend FOO) CONST) to
|
10858 |
|
|
(compare FOO CONST) if CONST fits in FOO's mode and we
|
10859 |
|
|
are either testing inequality or have an unsigned
|
10860 |
|
|
comparison with ZERO_EXTEND or a signed comparison with
|
10861 |
|
|
SIGN_EXTEND. But don't do it if we don't have a compare
|
10862 |
|
|
insn of the given mode, since we'd have to revert it
|
10863 |
|
|
later on, and then we wouldn't know whether to sign- or
|
10864 |
|
|
zero-extend. */
|
10865 |
|
|
mode = GET_MODE (XEXP (op0, 0));
|
10866 |
|
|
if (mode != VOIDmode && GET_MODE_CLASS (mode) == MODE_INT
|
10867 |
|
|
&& ! unsigned_comparison_p
|
10868 |
|
|
&& (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
10869 |
|
|
&& ((unsigned HOST_WIDE_INT) const_op
|
10870 |
|
|
< (((unsigned HOST_WIDE_INT) 1
|
10871 |
|
|
<< (GET_MODE_BITSIZE (mode) - 1))))
|
10872 |
|
|
&& have_insn_for (COMPARE, mode))
|
10873 |
|
|
{
|
10874 |
|
|
op0 = XEXP (op0, 0);
|
10875 |
|
|
continue;
|
10876 |
|
|
}
|
10877 |
|
|
break;
|
10878 |
|
|
|
10879 |
|
|
case SUBREG:
|
10880 |
|
|
/* Check for the case where we are comparing A - C1 with C2, that is
|
10881 |
|
|
|
10882 |
|
|
(subreg:MODE (plus (A) (-C1))) op (C2)
|
10883 |
|
|
|
10884 |
|
|
with C1 a constant, and try to lift the SUBREG, i.e. to do the
|
10885 |
|
|
comparison in the wider mode. One of the following two conditions
|
10886 |
|
|
must be true in order for this to be valid:
|
10887 |
|
|
|
10888 |
|
|
1. The mode extension results in the same bit pattern being added
|
10889 |
|
|
on both sides and the comparison is equality or unsigned. As
|
10890 |
|
|
C2 has been truncated to fit in MODE, the pattern can only be
|
10891 |
|
|
all 0s or all 1s.
|
10892 |
|
|
|
10893 |
|
|
2. The mode extension results in the sign bit being copied on
|
10894 |
|
|
each side.
|
10895 |
|
|
|
10896 |
|
|
The difficulty here is that we have predicates for A but not for
|
10897 |
|
|
(A - C1) so we need to check that C1 is within proper bounds so
|
10898 |
|
|
as to perturbate A as little as possible. */
|
10899 |
|
|
|
10900 |
|
|
if (mode_width <= HOST_BITS_PER_WIDE_INT
|
10901 |
|
|
&& subreg_lowpart_p (op0)
|
10902 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (op0))) > mode_width
|
10903 |
|
|
&& GET_CODE (SUBREG_REG (op0)) == PLUS
|
10904 |
|
|
&& CONST_INT_P (XEXP (SUBREG_REG (op0), 1)))
|
10905 |
|
|
{
|
10906 |
|
|
enum machine_mode inner_mode = GET_MODE (SUBREG_REG (op0));
|
10907 |
|
|
rtx a = XEXP (SUBREG_REG (op0), 0);
|
10908 |
|
|
HOST_WIDE_INT c1 = -INTVAL (XEXP (SUBREG_REG (op0), 1));
|
10909 |
|
|
|
10910 |
|
|
if ((c1 > 0
|
10911 |
|
|
&& (unsigned HOST_WIDE_INT) c1
|
10912 |
|
|
< (unsigned HOST_WIDE_INT) 1 << (mode_width - 1)
|
10913 |
|
|
&& (equality_comparison_p || unsigned_comparison_p)
|
10914 |
|
|
/* (A - C1) zero-extends if it is positive and sign-extends
|
10915 |
|
|
if it is negative, C2 both zero- and sign-extends. */
|
10916 |
|
|
&& ((0 == (nonzero_bits (a, inner_mode)
|
10917 |
|
|
& ~GET_MODE_MASK (mode))
|
10918 |
|
|
&& const_op >= 0)
|
10919 |
|
|
/* (A - C1) sign-extends if it is positive and 1-extends
|
10920 |
|
|
if it is negative, C2 both sign- and 1-extends. */
|
10921 |
|
|
|| (num_sign_bit_copies (a, inner_mode)
|
10922 |
|
|
> (unsigned int) (GET_MODE_BITSIZE (inner_mode)
|
10923 |
|
|
- mode_width)
|
10924 |
|
|
&& const_op < 0)))
|
10925 |
|
|
|| ((unsigned HOST_WIDE_INT) c1
|
10926 |
|
|
< (unsigned HOST_WIDE_INT) 1 << (mode_width - 2)
|
10927 |
|
|
/* (A - C1) always sign-extends, like C2. */
|
10928 |
|
|
&& num_sign_bit_copies (a, inner_mode)
|
10929 |
|
|
> (unsigned int) (GET_MODE_BITSIZE (inner_mode)
|
10930 |
|
|
- (mode_width - 1))))
|
10931 |
|
|
{
|
10932 |
|
|
op0 = SUBREG_REG (op0);
|
10933 |
|
|
continue;
|
10934 |
|
|
}
|
10935 |
|
|
}
|
10936 |
|
|
|
10937 |
|
|
/* If the inner mode is narrower and we are extracting the low part,
|
10938 |
|
|
we can treat the SUBREG as if it were a ZERO_EXTEND. */
|
10939 |
|
|
if (subreg_lowpart_p (op0)
|
10940 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (op0))) < mode_width)
|
10941 |
|
|
/* Fall through */ ;
|
10942 |
|
|
else
|
10943 |
|
|
break;
|
10944 |
|
|
|
10945 |
|
|
/* ... fall through ... */
|
10946 |
|
|
|
10947 |
|
|
case ZERO_EXTEND:
|
10948 |
|
|
mode = GET_MODE (XEXP (op0, 0));
|
10949 |
|
|
if (mode != VOIDmode && GET_MODE_CLASS (mode) == MODE_INT
|
10950 |
|
|
&& (unsigned_comparison_p || equality_comparison_p)
|
10951 |
|
|
&& (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
10952 |
|
|
&& ((unsigned HOST_WIDE_INT) const_op < GET_MODE_MASK (mode))
|
10953 |
|
|
&& have_insn_for (COMPARE, mode))
|
10954 |
|
|
{
|
10955 |
|
|
op0 = XEXP (op0, 0);
|
10956 |
|
|
continue;
|
10957 |
|
|
}
|
10958 |
|
|
break;
|
10959 |
|
|
|
10960 |
|
|
case PLUS:
|
10961 |
|
|
/* (eq (plus X A) B) -> (eq X (minus B A)). We can only do
|
10962 |
|
|
this for equality comparisons due to pathological cases involving
|
10963 |
|
|
overflows. */
|
10964 |
|
|
if (equality_comparison_p
|
10965 |
|
|
&& 0 != (tem = simplify_binary_operation (MINUS, mode,
|
10966 |
|
|
op1, XEXP (op0, 1))))
|
10967 |
|
|
{
|
10968 |
|
|
op0 = XEXP (op0, 0);
|
10969 |
|
|
op1 = tem;
|
10970 |
|
|
continue;
|
10971 |
|
|
}
|
10972 |
|
|
|
10973 |
|
|
/* (plus (abs X) (const_int -1)) is < 0 if and only if X == 0. */
|
10974 |
|
|
if (const_op == 0 && XEXP (op0, 1) == constm1_rtx
|
10975 |
|
|
&& GET_CODE (XEXP (op0, 0)) == ABS && sign_bit_comparison_p)
|
10976 |
|
|
{
|
10977 |
|
|
op0 = XEXP (XEXP (op0, 0), 0);
|
10978 |
|
|
code = (code == LT ? EQ : NE);
|
10979 |
|
|
continue;
|
10980 |
|
|
}
|
10981 |
|
|
break;
|
10982 |
|
|
|
10983 |
|
|
case MINUS:
|
10984 |
|
|
/* We used to optimize signed comparisons against zero, but that
|
10985 |
|
|
was incorrect. Unsigned comparisons against zero (GTU, LEU)
|
10986 |
|
|
arrive here as equality comparisons, or (GEU, LTU) are
|
10987 |
|
|
optimized away. No need to special-case them. */
|
10988 |
|
|
|
10989 |
|
|
/* (eq (minus A B) C) -> (eq A (plus B C)) or
|
10990 |
|
|
(eq B (minus A C)), whichever simplifies. We can only do
|
10991 |
|
|
this for equality comparisons due to pathological cases involving
|
10992 |
|
|
overflows. */
|
10993 |
|
|
if (equality_comparison_p
|
10994 |
|
|
&& 0 != (tem = simplify_binary_operation (PLUS, mode,
|
10995 |
|
|
XEXP (op0, 1), op1)))
|
10996 |
|
|
{
|
10997 |
|
|
op0 = XEXP (op0, 0);
|
10998 |
|
|
op1 = tem;
|
10999 |
|
|
continue;
|
11000 |
|
|
}
|
11001 |
|
|
|
11002 |
|
|
if (equality_comparison_p
|
11003 |
|
|
&& 0 != (tem = simplify_binary_operation (MINUS, mode,
|
11004 |
|
|
XEXP (op0, 0), op1)))
|
11005 |
|
|
{
|
11006 |
|
|
op0 = XEXP (op0, 1);
|
11007 |
|
|
op1 = tem;
|
11008 |
|
|
continue;
|
11009 |
|
|
}
|
11010 |
|
|
|
11011 |
|
|
/* The sign bit of (minus (ashiftrt X C) X), where C is the number
|
11012 |
|
|
of bits in X minus 1, is one iff X > 0. */
|
11013 |
|
|
if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == ASHIFTRT
|
11014 |
|
|
&& CONST_INT_P (XEXP (XEXP (op0, 0), 1))
|
11015 |
|
|
&& (unsigned HOST_WIDE_INT) INTVAL (XEXP (XEXP (op0, 0), 1))
|
11016 |
|
|
== mode_width - 1
|
11017 |
|
|
&& rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
|
11018 |
|
|
{
|
11019 |
|
|
op0 = XEXP (op0, 1);
|
11020 |
|
|
code = (code == GE ? LE : GT);
|
11021 |
|
|
continue;
|
11022 |
|
|
}
|
11023 |
|
|
break;
|
11024 |
|
|
|
11025 |
|
|
case XOR:
|
11026 |
|
|
/* (eq (xor A B) C) -> (eq A (xor B C)). This is a simplification
|
11027 |
|
|
if C is zero or B is a constant. */
|
11028 |
|
|
if (equality_comparison_p
|
11029 |
|
|
&& 0 != (tem = simplify_binary_operation (XOR, mode,
|
11030 |
|
|
XEXP (op0, 1), op1)))
|
11031 |
|
|
{
|
11032 |
|
|
op0 = XEXP (op0, 0);
|
11033 |
|
|
op1 = tem;
|
11034 |
|
|
continue;
|
11035 |
|
|
}
|
11036 |
|
|
break;
|
11037 |
|
|
|
11038 |
|
|
case EQ: case NE:
|
11039 |
|
|
case UNEQ: case LTGT:
|
11040 |
|
|
case LT: case LTU: case UNLT: case LE: case LEU: case UNLE:
|
11041 |
|
|
case GT: case GTU: case UNGT: case GE: case GEU: case UNGE:
|
11042 |
|
|
case UNORDERED: case ORDERED:
|
11043 |
|
|
/* We can't do anything if OP0 is a condition code value, rather
|
11044 |
|
|
than an actual data value. */
|
11045 |
|
|
if (const_op != 0
|
11046 |
|
|
|| CC0_P (XEXP (op0, 0))
|
11047 |
|
|
|| GET_MODE_CLASS (GET_MODE (XEXP (op0, 0))) == MODE_CC)
|
11048 |
|
|
break;
|
11049 |
|
|
|
11050 |
|
|
/* Get the two operands being compared. */
|
11051 |
|
|
if (GET_CODE (XEXP (op0, 0)) == COMPARE)
|
11052 |
|
|
tem = XEXP (XEXP (op0, 0), 0), tem1 = XEXP (XEXP (op0, 0), 1);
|
11053 |
|
|
else
|
11054 |
|
|
tem = XEXP (op0, 0), tem1 = XEXP (op0, 1);
|
11055 |
|
|
|
11056 |
|
|
/* Check for the cases where we simply want the result of the
|
11057 |
|
|
earlier test or the opposite of that result. */
|
11058 |
|
|
if (code == NE || code == EQ
|
11059 |
|
|
|| (GET_MODE_BITSIZE (GET_MODE (op0)) <= HOST_BITS_PER_WIDE_INT
|
11060 |
|
|
&& GET_MODE_CLASS (GET_MODE (op0)) == MODE_INT
|
11061 |
|
|
&& (STORE_FLAG_VALUE
|
11062 |
|
|
& (((HOST_WIDE_INT) 1
|
11063 |
|
|
<< (GET_MODE_BITSIZE (GET_MODE (op0)) - 1))))
|
11064 |
|
|
&& (code == LT || code == GE)))
|
11065 |
|
|
{
|
11066 |
|
|
enum rtx_code new_code;
|
11067 |
|
|
if (code == LT || code == NE)
|
11068 |
|
|
new_code = GET_CODE (op0);
|
11069 |
|
|
else
|
11070 |
|
|
new_code = reversed_comparison_code (op0, NULL);
|
11071 |
|
|
|
11072 |
|
|
if (new_code != UNKNOWN)
|
11073 |
|
|
{
|
11074 |
|
|
code = new_code;
|
11075 |
|
|
op0 = tem;
|
11076 |
|
|
op1 = tem1;
|
11077 |
|
|
continue;
|
11078 |
|
|
}
|
11079 |
|
|
}
|
11080 |
|
|
break;
|
11081 |
|
|
|
11082 |
|
|
case IOR:
|
11083 |
|
|
/* The sign bit of (ior (plus X (const_int -1)) X) is nonzero
|
11084 |
|
|
iff X <= 0. */
|
11085 |
|
|
if (sign_bit_comparison_p && GET_CODE (XEXP (op0, 0)) == PLUS
|
11086 |
|
|
&& XEXP (XEXP (op0, 0), 1) == constm1_rtx
|
11087 |
|
|
&& rtx_equal_p (XEXP (XEXP (op0, 0), 0), XEXP (op0, 1)))
|
11088 |
|
|
{
|
11089 |
|
|
op0 = XEXP (op0, 1);
|
11090 |
|
|
code = (code == GE ? GT : LE);
|
11091 |
|
|
continue;
|
11092 |
|
|
}
|
11093 |
|
|
break;
|
11094 |
|
|
|
11095 |
|
|
case AND:
|
11096 |
|
|
/* Convert (and (xshift 1 X) Y) to (and (lshiftrt Y X) 1). This
|
11097 |
|
|
will be converted to a ZERO_EXTRACT later. */
|
11098 |
|
|
if (const_op == 0 && equality_comparison_p
|
11099 |
|
|
&& GET_CODE (XEXP (op0, 0)) == ASHIFT
|
11100 |
|
|
&& XEXP (XEXP (op0, 0), 0) == const1_rtx)
|
11101 |
|
|
{
|
11102 |
|
|
op0 = simplify_and_const_int
|
11103 |
|
|
(NULL_RTX, mode, gen_rtx_LSHIFTRT (mode,
|
11104 |
|
|
XEXP (op0, 1),
|
11105 |
|
|
XEXP (XEXP (op0, 0), 1)),
|
11106 |
|
|
(HOST_WIDE_INT) 1);
|
11107 |
|
|
continue;
|
11108 |
|
|
}
|
11109 |
|
|
|
11110 |
|
|
/* If we are comparing (and (lshiftrt X C1) C2) for equality with
|
11111 |
|
|
zero and X is a comparison and C1 and C2 describe only bits set
|
11112 |
|
|
in STORE_FLAG_VALUE, we can compare with X. */
|
11113 |
|
|
if (const_op == 0 && equality_comparison_p
|
11114 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
11115 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11116 |
|
|
&& GET_CODE (XEXP (op0, 0)) == LSHIFTRT
|
11117 |
|
|
&& CONST_INT_P (XEXP (XEXP (op0, 0), 1))
|
11118 |
|
|
&& INTVAL (XEXP (XEXP (op0, 0), 1)) >= 0
|
11119 |
|
|
&& INTVAL (XEXP (XEXP (op0, 0), 1)) < HOST_BITS_PER_WIDE_INT)
|
11120 |
|
|
{
|
11121 |
|
|
mask = ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
|
11122 |
|
|
<< INTVAL (XEXP (XEXP (op0, 0), 1)));
|
11123 |
|
|
if ((~STORE_FLAG_VALUE & mask) == 0
|
11124 |
|
|
&& (COMPARISON_P (XEXP (XEXP (op0, 0), 0))
|
11125 |
|
|
|| ((tem = get_last_value (XEXP (XEXP (op0, 0), 0))) != 0
|
11126 |
|
|
&& COMPARISON_P (tem))))
|
11127 |
|
|
{
|
11128 |
|
|
op0 = XEXP (XEXP (op0, 0), 0);
|
11129 |
|
|
continue;
|
11130 |
|
|
}
|
11131 |
|
|
}
|
11132 |
|
|
|
11133 |
|
|
/* If we are doing an equality comparison of an AND of a bit equal
|
11134 |
|
|
to the sign bit, replace this with a LT or GE comparison of
|
11135 |
|
|
the underlying value. */
|
11136 |
|
|
if (equality_comparison_p
|
11137 |
|
|
&& const_op == 0
|
11138 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11139 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
11140 |
|
|
&& ((INTVAL (XEXP (op0, 1)) & GET_MODE_MASK (mode))
|
11141 |
|
|
== (unsigned HOST_WIDE_INT) 1 << (mode_width - 1)))
|
11142 |
|
|
{
|
11143 |
|
|
op0 = XEXP (op0, 0);
|
11144 |
|
|
code = (code == EQ ? GE : LT);
|
11145 |
|
|
continue;
|
11146 |
|
|
}
|
11147 |
|
|
|
11148 |
|
|
/* If this AND operation is really a ZERO_EXTEND from a narrower
|
11149 |
|
|
mode, the constant fits within that mode, and this is either an
|
11150 |
|
|
equality or unsigned comparison, try to do this comparison in
|
11151 |
|
|
the narrower mode.
|
11152 |
|
|
|
11153 |
|
|
Note that in:
|
11154 |
|
|
|
11155 |
|
|
(ne:DI (and:DI (reg:DI 4) (const_int 0xffffffff)) (const_int 0))
|
11156 |
|
|
-> (ne:DI (reg:SI 4) (const_int 0))
|
11157 |
|
|
|
11158 |
|
|
unless TRULY_NOOP_TRUNCATION allows it or the register is
|
11159 |
|
|
known to hold a value of the required mode the
|
11160 |
|
|
transformation is invalid. */
|
11161 |
|
|
if ((equality_comparison_p || unsigned_comparison_p)
|
11162 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11163 |
|
|
&& (i = exact_log2 ((INTVAL (XEXP (op0, 1))
|
11164 |
|
|
& GET_MODE_MASK (mode))
|
11165 |
|
|
+ 1)) >= 0
|
11166 |
|
|
&& const_op >> i == 0
|
11167 |
|
|
&& (tmode = mode_for_size (i, MODE_INT, 1)) != BLKmode
|
11168 |
|
|
&& (TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (tmode),
|
11169 |
|
|
GET_MODE_BITSIZE (GET_MODE (op0)))
|
11170 |
|
|
|| (REG_P (XEXP (op0, 0))
|
11171 |
|
|
&& reg_truncated_to_mode (tmode, XEXP (op0, 0)))))
|
11172 |
|
|
{
|
11173 |
|
|
op0 = gen_lowpart (tmode, XEXP (op0, 0));
|
11174 |
|
|
continue;
|
11175 |
|
|
}
|
11176 |
|
|
|
11177 |
|
|
/* If this is (and:M1 (subreg:M2 X 0) (const_int C1)) where C1
|
11178 |
|
|
fits in both M1 and M2 and the SUBREG is either paradoxical
|
11179 |
|
|
or represents the low part, permute the SUBREG and the AND
|
11180 |
|
|
and try again. */
|
11181 |
|
|
if (GET_CODE (XEXP (op0, 0)) == SUBREG)
|
11182 |
|
|
{
|
11183 |
|
|
unsigned HOST_WIDE_INT c1;
|
11184 |
|
|
tmode = GET_MODE (SUBREG_REG (XEXP (op0, 0)));
|
11185 |
|
|
/* Require an integral mode, to avoid creating something like
|
11186 |
|
|
(AND:SF ...). */
|
11187 |
|
|
if (SCALAR_INT_MODE_P (tmode)
|
11188 |
|
|
/* It is unsafe to commute the AND into the SUBREG if the
|
11189 |
|
|
SUBREG is paradoxical and WORD_REGISTER_OPERATIONS is
|
11190 |
|
|
not defined. As originally written the upper bits
|
11191 |
|
|
have a defined value due to the AND operation.
|
11192 |
|
|
However, if we commute the AND inside the SUBREG then
|
11193 |
|
|
they no longer have defined values and the meaning of
|
11194 |
|
|
the code has been changed. */
|
11195 |
|
|
&& (0
|
11196 |
|
|
#ifdef WORD_REGISTER_OPERATIONS
|
11197 |
|
|
|| (mode_width > GET_MODE_BITSIZE (tmode)
|
11198 |
|
|
&& mode_width <= BITS_PER_WORD)
|
11199 |
|
|
#endif
|
11200 |
|
|
|| (mode_width <= GET_MODE_BITSIZE (tmode)
|
11201 |
|
|
&& subreg_lowpart_p (XEXP (op0, 0))))
|
11202 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11203 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
11204 |
|
|
&& GET_MODE_BITSIZE (tmode) <= HOST_BITS_PER_WIDE_INT
|
11205 |
|
|
&& ((c1 = INTVAL (XEXP (op0, 1))) & ~mask) == 0
|
11206 |
|
|
&& (c1 & ~GET_MODE_MASK (tmode)) == 0
|
11207 |
|
|
&& c1 != mask
|
11208 |
|
|
&& c1 != GET_MODE_MASK (tmode))
|
11209 |
|
|
{
|
11210 |
|
|
op0 = simplify_gen_binary (AND, tmode,
|
11211 |
|
|
SUBREG_REG (XEXP (op0, 0)),
|
11212 |
|
|
gen_int_mode (c1, tmode));
|
11213 |
|
|
op0 = gen_lowpart (mode, op0);
|
11214 |
|
|
continue;
|
11215 |
|
|
}
|
11216 |
|
|
}
|
11217 |
|
|
|
11218 |
|
|
/* Convert (ne (and (not X) 1) 0) to (eq (and X 1) 0). */
|
11219 |
|
|
if (const_op == 0 && equality_comparison_p
|
11220 |
|
|
&& XEXP (op0, 1) == const1_rtx
|
11221 |
|
|
&& GET_CODE (XEXP (op0, 0)) == NOT)
|
11222 |
|
|
{
|
11223 |
|
|
op0 = simplify_and_const_int
|
11224 |
|
|
(NULL_RTX, mode, XEXP (XEXP (op0, 0), 0), (HOST_WIDE_INT) 1);
|
11225 |
|
|
code = (code == NE ? EQ : NE);
|
11226 |
|
|
continue;
|
11227 |
|
|
}
|
11228 |
|
|
|
11229 |
|
|
/* Convert (ne (and (lshiftrt (not X)) 1) 0) to
|
11230 |
|
|
(eq (and (lshiftrt X) 1) 0).
|
11231 |
|
|
Also handle the case where (not X) is expressed using xor. */
|
11232 |
|
|
if (const_op == 0 && equality_comparison_p
|
11233 |
|
|
&& XEXP (op0, 1) == const1_rtx
|
11234 |
|
|
&& GET_CODE (XEXP (op0, 0)) == LSHIFTRT)
|
11235 |
|
|
{
|
11236 |
|
|
rtx shift_op = XEXP (XEXP (op0, 0), 0);
|
11237 |
|
|
rtx shift_count = XEXP (XEXP (op0, 0), 1);
|
11238 |
|
|
|
11239 |
|
|
if (GET_CODE (shift_op) == NOT
|
11240 |
|
|
|| (GET_CODE (shift_op) == XOR
|
11241 |
|
|
&& CONST_INT_P (XEXP (shift_op, 1))
|
11242 |
|
|
&& CONST_INT_P (shift_count)
|
11243 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
|
11244 |
|
|
&& (INTVAL (XEXP (shift_op, 1))
|
11245 |
|
|
== (HOST_WIDE_INT) 1 << INTVAL (shift_count))))
|
11246 |
|
|
{
|
11247 |
|
|
op0 = simplify_and_const_int
|
11248 |
|
|
(NULL_RTX, mode,
|
11249 |
|
|
gen_rtx_LSHIFTRT (mode, XEXP (shift_op, 0), shift_count),
|
11250 |
|
|
(HOST_WIDE_INT) 1);
|
11251 |
|
|
code = (code == NE ? EQ : NE);
|
11252 |
|
|
continue;
|
11253 |
|
|
}
|
11254 |
|
|
}
|
11255 |
|
|
break;
|
11256 |
|
|
|
11257 |
|
|
case ASHIFT:
|
11258 |
|
|
/* If we have (compare (ashift FOO N) (const_int C)) and
|
11259 |
|
|
the high order N bits of FOO (N+1 if an inequality comparison)
|
11260 |
|
|
are known to be zero, we can do this by comparing FOO with C
|
11261 |
|
|
shifted right N bits so long as the low-order N bits of C are
|
11262 |
|
|
zero. */
|
11263 |
|
|
if (CONST_INT_P (XEXP (op0, 1))
|
11264 |
|
|
&& INTVAL (XEXP (op0, 1)) >= 0
|
11265 |
|
|
&& ((INTVAL (XEXP (op0, 1)) + ! equality_comparison_p)
|
11266 |
|
|
< HOST_BITS_PER_WIDE_INT)
|
11267 |
|
|
&& ((const_op
|
11268 |
|
|
& (((HOST_WIDE_INT) 1 << INTVAL (XEXP (op0, 1))) - 1)) == 0)
|
11269 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
11270 |
|
|
&& (nonzero_bits (XEXP (op0, 0), mode)
|
11271 |
|
|
& ~(mask >> (INTVAL (XEXP (op0, 1))
|
11272 |
|
|
+ ! equality_comparison_p))) == 0)
|
11273 |
|
|
{
|
11274 |
|
|
/* We must perform a logical shift, not an arithmetic one,
|
11275 |
|
|
as we want the top N bits of C to be zero. */
|
11276 |
|
|
unsigned HOST_WIDE_INT temp = const_op & GET_MODE_MASK (mode);
|
11277 |
|
|
|
11278 |
|
|
temp >>= INTVAL (XEXP (op0, 1));
|
11279 |
|
|
op1 = gen_int_mode (temp, mode);
|
11280 |
|
|
op0 = XEXP (op0, 0);
|
11281 |
|
|
continue;
|
11282 |
|
|
}
|
11283 |
|
|
|
11284 |
|
|
/* If we are doing a sign bit comparison, it means we are testing
|
11285 |
|
|
a particular bit. Convert it to the appropriate AND. */
|
11286 |
|
|
if (sign_bit_comparison_p && CONST_INT_P (XEXP (op0, 1))
|
11287 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT)
|
11288 |
|
|
{
|
11289 |
|
|
op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
|
11290 |
|
|
((HOST_WIDE_INT) 1
|
11291 |
|
|
<< (mode_width - 1
|
11292 |
|
|
- INTVAL (XEXP (op0, 1)))));
|
11293 |
|
|
code = (code == LT ? NE : EQ);
|
11294 |
|
|
continue;
|
11295 |
|
|
}
|
11296 |
|
|
|
11297 |
|
|
/* If this an equality comparison with zero and we are shifting
|
11298 |
|
|
the low bit to the sign bit, we can convert this to an AND of the
|
11299 |
|
|
low-order bit. */
|
11300 |
|
|
if (const_op == 0 && equality_comparison_p
|
11301 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11302 |
|
|
&& (unsigned HOST_WIDE_INT) INTVAL (XEXP (op0, 1))
|
11303 |
|
|
== mode_width - 1)
|
11304 |
|
|
{
|
11305 |
|
|
op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
|
11306 |
|
|
(HOST_WIDE_INT) 1);
|
11307 |
|
|
continue;
|
11308 |
|
|
}
|
11309 |
|
|
break;
|
11310 |
|
|
|
11311 |
|
|
case ASHIFTRT:
|
11312 |
|
|
/* If this is an equality comparison with zero, we can do this
|
11313 |
|
|
as a logical shift, which might be much simpler. */
|
11314 |
|
|
if (equality_comparison_p && const_op == 0
|
11315 |
|
|
&& CONST_INT_P (XEXP (op0, 1)))
|
11316 |
|
|
{
|
11317 |
|
|
op0 = simplify_shift_const (NULL_RTX, LSHIFTRT, mode,
|
11318 |
|
|
XEXP (op0, 0),
|
11319 |
|
|
INTVAL (XEXP (op0, 1)));
|
11320 |
|
|
continue;
|
11321 |
|
|
}
|
11322 |
|
|
|
11323 |
|
|
/* If OP0 is a sign extension and CODE is not an unsigned comparison,
|
11324 |
|
|
do the comparison in a narrower mode. */
|
11325 |
|
|
if (! unsigned_comparison_p
|
11326 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11327 |
|
|
&& GET_CODE (XEXP (op0, 0)) == ASHIFT
|
11328 |
|
|
&& XEXP (op0, 1) == XEXP (XEXP (op0, 0), 1)
|
11329 |
|
|
&& (tmode = mode_for_size (mode_width - INTVAL (XEXP (op0, 1)),
|
11330 |
|
|
MODE_INT, 1)) != BLKmode
|
11331 |
|
|
&& (((unsigned HOST_WIDE_INT) const_op
|
11332 |
|
|
+ (GET_MODE_MASK (tmode) >> 1) + 1)
|
11333 |
|
|
<= GET_MODE_MASK (tmode)))
|
11334 |
|
|
{
|
11335 |
|
|
op0 = gen_lowpart (tmode, XEXP (XEXP (op0, 0), 0));
|
11336 |
|
|
continue;
|
11337 |
|
|
}
|
11338 |
|
|
|
11339 |
|
|
/* Likewise if OP0 is a PLUS of a sign extension with a
|
11340 |
|
|
constant, which is usually represented with the PLUS
|
11341 |
|
|
between the shifts. */
|
11342 |
|
|
if (! unsigned_comparison_p
|
11343 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11344 |
|
|
&& GET_CODE (XEXP (op0, 0)) == PLUS
|
11345 |
|
|
&& CONST_INT_P (XEXP (XEXP (op0, 0), 1))
|
11346 |
|
|
&& GET_CODE (XEXP (XEXP (op0, 0), 0)) == ASHIFT
|
11347 |
|
|
&& XEXP (op0, 1) == XEXP (XEXP (XEXP (op0, 0), 0), 1)
|
11348 |
|
|
&& (tmode = mode_for_size (mode_width - INTVAL (XEXP (op0, 1)),
|
11349 |
|
|
MODE_INT, 1)) != BLKmode
|
11350 |
|
|
&& (((unsigned HOST_WIDE_INT) const_op
|
11351 |
|
|
+ (GET_MODE_MASK (tmode) >> 1) + 1)
|
11352 |
|
|
<= GET_MODE_MASK (tmode)))
|
11353 |
|
|
{
|
11354 |
|
|
rtx inner = XEXP (XEXP (XEXP (op0, 0), 0), 0);
|
11355 |
|
|
rtx add_const = XEXP (XEXP (op0, 0), 1);
|
11356 |
|
|
rtx new_const = simplify_gen_binary (ASHIFTRT, GET_MODE (op0),
|
11357 |
|
|
add_const, XEXP (op0, 1));
|
11358 |
|
|
|
11359 |
|
|
op0 = simplify_gen_binary (PLUS, tmode,
|
11360 |
|
|
gen_lowpart (tmode, inner),
|
11361 |
|
|
new_const);
|
11362 |
|
|
continue;
|
11363 |
|
|
}
|
11364 |
|
|
|
11365 |
|
|
/* ... fall through ... */
|
11366 |
|
|
case LSHIFTRT:
|
11367 |
|
|
/* If we have (compare (xshiftrt FOO N) (const_int C)) and
|
11368 |
|
|
the low order N bits of FOO are known to be zero, we can do this
|
11369 |
|
|
by comparing FOO with C shifted left N bits so long as no
|
11370 |
|
|
overflow occurs. */
|
11371 |
|
|
if (CONST_INT_P (XEXP (op0, 1))
|
11372 |
|
|
&& INTVAL (XEXP (op0, 1)) >= 0
|
11373 |
|
|
&& INTVAL (XEXP (op0, 1)) < HOST_BITS_PER_WIDE_INT
|
11374 |
|
|
&& mode_width <= HOST_BITS_PER_WIDE_INT
|
11375 |
|
|
&& (nonzero_bits (XEXP (op0, 0), mode)
|
11376 |
|
|
& (((HOST_WIDE_INT) 1 << INTVAL (XEXP (op0, 1))) - 1)) == 0
|
11377 |
|
|
&& (((unsigned HOST_WIDE_INT) const_op
|
11378 |
|
|
+ (GET_CODE (op0) != LSHIFTRT
|
11379 |
|
|
? ((GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1)) >> 1)
|
11380 |
|
|
+ 1)
|
11381 |
|
|
: 0))
|
11382 |
|
|
<= GET_MODE_MASK (mode) >> INTVAL (XEXP (op0, 1))))
|
11383 |
|
|
{
|
11384 |
|
|
/* If the shift was logical, then we must make the condition
|
11385 |
|
|
unsigned. */
|
11386 |
|
|
if (GET_CODE (op0) == LSHIFTRT)
|
11387 |
|
|
code = unsigned_condition (code);
|
11388 |
|
|
|
11389 |
|
|
const_op <<= INTVAL (XEXP (op0, 1));
|
11390 |
|
|
op1 = GEN_INT (const_op);
|
11391 |
|
|
op0 = XEXP (op0, 0);
|
11392 |
|
|
continue;
|
11393 |
|
|
}
|
11394 |
|
|
|
11395 |
|
|
/* If we are using this shift to extract just the sign bit, we
|
11396 |
|
|
can replace this with an LT or GE comparison. */
|
11397 |
|
|
if (const_op == 0
|
11398 |
|
|
&& (equality_comparison_p || sign_bit_comparison_p)
|
11399 |
|
|
&& CONST_INT_P (XEXP (op0, 1))
|
11400 |
|
|
&& (unsigned HOST_WIDE_INT) INTVAL (XEXP (op0, 1))
|
11401 |
|
|
== mode_width - 1)
|
11402 |
|
|
{
|
11403 |
|
|
op0 = XEXP (op0, 0);
|
11404 |
|
|
code = (code == NE || code == GT ? LT : GE);
|
11405 |
|
|
continue;
|
11406 |
|
|
}
|
11407 |
|
|
break;
|
11408 |
|
|
|
11409 |
|
|
default:
|
11410 |
|
|
break;
|
11411 |
|
|
}
|
11412 |
|
|
|
11413 |
|
|
break;
|
11414 |
|
|
}
|
11415 |
|
|
|
11416 |
|
|
/* Now make any compound operations involved in this comparison. Then,
|
11417 |
|
|
check for an outmost SUBREG on OP0 that is not doing anything or is
|
11418 |
|
|
paradoxical. The latter transformation must only be performed when
|
11419 |
|
|
it is known that the "extra" bits will be the same in op0 and op1 or
|
11420 |
|
|
that they don't matter. There are three cases to consider:
|
11421 |
|
|
|
11422 |
|
|
1. SUBREG_REG (op0) is a register. In this case the bits are don't
|
11423 |
|
|
care bits and we can assume they have any convenient value. So
|
11424 |
|
|
making the transformation is safe.
|
11425 |
|
|
|
11426 |
|
|
2. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is not defined.
|
11427 |
|
|
In this case the upper bits of op0 are undefined. We should not make
|
11428 |
|
|
the simplification in that case as we do not know the contents of
|
11429 |
|
|
those bits.
|
11430 |
|
|
|
11431 |
|
|
3. SUBREG_REG (op0) is a memory and LOAD_EXTEND_OP is defined and not
|
11432 |
|
|
UNKNOWN. In that case we know those bits are zeros or ones. We must
|
11433 |
|
|
also be sure that they are the same as the upper bits of op1.
|
11434 |
|
|
|
11435 |
|
|
We can never remove a SUBREG for a non-equality comparison because
|
11436 |
|
|
the sign bit is in a different place in the underlying object. */
|
11437 |
|
|
|
11438 |
|
|
op0 = make_compound_operation (op0, op1 == const0_rtx ? COMPARE : SET);
|
11439 |
|
|
op1 = make_compound_operation (op1, SET);
|
11440 |
|
|
|
11441 |
|
|
if (GET_CODE (op0) == SUBREG && subreg_lowpart_p (op0)
|
11442 |
|
|
&& GET_MODE_CLASS (GET_MODE (op0)) == MODE_INT
|
11443 |
|
|
&& GET_MODE_CLASS (GET_MODE (SUBREG_REG (op0))) == MODE_INT
|
11444 |
|
|
&& (code == NE || code == EQ))
|
11445 |
|
|
{
|
11446 |
|
|
if (GET_MODE_SIZE (GET_MODE (op0))
|
11447 |
|
|
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0))))
|
11448 |
|
|
{
|
11449 |
|
|
/* For paradoxical subregs, allow case 1 as above. Case 3 isn't
|
11450 |
|
|
implemented. */
|
11451 |
|
|
if (REG_P (SUBREG_REG (op0)))
|
11452 |
|
|
{
|
11453 |
|
|
op0 = SUBREG_REG (op0);
|
11454 |
|
|
op1 = gen_lowpart (GET_MODE (op0), op1);
|
11455 |
|
|
}
|
11456 |
|
|
}
|
11457 |
|
|
else if ((GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (op0)))
|
11458 |
|
|
<= HOST_BITS_PER_WIDE_INT)
|
11459 |
|
|
&& (nonzero_bits (SUBREG_REG (op0),
|
11460 |
|
|
GET_MODE (SUBREG_REG (op0)))
|
11461 |
|
|
& ~GET_MODE_MASK (GET_MODE (op0))) == 0)
|
11462 |
|
|
{
|
11463 |
|
|
tem = gen_lowpart (GET_MODE (SUBREG_REG (op0)), op1);
|
11464 |
|
|
|
11465 |
|
|
if ((nonzero_bits (tem, GET_MODE (SUBREG_REG (op0)))
|
11466 |
|
|
& ~GET_MODE_MASK (GET_MODE (op0))) == 0)
|
11467 |
|
|
op0 = SUBREG_REG (op0), op1 = tem;
|
11468 |
|
|
}
|
11469 |
|
|
}
|
11470 |
|
|
|
11471 |
|
|
/* We now do the opposite procedure: Some machines don't have compare
|
11472 |
|
|
insns in all modes. If OP0's mode is an integer mode smaller than a
|
11473 |
|
|
word and we can't do a compare in that mode, see if there is a larger
|
11474 |
|
|
mode for which we can do the compare. There are a number of cases in
|
11475 |
|
|
which we can use the wider mode. */
|
11476 |
|
|
|
11477 |
|
|
mode = GET_MODE (op0);
|
11478 |
|
|
if (mode != VOIDmode && GET_MODE_CLASS (mode) == MODE_INT
|
11479 |
|
|
&& GET_MODE_SIZE (mode) < UNITS_PER_WORD
|
11480 |
|
|
&& ! have_insn_for (COMPARE, mode))
|
11481 |
|
|
for (tmode = GET_MODE_WIDER_MODE (mode);
|
11482 |
|
|
(tmode != VOIDmode
|
11483 |
|
|
&& GET_MODE_BITSIZE (tmode) <= HOST_BITS_PER_WIDE_INT);
|
11484 |
|
|
tmode = GET_MODE_WIDER_MODE (tmode))
|
11485 |
|
|
if (have_insn_for (COMPARE, tmode))
|
11486 |
|
|
{
|
11487 |
|
|
int zero_extended;
|
11488 |
|
|
|
11489 |
|
|
/* If this is a test for negative, we can make an explicit
|
11490 |
|
|
test of the sign bit. Test this first so we can use
|
11491 |
|
|
a paradoxical subreg to extend OP0. */
|
11492 |
|
|
|
11493 |
|
|
if (op1 == const0_rtx && (code == LT || code == GE)
|
11494 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
11495 |
|
|
{
|
11496 |
|
|
op0 = simplify_gen_binary (AND, tmode,
|
11497 |
|
|
gen_lowpart (tmode, op0),
|
11498 |
|
|
GEN_INT ((HOST_WIDE_INT) 1
|
11499 |
|
|
<< (GET_MODE_BITSIZE (mode)
|
11500 |
|
|
- 1)));
|
11501 |
|
|
code = (code == LT) ? NE : EQ;
|
11502 |
|
|
break;
|
11503 |
|
|
}
|
11504 |
|
|
|
11505 |
|
|
/* If the only nonzero bits in OP0 and OP1 are those in the
|
11506 |
|
|
narrower mode and this is an equality or unsigned comparison,
|
11507 |
|
|
we can use the wider mode. Similarly for sign-extended
|
11508 |
|
|
values, in which case it is true for all comparisons. */
|
11509 |
|
|
zero_extended = ((code == EQ || code == NE
|
11510 |
|
|
|| code == GEU || code == GTU
|
11511 |
|
|
|| code == LEU || code == LTU)
|
11512 |
|
|
&& (nonzero_bits (op0, tmode)
|
11513 |
|
|
& ~GET_MODE_MASK (mode)) == 0
|
11514 |
|
|
&& ((CONST_INT_P (op1)
|
11515 |
|
|
|| (nonzero_bits (op1, tmode)
|
11516 |
|
|
& ~GET_MODE_MASK (mode)) == 0)));
|
11517 |
|
|
|
11518 |
|
|
if (zero_extended
|
11519 |
|
|
|| ((num_sign_bit_copies (op0, tmode)
|
11520 |
|
|
> (unsigned int) (GET_MODE_BITSIZE (tmode)
|
11521 |
|
|
- GET_MODE_BITSIZE (mode)))
|
11522 |
|
|
&& (num_sign_bit_copies (op1, tmode)
|
11523 |
|
|
> (unsigned int) (GET_MODE_BITSIZE (tmode)
|
11524 |
|
|
- GET_MODE_BITSIZE (mode)))))
|
11525 |
|
|
{
|
11526 |
|
|
/* If OP0 is an AND and we don't have an AND in MODE either,
|
11527 |
|
|
make a new AND in the proper mode. */
|
11528 |
|
|
if (GET_CODE (op0) == AND
|
11529 |
|
|
&& !have_insn_for (AND, mode))
|
11530 |
|
|
op0 = simplify_gen_binary (AND, tmode,
|
11531 |
|
|
gen_lowpart (tmode,
|
11532 |
|
|
XEXP (op0, 0)),
|
11533 |
|
|
gen_lowpart (tmode,
|
11534 |
|
|
XEXP (op0, 1)));
|
11535 |
|
|
else
|
11536 |
|
|
{
|
11537 |
|
|
if (zero_extended)
|
11538 |
|
|
{
|
11539 |
|
|
op0 = simplify_gen_unary (ZERO_EXTEND, tmode, op0, mode);
|
11540 |
|
|
op1 = simplify_gen_unary (ZERO_EXTEND, tmode, op1, mode);
|
11541 |
|
|
}
|
11542 |
|
|
else
|
11543 |
|
|
{
|
11544 |
|
|
op0 = simplify_gen_unary (SIGN_EXTEND, tmode, op0, mode);
|
11545 |
|
|
op1 = simplify_gen_unary (SIGN_EXTEND, tmode, op1, mode);
|
11546 |
|
|
}
|
11547 |
|
|
break;
|
11548 |
|
|
}
|
11549 |
|
|
}
|
11550 |
|
|
}
|
11551 |
|
|
|
11552 |
|
|
#ifdef CANONICALIZE_COMPARISON
|
11553 |
|
|
/* If this machine only supports a subset of valid comparisons, see if we
|
11554 |
|
|
can convert an unsupported one into a supported one. */
|
11555 |
|
|
CANONICALIZE_COMPARISON (code, op0, op1);
|
11556 |
|
|
#endif
|
11557 |
|
|
|
11558 |
|
|
*pop0 = op0;
|
11559 |
|
|
*pop1 = op1;
|
11560 |
|
|
|
11561 |
|
|
return code;
|
11562 |
|
|
}
|
11563 |
|
|
|
11564 |
|
|
/* Utility function for record_value_for_reg. Count number of
|
11565 |
|
|
rtxs in X. */
|
11566 |
|
|
static int
|
11567 |
|
|
count_rtxs (rtx x)
|
11568 |
|
|
{
|
11569 |
|
|
enum rtx_code code = GET_CODE (x);
|
11570 |
|
|
const char *fmt;
|
11571 |
|
|
int i, j, ret = 1;
|
11572 |
|
|
|
11573 |
|
|
if (GET_RTX_CLASS (code) == '2'
|
11574 |
|
|
|| GET_RTX_CLASS (code) == 'c')
|
11575 |
|
|
{
|
11576 |
|
|
rtx x0 = XEXP (x, 0);
|
11577 |
|
|
rtx x1 = XEXP (x, 1);
|
11578 |
|
|
|
11579 |
|
|
if (x0 == x1)
|
11580 |
|
|
return 1 + 2 * count_rtxs (x0);
|
11581 |
|
|
|
11582 |
|
|
if ((GET_RTX_CLASS (GET_CODE (x1)) == '2'
|
11583 |
|
|
|| GET_RTX_CLASS (GET_CODE (x1)) == 'c')
|
11584 |
|
|
&& (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
|
11585 |
|
|
return 2 + 2 * count_rtxs (x0)
|
11586 |
|
|
+ count_rtxs (x == XEXP (x1, 0)
|
11587 |
|
|
? XEXP (x1, 1) : XEXP (x1, 0));
|
11588 |
|
|
|
11589 |
|
|
if ((GET_RTX_CLASS (GET_CODE (x0)) == '2'
|
11590 |
|
|
|| GET_RTX_CLASS (GET_CODE (x0)) == 'c')
|
11591 |
|
|
&& (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
|
11592 |
|
|
return 2 + 2 * count_rtxs (x1)
|
11593 |
|
|
+ count_rtxs (x == XEXP (x0, 0)
|
11594 |
|
|
? XEXP (x0, 1) : XEXP (x0, 0));
|
11595 |
|
|
}
|
11596 |
|
|
|
11597 |
|
|
fmt = GET_RTX_FORMAT (code);
|
11598 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
11599 |
|
|
if (fmt[i] == 'e')
|
11600 |
|
|
ret += count_rtxs (XEXP (x, i));
|
11601 |
|
|
else if (fmt[i] == 'E')
|
11602 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
11603 |
|
|
ret += count_rtxs (XVECEXP (x, i, j));
|
11604 |
|
|
|
11605 |
|
|
return ret;
|
11606 |
|
|
}
|
11607 |
|
|
|
11608 |
|
|
/* Utility function for following routine. Called when X is part of a value
|
11609 |
|
|
being stored into last_set_value. Sets last_set_table_tick
|
11610 |
|
|
for each register mentioned. Similar to mention_regs in cse.c */
|
11611 |
|
|
|
11612 |
|
|
static void
|
11613 |
|
|
update_table_tick (rtx x)
|
11614 |
|
|
{
|
11615 |
|
|
enum rtx_code code = GET_CODE (x);
|
11616 |
|
|
const char *fmt = GET_RTX_FORMAT (code);
|
11617 |
|
|
int i, j;
|
11618 |
|
|
|
11619 |
|
|
if (code == REG)
|
11620 |
|
|
{
|
11621 |
|
|
unsigned int regno = REGNO (x);
|
11622 |
|
|
unsigned int endregno = END_REGNO (x);
|
11623 |
|
|
unsigned int r;
|
11624 |
|
|
|
11625 |
|
|
for (r = regno; r < endregno; r++)
|
11626 |
|
|
{
|
11627 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, r);
|
11628 |
|
|
rsp->last_set_table_tick = label_tick;
|
11629 |
|
|
}
|
11630 |
|
|
|
11631 |
|
|
return;
|
11632 |
|
|
}
|
11633 |
|
|
|
11634 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
11635 |
|
|
if (fmt[i] == 'e')
|
11636 |
|
|
{
|
11637 |
|
|
/* Check for identical subexpressions. If x contains
|
11638 |
|
|
identical subexpression we only have to traverse one of
|
11639 |
|
|
them. */
|
11640 |
|
|
if (i == 0 && ARITHMETIC_P (x))
|
11641 |
|
|
{
|
11642 |
|
|
/* Note that at this point x1 has already been
|
11643 |
|
|
processed. */
|
11644 |
|
|
rtx x0 = XEXP (x, 0);
|
11645 |
|
|
rtx x1 = XEXP (x, 1);
|
11646 |
|
|
|
11647 |
|
|
/* If x0 and x1 are identical then there is no need to
|
11648 |
|
|
process x0. */
|
11649 |
|
|
if (x0 == x1)
|
11650 |
|
|
break;
|
11651 |
|
|
|
11652 |
|
|
/* If x0 is identical to a subexpression of x1 then while
|
11653 |
|
|
processing x1, x0 has already been processed. Thus we
|
11654 |
|
|
are done with x. */
|
11655 |
|
|
if (ARITHMETIC_P (x1)
|
11656 |
|
|
&& (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
|
11657 |
|
|
break;
|
11658 |
|
|
|
11659 |
|
|
/* If x1 is identical to a subexpression of x0 then we
|
11660 |
|
|
still have to process the rest of x0. */
|
11661 |
|
|
if (ARITHMETIC_P (x0)
|
11662 |
|
|
&& (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
|
11663 |
|
|
{
|
11664 |
|
|
update_table_tick (XEXP (x0, x1 == XEXP (x0, 0) ? 1 : 0));
|
11665 |
|
|
break;
|
11666 |
|
|
}
|
11667 |
|
|
}
|
11668 |
|
|
|
11669 |
|
|
update_table_tick (XEXP (x, i));
|
11670 |
|
|
}
|
11671 |
|
|
else if (fmt[i] == 'E')
|
11672 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
11673 |
|
|
update_table_tick (XVECEXP (x, i, j));
|
11674 |
|
|
}
|
11675 |
|
|
|
11676 |
|
|
/* Record that REG is set to VALUE in insn INSN. If VALUE is zero, we
|
11677 |
|
|
are saying that the register is clobbered and we no longer know its
|
11678 |
|
|
value. If INSN is zero, don't update reg_stat[].last_set; this is
|
11679 |
|
|
only permitted with VALUE also zero and is used to invalidate the
|
11680 |
|
|
register. */
|
11681 |
|
|
|
11682 |
|
|
static void
|
11683 |
|
|
record_value_for_reg (rtx reg, rtx insn, rtx value)
|
11684 |
|
|
{
|
11685 |
|
|
unsigned int regno = REGNO (reg);
|
11686 |
|
|
unsigned int endregno = END_REGNO (reg);
|
11687 |
|
|
unsigned int i;
|
11688 |
|
|
reg_stat_type *rsp;
|
11689 |
|
|
|
11690 |
|
|
/* If VALUE contains REG and we have a previous value for REG, substitute
|
11691 |
|
|
the previous value. */
|
11692 |
|
|
if (value && insn && reg_overlap_mentioned_p (reg, value))
|
11693 |
|
|
{
|
11694 |
|
|
rtx tem;
|
11695 |
|
|
|
11696 |
|
|
/* Set things up so get_last_value is allowed to see anything set up to
|
11697 |
|
|
our insn. */
|
11698 |
|
|
subst_low_luid = DF_INSN_LUID (insn);
|
11699 |
|
|
tem = get_last_value (reg);
|
11700 |
|
|
|
11701 |
|
|
/* If TEM is simply a binary operation with two CLOBBERs as operands,
|
11702 |
|
|
it isn't going to be useful and will take a lot of time to process,
|
11703 |
|
|
so just use the CLOBBER. */
|
11704 |
|
|
|
11705 |
|
|
if (tem)
|
11706 |
|
|
{
|
11707 |
|
|
if (ARITHMETIC_P (tem)
|
11708 |
|
|
&& GET_CODE (XEXP (tem, 0)) == CLOBBER
|
11709 |
|
|
&& GET_CODE (XEXP (tem, 1)) == CLOBBER)
|
11710 |
|
|
tem = XEXP (tem, 0);
|
11711 |
|
|
else if (count_occurrences (value, reg, 1) >= 2)
|
11712 |
|
|
{
|
11713 |
|
|
/* If there are two or more occurrences of REG in VALUE,
|
11714 |
|
|
prevent the value from growing too much. */
|
11715 |
|
|
if (count_rtxs (tem) > MAX_LAST_VALUE_RTL)
|
11716 |
|
|
tem = gen_rtx_CLOBBER (GET_MODE (tem), const0_rtx);
|
11717 |
|
|
}
|
11718 |
|
|
|
11719 |
|
|
value = replace_rtx (copy_rtx (value), reg, tem);
|
11720 |
|
|
}
|
11721 |
|
|
}
|
11722 |
|
|
|
11723 |
|
|
/* For each register modified, show we don't know its value, that
|
11724 |
|
|
we don't know about its bitwise content, that its value has been
|
11725 |
|
|
updated, and that we don't know the location of the death of the
|
11726 |
|
|
register. */
|
11727 |
|
|
for (i = regno; i < endregno; i++)
|
11728 |
|
|
{
|
11729 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, i);
|
11730 |
|
|
|
11731 |
|
|
if (insn)
|
11732 |
|
|
rsp->last_set = insn;
|
11733 |
|
|
|
11734 |
|
|
rsp->last_set_value = 0;
|
11735 |
|
|
rsp->last_set_mode = VOIDmode;
|
11736 |
|
|
rsp->last_set_nonzero_bits = 0;
|
11737 |
|
|
rsp->last_set_sign_bit_copies = 0;
|
11738 |
|
|
rsp->last_death = 0;
|
11739 |
|
|
rsp->truncated_to_mode = VOIDmode;
|
11740 |
|
|
}
|
11741 |
|
|
|
11742 |
|
|
/* Mark registers that are being referenced in this value. */
|
11743 |
|
|
if (value)
|
11744 |
|
|
update_table_tick (value);
|
11745 |
|
|
|
11746 |
|
|
/* Now update the status of each register being set.
|
11747 |
|
|
If someone is using this register in this block, set this register
|
11748 |
|
|
to invalid since we will get confused between the two lives in this
|
11749 |
|
|
basic block. This makes using this register always invalid. In cse, we
|
11750 |
|
|
scan the table to invalidate all entries using this register, but this
|
11751 |
|
|
is too much work for us. */
|
11752 |
|
|
|
11753 |
|
|
for (i = regno; i < endregno; i++)
|
11754 |
|
|
{
|
11755 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, i);
|
11756 |
|
|
rsp->last_set_label = label_tick;
|
11757 |
|
|
if (!insn
|
11758 |
|
|
|| (value && rsp->last_set_table_tick >= label_tick_ebb_start))
|
11759 |
|
|
rsp->last_set_invalid = 1;
|
11760 |
|
|
else
|
11761 |
|
|
rsp->last_set_invalid = 0;
|
11762 |
|
|
}
|
11763 |
|
|
|
11764 |
|
|
/* The value being assigned might refer to X (like in "x++;"). In that
|
11765 |
|
|
case, we must replace it with (clobber (const_int 0)) to prevent
|
11766 |
|
|
infinite loops. */
|
11767 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, regno);
|
11768 |
|
|
if (value && !get_last_value_validate (&value, insn, label_tick, 0))
|
11769 |
|
|
{
|
11770 |
|
|
value = copy_rtx (value);
|
11771 |
|
|
if (!get_last_value_validate (&value, insn, label_tick, 1))
|
11772 |
|
|
value = 0;
|
11773 |
|
|
}
|
11774 |
|
|
|
11775 |
|
|
/* For the main register being modified, update the value, the mode, the
|
11776 |
|
|
nonzero bits, and the number of sign bit copies. */
|
11777 |
|
|
|
11778 |
|
|
rsp->last_set_value = value;
|
11779 |
|
|
|
11780 |
|
|
if (value)
|
11781 |
|
|
{
|
11782 |
|
|
enum machine_mode mode = GET_MODE (reg);
|
11783 |
|
|
subst_low_luid = DF_INSN_LUID (insn);
|
11784 |
|
|
rsp->last_set_mode = mode;
|
11785 |
|
|
if (GET_MODE_CLASS (mode) == MODE_INT
|
11786 |
|
|
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
|
11787 |
|
|
mode = nonzero_bits_mode;
|
11788 |
|
|
rsp->last_set_nonzero_bits = nonzero_bits (value, mode);
|
11789 |
|
|
rsp->last_set_sign_bit_copies
|
11790 |
|
|
= num_sign_bit_copies (value, GET_MODE (reg));
|
11791 |
|
|
}
|
11792 |
|
|
}
|
11793 |
|
|
|
11794 |
|
|
/* Called via note_stores from record_dead_and_set_regs to handle one
|
11795 |
|
|
SET or CLOBBER in an insn. DATA is the instruction in which the
|
11796 |
|
|
set is occurring. */
|
11797 |
|
|
|
11798 |
|
|
static void
|
11799 |
|
|
record_dead_and_set_regs_1 (rtx dest, const_rtx setter, void *data)
|
11800 |
|
|
{
|
11801 |
|
|
rtx record_dead_insn = (rtx) data;
|
11802 |
|
|
|
11803 |
|
|
if (GET_CODE (dest) == SUBREG)
|
11804 |
|
|
dest = SUBREG_REG (dest);
|
11805 |
|
|
|
11806 |
|
|
if (!record_dead_insn)
|
11807 |
|
|
{
|
11808 |
|
|
if (REG_P (dest))
|
11809 |
|
|
record_value_for_reg (dest, NULL_RTX, NULL_RTX);
|
11810 |
|
|
return;
|
11811 |
|
|
}
|
11812 |
|
|
|
11813 |
|
|
if (REG_P (dest))
|
11814 |
|
|
{
|
11815 |
|
|
/* If we are setting the whole register, we know its value. Otherwise
|
11816 |
|
|
show that we don't know the value. We can handle SUBREG in
|
11817 |
|
|
some cases. */
|
11818 |
|
|
if (GET_CODE (setter) == SET && dest == SET_DEST (setter))
|
11819 |
|
|
record_value_for_reg (dest, record_dead_insn, SET_SRC (setter));
|
11820 |
|
|
else if (GET_CODE (setter) == SET
|
11821 |
|
|
&& GET_CODE (SET_DEST (setter)) == SUBREG
|
11822 |
|
|
&& SUBREG_REG (SET_DEST (setter)) == dest
|
11823 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (dest)) <= BITS_PER_WORD
|
11824 |
|
|
&& subreg_lowpart_p (SET_DEST (setter)))
|
11825 |
|
|
record_value_for_reg (dest, record_dead_insn,
|
11826 |
|
|
gen_lowpart (GET_MODE (dest),
|
11827 |
|
|
SET_SRC (setter)));
|
11828 |
|
|
else
|
11829 |
|
|
record_value_for_reg (dest, record_dead_insn, NULL_RTX);
|
11830 |
|
|
}
|
11831 |
|
|
else if (MEM_P (dest)
|
11832 |
|
|
/* Ignore pushes, they clobber nothing. */
|
11833 |
|
|
&& ! push_operand (dest, GET_MODE (dest)))
|
11834 |
|
|
mem_last_set = DF_INSN_LUID (record_dead_insn);
|
11835 |
|
|
}
|
11836 |
|
|
|
11837 |
|
|
/* Update the records of when each REG was most recently set or killed
|
11838 |
|
|
for the things done by INSN. This is the last thing done in processing
|
11839 |
|
|
INSN in the combiner loop.
|
11840 |
|
|
|
11841 |
|
|
We update reg_stat[], in particular fields last_set, last_set_value,
|
11842 |
|
|
last_set_mode, last_set_nonzero_bits, last_set_sign_bit_copies,
|
11843 |
|
|
last_death, and also the similar information mem_last_set (which insn
|
11844 |
|
|
most recently modified memory) and last_call_luid (which insn was the
|
11845 |
|
|
most recent subroutine call). */
|
11846 |
|
|
|
11847 |
|
|
static void
|
11848 |
|
|
record_dead_and_set_regs (rtx insn)
|
11849 |
|
|
{
|
11850 |
|
|
rtx link;
|
11851 |
|
|
unsigned int i;
|
11852 |
|
|
|
11853 |
|
|
for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
|
11854 |
|
|
{
|
11855 |
|
|
if (REG_NOTE_KIND (link) == REG_DEAD
|
11856 |
|
|
&& REG_P (XEXP (link, 0)))
|
11857 |
|
|
{
|
11858 |
|
|
unsigned int regno = REGNO (XEXP (link, 0));
|
11859 |
|
|
unsigned int endregno = END_REGNO (XEXP (link, 0));
|
11860 |
|
|
|
11861 |
|
|
for (i = regno; i < endregno; i++)
|
11862 |
|
|
{
|
11863 |
|
|
reg_stat_type *rsp;
|
11864 |
|
|
|
11865 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, i);
|
11866 |
|
|
rsp->last_death = insn;
|
11867 |
|
|
}
|
11868 |
|
|
}
|
11869 |
|
|
else if (REG_NOTE_KIND (link) == REG_INC)
|
11870 |
|
|
record_value_for_reg (XEXP (link, 0), insn, NULL_RTX);
|
11871 |
|
|
}
|
11872 |
|
|
|
11873 |
|
|
if (CALL_P (insn))
|
11874 |
|
|
{
|
11875 |
|
|
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
|
11876 |
|
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
|
11877 |
|
|
{
|
11878 |
|
|
reg_stat_type *rsp;
|
11879 |
|
|
|
11880 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, i);
|
11881 |
|
|
rsp->last_set_invalid = 1;
|
11882 |
|
|
rsp->last_set = insn;
|
11883 |
|
|
rsp->last_set_value = 0;
|
11884 |
|
|
rsp->last_set_mode = VOIDmode;
|
11885 |
|
|
rsp->last_set_nonzero_bits = 0;
|
11886 |
|
|
rsp->last_set_sign_bit_copies = 0;
|
11887 |
|
|
rsp->last_death = 0;
|
11888 |
|
|
rsp->truncated_to_mode = VOIDmode;
|
11889 |
|
|
}
|
11890 |
|
|
|
11891 |
|
|
last_call_luid = mem_last_set = DF_INSN_LUID (insn);
|
11892 |
|
|
|
11893 |
|
|
/* We can't combine into a call pattern. Remember, though, that
|
11894 |
|
|
the return value register is set at this LUID. We could
|
11895 |
|
|
still replace a register with the return value from the
|
11896 |
|
|
wrong subroutine call! */
|
11897 |
|
|
note_stores (PATTERN (insn), record_dead_and_set_regs_1, NULL_RTX);
|
11898 |
|
|
}
|
11899 |
|
|
else
|
11900 |
|
|
note_stores (PATTERN (insn), record_dead_and_set_regs_1, insn);
|
11901 |
|
|
}
|
11902 |
|
|
|
11903 |
|
|
/* If a SUBREG has the promoted bit set, it is in fact a property of the
|
11904 |
|
|
register present in the SUBREG, so for each such SUBREG go back and
|
11905 |
|
|
adjust nonzero and sign bit information of the registers that are
|
11906 |
|
|
known to have some zero/sign bits set.
|
11907 |
|
|
|
11908 |
|
|
This is needed because when combine blows the SUBREGs away, the
|
11909 |
|
|
information on zero/sign bits is lost and further combines can be
|
11910 |
|
|
missed because of that. */
|
11911 |
|
|
|
11912 |
|
|
static void
|
11913 |
|
|
record_promoted_value (rtx insn, rtx subreg)
|
11914 |
|
|
{
|
11915 |
|
|
rtx links, set;
|
11916 |
|
|
unsigned int regno = REGNO (SUBREG_REG (subreg));
|
11917 |
|
|
enum machine_mode mode = GET_MODE (subreg);
|
11918 |
|
|
|
11919 |
|
|
if (GET_MODE_BITSIZE (mode) > HOST_BITS_PER_WIDE_INT)
|
11920 |
|
|
return;
|
11921 |
|
|
|
11922 |
|
|
for (links = LOG_LINKS (insn); links;)
|
11923 |
|
|
{
|
11924 |
|
|
reg_stat_type *rsp;
|
11925 |
|
|
|
11926 |
|
|
insn = XEXP (links, 0);
|
11927 |
|
|
set = single_set (insn);
|
11928 |
|
|
|
11929 |
|
|
if (! set || !REG_P (SET_DEST (set))
|
11930 |
|
|
|| REGNO (SET_DEST (set)) != regno
|
11931 |
|
|
|| GET_MODE (SET_DEST (set)) != GET_MODE (SUBREG_REG (subreg)))
|
11932 |
|
|
{
|
11933 |
|
|
links = XEXP (links, 1);
|
11934 |
|
|
continue;
|
11935 |
|
|
}
|
11936 |
|
|
|
11937 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, regno);
|
11938 |
|
|
if (rsp->last_set == insn)
|
11939 |
|
|
{
|
11940 |
|
|
if (SUBREG_PROMOTED_UNSIGNED_P (subreg) > 0)
|
11941 |
|
|
rsp->last_set_nonzero_bits &= GET_MODE_MASK (mode);
|
11942 |
|
|
}
|
11943 |
|
|
|
11944 |
|
|
if (REG_P (SET_SRC (set)))
|
11945 |
|
|
{
|
11946 |
|
|
regno = REGNO (SET_SRC (set));
|
11947 |
|
|
links = LOG_LINKS (insn);
|
11948 |
|
|
}
|
11949 |
|
|
else
|
11950 |
|
|
break;
|
11951 |
|
|
}
|
11952 |
|
|
}
|
11953 |
|
|
|
11954 |
|
|
/* Check if X, a register, is known to contain a value already
|
11955 |
|
|
truncated to MODE. In this case we can use a subreg to refer to
|
11956 |
|
|
the truncated value even though in the generic case we would need
|
11957 |
|
|
an explicit truncation. */
|
11958 |
|
|
|
11959 |
|
|
static bool
|
11960 |
|
|
reg_truncated_to_mode (enum machine_mode mode, const_rtx x)
|
11961 |
|
|
{
|
11962 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
|
11963 |
|
|
enum machine_mode truncated = rsp->truncated_to_mode;
|
11964 |
|
|
|
11965 |
|
|
if (truncated == 0
|
11966 |
|
|
|| rsp->truncation_label < label_tick_ebb_start)
|
11967 |
|
|
return false;
|
11968 |
|
|
if (GET_MODE_SIZE (truncated) <= GET_MODE_SIZE (mode))
|
11969 |
|
|
return true;
|
11970 |
|
|
if (TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (mode),
|
11971 |
|
|
GET_MODE_BITSIZE (truncated)))
|
11972 |
|
|
return true;
|
11973 |
|
|
return false;
|
11974 |
|
|
}
|
11975 |
|
|
|
11976 |
|
|
/* Callback for for_each_rtx. If *P is a hard reg or a subreg record the mode
|
11977 |
|
|
that the register is accessed in. For non-TRULY_NOOP_TRUNCATION targets we
|
11978 |
|
|
might be able to turn a truncate into a subreg using this information.
|
11979 |
|
|
Return -1 if traversing *P is complete or 0 otherwise. */
|
11980 |
|
|
|
11981 |
|
|
static int
|
11982 |
|
|
record_truncated_value (rtx *p, void *data ATTRIBUTE_UNUSED)
|
11983 |
|
|
{
|
11984 |
|
|
rtx x = *p;
|
11985 |
|
|
enum machine_mode truncated_mode;
|
11986 |
|
|
reg_stat_type *rsp;
|
11987 |
|
|
|
11988 |
|
|
if (GET_CODE (x) == SUBREG && REG_P (SUBREG_REG (x)))
|
11989 |
|
|
{
|
11990 |
|
|
enum machine_mode original_mode = GET_MODE (SUBREG_REG (x));
|
11991 |
|
|
truncated_mode = GET_MODE (x);
|
11992 |
|
|
|
11993 |
|
|
if (GET_MODE_SIZE (original_mode) <= GET_MODE_SIZE (truncated_mode))
|
11994 |
|
|
return -1;
|
11995 |
|
|
|
11996 |
|
|
if (TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (truncated_mode),
|
11997 |
|
|
GET_MODE_BITSIZE (original_mode)))
|
11998 |
|
|
return -1;
|
11999 |
|
|
|
12000 |
|
|
x = SUBREG_REG (x);
|
12001 |
|
|
}
|
12002 |
|
|
/* ??? For hard-regs we now record everything. We might be able to
|
12003 |
|
|
optimize this using last_set_mode. */
|
12004 |
|
|
else if (REG_P (x) && REGNO (x) < FIRST_PSEUDO_REGISTER)
|
12005 |
|
|
truncated_mode = GET_MODE (x);
|
12006 |
|
|
else
|
12007 |
|
|
return 0;
|
12008 |
|
|
|
12009 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, REGNO (x));
|
12010 |
|
|
if (rsp->truncated_to_mode == 0
|
12011 |
|
|
|| rsp->truncation_label < label_tick_ebb_start
|
12012 |
|
|
|| (GET_MODE_SIZE (truncated_mode)
|
12013 |
|
|
< GET_MODE_SIZE (rsp->truncated_to_mode)))
|
12014 |
|
|
{
|
12015 |
|
|
rsp->truncated_to_mode = truncated_mode;
|
12016 |
|
|
rsp->truncation_label = label_tick;
|
12017 |
|
|
}
|
12018 |
|
|
|
12019 |
|
|
return -1;
|
12020 |
|
|
}
|
12021 |
|
|
|
12022 |
|
|
/* Callback for note_uses. Find hardregs and subregs of pseudos and
|
12023 |
|
|
the modes they are used in. This can help truning TRUNCATEs into
|
12024 |
|
|
SUBREGs. */
|
12025 |
|
|
|
12026 |
|
|
static void
|
12027 |
|
|
record_truncated_values (rtx *x, void *data ATTRIBUTE_UNUSED)
|
12028 |
|
|
{
|
12029 |
|
|
for_each_rtx (x, record_truncated_value, NULL);
|
12030 |
|
|
}
|
12031 |
|
|
|
12032 |
|
|
/* Scan X for promoted SUBREGs. For each one found,
|
12033 |
|
|
note what it implies to the registers used in it. */
|
12034 |
|
|
|
12035 |
|
|
static void
|
12036 |
|
|
check_promoted_subreg (rtx insn, rtx x)
|
12037 |
|
|
{
|
12038 |
|
|
if (GET_CODE (x) == SUBREG
|
12039 |
|
|
&& SUBREG_PROMOTED_VAR_P (x)
|
12040 |
|
|
&& REG_P (SUBREG_REG (x)))
|
12041 |
|
|
record_promoted_value (insn, x);
|
12042 |
|
|
else
|
12043 |
|
|
{
|
12044 |
|
|
const char *format = GET_RTX_FORMAT (GET_CODE (x));
|
12045 |
|
|
int i, j;
|
12046 |
|
|
|
12047 |
|
|
for (i = 0; i < GET_RTX_LENGTH (GET_CODE (x)); i++)
|
12048 |
|
|
switch (format[i])
|
12049 |
|
|
{
|
12050 |
|
|
case 'e':
|
12051 |
|
|
check_promoted_subreg (insn, XEXP (x, i));
|
12052 |
|
|
break;
|
12053 |
|
|
case 'V':
|
12054 |
|
|
case 'E':
|
12055 |
|
|
if (XVEC (x, i) != 0)
|
12056 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
12057 |
|
|
check_promoted_subreg (insn, XVECEXP (x, i, j));
|
12058 |
|
|
break;
|
12059 |
|
|
}
|
12060 |
|
|
}
|
12061 |
|
|
}
|
12062 |
|
|
|
12063 |
|
|
/* Verify that all the registers and memory references mentioned in *LOC are
|
12064 |
|
|
still valid. *LOC was part of a value set in INSN when label_tick was
|
12065 |
|
|
equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
|
12066 |
|
|
the invalid references with (clobber (const_int 0)) and return 1. This
|
12067 |
|
|
replacement is useful because we often can get useful information about
|
12068 |
|
|
the form of a value (e.g., if it was produced by a shift that always
|
12069 |
|
|
produces -1 or 0) even though we don't know exactly what registers it
|
12070 |
|
|
was produced from. */
|
12071 |
|
|
|
12072 |
|
|
static int
|
12073 |
|
|
get_last_value_validate (rtx *loc, rtx insn, int tick, int replace)
|
12074 |
|
|
{
|
12075 |
|
|
rtx x = *loc;
|
12076 |
|
|
const char *fmt = GET_RTX_FORMAT (GET_CODE (x));
|
12077 |
|
|
int len = GET_RTX_LENGTH (GET_CODE (x));
|
12078 |
|
|
int i, j;
|
12079 |
|
|
|
12080 |
|
|
if (REG_P (x))
|
12081 |
|
|
{
|
12082 |
|
|
unsigned int regno = REGNO (x);
|
12083 |
|
|
unsigned int endregno = END_REGNO (x);
|
12084 |
|
|
unsigned int j;
|
12085 |
|
|
|
12086 |
|
|
for (j = regno; j < endregno; j++)
|
12087 |
|
|
{
|
12088 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, j);
|
12089 |
|
|
if (rsp->last_set_invalid
|
12090 |
|
|
/* If this is a pseudo-register that was only set once and not
|
12091 |
|
|
live at the beginning of the function, it is always valid. */
|
12092 |
|
|
|| (! (regno >= FIRST_PSEUDO_REGISTER
|
12093 |
|
|
&& REG_N_SETS (regno) == 1
|
12094 |
|
|
&& (!REGNO_REG_SET_P
|
12095 |
|
|
(DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), regno)))
|
12096 |
|
|
&& rsp->last_set_label > tick))
|
12097 |
|
|
{
|
12098 |
|
|
if (replace)
|
12099 |
|
|
*loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
|
12100 |
|
|
return replace;
|
12101 |
|
|
}
|
12102 |
|
|
}
|
12103 |
|
|
|
12104 |
|
|
return 1;
|
12105 |
|
|
}
|
12106 |
|
|
/* If this is a memory reference, make sure that there were no stores after
|
12107 |
|
|
it that might have clobbered the value. We don't have alias info, so we
|
12108 |
|
|
assume any store invalidates it. Moreover, we only have local UIDs, so
|
12109 |
|
|
we also assume that there were stores in the intervening basic blocks. */
|
12110 |
|
|
else if (MEM_P (x) && !MEM_READONLY_P (x)
|
12111 |
|
|
&& (tick != label_tick || DF_INSN_LUID (insn) <= mem_last_set))
|
12112 |
|
|
{
|
12113 |
|
|
if (replace)
|
12114 |
|
|
*loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
|
12115 |
|
|
return replace;
|
12116 |
|
|
}
|
12117 |
|
|
|
12118 |
|
|
for (i = 0; i < len; i++)
|
12119 |
|
|
{
|
12120 |
|
|
if (fmt[i] == 'e')
|
12121 |
|
|
{
|
12122 |
|
|
/* Check for identical subexpressions. If x contains
|
12123 |
|
|
identical subexpression we only have to traverse one of
|
12124 |
|
|
them. */
|
12125 |
|
|
if (i == 1 && ARITHMETIC_P (x))
|
12126 |
|
|
{
|
12127 |
|
|
/* Note that at this point x0 has already been checked
|
12128 |
|
|
and found valid. */
|
12129 |
|
|
rtx x0 = XEXP (x, 0);
|
12130 |
|
|
rtx x1 = XEXP (x, 1);
|
12131 |
|
|
|
12132 |
|
|
/* If x0 and x1 are identical then x is also valid. */
|
12133 |
|
|
if (x0 == x1)
|
12134 |
|
|
return 1;
|
12135 |
|
|
|
12136 |
|
|
/* If x1 is identical to a subexpression of x0 then
|
12137 |
|
|
while checking x0, x1 has already been checked. Thus
|
12138 |
|
|
it is valid and so as x. */
|
12139 |
|
|
if (ARITHMETIC_P (x0)
|
12140 |
|
|
&& (x1 == XEXP (x0, 0) || x1 == XEXP (x0, 1)))
|
12141 |
|
|
return 1;
|
12142 |
|
|
|
12143 |
|
|
/* If x0 is identical to a subexpression of x1 then x is
|
12144 |
|
|
valid iff the rest of x1 is valid. */
|
12145 |
|
|
if (ARITHMETIC_P (x1)
|
12146 |
|
|
&& (x0 == XEXP (x1, 0) || x0 == XEXP (x1, 1)))
|
12147 |
|
|
return
|
12148 |
|
|
get_last_value_validate (&XEXP (x1,
|
12149 |
|
|
x0 == XEXP (x1, 0) ? 1 : 0),
|
12150 |
|
|
insn, tick, replace);
|
12151 |
|
|
}
|
12152 |
|
|
|
12153 |
|
|
if (get_last_value_validate (&XEXP (x, i), insn, tick,
|
12154 |
|
|
replace) == 0)
|
12155 |
|
|
return 0;
|
12156 |
|
|
}
|
12157 |
|
|
else if (fmt[i] == 'E')
|
12158 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
12159 |
|
|
if (get_last_value_validate (&XVECEXP (x, i, j),
|
12160 |
|
|
insn, tick, replace) == 0)
|
12161 |
|
|
return 0;
|
12162 |
|
|
}
|
12163 |
|
|
|
12164 |
|
|
/* If we haven't found a reason for it to be invalid, it is valid. */
|
12165 |
|
|
return 1;
|
12166 |
|
|
}
|
12167 |
|
|
|
12168 |
|
|
/* Get the last value assigned to X, if known. Some registers
|
12169 |
|
|
in the value may be replaced with (clobber (const_int 0)) if their value
|
12170 |
|
|
is known longer known reliably. */
|
12171 |
|
|
|
12172 |
|
|
static rtx
|
12173 |
|
|
get_last_value (const_rtx x)
|
12174 |
|
|
{
|
12175 |
|
|
unsigned int regno;
|
12176 |
|
|
rtx value;
|
12177 |
|
|
reg_stat_type *rsp;
|
12178 |
|
|
|
12179 |
|
|
/* If this is a non-paradoxical SUBREG, get the value of its operand and
|
12180 |
|
|
then convert it to the desired mode. If this is a paradoxical SUBREG,
|
12181 |
|
|
we cannot predict what values the "extra" bits might have. */
|
12182 |
|
|
if (GET_CODE (x) == SUBREG
|
12183 |
|
|
&& subreg_lowpart_p (x)
|
12184 |
|
|
&& (GET_MODE_SIZE (GET_MODE (x))
|
12185 |
|
|
<= GET_MODE_SIZE (GET_MODE (SUBREG_REG (x))))
|
12186 |
|
|
&& (value = get_last_value (SUBREG_REG (x))) != 0)
|
12187 |
|
|
return gen_lowpart (GET_MODE (x), value);
|
12188 |
|
|
|
12189 |
|
|
if (!REG_P (x))
|
12190 |
|
|
return 0;
|
12191 |
|
|
|
12192 |
|
|
regno = REGNO (x);
|
12193 |
|
|
rsp = VEC_index (reg_stat_type, reg_stat, regno);
|
12194 |
|
|
value = rsp->last_set_value;
|
12195 |
|
|
|
12196 |
|
|
/* If we don't have a value, or if it isn't for this basic block and
|
12197 |
|
|
it's either a hard register, set more than once, or it's a live
|
12198 |
|
|
at the beginning of the function, return 0.
|
12199 |
|
|
|
12200 |
|
|
Because if it's not live at the beginning of the function then the reg
|
12201 |
|
|
is always set before being used (is never used without being set).
|
12202 |
|
|
And, if it's set only once, and it's always set before use, then all
|
12203 |
|
|
uses must have the same last value, even if it's not from this basic
|
12204 |
|
|
block. */
|
12205 |
|
|
|
12206 |
|
|
if (value == 0
|
12207 |
|
|
|| (rsp->last_set_label < label_tick_ebb_start
|
12208 |
|
|
&& (regno < FIRST_PSEUDO_REGISTER
|
12209 |
|
|
|| REG_N_SETS (regno) != 1
|
12210 |
|
|
|| REGNO_REG_SET_P
|
12211 |
|
|
(DF_LR_IN (ENTRY_BLOCK_PTR->next_bb), regno))))
|
12212 |
|
|
return 0;
|
12213 |
|
|
|
12214 |
|
|
/* If the value was set in a later insn than the ones we are processing,
|
12215 |
|
|
we can't use it even if the register was only set once. */
|
12216 |
|
|
if (rsp->last_set_label == label_tick
|
12217 |
|
|
&& DF_INSN_LUID (rsp->last_set) >= subst_low_luid)
|
12218 |
|
|
return 0;
|
12219 |
|
|
|
12220 |
|
|
/* If the value has all its registers valid, return it. */
|
12221 |
|
|
if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 0))
|
12222 |
|
|
return value;
|
12223 |
|
|
|
12224 |
|
|
/* Otherwise, make a copy and replace any invalid register with
|
12225 |
|
|
(clobber (const_int 0)). If that fails for some reason, return 0. */
|
12226 |
|
|
|
12227 |
|
|
value = copy_rtx (value);
|
12228 |
|
|
if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 1))
|
12229 |
|
|
return value;
|
12230 |
|
|
|
12231 |
|
|
return 0;
|
12232 |
|
|
}
|
12233 |
|
|
|
12234 |
|
|
/* Return nonzero if expression X refers to a REG or to memory
|
12235 |
|
|
that is set in an instruction more recent than FROM_LUID. */
|
12236 |
|
|
|
12237 |
|
|
static int
|
12238 |
|
|
use_crosses_set_p (const_rtx x, int from_luid)
|
12239 |
|
|
{
|
12240 |
|
|
const char *fmt;
|
12241 |
|
|
int i;
|
12242 |
|
|
enum rtx_code code = GET_CODE (x);
|
12243 |
|
|
|
12244 |
|
|
if (code == REG)
|
12245 |
|
|
{
|
12246 |
|
|
unsigned int regno = REGNO (x);
|
12247 |
|
|
unsigned endreg = END_REGNO (x);
|
12248 |
|
|
|
12249 |
|
|
#ifdef PUSH_ROUNDING
|
12250 |
|
|
/* Don't allow uses of the stack pointer to be moved,
|
12251 |
|
|
because we don't know whether the move crosses a push insn. */
|
12252 |
|
|
if (regno == STACK_POINTER_REGNUM && PUSH_ARGS)
|
12253 |
|
|
return 1;
|
12254 |
|
|
#endif
|
12255 |
|
|
for (; regno < endreg; regno++)
|
12256 |
|
|
{
|
12257 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, regno);
|
12258 |
|
|
if (rsp->last_set
|
12259 |
|
|
&& rsp->last_set_label == label_tick
|
12260 |
|
|
&& DF_INSN_LUID (rsp->last_set) > from_luid)
|
12261 |
|
|
return 1;
|
12262 |
|
|
}
|
12263 |
|
|
return 0;
|
12264 |
|
|
}
|
12265 |
|
|
|
12266 |
|
|
if (code == MEM && mem_last_set > from_luid)
|
12267 |
|
|
return 1;
|
12268 |
|
|
|
12269 |
|
|
fmt = GET_RTX_FORMAT (code);
|
12270 |
|
|
|
12271 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
12272 |
|
|
{
|
12273 |
|
|
if (fmt[i] == 'E')
|
12274 |
|
|
{
|
12275 |
|
|
int j;
|
12276 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
12277 |
|
|
if (use_crosses_set_p (XVECEXP (x, i, j), from_luid))
|
12278 |
|
|
return 1;
|
12279 |
|
|
}
|
12280 |
|
|
else if (fmt[i] == 'e'
|
12281 |
|
|
&& use_crosses_set_p (XEXP (x, i), from_luid))
|
12282 |
|
|
return 1;
|
12283 |
|
|
}
|
12284 |
|
|
return 0;
|
12285 |
|
|
}
|
12286 |
|
|
|
12287 |
|
|
/* Define three variables used for communication between the following
|
12288 |
|
|
routines. */
|
12289 |
|
|
|
12290 |
|
|
static unsigned int reg_dead_regno, reg_dead_endregno;
|
12291 |
|
|
static int reg_dead_flag;
|
12292 |
|
|
|
12293 |
|
|
/* Function called via note_stores from reg_dead_at_p.
|
12294 |
|
|
|
12295 |
|
|
If DEST is within [reg_dead_regno, reg_dead_endregno), set
|
12296 |
|
|
reg_dead_flag to 1 if X is a CLOBBER and to -1 it is a SET. */
|
12297 |
|
|
|
12298 |
|
|
static void
|
12299 |
|
|
reg_dead_at_p_1 (rtx dest, const_rtx x, void *data ATTRIBUTE_UNUSED)
|
12300 |
|
|
{
|
12301 |
|
|
unsigned int regno, endregno;
|
12302 |
|
|
|
12303 |
|
|
if (!REG_P (dest))
|
12304 |
|
|
return;
|
12305 |
|
|
|
12306 |
|
|
regno = REGNO (dest);
|
12307 |
|
|
endregno = END_REGNO (dest);
|
12308 |
|
|
if (reg_dead_endregno > regno && reg_dead_regno < endregno)
|
12309 |
|
|
reg_dead_flag = (GET_CODE (x) == CLOBBER) ? 1 : -1;
|
12310 |
|
|
}
|
12311 |
|
|
|
12312 |
|
|
/* Return nonzero if REG is known to be dead at INSN.
|
12313 |
|
|
|
12314 |
|
|
We scan backwards from INSN. If we hit a REG_DEAD note or a CLOBBER
|
12315 |
|
|
referencing REG, it is dead. If we hit a SET referencing REG, it is
|
12316 |
|
|
live. Otherwise, see if it is live or dead at the start of the basic
|
12317 |
|
|
block we are in. Hard regs marked as being live in NEWPAT_USED_REGS
|
12318 |
|
|
must be assumed to be always live. */
|
12319 |
|
|
|
12320 |
|
|
static int
|
12321 |
|
|
reg_dead_at_p (rtx reg, rtx insn)
|
12322 |
|
|
{
|
12323 |
|
|
basic_block block;
|
12324 |
|
|
unsigned int i;
|
12325 |
|
|
|
12326 |
|
|
/* Set variables for reg_dead_at_p_1. */
|
12327 |
|
|
reg_dead_regno = REGNO (reg);
|
12328 |
|
|
reg_dead_endregno = END_REGNO (reg);
|
12329 |
|
|
|
12330 |
|
|
reg_dead_flag = 0;
|
12331 |
|
|
|
12332 |
|
|
/* Check that reg isn't mentioned in NEWPAT_USED_REGS. For fixed registers
|
12333 |
|
|
we allow the machine description to decide whether use-and-clobber
|
12334 |
|
|
patterns are OK. */
|
12335 |
|
|
if (reg_dead_regno < FIRST_PSEUDO_REGISTER)
|
12336 |
|
|
{
|
12337 |
|
|
for (i = reg_dead_regno; i < reg_dead_endregno; i++)
|
12338 |
|
|
if (!fixed_regs[i] && TEST_HARD_REG_BIT (newpat_used_regs, i))
|
12339 |
|
|
return 0;
|
12340 |
|
|
}
|
12341 |
|
|
|
12342 |
|
|
/* Scan backwards until we find a REG_DEAD note, SET, CLOBBER, or
|
12343 |
|
|
beginning of basic block. */
|
12344 |
|
|
block = BLOCK_FOR_INSN (insn);
|
12345 |
|
|
for (;;)
|
12346 |
|
|
{
|
12347 |
|
|
if (INSN_P (insn))
|
12348 |
|
|
{
|
12349 |
|
|
note_stores (PATTERN (insn), reg_dead_at_p_1, NULL);
|
12350 |
|
|
if (reg_dead_flag)
|
12351 |
|
|
return reg_dead_flag == 1 ? 1 : 0;
|
12352 |
|
|
|
12353 |
|
|
if (find_regno_note (insn, REG_DEAD, reg_dead_regno))
|
12354 |
|
|
return 1;
|
12355 |
|
|
}
|
12356 |
|
|
|
12357 |
|
|
if (insn == BB_HEAD (block))
|
12358 |
|
|
break;
|
12359 |
|
|
|
12360 |
|
|
insn = PREV_INSN (insn);
|
12361 |
|
|
}
|
12362 |
|
|
|
12363 |
|
|
/* Look at live-in sets for the basic block that we were in. */
|
12364 |
|
|
for (i = reg_dead_regno; i < reg_dead_endregno; i++)
|
12365 |
|
|
if (REGNO_REG_SET_P (df_get_live_in (block), i))
|
12366 |
|
|
return 0;
|
12367 |
|
|
|
12368 |
|
|
return 1;
|
12369 |
|
|
}
|
12370 |
|
|
|
12371 |
|
|
/* Note hard registers in X that are used. */
|
12372 |
|
|
|
12373 |
|
|
static void
|
12374 |
|
|
mark_used_regs_combine (rtx x)
|
12375 |
|
|
{
|
12376 |
|
|
RTX_CODE code = GET_CODE (x);
|
12377 |
|
|
unsigned int regno;
|
12378 |
|
|
int i;
|
12379 |
|
|
|
12380 |
|
|
switch (code)
|
12381 |
|
|
{
|
12382 |
|
|
case LABEL_REF:
|
12383 |
|
|
case SYMBOL_REF:
|
12384 |
|
|
case CONST_INT:
|
12385 |
|
|
case CONST:
|
12386 |
|
|
case CONST_DOUBLE:
|
12387 |
|
|
case CONST_VECTOR:
|
12388 |
|
|
case PC:
|
12389 |
|
|
case ADDR_VEC:
|
12390 |
|
|
case ADDR_DIFF_VEC:
|
12391 |
|
|
case ASM_INPUT:
|
12392 |
|
|
#ifdef HAVE_cc0
|
12393 |
|
|
/* CC0 must die in the insn after it is set, so we don't need to take
|
12394 |
|
|
special note of it here. */
|
12395 |
|
|
case CC0:
|
12396 |
|
|
#endif
|
12397 |
|
|
return;
|
12398 |
|
|
|
12399 |
|
|
case CLOBBER:
|
12400 |
|
|
/* If we are clobbering a MEM, mark any hard registers inside the
|
12401 |
|
|
address as used. */
|
12402 |
|
|
if (MEM_P (XEXP (x, 0)))
|
12403 |
|
|
mark_used_regs_combine (XEXP (XEXP (x, 0), 0));
|
12404 |
|
|
return;
|
12405 |
|
|
|
12406 |
|
|
case REG:
|
12407 |
|
|
regno = REGNO (x);
|
12408 |
|
|
/* A hard reg in a wide mode may really be multiple registers.
|
12409 |
|
|
If so, mark all of them just like the first. */
|
12410 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
12411 |
|
|
{
|
12412 |
|
|
/* None of this applies to the stack, frame or arg pointers. */
|
12413 |
|
|
if (regno == STACK_POINTER_REGNUM
|
12414 |
|
|
#if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
|
12415 |
|
|
|| regno == HARD_FRAME_POINTER_REGNUM
|
12416 |
|
|
#endif
|
12417 |
|
|
#if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
|
12418 |
|
|
|| (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
|
12419 |
|
|
#endif
|
12420 |
|
|
|| regno == FRAME_POINTER_REGNUM)
|
12421 |
|
|
return;
|
12422 |
|
|
|
12423 |
|
|
add_to_hard_reg_set (&newpat_used_regs, GET_MODE (x), regno);
|
12424 |
|
|
}
|
12425 |
|
|
return;
|
12426 |
|
|
|
12427 |
|
|
case SET:
|
12428 |
|
|
{
|
12429 |
|
|
/* If setting a MEM, or a SUBREG of a MEM, then note any hard regs in
|
12430 |
|
|
the address. */
|
12431 |
|
|
rtx testreg = SET_DEST (x);
|
12432 |
|
|
|
12433 |
|
|
while (GET_CODE (testreg) == SUBREG
|
12434 |
|
|
|| GET_CODE (testreg) == ZERO_EXTRACT
|
12435 |
|
|
|| GET_CODE (testreg) == STRICT_LOW_PART)
|
12436 |
|
|
testreg = XEXP (testreg, 0);
|
12437 |
|
|
|
12438 |
|
|
if (MEM_P (testreg))
|
12439 |
|
|
mark_used_regs_combine (XEXP (testreg, 0));
|
12440 |
|
|
|
12441 |
|
|
mark_used_regs_combine (SET_SRC (x));
|
12442 |
|
|
}
|
12443 |
|
|
return;
|
12444 |
|
|
|
12445 |
|
|
default:
|
12446 |
|
|
break;
|
12447 |
|
|
}
|
12448 |
|
|
|
12449 |
|
|
/* Recursively scan the operands of this expression. */
|
12450 |
|
|
|
12451 |
|
|
{
|
12452 |
|
|
const char *fmt = GET_RTX_FORMAT (code);
|
12453 |
|
|
|
12454 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
12455 |
|
|
{
|
12456 |
|
|
if (fmt[i] == 'e')
|
12457 |
|
|
mark_used_regs_combine (XEXP (x, i));
|
12458 |
|
|
else if (fmt[i] == 'E')
|
12459 |
|
|
{
|
12460 |
|
|
int j;
|
12461 |
|
|
|
12462 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
12463 |
|
|
mark_used_regs_combine (XVECEXP (x, i, j));
|
12464 |
|
|
}
|
12465 |
|
|
}
|
12466 |
|
|
}
|
12467 |
|
|
}
|
12468 |
|
|
|
12469 |
|
|
/* Remove register number REGNO from the dead registers list of INSN.
|
12470 |
|
|
|
12471 |
|
|
Return the note used to record the death, if there was one. */
|
12472 |
|
|
|
12473 |
|
|
rtx
|
12474 |
|
|
remove_death (unsigned int regno, rtx insn)
|
12475 |
|
|
{
|
12476 |
|
|
rtx note = find_regno_note (insn, REG_DEAD, regno);
|
12477 |
|
|
|
12478 |
|
|
if (note)
|
12479 |
|
|
remove_note (insn, note);
|
12480 |
|
|
|
12481 |
|
|
return note;
|
12482 |
|
|
}
|
12483 |
|
|
|
12484 |
|
|
/* For each register (hardware or pseudo) used within expression X, if its
|
12485 |
|
|
death is in an instruction with luid between FROM_LUID (inclusive) and
|
12486 |
|
|
TO_INSN (exclusive), put a REG_DEAD note for that register in the
|
12487 |
|
|
list headed by PNOTES.
|
12488 |
|
|
|
12489 |
|
|
That said, don't move registers killed by maybe_kill_insn.
|
12490 |
|
|
|
12491 |
|
|
This is done when X is being merged by combination into TO_INSN. These
|
12492 |
|
|
notes will then be distributed as needed. */
|
12493 |
|
|
|
12494 |
|
|
static void
|
12495 |
|
|
move_deaths (rtx x, rtx maybe_kill_insn, int from_luid, rtx to_insn,
|
12496 |
|
|
rtx *pnotes)
|
12497 |
|
|
{
|
12498 |
|
|
const char *fmt;
|
12499 |
|
|
int len, i;
|
12500 |
|
|
enum rtx_code code = GET_CODE (x);
|
12501 |
|
|
|
12502 |
|
|
if (code == REG)
|
12503 |
|
|
{
|
12504 |
|
|
unsigned int regno = REGNO (x);
|
12505 |
|
|
rtx where_dead = VEC_index (reg_stat_type, reg_stat, regno)->last_death;
|
12506 |
|
|
|
12507 |
|
|
/* Don't move the register if it gets killed in between from and to. */
|
12508 |
|
|
if (maybe_kill_insn && reg_set_p (x, maybe_kill_insn)
|
12509 |
|
|
&& ! reg_referenced_p (x, maybe_kill_insn))
|
12510 |
|
|
return;
|
12511 |
|
|
|
12512 |
|
|
if (where_dead
|
12513 |
|
|
&& BLOCK_FOR_INSN (where_dead) == BLOCK_FOR_INSN (to_insn)
|
12514 |
|
|
&& DF_INSN_LUID (where_dead) >= from_luid
|
12515 |
|
|
&& DF_INSN_LUID (where_dead) < DF_INSN_LUID (to_insn))
|
12516 |
|
|
{
|
12517 |
|
|
rtx note = remove_death (regno, where_dead);
|
12518 |
|
|
|
12519 |
|
|
/* It is possible for the call above to return 0. This can occur
|
12520 |
|
|
when last_death points to I2 or I1 that we combined with.
|
12521 |
|
|
In that case make a new note.
|
12522 |
|
|
|
12523 |
|
|
We must also check for the case where X is a hard register
|
12524 |
|
|
and NOTE is a death note for a range of hard registers
|
12525 |
|
|
including X. In that case, we must put REG_DEAD notes for
|
12526 |
|
|
the remaining registers in place of NOTE. */
|
12527 |
|
|
|
12528 |
|
|
if (note != 0 && regno < FIRST_PSEUDO_REGISTER
|
12529 |
|
|
&& (GET_MODE_SIZE (GET_MODE (XEXP (note, 0)))
|
12530 |
|
|
> GET_MODE_SIZE (GET_MODE (x))))
|
12531 |
|
|
{
|
12532 |
|
|
unsigned int deadregno = REGNO (XEXP (note, 0));
|
12533 |
|
|
unsigned int deadend = END_HARD_REGNO (XEXP (note, 0));
|
12534 |
|
|
unsigned int ourend = END_HARD_REGNO (x);
|
12535 |
|
|
unsigned int i;
|
12536 |
|
|
|
12537 |
|
|
for (i = deadregno; i < deadend; i++)
|
12538 |
|
|
if (i < regno || i >= ourend)
|
12539 |
|
|
add_reg_note (where_dead, REG_DEAD, regno_reg_rtx[i]);
|
12540 |
|
|
}
|
12541 |
|
|
|
12542 |
|
|
/* If we didn't find any note, or if we found a REG_DEAD note that
|
12543 |
|
|
covers only part of the given reg, and we have a multi-reg hard
|
12544 |
|
|
register, then to be safe we must check for REG_DEAD notes
|
12545 |
|
|
for each register other than the first. They could have
|
12546 |
|
|
their own REG_DEAD notes lying around. */
|
12547 |
|
|
else if ((note == 0
|
12548 |
|
|
|| (note != 0
|
12549 |
|
|
&& (GET_MODE_SIZE (GET_MODE (XEXP (note, 0)))
|
12550 |
|
|
< GET_MODE_SIZE (GET_MODE (x)))))
|
12551 |
|
|
&& regno < FIRST_PSEUDO_REGISTER
|
12552 |
|
|
&& hard_regno_nregs[regno][GET_MODE (x)] > 1)
|
12553 |
|
|
{
|
12554 |
|
|
unsigned int ourend = END_HARD_REGNO (x);
|
12555 |
|
|
unsigned int i, offset;
|
12556 |
|
|
rtx oldnotes = 0;
|
12557 |
|
|
|
12558 |
|
|
if (note)
|
12559 |
|
|
offset = hard_regno_nregs[regno][GET_MODE (XEXP (note, 0))];
|
12560 |
|
|
else
|
12561 |
|
|
offset = 1;
|
12562 |
|
|
|
12563 |
|
|
for (i = regno + offset; i < ourend; i++)
|
12564 |
|
|
move_deaths (regno_reg_rtx[i],
|
12565 |
|
|
maybe_kill_insn, from_luid, to_insn, &oldnotes);
|
12566 |
|
|
}
|
12567 |
|
|
|
12568 |
|
|
if (note != 0 && GET_MODE (XEXP (note, 0)) == GET_MODE (x))
|
12569 |
|
|
{
|
12570 |
|
|
XEXP (note, 1) = *pnotes;
|
12571 |
|
|
*pnotes = note;
|
12572 |
|
|
}
|
12573 |
|
|
else
|
12574 |
|
|
*pnotes = alloc_reg_note (REG_DEAD, x, *pnotes);
|
12575 |
|
|
}
|
12576 |
|
|
|
12577 |
|
|
return;
|
12578 |
|
|
}
|
12579 |
|
|
|
12580 |
|
|
else if (GET_CODE (x) == SET)
|
12581 |
|
|
{
|
12582 |
|
|
rtx dest = SET_DEST (x);
|
12583 |
|
|
|
12584 |
|
|
move_deaths (SET_SRC (x), maybe_kill_insn, from_luid, to_insn, pnotes);
|
12585 |
|
|
|
12586 |
|
|
/* In the case of a ZERO_EXTRACT, a STRICT_LOW_PART, or a SUBREG
|
12587 |
|
|
that accesses one word of a multi-word item, some
|
12588 |
|
|
piece of everything register in the expression is used by
|
12589 |
|
|
this insn, so remove any old death. */
|
12590 |
|
|
/* ??? So why do we test for equality of the sizes? */
|
12591 |
|
|
|
12592 |
|
|
if (GET_CODE (dest) == ZERO_EXTRACT
|
12593 |
|
|
|| GET_CODE (dest) == STRICT_LOW_PART
|
12594 |
|
|
|| (GET_CODE (dest) == SUBREG
|
12595 |
|
|
&& (((GET_MODE_SIZE (GET_MODE (dest))
|
12596 |
|
|
+ UNITS_PER_WORD - 1) / UNITS_PER_WORD)
|
12597 |
|
|
== ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest)))
|
12598 |
|
|
+ UNITS_PER_WORD - 1) / UNITS_PER_WORD))))
|
12599 |
|
|
{
|
12600 |
|
|
move_deaths (dest, maybe_kill_insn, from_luid, to_insn, pnotes);
|
12601 |
|
|
return;
|
12602 |
|
|
}
|
12603 |
|
|
|
12604 |
|
|
/* If this is some other SUBREG, we know it replaces the entire
|
12605 |
|
|
value, so use that as the destination. */
|
12606 |
|
|
if (GET_CODE (dest) == SUBREG)
|
12607 |
|
|
dest = SUBREG_REG (dest);
|
12608 |
|
|
|
12609 |
|
|
/* If this is a MEM, adjust deaths of anything used in the address.
|
12610 |
|
|
For a REG (the only other possibility), the entire value is
|
12611 |
|
|
being replaced so the old value is not used in this insn. */
|
12612 |
|
|
|
12613 |
|
|
if (MEM_P (dest))
|
12614 |
|
|
move_deaths (XEXP (dest, 0), maybe_kill_insn, from_luid,
|
12615 |
|
|
to_insn, pnotes);
|
12616 |
|
|
return;
|
12617 |
|
|
}
|
12618 |
|
|
|
12619 |
|
|
else if (GET_CODE (x) == CLOBBER)
|
12620 |
|
|
return;
|
12621 |
|
|
|
12622 |
|
|
len = GET_RTX_LENGTH (code);
|
12623 |
|
|
fmt = GET_RTX_FORMAT (code);
|
12624 |
|
|
|
12625 |
|
|
for (i = 0; i < len; i++)
|
12626 |
|
|
{
|
12627 |
|
|
if (fmt[i] == 'E')
|
12628 |
|
|
{
|
12629 |
|
|
int j;
|
12630 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
12631 |
|
|
move_deaths (XVECEXP (x, i, j), maybe_kill_insn, from_luid,
|
12632 |
|
|
to_insn, pnotes);
|
12633 |
|
|
}
|
12634 |
|
|
else if (fmt[i] == 'e')
|
12635 |
|
|
move_deaths (XEXP (x, i), maybe_kill_insn, from_luid, to_insn, pnotes);
|
12636 |
|
|
}
|
12637 |
|
|
}
|
12638 |
|
|
|
12639 |
|
|
/* Return 1 if X is the target of a bit-field assignment in BODY, the
|
12640 |
|
|
pattern of an insn. X must be a REG. */
|
12641 |
|
|
|
12642 |
|
|
static int
|
12643 |
|
|
reg_bitfield_target_p (rtx x, rtx body)
|
12644 |
|
|
{
|
12645 |
|
|
int i;
|
12646 |
|
|
|
12647 |
|
|
if (GET_CODE (body) == SET)
|
12648 |
|
|
{
|
12649 |
|
|
rtx dest = SET_DEST (body);
|
12650 |
|
|
rtx target;
|
12651 |
|
|
unsigned int regno, tregno, endregno, endtregno;
|
12652 |
|
|
|
12653 |
|
|
if (GET_CODE (dest) == ZERO_EXTRACT)
|
12654 |
|
|
target = XEXP (dest, 0);
|
12655 |
|
|
else if (GET_CODE (dest) == STRICT_LOW_PART)
|
12656 |
|
|
target = SUBREG_REG (XEXP (dest, 0));
|
12657 |
|
|
else
|
12658 |
|
|
return 0;
|
12659 |
|
|
|
12660 |
|
|
if (GET_CODE (target) == SUBREG)
|
12661 |
|
|
target = SUBREG_REG (target);
|
12662 |
|
|
|
12663 |
|
|
if (!REG_P (target))
|
12664 |
|
|
return 0;
|
12665 |
|
|
|
12666 |
|
|
tregno = REGNO (target), regno = REGNO (x);
|
12667 |
|
|
if (tregno >= FIRST_PSEUDO_REGISTER || regno >= FIRST_PSEUDO_REGISTER)
|
12668 |
|
|
return target == x;
|
12669 |
|
|
|
12670 |
|
|
endtregno = end_hard_regno (GET_MODE (target), tregno);
|
12671 |
|
|
endregno = end_hard_regno (GET_MODE (x), regno);
|
12672 |
|
|
|
12673 |
|
|
return endregno > tregno && regno < endtregno;
|
12674 |
|
|
}
|
12675 |
|
|
|
12676 |
|
|
else if (GET_CODE (body) == PARALLEL)
|
12677 |
|
|
for (i = XVECLEN (body, 0) - 1; i >= 0; i--)
|
12678 |
|
|
if (reg_bitfield_target_p (x, XVECEXP (body, 0, i)))
|
12679 |
|
|
return 1;
|
12680 |
|
|
|
12681 |
|
|
return 0;
|
12682 |
|
|
}
|
12683 |
|
|
|
12684 |
|
|
/* Return the next insn after INSN that is neither a NOTE nor a
|
12685 |
|
|
DEBUG_INSN. This routine does not look inside SEQUENCEs. */
|
12686 |
|
|
|
12687 |
|
|
static rtx
|
12688 |
|
|
next_nonnote_nondebug_insn (rtx insn)
|
12689 |
|
|
{
|
12690 |
|
|
while (insn)
|
12691 |
|
|
{
|
12692 |
|
|
insn = NEXT_INSN (insn);
|
12693 |
|
|
if (insn == 0)
|
12694 |
|
|
break;
|
12695 |
|
|
if (NOTE_P (insn))
|
12696 |
|
|
continue;
|
12697 |
|
|
if (DEBUG_INSN_P (insn))
|
12698 |
|
|
continue;
|
12699 |
|
|
break;
|
12700 |
|
|
}
|
12701 |
|
|
|
12702 |
|
|
return insn;
|
12703 |
|
|
}
|
12704 |
|
|
|
12705 |
|
|
|
12706 |
|
|
|
12707 |
|
|
/* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
|
12708 |
|
|
as appropriate. I3 and I2 are the insns resulting from the combination
|
12709 |
|
|
insns including FROM (I2 may be zero).
|
12710 |
|
|
|
12711 |
|
|
ELIM_I2 and ELIM_I1 are either zero or registers that we know will
|
12712 |
|
|
not need REG_DEAD notes because they are being substituted for. This
|
12713 |
|
|
saves searching in the most common cases.
|
12714 |
|
|
|
12715 |
|
|
Each note in the list is either ignored or placed on some insns, depending
|
12716 |
|
|
on the type of note. */
|
12717 |
|
|
|
12718 |
|
|
static void
|
12719 |
|
|
distribute_notes (rtx notes, rtx from_insn, rtx i3, rtx i2, rtx elim_i2,
|
12720 |
|
|
rtx elim_i1)
|
12721 |
|
|
{
|
12722 |
|
|
rtx note, next_note;
|
12723 |
|
|
rtx tem;
|
12724 |
|
|
|
12725 |
|
|
for (note = notes; note; note = next_note)
|
12726 |
|
|
{
|
12727 |
|
|
rtx place = 0, place2 = 0;
|
12728 |
|
|
|
12729 |
|
|
next_note = XEXP (note, 1);
|
12730 |
|
|
switch (REG_NOTE_KIND (note))
|
12731 |
|
|
{
|
12732 |
|
|
case REG_BR_PROB:
|
12733 |
|
|
case REG_BR_PRED:
|
12734 |
|
|
/* Doesn't matter much where we put this, as long as it's somewhere.
|
12735 |
|
|
It is preferable to keep these notes on branches, which is most
|
12736 |
|
|
likely to be i3. */
|
12737 |
|
|
place = i3;
|
12738 |
|
|
break;
|
12739 |
|
|
|
12740 |
|
|
case REG_VALUE_PROFILE:
|
12741 |
|
|
/* Just get rid of this note, as it is unused later anyway. */
|
12742 |
|
|
break;
|
12743 |
|
|
|
12744 |
|
|
case REG_NON_LOCAL_GOTO:
|
12745 |
|
|
if (JUMP_P (i3))
|
12746 |
|
|
place = i3;
|
12747 |
|
|
else
|
12748 |
|
|
{
|
12749 |
|
|
gcc_assert (i2 && JUMP_P (i2));
|
12750 |
|
|
place = i2;
|
12751 |
|
|
}
|
12752 |
|
|
break;
|
12753 |
|
|
|
12754 |
|
|
case REG_EH_REGION:
|
12755 |
|
|
/* These notes must remain with the call or trapping instruction. */
|
12756 |
|
|
if (CALL_P (i3))
|
12757 |
|
|
place = i3;
|
12758 |
|
|
else if (i2 && CALL_P (i2))
|
12759 |
|
|
place = i2;
|
12760 |
|
|
else
|
12761 |
|
|
{
|
12762 |
|
|
gcc_assert (flag_non_call_exceptions);
|
12763 |
|
|
if (may_trap_p (i3))
|
12764 |
|
|
place = i3;
|
12765 |
|
|
else if (i2 && may_trap_p (i2))
|
12766 |
|
|
place = i2;
|
12767 |
|
|
/* ??? Otherwise assume we've combined things such that we
|
12768 |
|
|
can now prove that the instructions can't trap. Drop the
|
12769 |
|
|
note in this case. */
|
12770 |
|
|
}
|
12771 |
|
|
break;
|
12772 |
|
|
|
12773 |
|
|
case REG_NORETURN:
|
12774 |
|
|
case REG_SETJMP:
|
12775 |
|
|
/* These notes must remain with the call. It should not be
|
12776 |
|
|
possible for both I2 and I3 to be a call. */
|
12777 |
|
|
if (CALL_P (i3))
|
12778 |
|
|
place = i3;
|
12779 |
|
|
else
|
12780 |
|
|
{
|
12781 |
|
|
gcc_assert (i2 && CALL_P (i2));
|
12782 |
|
|
place = i2;
|
12783 |
|
|
}
|
12784 |
|
|
break;
|
12785 |
|
|
|
12786 |
|
|
case REG_UNUSED:
|
12787 |
|
|
/* Any clobbers for i3 may still exist, and so we must process
|
12788 |
|
|
REG_UNUSED notes from that insn.
|
12789 |
|
|
|
12790 |
|
|
Any clobbers from i2 or i1 can only exist if they were added by
|
12791 |
|
|
recog_for_combine. In that case, recog_for_combine created the
|
12792 |
|
|
necessary REG_UNUSED notes. Trying to keep any original
|
12793 |
|
|
REG_UNUSED notes from these insns can cause incorrect output
|
12794 |
|
|
if it is for the same register as the original i3 dest.
|
12795 |
|
|
In that case, we will notice that the register is set in i3,
|
12796 |
|
|
and then add a REG_UNUSED note for the destination of i3, which
|
12797 |
|
|
is wrong. However, it is possible to have REG_UNUSED notes from
|
12798 |
|
|
i2 or i1 for register which were both used and clobbered, so
|
12799 |
|
|
we keep notes from i2 or i1 if they will turn into REG_DEAD
|
12800 |
|
|
notes. */
|
12801 |
|
|
|
12802 |
|
|
/* If this register is set or clobbered in I3, put the note there
|
12803 |
|
|
unless there is one already. */
|
12804 |
|
|
if (reg_set_p (XEXP (note, 0), PATTERN (i3)))
|
12805 |
|
|
{
|
12806 |
|
|
if (from_insn != i3)
|
12807 |
|
|
break;
|
12808 |
|
|
|
12809 |
|
|
if (! (REG_P (XEXP (note, 0))
|
12810 |
|
|
? find_regno_note (i3, REG_UNUSED, REGNO (XEXP (note, 0)))
|
12811 |
|
|
: find_reg_note (i3, REG_UNUSED, XEXP (note, 0))))
|
12812 |
|
|
place = i3;
|
12813 |
|
|
}
|
12814 |
|
|
/* Otherwise, if this register is used by I3, then this register
|
12815 |
|
|
now dies here, so we must put a REG_DEAD note here unless there
|
12816 |
|
|
is one already. */
|
12817 |
|
|
else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3))
|
12818 |
|
|
&& ! (REG_P (XEXP (note, 0))
|
12819 |
|
|
? find_regno_note (i3, REG_DEAD,
|
12820 |
|
|
REGNO (XEXP (note, 0)))
|
12821 |
|
|
: find_reg_note (i3, REG_DEAD, XEXP (note, 0))))
|
12822 |
|
|
{
|
12823 |
|
|
PUT_REG_NOTE_KIND (note, REG_DEAD);
|
12824 |
|
|
place = i3;
|
12825 |
|
|
}
|
12826 |
|
|
break;
|
12827 |
|
|
|
12828 |
|
|
case REG_EQUAL:
|
12829 |
|
|
case REG_EQUIV:
|
12830 |
|
|
case REG_NOALIAS:
|
12831 |
|
|
/* These notes say something about results of an insn. We can
|
12832 |
|
|
only support them if they used to be on I3 in which case they
|
12833 |
|
|
remain on I3. Otherwise they are ignored.
|
12834 |
|
|
|
12835 |
|
|
If the note refers to an expression that is not a constant, we
|
12836 |
|
|
must also ignore the note since we cannot tell whether the
|
12837 |
|
|
equivalence is still true. It might be possible to do
|
12838 |
|
|
slightly better than this (we only have a problem if I2DEST
|
12839 |
|
|
or I1DEST is present in the expression), but it doesn't
|
12840 |
|
|
seem worth the trouble. */
|
12841 |
|
|
|
12842 |
|
|
if (from_insn == i3
|
12843 |
|
|
&& (XEXP (note, 0) == 0 || CONSTANT_P (XEXP (note, 0))))
|
12844 |
|
|
place = i3;
|
12845 |
|
|
break;
|
12846 |
|
|
|
12847 |
|
|
case REG_INC:
|
12848 |
|
|
/* These notes say something about how a register is used. They must
|
12849 |
|
|
be present on any use of the register in I2 or I3. */
|
12850 |
|
|
if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3)))
|
12851 |
|
|
place = i3;
|
12852 |
|
|
|
12853 |
|
|
if (i2 && reg_mentioned_p (XEXP (note, 0), PATTERN (i2)))
|
12854 |
|
|
{
|
12855 |
|
|
if (place)
|
12856 |
|
|
place2 = i2;
|
12857 |
|
|
else
|
12858 |
|
|
place = i2;
|
12859 |
|
|
}
|
12860 |
|
|
break;
|
12861 |
|
|
|
12862 |
|
|
case REG_LABEL_TARGET:
|
12863 |
|
|
case REG_LABEL_OPERAND:
|
12864 |
|
|
/* This can show up in several ways -- either directly in the
|
12865 |
|
|
pattern, or hidden off in the constant pool with (or without?)
|
12866 |
|
|
a REG_EQUAL note. */
|
12867 |
|
|
/* ??? Ignore the without-reg_equal-note problem for now. */
|
12868 |
|
|
if (reg_mentioned_p (XEXP (note, 0), PATTERN (i3))
|
12869 |
|
|
|| ((tem = find_reg_note (i3, REG_EQUAL, NULL_RTX))
|
12870 |
|
|
&& GET_CODE (XEXP (tem, 0)) == LABEL_REF
|
12871 |
|
|
&& XEXP (XEXP (tem, 0), 0) == XEXP (note, 0)))
|
12872 |
|
|
place = i3;
|
12873 |
|
|
|
12874 |
|
|
if (i2
|
12875 |
|
|
&& (reg_mentioned_p (XEXP (note, 0), PATTERN (i2))
|
12876 |
|
|
|| ((tem = find_reg_note (i2, REG_EQUAL, NULL_RTX))
|
12877 |
|
|
&& GET_CODE (XEXP (tem, 0)) == LABEL_REF
|
12878 |
|
|
&& XEXP (XEXP (tem, 0), 0) == XEXP (note, 0))))
|
12879 |
|
|
{
|
12880 |
|
|
if (place)
|
12881 |
|
|
place2 = i2;
|
12882 |
|
|
else
|
12883 |
|
|
place = i2;
|
12884 |
|
|
}
|
12885 |
|
|
|
12886 |
|
|
/* For REG_LABEL_TARGET on a JUMP_P, we prefer to put the note
|
12887 |
|
|
as a JUMP_LABEL or decrement LABEL_NUSES if it's already
|
12888 |
|
|
there. */
|
12889 |
|
|
if (place && JUMP_P (place)
|
12890 |
|
|
&& REG_NOTE_KIND (note) == REG_LABEL_TARGET
|
12891 |
|
|
&& (JUMP_LABEL (place) == NULL
|
12892 |
|
|
|| JUMP_LABEL (place) == XEXP (note, 0)))
|
12893 |
|
|
{
|
12894 |
|
|
rtx label = JUMP_LABEL (place);
|
12895 |
|
|
|
12896 |
|
|
if (!label)
|
12897 |
|
|
JUMP_LABEL (place) = XEXP (note, 0);
|
12898 |
|
|
else if (LABEL_P (label))
|
12899 |
|
|
LABEL_NUSES (label)--;
|
12900 |
|
|
}
|
12901 |
|
|
|
12902 |
|
|
if (place2 && JUMP_P (place2)
|
12903 |
|
|
&& REG_NOTE_KIND (note) == REG_LABEL_TARGET
|
12904 |
|
|
&& (JUMP_LABEL (place2) == NULL
|
12905 |
|
|
|| JUMP_LABEL (place2) == XEXP (note, 0)))
|
12906 |
|
|
{
|
12907 |
|
|
rtx label = JUMP_LABEL (place2);
|
12908 |
|
|
|
12909 |
|
|
if (!label)
|
12910 |
|
|
JUMP_LABEL (place2) = XEXP (note, 0);
|
12911 |
|
|
else if (LABEL_P (label))
|
12912 |
|
|
LABEL_NUSES (label)--;
|
12913 |
|
|
place2 = 0;
|
12914 |
|
|
}
|
12915 |
|
|
break;
|
12916 |
|
|
|
12917 |
|
|
case REG_NONNEG:
|
12918 |
|
|
/* This note says something about the value of a register prior
|
12919 |
|
|
to the execution of an insn. It is too much trouble to see
|
12920 |
|
|
if the note is still correct in all situations. It is better
|
12921 |
|
|
to simply delete it. */
|
12922 |
|
|
break;
|
12923 |
|
|
|
12924 |
|
|
case REG_DEAD:
|
12925 |
|
|
/* If we replaced the right hand side of FROM_INSN with a
|
12926 |
|
|
REG_EQUAL note, the original use of the dying register
|
12927 |
|
|
will not have been combined into I3 and I2. In such cases,
|
12928 |
|
|
FROM_INSN is guaranteed to be the first of the combined
|
12929 |
|
|
instructions, so we simply need to search back before
|
12930 |
|
|
FROM_INSN for the previous use or set of this register,
|
12931 |
|
|
then alter the notes there appropriately.
|
12932 |
|
|
|
12933 |
|
|
If the register is used as an input in I3, it dies there.
|
12934 |
|
|
Similarly for I2, if it is nonzero and adjacent to I3.
|
12935 |
|
|
|
12936 |
|
|
If the register is not used as an input in either I3 or I2
|
12937 |
|
|
and it is not one of the registers we were supposed to eliminate,
|
12938 |
|
|
there are two possibilities. We might have a non-adjacent I2
|
12939 |
|
|
or we might have somehow eliminated an additional register
|
12940 |
|
|
from a computation. For example, we might have had A & B where
|
12941 |
|
|
we discover that B will always be zero. In this case we will
|
12942 |
|
|
eliminate the reference to A.
|
12943 |
|
|
|
12944 |
|
|
In both cases, we must search to see if we can find a previous
|
12945 |
|
|
use of A and put the death note there. */
|
12946 |
|
|
|
12947 |
|
|
if (from_insn
|
12948 |
|
|
&& from_insn == i2mod
|
12949 |
|
|
&& !reg_overlap_mentioned_p (XEXP (note, 0), i2mod_new_rhs))
|
12950 |
|
|
tem = from_insn;
|
12951 |
|
|
else
|
12952 |
|
|
{
|
12953 |
|
|
if (from_insn
|
12954 |
|
|
&& CALL_P (from_insn)
|
12955 |
|
|
&& find_reg_fusage (from_insn, USE, XEXP (note, 0)))
|
12956 |
|
|
place = from_insn;
|
12957 |
|
|
else if (reg_referenced_p (XEXP (note, 0), PATTERN (i3)))
|
12958 |
|
|
place = i3;
|
12959 |
|
|
else if (i2 != 0 && next_nonnote_nondebug_insn (i2) == i3
|
12960 |
|
|
&& reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
|
12961 |
|
|
place = i2;
|
12962 |
|
|
else if ((rtx_equal_p (XEXP (note, 0), elim_i2)
|
12963 |
|
|
&& !(i2mod
|
12964 |
|
|
&& reg_overlap_mentioned_p (XEXP (note, 0),
|
12965 |
|
|
i2mod_old_rhs)))
|
12966 |
|
|
|| rtx_equal_p (XEXP (note, 0), elim_i1))
|
12967 |
|
|
break;
|
12968 |
|
|
tem = i3;
|
12969 |
|
|
}
|
12970 |
|
|
|
12971 |
|
|
if (place == 0)
|
12972 |
|
|
{
|
12973 |
|
|
basic_block bb = this_basic_block;
|
12974 |
|
|
|
12975 |
|
|
for (tem = PREV_INSN (tem); place == 0; tem = PREV_INSN (tem))
|
12976 |
|
|
{
|
12977 |
|
|
if (!NONDEBUG_INSN_P (tem))
|
12978 |
|
|
{
|
12979 |
|
|
if (tem == BB_HEAD (bb))
|
12980 |
|
|
break;
|
12981 |
|
|
continue;
|
12982 |
|
|
}
|
12983 |
|
|
|
12984 |
|
|
/* If the register is being set at TEM, see if that is all
|
12985 |
|
|
TEM is doing. If so, delete TEM. Otherwise, make this
|
12986 |
|
|
into a REG_UNUSED note instead. Don't delete sets to
|
12987 |
|
|
global register vars. */
|
12988 |
|
|
if ((REGNO (XEXP (note, 0)) >= FIRST_PSEUDO_REGISTER
|
12989 |
|
|
|| !global_regs[REGNO (XEXP (note, 0))])
|
12990 |
|
|
&& reg_set_p (XEXP (note, 0), PATTERN (tem)))
|
12991 |
|
|
{
|
12992 |
|
|
rtx set = single_set (tem);
|
12993 |
|
|
rtx inner_dest = 0;
|
12994 |
|
|
#ifdef HAVE_cc0
|
12995 |
|
|
rtx cc0_setter = NULL_RTX;
|
12996 |
|
|
#endif
|
12997 |
|
|
|
12998 |
|
|
if (set != 0)
|
12999 |
|
|
for (inner_dest = SET_DEST (set);
|
13000 |
|
|
(GET_CODE (inner_dest) == STRICT_LOW_PART
|
13001 |
|
|
|| GET_CODE (inner_dest) == SUBREG
|
13002 |
|
|
|| GET_CODE (inner_dest) == ZERO_EXTRACT);
|
13003 |
|
|
inner_dest = XEXP (inner_dest, 0))
|
13004 |
|
|
;
|
13005 |
|
|
|
13006 |
|
|
/* Verify that it was the set, and not a clobber that
|
13007 |
|
|
modified the register.
|
13008 |
|
|
|
13009 |
|
|
CC0 targets must be careful to maintain setter/user
|
13010 |
|
|
pairs. If we cannot delete the setter due to side
|
13011 |
|
|
effects, mark the user with an UNUSED note instead
|
13012 |
|
|
of deleting it. */
|
13013 |
|
|
|
13014 |
|
|
if (set != 0 && ! side_effects_p (SET_SRC (set))
|
13015 |
|
|
&& rtx_equal_p (XEXP (note, 0), inner_dest)
|
13016 |
|
|
#ifdef HAVE_cc0
|
13017 |
|
|
&& (! reg_mentioned_p (cc0_rtx, SET_SRC (set))
|
13018 |
|
|
|| ((cc0_setter = prev_cc0_setter (tem)) != NULL
|
13019 |
|
|
&& sets_cc0_p (PATTERN (cc0_setter)) > 0))
|
13020 |
|
|
#endif
|
13021 |
|
|
)
|
13022 |
|
|
{
|
13023 |
|
|
/* Move the notes and links of TEM elsewhere.
|
13024 |
|
|
This might delete other dead insns recursively.
|
13025 |
|
|
First set the pattern to something that won't use
|
13026 |
|
|
any register. */
|
13027 |
|
|
rtx old_notes = REG_NOTES (tem);
|
13028 |
|
|
|
13029 |
|
|
PATTERN (tem) = pc_rtx;
|
13030 |
|
|
REG_NOTES (tem) = NULL;
|
13031 |
|
|
|
13032 |
|
|
distribute_notes (old_notes, tem, tem, NULL_RTX,
|
13033 |
|
|
NULL_RTX, NULL_RTX);
|
13034 |
|
|
distribute_links (LOG_LINKS (tem));
|
13035 |
|
|
|
13036 |
|
|
SET_INSN_DELETED (tem);
|
13037 |
|
|
if (tem == i2)
|
13038 |
|
|
i2 = NULL_RTX;
|
13039 |
|
|
|
13040 |
|
|
#ifdef HAVE_cc0
|
13041 |
|
|
/* Delete the setter too. */
|
13042 |
|
|
if (cc0_setter)
|
13043 |
|
|
{
|
13044 |
|
|
PATTERN (cc0_setter) = pc_rtx;
|
13045 |
|
|
old_notes = REG_NOTES (cc0_setter);
|
13046 |
|
|
REG_NOTES (cc0_setter) = NULL;
|
13047 |
|
|
|
13048 |
|
|
distribute_notes (old_notes, cc0_setter,
|
13049 |
|
|
cc0_setter, NULL_RTX,
|
13050 |
|
|
NULL_RTX, NULL_RTX);
|
13051 |
|
|
distribute_links (LOG_LINKS (cc0_setter));
|
13052 |
|
|
|
13053 |
|
|
SET_INSN_DELETED (cc0_setter);
|
13054 |
|
|
if (cc0_setter == i2)
|
13055 |
|
|
i2 = NULL_RTX;
|
13056 |
|
|
}
|
13057 |
|
|
#endif
|
13058 |
|
|
}
|
13059 |
|
|
else
|
13060 |
|
|
{
|
13061 |
|
|
PUT_REG_NOTE_KIND (note, REG_UNUSED);
|
13062 |
|
|
|
13063 |
|
|
/* If there isn't already a REG_UNUSED note, put one
|
13064 |
|
|
here. Do not place a REG_DEAD note, even if
|
13065 |
|
|
the register is also used here; that would not
|
13066 |
|
|
match the algorithm used in lifetime analysis
|
13067 |
|
|
and can cause the consistency check in the
|
13068 |
|
|
scheduler to fail. */
|
13069 |
|
|
if (! find_regno_note (tem, REG_UNUSED,
|
13070 |
|
|
REGNO (XEXP (note, 0))))
|
13071 |
|
|
place = tem;
|
13072 |
|
|
break;
|
13073 |
|
|
}
|
13074 |
|
|
}
|
13075 |
|
|
else if (reg_referenced_p (XEXP (note, 0), PATTERN (tem))
|
13076 |
|
|
|| (CALL_P (tem)
|
13077 |
|
|
&& find_reg_fusage (tem, USE, XEXP (note, 0))))
|
13078 |
|
|
{
|
13079 |
|
|
place = tem;
|
13080 |
|
|
|
13081 |
|
|
/* If we are doing a 3->2 combination, and we have a
|
13082 |
|
|
register which formerly died in i3 and was not used
|
13083 |
|
|
by i2, which now no longer dies in i3 and is used in
|
13084 |
|
|
i2 but does not die in i2, and place is between i2
|
13085 |
|
|
and i3, then we may need to move a link from place to
|
13086 |
|
|
i2. */
|
13087 |
|
|
if (i2 && DF_INSN_LUID (place) > DF_INSN_LUID (i2)
|
13088 |
|
|
&& from_insn
|
13089 |
|
|
&& DF_INSN_LUID (from_insn) > DF_INSN_LUID (i2)
|
13090 |
|
|
&& reg_referenced_p (XEXP (note, 0), PATTERN (i2)))
|
13091 |
|
|
{
|
13092 |
|
|
rtx links = LOG_LINKS (place);
|
13093 |
|
|
LOG_LINKS (place) = 0;
|
13094 |
|
|
distribute_links (links);
|
13095 |
|
|
}
|
13096 |
|
|
break;
|
13097 |
|
|
}
|
13098 |
|
|
|
13099 |
|
|
if (tem == BB_HEAD (bb))
|
13100 |
|
|
break;
|
13101 |
|
|
}
|
13102 |
|
|
|
13103 |
|
|
}
|
13104 |
|
|
|
13105 |
|
|
/* If the register is set or already dead at PLACE, we needn't do
|
13106 |
|
|
anything with this note if it is still a REG_DEAD note.
|
13107 |
|
|
We check here if it is set at all, not if is it totally replaced,
|
13108 |
|
|
which is what `dead_or_set_p' checks, so also check for it being
|
13109 |
|
|
set partially. */
|
13110 |
|
|
|
13111 |
|
|
if (place && REG_NOTE_KIND (note) == REG_DEAD)
|
13112 |
|
|
{
|
13113 |
|
|
unsigned int regno = REGNO (XEXP (note, 0));
|
13114 |
|
|
reg_stat_type *rsp = VEC_index (reg_stat_type, reg_stat, regno);
|
13115 |
|
|
|
13116 |
|
|
if (dead_or_set_p (place, XEXP (note, 0))
|
13117 |
|
|
|| reg_bitfield_target_p (XEXP (note, 0), PATTERN (place)))
|
13118 |
|
|
{
|
13119 |
|
|
/* Unless the register previously died in PLACE, clear
|
13120 |
|
|
last_death. [I no longer understand why this is
|
13121 |
|
|
being done.] */
|
13122 |
|
|
if (rsp->last_death != place)
|
13123 |
|
|
rsp->last_death = 0;
|
13124 |
|
|
place = 0;
|
13125 |
|
|
}
|
13126 |
|
|
else
|
13127 |
|
|
rsp->last_death = place;
|
13128 |
|
|
|
13129 |
|
|
/* If this is a death note for a hard reg that is occupying
|
13130 |
|
|
multiple registers, ensure that we are still using all
|
13131 |
|
|
parts of the object. If we find a piece of the object
|
13132 |
|
|
that is unused, we must arrange for an appropriate REG_DEAD
|
13133 |
|
|
note to be added for it. However, we can't just emit a USE
|
13134 |
|
|
and tag the note to it, since the register might actually
|
13135 |
|
|
be dead; so we recourse, and the recursive call then finds
|
13136 |
|
|
the previous insn that used this register. */
|
13137 |
|
|
|
13138 |
|
|
if (place && regno < FIRST_PSEUDO_REGISTER
|
13139 |
|
|
&& hard_regno_nregs[regno][GET_MODE (XEXP (note, 0))] > 1)
|
13140 |
|
|
{
|
13141 |
|
|
unsigned int endregno = END_HARD_REGNO (XEXP (note, 0));
|
13142 |
|
|
int all_used = 1;
|
13143 |
|
|
unsigned int i;
|
13144 |
|
|
|
13145 |
|
|
for (i = regno; i < endregno; i++)
|
13146 |
|
|
if ((! refers_to_regno_p (i, i + 1, PATTERN (place), 0)
|
13147 |
|
|
&& ! find_regno_fusage (place, USE, i))
|
13148 |
|
|
|| dead_or_set_regno_p (place, i))
|
13149 |
|
|
all_used = 0;
|
13150 |
|
|
|
13151 |
|
|
if (! all_used)
|
13152 |
|
|
{
|
13153 |
|
|
/* Put only REG_DEAD notes for pieces that are
|
13154 |
|
|
not already dead or set. */
|
13155 |
|
|
|
13156 |
|
|
for (i = regno; i < endregno;
|
13157 |
|
|
i += hard_regno_nregs[i][reg_raw_mode[i]])
|
13158 |
|
|
{
|
13159 |
|
|
rtx piece = regno_reg_rtx[i];
|
13160 |
|
|
basic_block bb = this_basic_block;
|
13161 |
|
|
|
13162 |
|
|
if (! dead_or_set_p (place, piece)
|
13163 |
|
|
&& ! reg_bitfield_target_p (piece,
|
13164 |
|
|
PATTERN (place)))
|
13165 |
|
|
{
|
13166 |
|
|
rtx new_note = alloc_reg_note (REG_DEAD, piece,
|
13167 |
|
|
NULL_RTX);
|
13168 |
|
|
|
13169 |
|
|
distribute_notes (new_note, place, place,
|
13170 |
|
|
NULL_RTX, NULL_RTX, NULL_RTX);
|
13171 |
|
|
}
|
13172 |
|
|
else if (! refers_to_regno_p (i, i + 1,
|
13173 |
|
|
PATTERN (place), 0)
|
13174 |
|
|
&& ! find_regno_fusage (place, USE, i))
|
13175 |
|
|
for (tem = PREV_INSN (place); ;
|
13176 |
|
|
tem = PREV_INSN (tem))
|
13177 |
|
|
{
|
13178 |
|
|
if (!NONDEBUG_INSN_P (tem))
|
13179 |
|
|
{
|
13180 |
|
|
if (tem == BB_HEAD (bb))
|
13181 |
|
|
break;
|
13182 |
|
|
continue;
|
13183 |
|
|
}
|
13184 |
|
|
if (dead_or_set_p (tem, piece)
|
13185 |
|
|
|| reg_bitfield_target_p (piece,
|
13186 |
|
|
PATTERN (tem)))
|
13187 |
|
|
{
|
13188 |
|
|
add_reg_note (tem, REG_UNUSED, piece);
|
13189 |
|
|
break;
|
13190 |
|
|
}
|
13191 |
|
|
}
|
13192 |
|
|
|
13193 |
|
|
}
|
13194 |
|
|
|
13195 |
|
|
place = 0;
|
13196 |
|
|
}
|
13197 |
|
|
}
|
13198 |
|
|
}
|
13199 |
|
|
break;
|
13200 |
|
|
|
13201 |
|
|
default:
|
13202 |
|
|
/* Any other notes should not be present at this point in the
|
13203 |
|
|
compilation. */
|
13204 |
|
|
gcc_unreachable ();
|
13205 |
|
|
}
|
13206 |
|
|
|
13207 |
|
|
if (place)
|
13208 |
|
|
{
|
13209 |
|
|
XEXP (note, 1) = REG_NOTES (place);
|
13210 |
|
|
REG_NOTES (place) = note;
|
13211 |
|
|
}
|
13212 |
|
|
|
13213 |
|
|
if (place2)
|
13214 |
|
|
add_reg_note (place2, REG_NOTE_KIND (note), XEXP (note, 0));
|
13215 |
|
|
}
|
13216 |
|
|
}
|
13217 |
|
|
|
13218 |
|
|
/* Similarly to above, distribute the LOG_LINKS that used to be present on
|
13219 |
|
|
I3, I2, and I1 to new locations. This is also called to add a link
|
13220 |
|
|
pointing at I3 when I3's destination is changed. */
|
13221 |
|
|
|
13222 |
|
|
static void
|
13223 |
|
|
distribute_links (rtx links)
|
13224 |
|
|
{
|
13225 |
|
|
rtx link, next_link;
|
13226 |
|
|
|
13227 |
|
|
for (link = links; link; link = next_link)
|
13228 |
|
|
{
|
13229 |
|
|
rtx place = 0;
|
13230 |
|
|
rtx insn;
|
13231 |
|
|
rtx set, reg;
|
13232 |
|
|
|
13233 |
|
|
next_link = XEXP (link, 1);
|
13234 |
|
|
|
13235 |
|
|
/* If the insn that this link points to is a NOTE or isn't a single
|
13236 |
|
|
set, ignore it. In the latter case, it isn't clear what we
|
13237 |
|
|
can do other than ignore the link, since we can't tell which
|
13238 |
|
|
register it was for. Such links wouldn't be used by combine
|
13239 |
|
|
anyway.
|
13240 |
|
|
|
13241 |
|
|
It is not possible for the destination of the target of the link to
|
13242 |
|
|
have been changed by combine. The only potential of this is if we
|
13243 |
|
|
replace I3, I2, and I1 by I3 and I2. But in that case the
|
13244 |
|
|
destination of I2 also remains unchanged. */
|
13245 |
|
|
|
13246 |
|
|
if (NOTE_P (XEXP (link, 0))
|
13247 |
|
|
|| (set = single_set (XEXP (link, 0))) == 0)
|
13248 |
|
|
continue;
|
13249 |
|
|
|
13250 |
|
|
reg = SET_DEST (set);
|
13251 |
|
|
while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
|
13252 |
|
|
|| GET_CODE (reg) == STRICT_LOW_PART)
|
13253 |
|
|
reg = XEXP (reg, 0);
|
13254 |
|
|
|
13255 |
|
|
/* A LOG_LINK is defined as being placed on the first insn that uses
|
13256 |
|
|
a register and points to the insn that sets the register. Start
|
13257 |
|
|
searching at the next insn after the target of the link and stop
|
13258 |
|
|
when we reach a set of the register or the end of the basic block.
|
13259 |
|
|
|
13260 |
|
|
Note that this correctly handles the link that used to point from
|
13261 |
|
|
I3 to I2. Also note that not much searching is typically done here
|
13262 |
|
|
since most links don't point very far away. */
|
13263 |
|
|
|
13264 |
|
|
for (insn = NEXT_INSN (XEXP (link, 0));
|
13265 |
|
|
(insn && (this_basic_block->next_bb == EXIT_BLOCK_PTR
|
13266 |
|
|
|| BB_HEAD (this_basic_block->next_bb) != insn));
|
13267 |
|
|
insn = NEXT_INSN (insn))
|
13268 |
|
|
if (DEBUG_INSN_P (insn))
|
13269 |
|
|
continue;
|
13270 |
|
|
else if (INSN_P (insn) && reg_overlap_mentioned_p (reg, PATTERN (insn)))
|
13271 |
|
|
{
|
13272 |
|
|
if (reg_referenced_p (reg, PATTERN (insn)))
|
13273 |
|
|
place = insn;
|
13274 |
|
|
break;
|
13275 |
|
|
}
|
13276 |
|
|
else if (CALL_P (insn)
|
13277 |
|
|
&& find_reg_fusage (insn, USE, reg))
|
13278 |
|
|
{
|
13279 |
|
|
place = insn;
|
13280 |
|
|
break;
|
13281 |
|
|
}
|
13282 |
|
|
else if (INSN_P (insn) && reg_set_p (reg, insn))
|
13283 |
|
|
break;
|
13284 |
|
|
|
13285 |
|
|
/* If we found a place to put the link, place it there unless there
|
13286 |
|
|
is already a link to the same insn as LINK at that point. */
|
13287 |
|
|
|
13288 |
|
|
if (place)
|
13289 |
|
|
{
|
13290 |
|
|
rtx link2;
|
13291 |
|
|
|
13292 |
|
|
for (link2 = LOG_LINKS (place); link2; link2 = XEXP (link2, 1))
|
13293 |
|
|
if (XEXP (link2, 0) == XEXP (link, 0))
|
13294 |
|
|
break;
|
13295 |
|
|
|
13296 |
|
|
if (link2 == 0)
|
13297 |
|
|
{
|
13298 |
|
|
XEXP (link, 1) = LOG_LINKS (place);
|
13299 |
|
|
LOG_LINKS (place) = link;
|
13300 |
|
|
|
13301 |
|
|
/* Set added_links_insn to the earliest insn we added a
|
13302 |
|
|
link to. */
|
13303 |
|
|
if (added_links_insn == 0
|
13304 |
|
|
|| DF_INSN_LUID (added_links_insn) > DF_INSN_LUID (place))
|
13305 |
|
|
added_links_insn = place;
|
13306 |
|
|
}
|
13307 |
|
|
}
|
13308 |
|
|
}
|
13309 |
|
|
}
|
13310 |
|
|
|
13311 |
|
|
/* Subroutine of unmentioned_reg_p and callback from for_each_rtx.
|
13312 |
|
|
Check whether the expression pointer to by LOC is a register or
|
13313 |
|
|
memory, and if so return 1 if it isn't mentioned in the rtx EXPR.
|
13314 |
|
|
Otherwise return zero. */
|
13315 |
|
|
|
13316 |
|
|
static int
|
13317 |
|
|
unmentioned_reg_p_1 (rtx *loc, void *expr)
|
13318 |
|
|
{
|
13319 |
|
|
rtx x = *loc;
|
13320 |
|
|
|
13321 |
|
|
if (x != NULL_RTX
|
13322 |
|
|
&& (REG_P (x) || MEM_P (x))
|
13323 |
|
|
&& ! reg_mentioned_p (x, (rtx) expr))
|
13324 |
|
|
return 1;
|
13325 |
|
|
return 0;
|
13326 |
|
|
}
|
13327 |
|
|
|
13328 |
|
|
/* Check for any register or memory mentioned in EQUIV that is not
|
13329 |
|
|
mentioned in EXPR. This is used to restrict EQUIV to "specializations"
|
13330 |
|
|
of EXPR where some registers may have been replaced by constants. */
|
13331 |
|
|
|
13332 |
|
|
static bool
|
13333 |
|
|
unmentioned_reg_p (rtx equiv, rtx expr)
|
13334 |
|
|
{
|
13335 |
|
|
return for_each_rtx (&equiv, unmentioned_reg_p_1, expr);
|
13336 |
|
|
}
|
13337 |
|
|
|
13338 |
|
|
void
|
13339 |
|
|
dump_combine_stats (FILE *file)
|
13340 |
|
|
{
|
13341 |
|
|
fprintf
|
13342 |
|
|
(file,
|
13343 |
|
|
";; Combiner statistics: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n\n",
|
13344 |
|
|
combine_attempts, combine_merges, combine_extras, combine_successes);
|
13345 |
|
|
}
|
13346 |
|
|
|
13347 |
|
|
void
|
13348 |
|
|
dump_combine_total_stats (FILE *file)
|
13349 |
|
|
{
|
13350 |
|
|
fprintf
|
13351 |
|
|
(file,
|
13352 |
|
|
"\n;; Combiner totals: %d attempts, %d substitutions (%d requiring new space),\n;; %d successes.\n",
|
13353 |
|
|
total_attempts, total_merges, total_extras, total_successes);
|
13354 |
|
|
}
|
13355 |
|
|
|
13356 |
|
|
static bool
|
13357 |
|
|
gate_handle_combine (void)
|
13358 |
|
|
{
|
13359 |
|
|
return (optimize > 0);
|
13360 |
|
|
}
|
13361 |
|
|
|
13362 |
|
|
/* Try combining insns through substitution. */
|
13363 |
|
|
static unsigned int
|
13364 |
|
|
rest_of_handle_combine (void)
|
13365 |
|
|
{
|
13366 |
|
|
int rebuild_jump_labels_after_combine;
|
13367 |
|
|
|
13368 |
|
|
df_set_flags (DF_LR_RUN_DCE + DF_DEFER_INSN_RESCAN);
|
13369 |
|
|
df_note_add_problem ();
|
13370 |
|
|
df_analyze ();
|
13371 |
|
|
|
13372 |
|
|
regstat_init_n_sets_and_refs ();
|
13373 |
|
|
|
13374 |
|
|
rebuild_jump_labels_after_combine
|
13375 |
|
|
= combine_instructions (get_insns (), max_reg_num ());
|
13376 |
|
|
|
13377 |
|
|
/* Combining insns may have turned an indirect jump into a
|
13378 |
|
|
direct jump. Rebuild the JUMP_LABEL fields of jumping
|
13379 |
|
|
instructions. */
|
13380 |
|
|
if (rebuild_jump_labels_after_combine)
|
13381 |
|
|
{
|
13382 |
|
|
timevar_push (TV_JUMP);
|
13383 |
|
|
rebuild_jump_labels (get_insns ());
|
13384 |
|
|
cleanup_cfg (0);
|
13385 |
|
|
timevar_pop (TV_JUMP);
|
13386 |
|
|
}
|
13387 |
|
|
|
13388 |
|
|
regstat_free_n_sets_and_refs ();
|
13389 |
|
|
return 0;
|
13390 |
|
|
}
|
13391 |
|
|
|
13392 |
|
|
struct rtl_opt_pass pass_combine =
|
13393 |
|
|
{
|
13394 |
|
|
{
|
13395 |
|
|
RTL_PASS,
|
13396 |
|
|
"combine", /* name */
|
13397 |
|
|
gate_handle_combine, /* gate */
|
13398 |
|
|
rest_of_handle_combine, /* execute */
|
13399 |
|
|
NULL, /* sub */
|
13400 |
|
|
NULL, /* next */
|
13401 |
|
|
0, /* static_pass_number */
|
13402 |
|
|
TV_COMBINE, /* tv_id */
|
13403 |
|
|
PROP_cfglayout, /* properties_required */
|
13404 |
|
|
0, /* properties_provided */
|
13405 |
|
|
0, /* properties_destroyed */
|
13406 |
|
|
0, /* todo_flags_start */
|
13407 |
|
|
TODO_dump_func |
|
13408 |
|
|
TODO_df_finish | TODO_verify_rtl_sharing |
|
13409 |
|
|
TODO_ggc_collect, /* todo_flags_finish */
|
13410 |
|
|
}
|
13411 |
|
|
};
|