Mercurial > jdk8-mips64-public > hotspot / file revision
src/share/vm/opto/phaseX.cpp@69ea58782b1a
src/share/vm/opto/phaseX.cpp
Thu, 31 Jul 2014 19:59:36 +0200
- author
- roland
- date
- Thu, 31 Jul 2014 19:59:36 +0200
- changeset 7003
- 69ea58782b1a
- parent 6681
- 1555c0843770
- child 6876
- 710a3c8b516e
- child 7394
- 5b8e0f84f00f
- permissions
- -rw-r--r--
8054054: 8040121 is broken
Summary: C++ code pattern from 8040121 is incorrect
Reviewed-by: kvn
1 /*
2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "memory/allocation.inline.hpp"
27 #include "opto/block.hpp"
28 #include "opto/callnode.hpp"
29 #include "opto/cfgnode.hpp"
30 #include "opto/connode.hpp"
31 #include "opto/idealGraphPrinter.hpp"
32 #include "opto/loopnode.hpp"
33 #include "opto/machnode.hpp"
34 #include "opto/opcodes.hpp"
35 #include "opto/phaseX.hpp"
36 #include "opto/regalloc.hpp"
37 #include "opto/rootnode.hpp"
39 //=============================================================================
40 #define NODE_HASH_MINIMUM_SIZE 255
41 //------------------------------NodeHash---------------------------------------
42 NodeHash::NodeHash(uint est_max_size) :
43 _max( round_up(est_max_size < NODE_HASH_MINIMUM_SIZE ? NODE_HASH_MINIMUM_SIZE : est_max_size) ),
44 _a(Thread::current()->resource_area()),
45 _table( NEW_ARENA_ARRAY( _a , Node* , _max ) ), // (Node**)_a->Amalloc(_max * sizeof(Node*)) ),
46 _inserts(0), _insert_limit( insert_limit() ),
47 _look_probes(0), _lookup_hits(0), _lookup_misses(0),
48 _total_insert_probes(0), _total_inserts(0),
49 _insert_probes(0), _grows(0) {
50 // _sentinel must be in the current node space
51 _sentinel = new (Compile::current()) ProjNode(NULL, TypeFunc::Control);
52 memset(_table,0,sizeof(Node*)*_max);
53 }
55 //------------------------------NodeHash---------------------------------------
56 NodeHash::NodeHash(Arena *arena, uint est_max_size) :
57 _max( round_up(est_max_size < NODE_HASH_MINIMUM_SIZE ? NODE_HASH_MINIMUM_SIZE : est_max_size) ),
58 _a(arena),
59 _table( NEW_ARENA_ARRAY( _a , Node* , _max ) ),
60 _inserts(0), _insert_limit( insert_limit() ),
61 _look_probes(0), _lookup_hits(0), _lookup_misses(0),
62 _delete_probes(0), _delete_hits(0), _delete_misses(0),
63 _total_insert_probes(0), _total_inserts(0),
64 _insert_probes(0), _grows(0) {
65 // _sentinel must be in the current node space
66 _sentinel = new (Compile::current()) ProjNode(NULL, TypeFunc::Control);
67 memset(_table,0,sizeof(Node*)*_max);
68 }
70 //------------------------------NodeHash---------------------------------------
71 NodeHash::NodeHash(NodeHash *nh) {
72 debug_only(_table = (Node**)badAddress); // interact correctly w/ operator=
73 // just copy in all the fields
74 *this = *nh;
75 // nh->_sentinel must be in the current node space
76 }
78 void NodeHash::replace_with(NodeHash *nh) {
79 debug_only(_table = (Node**)badAddress); // interact correctly w/ operator=
80 // just copy in all the fields
81 *this = *nh;
82 // nh->_sentinel must be in the current node space
83 }
85 //------------------------------hash_find--------------------------------------
86 // Find in hash table
87 Node *NodeHash::hash_find( const Node *n ) {
88 // ((Node*)n)->set_hash( n->hash() );
89 uint hash = n->hash();
90 if (hash == Node::NO_HASH) {
91 debug_only( _lookup_misses++ );
92 return NULL;
93 }
94 uint key = hash & (_max-1);
95 uint stride = key | 0x01;
96 debug_only( _look_probes++ );
97 Node *k = _table[key]; // Get hashed value
98 if( !k ) { // ?Miss?
99 debug_only( _lookup_misses++ );
100 return NULL; // Miss!
101 }
103 int op = n->Opcode();
104 uint req = n->req();
105 while( 1 ) { // While probing hash table
106 if( k->req() == req && // Same count of inputs
107 k->Opcode() == op ) { // Same Opcode
108 for( uint i=0; i<req; i++ )
109 if( n->in(i)!=k->in(i)) // Different inputs?
110 goto collision; // "goto" is a speed hack...
111 if( n->cmp(*k) ) { // Check for any special bits
112 debug_only( _lookup_hits++ );
113 return k; // Hit!
114 }
115 }
116 collision:
117 debug_only( _look_probes++ );
118 key = (key + stride/*7*/) & (_max-1); // Stride through table with relative prime
119 k = _table[key]; // Get hashed value
120 if( !k ) { // ?Miss?
121 debug_only( _lookup_misses++ );
122 return NULL; // Miss!
123 }
124 }
125 ShouldNotReachHere();
126 return NULL;
127 }
129 //------------------------------hash_find_insert-------------------------------
130 // Find in hash table, insert if not already present
131 // Used to preserve unique entries in hash table
132 Node *NodeHash::hash_find_insert( Node *n ) {
133 // n->set_hash( );
134 uint hash = n->hash();
135 if (hash == Node::NO_HASH) {
136 debug_only( _lookup_misses++ );
137 return NULL;
138 }
139 uint key = hash & (_max-1);
140 uint stride = key | 0x01; // stride must be relatively prime to table siz
141 uint first_sentinel = 0; // replace a sentinel if seen.
142 debug_only( _look_probes++ );
143 Node *k = _table[key]; // Get hashed value
144 if( !k ) { // ?Miss?
145 debug_only( _lookup_misses++ );
146 _table[key] = n; // Insert into table!
147 debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
148 check_grow(); // Grow table if insert hit limit
149 return NULL; // Miss!
150 }
151 else if( k == _sentinel ) {
152 first_sentinel = key; // Can insert here
153 }
155 int op = n->Opcode();
156 uint req = n->req();
157 while( 1 ) { // While probing hash table
158 if( k->req() == req && // Same count of inputs
159 k->Opcode() == op ) { // Same Opcode
160 for( uint i=0; i<req; i++ )
161 if( n->in(i)!=k->in(i)) // Different inputs?
162 goto collision; // "goto" is a speed hack...
163 if( n->cmp(*k) ) { // Check for any special bits
164 debug_only( _lookup_hits++ );
165 return k; // Hit!
166 }
167 }
168 collision:
169 debug_only( _look_probes++ );
170 key = (key + stride) & (_max-1); // Stride through table w/ relative prime
171 k = _table[key]; // Get hashed value
172 if( !k ) { // ?Miss?
173 debug_only( _lookup_misses++ );
174 key = (first_sentinel == 0) ? key : first_sentinel; // ?saw sentinel?
175 _table[key] = n; // Insert into table!
176 debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
177 check_grow(); // Grow table if insert hit limit
178 return NULL; // Miss!
179 }
180 else if( first_sentinel == 0 && k == _sentinel ) {
181 first_sentinel = key; // Can insert here
182 }
184 }
185 ShouldNotReachHere();
186 return NULL;
187 }
189 //------------------------------hash_insert------------------------------------
190 // Insert into hash table
191 void NodeHash::hash_insert( Node *n ) {
192 // // "conflict" comments -- print nodes that conflict
193 // bool conflict = false;
194 // n->set_hash();
195 uint hash = n->hash();
196 if (hash == Node::NO_HASH) {
197 return;
198 }
199 check_grow();
200 uint key = hash & (_max-1);
201 uint stride = key | 0x01;
203 while( 1 ) { // While probing hash table
204 debug_only( _insert_probes++ );
205 Node *k = _table[key]; // Get hashed value
206 if( !k || (k == _sentinel) ) break; // Found a slot
207 assert( k != n, "already inserted" );
208 // if( PrintCompilation && PrintOptoStatistics && Verbose ) { tty->print(" conflict: "); k->dump(); conflict = true; }
209 key = (key + stride) & (_max-1); // Stride through table w/ relative prime
210 }
211 _table[key] = n; // Insert into table!
212 debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
213 // if( conflict ) { n->dump(); }
214 }
216 //------------------------------hash_delete------------------------------------
217 // Replace in hash table with sentinel
218 bool NodeHash::hash_delete( const Node *n ) {
219 Node *k;
220 uint hash = n->hash();
221 if (hash == Node::NO_HASH) {
222 debug_only( _delete_misses++ );
223 return false;
224 }
225 uint key = hash & (_max-1);
226 uint stride = key | 0x01;
227 debug_only( uint counter = 0; );
228 for( ; /* (k != NULL) && (k != _sentinel) */; ) {
229 debug_only( counter++ );
230 debug_only( _delete_probes++ );
231 k = _table[key]; // Get hashed value
232 if( !k ) { // Miss?
233 debug_only( _delete_misses++ );
234 #ifdef ASSERT
235 if( VerifyOpto ) {
236 for( uint i=0; i < _max; i++ )
237 assert( _table[i] != n, "changed edges with rehashing" );
238 }
239 #endif
240 return false; // Miss! Not in chain
241 }
242 else if( n == k ) {
243 debug_only( _delete_hits++ );
244 _table[key] = _sentinel; // Hit! Label as deleted entry
245 debug_only(((Node*)n)->exit_hash_lock()); // Unlock the node upon removal from table.
246 return true;
247 }
248 else {
249 // collision: move through table with prime offset
250 key = (key + stride/*7*/) & (_max-1);
251 assert( counter <= _insert_limit, "Cycle in hash-table");
252 }
253 }
254 ShouldNotReachHere();
255 return false;
256 }
258 //------------------------------round_up---------------------------------------
259 // Round up to nearest power of 2
260 uint NodeHash::round_up( uint x ) {
261 x += (x>>2); // Add 25% slop
262 if( x <16 ) return 16; // Small stuff
263 uint i=16;
264 while( i < x ) i <<= 1; // Double to fit
265 return i; // Return hash table size
266 }
268 //------------------------------grow-------------------------------------------
269 // Grow _table to next power of 2 and insert old entries
270 void NodeHash::grow() {
271 // Record old state
272 uint old_max = _max;
273 Node **old_table = _table;
274 // Construct new table with twice the space
275 _grows++;
276 _total_inserts += _inserts;
277 _total_insert_probes += _insert_probes;
278 _inserts = 0;
279 _insert_probes = 0;
280 _max = _max << 1;
281 _table = NEW_ARENA_ARRAY( _a , Node* , _max ); // (Node**)_a->Amalloc( _max * sizeof(Node*) );
282 memset(_table,0,sizeof(Node*)*_max);
283 _insert_limit = insert_limit();
284 // Insert old entries into the new table
285 for( uint i = 0; i < old_max; i++ ) {
286 Node *m = *old_table++;
287 if( !m || m == _sentinel ) continue;
288 debug_only(m->exit_hash_lock()); // Unlock the node upon removal from old table.
289 hash_insert(m);
290 }
291 }
293 //------------------------------clear------------------------------------------
294 // Clear all entries in _table to NULL but keep storage
295 void NodeHash::clear() {
296 #ifdef ASSERT
297 // Unlock all nodes upon removal from table.
298 for (uint i = 0; i < _max; i++) {
299 Node* n = _table[i];
300 if (!n || n == _sentinel) continue;
301 n->exit_hash_lock();
302 }
303 #endif
305 memset( _table, 0, _max * sizeof(Node*) );
306 }
308 //-----------------------remove_useless_nodes----------------------------------
309 // Remove useless nodes from value table,
310 // implementation does not depend on hash function
311 void NodeHash::remove_useless_nodes(VectorSet &useful) {
313 // Dead nodes in the hash table inherited from GVN should not replace
314 // existing nodes, remove dead nodes.
315 uint max = size();
316 Node *sentinel_node = sentinel();
317 for( uint i = 0; i < max; ++i ) {
318 Node *n = at(i);
319 if(n != NULL && n != sentinel_node && !useful.test(n->_idx)) {
320 debug_only(n->exit_hash_lock()); // Unlock the node when removed
321 _table[i] = sentinel_node; // Replace with placeholder
322 }
323 }
324 }
327 void NodeHash::check_no_speculative_types() {
328 #ifdef ASSERT
329 uint max = size();
330 Node *sentinel_node = sentinel();
331 for (uint i = 0; i < max; ++i) {
332 Node *n = at(i);
333 if(n != NULL && n != sentinel_node && n->is_Type()) {
334 TypeNode* tn = n->as_Type();
335 const Type* t = tn->type();
336 const Type* t_no_spec = t->remove_speculative();
337 assert(t == t_no_spec, "dead node in hash table or missed node during speculative cleanup");
338 }
339 }
340 #endif
341 }
343 #ifndef PRODUCT
344 //------------------------------dump-------------------------------------------
345 // Dump statistics for the hash table
346 void NodeHash::dump() {
347 _total_inserts += _inserts;
348 _total_insert_probes += _insert_probes;
349 if (PrintCompilation && PrintOptoStatistics && Verbose && (_inserts > 0)) {
350 if (WizardMode) {
351 for (uint i=0; i<_max; i++) {
352 if (_table[i])
353 tty->print("%d/%d/%d ",i,_table[i]->hash()&(_max-1),_table[i]->_idx);
354 }
355 }
356 tty->print("\nGVN Hash stats: %d grows to %d max_size\n", _grows, _max);
357 tty->print(" %d/%d (%8.1f%% full)\n", _inserts, _max, (double)_inserts/_max*100.0);
358 tty->print(" %dp/(%dh+%dm) (%8.2f probes/lookup)\n", _look_probes, _lookup_hits, _lookup_misses, (double)_look_probes/(_lookup_hits+_lookup_misses));
359 tty->print(" %dp/%di (%8.2f probes/insert)\n", _total_insert_probes, _total_inserts, (double)_total_insert_probes/_total_inserts);
360 // sentinels increase lookup cost, but not insert cost
361 assert((_lookup_misses+_lookup_hits)*4+100 >= _look_probes, "bad hash function");
362 assert( _inserts+(_inserts>>3) < _max, "table too full" );
363 assert( _inserts*3+100 >= _insert_probes, "bad hash function" );
364 }
365 }
367 Node *NodeHash::find_index(uint idx) { // For debugging
368 // Find an entry by its index value
369 for( uint i = 0; i < _max; i++ ) {
370 Node *m = _table[i];
371 if( !m || m == _sentinel ) continue;
372 if( m->_idx == (uint)idx ) return m;
373 }
374 return NULL;
375 }
376 #endif
378 #ifdef ASSERT
379 NodeHash::~NodeHash() {
380 // Unlock all nodes upon destruction of table.
381 if (_table != (Node**)badAddress) clear();
382 }
384 void NodeHash::operator=(const NodeHash& nh) {
385 // Unlock all nodes upon replacement of table.
386 if (&nh == this) return;
387 if (_table != (Node**)badAddress) clear();
388 memcpy(this, &nh, sizeof(*this));
389 // Do not increment hash_lock counts again.
390 // Instead, be sure we never again use the source table.
391 ((NodeHash*)&nh)->_table = (Node**)badAddress;
392 }
395 #endif
398 //=============================================================================
399 //------------------------------PhaseRemoveUseless-----------------------------
400 // 1) Use a breadthfirst walk to collect useful nodes reachable from root.
401 PhaseRemoveUseless::PhaseRemoveUseless( PhaseGVN *gvn, Unique_Node_List *worklist ) : Phase(Remove_Useless),
402 _useful(Thread::current()->resource_area()) {
404 // Implementation requires 'UseLoopSafepoints == true' and an edge from root
405 // to each SafePointNode at a backward branch. Inserted in add_safepoint().
406 if( !UseLoopSafepoints || !OptoRemoveUseless ) return;
408 // Identify nodes that are reachable from below, useful.
409 C->identify_useful_nodes(_useful);
410 // Update dead node list
411 C->update_dead_node_list(_useful);
413 // Remove all useless nodes from PhaseValues' recorded types
414 // Must be done before disconnecting nodes to preserve hash-table-invariant
415 gvn->remove_useless_nodes(_useful.member_set());
417 // Remove all useless nodes from future worklist
418 worklist->remove_useless_nodes(_useful.member_set());
420 // Disconnect 'useless' nodes that are adjacent to useful nodes
421 C->remove_useless_nodes(_useful);
423 // Remove edges from "root" to each SafePoint at a backward branch.
424 // They were inserted during parsing (see add_safepoint()) to make infinite
425 // loops without calls or exceptions visible to root, i.e., useful.
426 Node *root = C->root();
427 if( root != NULL ) {
428 for( uint i = root->req(); i < root->len(); ++i ) {
429 Node *n = root->in(i);
430 if( n != NULL && n->is_SafePoint() ) {
431 root->rm_prec(i);
432 --i;
433 }
434 }
435 }
436 }
439 //=============================================================================
440 //------------------------------PhaseTransform---------------------------------
441 PhaseTransform::PhaseTransform( PhaseNumber pnum ) : Phase(pnum),
442 _arena(Thread::current()->resource_area()),
443 _nodes(_arena),
444 _types(_arena)
445 {
446 init_con_caches();
447 #ifndef PRODUCT
448 clear_progress();
449 clear_transforms();
450 set_allow_progress(true);
451 #endif
452 // Force allocation for currently existing nodes
453 _types.map(C->unique(), NULL);
454 }
456 //------------------------------PhaseTransform---------------------------------
457 PhaseTransform::PhaseTransform( Arena *arena, PhaseNumber pnum ) : Phase(pnum),
458 _arena(arena),
459 _nodes(arena),
460 _types(arena)
461 {
462 init_con_caches();
463 #ifndef PRODUCT
464 clear_progress();
465 clear_transforms();
466 set_allow_progress(true);
467 #endif
468 // Force allocation for currently existing nodes
469 _types.map(C->unique(), NULL);
470 }
472 //------------------------------PhaseTransform---------------------------------
473 // Initialize with previously generated type information
474 PhaseTransform::PhaseTransform( PhaseTransform *pt, PhaseNumber pnum ) : Phase(pnum),
475 _arena(pt->_arena),
476 _nodes(pt->_nodes),
477 _types(pt->_types)
478 {
479 init_con_caches();
480 #ifndef PRODUCT
481 clear_progress();
482 clear_transforms();
483 set_allow_progress(true);
484 #endif
485 }
487 void PhaseTransform::init_con_caches() {
488 memset(_icons,0,sizeof(_icons));
489 memset(_lcons,0,sizeof(_lcons));
490 memset(_zcons,0,sizeof(_zcons));
491 }
494 //--------------------------------find_int_type--------------------------------
495 const TypeInt* PhaseTransform::find_int_type(Node* n) {
496 if (n == NULL) return NULL;
497 // Call type_or_null(n) to determine node's type since we might be in
498 // parse phase and call n->Value() may return wrong type.
499 // (For example, a phi node at the beginning of loop parsing is not ready.)
500 const Type* t = type_or_null(n);
501 if (t == NULL) return NULL;
502 return t->isa_int();
503 }
506 //-------------------------------find_long_type--------------------------------
507 const TypeLong* PhaseTransform::find_long_type(Node* n) {
508 if (n == NULL) return NULL;
509 // (See comment above on type_or_null.)
510 const Type* t = type_or_null(n);
511 if (t == NULL) return NULL;
512 return t->isa_long();
513 }
516 #ifndef PRODUCT
517 void PhaseTransform::dump_old2new_map() const {
518 _nodes.dump();
519 }
521 void PhaseTransform::dump_new( uint nidx ) const {
522 for( uint i=0; i<_nodes.Size(); i++ )
523 if( _nodes[i] && _nodes[i]->_idx == nidx ) {
524 _nodes[i]->dump();
525 tty->cr();
526 tty->print_cr("Old index= %d",i);
527 return;
528 }
529 tty->print_cr("Node %d not found in the new indices", nidx);
530 }
532 //------------------------------dump_types-------------------------------------
533 void PhaseTransform::dump_types( ) const {
534 _types.dump();
535 }
537 //------------------------------dump_nodes_and_types---------------------------
538 void PhaseTransform::dump_nodes_and_types(const Node *root, uint depth, bool only_ctrl) {
539 VectorSet visited(Thread::current()->resource_area());
540 dump_nodes_and_types_recur( root, depth, only_ctrl, visited );
541 }
543 //------------------------------dump_nodes_and_types_recur---------------------
544 void PhaseTransform::dump_nodes_and_types_recur( const Node *n, uint depth, bool only_ctrl, VectorSet &visited) {
545 if( !n ) return;
546 if( depth == 0 ) return;
547 if( visited.test_set(n->_idx) ) return;
548 for( uint i=0; i<n->len(); i++ ) {
549 if( only_ctrl && !(n->is_Region()) && i != TypeFunc::Control ) continue;
550 dump_nodes_and_types_recur( n->in(i), depth-1, only_ctrl, visited );
551 }
552 n->dump();
553 if (type_or_null(n) != NULL) {
554 tty->print(" "); type(n)->dump(); tty->cr();
555 }
556 }
558 #endif
561 //=============================================================================
562 //------------------------------PhaseValues------------------------------------
563 // Set minimum table size to "255"
564 PhaseValues::PhaseValues( Arena *arena, uint est_max_size ) : PhaseTransform(arena, GVN), _table(arena, est_max_size) {
565 NOT_PRODUCT( clear_new_values(); )
566 }
568 //------------------------------PhaseValues------------------------------------
569 // Set minimum table size to "255"
570 PhaseValues::PhaseValues( PhaseValues *ptv ) : PhaseTransform( ptv, GVN ),
571 _table(&ptv->_table) {
572 NOT_PRODUCT( clear_new_values(); )
573 }
575 //------------------------------PhaseValues------------------------------------
576 // Used by +VerifyOpto. Clear out hash table but copy _types array.
577 PhaseValues::PhaseValues( PhaseValues *ptv, const char *dummy ) : PhaseTransform( ptv, GVN ),
578 _table(ptv->arena(),ptv->_table.size()) {
579 NOT_PRODUCT( clear_new_values(); )
580 }
582 //------------------------------~PhaseValues-----------------------------------
583 #ifndef PRODUCT
584 PhaseValues::~PhaseValues() {
585 _table.dump();
587 // Statistics for value progress and efficiency
588 if( PrintCompilation && Verbose && WizardMode ) {
589 tty->print("\n%sValues: %d nodes ---> %d/%d (%d)",
590 is_IterGVN() ? "Iter" : " ", C->unique(), made_progress(), made_transforms(), made_new_values());
591 if( made_transforms() != 0 ) {
592 tty->print_cr(" ratio %f", made_progress()/(float)made_transforms() );
593 } else {
594 tty->cr();
595 }
596 }
597 }
598 #endif
600 //------------------------------makecon----------------------------------------
601 ConNode* PhaseTransform::makecon(const Type *t) {
602 assert(t->singleton(), "must be a constant");
603 assert(!t->empty() || t == Type::TOP, "must not be vacuous range");
604 switch (t->base()) { // fast paths
605 case Type::Half:
606 case Type::Top: return (ConNode*) C->top();
607 case Type::Int: return intcon( t->is_int()->get_con() );
608 case Type::Long: return longcon( t->is_long()->get_con() );
609 }
610 if (t->is_zero_type())
611 return zerocon(t->basic_type());
612 return uncached_makecon(t);
613 }
615 //--------------------------uncached_makecon-----------------------------------
616 // Make an idealized constant - one of ConINode, ConPNode, etc.
617 ConNode* PhaseValues::uncached_makecon(const Type *t) {
618 assert(t->singleton(), "must be a constant");
619 ConNode* x = ConNode::make(C, t);
620 ConNode* k = (ConNode*)hash_find_insert(x); // Value numbering
621 if (k == NULL) {
622 set_type(x, t); // Missed, provide type mapping
623 GrowableArray<Node_Notes*>* nna = C->node_note_array();
624 if (nna != NULL) {
625 Node_Notes* loc = C->locate_node_notes(nna, x->_idx, true);
626 loc->clear(); // do not put debug info on constants
627 }
628 } else {
629 x->destruct(); // Hit, destroy duplicate constant
630 x = k; // use existing constant
631 }
632 return x;
633 }
635 //------------------------------intcon-----------------------------------------
636 // Fast integer constant. Same as "transform(new ConINode(TypeInt::make(i)))"
637 ConINode* PhaseTransform::intcon(int i) {
638 // Small integer? Check cache! Check that cached node is not dead
639 if (i >= _icon_min && i <= _icon_max) {
640 ConINode* icon = _icons[i-_icon_min];
641 if (icon != NULL && icon->in(TypeFunc::Control) != NULL)
642 return icon;
643 }
644 ConINode* icon = (ConINode*) uncached_makecon(TypeInt::make(i));
645 assert(icon->is_Con(), "");
646 if (i >= _icon_min && i <= _icon_max)
647 _icons[i-_icon_min] = icon; // Cache small integers
648 return icon;
649 }
651 //------------------------------longcon----------------------------------------
652 // Fast long constant.
653 ConLNode* PhaseTransform::longcon(jlong l) {
654 // Small integer? Check cache! Check that cached node is not dead
655 if (l >= _lcon_min && l <= _lcon_max) {
656 ConLNode* lcon = _lcons[l-_lcon_min];
657 if (lcon != NULL && lcon->in(TypeFunc::Control) != NULL)
658 return lcon;
659 }
660 ConLNode* lcon = (ConLNode*) uncached_makecon(TypeLong::make(l));
661 assert(lcon->is_Con(), "");
662 if (l >= _lcon_min && l <= _lcon_max)
663 _lcons[l-_lcon_min] = lcon; // Cache small integers
664 return lcon;
665 }
667 //------------------------------zerocon-----------------------------------------
668 // Fast zero or null constant. Same as "transform(ConNode::make(Type::get_zero_type(bt)))"
669 ConNode* PhaseTransform::zerocon(BasicType bt) {
670 assert((uint)bt <= _zcon_max, "domain check");
671 ConNode* zcon = _zcons[bt];
672 if (zcon != NULL && zcon->in(TypeFunc::Control) != NULL)
673 return zcon;
674 zcon = (ConNode*) uncached_makecon(Type::get_zero_type(bt));
675 _zcons[bt] = zcon;
676 return zcon;
677 }
681 //=============================================================================
682 //------------------------------transform--------------------------------------
683 // Return a node which computes the same function as this node, but in a
684 // faster or cheaper fashion.
685 Node *PhaseGVN::transform( Node *n ) {
686 return transform_no_reclaim(n);
687 }
689 //------------------------------transform--------------------------------------
690 // Return a node which computes the same function as this node, but
691 // in a faster or cheaper fashion.
692 Node *PhaseGVN::transform_no_reclaim( Node *n ) {
693 NOT_PRODUCT( set_transforms(); )
695 // Apply the Ideal call in a loop until it no longer applies
696 Node *k = n;
697 NOT_PRODUCT( uint loop_count = 0; )
698 while( 1 ) {
699 Node *i = k->Ideal(this, /*can_reshape=*/false);
700 if( !i ) break;
701 assert( i->_idx >= k->_idx, "Idealize should return new nodes, use Identity to return old nodes" );
702 k = i;
703 assert(loop_count++ < K, "infinite loop in PhaseGVN::transform");
704 }
705 NOT_PRODUCT( if( loop_count != 0 ) { set_progress(); } )
708 // If brand new node, make space in type array.
709 ensure_type_or_null(k);
711 // Since I just called 'Value' to compute the set of run-time values
712 // for this Node, and 'Value' is non-local (and therefore expensive) I'll
713 // cache Value. Later requests for the local phase->type of this Node can
714 // use the cached Value instead of suffering with 'bottom_type'.
715 const Type *t = k->Value(this); // Get runtime Value set
716 assert(t != NULL, "value sanity");
717 if (type_or_null(k) != t) {
718 #ifndef PRODUCT
719 // Do not count initial visit to node as a transformation
720 if (type_or_null(k) == NULL) {
721 inc_new_values();
722 set_progress();
723 }
724 #endif
725 set_type(k, t);
726 // If k is a TypeNode, capture any more-precise type permanently into Node
727 k->raise_bottom_type(t);
728 }
730 if( t->singleton() && !k->is_Con() ) {
731 NOT_PRODUCT( set_progress(); )
732 return makecon(t); // Turn into a constant
733 }
735 // Now check for Identities
736 Node *i = k->Identity(this); // Look for a nearby replacement
737 if( i != k ) { // Found? Return replacement!
738 NOT_PRODUCT( set_progress(); )
739 return i;
740 }
742 // Global Value Numbering
743 i = hash_find_insert(k); // Insert if new
744 if( i && (i != k) ) {
745 // Return the pre-existing node
746 NOT_PRODUCT( set_progress(); )
747 return i;
748 }
750 // Return Idealized original
751 return k;
752 }
754 #ifdef ASSERT
755 //------------------------------dead_loop_check--------------------------------
756 // Check for a simple dead loop when a data node references itself directly
757 // or through an other data node excluding cons and phis.
758 void PhaseGVN::dead_loop_check( Node *n ) {
759 // Phi may reference itself in a loop
760 if (n != NULL && !n->is_dead_loop_safe() && !n->is_CFG()) {
761 // Do 2 levels check and only data inputs.
762 bool no_dead_loop = true;
763 uint cnt = n->req();
764 for (uint i = 1; i < cnt && no_dead_loop; i++) {
765 Node *in = n->in(i);
766 if (in == n) {
767 no_dead_loop = false;
768 } else if (in != NULL && !in->is_dead_loop_safe()) {
769 uint icnt = in->req();
770 for (uint j = 1; j < icnt && no_dead_loop; j++) {
771 if (in->in(j) == n || in->in(j) == in)
772 no_dead_loop = false;
773 }
774 }
775 }
776 if (!no_dead_loop) n->dump(3);
777 assert(no_dead_loop, "dead loop detected");
778 }
779 }
780 #endif
782 //=============================================================================
783 //------------------------------PhaseIterGVN-----------------------------------
784 // Initialize hash table to fresh and clean for +VerifyOpto
785 PhaseIterGVN::PhaseIterGVN( PhaseIterGVN *igvn, const char *dummy ) : PhaseGVN(igvn,dummy), _worklist( ),
786 _stack(C->unique() >> 1),
787 _delay_transform(false) {
788 }
790 //------------------------------PhaseIterGVN-----------------------------------
791 // Initialize with previous PhaseIterGVN info; used by PhaseCCP
792 PhaseIterGVN::PhaseIterGVN( PhaseIterGVN *igvn ) : PhaseGVN(igvn),
793 _worklist( igvn->_worklist ),
794 _stack( igvn->_stack ),
795 _delay_transform(igvn->_delay_transform)
796 {
797 }
799 //------------------------------PhaseIterGVN-----------------------------------
800 // Initialize with previous PhaseGVN info from Parser
801 PhaseIterGVN::PhaseIterGVN( PhaseGVN *gvn ) : PhaseGVN(gvn),
802 _worklist(*C->for_igvn()),
803 _stack(C->unique() >> 1),
804 _delay_transform(false)
805 {
806 uint max;
808 // Dead nodes in the hash table inherited from GVN were not treated as
809 // roots during def-use info creation; hence they represent an invisible
810 // use. Clear them out.
811 max = _table.size();
812 for( uint i = 0; i < max; ++i ) {
813 Node *n = _table.at(i);
814 if(n != NULL && n != _table.sentinel() && n->outcnt() == 0) {
815 if( n->is_top() ) continue;
816 assert( false, "Parse::remove_useless_nodes missed this node");
817 hash_delete(n);
818 }
819 }
821 // Any Phis or Regions on the worklist probably had uses that could not
822 // make more progress because the uses were made while the Phis and Regions
823 // were in half-built states. Put all uses of Phis and Regions on worklist.
824 max = _worklist.size();
825 for( uint j = 0; j < max; j++ ) {
826 Node *n = _worklist.at(j);
827 uint uop = n->Opcode();
828 if( uop == Op_Phi || uop == Op_Region ||
829 n->is_Type() ||
830 n->is_Mem() )
831 add_users_to_worklist(n);
832 }
833 }
836 #ifndef PRODUCT
837 void PhaseIterGVN::verify_step(Node* n) {
838 _verify_window[_verify_counter % _verify_window_size] = n;
839 ++_verify_counter;
840 ResourceMark rm;
841 ResourceArea *area = Thread::current()->resource_area();
842 VectorSet old_space(area), new_space(area);
843 if (C->unique() < 1000 ||
844 0 == _verify_counter % (C->unique() < 10000 ? 10 : 100)) {
845 ++_verify_full_passes;
846 Node::verify_recur(C->root(), -1, old_space, new_space);
847 }
848 const int verify_depth = 4;
849 for ( int i = 0; i < _verify_window_size; i++ ) {
850 Node* n = _verify_window[i];
851 if ( n == NULL ) continue;
852 if( n->in(0) == NodeSentinel ) { // xform_idom
853 _verify_window[i] = n->in(1);
854 --i; continue;
855 }
856 // Typical fanout is 1-2, so this call visits about 6 nodes.
857 Node::verify_recur(n, verify_depth, old_space, new_space);
858 }
859 }
860 #endif
863 //------------------------------init_worklist----------------------------------
864 // Initialize worklist for each node.
865 void PhaseIterGVN::init_worklist( Node *n ) {
866 if( _worklist.member(n) ) return;
867 _worklist.push(n);
868 uint cnt = n->req();
869 for( uint i =0 ; i < cnt; i++ ) {
870 Node *m = n->in(i);
871 if( m ) init_worklist(m);
872 }
873 }
875 //------------------------------optimize---------------------------------------
876 void PhaseIterGVN::optimize() {
877 debug_only(uint num_processed = 0;);
878 #ifndef PRODUCT
879 {
880 _verify_counter = 0;
881 _verify_full_passes = 0;
882 for ( int i = 0; i < _verify_window_size; i++ ) {
883 _verify_window[i] = NULL;
884 }
885 }
886 #endif
888 #ifdef ASSERT
889 Node* prev = NULL;
890 uint rep_cnt = 0;
891 #endif
892 uint loop_count = 0;
894 // Pull from worklist; transform node;
895 // If node has changed: update edge info and put uses on worklist.
896 while( _worklist.size() ) {
897 if (C->check_node_count(NodeLimitFudgeFactor * 2,
898 "out of nodes optimizing method")) {
899 return;
900 }
901 Node *n = _worklist.pop();
902 if (++loop_count >= K * C->live_nodes()) {
903 debug_only(n->dump(4);)
904 assert(false, "infinite loop in PhaseIterGVN::optimize");
905 C->record_method_not_compilable("infinite loop in PhaseIterGVN::optimize");
906 return;
907 }
908 #ifdef ASSERT
909 if (n == prev) {
910 if (++rep_cnt > 3) {
911 n->dump(4);
912 assert(false, "loop in Ideal transformation");
913 }
914 } else {
915 rep_cnt = 0;
916 }
917 prev = n;
918 #endif
919 if (TraceIterativeGVN && Verbose) {
920 tty->print(" Pop ");
921 NOT_PRODUCT( n->dump(); )
922 debug_only(if( (num_processed++ % 100) == 0 ) _worklist.print_set();)
923 }
925 if (n->outcnt() != 0) {
927 #ifndef PRODUCT
928 uint wlsize = _worklist.size();
929 const Type* oldtype = type_or_null(n);
930 #endif //PRODUCT
932 Node *nn = transform_old(n);
934 #ifndef PRODUCT
935 if (TraceIterativeGVN) {
936 const Type* newtype = type_or_null(n);
937 if (nn != n) {
938 // print old node
939 tty->print("< ");
940 if (oldtype != newtype && oldtype != NULL) {
941 oldtype->dump();
942 }
943 do { tty->print("\t"); } while (tty->position() < 16);
944 tty->print("<");
945 n->dump();
946 }
947 if (oldtype != newtype || nn != n) {
948 // print new node and/or new type
949 if (oldtype == NULL) {
950 tty->print("* ");
951 } else if (nn != n) {
952 tty->print("> ");
953 } else {
954 tty->print("= ");
955 }
956 if (newtype == NULL) {
957 tty->print("null");
958 } else {
959 newtype->dump();
960 }
961 do { tty->print("\t"); } while (tty->position() < 16);
962 nn->dump();
963 }
964 if (Verbose && wlsize < _worklist.size()) {
965 tty->print(" Push {");
966 while (wlsize != _worklist.size()) {
967 Node* pushed = _worklist.at(wlsize++);
968 tty->print(" %d", pushed->_idx);
969 }
970 tty->print_cr(" }");
971 }
972 }
973 if( VerifyIterativeGVN && nn != n ) {
974 verify_step((Node*) NULL); // ignore n, it might be subsumed
975 }
976 #endif
977 } else if (!n->is_top()) {
978 remove_dead_node(n);
979 }
980 }
982 #ifndef PRODUCT
983 C->verify_graph_edges();
984 if( VerifyOpto && allow_progress() ) {
985 // Must turn off allow_progress to enable assert and break recursion
986 C->root()->verify();
987 { // Check if any progress was missed using IterGVN
988 // Def-Use info enables transformations not attempted in wash-pass
989 // e.g. Region/Phi cleanup, ...
990 // Null-check elision -- may not have reached fixpoint
991 // do not propagate to dominated nodes
992 ResourceMark rm;
993 PhaseIterGVN igvn2(this,"Verify"); // Fresh and clean!
994 // Fill worklist completely
995 igvn2.init_worklist(C->root());
997 igvn2.set_allow_progress(false);
998 igvn2.optimize();
999 igvn2.set_allow_progress(true);
1000 }
1001 }
1002 if ( VerifyIterativeGVN && PrintOpto ) {
1003 if ( _verify_counter == _verify_full_passes )
1004 tty->print_cr("VerifyIterativeGVN: %d transforms and verify passes",
1005 (int) _verify_full_passes);
1006 else
1007 tty->print_cr("VerifyIterativeGVN: %d transforms, %d full verify passes",
1008 (int) _verify_counter, (int) _verify_full_passes);
1009 }
1010 #endif
1011 }
1014 //------------------register_new_node_with_optimizer---------------------------
1015 // Register a new node with the optimizer. Update the types array, the def-use
1016 // info. Put on worklist.
1017 Node* PhaseIterGVN::register_new_node_with_optimizer(Node* n, Node* orig) {
1018 set_type_bottom(n);
1019 _worklist.push(n);
1020 if (orig != NULL) C->copy_node_notes_to(n, orig);
1021 return n;
1022 }
1024 //------------------------------transform--------------------------------------
1025 // Non-recursive: idealize Node 'n' with respect to its inputs and its value
1026 Node *PhaseIterGVN::transform( Node *n ) {
1027 if (_delay_transform) {
1028 // Register the node but don't optimize for now
1029 register_new_node_with_optimizer(n);
1030 return n;
1031 }
1033 // If brand new node, make space in type array, and give it a type.
1034 ensure_type_or_null(n);
1035 if (type_or_null(n) == NULL) {
1036 set_type_bottom(n);
1037 }
1039 return transform_old(n);
1040 }
1042 //------------------------------transform_old----------------------------------
1043 Node *PhaseIterGVN::transform_old( Node *n ) {
1044 #ifndef PRODUCT
1045 debug_only(uint loop_count = 0;);
1046 set_transforms();
1047 #endif
1048 // Remove 'n' from hash table in case it gets modified
1049 _table.hash_delete(n);
1050 if( VerifyIterativeGVN ) {
1051 assert( !_table.find_index(n->_idx), "found duplicate entry in table");
1052 }
1054 // Apply the Ideal call in a loop until it no longer applies
1055 Node *k = n;
1056 DEBUG_ONLY(dead_loop_check(k);)
1057 DEBUG_ONLY(bool is_new = (k->outcnt() == 0);)
1058 Node *i = k->Ideal(this, /*can_reshape=*/true);
1059 assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes");
1060 #ifndef PRODUCT
1061 if( VerifyIterativeGVN )
1062 verify_step(k);
1063 if( i && VerifyOpto ) {
1064 if( !allow_progress() ) {
1065 if (i->is_Add() && i->outcnt() == 1) {
1066 // Switched input to left side because this is the only use
1067 } else if( i->is_If() && (i->in(0) == NULL) ) {
1068 // This IF is dead because it is dominated by an equivalent IF When
1069 // dominating if changed, info is not propagated sparsely to 'this'
1070 // Propagating this info further will spuriously identify other
1071 // progress.
1072 return i;
1073 } else
1074 set_progress();
1075 } else
1076 set_progress();
1077 }
1078 #endif
1080 while( i ) {
1081 #ifndef PRODUCT
1082 debug_only( if( loop_count >= K ) i->dump(4); )
1083 assert(loop_count < K, "infinite loop in PhaseIterGVN::transform");
1084 debug_only( loop_count++; )
1085 #endif
1086 assert((i->_idx >= k->_idx) || i->is_top(), "Idealize should return new nodes, use Identity to return old nodes");
1087 // Made a change; put users of original Node on worklist
1088 add_users_to_worklist( k );
1089 // Replacing root of transform tree?
1090 if( k != i ) {
1091 // Make users of old Node now use new.
1092 subsume_node( k, i );
1093 k = i;
1094 }
1095 DEBUG_ONLY(dead_loop_check(k);)
1096 // Try idealizing again
1097 DEBUG_ONLY(is_new = (k->outcnt() == 0);)
1098 i = k->Ideal(this, /*can_reshape=*/true);
1099 assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes");
1100 #ifndef PRODUCT
1101 if( VerifyIterativeGVN )
1102 verify_step(k);
1103 if( i && VerifyOpto ) set_progress();
1104 #endif
1105 }
1107 // If brand new node, make space in type array.
1108 ensure_type_or_null(k);
1110 // See what kind of values 'k' takes on at runtime
1111 const Type *t = k->Value(this);
1112 assert(t != NULL, "value sanity");
1114 // Since I just called 'Value' to compute the set of run-time values
1115 // for this Node, and 'Value' is non-local (and therefore expensive) I'll
1116 // cache Value. Later requests for the local phase->type of this Node can
1117 // use the cached Value instead of suffering with 'bottom_type'.
1118 if (t != type_or_null(k)) {
1119 NOT_PRODUCT( set_progress(); )
1120 NOT_PRODUCT( inc_new_values();)
1121 set_type(k, t);
1122 // If k is a TypeNode, capture any more-precise type permanently into Node
1123 k->raise_bottom_type(t);
1124 // Move users of node to worklist
1125 add_users_to_worklist( k );
1126 }
1128 // If 'k' computes a constant, replace it with a constant
1129 if( t->singleton() && !k->is_Con() ) {
1130 NOT_PRODUCT( set_progress(); )
1131 Node *con = makecon(t); // Make a constant
1132 add_users_to_worklist( k );
1133 subsume_node( k, con ); // Everybody using k now uses con
1134 return con;
1135 }
1137 // Now check for Identities
1138 i = k->Identity(this); // Look for a nearby replacement
1139 if( i != k ) { // Found? Return replacement!
1140 NOT_PRODUCT( set_progress(); )
1141 add_users_to_worklist( k );
1142 subsume_node( k, i ); // Everybody using k now uses i
1143 return i;
1144 }
1146 // Global Value Numbering
1147 i = hash_find_insert(k); // Check for pre-existing node
1148 if( i && (i != k) ) {
1149 // Return the pre-existing node if it isn't dead
1150 NOT_PRODUCT( set_progress(); )
1151 add_users_to_worklist( k );
1152 subsume_node( k, i ); // Everybody using k now uses i
1153 return i;
1154 }
1156 // Return Idealized original
1157 return k;
1158 }
1160 //---------------------------------saturate------------------------------------
1161 const Type* PhaseIterGVN::saturate(const Type* new_type, const Type* old_type,
1162 const Type* limit_type) const {
1163 return new_type->narrow(old_type);
1164 }
1166 //------------------------------remove_globally_dead_node----------------------
1167 // Kill a globally dead Node. All uses are also globally dead and are
1168 // aggressively trimmed.
1169 void PhaseIterGVN::remove_globally_dead_node( Node *dead ) {
1170 enum DeleteProgress {
1171 PROCESS_INPUTS,
1172 PROCESS_OUTPUTS
1173 };
1174 assert(_stack.is_empty(), "not empty");
1175 _stack.push(dead, PROCESS_INPUTS);
1177 while (_stack.is_nonempty()) {
1178 dead = _stack.node();
1179 uint progress_state = _stack.index();
1180 assert(dead != C->root(), "killing root, eh?");
1181 assert(!dead->is_top(), "add check for top when pushing");
1182 NOT_PRODUCT( set_progress(); )
1183 if (progress_state == PROCESS_INPUTS) {
1184 // After following inputs, continue to outputs
1185 _stack.set_index(PROCESS_OUTPUTS);
1186 if (!dead->is_Con()) { // Don't kill cons but uses
1187 bool recurse = false;
1188 // Remove from hash table
1189 _table.hash_delete( dead );
1190 // Smash all inputs to 'dead', isolating him completely
1191 for (uint i = 0; i < dead->req(); i++) {
1192 Node *in = dead->in(i);
1193 if (in != NULL && in != C->top()) { // Points to something?
1194 int nrep = dead->replace_edge(in, NULL); // Kill edges
1195 assert((nrep > 0), "sanity");
1196 if (in->outcnt() == 0) { // Made input go dead?
1197 _stack.push(in, PROCESS_INPUTS); // Recursively remove
1198 recurse = true;
1199 } else if (in->outcnt() == 1 &&
1200 in->has_special_unique_user()) {
1201 _worklist.push(in->unique_out());
1202 } else if (in->outcnt() <= 2 && dead->is_Phi()) {
1203 if (in->Opcode() == Op_Region) {
1204 _worklist.push(in);
1205 } else if (in->is_Store()) {
1206 DUIterator_Fast imax, i = in->fast_outs(imax);
1207 _worklist.push(in->fast_out(i));
1208 i++;
1209 if (in->outcnt() == 2) {
1210 _worklist.push(in->fast_out(i));
1211 i++;
1212 }
1213 assert(!(i < imax), "sanity");
1214 }
1215 }
1216 if (ReduceFieldZeroing && dead->is_Load() && i == MemNode::Memory &&
1217 in->is_Proj() && in->in(0) != NULL && in->in(0)->is_Initialize()) {
1218 // A Load that directly follows an InitializeNode is
1219 // going away. The Stores that follow are candidates
1220 // again to be captured by the InitializeNode.
1221 for (DUIterator_Fast jmax, j = in->fast_outs(jmax); j < jmax; j++) {
1222 Node *n = in->fast_out(j);
1223 if (n->is_Store()) {
1224 _worklist.push(n);
1225 }
1226 }
1227 }
1228 } // if (in != NULL && in != C->top())
1229 } // for (uint i = 0; i < dead->req(); i++)
1230 if (recurse) {
1231 continue;
1232 }
1233 } // if (!dead->is_Con())
1234 } // if (progress_state == PROCESS_INPUTS)
1236 // Aggressively kill globally dead uses
1237 // (Rather than pushing all the outs at once, we push one at a time,
1238 // plus the parent to resume later, because of the indefinite number
1239 // of edge deletions per loop trip.)
1240 if (dead->outcnt() > 0) {
1241 // Recursively remove output edges
1242 _stack.push(dead->raw_out(0), PROCESS_INPUTS);
1243 } else {
1244 // Finished disconnecting all input and output edges.
1245 _stack.pop();
1246 // Remove dead node from iterative worklist
1247 _worklist.remove(dead);
1248 // Constant node that has no out-edges and has only one in-edge from
1249 // root is usually dead. However, sometimes reshaping walk makes
1250 // it reachable by adding use edges. So, we will NOT count Con nodes
1251 // as dead to be conservative about the dead node count at any
1252 // given time.
1253 if (!dead->is_Con()) {
1254 C->record_dead_node(dead->_idx);
1255 }
1256 if (dead->is_macro()) {
1257 C->remove_macro_node(dead);
1258 }
1259 if (dead->is_expensive()) {
1260 C->remove_expensive_node(dead);
1261 }
1262 }
1263 } // while (_stack.is_nonempty())
1264 }
1266 //------------------------------subsume_node-----------------------------------
1267 // Remove users from node 'old' and add them to node 'nn'.
1268 void PhaseIterGVN::subsume_node( Node *old, Node *nn ) {
1269 assert( old != hash_find(old), "should already been removed" );
1270 assert( old != C->top(), "cannot subsume top node");
1271 // Copy debug or profile information to the new version:
1272 C->copy_node_notes_to(nn, old);
1273 // Move users of node 'old' to node 'nn'
1274 for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin; ) {
1275 Node* use = old->last_out(i); // for each use...
1276 // use might need re-hashing (but it won't if it's a new node)
1277 bool is_in_table = _table.hash_delete( use );
1278 // Update use-def info as well
1279 // We remove all occurrences of old within use->in,
1280 // so as to avoid rehashing any node more than once.
1281 // The hash table probe swamps any outer loop overhead.
1282 uint num_edges = 0;
1283 for (uint jmax = use->len(), j = 0; j < jmax; j++) {
1284 if (use->in(j) == old) {
1285 use->set_req(j, nn);
1286 ++num_edges;
1287 }
1288 }
1289 // Insert into GVN hash table if unique
1290 // If a duplicate, 'use' will be cleaned up when pulled off worklist
1291 if( is_in_table ) {
1292 hash_find_insert(use);
1293 }
1294 i -= num_edges; // we deleted 1 or more copies of this edge
1295 }
1297 // Smash all inputs to 'old', isolating him completely
1298 Node *temp = new (C) Node(1);
1299 temp->init_req(0,nn); // Add a use to nn to prevent him from dying
1300 remove_dead_node( old );
1301 temp->del_req(0); // Yank bogus edge
1302 #ifndef PRODUCT
1303 if( VerifyIterativeGVN ) {
1304 for ( int i = 0; i < _verify_window_size; i++ ) {
1305 if ( _verify_window[i] == old )
1306 _verify_window[i] = nn;
1307 }
1308 }
1309 #endif
1310 _worklist.remove(temp); // this can be necessary
1311 temp->destruct(); // reuse the _idx of this little guy
1312 }
1314 //------------------------------add_users_to_worklist--------------------------
1315 void PhaseIterGVN::add_users_to_worklist0( Node *n ) {
1316 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1317 _worklist.push(n->fast_out(i)); // Push on worklist
1318 }
1319 }
1321 void PhaseIterGVN::add_users_to_worklist( Node *n ) {
1322 add_users_to_worklist0(n);
1324 // Move users of node to worklist
1325 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1326 Node* use = n->fast_out(i); // Get use
1328 if( use->is_Multi() || // Multi-definer? Push projs on worklist
1329 use->is_Store() ) // Enable store/load same address
1330 add_users_to_worklist0(use);
1332 // If we changed the receiver type to a call, we need to revisit
1333 // the Catch following the call. It's looking for a non-NULL
1334 // receiver to know when to enable the regular fall-through path
1335 // in addition to the NullPtrException path.
1336 if (use->is_CallDynamicJava() && n == use->in(TypeFunc::Parms)) {
1337 Node* p = use->as_CallDynamicJava()->proj_out(TypeFunc::Control);
1338 if (p != NULL) {
1339 add_users_to_worklist0(p);
1340 }
1341 }
1343 if( use->is_Cmp() ) { // Enable CMP/BOOL optimization
1344 add_users_to_worklist(use); // Put Bool on worklist
1345 // Look for the 'is_x2logic' pattern: "x ? : 0 : 1" and put the
1346 // phi merging either 0 or 1 onto the worklist
1347 if (use->outcnt() > 0) {
1348 Node* bol = use->raw_out(0);
1349 if (bol->outcnt() > 0) {
1350 Node* iff = bol->raw_out(0);
1351 if (iff->outcnt() == 2) {
1352 Node* ifproj0 = iff->raw_out(0);
1353 Node* ifproj1 = iff->raw_out(1);
1354 if (ifproj0->outcnt() > 0 && ifproj1->outcnt() > 0) {
1355 Node* region0 = ifproj0->raw_out(0);
1356 Node* region1 = ifproj1->raw_out(0);
1357 if( region0 == region1 )
1358 add_users_to_worklist0(region0);
1359 }
1360 }
1361 }
1362 }
1363 }
1365 uint use_op = use->Opcode();
1366 // If changed Cast input, check Phi users for simple cycles
1367 if( use->is_ConstraintCast() || use->is_CheckCastPP() ) {
1368 for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
1369 Node* u = use->fast_out(i2);
1370 if (u->is_Phi())
1371 _worklist.push(u);
1372 }
1373 }
1374 // If changed LShift inputs, check RShift users for useless sign-ext
1375 if( use_op == Op_LShiftI ) {
1376 for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
1377 Node* u = use->fast_out(i2);
1378 if (u->Opcode() == Op_RShiftI)
1379 _worklist.push(u);
1380 }
1381 }
1382 // If changed AddI/SubI inputs, check CmpU for range check optimization.
1383 if (use_op == Op_AddI || use_op == Op_SubI) {
1384 for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
1385 Node* u = use->fast_out(i2);
1386 if (u->is_Cmp() && (u->Opcode() == Op_CmpU)) {
1387 _worklist.push(u);
1388 }
1389 }
1390 }
1391 // If changed AddP inputs, check Stores for loop invariant
1392 if( use_op == Op_AddP ) {
1393 for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
1394 Node* u = use->fast_out(i2);
1395 if (u->is_Mem())
1396 _worklist.push(u);
1397 }
1398 }
1399 // If changed initialization activity, check dependent Stores
1400 if (use_op == Op_Allocate || use_op == Op_AllocateArray) {
1401 InitializeNode* init = use->as_Allocate()->initialization();
1402 if (init != NULL) {
1403 Node* imem = init->proj_out(TypeFunc::Memory);
1404 if (imem != NULL) add_users_to_worklist0(imem);
1405 }
1406 }
1407 if (use_op == Op_Initialize) {
1408 Node* imem = use->as_Initialize()->proj_out(TypeFunc::Memory);
1409 if (imem != NULL) add_users_to_worklist0(imem);
1410 }
1411 }
1412 }
1414 /**
1415 * Remove the speculative part of all types that we know of
1416 */
1417 void PhaseIterGVN::remove_speculative_types() {
1418 assert(UseTypeSpeculation, "speculation is off");
1419 for (uint i = 0; i < _types.Size(); i++) {
1420 const Type* t = _types.fast_lookup(i);
1421 if (t != NULL) {
1422 _types.map(i, t->remove_speculative());
1423 }
1424 }
1425 _table.check_no_speculative_types();
1426 }
1428 //=============================================================================
1429 #ifndef PRODUCT
1430 uint PhaseCCP::_total_invokes = 0;
1431 uint PhaseCCP::_total_constants = 0;
1432 #endif
1433 //------------------------------PhaseCCP---------------------------------------
1434 // Conditional Constant Propagation, ala Wegman & Zadeck
1435 PhaseCCP::PhaseCCP( PhaseIterGVN *igvn ) : PhaseIterGVN(igvn) {
1436 NOT_PRODUCT( clear_constants(); )
1437 assert( _worklist.size() == 0, "" );
1438 // Clear out _nodes from IterGVN. Must be clear to transform call.
1439 _nodes.clear(); // Clear out from IterGVN
1440 analyze();
1441 }
1443 #ifndef PRODUCT
1444 //------------------------------~PhaseCCP--------------------------------------
1445 PhaseCCP::~PhaseCCP() {
1446 inc_invokes();
1447 _total_constants += count_constants();
1448 }
1449 #endif
1452 #ifdef ASSERT
1453 static bool ccp_type_widens(const Type* t, const Type* t0) {
1454 assert(t->meet(t0) == t, "Not monotonic");
1455 switch (t->base() == t0->base() ? t->base() : Type::Top) {
1456 case Type::Int:
1457 assert(t0->isa_int()->_widen <= t->isa_int()->_widen, "widen increases");
1458 break;
1459 case Type::Long:
1460 assert(t0->isa_long()->_widen <= t->isa_long()->_widen, "widen increases");
1461 break;
1462 }
1463 return true;
1464 }
1465 #endif //ASSERT
1467 //------------------------------analyze----------------------------------------
1468 void PhaseCCP::analyze() {
1469 // Initialize all types to TOP, optimistic analysis
1470 for (int i = C->unique() - 1; i >= 0; i--) {
1471 _types.map(i,Type::TOP);
1472 }
1474 // Push root onto worklist
1475 Unique_Node_List worklist;
1476 worklist.push(C->root());
1478 // Pull from worklist; compute new value; push changes out.
1479 // This loop is the meat of CCP.
1480 while( worklist.size() ) {
1481 Node *n = worklist.pop();
1482 const Type *t = n->Value(this);
1483 if (t != type(n)) {
1484 assert(ccp_type_widens(t, type(n)), "ccp type must widen");
1485 #ifndef PRODUCT
1486 if( TracePhaseCCP ) {
1487 t->dump();
1488 do { tty->print("\t"); } while (tty->position() < 16);
1489 n->dump();
1490 }
1491 #endif
1492 set_type(n, t);
1493 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1494 Node* m = n->fast_out(i); // Get user
1495 if( m->is_Region() ) { // New path to Region? Must recheck Phis too
1496 for (DUIterator_Fast i2max, i2 = m->fast_outs(i2max); i2 < i2max; i2++) {
1497 Node* p = m->fast_out(i2); // Propagate changes to uses
1498 if( p->bottom_type() != type(p) ) // If not already bottomed out
1499 worklist.push(p); // Propagate change to user
1500 }
1501 }
1502 // If we changed the receiver type to a call, we need to revisit
1503 // the Catch following the call. It's looking for a non-NULL
1504 // receiver to know when to enable the regular fall-through path
1505 // in addition to the NullPtrException path
1506 if (m->is_Call()) {
1507 for (DUIterator_Fast i2max, i2 = m->fast_outs(i2max); i2 < i2max; i2++) {
1508 Node* p = m->fast_out(i2); // Propagate changes to uses
1509 if (p->is_Proj() && p->as_Proj()->_con == TypeFunc::Control && p->outcnt() == 1)
1510 worklist.push(p->unique_out());
1511 }
1512 }
1513 if( m->bottom_type() != type(m) ) // If not already bottomed out
1514 worklist.push(m); // Propagate change to user
1515 }
1516 }
1517 }
1518 }
1520 //------------------------------do_transform-----------------------------------
1521 // Top level driver for the recursive transformer
1522 void PhaseCCP::do_transform() {
1523 // Correct leaves of new-space Nodes; they point to old-space.
1524 C->set_root( transform(C->root())->as_Root() );
1525 assert( C->top(), "missing TOP node" );
1526 assert( C->root(), "missing root" );
1527 }
1529 //------------------------------transform--------------------------------------
1530 // Given a Node in old-space, clone him into new-space.
1531 // Convert any of his old-space children into new-space children.
1532 Node *PhaseCCP::transform( Node *n ) {
1533 Node *new_node = _nodes[n->_idx]; // Check for transformed node
1534 if( new_node != NULL )
1535 return new_node; // Been there, done that, return old answer
1536 new_node = transform_once(n); // Check for constant
1537 _nodes.map( n->_idx, new_node ); // Flag as having been cloned
1539 // Allocate stack of size _nodes.Size()/2 to avoid frequent realloc
1540 GrowableArray <Node *> trstack(C->unique() >> 1);
1542 trstack.push(new_node); // Process children of cloned node
1543 while ( trstack.is_nonempty() ) {
1544 Node *clone = trstack.pop();
1545 uint cnt = clone->req();
1546 for( uint i = 0; i < cnt; i++ ) { // For all inputs do
1547 Node *input = clone->in(i);
1548 if( input != NULL ) { // Ignore NULLs
1549 Node *new_input = _nodes[input->_idx]; // Check for cloned input node
1550 if( new_input == NULL ) {
1551 new_input = transform_once(input); // Check for constant
1552 _nodes.map( input->_idx, new_input );// Flag as having been cloned
1553 trstack.push(new_input);
1554 }
1555 assert( new_input == clone->in(i), "insanity check");
1556 }
1557 }
1558 }
1559 return new_node;
1560 }
1563 //------------------------------transform_once---------------------------------
1564 // For PhaseCCP, transformation is IDENTITY unless Node computed a constant.
1565 Node *PhaseCCP::transform_once( Node *n ) {
1566 const Type *t = type(n);
1567 // Constant? Use constant Node instead
1568 if( t->singleton() ) {
1569 Node *nn = n; // Default is to return the original constant
1570 if( t == Type::TOP ) {
1571 // cache my top node on the Compile instance
1572 if( C->cached_top_node() == NULL || C->cached_top_node()->in(0) == NULL ) {
1573 C->set_cached_top_node( ConNode::make(C, Type::TOP) );
1574 set_type(C->top(), Type::TOP);
1575 }
1576 nn = C->top();
1577 }
1578 if( !n->is_Con() ) {
1579 if( t != Type::TOP ) {
1580 nn = makecon(t); // ConNode::make(t);
1581 NOT_PRODUCT( inc_constants(); )
1582 } else if( n->is_Region() ) { // Unreachable region
1583 // Note: nn == C->top()
1584 n->set_req(0, NULL); // Cut selfreference
1585 // Eagerly remove dead phis to avoid phis copies creation.
1586 for (DUIterator i = n->outs(); n->has_out(i); i++) {
1587 Node* m = n->out(i);
1588 if( m->is_Phi() ) {
1589 assert(type(m) == Type::TOP, "Unreachable region should not have live phis.");
1590 replace_node(m, nn);
1591 --i; // deleted this phi; rescan starting with next position
1592 }
1593 }
1594 }
1595 replace_node(n,nn); // Update DefUse edges for new constant
1596 }
1597 return nn;
1598 }
1600 // If x is a TypeNode, capture any more-precise type permanently into Node
1601 if (t != n->bottom_type()) {
1602 hash_delete(n); // changing bottom type may force a rehash
1603 n->raise_bottom_type(t);
1604 _worklist.push(n); // n re-enters the hash table via the worklist
1605 }
1607 // Idealize graph using DU info. Must clone() into new-space.
1608 // DU info is generally used to show profitability, progress or safety
1609 // (but generally not needed for correctness).
1610 Node *nn = n->Ideal_DU_postCCP(this);
1612 // TEMPORARY fix to ensure that 2nd GVN pass eliminates NULL checks
1613 switch( n->Opcode() ) {
1614 case Op_FastLock: // Revisit FastLocks for lock coarsening
1615 case Op_If:
1616 case Op_CountedLoopEnd:
1617 case Op_Region:
1618 case Op_Loop:
1619 case Op_CountedLoop:
1620 case Op_Conv2B:
1621 case Op_Opaque1:
1622 case Op_Opaque2:
1623 _worklist.push(n);
1624 break;
1625 default:
1626 break;
1627 }
1628 if( nn ) {
1629 _worklist.push(n);
1630 // Put users of 'n' onto worklist for second igvn transform
1631 add_users_to_worklist(n);
1632 return nn;
1633 }
1635 return n;
1636 }
1638 //---------------------------------saturate------------------------------------
1639 const Type* PhaseCCP::saturate(const Type* new_type, const Type* old_type,
1640 const Type* limit_type) const {
1641 const Type* wide_type = new_type->widen(old_type, limit_type);
1642 if (wide_type != new_type) { // did we widen?
1643 // If so, we may have widened beyond the limit type. Clip it back down.
1644 new_type = wide_type->filter(limit_type);
1645 }
1646 return new_type;
1647 }
1649 //------------------------------print_statistics-------------------------------
1650 #ifndef PRODUCT
1651 void PhaseCCP::print_statistics() {
1652 tty->print_cr("CCP: %d constants found: %d", _total_invokes, _total_constants);
1653 }
1654 #endif
1657 //=============================================================================
1658 #ifndef PRODUCT
1659 uint PhasePeephole::_total_peepholes = 0;
1660 #endif
1661 //------------------------------PhasePeephole----------------------------------
1662 // Conditional Constant Propagation, ala Wegman & Zadeck
1663 PhasePeephole::PhasePeephole( PhaseRegAlloc *regalloc, PhaseCFG &cfg )
1664 : PhaseTransform(Peephole), _regalloc(regalloc), _cfg(cfg) {
1665 NOT_PRODUCT( clear_peepholes(); )
1666 }
1668 #ifndef PRODUCT
1669 //------------------------------~PhasePeephole---------------------------------
1670 PhasePeephole::~PhasePeephole() {
1671 _total_peepholes += count_peepholes();
1672 }
1673 #endif
1675 //------------------------------transform--------------------------------------
1676 Node *PhasePeephole::transform( Node *n ) {
1677 ShouldNotCallThis();
1678 return NULL;
1679 }
1681 //------------------------------do_transform-----------------------------------
1682 void PhasePeephole::do_transform() {
1683 bool method_name_not_printed = true;
1685 // Examine each basic block
1686 for (uint block_number = 1; block_number < _cfg.number_of_blocks(); ++block_number) {
1687 Block* block = _cfg.get_block(block_number);
1688 bool block_not_printed = true;
1690 // and each instruction within a block
1691 uint end_index = block->number_of_nodes();
1692 // block->end_idx() not valid after PhaseRegAlloc
1693 for( uint instruction_index = 1; instruction_index < end_index; ++instruction_index ) {
1694 Node *n = block->get_node(instruction_index);
1695 if( n->is_Mach() ) {
1696 MachNode *m = n->as_Mach();
1697 int deleted_count = 0;
1698 // check for peephole opportunities
1699 MachNode *m2 = m->peephole( block, instruction_index, _regalloc, deleted_count, C );
1700 if( m2 != NULL ) {
1701 #ifndef PRODUCT
1702 if( PrintOptoPeephole ) {
1703 // Print method, first time only
1704 if( C->method() && method_name_not_printed ) {
1705 C->method()->print_short_name(); tty->cr();
1706 method_name_not_printed = false;
1707 }
1708 // Print this block
1709 if( Verbose && block_not_printed) {
1710 tty->print_cr("in block");
1711 block->dump();
1712 block_not_printed = false;
1713 }
1714 // Print instructions being deleted
1715 for( int i = (deleted_count - 1); i >= 0; --i ) {
1716 block->get_node(instruction_index-i)->as_Mach()->format(_regalloc); tty->cr();
1717 }
1718 tty->print_cr("replaced with");
1719 // Print new instruction
1720 m2->format(_regalloc);
1721 tty->print("\n\n");
1722 }
1723 #endif
1724 // Remove old nodes from basic block and update instruction_index
1725 // (old nodes still exist and may have edges pointing to them
1726 // as register allocation info is stored in the allocator using
1727 // the node index to live range mappings.)
1728 uint safe_instruction_index = (instruction_index - deleted_count);
1729 for( ; (instruction_index > safe_instruction_index); --instruction_index ) {
1730 block->remove_node( instruction_index );
1731 }
1732 // install new node after safe_instruction_index
1733 block->insert_node(m2, safe_instruction_index + 1);
1734 end_index = block->number_of_nodes() - 1; // Recompute new block size
1735 NOT_PRODUCT( inc_peepholes(); )
1736 }
1737 }
1738 }
1739 }
1740 }
1742 //------------------------------print_statistics-------------------------------
1743 #ifndef PRODUCT
1744 void PhasePeephole::print_statistics() {
1745 tty->print_cr("Peephole: peephole rules applied: %d", _total_peepholes);
1746 }
1747 #endif
1750 //=============================================================================
1751 //------------------------------set_req_X--------------------------------------
1752 void Node::set_req_X( uint i, Node *n, PhaseIterGVN *igvn ) {
1753 assert( is_not_dead(n), "can not use dead node");
1754 assert( igvn->hash_find(this) != this, "Need to remove from hash before changing edges" );
1755 Node *old = in(i);
1756 set_req(i, n);
1758 // old goes dead?
1759 if( old ) {
1760 switch (old->outcnt()) {
1761 case 0:
1762 // Put into the worklist to kill later. We do not kill it now because the
1763 // recursive kill will delete the current node (this) if dead-loop exists
1764 if (!old->is_top())
1765 igvn->_worklist.push( old );
1766 break;
1767 case 1:
1768 if( old->is_Store() || old->has_special_unique_user() )
1769 igvn->add_users_to_worklist( old );
1770 break;
1771 case 2:
1772 if( old->is_Store() )
1773 igvn->add_users_to_worklist( old );
1774 if( old->Opcode() == Op_Region )
1775 igvn->_worklist.push(old);
1776 break;
1777 case 3:
1778 if( old->Opcode() == Op_Region ) {
1779 igvn->_worklist.push(old);
1780 igvn->add_users_to_worklist( old );
1781 }
1782 break;
1783 default:
1784 break;
1785 }
1786 }
1788 }
1790 //-------------------------------replace_by-----------------------------------
1791 // Using def-use info, replace one node for another. Follow the def-use info
1792 // to all users of the OLD node. Then make all uses point to the NEW node.
1793 void Node::replace_by(Node *new_node) {
1794 assert(!is_top(), "top node has no DU info");
1795 for (DUIterator_Last imin, i = last_outs(imin); i >= imin; ) {
1796 Node* use = last_out(i);
1797 uint uses_found = 0;
1798 for (uint j = 0; j < use->len(); j++) {
1799 if (use->in(j) == this) {
1800 if (j < use->req())
1801 use->set_req(j, new_node);
1802 else use->set_prec(j, new_node);
1803 uses_found++;
1804 }
1805 }
1806 i -= uses_found; // we deleted 1 or more copies of this edge
1807 }
1808 }
1810 //=============================================================================
1811 //-----------------------------------------------------------------------------
1812 void Type_Array::grow( uint i ) {
1813 if( !_max ) {
1814 _max = 1;
1815 _types = (const Type**)_a->Amalloc( _max * sizeof(Type*) );
1816 _types[0] = NULL;
1817 }
1818 uint old = _max;
1819 while( i >= _max ) _max <<= 1; // Double to fit
1820 _types = (const Type**)_a->Arealloc( _types, old*sizeof(Type*),_max*sizeof(Type*));
1821 memset( &_types[old], 0, (_max-old)*sizeof(Type*) );
1822 }
1824 //------------------------------dump-------------------------------------------
1825 #ifndef PRODUCT
1826 void Type_Array::dump() const {
1827 uint max = Size();
1828 for( uint i = 0; i < max; i++ ) {
1829 if( _types[i] != NULL ) {
1830 tty->print(" %d\t== ", i); _types[i]->dump(); tty->cr();
1831 }
1832 }
1833 }
1834 #endif
