1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
|
/* Generic routines for manipulating PHIs
Copyright (C) 2003-2015 Free Software Foundation, Inc.
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.
GCC is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "backend.h"
#include "tree.h"
#include "gimple.h"
#include "ssa.h"
#include "fold-const.h"
#include "gimple-iterator.h"
#include "tree-ssa.h"
/* Rewriting a function into SSA form can create a huge number of PHIs
many of which may be thrown away shortly after their creation if jumps
were threaded through PHI nodes.
While our garbage collection mechanisms will handle this situation, it
is extremely wasteful to create nodes and throw them away, especially
when the nodes can be reused.
For PR 8361, we can significantly reduce the number of nodes allocated
and thus the total amount of memory allocated by managing PHIs a
little. This additionally helps reduce the amount of work done by the
garbage collector. Similar results have been seen on a wider variety
of tests (such as the compiler itself).
PHI nodes have different sizes, so we can't have a single list of all
the PHI nodes as it would be too expensive to walk down that list to
find a PHI of a suitable size.
Instead we have an array of lists of free PHI nodes. The array is
indexed by the number of PHI alternatives that PHI node can hold.
Except for the last array member, which holds all remaining PHI
nodes.
So to find a free PHI node, we compute its index into the free PHI
node array and see if there are any elements with an exact match.
If so, then we are done. Otherwise, we test the next larger size
up and continue until we are in the last array element.
We do not actually walk members of the last array element. While it
might allow us to pick up a few reusable PHI nodes, it could potentially
be very expensive if the program has released a bunch of large PHI nodes,
but keeps asking for even larger PHI nodes. Experiments have shown that
walking the elements of the last array entry would result in finding less
than .1% additional reusable PHI nodes.
Note that we can never have less than two PHI argument slots. Thus,
the -2 on all the calculations below. */
#define NUM_BUCKETS 10
static GTY ((deletable (""))) vec<gimple *, va_gc> *free_phinodes[NUM_BUCKETS - 2];
static unsigned long free_phinode_count;
static int ideal_phi_node_len (int);
unsigned int phi_nodes_reused;
unsigned int phi_nodes_created;
/* Dump some simple statistics regarding the re-use of PHI nodes. */
void
phinodes_print_statistics (void)
{
fprintf (stderr, "PHI nodes allocated: %u\n", phi_nodes_created);
fprintf (stderr, "PHI nodes reused: %u\n", phi_nodes_reused);
}
/* Allocate a PHI node with at least LEN arguments. If the free list
happens to contain a PHI node with LEN arguments or more, return
that one. */
static inline gphi *
allocate_phi_node (size_t len)
{
gphi *phi;
size_t bucket = NUM_BUCKETS - 2;
size_t size = sizeof (struct gphi)
+ (len - 1) * sizeof (struct phi_arg_d);
if (free_phinode_count)
for (bucket = len - 2; bucket < NUM_BUCKETS - 2; bucket++)
if (free_phinodes[bucket])
break;
/* If our free list has an element, then use it. */
if (bucket < NUM_BUCKETS - 2
&& gimple_phi_capacity ((*free_phinodes[bucket])[0]) >= len)
{
free_phinode_count--;
phi = as_a <gphi *> (free_phinodes[bucket]->pop ());
if (free_phinodes[bucket]->is_empty ())
vec_free (free_phinodes[bucket]);
if (GATHER_STATISTICS)
phi_nodes_reused++;
}
else
{
phi = static_cast <gphi *> (ggc_internal_alloc (size));
if (GATHER_STATISTICS)
{
enum gimple_alloc_kind kind = gimple_alloc_kind (GIMPLE_PHI);
phi_nodes_created++;
gimple_alloc_counts[(int) kind]++;
gimple_alloc_sizes[(int) kind] += size;
}
}
return phi;
}
/* Given LEN, the original number of requested PHI arguments, return
a new, "ideal" length for the PHI node. The "ideal" length rounds
the total size of the PHI node up to the next power of two bytes.
Rounding up will not result in wasting any memory since the size request
will be rounded up by the GC system anyway. [ Note this is not entirely
true since the original length might have fit on one of the special
GC pages. ] By rounding up, we may avoid the need to reallocate the
PHI node later if we increase the number of arguments for the PHI. */
static int
ideal_phi_node_len (int len)
{
size_t size, new_size;
int log2, new_len;
/* We do not support allocations of less than two PHI argument slots. */
if (len < 2)
len = 2;
/* Compute the number of bytes of the original request. */
size = sizeof (struct gphi)
+ (len - 1) * sizeof (struct phi_arg_d);
/* Round it up to the next power of two. */
log2 = ceil_log2 (size);
new_size = 1 << log2;
/* Now compute and return the number of PHI argument slots given an
ideal size allocation. */
new_len = len + (new_size - size) / sizeof (struct phi_arg_d);
return new_len;
}
/* Return a PHI node with LEN argument slots for variable VAR. */
static gphi *
make_phi_node (tree var, int len)
{
gphi *phi;
int capacity, i;
capacity = ideal_phi_node_len (len);
phi = allocate_phi_node (capacity);
/* We need to clear the entire PHI node, including the argument
portion, because we represent a "missing PHI argument" by placing
NULL_TREE in PHI_ARG_DEF. */
memset (phi, 0, (sizeof (struct gphi)
- sizeof (struct phi_arg_d)
+ sizeof (struct phi_arg_d) * len));
phi->code = GIMPLE_PHI;
gimple_init_singleton (phi);
phi->nargs = len;
phi->capacity = capacity;
if (!var)
;
else if (TREE_CODE (var) == SSA_NAME)
gimple_phi_set_result (phi, var);
else
gimple_phi_set_result (phi, make_ssa_name (var, phi));
for (i = 0; i < capacity; i++)
{
use_operand_p imm;
gimple_phi_arg_set_location (phi, i, UNKNOWN_LOCATION);
imm = gimple_phi_arg_imm_use_ptr (phi, i);
imm->use = gimple_phi_arg_def_ptr (phi, i);
imm->prev = NULL;
imm->next = NULL;
imm->loc.stmt = phi;
}
return phi;
}
/* We no longer need PHI, release it so that it may be reused. */
void
release_phi_node (gimple *phi)
{
size_t bucket;
size_t len = gimple_phi_capacity (phi);
size_t x;
for (x = 0; x < gimple_phi_num_args (phi); x++)
{
use_operand_p imm;
imm = gimple_phi_arg_imm_use_ptr (phi, x);
delink_imm_use (imm);
}
bucket = len > NUM_BUCKETS - 1 ? NUM_BUCKETS - 1 : len;
bucket -= 2;
vec_safe_push (free_phinodes[bucket], phi);
free_phinode_count++;
}
/* Resize an existing PHI node. The only way is up. Return the
possibly relocated phi. */
static gphi *
resize_phi_node (gphi *phi, size_t len)
{
size_t old_size, i;
gphi *new_phi;
gcc_assert (len > gimple_phi_capacity (phi));
/* The garbage collector will not look at the PHI node beyond the
first PHI_NUM_ARGS elements. Therefore, all we have to copy is a
portion of the PHI node currently in use. */
old_size = sizeof (struct gphi)
+ (gimple_phi_num_args (phi) - 1) * sizeof (struct phi_arg_d);
new_phi = allocate_phi_node (len);
memcpy (new_phi, phi, old_size);
for (i = 0; i < gimple_phi_num_args (new_phi); i++)
{
use_operand_p imm, old_imm;
imm = gimple_phi_arg_imm_use_ptr (new_phi, i);
old_imm = gimple_phi_arg_imm_use_ptr (phi, i);
imm->use = gimple_phi_arg_def_ptr (new_phi, i);
relink_imm_use_stmt (imm, old_imm, new_phi);
}
new_phi->capacity = len;
for (i = gimple_phi_num_args (new_phi); i < len; i++)
{
use_operand_p imm;
gimple_phi_arg_set_location (new_phi, i, UNKNOWN_LOCATION);
imm = gimple_phi_arg_imm_use_ptr (new_phi, i);
imm->use = gimple_phi_arg_def_ptr (new_phi, i);
imm->prev = NULL;
imm->next = NULL;
imm->loc.stmt = new_phi;
}
return new_phi;
}
/* Reserve PHI arguments for a new edge to basic block BB. */
void
reserve_phi_args_for_new_edge (basic_block bb)
{
size_t len = EDGE_COUNT (bb->preds);
size_t cap = ideal_phi_node_len (len + 4);
gphi_iterator gsi;
for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi))
{
gphi *stmt = gsi.phi ();
if (len > gimple_phi_capacity (stmt))
{
gphi *new_phi = resize_phi_node (stmt, cap);
/* The result of the PHI is defined by this PHI node. */
SSA_NAME_DEF_STMT (gimple_phi_result (new_phi)) = new_phi;
gsi_set_stmt (&gsi, new_phi);
release_phi_node (stmt);
stmt = new_phi;
}
/* We represent a "missing PHI argument" by placing NULL_TREE in
the corresponding slot. If PHI arguments were added
immediately after an edge is created, this zeroing would not
be necessary, but unfortunately this is not the case. For
example, the loop optimizer duplicates several basic blocks,
redirects edges, and then fixes up PHI arguments later in
batch. */
SET_PHI_ARG_DEF (stmt, len - 1, NULL_TREE);
gimple_phi_arg_set_location (stmt, len - 1, UNKNOWN_LOCATION);
stmt->nargs++;
}
}
/* Adds PHI to BB. */
void
add_phi_node_to_bb (gphi *phi, basic_block bb)
{
gimple_seq seq = phi_nodes (bb);
/* Add the new PHI node to the list of PHI nodes for block BB. */
if (seq == NULL)
set_phi_nodes (bb, gimple_seq_alloc_with_stmt (phi));
else
{
gimple_seq_add_stmt (&seq, phi);
gcc_assert (seq == phi_nodes (bb));
}
/* Associate BB to the PHI node. */
gimple_set_bb (phi, bb);
}
/* Create a new PHI node for variable VAR at basic block BB. */
gphi *
create_phi_node (tree var, basic_block bb)
{
gphi *phi = make_phi_node (var, EDGE_COUNT (bb->preds));
add_phi_node_to_bb (phi, bb);
return phi;
}
/* Add a new argument to PHI node PHI. DEF is the incoming reaching
definition and E is the edge through which DEF reaches PHI. The new
argument is added at the end of the argument list.
If PHI has reached its maximum capacity, add a few slots. In this case,
PHI points to the reallocated phi node when we return. */
void
add_phi_arg (gphi *phi, tree def, edge e, source_location locus)
{
basic_block bb = e->dest;
gcc_assert (bb == gimple_bb (phi));
/* We resize PHI nodes upon edge creation. We should always have
enough room at this point. */
gcc_assert (gimple_phi_num_args (phi) <= gimple_phi_capacity (phi));
/* We resize PHI nodes upon edge creation. We should always have
enough room at this point. */
gcc_assert (e->dest_idx < gimple_phi_num_args (phi));
/* Copy propagation needs to know what object occur in abnormal
PHI nodes. This is a convenient place to record such information. */
if (e->flags & EDGE_ABNORMAL)
{
SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def) = 1;
SSA_NAME_OCCURS_IN_ABNORMAL_PHI (PHI_RESULT (phi)) = 1;
}
SET_PHI_ARG_DEF (phi, e->dest_idx, def);
gimple_phi_arg_set_location (phi, e->dest_idx, locus);
}
/* Remove the Ith argument from PHI's argument list. This routine
implements removal by swapping the last alternative with the
alternative we want to delete and then shrinking the vector, which
is consistent with how we remove an edge from the edge vector. */
static void
remove_phi_arg_num (gphi *phi, int i)
{
int num_elem = gimple_phi_num_args (phi);
gcc_assert (i < num_elem);
/* Delink the item which is being removed. */
delink_imm_use (gimple_phi_arg_imm_use_ptr (phi, i));
/* If it is not the last element, move the last element
to the element we want to delete, resetting all the links. */
if (i != num_elem - 1)
{
use_operand_p old_p, new_p;
old_p = gimple_phi_arg_imm_use_ptr (phi, num_elem - 1);
new_p = gimple_phi_arg_imm_use_ptr (phi, i);
/* Set use on new node, and link into last element's place. */
*(new_p->use) = *(old_p->use);
relink_imm_use (new_p, old_p);
/* Move the location as well. */
gimple_phi_arg_set_location (phi, i,
gimple_phi_arg_location (phi, num_elem - 1));
}
/* Shrink the vector and return. Note that we do not have to clear
PHI_ARG_DEF because the garbage collector will not look at those
elements beyond the first PHI_NUM_ARGS elements of the array. */
phi->nargs--;
}
/* Remove all PHI arguments associated with edge E. */
void
remove_phi_args (edge e)
{
gphi_iterator gsi;
for (gsi = gsi_start_phis (e->dest); !gsi_end_p (gsi); gsi_next (&gsi))
remove_phi_arg_num (gsi.phi (),
e->dest_idx);
}
/* Remove the PHI node pointed-to by iterator GSI from basic block BB. After
removal, iterator GSI is updated to point to the next PHI node in the
sequence. If RELEASE_LHS_P is true, the LHS of this PHI node is released
into the free pool of SSA names. */
void
remove_phi_node (gimple_stmt_iterator *gsi, bool release_lhs_p)
{
gimple *phi = gsi_stmt (*gsi);
if (release_lhs_p)
insert_debug_temps_for_defs (gsi);
gsi_remove (gsi, false);
/* If we are deleting the PHI node, then we should release the
SSA_NAME node so that it can be reused. */
release_phi_node (phi);
if (release_lhs_p)
release_ssa_name (gimple_phi_result (phi));
}
/* Remove all the phi nodes from BB. */
void
remove_phi_nodes (basic_block bb)
{
gphi_iterator gsi;
for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); )
remove_phi_node (&gsi, true);
set_phi_nodes (bb, NULL);
}
/* Given PHI, return its RHS if the PHI is a degenerate, otherwise return
NULL. */
tree
degenerate_phi_result (gphi *phi)
{
tree lhs = gimple_phi_result (phi);
tree val = NULL;
size_t i;
/* Ignoring arguments which are the same as LHS, if all the remaining
arguments are the same, then the PHI is a degenerate and has the
value of that common argument. */
for (i = 0; i < gimple_phi_num_args (phi); i++)
{
tree arg = gimple_phi_arg_def (phi, i);
if (arg == lhs)
continue;
else if (!arg)
break;
else if (!val)
val = arg;
else if (arg == val)
continue;
/* We bring in some of operand_equal_p not only to speed things
up, but also to avoid crashing when dereferencing the type of
a released SSA name. */
else if (TREE_CODE (val) != TREE_CODE (arg)
|| TREE_CODE (val) == SSA_NAME
|| !operand_equal_p (arg, val, 0))
break;
}
return (i == gimple_phi_num_args (phi) ? val : NULL);
}
/* Set PHI nodes of a basic block BB to SEQ. */
void
set_phi_nodes (basic_block bb, gimple_seq seq)
{
gimple_stmt_iterator i;
gcc_checking_assert (!(bb->flags & BB_RTL));
bb->il.gimple.phi_nodes = seq;
if (seq)
for (i = gsi_start (seq); !gsi_end_p (i); gsi_next (&i))
gimple_set_bb (gsi_stmt (i), bb);
}
#include "gt-tree-phinodes.h"
|