1 |
280 |
jeremybenn |
/* Alias analysis for GNU C
|
2 |
|
|
Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006,
|
3 |
|
|
2007, 2008, 2009, 2010 Free Software Foundation, Inc.
|
4 |
|
|
Contributed by John Carr (jfc@mit.edu).
|
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 |
|
|
#include "config.h"
|
23 |
|
|
#include "system.h"
|
24 |
|
|
#include "coretypes.h"
|
25 |
|
|
#include "tm.h"
|
26 |
|
|
#include "rtl.h"
|
27 |
|
|
#include "tree.h"
|
28 |
|
|
#include "tm_p.h"
|
29 |
|
|
#include "function.h"
|
30 |
|
|
#include "alias.h"
|
31 |
|
|
#include "emit-rtl.h"
|
32 |
|
|
#include "regs.h"
|
33 |
|
|
#include "hard-reg-set.h"
|
34 |
|
|
#include "basic-block.h"
|
35 |
|
|
#include "flags.h"
|
36 |
|
|
#include "output.h"
|
37 |
|
|
#include "toplev.h"
|
38 |
|
|
#include "cselib.h"
|
39 |
|
|
#include "splay-tree.h"
|
40 |
|
|
#include "ggc.h"
|
41 |
|
|
#include "langhooks.h"
|
42 |
|
|
#include "timevar.h"
|
43 |
|
|
#include "target.h"
|
44 |
|
|
#include "cgraph.h"
|
45 |
|
|
#include "varray.h"
|
46 |
|
|
#include "tree-pass.h"
|
47 |
|
|
#include "ipa-type-escape.h"
|
48 |
|
|
#include "df.h"
|
49 |
|
|
#include "tree-ssa-alias.h"
|
50 |
|
|
#include "pointer-set.h"
|
51 |
|
|
#include "tree-flow.h"
|
52 |
|
|
|
53 |
|
|
/* The aliasing API provided here solves related but different problems:
|
54 |
|
|
|
55 |
|
|
Say there exists (in c)
|
56 |
|
|
|
57 |
|
|
struct X {
|
58 |
|
|
struct Y y1;
|
59 |
|
|
struct Z z2;
|
60 |
|
|
} x1, *px1, *px2;
|
61 |
|
|
|
62 |
|
|
struct Y y2, *py;
|
63 |
|
|
struct Z z2, *pz;
|
64 |
|
|
|
65 |
|
|
|
66 |
|
|
py = &px1.y1;
|
67 |
|
|
px2 = &x1;
|
68 |
|
|
|
69 |
|
|
Consider the four questions:
|
70 |
|
|
|
71 |
|
|
Can a store to x1 interfere with px2->y1?
|
72 |
|
|
Can a store to x1 interfere with px2->z2?
|
73 |
|
|
(*px2).z2
|
74 |
|
|
Can a store to x1 change the value pointed to by with py?
|
75 |
|
|
Can a store to x1 change the value pointed to by with pz?
|
76 |
|
|
|
77 |
|
|
The answer to these questions can be yes, yes, yes, and maybe.
|
78 |
|
|
|
79 |
|
|
The first two questions can be answered with a simple examination
|
80 |
|
|
of the type system. If structure X contains a field of type Y then
|
81 |
|
|
a store thru a pointer to an X can overwrite any field that is
|
82 |
|
|
contained (recursively) in an X (unless we know that px1 != px2).
|
83 |
|
|
|
84 |
|
|
The last two of the questions can be solved in the same way as the
|
85 |
|
|
first two questions but this is too conservative. The observation
|
86 |
|
|
is that in some cases analysis we can know if which (if any) fields
|
87 |
|
|
are addressed and if those addresses are used in bad ways. This
|
88 |
|
|
analysis may be language specific. In C, arbitrary operations may
|
89 |
|
|
be applied to pointers. However, there is some indication that
|
90 |
|
|
this may be too conservative for some C++ types.
|
91 |
|
|
|
92 |
|
|
The pass ipa-type-escape does this analysis for the types whose
|
93 |
|
|
instances do not escape across the compilation boundary.
|
94 |
|
|
|
95 |
|
|
Historically in GCC, these two problems were combined and a single
|
96 |
|
|
data structure was used to represent the solution to these
|
97 |
|
|
problems. We now have two similar but different data structures,
|
98 |
|
|
The data structure to solve the last two question is similar to the
|
99 |
|
|
first, but does not contain have the fields in it whose address are
|
100 |
|
|
never taken. For types that do escape the compilation unit, the
|
101 |
|
|
data structures will have identical information.
|
102 |
|
|
*/
|
103 |
|
|
|
104 |
|
|
/* The alias sets assigned to MEMs assist the back-end in determining
|
105 |
|
|
which MEMs can alias which other MEMs. In general, two MEMs in
|
106 |
|
|
different alias sets cannot alias each other, with one important
|
107 |
|
|
exception. Consider something like:
|
108 |
|
|
|
109 |
|
|
struct S { int i; double d; };
|
110 |
|
|
|
111 |
|
|
a store to an `S' can alias something of either type `int' or type
|
112 |
|
|
`double'. (However, a store to an `int' cannot alias a `double'
|
113 |
|
|
and vice versa.) We indicate this via a tree structure that looks
|
114 |
|
|
like:
|
115 |
|
|
struct S
|
116 |
|
|
/ \
|
117 |
|
|
/ \
|
118 |
|
|
|/_ _\|
|
119 |
|
|
int double
|
120 |
|
|
|
121 |
|
|
(The arrows are directed and point downwards.)
|
122 |
|
|
In this situation we say the alias set for `struct S' is the
|
123 |
|
|
`superset' and that those for `int' and `double' are `subsets'.
|
124 |
|
|
|
125 |
|
|
To see whether two alias sets can point to the same memory, we must
|
126 |
|
|
see if either alias set is a subset of the other. We need not trace
|
127 |
|
|
past immediate descendants, however, since we propagate all
|
128 |
|
|
grandchildren up one level.
|
129 |
|
|
|
130 |
|
|
Alias set zero is implicitly a superset of all other alias sets.
|
131 |
|
|
However, this is no actual entry for alias set zero. It is an
|
132 |
|
|
error to attempt to explicitly construct a subset of zero. */
|
133 |
|
|
|
134 |
|
|
struct GTY(()) alias_set_entry_d {
|
135 |
|
|
/* The alias set number, as stored in MEM_ALIAS_SET. */
|
136 |
|
|
alias_set_type alias_set;
|
137 |
|
|
|
138 |
|
|
/* Nonzero if would have a child of zero: this effectively makes this
|
139 |
|
|
alias set the same as alias set zero. */
|
140 |
|
|
int has_zero_child;
|
141 |
|
|
|
142 |
|
|
/* The children of the alias set. These are not just the immediate
|
143 |
|
|
children, but, in fact, all descendants. So, if we have:
|
144 |
|
|
|
145 |
|
|
struct T { struct S s; float f; }
|
146 |
|
|
|
147 |
|
|
continuing our example above, the children here will be all of
|
148 |
|
|
`int', `double', `float', and `struct S'. */
|
149 |
|
|
splay_tree GTY((param1_is (int), param2_is (int))) children;
|
150 |
|
|
};
|
151 |
|
|
typedef struct alias_set_entry_d *alias_set_entry;
|
152 |
|
|
|
153 |
|
|
static int rtx_equal_for_memref_p (const_rtx, const_rtx);
|
154 |
|
|
static int memrefs_conflict_p (int, rtx, int, rtx, HOST_WIDE_INT);
|
155 |
|
|
static void record_set (rtx, const_rtx, void *);
|
156 |
|
|
static int base_alias_check (rtx, rtx, enum machine_mode,
|
157 |
|
|
enum machine_mode);
|
158 |
|
|
static rtx find_base_value (rtx);
|
159 |
|
|
static int mems_in_disjoint_alias_sets_p (const_rtx, const_rtx);
|
160 |
|
|
static int insert_subset_children (splay_tree_node, void*);
|
161 |
|
|
static alias_set_entry get_alias_set_entry (alias_set_type);
|
162 |
|
|
static const_rtx fixed_scalar_and_varying_struct_p (const_rtx, const_rtx, rtx, rtx,
|
163 |
|
|
bool (*) (const_rtx, bool));
|
164 |
|
|
static int aliases_everything_p (const_rtx);
|
165 |
|
|
static bool nonoverlapping_component_refs_p (const_tree, const_tree);
|
166 |
|
|
static tree decl_for_component_ref (tree);
|
167 |
|
|
static rtx adjust_offset_for_component_ref (tree, rtx);
|
168 |
|
|
static int write_dependence_p (const_rtx, const_rtx, int);
|
169 |
|
|
|
170 |
|
|
static void memory_modified_1 (rtx, const_rtx, void *);
|
171 |
|
|
|
172 |
|
|
/* Set up all info needed to perform alias analysis on memory references. */
|
173 |
|
|
|
174 |
|
|
/* Returns the size in bytes of the mode of X. */
|
175 |
|
|
#define SIZE_FOR_MODE(X) (GET_MODE_SIZE (GET_MODE (X)))
|
176 |
|
|
|
177 |
|
|
/* Returns nonzero if MEM1 and MEM2 do not alias because they are in
|
178 |
|
|
different alias sets. We ignore alias sets in functions making use
|
179 |
|
|
of variable arguments because the va_arg macros on some systems are
|
180 |
|
|
not legal ANSI C. */
|
181 |
|
|
#define DIFFERENT_ALIAS_SETS_P(MEM1, MEM2) \
|
182 |
|
|
mems_in_disjoint_alias_sets_p (MEM1, MEM2)
|
183 |
|
|
|
184 |
|
|
/* Cap the number of passes we make over the insns propagating alias
|
185 |
|
|
information through set chains. 10 is a completely arbitrary choice. */
|
186 |
|
|
#define MAX_ALIAS_LOOP_PASSES 10
|
187 |
|
|
|
188 |
|
|
/* reg_base_value[N] gives an address to which register N is related.
|
189 |
|
|
If all sets after the first add or subtract to the current value
|
190 |
|
|
or otherwise modify it so it does not point to a different top level
|
191 |
|
|
object, reg_base_value[N] is equal to the address part of the source
|
192 |
|
|
of the first set.
|
193 |
|
|
|
194 |
|
|
A base address can be an ADDRESS, SYMBOL_REF, or LABEL_REF. ADDRESS
|
195 |
|
|
expressions represent certain special values: function arguments and
|
196 |
|
|
the stack, frame, and argument pointers.
|
197 |
|
|
|
198 |
|
|
The contents of an ADDRESS is not normally used, the mode of the
|
199 |
|
|
ADDRESS determines whether the ADDRESS is a function argument or some
|
200 |
|
|
other special value. Pointer equality, not rtx_equal_p, determines whether
|
201 |
|
|
two ADDRESS expressions refer to the same base address.
|
202 |
|
|
|
203 |
|
|
The only use of the contents of an ADDRESS is for determining if the
|
204 |
|
|
current function performs nonlocal memory memory references for the
|
205 |
|
|
purposes of marking the function as a constant function. */
|
206 |
|
|
|
207 |
|
|
static GTY(()) VEC(rtx,gc) *reg_base_value;
|
208 |
|
|
static rtx *new_reg_base_value;
|
209 |
|
|
|
210 |
|
|
/* We preserve the copy of old array around to avoid amount of garbage
|
211 |
|
|
produced. About 8% of garbage produced were attributed to this
|
212 |
|
|
array. */
|
213 |
|
|
static GTY((deletable)) VEC(rtx,gc) *old_reg_base_value;
|
214 |
|
|
|
215 |
|
|
/* Static hunks of RTL used by the aliasing code; these are initialized
|
216 |
|
|
once per function to avoid unnecessary RTL allocations. */
|
217 |
|
|
static GTY (()) rtx static_reg_base_value[FIRST_PSEUDO_REGISTER];
|
218 |
|
|
|
219 |
|
|
#define REG_BASE_VALUE(X) \
|
220 |
|
|
(REGNO (X) < VEC_length (rtx, reg_base_value) \
|
221 |
|
|
? VEC_index (rtx, reg_base_value, REGNO (X)) : 0)
|
222 |
|
|
|
223 |
|
|
/* Vector indexed by N giving the initial (unchanging) value known for
|
224 |
|
|
pseudo-register N. This array is initialized in init_alias_analysis,
|
225 |
|
|
and does not change until end_alias_analysis is called. */
|
226 |
|
|
static GTY((length("reg_known_value_size"))) rtx *reg_known_value;
|
227 |
|
|
|
228 |
|
|
/* Indicates number of valid entries in reg_known_value. */
|
229 |
|
|
static GTY(()) unsigned int reg_known_value_size;
|
230 |
|
|
|
231 |
|
|
/* Vector recording for each reg_known_value whether it is due to a
|
232 |
|
|
REG_EQUIV note. Future passes (viz., reload) may replace the
|
233 |
|
|
pseudo with the equivalent expression and so we account for the
|
234 |
|
|
dependences that would be introduced if that happens.
|
235 |
|
|
|
236 |
|
|
The REG_EQUIV notes created in assign_parms may mention the arg
|
237 |
|
|
pointer, and there are explicit insns in the RTL that modify the
|
238 |
|
|
arg pointer. Thus we must ensure that such insns don't get
|
239 |
|
|
scheduled across each other because that would invalidate the
|
240 |
|
|
REG_EQUIV notes. One could argue that the REG_EQUIV notes are
|
241 |
|
|
wrong, but solving the problem in the scheduler will likely give
|
242 |
|
|
better code, so we do it here. */
|
243 |
|
|
static bool *reg_known_equiv_p;
|
244 |
|
|
|
245 |
|
|
/* True when scanning insns from the start of the rtl to the
|
246 |
|
|
NOTE_INSN_FUNCTION_BEG note. */
|
247 |
|
|
static bool copying_arguments;
|
248 |
|
|
|
249 |
|
|
DEF_VEC_P(alias_set_entry);
|
250 |
|
|
DEF_VEC_ALLOC_P(alias_set_entry,gc);
|
251 |
|
|
|
252 |
|
|
/* The splay-tree used to store the various alias set entries. */
|
253 |
|
|
static GTY (()) VEC(alias_set_entry,gc) *alias_sets;
|
254 |
|
|
|
255 |
|
|
/* Build a decomposed reference object for querying the alias-oracle
|
256 |
|
|
from the MEM rtx and store it in *REF.
|
257 |
|
|
Returns false if MEM is not suitable for the alias-oracle. */
|
258 |
|
|
|
259 |
|
|
static bool
|
260 |
|
|
ao_ref_from_mem (ao_ref *ref, const_rtx mem)
|
261 |
|
|
{
|
262 |
|
|
tree expr = MEM_EXPR (mem);
|
263 |
|
|
tree base;
|
264 |
|
|
|
265 |
|
|
if (!expr)
|
266 |
|
|
return false;
|
267 |
|
|
|
268 |
|
|
/* If MEM_OFFSET or MEM_SIZE are NULL punt. */
|
269 |
|
|
if (!MEM_OFFSET (mem)
|
270 |
|
|
|| !MEM_SIZE (mem))
|
271 |
|
|
return false;
|
272 |
|
|
|
273 |
|
|
ao_ref_init (ref, expr);
|
274 |
|
|
|
275 |
|
|
/* Get the base of the reference and see if we have to reject or
|
276 |
|
|
adjust it. */
|
277 |
|
|
base = ao_ref_base (ref);
|
278 |
|
|
if (base == NULL_TREE)
|
279 |
|
|
return false;
|
280 |
|
|
|
281 |
|
|
/* If this is a pointer dereference of a non-SSA_NAME punt.
|
282 |
|
|
??? We could replace it with a pointer to anything. */
|
283 |
|
|
if (INDIRECT_REF_P (base)
|
284 |
|
|
&& TREE_CODE (TREE_OPERAND (base, 0)) != SSA_NAME)
|
285 |
|
|
return false;
|
286 |
|
|
|
287 |
|
|
/* The tree oracle doesn't like to have these. */
|
288 |
|
|
if (TREE_CODE (base) == FUNCTION_DECL
|
289 |
|
|
|| TREE_CODE (base) == LABEL_DECL)
|
290 |
|
|
return false;
|
291 |
|
|
|
292 |
|
|
/* If this is a reference based on a partitioned decl replace the
|
293 |
|
|
base with an INDIRECT_REF of the pointer representative we
|
294 |
|
|
created during stack slot partitioning. */
|
295 |
|
|
if (TREE_CODE (base) == VAR_DECL
|
296 |
|
|
&& ! TREE_STATIC (base)
|
297 |
|
|
&& cfun->gimple_df->decls_to_pointers != NULL)
|
298 |
|
|
{
|
299 |
|
|
void *namep;
|
300 |
|
|
namep = pointer_map_contains (cfun->gimple_df->decls_to_pointers, base);
|
301 |
|
|
if (namep)
|
302 |
|
|
{
|
303 |
|
|
ref->base_alias_set = get_alias_set (base);
|
304 |
|
|
ref->base = build1 (INDIRECT_REF, TREE_TYPE (base), *(tree *)namep);
|
305 |
|
|
}
|
306 |
|
|
}
|
307 |
|
|
|
308 |
|
|
ref->ref_alias_set = MEM_ALIAS_SET (mem);
|
309 |
|
|
|
310 |
|
|
/* If the base decl is a parameter we can have negative MEM_OFFSET in
|
311 |
|
|
case of promoted subregs on bigendian targets. Trust the MEM_EXPR
|
312 |
|
|
here. */
|
313 |
|
|
if (INTVAL (MEM_OFFSET (mem)) < 0
|
314 |
|
|
&& ((INTVAL (MEM_SIZE (mem)) + INTVAL (MEM_OFFSET (mem)))
|
315 |
|
|
* BITS_PER_UNIT) == ref->size)
|
316 |
|
|
return true;
|
317 |
|
|
|
318 |
|
|
ref->offset += INTVAL (MEM_OFFSET (mem)) * BITS_PER_UNIT;
|
319 |
|
|
ref->size = INTVAL (MEM_SIZE (mem)) * BITS_PER_UNIT;
|
320 |
|
|
|
321 |
|
|
/* The MEM may extend into adjacent fields, so adjust max_size if
|
322 |
|
|
necessary. */
|
323 |
|
|
if (ref->max_size != -1
|
324 |
|
|
&& ref->size > ref->max_size)
|
325 |
|
|
ref->max_size = ref->size;
|
326 |
|
|
|
327 |
|
|
/* If MEM_OFFSET and MEM_SIZE get us outside of the base object of
|
328 |
|
|
the MEM_EXPR punt. This happens for STRICT_ALIGNMENT targets a lot. */
|
329 |
|
|
if (MEM_EXPR (mem) != get_spill_slot_decl (false)
|
330 |
|
|
&& (ref->offset < 0
|
331 |
|
|
|| (DECL_P (ref->base)
|
332 |
|
|
&& (!host_integerp (DECL_SIZE (ref->base), 1)
|
333 |
|
|
|| (TREE_INT_CST_LOW (DECL_SIZE ((ref->base)))
|
334 |
|
|
< (unsigned HOST_WIDE_INT)(ref->offset + ref->size))))))
|
335 |
|
|
return false;
|
336 |
|
|
|
337 |
|
|
return true;
|
338 |
|
|
}
|
339 |
|
|
|
340 |
|
|
/* Query the alias-oracle on whether the two memory rtx X and MEM may
|
341 |
|
|
alias. If TBAA_P is set also apply TBAA. Returns true if the
|
342 |
|
|
two rtxen may alias, false otherwise. */
|
343 |
|
|
|
344 |
|
|
static bool
|
345 |
|
|
rtx_refs_may_alias_p (const_rtx x, const_rtx mem, bool tbaa_p)
|
346 |
|
|
{
|
347 |
|
|
ao_ref ref1, ref2;
|
348 |
|
|
|
349 |
|
|
if (!ao_ref_from_mem (&ref1, x)
|
350 |
|
|
|| !ao_ref_from_mem (&ref2, mem))
|
351 |
|
|
return true;
|
352 |
|
|
|
353 |
|
|
return refs_may_alias_p_1 (&ref1, &ref2, tbaa_p);
|
354 |
|
|
}
|
355 |
|
|
|
356 |
|
|
/* Returns a pointer to the alias set entry for ALIAS_SET, if there is
|
357 |
|
|
such an entry, or NULL otherwise. */
|
358 |
|
|
|
359 |
|
|
static inline alias_set_entry
|
360 |
|
|
get_alias_set_entry (alias_set_type alias_set)
|
361 |
|
|
{
|
362 |
|
|
return VEC_index (alias_set_entry, alias_sets, alias_set);
|
363 |
|
|
}
|
364 |
|
|
|
365 |
|
|
/* Returns nonzero if the alias sets for MEM1 and MEM2 are such that
|
366 |
|
|
the two MEMs cannot alias each other. */
|
367 |
|
|
|
368 |
|
|
static inline int
|
369 |
|
|
mems_in_disjoint_alias_sets_p (const_rtx mem1, const_rtx mem2)
|
370 |
|
|
{
|
371 |
|
|
/* Perform a basic sanity check. Namely, that there are no alias sets
|
372 |
|
|
if we're not using strict aliasing. This helps to catch bugs
|
373 |
|
|
whereby someone uses PUT_CODE, but doesn't clear MEM_ALIAS_SET, or
|
374 |
|
|
where a MEM is allocated in some way other than by the use of
|
375 |
|
|
gen_rtx_MEM, and the MEM_ALIAS_SET is not cleared. If we begin to
|
376 |
|
|
use alias sets to indicate that spilled registers cannot alias each
|
377 |
|
|
other, we might need to remove this check. */
|
378 |
|
|
gcc_assert (flag_strict_aliasing
|
379 |
|
|
|| (!MEM_ALIAS_SET (mem1) && !MEM_ALIAS_SET (mem2)));
|
380 |
|
|
|
381 |
|
|
return ! alias_sets_conflict_p (MEM_ALIAS_SET (mem1), MEM_ALIAS_SET (mem2));
|
382 |
|
|
}
|
383 |
|
|
|
384 |
|
|
/* Insert the NODE into the splay tree given by DATA. Used by
|
385 |
|
|
record_alias_subset via splay_tree_foreach. */
|
386 |
|
|
|
387 |
|
|
static int
|
388 |
|
|
insert_subset_children (splay_tree_node node, void *data)
|
389 |
|
|
{
|
390 |
|
|
splay_tree_insert ((splay_tree) data, node->key, node->value);
|
391 |
|
|
|
392 |
|
|
return 0;
|
393 |
|
|
}
|
394 |
|
|
|
395 |
|
|
/* Return true if the first alias set is a subset of the second. */
|
396 |
|
|
|
397 |
|
|
bool
|
398 |
|
|
alias_set_subset_of (alias_set_type set1, alias_set_type set2)
|
399 |
|
|
{
|
400 |
|
|
alias_set_entry ase;
|
401 |
|
|
|
402 |
|
|
/* Everything is a subset of the "aliases everything" set. */
|
403 |
|
|
if (set2 == 0)
|
404 |
|
|
return true;
|
405 |
|
|
|
406 |
|
|
/* Otherwise, check if set1 is a subset of set2. */
|
407 |
|
|
ase = get_alias_set_entry (set2);
|
408 |
|
|
if (ase != 0
|
409 |
|
|
&& (ase->has_zero_child
|
410 |
|
|
|| splay_tree_lookup (ase->children,
|
411 |
|
|
(splay_tree_key) set1)))
|
412 |
|
|
return true;
|
413 |
|
|
return false;
|
414 |
|
|
}
|
415 |
|
|
|
416 |
|
|
/* Return 1 if the two specified alias sets may conflict. */
|
417 |
|
|
|
418 |
|
|
int
|
419 |
|
|
alias_sets_conflict_p (alias_set_type set1, alias_set_type set2)
|
420 |
|
|
{
|
421 |
|
|
alias_set_entry ase;
|
422 |
|
|
|
423 |
|
|
/* The easy case. */
|
424 |
|
|
if (alias_sets_must_conflict_p (set1, set2))
|
425 |
|
|
return 1;
|
426 |
|
|
|
427 |
|
|
/* See if the first alias set is a subset of the second. */
|
428 |
|
|
ase = get_alias_set_entry (set1);
|
429 |
|
|
if (ase != 0
|
430 |
|
|
&& (ase->has_zero_child
|
431 |
|
|
|| splay_tree_lookup (ase->children,
|
432 |
|
|
(splay_tree_key) set2)))
|
433 |
|
|
return 1;
|
434 |
|
|
|
435 |
|
|
/* Now do the same, but with the alias sets reversed. */
|
436 |
|
|
ase = get_alias_set_entry (set2);
|
437 |
|
|
if (ase != 0
|
438 |
|
|
&& (ase->has_zero_child
|
439 |
|
|
|| splay_tree_lookup (ase->children,
|
440 |
|
|
(splay_tree_key) set1)))
|
441 |
|
|
return 1;
|
442 |
|
|
|
443 |
|
|
/* The two alias sets are distinct and neither one is the
|
444 |
|
|
child of the other. Therefore, they cannot conflict. */
|
445 |
|
|
return 0;
|
446 |
|
|
}
|
447 |
|
|
|
448 |
|
|
static int
|
449 |
|
|
walk_mems_2 (rtx *x, rtx mem)
|
450 |
|
|
{
|
451 |
|
|
if (MEM_P (*x))
|
452 |
|
|
{
|
453 |
|
|
if (alias_sets_conflict_p (MEM_ALIAS_SET(*x), MEM_ALIAS_SET(mem)))
|
454 |
|
|
return 1;
|
455 |
|
|
|
456 |
|
|
return -1;
|
457 |
|
|
}
|
458 |
|
|
return 0;
|
459 |
|
|
}
|
460 |
|
|
|
461 |
|
|
static int
|
462 |
|
|
walk_mems_1 (rtx *x, rtx *pat)
|
463 |
|
|
{
|
464 |
|
|
if (MEM_P (*x))
|
465 |
|
|
{
|
466 |
|
|
/* Visit all MEMs in *PAT and check indepedence. */
|
467 |
|
|
if (for_each_rtx (pat, (rtx_function) walk_mems_2, *x))
|
468 |
|
|
/* Indicate that dependence was determined and stop traversal. */
|
469 |
|
|
return 1;
|
470 |
|
|
|
471 |
|
|
return -1;
|
472 |
|
|
}
|
473 |
|
|
return 0;
|
474 |
|
|
}
|
475 |
|
|
|
476 |
|
|
/* Return 1 if two specified instructions have mem expr with conflict alias sets*/
|
477 |
|
|
bool
|
478 |
|
|
insn_alias_sets_conflict_p (rtx insn1, rtx insn2)
|
479 |
|
|
{
|
480 |
|
|
/* For each pair of MEMs in INSN1 and INSN2 check their independence. */
|
481 |
|
|
return for_each_rtx (&PATTERN (insn1), (rtx_function) walk_mems_1,
|
482 |
|
|
&PATTERN (insn2));
|
483 |
|
|
}
|
484 |
|
|
|
485 |
|
|
/* Return 1 if the two specified alias sets will always conflict. */
|
486 |
|
|
|
487 |
|
|
int
|
488 |
|
|
alias_sets_must_conflict_p (alias_set_type set1, alias_set_type set2)
|
489 |
|
|
{
|
490 |
|
|
if (set1 == 0 || set2 == 0 || set1 == set2)
|
491 |
|
|
return 1;
|
492 |
|
|
|
493 |
|
|
return 0;
|
494 |
|
|
}
|
495 |
|
|
|
496 |
|
|
/* Return 1 if any MEM object of type T1 will always conflict (using the
|
497 |
|
|
dependency routines in this file) with any MEM object of type T2.
|
498 |
|
|
This is used when allocating temporary storage. If T1 and/or T2 are
|
499 |
|
|
NULL_TREE, it means we know nothing about the storage. */
|
500 |
|
|
|
501 |
|
|
int
|
502 |
|
|
objects_must_conflict_p (tree t1, tree t2)
|
503 |
|
|
{
|
504 |
|
|
alias_set_type set1, set2;
|
505 |
|
|
|
506 |
|
|
/* If neither has a type specified, we don't know if they'll conflict
|
507 |
|
|
because we may be using them to store objects of various types, for
|
508 |
|
|
example the argument and local variables areas of inlined functions. */
|
509 |
|
|
if (t1 == 0 && t2 == 0)
|
510 |
|
|
return 0;
|
511 |
|
|
|
512 |
|
|
/* If they are the same type, they must conflict. */
|
513 |
|
|
if (t1 == t2
|
514 |
|
|
/* Likewise if both are volatile. */
|
515 |
|
|
|| (t1 != 0 && TYPE_VOLATILE (t1) && t2 != 0 && TYPE_VOLATILE (t2)))
|
516 |
|
|
return 1;
|
517 |
|
|
|
518 |
|
|
set1 = t1 ? get_alias_set (t1) : 0;
|
519 |
|
|
set2 = t2 ? get_alias_set (t2) : 0;
|
520 |
|
|
|
521 |
|
|
/* We can't use alias_sets_conflict_p because we must make sure
|
522 |
|
|
that every subtype of t1 will conflict with every subtype of
|
523 |
|
|
t2 for which a pair of subobjects of these respective subtypes
|
524 |
|
|
overlaps on the stack. */
|
525 |
|
|
return alias_sets_must_conflict_p (set1, set2);
|
526 |
|
|
}
|
527 |
|
|
|
528 |
|
|
/* Return true if all nested component references handled by
|
529 |
|
|
get_inner_reference in T are such that we should use the alias set
|
530 |
|
|
provided by the object at the heart of T.
|
531 |
|
|
|
532 |
|
|
This is true for non-addressable components (which don't have their
|
533 |
|
|
own alias set), as well as components of objects in alias set zero.
|
534 |
|
|
This later point is a special case wherein we wish to override the
|
535 |
|
|
alias set used by the component, but we don't have per-FIELD_DECL
|
536 |
|
|
assignable alias sets. */
|
537 |
|
|
|
538 |
|
|
bool
|
539 |
|
|
component_uses_parent_alias_set (const_tree t)
|
540 |
|
|
{
|
541 |
|
|
while (1)
|
542 |
|
|
{
|
543 |
|
|
/* If we're at the end, it vacuously uses its own alias set. */
|
544 |
|
|
if (!handled_component_p (t))
|
545 |
|
|
return false;
|
546 |
|
|
|
547 |
|
|
switch (TREE_CODE (t))
|
548 |
|
|
{
|
549 |
|
|
case COMPONENT_REF:
|
550 |
|
|
if (DECL_NONADDRESSABLE_P (TREE_OPERAND (t, 1)))
|
551 |
|
|
return true;
|
552 |
|
|
break;
|
553 |
|
|
|
554 |
|
|
case ARRAY_REF:
|
555 |
|
|
case ARRAY_RANGE_REF:
|
556 |
|
|
if (TYPE_NONALIASED_COMPONENT (TREE_TYPE (TREE_OPERAND (t, 0))))
|
557 |
|
|
return true;
|
558 |
|
|
break;
|
559 |
|
|
|
560 |
|
|
case REALPART_EXPR:
|
561 |
|
|
case IMAGPART_EXPR:
|
562 |
|
|
break;
|
563 |
|
|
|
564 |
|
|
default:
|
565 |
|
|
/* Bitfields and casts are never addressable. */
|
566 |
|
|
return true;
|
567 |
|
|
}
|
568 |
|
|
|
569 |
|
|
t = TREE_OPERAND (t, 0);
|
570 |
|
|
if (get_alias_set (TREE_TYPE (t)) == 0)
|
571 |
|
|
return true;
|
572 |
|
|
}
|
573 |
|
|
}
|
574 |
|
|
|
575 |
|
|
/* Return the alias set for the memory pointed to by T, which may be
|
576 |
|
|
either a type or an expression. Return -1 if there is nothing
|
577 |
|
|
special about dereferencing T. */
|
578 |
|
|
|
579 |
|
|
static alias_set_type
|
580 |
|
|
get_deref_alias_set_1 (tree t)
|
581 |
|
|
{
|
582 |
|
|
/* If we're not doing any alias analysis, just assume everything
|
583 |
|
|
aliases everything else. */
|
584 |
|
|
if (!flag_strict_aliasing)
|
585 |
|
|
return 0;
|
586 |
|
|
|
587 |
|
|
/* All we care about is the type. */
|
588 |
|
|
if (! TYPE_P (t))
|
589 |
|
|
t = TREE_TYPE (t);
|
590 |
|
|
|
591 |
|
|
/* If we have an INDIRECT_REF via a void pointer, we don't
|
592 |
|
|
know anything about what that might alias. Likewise if the
|
593 |
|
|
pointer is marked that way. */
|
594 |
|
|
if (TREE_CODE (TREE_TYPE (t)) == VOID_TYPE
|
595 |
|
|
|| TYPE_REF_CAN_ALIAS_ALL (t))
|
596 |
|
|
return 0;
|
597 |
|
|
|
598 |
|
|
return -1;
|
599 |
|
|
}
|
600 |
|
|
|
601 |
|
|
/* Return the alias set for the memory pointed to by T, which may be
|
602 |
|
|
either a type or an expression. */
|
603 |
|
|
|
604 |
|
|
alias_set_type
|
605 |
|
|
get_deref_alias_set (tree t)
|
606 |
|
|
{
|
607 |
|
|
alias_set_type set = get_deref_alias_set_1 (t);
|
608 |
|
|
|
609 |
|
|
/* Fall back to the alias-set of the pointed-to type. */
|
610 |
|
|
if (set == -1)
|
611 |
|
|
{
|
612 |
|
|
if (! TYPE_P (t))
|
613 |
|
|
t = TREE_TYPE (t);
|
614 |
|
|
set = get_alias_set (TREE_TYPE (t));
|
615 |
|
|
}
|
616 |
|
|
|
617 |
|
|
return set;
|
618 |
|
|
}
|
619 |
|
|
|
620 |
|
|
/* Return the alias set for T, which may be either a type or an
|
621 |
|
|
expression. Call language-specific routine for help, if needed. */
|
622 |
|
|
|
623 |
|
|
alias_set_type
|
624 |
|
|
get_alias_set (tree t)
|
625 |
|
|
{
|
626 |
|
|
alias_set_type set;
|
627 |
|
|
|
628 |
|
|
/* If we're not doing any alias analysis, just assume everything
|
629 |
|
|
aliases everything else. Also return 0 if this or its type is
|
630 |
|
|
an error. */
|
631 |
|
|
if (! flag_strict_aliasing || t == error_mark_node
|
632 |
|
|
|| (! TYPE_P (t)
|
633 |
|
|
&& (TREE_TYPE (t) == 0 || TREE_TYPE (t) == error_mark_node)))
|
634 |
|
|
return 0;
|
635 |
|
|
|
636 |
|
|
/* We can be passed either an expression or a type. This and the
|
637 |
|
|
language-specific routine may make mutually-recursive calls to each other
|
638 |
|
|
to figure out what to do. At each juncture, we see if this is a tree
|
639 |
|
|
that the language may need to handle specially. First handle things that
|
640 |
|
|
aren't types. */
|
641 |
|
|
if (! TYPE_P (t))
|
642 |
|
|
{
|
643 |
|
|
tree inner;
|
644 |
|
|
|
645 |
|
|
/* Remove any nops, then give the language a chance to do
|
646 |
|
|
something with this tree before we look at it. */
|
647 |
|
|
STRIP_NOPS (t);
|
648 |
|
|
set = lang_hooks.get_alias_set (t);
|
649 |
|
|
if (set != -1)
|
650 |
|
|
return set;
|
651 |
|
|
|
652 |
|
|
/* Retrieve the original memory reference if needed. */
|
653 |
|
|
if (TREE_CODE (t) == TARGET_MEM_REF)
|
654 |
|
|
t = TMR_ORIGINAL (t);
|
655 |
|
|
|
656 |
|
|
/* First see if the actual object referenced is an INDIRECT_REF from a
|
657 |
|
|
restrict-qualified pointer or a "void *". */
|
658 |
|
|
inner = t;
|
659 |
|
|
while (handled_component_p (inner))
|
660 |
|
|
{
|
661 |
|
|
inner = TREE_OPERAND (inner, 0);
|
662 |
|
|
STRIP_NOPS (inner);
|
663 |
|
|
}
|
664 |
|
|
|
665 |
|
|
if (INDIRECT_REF_P (inner))
|
666 |
|
|
{
|
667 |
|
|
set = get_deref_alias_set_1 (TREE_OPERAND (inner, 0));
|
668 |
|
|
if (set != -1)
|
669 |
|
|
return set;
|
670 |
|
|
}
|
671 |
|
|
|
672 |
|
|
/* Otherwise, pick up the outermost object that we could have a pointer
|
673 |
|
|
to, processing conversions as above. */
|
674 |
|
|
while (component_uses_parent_alias_set (t))
|
675 |
|
|
{
|
676 |
|
|
t = TREE_OPERAND (t, 0);
|
677 |
|
|
STRIP_NOPS (t);
|
678 |
|
|
}
|
679 |
|
|
|
680 |
|
|
/* If we've already determined the alias set for a decl, just return
|
681 |
|
|
it. This is necessary for C++ anonymous unions, whose component
|
682 |
|
|
variables don't look like union members (boo!). */
|
683 |
|
|
if (TREE_CODE (t) == VAR_DECL
|
684 |
|
|
&& DECL_RTL_SET_P (t) && MEM_P (DECL_RTL (t)))
|
685 |
|
|
return MEM_ALIAS_SET (DECL_RTL (t));
|
686 |
|
|
|
687 |
|
|
/* Now all we care about is the type. */
|
688 |
|
|
t = TREE_TYPE (t);
|
689 |
|
|
}
|
690 |
|
|
|
691 |
|
|
/* Variant qualifiers don't affect the alias set, so get the main
|
692 |
|
|
variant. */
|
693 |
|
|
t = TYPE_MAIN_VARIANT (t);
|
694 |
|
|
|
695 |
|
|
/* Always use the canonical type as well. If this is a type that
|
696 |
|
|
requires structural comparisons to identify compatible types
|
697 |
|
|
use alias set zero. */
|
698 |
|
|
if (TYPE_STRUCTURAL_EQUALITY_P (t))
|
699 |
|
|
{
|
700 |
|
|
/* Allow the language to specify another alias set for this
|
701 |
|
|
type. */
|
702 |
|
|
set = lang_hooks.get_alias_set (t);
|
703 |
|
|
if (set != -1)
|
704 |
|
|
return set;
|
705 |
|
|
return 0;
|
706 |
|
|
}
|
707 |
|
|
t = TYPE_CANONICAL (t);
|
708 |
|
|
/* Canonical types shouldn't form a tree nor should the canonical
|
709 |
|
|
type require structural equality checks. */
|
710 |
|
|
gcc_assert (!TYPE_STRUCTURAL_EQUALITY_P (t) && TYPE_CANONICAL (t) == t);
|
711 |
|
|
|
712 |
|
|
/* If this is a type with a known alias set, return it. */
|
713 |
|
|
if (TYPE_ALIAS_SET_KNOWN_P (t))
|
714 |
|
|
return TYPE_ALIAS_SET (t);
|
715 |
|
|
|
716 |
|
|
/* We don't want to set TYPE_ALIAS_SET for incomplete types. */
|
717 |
|
|
if (!COMPLETE_TYPE_P (t))
|
718 |
|
|
{
|
719 |
|
|
/* For arrays with unknown size the conservative answer is the
|
720 |
|
|
alias set of the element type. */
|
721 |
|
|
if (TREE_CODE (t) == ARRAY_TYPE)
|
722 |
|
|
return get_alias_set (TREE_TYPE (t));
|
723 |
|
|
|
724 |
|
|
/* But return zero as a conservative answer for incomplete types. */
|
725 |
|
|
return 0;
|
726 |
|
|
}
|
727 |
|
|
|
728 |
|
|
/* See if the language has special handling for this type. */
|
729 |
|
|
set = lang_hooks.get_alias_set (t);
|
730 |
|
|
if (set != -1)
|
731 |
|
|
return set;
|
732 |
|
|
|
733 |
|
|
/* There are no objects of FUNCTION_TYPE, so there's no point in
|
734 |
|
|
using up an alias set for them. (There are, of course, pointers
|
735 |
|
|
and references to functions, but that's different.) */
|
736 |
|
|
else if (TREE_CODE (t) == FUNCTION_TYPE
|
737 |
|
|
|| TREE_CODE (t) == METHOD_TYPE)
|
738 |
|
|
set = 0;
|
739 |
|
|
|
740 |
|
|
/* Unless the language specifies otherwise, let vector types alias
|
741 |
|
|
their components. This avoids some nasty type punning issues in
|
742 |
|
|
normal usage. And indeed lets vectors be treated more like an
|
743 |
|
|
array slice. */
|
744 |
|
|
else if (TREE_CODE (t) == VECTOR_TYPE)
|
745 |
|
|
set = get_alias_set (TREE_TYPE (t));
|
746 |
|
|
|
747 |
|
|
/* Unless the language specifies otherwise, treat array types the
|
748 |
|
|
same as their components. This avoids the asymmetry we get
|
749 |
|
|
through recording the components. Consider accessing a
|
750 |
|
|
character(kind=1) through a reference to a character(kind=1)[1:1].
|
751 |
|
|
Or consider if we want to assign integer(kind=4)[0:D.1387] and
|
752 |
|
|
integer(kind=4)[4] the same alias set or not.
|
753 |
|
|
Just be pragmatic here and make sure the array and its element
|
754 |
|
|
type get the same alias set assigned. */
|
755 |
|
|
else if (TREE_CODE (t) == ARRAY_TYPE
|
756 |
|
|
&& !TYPE_NONALIASED_COMPONENT (t))
|
757 |
|
|
set = get_alias_set (TREE_TYPE (t));
|
758 |
|
|
|
759 |
|
|
else
|
760 |
|
|
/* Otherwise make a new alias set for this type. */
|
761 |
|
|
set = new_alias_set ();
|
762 |
|
|
|
763 |
|
|
TYPE_ALIAS_SET (t) = set;
|
764 |
|
|
|
765 |
|
|
/* If this is an aggregate type, we must record any component aliasing
|
766 |
|
|
information. */
|
767 |
|
|
if (AGGREGATE_TYPE_P (t) || TREE_CODE (t) == COMPLEX_TYPE)
|
768 |
|
|
record_component_aliases (t);
|
769 |
|
|
|
770 |
|
|
return set;
|
771 |
|
|
}
|
772 |
|
|
|
773 |
|
|
/* Return a brand-new alias set. */
|
774 |
|
|
|
775 |
|
|
alias_set_type
|
776 |
|
|
new_alias_set (void)
|
777 |
|
|
{
|
778 |
|
|
if (flag_strict_aliasing)
|
779 |
|
|
{
|
780 |
|
|
if (alias_sets == 0)
|
781 |
|
|
VEC_safe_push (alias_set_entry, gc, alias_sets, 0);
|
782 |
|
|
VEC_safe_push (alias_set_entry, gc, alias_sets, 0);
|
783 |
|
|
return VEC_length (alias_set_entry, alias_sets) - 1;
|
784 |
|
|
}
|
785 |
|
|
else
|
786 |
|
|
return 0;
|
787 |
|
|
}
|
788 |
|
|
|
789 |
|
|
/* Indicate that things in SUBSET can alias things in SUPERSET, but that
|
790 |
|
|
not everything that aliases SUPERSET also aliases SUBSET. For example,
|
791 |
|
|
in C, a store to an `int' can alias a load of a structure containing an
|
792 |
|
|
`int', and vice versa. But it can't alias a load of a 'double' member
|
793 |
|
|
of the same structure. Here, the structure would be the SUPERSET and
|
794 |
|
|
`int' the SUBSET. This relationship is also described in the comment at
|
795 |
|
|
the beginning of this file.
|
796 |
|
|
|
797 |
|
|
This function should be called only once per SUPERSET/SUBSET pair.
|
798 |
|
|
|
799 |
|
|
It is illegal for SUPERSET to be zero; everything is implicitly a
|
800 |
|
|
subset of alias set zero. */
|
801 |
|
|
|
802 |
|
|
void
|
803 |
|
|
record_alias_subset (alias_set_type superset, alias_set_type subset)
|
804 |
|
|
{
|
805 |
|
|
alias_set_entry superset_entry;
|
806 |
|
|
alias_set_entry subset_entry;
|
807 |
|
|
|
808 |
|
|
/* It is possible in complex type situations for both sets to be the same,
|
809 |
|
|
in which case we can ignore this operation. */
|
810 |
|
|
if (superset == subset)
|
811 |
|
|
return;
|
812 |
|
|
|
813 |
|
|
gcc_assert (superset);
|
814 |
|
|
|
815 |
|
|
superset_entry = get_alias_set_entry (superset);
|
816 |
|
|
if (superset_entry == 0)
|
817 |
|
|
{
|
818 |
|
|
/* Create an entry for the SUPERSET, so that we have a place to
|
819 |
|
|
attach the SUBSET. */
|
820 |
|
|
superset_entry = GGC_NEW (struct alias_set_entry_d);
|
821 |
|
|
superset_entry->alias_set = superset;
|
822 |
|
|
superset_entry->children
|
823 |
|
|
= splay_tree_new_ggc (splay_tree_compare_ints);
|
824 |
|
|
superset_entry->has_zero_child = 0;
|
825 |
|
|
VEC_replace (alias_set_entry, alias_sets, superset, superset_entry);
|
826 |
|
|
}
|
827 |
|
|
|
828 |
|
|
if (subset == 0)
|
829 |
|
|
superset_entry->has_zero_child = 1;
|
830 |
|
|
else
|
831 |
|
|
{
|
832 |
|
|
subset_entry = get_alias_set_entry (subset);
|
833 |
|
|
/* If there is an entry for the subset, enter all of its children
|
834 |
|
|
(if they are not already present) as children of the SUPERSET. */
|
835 |
|
|
if (subset_entry)
|
836 |
|
|
{
|
837 |
|
|
if (subset_entry->has_zero_child)
|
838 |
|
|
superset_entry->has_zero_child = 1;
|
839 |
|
|
|
840 |
|
|
splay_tree_foreach (subset_entry->children, insert_subset_children,
|
841 |
|
|
superset_entry->children);
|
842 |
|
|
}
|
843 |
|
|
|
844 |
|
|
/* Enter the SUBSET itself as a child of the SUPERSET. */
|
845 |
|
|
splay_tree_insert (superset_entry->children,
|
846 |
|
|
(splay_tree_key) subset, 0);
|
847 |
|
|
}
|
848 |
|
|
}
|
849 |
|
|
|
850 |
|
|
/* Record that component types of TYPE, if any, are part of that type for
|
851 |
|
|
aliasing purposes. For record types, we only record component types
|
852 |
|
|
for fields that are not marked non-addressable. For array types, we
|
853 |
|
|
only record the component type if it is not marked non-aliased. */
|
854 |
|
|
|
855 |
|
|
void
|
856 |
|
|
record_component_aliases (tree type)
|
857 |
|
|
{
|
858 |
|
|
alias_set_type superset = get_alias_set (type);
|
859 |
|
|
tree field;
|
860 |
|
|
|
861 |
|
|
if (superset == 0)
|
862 |
|
|
return;
|
863 |
|
|
|
864 |
|
|
switch (TREE_CODE (type))
|
865 |
|
|
{
|
866 |
|
|
case RECORD_TYPE:
|
867 |
|
|
case UNION_TYPE:
|
868 |
|
|
case QUAL_UNION_TYPE:
|
869 |
|
|
/* Recursively record aliases for the base classes, if there are any. */
|
870 |
|
|
if (TYPE_BINFO (type))
|
871 |
|
|
{
|
872 |
|
|
int i;
|
873 |
|
|
tree binfo, base_binfo;
|
874 |
|
|
|
875 |
|
|
for (binfo = TYPE_BINFO (type), i = 0;
|
876 |
|
|
BINFO_BASE_ITERATE (binfo, i, base_binfo); i++)
|
877 |
|
|
record_alias_subset (superset,
|
878 |
|
|
get_alias_set (BINFO_TYPE (base_binfo)));
|
879 |
|
|
}
|
880 |
|
|
for (field = TYPE_FIELDS (type); field != 0; field = TREE_CHAIN (field))
|
881 |
|
|
if (TREE_CODE (field) == FIELD_DECL && !DECL_NONADDRESSABLE_P (field))
|
882 |
|
|
record_alias_subset (superset, get_alias_set (TREE_TYPE (field)));
|
883 |
|
|
break;
|
884 |
|
|
|
885 |
|
|
case COMPLEX_TYPE:
|
886 |
|
|
record_alias_subset (superset, get_alias_set (TREE_TYPE (type)));
|
887 |
|
|
break;
|
888 |
|
|
|
889 |
|
|
/* VECTOR_TYPE and ARRAY_TYPE share the alias set with their
|
890 |
|
|
element type. */
|
891 |
|
|
|
892 |
|
|
default:
|
893 |
|
|
break;
|
894 |
|
|
}
|
895 |
|
|
}
|
896 |
|
|
|
897 |
|
|
/* Allocate an alias set for use in storing and reading from the varargs
|
898 |
|
|
spill area. */
|
899 |
|
|
|
900 |
|
|
static GTY(()) alias_set_type varargs_set = -1;
|
901 |
|
|
|
902 |
|
|
alias_set_type
|
903 |
|
|
get_varargs_alias_set (void)
|
904 |
|
|
{
|
905 |
|
|
#if 1
|
906 |
|
|
/* We now lower VA_ARG_EXPR, and there's currently no way to attach the
|
907 |
|
|
varargs alias set to an INDIRECT_REF (FIXME!), so we can't
|
908 |
|
|
consistently use the varargs alias set for loads from the varargs
|
909 |
|
|
area. So don't use it anywhere. */
|
910 |
|
|
return 0;
|
911 |
|
|
#else
|
912 |
|
|
if (varargs_set == -1)
|
913 |
|
|
varargs_set = new_alias_set ();
|
914 |
|
|
|
915 |
|
|
return varargs_set;
|
916 |
|
|
#endif
|
917 |
|
|
}
|
918 |
|
|
|
919 |
|
|
/* Likewise, but used for the fixed portions of the frame, e.g., register
|
920 |
|
|
save areas. */
|
921 |
|
|
|
922 |
|
|
static GTY(()) alias_set_type frame_set = -1;
|
923 |
|
|
|
924 |
|
|
alias_set_type
|
925 |
|
|
get_frame_alias_set (void)
|
926 |
|
|
{
|
927 |
|
|
if (frame_set == -1)
|
928 |
|
|
frame_set = new_alias_set ();
|
929 |
|
|
|
930 |
|
|
return frame_set;
|
931 |
|
|
}
|
932 |
|
|
|
933 |
|
|
/* Inside SRC, the source of a SET, find a base address. */
|
934 |
|
|
|
935 |
|
|
static rtx
|
936 |
|
|
find_base_value (rtx src)
|
937 |
|
|
{
|
938 |
|
|
unsigned int regno;
|
939 |
|
|
|
940 |
|
|
#if defined (FIND_BASE_TERM)
|
941 |
|
|
/* Try machine-dependent ways to find the base term. */
|
942 |
|
|
src = FIND_BASE_TERM (src);
|
943 |
|
|
#endif
|
944 |
|
|
|
945 |
|
|
switch (GET_CODE (src))
|
946 |
|
|
{
|
947 |
|
|
case SYMBOL_REF:
|
948 |
|
|
case LABEL_REF:
|
949 |
|
|
return src;
|
950 |
|
|
|
951 |
|
|
case REG:
|
952 |
|
|
regno = REGNO (src);
|
953 |
|
|
/* At the start of a function, argument registers have known base
|
954 |
|
|
values which may be lost later. Returning an ADDRESS
|
955 |
|
|
expression here allows optimization based on argument values
|
956 |
|
|
even when the argument registers are used for other purposes. */
|
957 |
|
|
if (regno < FIRST_PSEUDO_REGISTER && copying_arguments)
|
958 |
|
|
return new_reg_base_value[regno];
|
959 |
|
|
|
960 |
|
|
/* If a pseudo has a known base value, return it. Do not do this
|
961 |
|
|
for non-fixed hard regs since it can result in a circular
|
962 |
|
|
dependency chain for registers which have values at function entry.
|
963 |
|
|
|
964 |
|
|
The test above is not sufficient because the scheduler may move
|
965 |
|
|
a copy out of an arg reg past the NOTE_INSN_FUNCTION_BEGIN. */
|
966 |
|
|
if ((regno >= FIRST_PSEUDO_REGISTER || fixed_regs[regno])
|
967 |
|
|
&& regno < VEC_length (rtx, reg_base_value))
|
968 |
|
|
{
|
969 |
|
|
/* If we're inside init_alias_analysis, use new_reg_base_value
|
970 |
|
|
to reduce the number of relaxation iterations. */
|
971 |
|
|
if (new_reg_base_value && new_reg_base_value[regno]
|
972 |
|
|
&& DF_REG_DEF_COUNT (regno) == 1)
|
973 |
|
|
return new_reg_base_value[regno];
|
974 |
|
|
|
975 |
|
|
if (VEC_index (rtx, reg_base_value, regno))
|
976 |
|
|
return VEC_index (rtx, reg_base_value, regno);
|
977 |
|
|
}
|
978 |
|
|
|
979 |
|
|
return 0;
|
980 |
|
|
|
981 |
|
|
case MEM:
|
982 |
|
|
/* Check for an argument passed in memory. Only record in the
|
983 |
|
|
copying-arguments block; it is too hard to track changes
|
984 |
|
|
otherwise. */
|
985 |
|
|
if (copying_arguments
|
986 |
|
|
&& (XEXP (src, 0) == arg_pointer_rtx
|
987 |
|
|
|| (GET_CODE (XEXP (src, 0)) == PLUS
|
988 |
|
|
&& XEXP (XEXP (src, 0), 0) == arg_pointer_rtx)))
|
989 |
|
|
return gen_rtx_ADDRESS (VOIDmode, src);
|
990 |
|
|
return 0;
|
991 |
|
|
|
992 |
|
|
case CONST:
|
993 |
|
|
src = XEXP (src, 0);
|
994 |
|
|
if (GET_CODE (src) != PLUS && GET_CODE (src) != MINUS)
|
995 |
|
|
break;
|
996 |
|
|
|
997 |
|
|
/* ... fall through ... */
|
998 |
|
|
|
999 |
|
|
case PLUS:
|
1000 |
|
|
case MINUS:
|
1001 |
|
|
{
|
1002 |
|
|
rtx temp, src_0 = XEXP (src, 0), src_1 = XEXP (src, 1);
|
1003 |
|
|
|
1004 |
|
|
/* If either operand is a REG that is a known pointer, then it
|
1005 |
|
|
is the base. */
|
1006 |
|
|
if (REG_P (src_0) && REG_POINTER (src_0))
|
1007 |
|
|
return find_base_value (src_0);
|
1008 |
|
|
if (REG_P (src_1) && REG_POINTER (src_1))
|
1009 |
|
|
return find_base_value (src_1);
|
1010 |
|
|
|
1011 |
|
|
/* If either operand is a REG, then see if we already have
|
1012 |
|
|
a known value for it. */
|
1013 |
|
|
if (REG_P (src_0))
|
1014 |
|
|
{
|
1015 |
|
|
temp = find_base_value (src_0);
|
1016 |
|
|
if (temp != 0)
|
1017 |
|
|
src_0 = temp;
|
1018 |
|
|
}
|
1019 |
|
|
|
1020 |
|
|
if (REG_P (src_1))
|
1021 |
|
|
{
|
1022 |
|
|
temp = find_base_value (src_1);
|
1023 |
|
|
if (temp!= 0)
|
1024 |
|
|
src_1 = temp;
|
1025 |
|
|
}
|
1026 |
|
|
|
1027 |
|
|
/* If either base is named object or a special address
|
1028 |
|
|
(like an argument or stack reference), then use it for the
|
1029 |
|
|
base term. */
|
1030 |
|
|
if (src_0 != 0
|
1031 |
|
|
&& (GET_CODE (src_0) == SYMBOL_REF
|
1032 |
|
|
|| GET_CODE (src_0) == LABEL_REF
|
1033 |
|
|
|| (GET_CODE (src_0) == ADDRESS
|
1034 |
|
|
&& GET_MODE (src_0) != VOIDmode)))
|
1035 |
|
|
return src_0;
|
1036 |
|
|
|
1037 |
|
|
if (src_1 != 0
|
1038 |
|
|
&& (GET_CODE (src_1) == SYMBOL_REF
|
1039 |
|
|
|| GET_CODE (src_1) == LABEL_REF
|
1040 |
|
|
|| (GET_CODE (src_1) == ADDRESS
|
1041 |
|
|
&& GET_MODE (src_1) != VOIDmode)))
|
1042 |
|
|
return src_1;
|
1043 |
|
|
|
1044 |
|
|
/* Guess which operand is the base address:
|
1045 |
|
|
If either operand is a symbol, then it is the base. If
|
1046 |
|
|
either operand is a CONST_INT, then the other is the base. */
|
1047 |
|
|
if (CONST_INT_P (src_1) || CONSTANT_P (src_0))
|
1048 |
|
|
return find_base_value (src_0);
|
1049 |
|
|
else if (CONST_INT_P (src_0) || CONSTANT_P (src_1))
|
1050 |
|
|
return find_base_value (src_1);
|
1051 |
|
|
|
1052 |
|
|
return 0;
|
1053 |
|
|
}
|
1054 |
|
|
|
1055 |
|
|
case LO_SUM:
|
1056 |
|
|
/* The standard form is (lo_sum reg sym) so look only at the
|
1057 |
|
|
second operand. */
|
1058 |
|
|
return find_base_value (XEXP (src, 1));
|
1059 |
|
|
|
1060 |
|
|
case AND:
|
1061 |
|
|
/* If the second operand is constant set the base
|
1062 |
|
|
address to the first operand. */
|
1063 |
|
|
if (CONST_INT_P (XEXP (src, 1)) && INTVAL (XEXP (src, 1)) != 0)
|
1064 |
|
|
return find_base_value (XEXP (src, 0));
|
1065 |
|
|
return 0;
|
1066 |
|
|
|
1067 |
|
|
case TRUNCATE:
|
1068 |
|
|
/* As we do not know which address space the pointer is refering to, we can
|
1069 |
|
|
handle this only if the target does not support different pointer or
|
1070 |
|
|
address modes depending on the address space. */
|
1071 |
|
|
if (!target_default_pointer_address_modes_p ())
|
1072 |
|
|
break;
|
1073 |
|
|
if (GET_MODE_SIZE (GET_MODE (src)) < GET_MODE_SIZE (Pmode))
|
1074 |
|
|
break;
|
1075 |
|
|
/* Fall through. */
|
1076 |
|
|
case HIGH:
|
1077 |
|
|
case PRE_INC:
|
1078 |
|
|
case PRE_DEC:
|
1079 |
|
|
case POST_INC:
|
1080 |
|
|
case POST_DEC:
|
1081 |
|
|
case PRE_MODIFY:
|
1082 |
|
|
case POST_MODIFY:
|
1083 |
|
|
return find_base_value (XEXP (src, 0));
|
1084 |
|
|
|
1085 |
|
|
case ZERO_EXTEND:
|
1086 |
|
|
case SIGN_EXTEND: /* used for NT/Alpha pointers */
|
1087 |
|
|
/* As we do not know which address space the pointer is refering to, we can
|
1088 |
|
|
handle this only if the target does not support different pointer or
|
1089 |
|
|
address modes depending on the address space. */
|
1090 |
|
|
if (!target_default_pointer_address_modes_p ())
|
1091 |
|
|
break;
|
1092 |
|
|
|
1093 |
|
|
{
|
1094 |
|
|
rtx temp = find_base_value (XEXP (src, 0));
|
1095 |
|
|
|
1096 |
|
|
if (temp != 0 && CONSTANT_P (temp))
|
1097 |
|
|
temp = convert_memory_address (Pmode, temp);
|
1098 |
|
|
|
1099 |
|
|
return temp;
|
1100 |
|
|
}
|
1101 |
|
|
|
1102 |
|
|
default:
|
1103 |
|
|
break;
|
1104 |
|
|
}
|
1105 |
|
|
|
1106 |
|
|
return 0;
|
1107 |
|
|
}
|
1108 |
|
|
|
1109 |
|
|
/* Called from init_alias_analysis indirectly through note_stores. */
|
1110 |
|
|
|
1111 |
|
|
/* While scanning insns to find base values, reg_seen[N] is nonzero if
|
1112 |
|
|
register N has been set in this function. */
|
1113 |
|
|
static char *reg_seen;
|
1114 |
|
|
|
1115 |
|
|
/* Addresses which are known not to alias anything else are identified
|
1116 |
|
|
by a unique integer. */
|
1117 |
|
|
static int unique_id;
|
1118 |
|
|
|
1119 |
|
|
static void
|
1120 |
|
|
record_set (rtx dest, const_rtx set, void *data ATTRIBUTE_UNUSED)
|
1121 |
|
|
{
|
1122 |
|
|
unsigned regno;
|
1123 |
|
|
rtx src;
|
1124 |
|
|
int n;
|
1125 |
|
|
|
1126 |
|
|
if (!REG_P (dest))
|
1127 |
|
|
return;
|
1128 |
|
|
|
1129 |
|
|
regno = REGNO (dest);
|
1130 |
|
|
|
1131 |
|
|
gcc_assert (regno < VEC_length (rtx, reg_base_value));
|
1132 |
|
|
|
1133 |
|
|
/* If this spans multiple hard registers, then we must indicate that every
|
1134 |
|
|
register has an unusable value. */
|
1135 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
1136 |
|
|
n = hard_regno_nregs[regno][GET_MODE (dest)];
|
1137 |
|
|
else
|
1138 |
|
|
n = 1;
|
1139 |
|
|
if (n != 1)
|
1140 |
|
|
{
|
1141 |
|
|
while (--n >= 0)
|
1142 |
|
|
{
|
1143 |
|
|
reg_seen[regno + n] = 1;
|
1144 |
|
|
new_reg_base_value[regno + n] = 0;
|
1145 |
|
|
}
|
1146 |
|
|
return;
|
1147 |
|
|
}
|
1148 |
|
|
|
1149 |
|
|
if (set)
|
1150 |
|
|
{
|
1151 |
|
|
/* A CLOBBER wipes out any old value but does not prevent a previously
|
1152 |
|
|
unset register from acquiring a base address (i.e. reg_seen is not
|
1153 |
|
|
set). */
|
1154 |
|
|
if (GET_CODE (set) == CLOBBER)
|
1155 |
|
|
{
|
1156 |
|
|
new_reg_base_value[regno] = 0;
|
1157 |
|
|
return;
|
1158 |
|
|
}
|
1159 |
|
|
src = SET_SRC (set);
|
1160 |
|
|
}
|
1161 |
|
|
else
|
1162 |
|
|
{
|
1163 |
|
|
if (reg_seen[regno])
|
1164 |
|
|
{
|
1165 |
|
|
new_reg_base_value[regno] = 0;
|
1166 |
|
|
return;
|
1167 |
|
|
}
|
1168 |
|
|
reg_seen[regno] = 1;
|
1169 |
|
|
new_reg_base_value[regno] = gen_rtx_ADDRESS (Pmode,
|
1170 |
|
|
GEN_INT (unique_id++));
|
1171 |
|
|
return;
|
1172 |
|
|
}
|
1173 |
|
|
|
1174 |
|
|
/* If this is not the first set of REGNO, see whether the new value
|
1175 |
|
|
is related to the old one. There are two cases of interest:
|
1176 |
|
|
|
1177 |
|
|
(1) The register might be assigned an entirely new value
|
1178 |
|
|
that has the same base term as the original set.
|
1179 |
|
|
|
1180 |
|
|
(2) The set might be a simple self-modification that
|
1181 |
|
|
cannot change REGNO's base value.
|
1182 |
|
|
|
1183 |
|
|
If neither case holds, reject the original base value as invalid.
|
1184 |
|
|
Note that the following situation is not detected:
|
1185 |
|
|
|
1186 |
|
|
extern int x, y; int *p = &x; p += (&y-&x);
|
1187 |
|
|
|
1188 |
|
|
ANSI C does not allow computing the difference of addresses
|
1189 |
|
|
of distinct top level objects. */
|
1190 |
|
|
if (new_reg_base_value[regno] != 0
|
1191 |
|
|
&& find_base_value (src) != new_reg_base_value[regno])
|
1192 |
|
|
switch (GET_CODE (src))
|
1193 |
|
|
{
|
1194 |
|
|
case LO_SUM:
|
1195 |
|
|
case MINUS:
|
1196 |
|
|
if (XEXP (src, 0) != dest && XEXP (src, 1) != dest)
|
1197 |
|
|
new_reg_base_value[regno] = 0;
|
1198 |
|
|
break;
|
1199 |
|
|
case PLUS:
|
1200 |
|
|
/* If the value we add in the PLUS is also a valid base value,
|
1201 |
|
|
this might be the actual base value, and the original value
|
1202 |
|
|
an index. */
|
1203 |
|
|
{
|
1204 |
|
|
rtx other = NULL_RTX;
|
1205 |
|
|
|
1206 |
|
|
if (XEXP (src, 0) == dest)
|
1207 |
|
|
other = XEXP (src, 1);
|
1208 |
|
|
else if (XEXP (src, 1) == dest)
|
1209 |
|
|
other = XEXP (src, 0);
|
1210 |
|
|
|
1211 |
|
|
if (! other || find_base_value (other))
|
1212 |
|
|
new_reg_base_value[regno] = 0;
|
1213 |
|
|
break;
|
1214 |
|
|
}
|
1215 |
|
|
case AND:
|
1216 |
|
|
if (XEXP (src, 0) != dest || !CONST_INT_P (XEXP (src, 1)))
|
1217 |
|
|
new_reg_base_value[regno] = 0;
|
1218 |
|
|
break;
|
1219 |
|
|
default:
|
1220 |
|
|
new_reg_base_value[regno] = 0;
|
1221 |
|
|
break;
|
1222 |
|
|
}
|
1223 |
|
|
/* If this is the first set of a register, record the value. */
|
1224 |
|
|
else if ((regno >= FIRST_PSEUDO_REGISTER || ! fixed_regs[regno])
|
1225 |
|
|
&& ! reg_seen[regno] && new_reg_base_value[regno] == 0)
|
1226 |
|
|
new_reg_base_value[regno] = find_base_value (src);
|
1227 |
|
|
|
1228 |
|
|
reg_seen[regno] = 1;
|
1229 |
|
|
}
|
1230 |
|
|
|
1231 |
|
|
/* If a value is known for REGNO, return it. */
|
1232 |
|
|
|
1233 |
|
|
rtx
|
1234 |
|
|
get_reg_known_value (unsigned int regno)
|
1235 |
|
|
{
|
1236 |
|
|
if (regno >= FIRST_PSEUDO_REGISTER)
|
1237 |
|
|
{
|
1238 |
|
|
regno -= FIRST_PSEUDO_REGISTER;
|
1239 |
|
|
if (regno < reg_known_value_size)
|
1240 |
|
|
return reg_known_value[regno];
|
1241 |
|
|
}
|
1242 |
|
|
return NULL;
|
1243 |
|
|
}
|
1244 |
|
|
|
1245 |
|
|
/* Set it. */
|
1246 |
|
|
|
1247 |
|
|
static void
|
1248 |
|
|
set_reg_known_value (unsigned int regno, rtx val)
|
1249 |
|
|
{
|
1250 |
|
|
if (regno >= FIRST_PSEUDO_REGISTER)
|
1251 |
|
|
{
|
1252 |
|
|
regno -= FIRST_PSEUDO_REGISTER;
|
1253 |
|
|
if (regno < reg_known_value_size)
|
1254 |
|
|
reg_known_value[regno] = val;
|
1255 |
|
|
}
|
1256 |
|
|
}
|
1257 |
|
|
|
1258 |
|
|
/* Similarly for reg_known_equiv_p. */
|
1259 |
|
|
|
1260 |
|
|
bool
|
1261 |
|
|
get_reg_known_equiv_p (unsigned int regno)
|
1262 |
|
|
{
|
1263 |
|
|
if (regno >= FIRST_PSEUDO_REGISTER)
|
1264 |
|
|
{
|
1265 |
|
|
regno -= FIRST_PSEUDO_REGISTER;
|
1266 |
|
|
if (regno < reg_known_value_size)
|
1267 |
|
|
return reg_known_equiv_p[regno];
|
1268 |
|
|
}
|
1269 |
|
|
return false;
|
1270 |
|
|
}
|
1271 |
|
|
|
1272 |
|
|
static void
|
1273 |
|
|
set_reg_known_equiv_p (unsigned int regno, bool val)
|
1274 |
|
|
{
|
1275 |
|
|
if (regno >= FIRST_PSEUDO_REGISTER)
|
1276 |
|
|
{
|
1277 |
|
|
regno -= FIRST_PSEUDO_REGISTER;
|
1278 |
|
|
if (regno < reg_known_value_size)
|
1279 |
|
|
reg_known_equiv_p[regno] = val;
|
1280 |
|
|
}
|
1281 |
|
|
}
|
1282 |
|
|
|
1283 |
|
|
|
1284 |
|
|
/* Returns a canonical version of X, from the point of view alias
|
1285 |
|
|
analysis. (For example, if X is a MEM whose address is a register,
|
1286 |
|
|
and the register has a known value (say a SYMBOL_REF), then a MEM
|
1287 |
|
|
whose address is the SYMBOL_REF is returned.) */
|
1288 |
|
|
|
1289 |
|
|
rtx
|
1290 |
|
|
canon_rtx (rtx x)
|
1291 |
|
|
{
|
1292 |
|
|
/* Recursively look for equivalences. */
|
1293 |
|
|
if (REG_P (x) && REGNO (x) >= FIRST_PSEUDO_REGISTER)
|
1294 |
|
|
{
|
1295 |
|
|
rtx t = get_reg_known_value (REGNO (x));
|
1296 |
|
|
if (t == x)
|
1297 |
|
|
return x;
|
1298 |
|
|
if (t)
|
1299 |
|
|
return canon_rtx (t);
|
1300 |
|
|
}
|
1301 |
|
|
|
1302 |
|
|
if (GET_CODE (x) == PLUS)
|
1303 |
|
|
{
|
1304 |
|
|
rtx x0 = canon_rtx (XEXP (x, 0));
|
1305 |
|
|
rtx x1 = canon_rtx (XEXP (x, 1));
|
1306 |
|
|
|
1307 |
|
|
if (x0 != XEXP (x, 0) || x1 != XEXP (x, 1))
|
1308 |
|
|
{
|
1309 |
|
|
if (CONST_INT_P (x0))
|
1310 |
|
|
return plus_constant (x1, INTVAL (x0));
|
1311 |
|
|
else if (CONST_INT_P (x1))
|
1312 |
|
|
return plus_constant (x0, INTVAL (x1));
|
1313 |
|
|
return gen_rtx_PLUS (GET_MODE (x), x0, x1);
|
1314 |
|
|
}
|
1315 |
|
|
}
|
1316 |
|
|
|
1317 |
|
|
/* This gives us much better alias analysis when called from
|
1318 |
|
|
the loop optimizer. Note we want to leave the original
|
1319 |
|
|
MEM alone, but need to return the canonicalized MEM with
|
1320 |
|
|
all the flags with their original values. */
|
1321 |
|
|
else if (MEM_P (x))
|
1322 |
|
|
x = replace_equiv_address_nv (x, canon_rtx (XEXP (x, 0)));
|
1323 |
|
|
|
1324 |
|
|
return x;
|
1325 |
|
|
}
|
1326 |
|
|
|
1327 |
|
|
/* Return 1 if X and Y are identical-looking rtx's.
|
1328 |
|
|
Expect that X and Y has been already canonicalized.
|
1329 |
|
|
|
1330 |
|
|
We use the data in reg_known_value above to see if two registers with
|
1331 |
|
|
different numbers are, in fact, equivalent. */
|
1332 |
|
|
|
1333 |
|
|
static int
|
1334 |
|
|
rtx_equal_for_memref_p (const_rtx x, const_rtx y)
|
1335 |
|
|
{
|
1336 |
|
|
int i;
|
1337 |
|
|
int j;
|
1338 |
|
|
enum rtx_code code;
|
1339 |
|
|
const char *fmt;
|
1340 |
|
|
|
1341 |
|
|
if (x == 0 && y == 0)
|
1342 |
|
|
return 1;
|
1343 |
|
|
if (x == 0 || y == 0)
|
1344 |
|
|
return 0;
|
1345 |
|
|
|
1346 |
|
|
if (x == y)
|
1347 |
|
|
return 1;
|
1348 |
|
|
|
1349 |
|
|
code = GET_CODE (x);
|
1350 |
|
|
/* Rtx's of different codes cannot be equal. */
|
1351 |
|
|
if (code != GET_CODE (y))
|
1352 |
|
|
return 0;
|
1353 |
|
|
|
1354 |
|
|
/* (MULT:SI x y) and (MULT:HI x y) are NOT equivalent.
|
1355 |
|
|
(REG:SI x) and (REG:HI x) are NOT equivalent. */
|
1356 |
|
|
|
1357 |
|
|
if (GET_MODE (x) != GET_MODE (y))
|
1358 |
|
|
return 0;
|
1359 |
|
|
|
1360 |
|
|
/* Some RTL can be compared without a recursive examination. */
|
1361 |
|
|
switch (code)
|
1362 |
|
|
{
|
1363 |
|
|
case REG:
|
1364 |
|
|
return REGNO (x) == REGNO (y);
|
1365 |
|
|
|
1366 |
|
|
case LABEL_REF:
|
1367 |
|
|
return XEXP (x, 0) == XEXP (y, 0);
|
1368 |
|
|
|
1369 |
|
|
case SYMBOL_REF:
|
1370 |
|
|
return XSTR (x, 0) == XSTR (y, 0);
|
1371 |
|
|
|
1372 |
|
|
case VALUE:
|
1373 |
|
|
case CONST_INT:
|
1374 |
|
|
case CONST_DOUBLE:
|
1375 |
|
|
case CONST_FIXED:
|
1376 |
|
|
/* There's no need to compare the contents of CONST_DOUBLEs or
|
1377 |
|
|
CONST_INTs because pointer equality is a good enough
|
1378 |
|
|
comparison for these nodes. */
|
1379 |
|
|
return 0;
|
1380 |
|
|
|
1381 |
|
|
default:
|
1382 |
|
|
break;
|
1383 |
|
|
}
|
1384 |
|
|
|
1385 |
|
|
/* canon_rtx knows how to handle plus. No need to canonicalize. */
|
1386 |
|
|
if (code == PLUS)
|
1387 |
|
|
return ((rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 0))
|
1388 |
|
|
&& rtx_equal_for_memref_p (XEXP (x, 1), XEXP (y, 1)))
|
1389 |
|
|
|| (rtx_equal_for_memref_p (XEXP (x, 0), XEXP (y, 1))
|
1390 |
|
|
&& rtx_equal_for_memref_p (XEXP (x, 1), XEXP (y, 0))));
|
1391 |
|
|
/* For commutative operations, the RTX match if the operand match in any
|
1392 |
|
|
order. Also handle the simple binary and unary cases without a loop. */
|
1393 |
|
|
if (COMMUTATIVE_P (x))
|
1394 |
|
|
{
|
1395 |
|
|
rtx xop0 = canon_rtx (XEXP (x, 0));
|
1396 |
|
|
rtx yop0 = canon_rtx (XEXP (y, 0));
|
1397 |
|
|
rtx yop1 = canon_rtx (XEXP (y, 1));
|
1398 |
|
|
|
1399 |
|
|
return ((rtx_equal_for_memref_p (xop0, yop0)
|
1400 |
|
|
&& rtx_equal_for_memref_p (canon_rtx (XEXP (x, 1)), yop1))
|
1401 |
|
|
|| (rtx_equal_for_memref_p (xop0, yop1)
|
1402 |
|
|
&& rtx_equal_for_memref_p (canon_rtx (XEXP (x, 1)), yop0)));
|
1403 |
|
|
}
|
1404 |
|
|
else if (NON_COMMUTATIVE_P (x))
|
1405 |
|
|
{
|
1406 |
|
|
return (rtx_equal_for_memref_p (canon_rtx (XEXP (x, 0)),
|
1407 |
|
|
canon_rtx (XEXP (y, 0)))
|
1408 |
|
|
&& rtx_equal_for_memref_p (canon_rtx (XEXP (x, 1)),
|
1409 |
|
|
canon_rtx (XEXP (y, 1))));
|
1410 |
|
|
}
|
1411 |
|
|
else if (UNARY_P (x))
|
1412 |
|
|
return rtx_equal_for_memref_p (canon_rtx (XEXP (x, 0)),
|
1413 |
|
|
canon_rtx (XEXP (y, 0)));
|
1414 |
|
|
|
1415 |
|
|
/* Compare the elements. If any pair of corresponding elements
|
1416 |
|
|
fail to match, return 0 for the whole things.
|
1417 |
|
|
|
1418 |
|
|
Limit cases to types which actually appear in addresses. */
|
1419 |
|
|
|
1420 |
|
|
fmt = GET_RTX_FORMAT (code);
|
1421 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
1422 |
|
|
{
|
1423 |
|
|
switch (fmt[i])
|
1424 |
|
|
{
|
1425 |
|
|
case 'i':
|
1426 |
|
|
if (XINT (x, i) != XINT (y, i))
|
1427 |
|
|
return 0;
|
1428 |
|
|
break;
|
1429 |
|
|
|
1430 |
|
|
case 'E':
|
1431 |
|
|
/* Two vectors must have the same length. */
|
1432 |
|
|
if (XVECLEN (x, i) != XVECLEN (y, i))
|
1433 |
|
|
return 0;
|
1434 |
|
|
|
1435 |
|
|
/* And the corresponding elements must match. */
|
1436 |
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
1437 |
|
|
if (rtx_equal_for_memref_p (canon_rtx (XVECEXP (x, i, j)),
|
1438 |
|
|
canon_rtx (XVECEXP (y, i, j))) == 0)
|
1439 |
|
|
return 0;
|
1440 |
|
|
break;
|
1441 |
|
|
|
1442 |
|
|
case 'e':
|
1443 |
|
|
if (rtx_equal_for_memref_p (canon_rtx (XEXP (x, i)),
|
1444 |
|
|
canon_rtx (XEXP (y, i))) == 0)
|
1445 |
|
|
return 0;
|
1446 |
|
|
break;
|
1447 |
|
|
|
1448 |
|
|
/* This can happen for asm operands. */
|
1449 |
|
|
case 's':
|
1450 |
|
|
if (strcmp (XSTR (x, i), XSTR (y, i)))
|
1451 |
|
|
return 0;
|
1452 |
|
|
break;
|
1453 |
|
|
|
1454 |
|
|
/* This can happen for an asm which clobbers memory. */
|
1455 |
|
|
case '0':
|
1456 |
|
|
break;
|
1457 |
|
|
|
1458 |
|
|
/* It is believed that rtx's at this level will never
|
1459 |
|
|
contain anything but integers and other rtx's,
|
1460 |
|
|
except for within LABEL_REFs and SYMBOL_REFs. */
|
1461 |
|
|
default:
|
1462 |
|
|
gcc_unreachable ();
|
1463 |
|
|
}
|
1464 |
|
|
}
|
1465 |
|
|
return 1;
|
1466 |
|
|
}
|
1467 |
|
|
|
1468 |
|
|
rtx
|
1469 |
|
|
find_base_term (rtx x)
|
1470 |
|
|
{
|
1471 |
|
|
cselib_val *val;
|
1472 |
|
|
struct elt_loc_list *l;
|
1473 |
|
|
|
1474 |
|
|
#if defined (FIND_BASE_TERM)
|
1475 |
|
|
/* Try machine-dependent ways to find the base term. */
|
1476 |
|
|
x = FIND_BASE_TERM (x);
|
1477 |
|
|
#endif
|
1478 |
|
|
|
1479 |
|
|
switch (GET_CODE (x))
|
1480 |
|
|
{
|
1481 |
|
|
case REG:
|
1482 |
|
|
return REG_BASE_VALUE (x);
|
1483 |
|
|
|
1484 |
|
|
case TRUNCATE:
|
1485 |
|
|
/* As we do not know which address space the pointer is refering to, we can
|
1486 |
|
|
handle this only if the target does not support different pointer or
|
1487 |
|
|
address modes depending on the address space. */
|
1488 |
|
|
if (!target_default_pointer_address_modes_p ())
|
1489 |
|
|
return 0;
|
1490 |
|
|
if (GET_MODE_SIZE (GET_MODE (x)) < GET_MODE_SIZE (Pmode))
|
1491 |
|
|
return 0;
|
1492 |
|
|
/* Fall through. */
|
1493 |
|
|
case HIGH:
|
1494 |
|
|
case PRE_INC:
|
1495 |
|
|
case PRE_DEC:
|
1496 |
|
|
case POST_INC:
|
1497 |
|
|
case POST_DEC:
|
1498 |
|
|
case PRE_MODIFY:
|
1499 |
|
|
case POST_MODIFY:
|
1500 |
|
|
return find_base_term (XEXP (x, 0));
|
1501 |
|
|
|
1502 |
|
|
case ZERO_EXTEND:
|
1503 |
|
|
case SIGN_EXTEND: /* Used for Alpha/NT pointers */
|
1504 |
|
|
/* As we do not know which address space the pointer is refering to, we can
|
1505 |
|
|
handle this only if the target does not support different pointer or
|
1506 |
|
|
address modes depending on the address space. */
|
1507 |
|
|
if (!target_default_pointer_address_modes_p ())
|
1508 |
|
|
return 0;
|
1509 |
|
|
|
1510 |
|
|
{
|
1511 |
|
|
rtx temp = find_base_term (XEXP (x, 0));
|
1512 |
|
|
|
1513 |
|
|
if (temp != 0 && CONSTANT_P (temp))
|
1514 |
|
|
temp = convert_memory_address (Pmode, temp);
|
1515 |
|
|
|
1516 |
|
|
return temp;
|
1517 |
|
|
}
|
1518 |
|
|
|
1519 |
|
|
case VALUE:
|
1520 |
|
|
val = CSELIB_VAL_PTR (x);
|
1521 |
|
|
if (!val)
|
1522 |
|
|
return 0;
|
1523 |
|
|
for (l = val->locs; l; l = l->next)
|
1524 |
|
|
if ((x = find_base_term (l->loc)) != 0)
|
1525 |
|
|
return x;
|
1526 |
|
|
return 0;
|
1527 |
|
|
|
1528 |
|
|
case LO_SUM:
|
1529 |
|
|
/* The standard form is (lo_sum reg sym) so look only at the
|
1530 |
|
|
second operand. */
|
1531 |
|
|
return find_base_term (XEXP (x, 1));
|
1532 |
|
|
|
1533 |
|
|
case CONST:
|
1534 |
|
|
x = XEXP (x, 0);
|
1535 |
|
|
if (GET_CODE (x) != PLUS && GET_CODE (x) != MINUS)
|
1536 |
|
|
return 0;
|
1537 |
|
|
/* Fall through. */
|
1538 |
|
|
case PLUS:
|
1539 |
|
|
case MINUS:
|
1540 |
|
|
{
|
1541 |
|
|
rtx tmp1 = XEXP (x, 0);
|
1542 |
|
|
rtx tmp2 = XEXP (x, 1);
|
1543 |
|
|
|
1544 |
|
|
/* This is a little bit tricky since we have to determine which of
|
1545 |
|
|
the two operands represents the real base address. Otherwise this
|
1546 |
|
|
routine may return the index register instead of the base register.
|
1547 |
|
|
|
1548 |
|
|
That may cause us to believe no aliasing was possible, when in
|
1549 |
|
|
fact aliasing is possible.
|
1550 |
|
|
|
1551 |
|
|
We use a few simple tests to guess the base register. Additional
|
1552 |
|
|
tests can certainly be added. For example, if one of the operands
|
1553 |
|
|
is a shift or multiply, then it must be the index register and the
|
1554 |
|
|
other operand is the base register. */
|
1555 |
|
|
|
1556 |
|
|
if (tmp1 == pic_offset_table_rtx && CONSTANT_P (tmp2))
|
1557 |
|
|
return find_base_term (tmp2);
|
1558 |
|
|
|
1559 |
|
|
/* If either operand is known to be a pointer, then use it
|
1560 |
|
|
to determine the base term. */
|
1561 |
|
|
if (REG_P (tmp1) && REG_POINTER (tmp1))
|
1562 |
|
|
{
|
1563 |
|
|
rtx base = find_base_term (tmp1);
|
1564 |
|
|
if (base)
|
1565 |
|
|
return base;
|
1566 |
|
|
}
|
1567 |
|
|
|
1568 |
|
|
if (REG_P (tmp2) && REG_POINTER (tmp2))
|
1569 |
|
|
{
|
1570 |
|
|
rtx base = find_base_term (tmp2);
|
1571 |
|
|
if (base)
|
1572 |
|
|
return base;
|
1573 |
|
|
}
|
1574 |
|
|
|
1575 |
|
|
/* Neither operand was known to be a pointer. Go ahead and find the
|
1576 |
|
|
base term for both operands. */
|
1577 |
|
|
tmp1 = find_base_term (tmp1);
|
1578 |
|
|
tmp2 = find_base_term (tmp2);
|
1579 |
|
|
|
1580 |
|
|
/* If either base term is named object or a special address
|
1581 |
|
|
(like an argument or stack reference), then use it for the
|
1582 |
|
|
base term. */
|
1583 |
|
|
if (tmp1 != 0
|
1584 |
|
|
&& (GET_CODE (tmp1) == SYMBOL_REF
|
1585 |
|
|
|| GET_CODE (tmp1) == LABEL_REF
|
1586 |
|
|
|| (GET_CODE (tmp1) == ADDRESS
|
1587 |
|
|
&& GET_MODE (tmp1) != VOIDmode)))
|
1588 |
|
|
return tmp1;
|
1589 |
|
|
|
1590 |
|
|
if (tmp2 != 0
|
1591 |
|
|
&& (GET_CODE (tmp2) == SYMBOL_REF
|
1592 |
|
|
|| GET_CODE (tmp2) == LABEL_REF
|
1593 |
|
|
|| (GET_CODE (tmp2) == ADDRESS
|
1594 |
|
|
&& GET_MODE (tmp2) != VOIDmode)))
|
1595 |
|
|
return tmp2;
|
1596 |
|
|
|
1597 |
|
|
/* We could not determine which of the two operands was the
|
1598 |
|
|
base register and which was the index. So we can determine
|
1599 |
|
|
nothing from the base alias check. */
|
1600 |
|
|
return 0;
|
1601 |
|
|
}
|
1602 |
|
|
|
1603 |
|
|
case AND:
|
1604 |
|
|
if (CONST_INT_P (XEXP (x, 1)) && INTVAL (XEXP (x, 1)) != 0)
|
1605 |
|
|
return find_base_term (XEXP (x, 0));
|
1606 |
|
|
return 0;
|
1607 |
|
|
|
1608 |
|
|
case SYMBOL_REF:
|
1609 |
|
|
case LABEL_REF:
|
1610 |
|
|
return x;
|
1611 |
|
|
|
1612 |
|
|
default:
|
1613 |
|
|
return 0;
|
1614 |
|
|
}
|
1615 |
|
|
}
|
1616 |
|
|
|
1617 |
|
|
/* Return 0 if the addresses X and Y are known to point to different
|
1618 |
|
|
objects, 1 if they might be pointers to the same object. */
|
1619 |
|
|
|
1620 |
|
|
static int
|
1621 |
|
|
base_alias_check (rtx x, rtx y, enum machine_mode x_mode,
|
1622 |
|
|
enum machine_mode y_mode)
|
1623 |
|
|
{
|
1624 |
|
|
rtx x_base = find_base_term (x);
|
1625 |
|
|
rtx y_base = find_base_term (y);
|
1626 |
|
|
|
1627 |
|
|
/* If the address itself has no known base see if a known equivalent
|
1628 |
|
|
value has one. If either address still has no known base, nothing
|
1629 |
|
|
is known about aliasing. */
|
1630 |
|
|
if (x_base == 0)
|
1631 |
|
|
{
|
1632 |
|
|
rtx x_c;
|
1633 |
|
|
|
1634 |
|
|
if (! flag_expensive_optimizations || (x_c = canon_rtx (x)) == x)
|
1635 |
|
|
return 1;
|
1636 |
|
|
|
1637 |
|
|
x_base = find_base_term (x_c);
|
1638 |
|
|
if (x_base == 0)
|
1639 |
|
|
return 1;
|
1640 |
|
|
}
|
1641 |
|
|
|
1642 |
|
|
if (y_base == 0)
|
1643 |
|
|
{
|
1644 |
|
|
rtx y_c;
|
1645 |
|
|
if (! flag_expensive_optimizations || (y_c = canon_rtx (y)) == y)
|
1646 |
|
|
return 1;
|
1647 |
|
|
|
1648 |
|
|
y_base = find_base_term (y_c);
|
1649 |
|
|
if (y_base == 0)
|
1650 |
|
|
return 1;
|
1651 |
|
|
}
|
1652 |
|
|
|
1653 |
|
|
/* If the base addresses are equal nothing is known about aliasing. */
|
1654 |
|
|
if (rtx_equal_p (x_base, y_base))
|
1655 |
|
|
return 1;
|
1656 |
|
|
|
1657 |
|
|
/* The base addresses are different expressions. If they are not accessed
|
1658 |
|
|
via AND, there is no conflict. We can bring knowledge of object
|
1659 |
|
|
alignment into play here. For example, on alpha, "char a, b;" can
|
1660 |
|
|
alias one another, though "char a; long b;" cannot. AND addesses may
|
1661 |
|
|
implicitly alias surrounding objects; i.e. unaligned access in DImode
|
1662 |
|
|
via AND address can alias all surrounding object types except those
|
1663 |
|
|
with aligment 8 or higher. */
|
1664 |
|
|
if (GET_CODE (x) == AND && GET_CODE (y) == AND)
|
1665 |
|
|
return 1;
|
1666 |
|
|
if (GET_CODE (x) == AND
|
1667 |
|
|
&& (!CONST_INT_P (XEXP (x, 1))
|
1668 |
|
|
|| (int) GET_MODE_UNIT_SIZE (y_mode) < -INTVAL (XEXP (x, 1))))
|
1669 |
|
|
return 1;
|
1670 |
|
|
if (GET_CODE (y) == AND
|
1671 |
|
|
&& (!CONST_INT_P (XEXP (y, 1))
|
1672 |
|
|
|| (int) GET_MODE_UNIT_SIZE (x_mode) < -INTVAL (XEXP (y, 1))))
|
1673 |
|
|
return 1;
|
1674 |
|
|
|
1675 |
|
|
/* Differing symbols not accessed via AND never alias. */
|
1676 |
|
|
if (GET_CODE (x_base) != ADDRESS && GET_CODE (y_base) != ADDRESS)
|
1677 |
|
|
return 0;
|
1678 |
|
|
|
1679 |
|
|
/* If one address is a stack reference there can be no alias:
|
1680 |
|
|
stack references using different base registers do not alias,
|
1681 |
|
|
a stack reference can not alias a parameter, and a stack reference
|
1682 |
|
|
can not alias a global. */
|
1683 |
|
|
if ((GET_CODE (x_base) == ADDRESS && GET_MODE (x_base) == Pmode)
|
1684 |
|
|
|| (GET_CODE (y_base) == ADDRESS && GET_MODE (y_base) == Pmode))
|
1685 |
|
|
return 0;
|
1686 |
|
|
|
1687 |
|
|
if (! flag_argument_noalias)
|
1688 |
|
|
return 1;
|
1689 |
|
|
|
1690 |
|
|
if (flag_argument_noalias > 1)
|
1691 |
|
|
return 0;
|
1692 |
|
|
|
1693 |
|
|
/* Weak noalias assertion (arguments are distinct, but may match globals). */
|
1694 |
|
|
return ! (GET_MODE (x_base) == VOIDmode && GET_MODE (y_base) == VOIDmode);
|
1695 |
|
|
}
|
1696 |
|
|
|
1697 |
|
|
/* Convert the address X into something we can use. This is done by returning
|
1698 |
|
|
it unchanged unless it is a value; in the latter case we call cselib to get
|
1699 |
|
|
a more useful rtx. */
|
1700 |
|
|
|
1701 |
|
|
rtx
|
1702 |
|
|
get_addr (rtx x)
|
1703 |
|
|
{
|
1704 |
|
|
cselib_val *v;
|
1705 |
|
|
struct elt_loc_list *l;
|
1706 |
|
|
|
1707 |
|
|
if (GET_CODE (x) != VALUE)
|
1708 |
|
|
return x;
|
1709 |
|
|
v = CSELIB_VAL_PTR (x);
|
1710 |
|
|
if (v)
|
1711 |
|
|
{
|
1712 |
|
|
for (l = v->locs; l; l = l->next)
|
1713 |
|
|
if (CONSTANT_P (l->loc))
|
1714 |
|
|
return l->loc;
|
1715 |
|
|
for (l = v->locs; l; l = l->next)
|
1716 |
|
|
if (!REG_P (l->loc) && !MEM_P (l->loc))
|
1717 |
|
|
return l->loc;
|
1718 |
|
|
if (v->locs)
|
1719 |
|
|
return v->locs->loc;
|
1720 |
|
|
}
|
1721 |
|
|
return x;
|
1722 |
|
|
}
|
1723 |
|
|
|
1724 |
|
|
/* Return the address of the (N_REFS + 1)th memory reference to ADDR
|
1725 |
|
|
where SIZE is the size in bytes of the memory reference. If ADDR
|
1726 |
|
|
is not modified by the memory reference then ADDR is returned. */
|
1727 |
|
|
|
1728 |
|
|
static rtx
|
1729 |
|
|
addr_side_effect_eval (rtx addr, int size, int n_refs)
|
1730 |
|
|
{
|
1731 |
|
|
int offset = 0;
|
1732 |
|
|
|
1733 |
|
|
switch (GET_CODE (addr))
|
1734 |
|
|
{
|
1735 |
|
|
case PRE_INC:
|
1736 |
|
|
offset = (n_refs + 1) * size;
|
1737 |
|
|
break;
|
1738 |
|
|
case PRE_DEC:
|
1739 |
|
|
offset = -(n_refs + 1) * size;
|
1740 |
|
|
break;
|
1741 |
|
|
case POST_INC:
|
1742 |
|
|
offset = n_refs * size;
|
1743 |
|
|
break;
|
1744 |
|
|
case POST_DEC:
|
1745 |
|
|
offset = -n_refs * size;
|
1746 |
|
|
break;
|
1747 |
|
|
|
1748 |
|
|
default:
|
1749 |
|
|
return addr;
|
1750 |
|
|
}
|
1751 |
|
|
|
1752 |
|
|
if (offset)
|
1753 |
|
|
addr = gen_rtx_PLUS (GET_MODE (addr), XEXP (addr, 0),
|
1754 |
|
|
GEN_INT (offset));
|
1755 |
|
|
else
|
1756 |
|
|
addr = XEXP (addr, 0);
|
1757 |
|
|
addr = canon_rtx (addr);
|
1758 |
|
|
|
1759 |
|
|
return addr;
|
1760 |
|
|
}
|
1761 |
|
|
|
1762 |
|
|
/* Return one if X and Y (memory addresses) reference the
|
1763 |
|
|
same location in memory or if the references overlap.
|
1764 |
|
|
Return zero if they do not overlap, else return
|
1765 |
|
|
minus one in which case they still might reference the same location.
|
1766 |
|
|
|
1767 |
|
|
C is an offset accumulator. When
|
1768 |
|
|
C is nonzero, we are testing aliases between X and Y + C.
|
1769 |
|
|
XSIZE is the size in bytes of the X reference,
|
1770 |
|
|
similarly YSIZE is the size in bytes for Y.
|
1771 |
|
|
Expect that canon_rtx has been already called for X and Y.
|
1772 |
|
|
|
1773 |
|
|
If XSIZE or YSIZE is zero, we do not know the amount of memory being
|
1774 |
|
|
referenced (the reference was BLKmode), so make the most pessimistic
|
1775 |
|
|
assumptions.
|
1776 |
|
|
|
1777 |
|
|
If XSIZE or YSIZE is negative, we may access memory outside the object
|
1778 |
|
|
being referenced as a side effect. This can happen when using AND to
|
1779 |
|
|
align memory references, as is done on the Alpha.
|
1780 |
|
|
|
1781 |
|
|
Nice to notice that varying addresses cannot conflict with fp if no
|
1782 |
|
|
local variables had their addresses taken, but that's too hard now.
|
1783 |
|
|
|
1784 |
|
|
??? Contrary to the tree alias oracle this does not return
|
1785 |
|
|
one for X + non-constant and Y + non-constant when X and Y are equal.
|
1786 |
|
|
If that is fixed the TBAA hack for union type-punning can be removed. */
|
1787 |
|
|
|
1788 |
|
|
static int
|
1789 |
|
|
memrefs_conflict_p (int xsize, rtx x, int ysize, rtx y, HOST_WIDE_INT c)
|
1790 |
|
|
{
|
1791 |
|
|
if (GET_CODE (x) == VALUE)
|
1792 |
|
|
x = get_addr (x);
|
1793 |
|
|
if (GET_CODE (y) == VALUE)
|
1794 |
|
|
y = get_addr (y);
|
1795 |
|
|
if (GET_CODE (x) == HIGH)
|
1796 |
|
|
x = XEXP (x, 0);
|
1797 |
|
|
else if (GET_CODE (x) == LO_SUM)
|
1798 |
|
|
x = XEXP (x, 1);
|
1799 |
|
|
else
|
1800 |
|
|
x = addr_side_effect_eval (x, xsize, 0);
|
1801 |
|
|
if (GET_CODE (y) == HIGH)
|
1802 |
|
|
y = XEXP (y, 0);
|
1803 |
|
|
else if (GET_CODE (y) == LO_SUM)
|
1804 |
|
|
y = XEXP (y, 1);
|
1805 |
|
|
else
|
1806 |
|
|
y = addr_side_effect_eval (y, ysize, 0);
|
1807 |
|
|
|
1808 |
|
|
if (rtx_equal_for_memref_p (x, y))
|
1809 |
|
|
{
|
1810 |
|
|
if (xsize <= 0 || ysize <= 0)
|
1811 |
|
|
return 1;
|
1812 |
|
|
if (c >= 0 && xsize > c)
|
1813 |
|
|
return 1;
|
1814 |
|
|
if (c < 0 && ysize+c > 0)
|
1815 |
|
|
return 1;
|
1816 |
|
|
return 0;
|
1817 |
|
|
}
|
1818 |
|
|
|
1819 |
|
|
/* This code used to check for conflicts involving stack references and
|
1820 |
|
|
globals but the base address alias code now handles these cases. */
|
1821 |
|
|
|
1822 |
|
|
if (GET_CODE (x) == PLUS)
|
1823 |
|
|
{
|
1824 |
|
|
/* The fact that X is canonicalized means that this
|
1825 |
|
|
PLUS rtx is canonicalized. */
|
1826 |
|
|
rtx x0 = XEXP (x, 0);
|
1827 |
|
|
rtx x1 = XEXP (x, 1);
|
1828 |
|
|
|
1829 |
|
|
if (GET_CODE (y) == PLUS)
|
1830 |
|
|
{
|
1831 |
|
|
/* The fact that Y is canonicalized means that this
|
1832 |
|
|
PLUS rtx is canonicalized. */
|
1833 |
|
|
rtx y0 = XEXP (y, 0);
|
1834 |
|
|
rtx y1 = XEXP (y, 1);
|
1835 |
|
|
|
1836 |
|
|
if (rtx_equal_for_memref_p (x1, y1))
|
1837 |
|
|
return memrefs_conflict_p (xsize, x0, ysize, y0, c);
|
1838 |
|
|
if (rtx_equal_for_memref_p (x0, y0))
|
1839 |
|
|
return memrefs_conflict_p (xsize, x1, ysize, y1, c);
|
1840 |
|
|
if (CONST_INT_P (x1))
|
1841 |
|
|
{
|
1842 |
|
|
if (CONST_INT_P (y1))
|
1843 |
|
|
return memrefs_conflict_p (xsize, x0, ysize, y0,
|
1844 |
|
|
c - INTVAL (x1) + INTVAL (y1));
|
1845 |
|
|
else
|
1846 |
|
|
return memrefs_conflict_p (xsize, x0, ysize, y,
|
1847 |
|
|
c - INTVAL (x1));
|
1848 |
|
|
}
|
1849 |
|
|
else if (CONST_INT_P (y1))
|
1850 |
|
|
return memrefs_conflict_p (xsize, x, ysize, y0, c + INTVAL (y1));
|
1851 |
|
|
|
1852 |
|
|
return -1;
|
1853 |
|
|
}
|
1854 |
|
|
else if (CONST_INT_P (x1))
|
1855 |
|
|
return memrefs_conflict_p (xsize, x0, ysize, y, c - INTVAL (x1));
|
1856 |
|
|
}
|
1857 |
|
|
else if (GET_CODE (y) == PLUS)
|
1858 |
|
|
{
|
1859 |
|
|
/* The fact that Y is canonicalized means that this
|
1860 |
|
|
PLUS rtx is canonicalized. */
|
1861 |
|
|
rtx y0 = XEXP (y, 0);
|
1862 |
|
|
rtx y1 = XEXP (y, 1);
|
1863 |
|
|
|
1864 |
|
|
if (CONST_INT_P (y1))
|
1865 |
|
|
return memrefs_conflict_p (xsize, x, ysize, y0, c + INTVAL (y1));
|
1866 |
|
|
else
|
1867 |
|
|
return -1;
|
1868 |
|
|
}
|
1869 |
|
|
|
1870 |
|
|
if (GET_CODE (x) == GET_CODE (y))
|
1871 |
|
|
switch (GET_CODE (x))
|
1872 |
|
|
{
|
1873 |
|
|
case MULT:
|
1874 |
|
|
{
|
1875 |
|
|
/* Handle cases where we expect the second operands to be the
|
1876 |
|
|
same, and check only whether the first operand would conflict
|
1877 |
|
|
or not. */
|
1878 |
|
|
rtx x0, y0;
|
1879 |
|
|
rtx x1 = canon_rtx (XEXP (x, 1));
|
1880 |
|
|
rtx y1 = canon_rtx (XEXP (y, 1));
|
1881 |
|
|
if (! rtx_equal_for_memref_p (x1, y1))
|
1882 |
|
|
return -1;
|
1883 |
|
|
x0 = canon_rtx (XEXP (x, 0));
|
1884 |
|
|
y0 = canon_rtx (XEXP (y, 0));
|
1885 |
|
|
if (rtx_equal_for_memref_p (x0, y0))
|
1886 |
|
|
return (xsize == 0 || ysize == 0
|
1887 |
|
|
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
|
1888 |
|
|
|
1889 |
|
|
/* Can't properly adjust our sizes. */
|
1890 |
|
|
if (!CONST_INT_P (x1))
|
1891 |
|
|
return -1;
|
1892 |
|
|
xsize /= INTVAL (x1);
|
1893 |
|
|
ysize /= INTVAL (x1);
|
1894 |
|
|
c /= INTVAL (x1);
|
1895 |
|
|
return memrefs_conflict_p (xsize, x0, ysize, y0, c);
|
1896 |
|
|
}
|
1897 |
|
|
|
1898 |
|
|
default:
|
1899 |
|
|
break;
|
1900 |
|
|
}
|
1901 |
|
|
|
1902 |
|
|
/* Treat an access through an AND (e.g. a subword access on an Alpha)
|
1903 |
|
|
as an access with indeterminate size. Assume that references
|
1904 |
|
|
besides AND are aligned, so if the size of the other reference is
|
1905 |
|
|
at least as large as the alignment, assume no other overlap. */
|
1906 |
|
|
if (GET_CODE (x) == AND && CONST_INT_P (XEXP (x, 1)))
|
1907 |
|
|
{
|
1908 |
|
|
if (GET_CODE (y) == AND || ysize < -INTVAL (XEXP (x, 1)))
|
1909 |
|
|
xsize = -1;
|
1910 |
|
|
return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)), ysize, y, c);
|
1911 |
|
|
}
|
1912 |
|
|
if (GET_CODE (y) == AND && CONST_INT_P (XEXP (y, 1)))
|
1913 |
|
|
{
|
1914 |
|
|
/* ??? If we are indexing far enough into the array/structure, we
|
1915 |
|
|
may yet be able to determine that we can not overlap. But we
|
1916 |
|
|
also need to that we are far enough from the end not to overlap
|
1917 |
|
|
a following reference, so we do nothing with that for now. */
|
1918 |
|
|
if (GET_CODE (x) == AND || xsize < -INTVAL (XEXP (y, 1)))
|
1919 |
|
|
ysize = -1;
|
1920 |
|
|
return memrefs_conflict_p (xsize, x, ysize, canon_rtx (XEXP (y, 0)), c);
|
1921 |
|
|
}
|
1922 |
|
|
|
1923 |
|
|
if (CONSTANT_P (x))
|
1924 |
|
|
{
|
1925 |
|
|
if (CONST_INT_P (x) && CONST_INT_P (y))
|
1926 |
|
|
{
|
1927 |
|
|
c += (INTVAL (y) - INTVAL (x));
|
1928 |
|
|
return (xsize <= 0 || ysize <= 0
|
1929 |
|
|
|| (c >= 0 && xsize > c) || (c < 0 && ysize+c > 0));
|
1930 |
|
|
}
|
1931 |
|
|
|
1932 |
|
|
if (GET_CODE (x) == CONST)
|
1933 |
|
|
{
|
1934 |
|
|
if (GET_CODE (y) == CONST)
|
1935 |
|
|
return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)),
|
1936 |
|
|
ysize, canon_rtx (XEXP (y, 0)), c);
|
1937 |
|
|
else
|
1938 |
|
|
return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)),
|
1939 |
|
|
ysize, y, c);
|
1940 |
|
|
}
|
1941 |
|
|
if (GET_CODE (y) == CONST)
|
1942 |
|
|
return memrefs_conflict_p (xsize, x, ysize,
|
1943 |
|
|
canon_rtx (XEXP (y, 0)), c);
|
1944 |
|
|
|
1945 |
|
|
if (CONSTANT_P (y))
|
1946 |
|
|
return (xsize <= 0 || ysize <= 0
|
1947 |
|
|
|| (rtx_equal_for_memref_p (x, y)
|
1948 |
|
|
&& ((c >= 0 && xsize > c) || (c < 0 && ysize+c > 0))));
|
1949 |
|
|
|
1950 |
|
|
return -1;
|
1951 |
|
|
}
|
1952 |
|
|
|
1953 |
|
|
return -1;
|
1954 |
|
|
}
|
1955 |
|
|
|
1956 |
|
|
/* Functions to compute memory dependencies.
|
1957 |
|
|
|
1958 |
|
|
Since we process the insns in execution order, we can build tables
|
1959 |
|
|
to keep track of what registers are fixed (and not aliased), what registers
|
1960 |
|
|
are varying in known ways, and what registers are varying in unknown
|
1961 |
|
|
ways.
|
1962 |
|
|
|
1963 |
|
|
If both memory references are volatile, then there must always be a
|
1964 |
|
|
dependence between the two references, since their order can not be
|
1965 |
|
|
changed. A volatile and non-volatile reference can be interchanged
|
1966 |
|
|
though.
|
1967 |
|
|
|
1968 |
|
|
A MEM_IN_STRUCT reference at a non-AND varying address can never
|
1969 |
|
|
conflict with a non-MEM_IN_STRUCT reference at a fixed address. We
|
1970 |
|
|
also must allow AND addresses, because they may generate accesses
|
1971 |
|
|
outside the object being referenced. This is used to generate
|
1972 |
|
|
aligned addresses from unaligned addresses, for instance, the alpha
|
1973 |
|
|
storeqi_unaligned pattern. */
|
1974 |
|
|
|
1975 |
|
|
/* Read dependence: X is read after read in MEM takes place. There can
|
1976 |
|
|
only be a dependence here if both reads are volatile. */
|
1977 |
|
|
|
1978 |
|
|
int
|
1979 |
|
|
read_dependence (const_rtx mem, const_rtx x)
|
1980 |
|
|
{
|
1981 |
|
|
return MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem);
|
1982 |
|
|
}
|
1983 |
|
|
|
1984 |
|
|
/* Returns MEM1 if and only if MEM1 is a scalar at a fixed address and
|
1985 |
|
|
MEM2 is a reference to a structure at a varying address, or returns
|
1986 |
|
|
MEM2 if vice versa. Otherwise, returns NULL_RTX. If a non-NULL
|
1987 |
|
|
value is returned MEM1 and MEM2 can never alias. VARIES_P is used
|
1988 |
|
|
to decide whether or not an address may vary; it should return
|
1989 |
|
|
nonzero whenever variation is possible.
|
1990 |
|
|
MEM1_ADDR and MEM2_ADDR are the addresses of MEM1 and MEM2. */
|
1991 |
|
|
|
1992 |
|
|
static const_rtx
|
1993 |
|
|
fixed_scalar_and_varying_struct_p (const_rtx mem1, const_rtx mem2, rtx mem1_addr,
|
1994 |
|
|
rtx mem2_addr,
|
1995 |
|
|
bool (*varies_p) (const_rtx, bool))
|
1996 |
|
|
{
|
1997 |
|
|
if (! flag_strict_aliasing)
|
1998 |
|
|
return NULL_RTX;
|
1999 |
|
|
|
2000 |
|
|
if (MEM_ALIAS_SET (mem2)
|
2001 |
|
|
&& MEM_SCALAR_P (mem1) && MEM_IN_STRUCT_P (mem2)
|
2002 |
|
|
&& !varies_p (mem1_addr, 1) && varies_p (mem2_addr, 1))
|
2003 |
|
|
/* MEM1 is a scalar at a fixed address; MEM2 is a struct at a
|
2004 |
|
|
varying address. */
|
2005 |
|
|
return mem1;
|
2006 |
|
|
|
2007 |
|
|
if (MEM_ALIAS_SET (mem1)
|
2008 |
|
|
&& MEM_IN_STRUCT_P (mem1) && MEM_SCALAR_P (mem2)
|
2009 |
|
|
&& varies_p (mem1_addr, 1) && !varies_p (mem2_addr, 1))
|
2010 |
|
|
/* MEM2 is a scalar at a fixed address; MEM1 is a struct at a
|
2011 |
|
|
varying address. */
|
2012 |
|
|
return mem2;
|
2013 |
|
|
|
2014 |
|
|
return NULL_RTX;
|
2015 |
|
|
}
|
2016 |
|
|
|
2017 |
|
|
/* Returns nonzero if something about the mode or address format MEM1
|
2018 |
|
|
indicates that it might well alias *anything*. */
|
2019 |
|
|
|
2020 |
|
|
static int
|
2021 |
|
|
aliases_everything_p (const_rtx mem)
|
2022 |
|
|
{
|
2023 |
|
|
if (GET_CODE (XEXP (mem, 0)) == AND)
|
2024 |
|
|
/* If the address is an AND, it's very hard to know at what it is
|
2025 |
|
|
actually pointing. */
|
2026 |
|
|
return 1;
|
2027 |
|
|
|
2028 |
|
|
return 0;
|
2029 |
|
|
}
|
2030 |
|
|
|
2031 |
|
|
/* Return true if we can determine that the fields referenced cannot
|
2032 |
|
|
overlap for any pair of objects. */
|
2033 |
|
|
|
2034 |
|
|
static bool
|
2035 |
|
|
nonoverlapping_component_refs_p (const_tree x, const_tree y)
|
2036 |
|
|
{
|
2037 |
|
|
const_tree fieldx, fieldy, typex, typey, orig_y;
|
2038 |
|
|
|
2039 |
|
|
if (!flag_strict_aliasing)
|
2040 |
|
|
return false;
|
2041 |
|
|
|
2042 |
|
|
do
|
2043 |
|
|
{
|
2044 |
|
|
/* The comparison has to be done at a common type, since we don't
|
2045 |
|
|
know how the inheritance hierarchy works. */
|
2046 |
|
|
orig_y = y;
|
2047 |
|
|
do
|
2048 |
|
|
{
|
2049 |
|
|
fieldx = TREE_OPERAND (x, 1);
|
2050 |
|
|
typex = TYPE_MAIN_VARIANT (DECL_FIELD_CONTEXT (fieldx));
|
2051 |
|
|
|
2052 |
|
|
y = orig_y;
|
2053 |
|
|
do
|
2054 |
|
|
{
|
2055 |
|
|
fieldy = TREE_OPERAND (y, 1);
|
2056 |
|
|
typey = TYPE_MAIN_VARIANT (DECL_FIELD_CONTEXT (fieldy));
|
2057 |
|
|
|
2058 |
|
|
if (typex == typey)
|
2059 |
|
|
goto found;
|
2060 |
|
|
|
2061 |
|
|
y = TREE_OPERAND (y, 0);
|
2062 |
|
|
}
|
2063 |
|
|
while (y && TREE_CODE (y) == COMPONENT_REF);
|
2064 |
|
|
|
2065 |
|
|
x = TREE_OPERAND (x, 0);
|
2066 |
|
|
}
|
2067 |
|
|
while (x && TREE_CODE (x) == COMPONENT_REF);
|
2068 |
|
|
/* Never found a common type. */
|
2069 |
|
|
return false;
|
2070 |
|
|
|
2071 |
|
|
found:
|
2072 |
|
|
/* If we're left with accessing different fields of a structure,
|
2073 |
|
|
then no overlap. */
|
2074 |
|
|
if (TREE_CODE (typex) == RECORD_TYPE
|
2075 |
|
|
&& fieldx != fieldy)
|
2076 |
|
|
return true;
|
2077 |
|
|
|
2078 |
|
|
/* The comparison on the current field failed. If we're accessing
|
2079 |
|
|
a very nested structure, look at the next outer level. */
|
2080 |
|
|
x = TREE_OPERAND (x, 0);
|
2081 |
|
|
y = TREE_OPERAND (y, 0);
|
2082 |
|
|
}
|
2083 |
|
|
while (x && y
|
2084 |
|
|
&& TREE_CODE (x) == COMPONENT_REF
|
2085 |
|
|
&& TREE_CODE (y) == COMPONENT_REF);
|
2086 |
|
|
|
2087 |
|
|
return false;
|
2088 |
|
|
}
|
2089 |
|
|
|
2090 |
|
|
/* Look at the bottom of the COMPONENT_REF list for a DECL, and return it. */
|
2091 |
|
|
|
2092 |
|
|
static tree
|
2093 |
|
|
decl_for_component_ref (tree x)
|
2094 |
|
|
{
|
2095 |
|
|
do
|
2096 |
|
|
{
|
2097 |
|
|
x = TREE_OPERAND (x, 0);
|
2098 |
|
|
}
|
2099 |
|
|
while (x && TREE_CODE (x) == COMPONENT_REF);
|
2100 |
|
|
|
2101 |
|
|
return x && DECL_P (x) ? x : NULL_TREE;
|
2102 |
|
|
}
|
2103 |
|
|
|
2104 |
|
|
/* Walk up the COMPONENT_REF list and adjust OFFSET to compensate for the
|
2105 |
|
|
offset of the field reference. */
|
2106 |
|
|
|
2107 |
|
|
static rtx
|
2108 |
|
|
adjust_offset_for_component_ref (tree x, rtx offset)
|
2109 |
|
|
{
|
2110 |
|
|
HOST_WIDE_INT ioffset;
|
2111 |
|
|
|
2112 |
|
|
if (! offset)
|
2113 |
|
|
return NULL_RTX;
|
2114 |
|
|
|
2115 |
|
|
ioffset = INTVAL (offset);
|
2116 |
|
|
do
|
2117 |
|
|
{
|
2118 |
|
|
tree offset = component_ref_field_offset (x);
|
2119 |
|
|
tree field = TREE_OPERAND (x, 1);
|
2120 |
|
|
|
2121 |
|
|
if (! host_integerp (offset, 1))
|
2122 |
|
|
return NULL_RTX;
|
2123 |
|
|
ioffset += (tree_low_cst (offset, 1)
|
2124 |
|
|
+ (tree_low_cst (DECL_FIELD_BIT_OFFSET (field), 1)
|
2125 |
|
|
/ BITS_PER_UNIT));
|
2126 |
|
|
|
2127 |
|
|
x = TREE_OPERAND (x, 0);
|
2128 |
|
|
}
|
2129 |
|
|
while (x && TREE_CODE (x) == COMPONENT_REF);
|
2130 |
|
|
|
2131 |
|
|
return GEN_INT (ioffset);
|
2132 |
|
|
}
|
2133 |
|
|
|
2134 |
|
|
/* Return nonzero if we can determine the exprs corresponding to memrefs
|
2135 |
|
|
X and Y and they do not overlap. */
|
2136 |
|
|
|
2137 |
|
|
int
|
2138 |
|
|
nonoverlapping_memrefs_p (const_rtx x, const_rtx y)
|
2139 |
|
|
{
|
2140 |
|
|
tree exprx = MEM_EXPR (x), expry = MEM_EXPR (y);
|
2141 |
|
|
rtx rtlx, rtly;
|
2142 |
|
|
rtx basex, basey;
|
2143 |
|
|
rtx moffsetx, moffsety;
|
2144 |
|
|
HOST_WIDE_INT offsetx = 0, offsety = 0, sizex, sizey, tem;
|
2145 |
|
|
|
2146 |
|
|
/* Unless both have exprs, we can't tell anything. */
|
2147 |
|
|
if (exprx == 0 || expry == 0)
|
2148 |
|
|
return 0;
|
2149 |
|
|
|
2150 |
|
|
/* For spill-slot accesses make sure we have valid offsets. */
|
2151 |
|
|
if ((exprx == get_spill_slot_decl (false)
|
2152 |
|
|
&& ! MEM_OFFSET (x))
|
2153 |
|
|
|| (expry == get_spill_slot_decl (false)
|
2154 |
|
|
&& ! MEM_OFFSET (y)))
|
2155 |
|
|
return 0;
|
2156 |
|
|
|
2157 |
|
|
/* If both are field references, we may be able to determine something. */
|
2158 |
|
|
if (TREE_CODE (exprx) == COMPONENT_REF
|
2159 |
|
|
&& TREE_CODE (expry) == COMPONENT_REF
|
2160 |
|
|
&& nonoverlapping_component_refs_p (exprx, expry))
|
2161 |
|
|
return 1;
|
2162 |
|
|
|
2163 |
|
|
|
2164 |
|
|
/* If the field reference test failed, look at the DECLs involved. */
|
2165 |
|
|
moffsetx = MEM_OFFSET (x);
|
2166 |
|
|
if (TREE_CODE (exprx) == COMPONENT_REF)
|
2167 |
|
|
{
|
2168 |
|
|
if (TREE_CODE (expry) == VAR_DECL
|
2169 |
|
|
&& POINTER_TYPE_P (TREE_TYPE (expry)))
|
2170 |
|
|
{
|
2171 |
|
|
tree field = TREE_OPERAND (exprx, 1);
|
2172 |
|
|
tree fieldcontext = DECL_FIELD_CONTEXT (field);
|
2173 |
|
|
if (ipa_type_escape_field_does_not_clobber_p (fieldcontext,
|
2174 |
|
|
TREE_TYPE (field)))
|
2175 |
|
|
return 1;
|
2176 |
|
|
}
|
2177 |
|
|
{
|
2178 |
|
|
tree t = decl_for_component_ref (exprx);
|
2179 |
|
|
if (! t)
|
2180 |
|
|
return 0;
|
2181 |
|
|
moffsetx = adjust_offset_for_component_ref (exprx, moffsetx);
|
2182 |
|
|
exprx = t;
|
2183 |
|
|
}
|
2184 |
|
|
}
|
2185 |
|
|
else if (INDIRECT_REF_P (exprx))
|
2186 |
|
|
{
|
2187 |
|
|
exprx = TREE_OPERAND (exprx, 0);
|
2188 |
|
|
if (flag_argument_noalias < 2
|
2189 |
|
|
|| TREE_CODE (exprx) != PARM_DECL)
|
2190 |
|
|
return 0;
|
2191 |
|
|
}
|
2192 |
|
|
|
2193 |
|
|
moffsety = MEM_OFFSET (y);
|
2194 |
|
|
if (TREE_CODE (expry) == COMPONENT_REF)
|
2195 |
|
|
{
|
2196 |
|
|
if (TREE_CODE (exprx) == VAR_DECL
|
2197 |
|
|
&& POINTER_TYPE_P (TREE_TYPE (exprx)))
|
2198 |
|
|
{
|
2199 |
|
|
tree field = TREE_OPERAND (expry, 1);
|
2200 |
|
|
tree fieldcontext = DECL_FIELD_CONTEXT (field);
|
2201 |
|
|
if (ipa_type_escape_field_does_not_clobber_p (fieldcontext,
|
2202 |
|
|
TREE_TYPE (field)))
|
2203 |
|
|
return 1;
|
2204 |
|
|
}
|
2205 |
|
|
{
|
2206 |
|
|
tree t = decl_for_component_ref (expry);
|
2207 |
|
|
if (! t)
|
2208 |
|
|
return 0;
|
2209 |
|
|
moffsety = adjust_offset_for_component_ref (expry, moffsety);
|
2210 |
|
|
expry = t;
|
2211 |
|
|
}
|
2212 |
|
|
}
|
2213 |
|
|
else if (INDIRECT_REF_P (expry))
|
2214 |
|
|
{
|
2215 |
|
|
expry = TREE_OPERAND (expry, 0);
|
2216 |
|
|
if (flag_argument_noalias < 2
|
2217 |
|
|
|| TREE_CODE (expry) != PARM_DECL)
|
2218 |
|
|
return 0;
|
2219 |
|
|
}
|
2220 |
|
|
|
2221 |
|
|
if (! DECL_P (exprx) || ! DECL_P (expry))
|
2222 |
|
|
return 0;
|
2223 |
|
|
|
2224 |
|
|
/* With invalid code we can end up storing into the constant pool.
|
2225 |
|
|
Bail out to avoid ICEing when creating RTL for this.
|
2226 |
|
|
See gfortran.dg/lto/20091028-2_0.f90. */
|
2227 |
|
|
if (TREE_CODE (exprx) == CONST_DECL
|
2228 |
|
|
|| TREE_CODE (expry) == CONST_DECL)
|
2229 |
|
|
return 1;
|
2230 |
|
|
|
2231 |
|
|
rtlx = DECL_RTL (exprx);
|
2232 |
|
|
rtly = DECL_RTL (expry);
|
2233 |
|
|
|
2234 |
|
|
/* If either RTL is not a MEM, it must be a REG or CONCAT, meaning they
|
2235 |
|
|
can't overlap unless they are the same because we never reuse that part
|
2236 |
|
|
of the stack frame used for locals for spilled pseudos. */
|
2237 |
|
|
if ((!MEM_P (rtlx) || !MEM_P (rtly))
|
2238 |
|
|
&& ! rtx_equal_p (rtlx, rtly))
|
2239 |
|
|
return 1;
|
2240 |
|
|
|
2241 |
|
|
/* If we have MEMs refering to different address spaces (which can
|
2242 |
|
|
potentially overlap), we cannot easily tell from the addresses
|
2243 |
|
|
whether the references overlap. */
|
2244 |
|
|
if (MEM_P (rtlx) && MEM_P (rtly)
|
2245 |
|
|
&& MEM_ADDR_SPACE (rtlx) != MEM_ADDR_SPACE (rtly))
|
2246 |
|
|
return 0;
|
2247 |
|
|
|
2248 |
|
|
/* Get the base and offsets of both decls. If either is a register, we
|
2249 |
|
|
know both are and are the same, so use that as the base. The only
|
2250 |
|
|
we can avoid overlap is if we can deduce that they are nonoverlapping
|
2251 |
|
|
pieces of that decl, which is very rare. */
|
2252 |
|
|
basex = MEM_P (rtlx) ? XEXP (rtlx, 0) : rtlx;
|
2253 |
|
|
if (GET_CODE (basex) == PLUS && CONST_INT_P (XEXP (basex, 1)))
|
2254 |
|
|
offsetx = INTVAL (XEXP (basex, 1)), basex = XEXP (basex, 0);
|
2255 |
|
|
|
2256 |
|
|
basey = MEM_P (rtly) ? XEXP (rtly, 0) : rtly;
|
2257 |
|
|
if (GET_CODE (basey) == PLUS && CONST_INT_P (XEXP (basey, 1)))
|
2258 |
|
|
offsety = INTVAL (XEXP (basey, 1)), basey = XEXP (basey, 0);
|
2259 |
|
|
|
2260 |
|
|
/* If the bases are different, we know they do not overlap if both
|
2261 |
|
|
are constants or if one is a constant and the other a pointer into the
|
2262 |
|
|
stack frame. Otherwise a different base means we can't tell if they
|
2263 |
|
|
overlap or not. */
|
2264 |
|
|
if (! rtx_equal_p (basex, basey))
|
2265 |
|
|
return ((CONSTANT_P (basex) && CONSTANT_P (basey))
|
2266 |
|
|
|| (CONSTANT_P (basex) && REG_P (basey)
|
2267 |
|
|
&& REGNO_PTR_FRAME_P (REGNO (basey)))
|
2268 |
|
|
|| (CONSTANT_P (basey) && REG_P (basex)
|
2269 |
|
|
&& REGNO_PTR_FRAME_P (REGNO (basex))));
|
2270 |
|
|
|
2271 |
|
|
sizex = (!MEM_P (rtlx) ? (int) GET_MODE_SIZE (GET_MODE (rtlx))
|
2272 |
|
|
: MEM_SIZE (rtlx) ? INTVAL (MEM_SIZE (rtlx))
|
2273 |
|
|
: -1);
|
2274 |
|
|
sizey = (!MEM_P (rtly) ? (int) GET_MODE_SIZE (GET_MODE (rtly))
|
2275 |
|
|
: MEM_SIZE (rtly) ? INTVAL (MEM_SIZE (rtly)) :
|
2276 |
|
|
-1);
|
2277 |
|
|
|
2278 |
|
|
/* If we have an offset for either memref, it can update the values computed
|
2279 |
|
|
above. */
|
2280 |
|
|
if (moffsetx)
|
2281 |
|
|
offsetx += INTVAL (moffsetx), sizex -= INTVAL (moffsetx);
|
2282 |
|
|
if (moffsety)
|
2283 |
|
|
offsety += INTVAL (moffsety), sizey -= INTVAL (moffsety);
|
2284 |
|
|
|
2285 |
|
|
/* If a memref has both a size and an offset, we can use the smaller size.
|
2286 |
|
|
We can't do this if the offset isn't known because we must view this
|
2287 |
|
|
memref as being anywhere inside the DECL's MEM. */
|
2288 |
|
|
if (MEM_SIZE (x) && moffsetx)
|
2289 |
|
|
sizex = INTVAL (MEM_SIZE (x));
|
2290 |
|
|
if (MEM_SIZE (y) && moffsety)
|
2291 |
|
|
sizey = INTVAL (MEM_SIZE (y));
|
2292 |
|
|
|
2293 |
|
|
/* Put the values of the memref with the lower offset in X's values. */
|
2294 |
|
|
if (offsetx > offsety)
|
2295 |
|
|
{
|
2296 |
|
|
tem = offsetx, offsetx = offsety, offsety = tem;
|
2297 |
|
|
tem = sizex, sizex = sizey, sizey = tem;
|
2298 |
|
|
}
|
2299 |
|
|
|
2300 |
|
|
/* If we don't know the size of the lower-offset value, we can't tell
|
2301 |
|
|
if they conflict. Otherwise, we do the test. */
|
2302 |
|
|
return sizex >= 0 && offsety >= offsetx + sizex;
|
2303 |
|
|
}
|
2304 |
|
|
|
2305 |
|
|
/* True dependence: X is read after store in MEM takes place. */
|
2306 |
|
|
|
2307 |
|
|
int
|
2308 |
|
|
true_dependence (const_rtx mem, enum machine_mode mem_mode, const_rtx x,
|
2309 |
|
|
bool (*varies) (const_rtx, bool))
|
2310 |
|
|
{
|
2311 |
|
|
rtx x_addr, mem_addr;
|
2312 |
|
|
rtx base;
|
2313 |
|
|
int ret;
|
2314 |
|
|
|
2315 |
|
|
if (MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|
2316 |
|
|
return 1;
|
2317 |
|
|
|
2318 |
|
|
/* (mem:BLK (scratch)) is a special mechanism to conflict with everything.
|
2319 |
|
|
This is used in epilogue deallocation functions, and in cselib. */
|
2320 |
|
|
if (GET_MODE (x) == BLKmode && GET_CODE (XEXP (x, 0)) == SCRATCH)
|
2321 |
|
|
return 1;
|
2322 |
|
|
if (GET_MODE (mem) == BLKmode && GET_CODE (XEXP (mem, 0)) == SCRATCH)
|
2323 |
|
|
return 1;
|
2324 |
|
|
if (MEM_ALIAS_SET (x) == ALIAS_SET_MEMORY_BARRIER
|
2325 |
|
|
|| MEM_ALIAS_SET (mem) == ALIAS_SET_MEMORY_BARRIER)
|
2326 |
|
|
return 1;
|
2327 |
|
|
|
2328 |
|
|
/* Read-only memory is by definition never modified, and therefore can't
|
2329 |
|
|
conflict with anything. We don't expect to find read-only set on MEM,
|
2330 |
|
|
but stupid user tricks can produce them, so don't die. */
|
2331 |
|
|
if (MEM_READONLY_P (x))
|
2332 |
|
|
return 0;
|
2333 |
|
|
|
2334 |
|
|
/* If we have MEMs refering to different address spaces (which can
|
2335 |
|
|
potentially overlap), we cannot easily tell from the addresses
|
2336 |
|
|
whether the references overlap. */
|
2337 |
|
|
if (MEM_ADDR_SPACE (mem) != MEM_ADDR_SPACE (x))
|
2338 |
|
|
return 1;
|
2339 |
|
|
|
2340 |
|
|
if (mem_mode == VOIDmode)
|
2341 |
|
|
mem_mode = GET_MODE (mem);
|
2342 |
|
|
|
2343 |
|
|
x_addr = XEXP (x, 0);
|
2344 |
|
|
mem_addr = XEXP (mem, 0);
|
2345 |
|
|
if (!((GET_CODE (x_addr) == VALUE
|
2346 |
|
|
&& GET_CODE (mem_addr) != VALUE
|
2347 |
|
|
&& reg_mentioned_p (x_addr, mem_addr))
|
2348 |
|
|
|| (GET_CODE (x_addr) != VALUE
|
2349 |
|
|
&& GET_CODE (mem_addr) == VALUE
|
2350 |
|
|
&& reg_mentioned_p (mem_addr, x_addr))))
|
2351 |
|
|
{
|
2352 |
|
|
x_addr = get_addr (x_addr);
|
2353 |
|
|
mem_addr = get_addr (mem_addr);
|
2354 |
|
|
}
|
2355 |
|
|
|
2356 |
|
|
base = find_base_term (x_addr);
|
2357 |
|
|
if (base && (GET_CODE (base) == LABEL_REF
|
2358 |
|
|
|| (GET_CODE (base) == SYMBOL_REF
|
2359 |
|
|
&& CONSTANT_POOL_ADDRESS_P (base))))
|
2360 |
|
|
return 0;
|
2361 |
|
|
|
2362 |
|
|
if (! base_alias_check (x_addr, mem_addr, GET_MODE (x), mem_mode))
|
2363 |
|
|
return 0;
|
2364 |
|
|
|
2365 |
|
|
x_addr = canon_rtx (x_addr);
|
2366 |
|
|
mem_addr = canon_rtx (mem_addr);
|
2367 |
|
|
|
2368 |
|
|
if ((ret = memrefs_conflict_p (GET_MODE_SIZE (mem_mode), mem_addr,
|
2369 |
|
|
SIZE_FOR_MODE (x), x_addr, 0)) != -1)
|
2370 |
|
|
return ret;
|
2371 |
|
|
|
2372 |
|
|
if (DIFFERENT_ALIAS_SETS_P (x, mem))
|
2373 |
|
|
return 0;
|
2374 |
|
|
|
2375 |
|
|
if (nonoverlapping_memrefs_p (mem, x))
|
2376 |
|
|
return 0;
|
2377 |
|
|
|
2378 |
|
|
if (aliases_everything_p (x))
|
2379 |
|
|
return 1;
|
2380 |
|
|
|
2381 |
|
|
/* We cannot use aliases_everything_p to test MEM, since we must look
|
2382 |
|
|
at MEM_MODE, rather than GET_MODE (MEM). */
|
2383 |
|
|
if (mem_mode == QImode || GET_CODE (mem_addr) == AND)
|
2384 |
|
|
return 1;
|
2385 |
|
|
|
2386 |
|
|
/* In true_dependence we also allow BLKmode to alias anything. Why
|
2387 |
|
|
don't we do this in anti_dependence and output_dependence? */
|
2388 |
|
|
if (mem_mode == BLKmode || GET_MODE (x) == BLKmode)
|
2389 |
|
|
return 1;
|
2390 |
|
|
|
2391 |
|
|
if (fixed_scalar_and_varying_struct_p (mem, x, mem_addr, x_addr, varies))
|
2392 |
|
|
return 0;
|
2393 |
|
|
|
2394 |
|
|
return rtx_refs_may_alias_p (x, mem, true);
|
2395 |
|
|
}
|
2396 |
|
|
|
2397 |
|
|
/* Canonical true dependence: X is read after store in MEM takes place.
|
2398 |
|
|
Variant of true_dependence which assumes MEM has already been
|
2399 |
|
|
canonicalized (hence we no longer do that here).
|
2400 |
|
|
The mem_addr argument has been added, since true_dependence computed
|
2401 |
|
|
this value prior to canonicalizing.
|
2402 |
|
|
If x_addr is non-NULL, it is used in preference of XEXP (x, 0). */
|
2403 |
|
|
|
2404 |
|
|
int
|
2405 |
|
|
canon_true_dependence (const_rtx mem, enum machine_mode mem_mode, rtx mem_addr,
|
2406 |
|
|
const_rtx x, rtx x_addr, bool (*varies) (const_rtx, bool))
|
2407 |
|
|
{
|
2408 |
|
|
int ret;
|
2409 |
|
|
|
2410 |
|
|
if (MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|
2411 |
|
|
return 1;
|
2412 |
|
|
|
2413 |
|
|
/* (mem:BLK (scratch)) is a special mechanism to conflict with everything.
|
2414 |
|
|
This is used in epilogue deallocation functions. */
|
2415 |
|
|
if (GET_MODE (x) == BLKmode && GET_CODE (XEXP (x, 0)) == SCRATCH)
|
2416 |
|
|
return 1;
|
2417 |
|
|
if (GET_MODE (mem) == BLKmode && GET_CODE (XEXP (mem, 0)) == SCRATCH)
|
2418 |
|
|
return 1;
|
2419 |
|
|
if (MEM_ALIAS_SET (x) == ALIAS_SET_MEMORY_BARRIER
|
2420 |
|
|
|| MEM_ALIAS_SET (mem) == ALIAS_SET_MEMORY_BARRIER)
|
2421 |
|
|
return 1;
|
2422 |
|
|
|
2423 |
|
|
/* Read-only memory is by definition never modified, and therefore can't
|
2424 |
|
|
conflict with anything. We don't expect to find read-only set on MEM,
|
2425 |
|
|
but stupid user tricks can produce them, so don't die. */
|
2426 |
|
|
if (MEM_READONLY_P (x))
|
2427 |
|
|
return 0;
|
2428 |
|
|
|
2429 |
|
|
/* If we have MEMs refering to different address spaces (which can
|
2430 |
|
|
potentially overlap), we cannot easily tell from the addresses
|
2431 |
|
|
whether the references overlap. */
|
2432 |
|
|
if (MEM_ADDR_SPACE (mem) != MEM_ADDR_SPACE (x))
|
2433 |
|
|
return 1;
|
2434 |
|
|
|
2435 |
|
|
if (! x_addr)
|
2436 |
|
|
{
|
2437 |
|
|
x_addr = XEXP (x, 0);
|
2438 |
|
|
if (!((GET_CODE (x_addr) == VALUE
|
2439 |
|
|
&& GET_CODE (mem_addr) != VALUE
|
2440 |
|
|
&& reg_mentioned_p (x_addr, mem_addr))
|
2441 |
|
|
|| (GET_CODE (x_addr) != VALUE
|
2442 |
|
|
&& GET_CODE (mem_addr) == VALUE
|
2443 |
|
|
&& reg_mentioned_p (mem_addr, x_addr))))
|
2444 |
|
|
x_addr = get_addr (x_addr);
|
2445 |
|
|
}
|
2446 |
|
|
|
2447 |
|
|
if (! base_alias_check (x_addr, mem_addr, GET_MODE (x), mem_mode))
|
2448 |
|
|
return 0;
|
2449 |
|
|
|
2450 |
|
|
x_addr = canon_rtx (x_addr);
|
2451 |
|
|
if ((ret = memrefs_conflict_p (GET_MODE_SIZE (mem_mode), mem_addr,
|
2452 |
|
|
SIZE_FOR_MODE (x), x_addr, 0)) != -1)
|
2453 |
|
|
return ret;
|
2454 |
|
|
|
2455 |
|
|
if (DIFFERENT_ALIAS_SETS_P (x, mem))
|
2456 |
|
|
return 0;
|
2457 |
|
|
|
2458 |
|
|
if (nonoverlapping_memrefs_p (x, mem))
|
2459 |
|
|
return 0;
|
2460 |
|
|
|
2461 |
|
|
if (aliases_everything_p (x))
|
2462 |
|
|
return 1;
|
2463 |
|
|
|
2464 |
|
|
/* We cannot use aliases_everything_p to test MEM, since we must look
|
2465 |
|
|
at MEM_MODE, rather than GET_MODE (MEM). */
|
2466 |
|
|
if (mem_mode == QImode || GET_CODE (mem_addr) == AND)
|
2467 |
|
|
return 1;
|
2468 |
|
|
|
2469 |
|
|
/* In true_dependence we also allow BLKmode to alias anything. Why
|
2470 |
|
|
don't we do this in anti_dependence and output_dependence? */
|
2471 |
|
|
if (mem_mode == BLKmode || GET_MODE (x) == BLKmode)
|
2472 |
|
|
return 1;
|
2473 |
|
|
|
2474 |
|
|
if (fixed_scalar_and_varying_struct_p (mem, x, mem_addr, x_addr, varies))
|
2475 |
|
|
return 0;
|
2476 |
|
|
|
2477 |
|
|
return rtx_refs_may_alias_p (x, mem, true);
|
2478 |
|
|
}
|
2479 |
|
|
|
2480 |
|
|
/* Returns nonzero if a write to X might alias a previous read from
|
2481 |
|
|
(or, if WRITEP is nonzero, a write to) MEM. */
|
2482 |
|
|
|
2483 |
|
|
static int
|
2484 |
|
|
write_dependence_p (const_rtx mem, const_rtx x, int writep)
|
2485 |
|
|
{
|
2486 |
|
|
rtx x_addr, mem_addr;
|
2487 |
|
|
const_rtx fixed_scalar;
|
2488 |
|
|
rtx base;
|
2489 |
|
|
int ret;
|
2490 |
|
|
|
2491 |
|
|
if (MEM_VOLATILE_P (x) && MEM_VOLATILE_P (mem))
|
2492 |
|
|
return 1;
|
2493 |
|
|
|
2494 |
|
|
/* (mem:BLK (scratch)) is a special mechanism to conflict with everything.
|
2495 |
|
|
This is used in epilogue deallocation functions. */
|
2496 |
|
|
if (GET_MODE (x) == BLKmode && GET_CODE (XEXP (x, 0)) == SCRATCH)
|
2497 |
|
|
return 1;
|
2498 |
|
|
if (GET_MODE (mem) == BLKmode && GET_CODE (XEXP (mem, 0)) == SCRATCH)
|
2499 |
|
|
return 1;
|
2500 |
|
|
if (MEM_ALIAS_SET (x) == ALIAS_SET_MEMORY_BARRIER
|
2501 |
|
|
|| MEM_ALIAS_SET (mem) == ALIAS_SET_MEMORY_BARRIER)
|
2502 |
|
|
return 1;
|
2503 |
|
|
|
2504 |
|
|
/* A read from read-only memory can't conflict with read-write memory. */
|
2505 |
|
|
if (!writep && MEM_READONLY_P (mem))
|
2506 |
|
|
return 0;
|
2507 |
|
|
|
2508 |
|
|
/* If we have MEMs refering to different address spaces (which can
|
2509 |
|
|
potentially overlap), we cannot easily tell from the addresses
|
2510 |
|
|
whether the references overlap. */
|
2511 |
|
|
if (MEM_ADDR_SPACE (mem) != MEM_ADDR_SPACE (x))
|
2512 |
|
|
return 1;
|
2513 |
|
|
|
2514 |
|
|
x_addr = XEXP (x, 0);
|
2515 |
|
|
mem_addr = XEXP (mem, 0);
|
2516 |
|
|
if (!((GET_CODE (x_addr) == VALUE
|
2517 |
|
|
&& GET_CODE (mem_addr) != VALUE
|
2518 |
|
|
&& reg_mentioned_p (x_addr, mem_addr))
|
2519 |
|
|
|| (GET_CODE (x_addr) != VALUE
|
2520 |
|
|
&& GET_CODE (mem_addr) == VALUE
|
2521 |
|
|
&& reg_mentioned_p (mem_addr, x_addr))))
|
2522 |
|
|
{
|
2523 |
|
|
x_addr = get_addr (x_addr);
|
2524 |
|
|
mem_addr = get_addr (mem_addr);
|
2525 |
|
|
}
|
2526 |
|
|
|
2527 |
|
|
if (! writep)
|
2528 |
|
|
{
|
2529 |
|
|
base = find_base_term (mem_addr);
|
2530 |
|
|
if (base && (GET_CODE (base) == LABEL_REF
|
2531 |
|
|
|| (GET_CODE (base) == SYMBOL_REF
|
2532 |
|
|
&& CONSTANT_POOL_ADDRESS_P (base))))
|
2533 |
|
|
return 0;
|
2534 |
|
|
}
|
2535 |
|
|
|
2536 |
|
|
if (! base_alias_check (x_addr, mem_addr, GET_MODE (x),
|
2537 |
|
|
GET_MODE (mem)))
|
2538 |
|
|
return 0;
|
2539 |
|
|
|
2540 |
|
|
x_addr = canon_rtx (x_addr);
|
2541 |
|
|
mem_addr = canon_rtx (mem_addr);
|
2542 |
|
|
|
2543 |
|
|
if ((ret = memrefs_conflict_p (SIZE_FOR_MODE (mem), mem_addr,
|
2544 |
|
|
SIZE_FOR_MODE (x), x_addr, 0)) != -1)
|
2545 |
|
|
return ret;
|
2546 |
|
|
|
2547 |
|
|
if (nonoverlapping_memrefs_p (x, mem))
|
2548 |
|
|
return 0;
|
2549 |
|
|
|
2550 |
|
|
fixed_scalar
|
2551 |
|
|
= fixed_scalar_and_varying_struct_p (mem, x, mem_addr, x_addr,
|
2552 |
|
|
rtx_addr_varies_p);
|
2553 |
|
|
|
2554 |
|
|
if ((fixed_scalar == mem && !aliases_everything_p (x))
|
2555 |
|
|
|| (fixed_scalar == x && !aliases_everything_p (mem)))
|
2556 |
|
|
return 0;
|
2557 |
|
|
|
2558 |
|
|
return rtx_refs_may_alias_p (x, mem, false);
|
2559 |
|
|
}
|
2560 |
|
|
|
2561 |
|
|
/* Anti dependence: X is written after read in MEM takes place. */
|
2562 |
|
|
|
2563 |
|
|
int
|
2564 |
|
|
anti_dependence (const_rtx mem, const_rtx x)
|
2565 |
|
|
{
|
2566 |
|
|
return write_dependence_p (mem, x, /*writep=*/0);
|
2567 |
|
|
}
|
2568 |
|
|
|
2569 |
|
|
/* Output dependence: X is written after store in MEM takes place. */
|
2570 |
|
|
|
2571 |
|
|
int
|
2572 |
|
|
output_dependence (const_rtx mem, const_rtx x)
|
2573 |
|
|
{
|
2574 |
|
|
return write_dependence_p (mem, x, /*writep=*/1);
|
2575 |
|
|
}
|
2576 |
|
|
|
2577 |
|
|
|
2578 |
|
|
void
|
2579 |
|
|
init_alias_target (void)
|
2580 |
|
|
{
|
2581 |
|
|
int i;
|
2582 |
|
|
|
2583 |
|
|
memset (static_reg_base_value, 0, sizeof static_reg_base_value);
|
2584 |
|
|
|
2585 |
|
|
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
|
2586 |
|
|
/* Check whether this register can hold an incoming pointer
|
2587 |
|
|
argument. FUNCTION_ARG_REGNO_P tests outgoing register
|
2588 |
|
|
numbers, so translate if necessary due to register windows. */
|
2589 |
|
|
if (FUNCTION_ARG_REGNO_P (OUTGOING_REGNO (i))
|
2590 |
|
|
&& HARD_REGNO_MODE_OK (i, Pmode))
|
2591 |
|
|
static_reg_base_value[i]
|
2592 |
|
|
= gen_rtx_ADDRESS (VOIDmode, gen_rtx_REG (Pmode, i));
|
2593 |
|
|
|
2594 |
|
|
static_reg_base_value[STACK_POINTER_REGNUM]
|
2595 |
|
|
= gen_rtx_ADDRESS (Pmode, stack_pointer_rtx);
|
2596 |
|
|
static_reg_base_value[ARG_POINTER_REGNUM]
|
2597 |
|
|
= gen_rtx_ADDRESS (Pmode, arg_pointer_rtx);
|
2598 |
|
|
static_reg_base_value[FRAME_POINTER_REGNUM]
|
2599 |
|
|
= gen_rtx_ADDRESS (Pmode, frame_pointer_rtx);
|
2600 |
|
|
#if HARD_FRAME_POINTER_REGNUM != FRAME_POINTER_REGNUM
|
2601 |
|
|
static_reg_base_value[HARD_FRAME_POINTER_REGNUM]
|
2602 |
|
|
= gen_rtx_ADDRESS (Pmode, hard_frame_pointer_rtx);
|
2603 |
|
|
#endif
|
2604 |
|
|
}
|
2605 |
|
|
|
2606 |
|
|
/* Set MEMORY_MODIFIED when X modifies DATA (that is assumed
|
2607 |
|
|
to be memory reference. */
|
2608 |
|
|
static bool memory_modified;
|
2609 |
|
|
static void
|
2610 |
|
|
memory_modified_1 (rtx x, const_rtx pat ATTRIBUTE_UNUSED, void *data)
|
2611 |
|
|
{
|
2612 |
|
|
if (MEM_P (x))
|
2613 |
|
|
{
|
2614 |
|
|
if (anti_dependence (x, (const_rtx)data) || output_dependence (x, (const_rtx)data))
|
2615 |
|
|
memory_modified = true;
|
2616 |
|
|
}
|
2617 |
|
|
}
|
2618 |
|
|
|
2619 |
|
|
|
2620 |
|
|
/* Return true when INSN possibly modify memory contents of MEM
|
2621 |
|
|
(i.e. address can be modified). */
|
2622 |
|
|
bool
|
2623 |
|
|
memory_modified_in_insn_p (const_rtx mem, const_rtx insn)
|
2624 |
|
|
{
|
2625 |
|
|
if (!INSN_P (insn))
|
2626 |
|
|
return false;
|
2627 |
|
|
memory_modified = false;
|
2628 |
|
|
note_stores (PATTERN (insn), memory_modified_1, CONST_CAST_RTX(mem));
|
2629 |
|
|
return memory_modified;
|
2630 |
|
|
}
|
2631 |
|
|
|
2632 |
|
|
/* Initialize the aliasing machinery. Initialize the REG_KNOWN_VALUE
|
2633 |
|
|
array. */
|
2634 |
|
|
|
2635 |
|
|
void
|
2636 |
|
|
init_alias_analysis (void)
|
2637 |
|
|
{
|
2638 |
|
|
unsigned int maxreg = max_reg_num ();
|
2639 |
|
|
int changed, pass;
|
2640 |
|
|
int i;
|
2641 |
|
|
unsigned int ui;
|
2642 |
|
|
rtx insn;
|
2643 |
|
|
|
2644 |
|
|
timevar_push (TV_ALIAS_ANALYSIS);
|
2645 |
|
|
|
2646 |
|
|
reg_known_value_size = maxreg - FIRST_PSEUDO_REGISTER;
|
2647 |
|
|
reg_known_value = GGC_CNEWVEC (rtx, reg_known_value_size);
|
2648 |
|
|
reg_known_equiv_p = XCNEWVEC (bool, reg_known_value_size);
|
2649 |
|
|
|
2650 |
|
|
/* If we have memory allocated from the previous run, use it. */
|
2651 |
|
|
if (old_reg_base_value)
|
2652 |
|
|
reg_base_value = old_reg_base_value;
|
2653 |
|
|
|
2654 |
|
|
if (reg_base_value)
|
2655 |
|
|
VEC_truncate (rtx, reg_base_value, 0);
|
2656 |
|
|
|
2657 |
|
|
VEC_safe_grow_cleared (rtx, gc, reg_base_value, maxreg);
|
2658 |
|
|
|
2659 |
|
|
new_reg_base_value = XNEWVEC (rtx, maxreg);
|
2660 |
|
|
reg_seen = XNEWVEC (char, maxreg);
|
2661 |
|
|
|
2662 |
|
|
/* The basic idea is that each pass through this loop will use the
|
2663 |
|
|
"constant" information from the previous pass to propagate alias
|
2664 |
|
|
information through another level of assignments.
|
2665 |
|
|
|
2666 |
|
|
This could get expensive if the assignment chains are long. Maybe
|
2667 |
|
|
we should throttle the number of iterations, possibly based on
|
2668 |
|
|
the optimization level or flag_expensive_optimizations.
|
2669 |
|
|
|
2670 |
|
|
We could propagate more information in the first pass by making use
|
2671 |
|
|
of DF_REG_DEF_COUNT to determine immediately that the alias information
|
2672 |
|
|
for a pseudo is "constant".
|
2673 |
|
|
|
2674 |
|
|
A program with an uninitialized variable can cause an infinite loop
|
2675 |
|
|
here. Instead of doing a full dataflow analysis to detect such problems
|
2676 |
|
|
we just cap the number of iterations for the loop.
|
2677 |
|
|
|
2678 |
|
|
The state of the arrays for the set chain in question does not matter
|
2679 |
|
|
since the program has undefined behavior. */
|
2680 |
|
|
|
2681 |
|
|
pass = 0;
|
2682 |
|
|
do
|
2683 |
|
|
{
|
2684 |
|
|
/* Assume nothing will change this iteration of the loop. */
|
2685 |
|
|
changed = 0;
|
2686 |
|
|
|
2687 |
|
|
/* We want to assign the same IDs each iteration of this loop, so
|
2688 |
|
|
start counting from zero each iteration of the loop. */
|
2689 |
|
|
unique_id = 0;
|
2690 |
|
|
|
2691 |
|
|
/* We're at the start of the function each iteration through the
|
2692 |
|
|
loop, so we're copying arguments. */
|
2693 |
|
|
copying_arguments = true;
|
2694 |
|
|
|
2695 |
|
|
/* Wipe the potential alias information clean for this pass. */
|
2696 |
|
|
memset (new_reg_base_value, 0, maxreg * sizeof (rtx));
|
2697 |
|
|
|
2698 |
|
|
/* Wipe the reg_seen array clean. */
|
2699 |
|
|
memset (reg_seen, 0, maxreg);
|
2700 |
|
|
|
2701 |
|
|
/* Mark all hard registers which may contain an address.
|
2702 |
|
|
The stack, frame and argument pointers may contain an address.
|
2703 |
|
|
An argument register which can hold a Pmode value may contain
|
2704 |
|
|
an address even if it is not in BASE_REGS.
|
2705 |
|
|
|
2706 |
|
|
The address expression is VOIDmode for an argument and
|
2707 |
|
|
Pmode for other registers. */
|
2708 |
|
|
|
2709 |
|
|
memcpy (new_reg_base_value, static_reg_base_value,
|
2710 |
|
|
FIRST_PSEUDO_REGISTER * sizeof (rtx));
|
2711 |
|
|
|
2712 |
|
|
/* Walk the insns adding values to the new_reg_base_value array. */
|
2713 |
|
|
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
|
2714 |
|
|
{
|
2715 |
|
|
if (INSN_P (insn))
|
2716 |
|
|
{
|
2717 |
|
|
rtx note, set;
|
2718 |
|
|
|
2719 |
|
|
#if defined (HAVE_prologue) || defined (HAVE_epilogue)
|
2720 |
|
|
/* The prologue/epilogue insns are not threaded onto the
|
2721 |
|
|
insn chain until after reload has completed. Thus,
|
2722 |
|
|
there is no sense wasting time checking if INSN is in
|
2723 |
|
|
the prologue/epilogue until after reload has completed. */
|
2724 |
|
|
if (reload_completed
|
2725 |
|
|
&& prologue_epilogue_contains (insn))
|
2726 |
|
|
continue;
|
2727 |
|
|
#endif
|
2728 |
|
|
|
2729 |
|
|
/* If this insn has a noalias note, process it, Otherwise,
|
2730 |
|
|
scan for sets. A simple set will have no side effects
|
2731 |
|
|
which could change the base value of any other register. */
|
2732 |
|
|
|
2733 |
|
|
if (GET_CODE (PATTERN (insn)) == SET
|
2734 |
|
|
&& REG_NOTES (insn) != 0
|
2735 |
|
|
&& find_reg_note (insn, REG_NOALIAS, NULL_RTX))
|
2736 |
|
|
record_set (SET_DEST (PATTERN (insn)), NULL_RTX, NULL);
|
2737 |
|
|
else
|
2738 |
|
|
note_stores (PATTERN (insn), record_set, NULL);
|
2739 |
|
|
|
2740 |
|
|
set = single_set (insn);
|
2741 |
|
|
|
2742 |
|
|
if (set != 0
|
2743 |
|
|
&& REG_P (SET_DEST (set))
|
2744 |
|
|
&& REGNO (SET_DEST (set)) >= FIRST_PSEUDO_REGISTER)
|
2745 |
|
|
{
|
2746 |
|
|
unsigned int regno = REGNO (SET_DEST (set));
|
2747 |
|
|
rtx src = SET_SRC (set);
|
2748 |
|
|
rtx t;
|
2749 |
|
|
|
2750 |
|
|
note = find_reg_equal_equiv_note (insn);
|
2751 |
|
|
if (note && REG_NOTE_KIND (note) == REG_EQUAL
|
2752 |
|
|
&& DF_REG_DEF_COUNT (regno) != 1)
|
2753 |
|
|
note = NULL_RTX;
|
2754 |
|
|
|
2755 |
|
|
if (note != NULL_RTX
|
2756 |
|
|
&& GET_CODE (XEXP (note, 0)) != EXPR_LIST
|
2757 |
|
|
&& ! rtx_varies_p (XEXP (note, 0), 1)
|
2758 |
|
|
&& ! reg_overlap_mentioned_p (SET_DEST (set),
|
2759 |
|
|
XEXP (note, 0)))
|
2760 |
|
|
{
|
2761 |
|
|
set_reg_known_value (regno, XEXP (note, 0));
|
2762 |
|
|
set_reg_known_equiv_p (regno,
|
2763 |
|
|
REG_NOTE_KIND (note) == REG_EQUIV);
|
2764 |
|
|
}
|
2765 |
|
|
else if (DF_REG_DEF_COUNT (regno) == 1
|
2766 |
|
|
&& GET_CODE (src) == PLUS
|
2767 |
|
|
&& REG_P (XEXP (src, 0))
|
2768 |
|
|
&& (t = get_reg_known_value (REGNO (XEXP (src, 0))))
|
2769 |
|
|
&& CONST_INT_P (XEXP (src, 1)))
|
2770 |
|
|
{
|
2771 |
|
|
t = plus_constant (t, INTVAL (XEXP (src, 1)));
|
2772 |
|
|
set_reg_known_value (regno, t);
|
2773 |
|
|
set_reg_known_equiv_p (regno, 0);
|
2774 |
|
|
}
|
2775 |
|
|
else if (DF_REG_DEF_COUNT (regno) == 1
|
2776 |
|
|
&& ! rtx_varies_p (src, 1))
|
2777 |
|
|
{
|
2778 |
|
|
set_reg_known_value (regno, src);
|
2779 |
|
|
set_reg_known_equiv_p (regno, 0);
|
2780 |
|
|
}
|
2781 |
|
|
}
|
2782 |
|
|
}
|
2783 |
|
|
else if (NOTE_P (insn)
|
2784 |
|
|
&& NOTE_KIND (insn) == NOTE_INSN_FUNCTION_BEG)
|
2785 |
|
|
copying_arguments = false;
|
2786 |
|
|
}
|
2787 |
|
|
|
2788 |
|
|
/* Now propagate values from new_reg_base_value to reg_base_value. */
|
2789 |
|
|
gcc_assert (maxreg == (unsigned int) max_reg_num ());
|
2790 |
|
|
|
2791 |
|
|
for (ui = 0; ui < maxreg; ui++)
|
2792 |
|
|
{
|
2793 |
|
|
if (new_reg_base_value[ui]
|
2794 |
|
|
&& new_reg_base_value[ui] != VEC_index (rtx, reg_base_value, ui)
|
2795 |
|
|
&& ! rtx_equal_p (new_reg_base_value[ui],
|
2796 |
|
|
VEC_index (rtx, reg_base_value, ui)))
|
2797 |
|
|
{
|
2798 |
|
|
VEC_replace (rtx, reg_base_value, ui, new_reg_base_value[ui]);
|
2799 |
|
|
changed = 1;
|
2800 |
|
|
}
|
2801 |
|
|
}
|
2802 |
|
|
}
|
2803 |
|
|
while (changed && ++pass < MAX_ALIAS_LOOP_PASSES);
|
2804 |
|
|
|
2805 |
|
|
/* Fill in the remaining entries. */
|
2806 |
|
|
for (i = 0; i < (int)reg_known_value_size; i++)
|
2807 |
|
|
if (reg_known_value[i] == 0)
|
2808 |
|
|
reg_known_value[i] = regno_reg_rtx[i + FIRST_PSEUDO_REGISTER];
|
2809 |
|
|
|
2810 |
|
|
/* Clean up. */
|
2811 |
|
|
free (new_reg_base_value);
|
2812 |
|
|
new_reg_base_value = 0;
|
2813 |
|
|
free (reg_seen);
|
2814 |
|
|
reg_seen = 0;
|
2815 |
|
|
timevar_pop (TV_ALIAS_ANALYSIS);
|
2816 |
|
|
}
|
2817 |
|
|
|
2818 |
|
|
void
|
2819 |
|
|
end_alias_analysis (void)
|
2820 |
|
|
{
|
2821 |
|
|
old_reg_base_value = reg_base_value;
|
2822 |
|
|
ggc_free (reg_known_value);
|
2823 |
|
|
reg_known_value = 0;
|
2824 |
|
|
reg_known_value_size = 0;
|
2825 |
|
|
free (reg_known_equiv_p);
|
2826 |
|
|
reg_known_equiv_p = 0;
|
2827 |
|
|
}
|
2828 |
|
|
|
2829 |
|
|
#include "gt-alias.h"
|