1 |
38 |
julius |
/* Copy propagation and SSA_NAME replacement support routines.
|
2 |
|
|
Copyright (C) 2004, 2005, 2007 Free Software Foundation, Inc.
|
3 |
|
|
|
4 |
|
|
This file is part of GCC.
|
5 |
|
|
|
6 |
|
|
GCC is free software; you can redistribute it and/or modify
|
7 |
|
|
it under the terms of the GNU General Public License as published by
|
8 |
|
|
the Free Software Foundation; either version 3, or (at your option)
|
9 |
|
|
any later version.
|
10 |
|
|
|
11 |
|
|
GCC is distributed in the hope that it will be useful,
|
12 |
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
14 |
|
|
GNU General Public License for more details.
|
15 |
|
|
|
16 |
|
|
You should have received a copy of the GNU General Public License
|
17 |
|
|
along with GCC; see the file COPYING3. If not see
|
18 |
|
|
<http://www.gnu.org/licenses/>. */
|
19 |
|
|
|
20 |
|
|
#include "config.h"
|
21 |
|
|
#include "system.h"
|
22 |
|
|
#include "coretypes.h"
|
23 |
|
|
#include "tm.h"
|
24 |
|
|
#include "tree.h"
|
25 |
|
|
#include "flags.h"
|
26 |
|
|
#include "rtl.h"
|
27 |
|
|
#include "tm_p.h"
|
28 |
|
|
#include "ggc.h"
|
29 |
|
|
#include "basic-block.h"
|
30 |
|
|
#include "output.h"
|
31 |
|
|
#include "expr.h"
|
32 |
|
|
#include "function.h"
|
33 |
|
|
#include "diagnostic.h"
|
34 |
|
|
#include "timevar.h"
|
35 |
|
|
#include "tree-dump.h"
|
36 |
|
|
#include "tree-flow.h"
|
37 |
|
|
#include "tree-pass.h"
|
38 |
|
|
#include "tree-ssa-propagate.h"
|
39 |
|
|
#include "langhooks.h"
|
40 |
|
|
|
41 |
|
|
/* This file implements the copy propagation pass and provides a
|
42 |
|
|
handful of interfaces for performing const/copy propagation and
|
43 |
|
|
simple expression replacement which keep variable annotations
|
44 |
|
|
up-to-date.
|
45 |
|
|
|
46 |
|
|
We require that for any copy operation where the RHS and LHS have
|
47 |
|
|
a non-null memory tag the memory tag be the same. It is OK
|
48 |
|
|
for one or both of the memory tags to be NULL.
|
49 |
|
|
|
50 |
|
|
We also require tracking if a variable is dereferenced in a load or
|
51 |
|
|
store operation.
|
52 |
|
|
|
53 |
|
|
We enforce these requirements by having all copy propagation and
|
54 |
|
|
replacements of one SSA_NAME with a different SSA_NAME to use the
|
55 |
|
|
APIs defined in this file. */
|
56 |
|
|
|
57 |
|
|
/* Return true if we may propagate ORIG into DEST, false otherwise. */
|
58 |
|
|
|
59 |
|
|
bool
|
60 |
|
|
may_propagate_copy (tree dest, tree orig)
|
61 |
|
|
{
|
62 |
|
|
tree type_d = TREE_TYPE (dest);
|
63 |
|
|
tree type_o = TREE_TYPE (orig);
|
64 |
|
|
|
65 |
|
|
/* Do not copy between types for which we *do* need a conversion. */
|
66 |
|
|
if (!tree_ssa_useless_type_conversion_1 (type_d, type_o))
|
67 |
|
|
return false;
|
68 |
|
|
|
69 |
|
|
/* FIXME. GIMPLE is allowing pointer assignments and comparisons of
|
70 |
|
|
pointers that have different alias sets. This means that these
|
71 |
|
|
pointers will have different memory tags associated to them.
|
72 |
|
|
|
73 |
|
|
If we allow copy propagation in these cases, statements de-referencing
|
74 |
|
|
the new pointer will now have a reference to a different memory tag
|
75 |
|
|
with potentially incorrect SSA information.
|
76 |
|
|
|
77 |
|
|
This was showing up in libjava/java/util/zip/ZipFile.java with code
|
78 |
|
|
like:
|
79 |
|
|
|
80 |
|
|
struct java.io.BufferedInputStream *T.660;
|
81 |
|
|
struct java.io.BufferedInputStream *T.647;
|
82 |
|
|
struct java.io.InputStream *is;
|
83 |
|
|
struct java.io.InputStream *is.662;
|
84 |
|
|
[ ... ]
|
85 |
|
|
T.660 = T.647;
|
86 |
|
|
is = T.660; <-- This ought to be type-casted
|
87 |
|
|
is.662 = is;
|
88 |
|
|
|
89 |
|
|
Also, f/name.c exposed a similar problem with a COND_EXPR predicate
|
90 |
|
|
that was causing DOM to generate and equivalence with two pointers of
|
91 |
|
|
alias-incompatible types:
|
92 |
|
|
|
93 |
|
|
struct _ffename_space *n;
|
94 |
|
|
struct _ffename *ns;
|
95 |
|
|
[ ... ]
|
96 |
|
|
if (n == ns)
|
97 |
|
|
goto lab;
|
98 |
|
|
...
|
99 |
|
|
lab:
|
100 |
|
|
return n;
|
101 |
|
|
|
102 |
|
|
I think that GIMPLE should emit the appropriate type-casts. For the
|
103 |
|
|
time being, blocking copy-propagation in these cases is the safe thing
|
104 |
|
|
to do. */
|
105 |
|
|
if (TREE_CODE (dest) == SSA_NAME
|
106 |
|
|
&& TREE_CODE (orig) == SSA_NAME
|
107 |
|
|
&& POINTER_TYPE_P (type_d)
|
108 |
|
|
&& POINTER_TYPE_P (type_o))
|
109 |
|
|
{
|
110 |
|
|
tree mt_dest = var_ann (SSA_NAME_VAR (dest))->symbol_mem_tag;
|
111 |
|
|
tree mt_orig = var_ann (SSA_NAME_VAR (orig))->symbol_mem_tag;
|
112 |
|
|
if (mt_dest && mt_orig && mt_dest != mt_orig)
|
113 |
|
|
return false;
|
114 |
|
|
else if (!lang_hooks.types_compatible_p (type_d, type_o))
|
115 |
|
|
return false;
|
116 |
|
|
else if (get_alias_set (TREE_TYPE (type_d)) !=
|
117 |
|
|
get_alias_set (TREE_TYPE (type_o)))
|
118 |
|
|
return false;
|
119 |
|
|
|
120 |
|
|
/* Also verify flow-sensitive information is compatible. */
|
121 |
|
|
if (SSA_NAME_PTR_INFO (orig) && SSA_NAME_PTR_INFO (dest))
|
122 |
|
|
{
|
123 |
|
|
struct ptr_info_def *orig_ptr_info = SSA_NAME_PTR_INFO (orig);
|
124 |
|
|
struct ptr_info_def *dest_ptr_info = SSA_NAME_PTR_INFO (dest);
|
125 |
|
|
|
126 |
|
|
if (orig_ptr_info->name_mem_tag
|
127 |
|
|
&& dest_ptr_info->name_mem_tag
|
128 |
|
|
&& orig_ptr_info->pt_vars
|
129 |
|
|
&& dest_ptr_info->pt_vars
|
130 |
|
|
&& !bitmap_intersect_p (dest_ptr_info->pt_vars,
|
131 |
|
|
orig_ptr_info->pt_vars))
|
132 |
|
|
return false;
|
133 |
|
|
}
|
134 |
|
|
}
|
135 |
|
|
|
136 |
|
|
/* If the destination is a SSA_NAME for a virtual operand, then we have
|
137 |
|
|
some special cases to handle. */
|
138 |
|
|
if (TREE_CODE (dest) == SSA_NAME && !is_gimple_reg (dest))
|
139 |
|
|
{
|
140 |
|
|
/* If both operands are SSA_NAMEs referring to virtual operands, then
|
141 |
|
|
we can always propagate. */
|
142 |
|
|
if (TREE_CODE (orig) == SSA_NAME
|
143 |
|
|
&& !is_gimple_reg (orig))
|
144 |
|
|
return true;
|
145 |
|
|
|
146 |
|
|
/* We have a "copy" from something like a constant into a virtual
|
147 |
|
|
operand. Reject these. */
|
148 |
|
|
return false;
|
149 |
|
|
}
|
150 |
|
|
|
151 |
|
|
/* If ORIG flows in from an abnormal edge, it cannot be propagated. */
|
152 |
|
|
if (TREE_CODE (orig) == SSA_NAME
|
153 |
|
|
&& SSA_NAME_OCCURS_IN_ABNORMAL_PHI (orig))
|
154 |
|
|
return false;
|
155 |
|
|
|
156 |
|
|
/* If DEST is an SSA_NAME that flows from an abnormal edge, then it
|
157 |
|
|
cannot be replaced. */
|
158 |
|
|
if (TREE_CODE (dest) == SSA_NAME
|
159 |
|
|
&& SSA_NAME_OCCURS_IN_ABNORMAL_PHI (dest))
|
160 |
|
|
return false;
|
161 |
|
|
|
162 |
|
|
/* Anything else is OK. */
|
163 |
|
|
return true;
|
164 |
|
|
}
|
165 |
|
|
|
166 |
|
|
/* Similarly, but we know that we're propagating into an ASM_EXPR. */
|
167 |
|
|
|
168 |
|
|
bool
|
169 |
|
|
may_propagate_copy_into_asm (tree dest)
|
170 |
|
|
{
|
171 |
|
|
/* Hard register operands of asms are special. Do not bypass. */
|
172 |
|
|
return !(TREE_CODE (dest) == SSA_NAME
|
173 |
|
|
&& TREE_CODE (SSA_NAME_VAR (dest)) == VAR_DECL
|
174 |
|
|
&& DECL_HARD_REGISTER (SSA_NAME_VAR (dest)));
|
175 |
|
|
}
|
176 |
|
|
|
177 |
|
|
|
178 |
|
|
/* Given two SSA_NAMEs pointers ORIG and NEW such that we are copy
|
179 |
|
|
propagating NEW into ORIG, consolidate aliasing information so that
|
180 |
|
|
they both share the same memory tags. */
|
181 |
|
|
|
182 |
|
|
void
|
183 |
|
|
merge_alias_info (tree orig, tree new)
|
184 |
|
|
{
|
185 |
|
|
tree new_sym = SSA_NAME_VAR (new);
|
186 |
|
|
tree orig_sym = SSA_NAME_VAR (orig);
|
187 |
|
|
var_ann_t new_ann = var_ann (new_sym);
|
188 |
|
|
var_ann_t orig_ann = var_ann (orig_sym);
|
189 |
|
|
|
190 |
|
|
gcc_assert (POINTER_TYPE_P (TREE_TYPE (orig)));
|
191 |
|
|
gcc_assert (POINTER_TYPE_P (TREE_TYPE (new)));
|
192 |
|
|
|
193 |
|
|
#if defined ENABLE_CHECKING
|
194 |
|
|
gcc_assert (lang_hooks.types_compatible_p (TREE_TYPE (orig),
|
195 |
|
|
TREE_TYPE (new)));
|
196 |
|
|
|
197 |
|
|
/* If the pointed-to alias sets are different, these two pointers
|
198 |
|
|
would never have the same memory tag. In this case, NEW should
|
199 |
|
|
not have been propagated into ORIG. */
|
200 |
|
|
gcc_assert (get_alias_set (TREE_TYPE (TREE_TYPE (new_sym)))
|
201 |
|
|
== get_alias_set (TREE_TYPE (TREE_TYPE (orig_sym))));
|
202 |
|
|
#endif
|
203 |
|
|
|
204 |
|
|
/* Synchronize the symbol tags. If both pointers had a tag and they
|
205 |
|
|
are different, then something has gone wrong. Symbol tags can
|
206 |
|
|
always be merged because they are flow insensitive, all the SSA
|
207 |
|
|
names of the same base DECL share the same symbol tag. */
|
208 |
|
|
if (new_ann->symbol_mem_tag == NULL_TREE)
|
209 |
|
|
new_ann->symbol_mem_tag = orig_ann->symbol_mem_tag;
|
210 |
|
|
else if (orig_ann->symbol_mem_tag == NULL_TREE)
|
211 |
|
|
orig_ann->symbol_mem_tag = new_ann->symbol_mem_tag;
|
212 |
|
|
else
|
213 |
|
|
gcc_assert (new_ann->symbol_mem_tag == orig_ann->symbol_mem_tag);
|
214 |
|
|
|
215 |
|
|
/* Check that flow-sensitive information is compatible. Notice that
|
216 |
|
|
we may not merge flow-sensitive information here. This function
|
217 |
|
|
is called when propagating equivalences dictated by the IL, like
|
218 |
|
|
a copy operation P_i = Q_j, and from equivalences dictated by
|
219 |
|
|
control-flow, like if (P_i == Q_j).
|
220 |
|
|
|
221 |
|
|
In the former case, P_i and Q_j are equivalent in every block
|
222 |
|
|
dominated by the assignment, so their flow-sensitive information
|
223 |
|
|
is always the same. However, in the latter case, the pointers
|
224 |
|
|
P_i and Q_j are only equivalent in one of the sub-graphs out of
|
225 |
|
|
the predicate, so their flow-sensitive information is not the
|
226 |
|
|
same in every block dominated by the predicate.
|
227 |
|
|
|
228 |
|
|
Since we cannot distinguish one case from another in this
|
229 |
|
|
function, we can only make sure that if P_i and Q_j have
|
230 |
|
|
flow-sensitive information, they should be compatible. */
|
231 |
|
|
if (SSA_NAME_PTR_INFO (orig) && SSA_NAME_PTR_INFO (new))
|
232 |
|
|
{
|
233 |
|
|
struct ptr_info_def *orig_ptr_info = SSA_NAME_PTR_INFO (orig);
|
234 |
|
|
struct ptr_info_def *new_ptr_info = SSA_NAME_PTR_INFO (new);
|
235 |
|
|
|
236 |
|
|
/* Note that pointer NEW and ORIG may actually have different
|
237 |
|
|
pointed-to variables (e.g., PR 18291 represented in
|
238 |
|
|
testsuite/gcc.c-torture/compile/pr18291.c). However, since
|
239 |
|
|
NEW is being copy-propagated into ORIG, it must always be
|
240 |
|
|
true that the pointed-to set for pointer NEW is the same, or
|
241 |
|
|
a subset, of the pointed-to set for pointer ORIG. If this
|
242 |
|
|
isn't the case, we shouldn't have been able to do the
|
243 |
|
|
propagation of NEW into ORIG. */
|
244 |
|
|
if (orig_ptr_info->name_mem_tag
|
245 |
|
|
&& new_ptr_info->name_mem_tag
|
246 |
|
|
&& orig_ptr_info->pt_vars
|
247 |
|
|
&& new_ptr_info->pt_vars)
|
248 |
|
|
gcc_assert (bitmap_intersect_p (new_ptr_info->pt_vars,
|
249 |
|
|
orig_ptr_info->pt_vars));
|
250 |
|
|
}
|
251 |
|
|
}
|
252 |
|
|
|
253 |
|
|
|
254 |
|
|
/* Common code for propagate_value and replace_exp.
|
255 |
|
|
|
256 |
|
|
Replace use operand OP_P with VAL. FOR_PROPAGATION indicates if the
|
257 |
|
|
replacement is done to propagate a value or not. */
|
258 |
|
|
|
259 |
|
|
static void
|
260 |
|
|
replace_exp_1 (use_operand_p op_p, tree val,
|
261 |
|
|
bool for_propagation ATTRIBUTE_UNUSED)
|
262 |
|
|
{
|
263 |
|
|
tree op = USE_FROM_PTR (op_p);
|
264 |
|
|
|
265 |
|
|
#if defined ENABLE_CHECKING
|
266 |
|
|
gcc_assert (!(for_propagation
|
267 |
|
|
&& TREE_CODE (op) == SSA_NAME
|
268 |
|
|
&& TREE_CODE (val) == SSA_NAME
|
269 |
|
|
&& !may_propagate_copy (op, val)));
|
270 |
|
|
#endif
|
271 |
|
|
|
272 |
|
|
if (TREE_CODE (val) == SSA_NAME)
|
273 |
|
|
{
|
274 |
|
|
if (TREE_CODE (op) == SSA_NAME && POINTER_TYPE_P (TREE_TYPE (op)))
|
275 |
|
|
merge_alias_info (op, val);
|
276 |
|
|
SET_USE (op_p, val);
|
277 |
|
|
}
|
278 |
|
|
else
|
279 |
|
|
SET_USE (op_p, unsave_expr_now (val));
|
280 |
|
|
}
|
281 |
|
|
|
282 |
|
|
|
283 |
|
|
/* Propagate the value VAL (assumed to be a constant or another SSA_NAME)
|
284 |
|
|
into the operand pointed to by OP_P.
|
285 |
|
|
|
286 |
|
|
Use this version for const/copy propagation as it will perform additional
|
287 |
|
|
checks to ensure validity of the const/copy propagation. */
|
288 |
|
|
|
289 |
|
|
void
|
290 |
|
|
propagate_value (use_operand_p op_p, tree val)
|
291 |
|
|
{
|
292 |
|
|
replace_exp_1 (op_p, val, true);
|
293 |
|
|
}
|
294 |
|
|
|
295 |
|
|
|
296 |
|
|
/* Propagate the value VAL (assumed to be a constant or another SSA_NAME)
|
297 |
|
|
into the tree pointed to by OP_P.
|
298 |
|
|
|
299 |
|
|
Use this version for const/copy propagation when SSA operands are not
|
300 |
|
|
available. It will perform the additional checks to ensure validity of
|
301 |
|
|
the const/copy propagation, but will not update any operand information.
|
302 |
|
|
Be sure to mark the stmt as modified. */
|
303 |
|
|
|
304 |
|
|
void
|
305 |
|
|
propagate_tree_value (tree *op_p, tree val)
|
306 |
|
|
{
|
307 |
|
|
#if defined ENABLE_CHECKING
|
308 |
|
|
gcc_assert (!(TREE_CODE (val) == SSA_NAME
|
309 |
|
|
&& TREE_CODE (*op_p) == SSA_NAME
|
310 |
|
|
&& !may_propagate_copy (*op_p, val)));
|
311 |
|
|
#endif
|
312 |
|
|
|
313 |
|
|
if (TREE_CODE (val) == SSA_NAME)
|
314 |
|
|
{
|
315 |
|
|
if (TREE_CODE (*op_p) == SSA_NAME && POINTER_TYPE_P (TREE_TYPE (*op_p)))
|
316 |
|
|
merge_alias_info (*op_p, val);
|
317 |
|
|
*op_p = val;
|
318 |
|
|
}
|
319 |
|
|
else
|
320 |
|
|
*op_p = unsave_expr_now (val);
|
321 |
|
|
}
|
322 |
|
|
|
323 |
|
|
|
324 |
|
|
/* Replace *OP_P with value VAL (assumed to be a constant or another SSA_NAME).
|
325 |
|
|
|
326 |
|
|
Use this version when not const/copy propagating values. For example,
|
327 |
|
|
PRE uses this version when building expressions as they would appear
|
328 |
|
|
in specific blocks taking into account actions of PHI nodes. */
|
329 |
|
|
|
330 |
|
|
void
|
331 |
|
|
replace_exp (use_operand_p op_p, tree val)
|
332 |
|
|
{
|
333 |
|
|
replace_exp_1 (op_p, val, false);
|
334 |
|
|
}
|
335 |
|
|
|
336 |
|
|
|
337 |
|
|
/*---------------------------------------------------------------------------
|
338 |
|
|
Copy propagation
|
339 |
|
|
---------------------------------------------------------------------------*/
|
340 |
|
|
/* During propagation, we keep chains of variables that are copies of
|
341 |
|
|
one another. If variable X_i is a copy of X_j and X_j is a copy of
|
342 |
|
|
X_k, COPY_OF will contain:
|
343 |
|
|
|
344 |
|
|
COPY_OF[i].VALUE = X_j
|
345 |
|
|
COPY_OF[j].VALUE = X_k
|
346 |
|
|
COPY_OF[k].VALUE = X_k
|
347 |
|
|
|
348 |
|
|
After propagation, the copy-of value for each variable X_i is
|
349 |
|
|
converted into the final value by walking the copy-of chains and
|
350 |
|
|
updating COPY_OF[i].VALUE to be the last element of the chain. */
|
351 |
|
|
static prop_value_t *copy_of;
|
352 |
|
|
|
353 |
|
|
/* Used in set_copy_of_val to determine if the last link of a copy-of
|
354 |
|
|
chain has changed. */
|
355 |
|
|
static tree *cached_last_copy_of;
|
356 |
|
|
|
357 |
|
|
/* True if we are doing copy propagation on loads and stores. */
|
358 |
|
|
static bool do_store_copy_prop;
|
359 |
|
|
|
360 |
|
|
|
361 |
|
|
/* Return true if this statement may generate a useful copy. */
|
362 |
|
|
|
363 |
|
|
static bool
|
364 |
|
|
stmt_may_generate_copy (tree stmt)
|
365 |
|
|
{
|
366 |
|
|
tree lhs, rhs;
|
367 |
|
|
stmt_ann_t ann;
|
368 |
|
|
|
369 |
|
|
if (TREE_CODE (stmt) == PHI_NODE)
|
370 |
|
|
return !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (PHI_RESULT (stmt));
|
371 |
|
|
|
372 |
|
|
if (TREE_CODE (stmt) != MODIFY_EXPR)
|
373 |
|
|
return false;
|
374 |
|
|
|
375 |
|
|
lhs = TREE_OPERAND (stmt, 0);
|
376 |
|
|
rhs = TREE_OPERAND (stmt, 1);
|
377 |
|
|
ann = stmt_ann (stmt);
|
378 |
|
|
|
379 |
|
|
/* If the statement has volatile operands, it won't generate a
|
380 |
|
|
useful copy. */
|
381 |
|
|
if (ann->has_volatile_ops)
|
382 |
|
|
return false;
|
383 |
|
|
|
384 |
|
|
/* If we are not doing store copy-prop, statements with loads and/or
|
385 |
|
|
stores will never generate a useful copy. */
|
386 |
|
|
if (!do_store_copy_prop
|
387 |
|
|
&& !ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS))
|
388 |
|
|
return false;
|
389 |
|
|
|
390 |
|
|
/* Otherwise, the only statements that generate useful copies are
|
391 |
|
|
assignments whose RHS is just an SSA name that doesn't flow
|
392 |
|
|
through abnormal edges. */
|
393 |
|
|
return (do_store_copy_prop
|
394 |
|
|
&& TREE_CODE (lhs) == SSA_NAME)
|
395 |
|
|
|| (TREE_CODE (rhs) == SSA_NAME
|
396 |
|
|
&& !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (rhs));
|
397 |
|
|
}
|
398 |
|
|
|
399 |
|
|
|
400 |
|
|
/* Return the copy-of value for VAR. */
|
401 |
|
|
|
402 |
|
|
static inline prop_value_t *
|
403 |
|
|
get_copy_of_val (tree var)
|
404 |
|
|
{
|
405 |
|
|
prop_value_t *val = ©_of[SSA_NAME_VERSION (var)];
|
406 |
|
|
|
407 |
|
|
if (val->value == NULL_TREE
|
408 |
|
|
&& !stmt_may_generate_copy (SSA_NAME_DEF_STMT (var)))
|
409 |
|
|
{
|
410 |
|
|
/* If the variable will never generate a useful copy relation,
|
411 |
|
|
make it its own copy. */
|
412 |
|
|
val->value = var;
|
413 |
|
|
val->mem_ref = NULL_TREE;
|
414 |
|
|
}
|
415 |
|
|
|
416 |
|
|
return val;
|
417 |
|
|
}
|
418 |
|
|
|
419 |
|
|
|
420 |
|
|
/* Return last link in the copy-of chain for VAR. */
|
421 |
|
|
|
422 |
|
|
static tree
|
423 |
|
|
get_last_copy_of (tree var)
|
424 |
|
|
{
|
425 |
|
|
tree last;
|
426 |
|
|
int i;
|
427 |
|
|
|
428 |
|
|
/* Traverse COPY_OF starting at VAR until we get to the last
|
429 |
|
|
link in the chain. Since it is possible to have cycles in PHI
|
430 |
|
|
nodes, the copy-of chain may also contain cycles.
|
431 |
|
|
|
432 |
|
|
To avoid infinite loops and to avoid traversing lengthy copy-of
|
433 |
|
|
chains, we artificially limit the maximum number of chains we are
|
434 |
|
|
willing to traverse.
|
435 |
|
|
|
436 |
|
|
The value 5 was taken from a compiler and runtime library
|
437 |
|
|
bootstrap and a mixture of C and C++ code from various sources.
|
438 |
|
|
More than 82% of all copy-of chains were shorter than 5 links. */
|
439 |
|
|
#define LIMIT 5
|
440 |
|
|
|
441 |
|
|
last = var;
|
442 |
|
|
for (i = 0; i < LIMIT; i++)
|
443 |
|
|
{
|
444 |
|
|
tree copy = copy_of[SSA_NAME_VERSION (last)].value;
|
445 |
|
|
if (copy == NULL_TREE || copy == last)
|
446 |
|
|
break;
|
447 |
|
|
last = copy;
|
448 |
|
|
}
|
449 |
|
|
|
450 |
|
|
/* If we have reached the limit, then we are either in a copy-of
|
451 |
|
|
cycle or the copy-of chain is too long. In this case, just
|
452 |
|
|
return VAR so that it is not considered a copy of anything. */
|
453 |
|
|
return (i < LIMIT ? last : var);
|
454 |
|
|
}
|
455 |
|
|
|
456 |
|
|
|
457 |
|
|
/* Set FIRST to be the first variable in the copy-of chain for DEST.
|
458 |
|
|
If DEST's copy-of value or its copy-of chain has changed, return
|
459 |
|
|
true.
|
460 |
|
|
|
461 |
|
|
MEM_REF is the memory reference where FIRST is stored. This is
|
462 |
|
|
used when DEST is a non-register and we are copy propagating loads
|
463 |
|
|
and stores. */
|
464 |
|
|
|
465 |
|
|
static inline bool
|
466 |
|
|
set_copy_of_val (tree dest, tree first, tree mem_ref)
|
467 |
|
|
{
|
468 |
|
|
unsigned int dest_ver = SSA_NAME_VERSION (dest);
|
469 |
|
|
tree old_first, old_last, new_last;
|
470 |
|
|
|
471 |
|
|
/* Set FIRST to be the first link in COPY_OF[DEST]. If that
|
472 |
|
|
changed, return true. */
|
473 |
|
|
old_first = copy_of[dest_ver].value;
|
474 |
|
|
copy_of[dest_ver].value = first;
|
475 |
|
|
copy_of[dest_ver].mem_ref = mem_ref;
|
476 |
|
|
|
477 |
|
|
if (old_first != first)
|
478 |
|
|
return true;
|
479 |
|
|
|
480 |
|
|
/* If FIRST and OLD_FIRST are the same, we need to check whether the
|
481 |
|
|
copy-of chain starting at FIRST ends in a different variable. If
|
482 |
|
|
the copy-of chain starting at FIRST ends up in a different
|
483 |
|
|
variable than the last cached value we had for DEST, then return
|
484 |
|
|
true because DEST is now a copy of a different variable.
|
485 |
|
|
|
486 |
|
|
This test is necessary because even though the first link in the
|
487 |
|
|
copy-of chain may not have changed, if any of the variables in
|
488 |
|
|
the copy-of chain changed its final value, DEST will now be the
|
489 |
|
|
copy of a different variable, so we have to do another round of
|
490 |
|
|
propagation for everything that depends on DEST. */
|
491 |
|
|
old_last = cached_last_copy_of[dest_ver];
|
492 |
|
|
new_last = get_last_copy_of (dest);
|
493 |
|
|
cached_last_copy_of[dest_ver] = new_last;
|
494 |
|
|
|
495 |
|
|
return (old_last != new_last);
|
496 |
|
|
}
|
497 |
|
|
|
498 |
|
|
|
499 |
|
|
/* Dump the copy-of value for variable VAR to FILE. */
|
500 |
|
|
|
501 |
|
|
static void
|
502 |
|
|
dump_copy_of (FILE *file, tree var)
|
503 |
|
|
{
|
504 |
|
|
tree val;
|
505 |
|
|
sbitmap visited;
|
506 |
|
|
|
507 |
|
|
print_generic_expr (file, var, dump_flags);
|
508 |
|
|
|
509 |
|
|
if (TREE_CODE (var) != SSA_NAME)
|
510 |
|
|
return;
|
511 |
|
|
|
512 |
|
|
visited = sbitmap_alloc (num_ssa_names);
|
513 |
|
|
sbitmap_zero (visited);
|
514 |
|
|
SET_BIT (visited, SSA_NAME_VERSION (var));
|
515 |
|
|
|
516 |
|
|
fprintf (file, " copy-of chain: ");
|
517 |
|
|
|
518 |
|
|
val = var;
|
519 |
|
|
print_generic_expr (file, val, 0);
|
520 |
|
|
fprintf (file, " ");
|
521 |
|
|
while (copy_of[SSA_NAME_VERSION (val)].value)
|
522 |
|
|
{
|
523 |
|
|
fprintf (file, "-> ");
|
524 |
|
|
val = copy_of[SSA_NAME_VERSION (val)].value;
|
525 |
|
|
print_generic_expr (file, val, 0);
|
526 |
|
|
fprintf (file, " ");
|
527 |
|
|
if (TEST_BIT (visited, SSA_NAME_VERSION (val)))
|
528 |
|
|
break;
|
529 |
|
|
SET_BIT (visited, SSA_NAME_VERSION (val));
|
530 |
|
|
}
|
531 |
|
|
|
532 |
|
|
val = get_copy_of_val (var)->value;
|
533 |
|
|
if (val == NULL_TREE)
|
534 |
|
|
fprintf (file, "[UNDEFINED]");
|
535 |
|
|
else if (val != var)
|
536 |
|
|
fprintf (file, "[COPY]");
|
537 |
|
|
else
|
538 |
|
|
fprintf (file, "[NOT A COPY]");
|
539 |
|
|
|
540 |
|
|
sbitmap_free (visited);
|
541 |
|
|
}
|
542 |
|
|
|
543 |
|
|
|
544 |
|
|
/* Evaluate the RHS of STMT. If it produces a valid copy, set the LHS
|
545 |
|
|
value and store the LHS into *RESULT_P. If STMT generates more
|
546 |
|
|
than one name (i.e., STMT is an aliased store), it is enough to
|
547 |
|
|
store the first name in the V_MAY_DEF list into *RESULT_P. After
|
548 |
|
|
all, the names generated will be VUSEd in the same statements. */
|
549 |
|
|
|
550 |
|
|
static enum ssa_prop_result
|
551 |
|
|
copy_prop_visit_assignment (tree stmt, tree *result_p)
|
552 |
|
|
{
|
553 |
|
|
tree lhs, rhs;
|
554 |
|
|
prop_value_t *rhs_val;
|
555 |
|
|
|
556 |
|
|
lhs = TREE_OPERAND (stmt, 0);
|
557 |
|
|
rhs = TREE_OPERAND (stmt, 1);
|
558 |
|
|
|
559 |
|
|
gcc_assert (TREE_CODE (rhs) == SSA_NAME);
|
560 |
|
|
|
561 |
|
|
rhs_val = get_copy_of_val (rhs);
|
562 |
|
|
|
563 |
|
|
if (TREE_CODE (lhs) == SSA_NAME)
|
564 |
|
|
{
|
565 |
|
|
/* Straight copy between two SSA names. First, make sure that
|
566 |
|
|
we can propagate the RHS into uses of LHS. */
|
567 |
|
|
if (!may_propagate_copy (lhs, rhs))
|
568 |
|
|
return SSA_PROP_VARYING;
|
569 |
|
|
|
570 |
|
|
/* Notice that in the case of assignments, we make the LHS be a
|
571 |
|
|
copy of RHS's value, not of RHS itself. This avoids keeping
|
572 |
|
|
unnecessary copy-of chains (assignments cannot be in a cycle
|
573 |
|
|
like PHI nodes), speeding up the propagation process.
|
574 |
|
|
This is different from what we do in copy_prop_visit_phi_node.
|
575 |
|
|
In those cases, we are interested in the copy-of chains. */
|
576 |
|
|
*result_p = lhs;
|
577 |
|
|
if (set_copy_of_val (*result_p, rhs_val->value, rhs_val->mem_ref))
|
578 |
|
|
return SSA_PROP_INTERESTING;
|
579 |
|
|
else
|
580 |
|
|
return SSA_PROP_NOT_INTERESTING;
|
581 |
|
|
}
|
582 |
|
|
else if (stmt_makes_single_store (stmt))
|
583 |
|
|
{
|
584 |
|
|
/* Otherwise, set the names in V_MAY_DEF/V_MUST_DEF operands
|
585 |
|
|
to be a copy of RHS. */
|
586 |
|
|
ssa_op_iter i;
|
587 |
|
|
tree vdef;
|
588 |
|
|
bool changed;
|
589 |
|
|
|
590 |
|
|
/* This should only be executed when doing store copy-prop. */
|
591 |
|
|
gcc_assert (do_store_copy_prop);
|
592 |
|
|
|
593 |
|
|
/* Set the value of every VDEF to RHS_VAL. */
|
594 |
|
|
changed = false;
|
595 |
|
|
FOR_EACH_SSA_TREE_OPERAND (vdef, stmt, i, SSA_OP_VIRTUAL_DEFS)
|
596 |
|
|
changed |= set_copy_of_val (vdef, rhs_val->value, lhs);
|
597 |
|
|
|
598 |
|
|
/* Note that for propagation purposes, we are only interested in
|
599 |
|
|
visiting statements that load the exact same memory reference
|
600 |
|
|
stored here. Those statements will have the exact same list
|
601 |
|
|
of virtual uses, so it is enough to set the output of this
|
602 |
|
|
statement to be its first virtual definition. */
|
603 |
|
|
*result_p = first_vdef (stmt);
|
604 |
|
|
|
605 |
|
|
if (changed)
|
606 |
|
|
return SSA_PROP_INTERESTING;
|
607 |
|
|
else
|
608 |
|
|
return SSA_PROP_NOT_INTERESTING;
|
609 |
|
|
}
|
610 |
|
|
|
611 |
|
|
|
612 |
|
|
return SSA_PROP_VARYING;
|
613 |
|
|
}
|
614 |
|
|
|
615 |
|
|
|
616 |
|
|
/* Visit the COND_EXPR STMT. Return SSA_PROP_INTERESTING
|
617 |
|
|
if it can determine which edge will be taken. Otherwise, return
|
618 |
|
|
SSA_PROP_VARYING. */
|
619 |
|
|
|
620 |
|
|
static enum ssa_prop_result
|
621 |
|
|
copy_prop_visit_cond_stmt (tree stmt, edge *taken_edge_p)
|
622 |
|
|
{
|
623 |
|
|
enum ssa_prop_result retval;
|
624 |
|
|
tree cond;
|
625 |
|
|
|
626 |
|
|
cond = COND_EXPR_COND (stmt);
|
627 |
|
|
retval = SSA_PROP_VARYING;
|
628 |
|
|
|
629 |
|
|
/* The only conditionals that we may be able to compute statically
|
630 |
|
|
are predicates involving two SSA_NAMEs. */
|
631 |
|
|
if (COMPARISON_CLASS_P (cond)
|
632 |
|
|
&& TREE_CODE (TREE_OPERAND (cond, 0)) == SSA_NAME
|
633 |
|
|
&& TREE_CODE (TREE_OPERAND (cond, 1)) == SSA_NAME)
|
634 |
|
|
{
|
635 |
|
|
tree op0 = get_last_copy_of (TREE_OPERAND (cond, 0));
|
636 |
|
|
tree op1 = get_last_copy_of (TREE_OPERAND (cond, 1));
|
637 |
|
|
|
638 |
|
|
/* See if we can determine the predicate's value. */
|
639 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
640 |
|
|
{
|
641 |
|
|
fprintf (dump_file, "Trying to determine truth value of ");
|
642 |
|
|
fprintf (dump_file, "predicate ");
|
643 |
|
|
print_generic_stmt (dump_file, cond, 0);
|
644 |
|
|
}
|
645 |
|
|
|
646 |
|
|
/* We can fold COND and get a useful result only when we have
|
647 |
|
|
the same SSA_NAME on both sides of a comparison operator. */
|
648 |
|
|
if (op0 == op1)
|
649 |
|
|
{
|
650 |
|
|
tree folded_cond = fold_binary (TREE_CODE (cond), boolean_type_node,
|
651 |
|
|
op0, op1);
|
652 |
|
|
if (folded_cond)
|
653 |
|
|
{
|
654 |
|
|
basic_block bb = bb_for_stmt (stmt);
|
655 |
|
|
*taken_edge_p = find_taken_edge (bb, folded_cond);
|
656 |
|
|
if (*taken_edge_p)
|
657 |
|
|
retval = SSA_PROP_INTERESTING;
|
658 |
|
|
}
|
659 |
|
|
}
|
660 |
|
|
}
|
661 |
|
|
|
662 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS) && *taken_edge_p)
|
663 |
|
|
fprintf (dump_file, "\nConditional will always take edge %d->%d\n",
|
664 |
|
|
(*taken_edge_p)->src->index, (*taken_edge_p)->dest->index);
|
665 |
|
|
|
666 |
|
|
return retval;
|
667 |
|
|
}
|
668 |
|
|
|
669 |
|
|
|
670 |
|
|
/* Evaluate statement STMT. If the statement produces a new output
|
671 |
|
|
value, return SSA_PROP_INTERESTING and store the SSA_NAME holding
|
672 |
|
|
the new value in *RESULT_P.
|
673 |
|
|
|
674 |
|
|
If STMT is a conditional branch and we can determine its truth
|
675 |
|
|
value, set *TAKEN_EDGE_P accordingly.
|
676 |
|
|
|
677 |
|
|
If the new value produced by STMT is varying, return
|
678 |
|
|
SSA_PROP_VARYING. */
|
679 |
|
|
|
680 |
|
|
static enum ssa_prop_result
|
681 |
|
|
copy_prop_visit_stmt (tree stmt, edge *taken_edge_p, tree *result_p)
|
682 |
|
|
{
|
683 |
|
|
enum ssa_prop_result retval;
|
684 |
|
|
|
685 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
686 |
|
|
{
|
687 |
|
|
fprintf (dump_file, "\nVisiting statement:\n");
|
688 |
|
|
print_generic_stmt (dump_file, stmt, dump_flags);
|
689 |
|
|
fprintf (dump_file, "\n");
|
690 |
|
|
}
|
691 |
|
|
|
692 |
|
|
if (TREE_CODE (stmt) == MODIFY_EXPR
|
693 |
|
|
&& TREE_CODE (TREE_OPERAND (stmt, 1)) == SSA_NAME
|
694 |
|
|
&& (do_store_copy_prop
|
695 |
|
|
|| TREE_CODE (TREE_OPERAND (stmt, 0)) == SSA_NAME))
|
696 |
|
|
{
|
697 |
|
|
/* If the statement is a copy assignment, evaluate its RHS to
|
698 |
|
|
see if the lattice value of its output has changed. */
|
699 |
|
|
retval = copy_prop_visit_assignment (stmt, result_p);
|
700 |
|
|
}
|
701 |
|
|
else if (TREE_CODE (stmt) == MODIFY_EXPR
|
702 |
|
|
&& TREE_CODE (TREE_OPERAND (stmt, 0)) == SSA_NAME
|
703 |
|
|
&& do_store_copy_prop
|
704 |
|
|
&& stmt_makes_single_load (stmt))
|
705 |
|
|
{
|
706 |
|
|
/* If the statement is a copy assignment with a memory load
|
707 |
|
|
on the RHS, see if we know the value of this load and
|
708 |
|
|
update the lattice accordingly. */
|
709 |
|
|
prop_value_t *val = get_value_loaded_by (stmt, copy_of);
|
710 |
|
|
if (val
|
711 |
|
|
&& val->mem_ref
|
712 |
|
|
&& is_gimple_reg (val->value)
|
713 |
|
|
&& operand_equal_p (val->mem_ref, TREE_OPERAND (stmt, 1), 0))
|
714 |
|
|
{
|
715 |
|
|
bool changed;
|
716 |
|
|
changed = set_copy_of_val (TREE_OPERAND (stmt, 0),
|
717 |
|
|
val->value, val->mem_ref);
|
718 |
|
|
if (changed)
|
719 |
|
|
{
|
720 |
|
|
*result_p = TREE_OPERAND (stmt, 0);
|
721 |
|
|
retval = SSA_PROP_INTERESTING;
|
722 |
|
|
}
|
723 |
|
|
else
|
724 |
|
|
retval = SSA_PROP_NOT_INTERESTING;
|
725 |
|
|
}
|
726 |
|
|
else
|
727 |
|
|
retval = SSA_PROP_VARYING;
|
728 |
|
|
}
|
729 |
|
|
else if (TREE_CODE (stmt) == COND_EXPR)
|
730 |
|
|
{
|
731 |
|
|
/* See if we can determine which edge goes out of a conditional
|
732 |
|
|
jump. */
|
733 |
|
|
retval = copy_prop_visit_cond_stmt (stmt, taken_edge_p);
|
734 |
|
|
}
|
735 |
|
|
else
|
736 |
|
|
retval = SSA_PROP_VARYING;
|
737 |
|
|
|
738 |
|
|
if (retval == SSA_PROP_VARYING)
|
739 |
|
|
{
|
740 |
|
|
tree def;
|
741 |
|
|
ssa_op_iter i;
|
742 |
|
|
|
743 |
|
|
/* Any other kind of statement is not interesting for constant
|
744 |
|
|
propagation and, therefore, not worth simulating. */
|
745 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
746 |
|
|
fprintf (dump_file, "No interesting values produced.\n");
|
747 |
|
|
|
748 |
|
|
/* The assignment is not a copy operation. Don't visit this
|
749 |
|
|
statement again and mark all the definitions in the statement
|
750 |
|
|
to be copies of nothing. */
|
751 |
|
|
FOR_EACH_SSA_TREE_OPERAND (def, stmt, i, SSA_OP_ALL_DEFS)
|
752 |
|
|
set_copy_of_val (def, def, NULL_TREE);
|
753 |
|
|
}
|
754 |
|
|
|
755 |
|
|
return retval;
|
756 |
|
|
}
|
757 |
|
|
|
758 |
|
|
|
759 |
|
|
/* Visit PHI node PHI. If all the arguments produce the same value,
|
760 |
|
|
set it to be the value of the LHS of PHI. */
|
761 |
|
|
|
762 |
|
|
static enum ssa_prop_result
|
763 |
|
|
copy_prop_visit_phi_node (tree phi)
|
764 |
|
|
{
|
765 |
|
|
enum ssa_prop_result retval;
|
766 |
|
|
int i;
|
767 |
|
|
tree lhs;
|
768 |
|
|
prop_value_t phi_val = { 0, NULL_TREE, NULL_TREE };
|
769 |
|
|
|
770 |
|
|
lhs = PHI_RESULT (phi);
|
771 |
|
|
|
772 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
773 |
|
|
{
|
774 |
|
|
fprintf (dump_file, "\nVisiting PHI node: ");
|
775 |
|
|
print_generic_expr (dump_file, phi, dump_flags);
|
776 |
|
|
fprintf (dump_file, "\n\n");
|
777 |
|
|
}
|
778 |
|
|
|
779 |
|
|
for (i = 0; i < PHI_NUM_ARGS (phi); i++)
|
780 |
|
|
{
|
781 |
|
|
prop_value_t *arg_val;
|
782 |
|
|
tree arg = PHI_ARG_DEF (phi, i);
|
783 |
|
|
edge e = PHI_ARG_EDGE (phi, i);
|
784 |
|
|
|
785 |
|
|
/* We don't care about values flowing through non-executable
|
786 |
|
|
edges. */
|
787 |
|
|
if (!(e->flags & EDGE_EXECUTABLE))
|
788 |
|
|
continue;
|
789 |
|
|
|
790 |
|
|
/* Constants in the argument list never generate a useful copy.
|
791 |
|
|
Similarly, names that flow through abnormal edges cannot be
|
792 |
|
|
used to derive copies. */
|
793 |
|
|
if (TREE_CODE (arg) != SSA_NAME || SSA_NAME_OCCURS_IN_ABNORMAL_PHI (arg))
|
794 |
|
|
{
|
795 |
|
|
phi_val.value = lhs;
|
796 |
|
|
break;
|
797 |
|
|
}
|
798 |
|
|
|
799 |
|
|
/* Avoid copy propagation from an inner into an outer loop.
|
800 |
|
|
Otherwise, this may move loop variant variables outside of
|
801 |
|
|
their loops and prevent coalescing opportunities. If the
|
802 |
|
|
value was loop invariant, it will be hoisted by LICM and
|
803 |
|
|
exposed for copy propagation. */
|
804 |
|
|
if (loop_depth_of_name (arg) > loop_depth_of_name (lhs))
|
805 |
|
|
{
|
806 |
|
|
phi_val.value = lhs;
|
807 |
|
|
break;
|
808 |
|
|
}
|
809 |
|
|
|
810 |
|
|
/* If the LHS appears in the argument list, ignore it. It is
|
811 |
|
|
irrelevant as a copy. */
|
812 |
|
|
if (arg == lhs || get_last_copy_of (arg) == lhs)
|
813 |
|
|
continue;
|
814 |
|
|
|
815 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
816 |
|
|
{
|
817 |
|
|
fprintf (dump_file, "\tArgument #%d: ", i);
|
818 |
|
|
dump_copy_of (dump_file, arg);
|
819 |
|
|
fprintf (dump_file, "\n");
|
820 |
|
|
}
|
821 |
|
|
|
822 |
|
|
arg_val = get_copy_of_val (arg);
|
823 |
|
|
|
824 |
|
|
/* If the LHS didn't have a value yet, make it a copy of the
|
825 |
|
|
first argument we find. Notice that while we make the LHS be
|
826 |
|
|
a copy of the argument itself, we take the memory reference
|
827 |
|
|
from the argument's value so that we can compare it to the
|
828 |
|
|
memory reference of all the other arguments. */
|
829 |
|
|
if (phi_val.value == NULL_TREE)
|
830 |
|
|
{
|
831 |
|
|
phi_val.value = arg;
|
832 |
|
|
phi_val.mem_ref = arg_val->mem_ref;
|
833 |
|
|
continue;
|
834 |
|
|
}
|
835 |
|
|
|
836 |
|
|
/* If PHI_VAL and ARG don't have a common copy-of chain, then
|
837 |
|
|
this PHI node cannot be a copy operation. Also, if we are
|
838 |
|
|
copy propagating stores and these two arguments came from
|
839 |
|
|
different memory references, they cannot be considered
|
840 |
|
|
copies. */
|
841 |
|
|
if (get_last_copy_of (phi_val.value) != get_last_copy_of (arg)
|
842 |
|
|
|| (do_store_copy_prop
|
843 |
|
|
&& phi_val.mem_ref
|
844 |
|
|
&& arg_val->mem_ref
|
845 |
|
|
&& simple_cst_equal (phi_val.mem_ref, arg_val->mem_ref) != 1))
|
846 |
|
|
{
|
847 |
|
|
phi_val.value = lhs;
|
848 |
|
|
break;
|
849 |
|
|
}
|
850 |
|
|
}
|
851 |
|
|
|
852 |
|
|
if (phi_val.value && set_copy_of_val (lhs, phi_val.value, phi_val.mem_ref))
|
853 |
|
|
retval = (phi_val.value != lhs) ? SSA_PROP_INTERESTING : SSA_PROP_VARYING;
|
854 |
|
|
else
|
855 |
|
|
retval = SSA_PROP_NOT_INTERESTING;
|
856 |
|
|
|
857 |
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
858 |
|
|
{
|
859 |
|
|
fprintf (dump_file, "\nPHI node ");
|
860 |
|
|
dump_copy_of (dump_file, lhs);
|
861 |
|
|
fprintf (dump_file, "\nTelling the propagator to ");
|
862 |
|
|
if (retval == SSA_PROP_INTERESTING)
|
863 |
|
|
fprintf (dump_file, "add SSA edges out of this PHI and continue.");
|
864 |
|
|
else if (retval == SSA_PROP_VARYING)
|
865 |
|
|
fprintf (dump_file, "add SSA edges out of this PHI and never visit again.");
|
866 |
|
|
else
|
867 |
|
|
fprintf (dump_file, "do nothing with SSA edges and keep iterating.");
|
868 |
|
|
fprintf (dump_file, "\n\n");
|
869 |
|
|
}
|
870 |
|
|
|
871 |
|
|
return retval;
|
872 |
|
|
}
|
873 |
|
|
|
874 |
|
|
|
875 |
|
|
/* Initialize structures used for copy propagation. PHIS_ONLY is true
|
876 |
|
|
if we should only consider PHI nodes as generating copy propagation
|
877 |
|
|
opportunities. */
|
878 |
|
|
|
879 |
|
|
static void
|
880 |
|
|
init_copy_prop (void)
|
881 |
|
|
{
|
882 |
|
|
basic_block bb;
|
883 |
|
|
|
884 |
|
|
copy_of = XNEWVEC (prop_value_t, num_ssa_names);
|
885 |
|
|
memset (copy_of, 0, num_ssa_names * sizeof (*copy_of));
|
886 |
|
|
|
887 |
|
|
cached_last_copy_of = XNEWVEC (tree, num_ssa_names);
|
888 |
|
|
memset (cached_last_copy_of, 0, num_ssa_names * sizeof (*cached_last_copy_of));
|
889 |
|
|
|
890 |
|
|
FOR_EACH_BB (bb)
|
891 |
|
|
{
|
892 |
|
|
block_stmt_iterator si;
|
893 |
|
|
tree phi, def;
|
894 |
|
|
int depth = bb->loop_depth;
|
895 |
|
|
|
896 |
|
|
for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
|
897 |
|
|
{
|
898 |
|
|
tree stmt = bsi_stmt (si);
|
899 |
|
|
ssa_op_iter iter;
|
900 |
|
|
|
901 |
|
|
/* The only statements that we care about are those that may
|
902 |
|
|
generate useful copies. We also need to mark conditional
|
903 |
|
|
jumps so that their outgoing edges are added to the work
|
904 |
|
|
lists of the propagator.
|
905 |
|
|
|
906 |
|
|
Avoid copy propagation from an inner into an outer loop.
|
907 |
|
|
Otherwise, this may move loop variant variables outside of
|
908 |
|
|
their loops and prevent coalescing opportunities. If the
|
909 |
|
|
value was loop invariant, it will be hoisted by LICM and
|
910 |
|
|
exposed for copy propagation. */
|
911 |
|
|
if (stmt_ends_bb_p (stmt))
|
912 |
|
|
DONT_SIMULATE_AGAIN (stmt) = false;
|
913 |
|
|
else if (stmt_may_generate_copy (stmt)
|
914 |
|
|
&& loop_depth_of_name (TREE_OPERAND (stmt, 1)) <= depth)
|
915 |
|
|
DONT_SIMULATE_AGAIN (stmt) = false;
|
916 |
|
|
else
|
917 |
|
|
DONT_SIMULATE_AGAIN (stmt) = true;
|
918 |
|
|
|
919 |
|
|
/* Mark all the outputs of this statement as not being
|
920 |
|
|
the copy of anything. */
|
921 |
|
|
FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_ALL_DEFS)
|
922 |
|
|
if (DONT_SIMULATE_AGAIN (stmt))
|
923 |
|
|
set_copy_of_val (def, def, NULL_TREE);
|
924 |
|
|
else
|
925 |
|
|
cached_last_copy_of[SSA_NAME_VERSION (def)] = def;
|
926 |
|
|
}
|
927 |
|
|
|
928 |
|
|
for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
|
929 |
|
|
{
|
930 |
|
|
def = PHI_RESULT (phi);
|
931 |
|
|
if (!do_store_copy_prop && !is_gimple_reg (def))
|
932 |
|
|
DONT_SIMULATE_AGAIN (phi) = true;
|
933 |
|
|
else
|
934 |
|
|
DONT_SIMULATE_AGAIN (phi) = false;
|
935 |
|
|
|
936 |
|
|
if (DONT_SIMULATE_AGAIN (phi))
|
937 |
|
|
set_copy_of_val (def, def, NULL_TREE);
|
938 |
|
|
else
|
939 |
|
|
cached_last_copy_of[SSA_NAME_VERSION (def)] = def;
|
940 |
|
|
}
|
941 |
|
|
}
|
942 |
|
|
}
|
943 |
|
|
|
944 |
|
|
|
945 |
|
|
/* Deallocate memory used in copy propagation and do final
|
946 |
|
|
substitution. */
|
947 |
|
|
|
948 |
|
|
static void
|
949 |
|
|
fini_copy_prop (void)
|
950 |
|
|
{
|
951 |
|
|
size_t i;
|
952 |
|
|
prop_value_t *tmp;
|
953 |
|
|
|
954 |
|
|
/* Set the final copy-of value for each variable by traversing the
|
955 |
|
|
copy-of chains. */
|
956 |
|
|
tmp = XNEWVEC (prop_value_t, num_ssa_names);
|
957 |
|
|
memset (tmp, 0, num_ssa_names * sizeof (*tmp));
|
958 |
|
|
for (i = 1; i < num_ssa_names; i++)
|
959 |
|
|
{
|
960 |
|
|
tree var = ssa_name (i);
|
961 |
|
|
if (var && copy_of[i].value && copy_of[i].value != var)
|
962 |
|
|
tmp[i].value = get_last_copy_of (var);
|
963 |
|
|
}
|
964 |
|
|
|
965 |
|
|
substitute_and_fold (tmp, false);
|
966 |
|
|
|
967 |
|
|
free (cached_last_copy_of);
|
968 |
|
|
free (copy_of);
|
969 |
|
|
free (tmp);
|
970 |
|
|
}
|
971 |
|
|
|
972 |
|
|
|
973 |
|
|
/* Main entry point to the copy propagator.
|
974 |
|
|
|
975 |
|
|
PHIS_ONLY is true if we should only consider PHI nodes as generating
|
976 |
|
|
copy propagation opportunities.
|
977 |
|
|
|
978 |
|
|
The algorithm propagates the value COPY-OF using ssa_propagate. For
|
979 |
|
|
every variable X_i, COPY-OF(X_i) indicates which variable is X_i created
|
980 |
|
|
from. The following example shows how the algorithm proceeds at a
|
981 |
|
|
high level:
|
982 |
|
|
|
983 |
|
|
1 a_24 = x_1
|
984 |
|
|
2 a_2 = PHI <a_24, x_1>
|
985 |
|
|
3 a_5 = PHI <a_2>
|
986 |
|
|
4 x_1 = PHI <x_298, a_5, a_2>
|
987 |
|
|
|
988 |
|
|
The end result should be that a_2, a_5, a_24 and x_1 are a copy of
|
989 |
|
|
x_298. Propagation proceeds as follows.
|
990 |
|
|
|
991 |
|
|
Visit #1: a_24 is copy-of x_1. Value changed.
|
992 |
|
|
Visit #2: a_2 is copy-of x_1. Value changed.
|
993 |
|
|
Visit #3: a_5 is copy-of x_1. Value changed.
|
994 |
|
|
Visit #4: x_1 is copy-of x_298. Value changed.
|
995 |
|
|
Visit #1: a_24 is copy-of x_298. Value changed.
|
996 |
|
|
Visit #2: a_2 is copy-of x_298. Value changed.
|
997 |
|
|
Visit #3: a_5 is copy-of x_298. Value changed.
|
998 |
|
|
Visit #4: x_1 is copy-of x_298. Stable state reached.
|
999 |
|
|
|
1000 |
|
|
When visiting PHI nodes, we only consider arguments that flow
|
1001 |
|
|
through edges marked executable by the propagation engine. So,
|
1002 |
|
|
when visiting statement #2 for the first time, we will only look at
|
1003 |
|
|
the first argument (a_24) and optimistically assume that its value
|
1004 |
|
|
is the copy of a_24 (x_1).
|
1005 |
|
|
|
1006 |
|
|
The problem with this approach is that it may fail to discover copy
|
1007 |
|
|
relations in PHI cycles. Instead of propagating copy-of
|
1008 |
|
|
values, we actually propagate copy-of chains. For instance:
|
1009 |
|
|
|
1010 |
|
|
A_3 = B_1;
|
1011 |
|
|
C_9 = A_3;
|
1012 |
|
|
D_4 = C_9;
|
1013 |
|
|
X_i = D_4;
|
1014 |
|
|
|
1015 |
|
|
In this code fragment, COPY-OF (X_i) = { D_4, C_9, A_3, B_1 }.
|
1016 |
|
|
Obviously, we are only really interested in the last value of the
|
1017 |
|
|
chain, however the propagator needs to access the copy-of chain
|
1018 |
|
|
when visiting PHI nodes.
|
1019 |
|
|
|
1020 |
|
|
To represent the copy-of chain, we use the array COPY_CHAINS, which
|
1021 |
|
|
holds the first link in the copy-of chain for every variable.
|
1022 |
|
|
If variable X_i is a copy of X_j, which in turn is a copy of X_k,
|
1023 |
|
|
the array will contain:
|
1024 |
|
|
|
1025 |
|
|
COPY_CHAINS[i] = X_j
|
1026 |
|
|
COPY_CHAINS[j] = X_k
|
1027 |
|
|
COPY_CHAINS[k] = X_k
|
1028 |
|
|
|
1029 |
|
|
Keeping copy-of chains instead of copy-of values directly becomes
|
1030 |
|
|
important when visiting PHI nodes. Suppose that we had the
|
1031 |
|
|
following PHI cycle, such that x_52 is already considered a copy of
|
1032 |
|
|
x_53:
|
1033 |
|
|
|
1034 |
|
|
1 x_54 = PHI <x_53, x_52>
|
1035 |
|
|
2 x_53 = PHI <x_898, x_54>
|
1036 |
|
|
|
1037 |
|
|
Visit #1: x_54 is copy-of x_53 (because x_52 is copy-of x_53)
|
1038 |
|
|
Visit #2: x_53 is copy-of x_898 (because x_54 is a copy of x_53,
|
1039 |
|
|
so it is considered irrelevant
|
1040 |
|
|
as a copy).
|
1041 |
|
|
Visit #1: x_54 is copy-of nothing (x_53 is a copy-of x_898 and
|
1042 |
|
|
x_52 is a copy of x_53, so
|
1043 |
|
|
they don't match)
|
1044 |
|
|
Visit #2: x_53 is copy-of nothing
|
1045 |
|
|
|
1046 |
|
|
This problem is avoided by keeping a chain of copies, instead of
|
1047 |
|
|
the final copy-of value. Propagation will now only keep the first
|
1048 |
|
|
element of a variable's copy-of chain. When visiting PHI nodes,
|
1049 |
|
|
arguments are considered equal if their copy-of chains end in the
|
1050 |
|
|
same variable. So, as long as their copy-of chains overlap, we
|
1051 |
|
|
know that they will be a copy of the same variable, regardless of
|
1052 |
|
|
which variable that may be).
|
1053 |
|
|
|
1054 |
|
|
Propagation would then proceed as follows (the notation a -> b
|
1055 |
|
|
means that a is a copy-of b):
|
1056 |
|
|
|
1057 |
|
|
Visit #1: x_54 = PHI <x_53, x_52>
|
1058 |
|
|
x_53 -> x_53
|
1059 |
|
|
x_52 -> x_53
|
1060 |
|
|
Result: x_54 -> x_53. Value changed. Add SSA edges.
|
1061 |
|
|
|
1062 |
|
|
Visit #1: x_53 = PHI <x_898, x_54>
|
1063 |
|
|
x_898 -> x_898
|
1064 |
|
|
x_54 -> x_53
|
1065 |
|
|
Result: x_53 -> x_898. Value changed. Add SSA edges.
|
1066 |
|
|
|
1067 |
|
|
Visit #2: x_54 = PHI <x_53, x_52>
|
1068 |
|
|
x_53 -> x_898
|
1069 |
|
|
x_52 -> x_53 -> x_898
|
1070 |
|
|
Result: x_54 -> x_898. Value changed. Add SSA edges.
|
1071 |
|
|
|
1072 |
|
|
Visit #2: x_53 = PHI <x_898, x_54>
|
1073 |
|
|
x_898 -> x_898
|
1074 |
|
|
x_54 -> x_898
|
1075 |
|
|
Result: x_53 -> x_898. Value didn't change. Stable state
|
1076 |
|
|
|
1077 |
|
|
Once the propagator stabilizes, we end up with the desired result
|
1078 |
|
|
x_53 and x_54 are both copies of x_898. */
|
1079 |
|
|
|
1080 |
|
|
static void
|
1081 |
|
|
execute_copy_prop (bool store_copy_prop)
|
1082 |
|
|
{
|
1083 |
|
|
do_store_copy_prop = store_copy_prop;
|
1084 |
|
|
init_copy_prop ();
|
1085 |
|
|
ssa_propagate (copy_prop_visit_stmt, copy_prop_visit_phi_node);
|
1086 |
|
|
fini_copy_prop ();
|
1087 |
|
|
}
|
1088 |
|
|
|
1089 |
|
|
|
1090 |
|
|
static bool
|
1091 |
|
|
gate_copy_prop (void)
|
1092 |
|
|
{
|
1093 |
|
|
return flag_tree_copy_prop != 0;
|
1094 |
|
|
}
|
1095 |
|
|
|
1096 |
|
|
static unsigned int
|
1097 |
|
|
do_copy_prop (void)
|
1098 |
|
|
{
|
1099 |
|
|
execute_copy_prop (false);
|
1100 |
|
|
return 0;
|
1101 |
|
|
}
|
1102 |
|
|
|
1103 |
|
|
struct tree_opt_pass pass_copy_prop =
|
1104 |
|
|
{
|
1105 |
|
|
"copyprop", /* name */
|
1106 |
|
|
gate_copy_prop, /* gate */
|
1107 |
|
|
do_copy_prop, /* execute */
|
1108 |
|
|
NULL, /* sub */
|
1109 |
|
|
NULL, /* next */
|
1110 |
|
|
0, /* static_pass_number */
|
1111 |
|
|
TV_TREE_COPY_PROP, /* tv_id */
|
1112 |
|
|
PROP_ssa | PROP_alias | PROP_cfg, /* properties_required */
|
1113 |
|
|
0, /* properties_provided */
|
1114 |
|
|
0, /* properties_destroyed */
|
1115 |
|
|
0, /* todo_flags_start */
|
1116 |
|
|
TODO_cleanup_cfg
|
1117 |
|
|
| TODO_dump_func
|
1118 |
|
|
| TODO_ggc_collect
|
1119 |
|
|
| TODO_verify_ssa
|
1120 |
|
|
| TODO_update_ssa, /* todo_flags_finish */
|
1121 |
|
|
|
1122 |
|
|
};
|
1123 |
|
|
|
1124 |
|
|
static bool
|
1125 |
|
|
gate_store_copy_prop (void)
|
1126 |
|
|
{
|
1127 |
|
|
/* STORE-COPY-PROP is enabled only with -ftree-store-copy-prop, but
|
1128 |
|
|
when -fno-tree-store-copy-prop is specified, we should run
|
1129 |
|
|
regular COPY-PROP. That's why the pass is enabled with either
|
1130 |
|
|
flag. */
|
1131 |
|
|
return flag_tree_store_copy_prop != 0 || flag_tree_copy_prop != 0;
|
1132 |
|
|
}
|
1133 |
|
|
|
1134 |
|
|
static unsigned int
|
1135 |
|
|
store_copy_prop (void)
|
1136 |
|
|
{
|
1137 |
|
|
/* If STORE-COPY-PROP is not enabled, we just run regular COPY-PROP. */
|
1138 |
|
|
execute_copy_prop (flag_tree_store_copy_prop != 0);
|
1139 |
|
|
return 0;
|
1140 |
|
|
}
|
1141 |
|
|
|
1142 |
|
|
struct tree_opt_pass pass_store_copy_prop =
|
1143 |
|
|
{
|
1144 |
|
|
"store_copyprop", /* name */
|
1145 |
|
|
gate_store_copy_prop, /* gate */
|
1146 |
|
|
store_copy_prop, /* execute */
|
1147 |
|
|
NULL, /* sub */
|
1148 |
|
|
NULL, /* next */
|
1149 |
|
|
0, /* static_pass_number */
|
1150 |
|
|
TV_TREE_STORE_COPY_PROP, /* tv_id */
|
1151 |
|
|
PROP_ssa | PROP_alias | PROP_cfg, /* properties_required */
|
1152 |
|
|
0, /* properties_provided */
|
1153 |
|
|
0, /* properties_destroyed */
|
1154 |
|
|
0, /* todo_flags_start */
|
1155 |
|
|
TODO_dump_func
|
1156 |
|
|
| TODO_cleanup_cfg
|
1157 |
|
|
| TODO_ggc_collect
|
1158 |
|
|
| TODO_verify_ssa
|
1159 |
|
|
| TODO_update_ssa, /* todo_flags_finish */
|
1160 |
|
|
|
1161 |
|
|
};
|