1 /*
2 * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2024, 2025, Alibaba Group Holding Limited. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #ifndef SHARE_OPTO_NODE_HPP
27 #define SHARE_OPTO_NODE_HPP
28
29 #include "libadt/vectset.hpp"
30 #include "opto/compile.hpp"
31 #include "opto/type.hpp"
32 #include "utilities/copy.hpp"
33
34 // Portions of code courtesy of Clifford Click
35
36 // Optimization - Graph Style
37
38
39 class AbstractLockNode;
40 class AddNode;
41 class AddPNode;
42 class AliasInfo;
43 class AllocateArrayNode;
44 class AllocateNode;
45 class ArrayCopyNode;
46 class BaseCountedLoopNode;
47 class BaseCountedLoopEndNode;
48 class BlackholeNode;
49 class Block;
50 class BoolNode;
51 class BoxLockNode;
52 class CMoveNode;
53 class CallDynamicJavaNode;
54 class CallJavaNode;
55 class CallLeafNode;
56 class CallLeafNoFPNode;
57 class CallLeafPureNode;
58 class CallNode;
59 class CallRuntimeNode;
60 class CallStaticJavaNode;
61 class CastFFNode;
62 class CastHHNode;
63 class CastDDNode;
64 class CastVVNode;
65 class CastIINode;
66 class CastLLNode;
67 class CastPPNode;
68 class CatchNode;
69 class CatchProjNode;
70 class CheckCastPPNode;
71 class ClearArrayNode;
72 class CmpNode;
73 class CodeBuffer;
74 class ConstraintCastNode;
75 class ConNode;
76 class ConINode;
77 class ConvertNode;
78 class CompareAndSwapNode;
79 class CompareAndExchangeNode;
80 class CountedLoopNode;
81 class CountedLoopEndNode;
82 class DecodeNarrowPtrNode;
83 class DecodeNNode;
84 class DecodeNKlassNode;
85 class EncodeNarrowPtrNode;
86 class EncodePNode;
87 class EncodePKlassNode;
88 class FastLockNode;
89 class FastUnlockNode;
90 class FlatArrayCheckNode;
91 class HaltNode;
92 class IfNode;
93 class IfProjNode;
94 class IfFalseNode;
95 class IfTrueNode;
96 class InitializeNode;
97 class JVMState;
98 class JumpNode;
99 class JumpProjNode;
100 class LoadNode;
101 class LoadStoreNode;
102 class LoadStoreConditionalNode;
103 class LockNode;
104 class LongCountedLoopNode;
105 class LongCountedLoopEndNode;
106 class LoopNode;
107 class LShiftNode;
108 class MachBranchNode;
109 class MachCallDynamicJavaNode;
110 class MachCallJavaNode;
111 class MachCallLeafNode;
112 class MachCallNode;
113 class MachCallRuntimeNode;
114 class MachCallStaticJavaNode;
115 class MachConstantBaseNode;
116 class MachConstantNode;
117 class MachGotoNode;
118 class MachIfNode;
119 class MachJumpNode;
120 class MachNode;
121 class MachNullCheckNode;
122 class MachProjNode;
123 class MachPrologNode;
124 class MachReturnNode;
125 class MachSafePointNode;
126 class MachSpillCopyNode;
127 class MachTempNode;
128 class MachMergeNode;
129 class MachMemBarNode;
130 class MachVEPNode;
131 class Matcher;
132 class MemBarNode;
133 class MemBarStoreStoreNode;
134 class MemNode;
135 class MergeMemNode;
136 class MinMaxNode;
137 class MoveNode;
138 class MulNode;
139 class MultiNode;
140 class MultiBranchNode;
141 class NarrowMemProjNode;
142 class NegNode;
143 class NegVNode;
144 class NeverBranchNode;
145 class Opaque1Node;
146 class OpaqueLoopInitNode;
147 class OpaqueLoopStrideNode;
148 class OpaqueMultiversioningNode;
149 class OpaqueConstantBoolNode;
150 class OpaqueInitializedAssertionPredicateNode;
151 class OpaqueTemplateAssertionPredicateNode;
152 class OuterStripMinedLoopNode;
153 class OuterStripMinedLoopEndNode;
154 class Node;
155 class Node_Array;
156 class Node_List;
157 class Node_Stack;
158 class OopMap;
159 class ParmNode;
160 class ParsePredicateNode;
161 class PCTableNode;
162 class PhaseCCP;
163 class PhaseGVN;
164 class PhaseIdealLoop;
165 class PhaseIterGVN;
166 class PhaseRegAlloc;
167 class PhaseTransform;
168 class PhaseValues;
169 class PhiNode;
170 class Pipeline;
171 class PopulateIndexNode;
172 class ProjNode;
173 class RangeCheckNode;
174 class ReachabilityFenceNode;
175 class ReductionNode;
176 class RegMask;
177 class RegionNode;
178 class RootNode;
179 class SafePointNode;
180 class SafePointScalarObjectNode;
181 class SafePointScalarMergeNode;
182 class SaturatingVectorNode;
183 class StartNode;
184 class State;
185 class StoreNode;
186 class SubNode;
187 class SubTypeCheckNode;
188 class Type;
189 class TypeNode;
190 class UnlockNode;
191 class InlineTypeNode;
192 class LoadFlatNode;
193 class StoreFlatNode;
194 class VectorNode;
195 class LoadVectorNode;
196 class LoadVectorMaskedNode;
197 class StoreVectorMaskedNode;
198 class LoadVectorGatherNode;
199 class LoadVectorGatherMaskedNode;
200 class StoreVectorNode;
201 class StoreVectorScatterNode;
202 class StoreVectorScatterMaskedNode;
203 class VerifyVectorAlignmentNode;
204 class VectorMaskCmpNode;
205 class VectorUnboxNode;
206 class VectorSet;
207 class VectorReinterpretNode;
208 class ShiftVNode;
209 class MulVLNode;
210 class ExpandVNode;
211 class CompressVNode;
212 class CompressMNode;
213 class C2_MacroAssembler;
214
215
216 #ifndef OPTO_DU_ITERATOR_ASSERT
217 #ifdef ASSERT
218 #define OPTO_DU_ITERATOR_ASSERT 1
219 #else
220 #define OPTO_DU_ITERATOR_ASSERT 0
221 #endif
222 #endif //OPTO_DU_ITERATOR_ASSERT
223
224 #if OPTO_DU_ITERATOR_ASSERT
225 class DUIterator;
226 class DUIterator_Fast;
227 class DUIterator_Last;
228 #else
229 typedef uint DUIterator;
230 typedef Node** DUIterator_Fast;
231 typedef Node** DUIterator_Last;
232 #endif
233
234 typedef ResizeableHashTable<Node*, Node*, AnyObj::RESOURCE_AREA, mtCompiler> OrigToNewHashtable;
235
236 // Node Sentinel
237 #define NodeSentinel (Node*)-1
238
239 // Unknown count frequency
240 #define COUNT_UNKNOWN (-1.0f)
241
242 //------------------------------Node-------------------------------------------
243 // Nodes define actions in the program. They create values, which have types.
244 // They are both vertices in a directed graph and program primitives. Nodes
245 // are labeled; the label is the "opcode", the primitive function in the lambda
246 // calculus sense that gives meaning to the Node. Node inputs are ordered (so
247 // that "a-b" is different from "b-a"). The inputs to a Node are the inputs to
248 // the Node's function. These inputs also define a Type equation for the Node.
249 // Solving these Type equations amounts to doing dataflow analysis.
250 // Control and data are uniformly represented in the graph. Finally, Nodes
251 // have a unique dense integer index which is used to index into side arrays
252 // whenever I have phase-specific information.
253
254 class Node {
255
256 // Lots of restrictions on cloning Nodes
257 NONCOPYABLE(Node);
258
259 public:
260 friend class Compile;
261 #if OPTO_DU_ITERATOR_ASSERT
262 friend class DUIterator_Common;
263 friend class DUIterator;
264 friend class DUIterator_Fast;
265 friend class DUIterator_Last;
266 #endif
267
268 // Because Nodes come and go, I define an Arena of Node structures to pull
269 // from. This should allow fast access to node creation & deletion. This
270 // field is a local cache of a value defined in some "program fragment" for
271 // which these Nodes are just a part of.
272
273 inline void* operator new(size_t x) throw() {
274 Compile* C = Compile::current();
275 Node* n = (Node*)C->node_arena()->AmallocWords(x);
276 return (void*)n;
277 }
278
279 // Delete is a NOP
280 void operator delete( void *ptr ) {}
281 // Fancy destructor; eagerly attempt to reclaim Node numberings and storage
282 void destruct(PhaseValues* phase);
283
284 // Create a new Node. Required is the number is of inputs required for
285 // semantic correctness.
286 Node( uint required );
287
288 // Create a new Node with given input edges.
289 // This version requires use of the "edge-count" new.
290 // E.g. new (C,3) FooNode( C, nullptr, left, right );
291 Node( Node *n0 );
292 Node( Node *n0, Node *n1 );
293 Node( Node *n0, Node *n1, Node *n2 );
294 Node( Node *n0, Node *n1, Node *n2, Node *n3 );
295 Node( Node *n0, Node *n1, Node *n2, Node *n3, Node *n4 );
296 Node( Node *n0, Node *n1, Node *n2, Node *n3, Node *n4, Node *n5 );
297 Node( Node *n0, Node *n1, Node *n2, Node *n3,
298 Node *n4, Node *n5, Node *n6 );
299
300 // Clone an inherited Node given only the base Node type.
301 Node* clone() const;
302
303 // Clone a Node, immediately supplying one or two new edges.
304 // The first and second arguments, if non-null, replace in(1) and in(2),
305 // respectively.
306 Node* clone_with_data_edge(Node* in1, Node* in2 = nullptr) const {
307 Node* nn = clone();
308 if (in1 != nullptr) nn->set_req(1, in1);
309 if (in2 != nullptr) nn->set_req(2, in2);
310 return nn;
311 }
312
313 private:
314 // Shared setup for the above constructors.
315 // Handles all interactions with Compile::current.
316 // Puts initial values in all Node fields except _idx.
317 // Returns the initial value for _idx, which cannot
318 // be initialized by assignment.
319 inline int Init(int req);
320
321 //----------------- input edge handling
322 protected:
323 friend class PhaseCFG; // Access to address of _in array elements
324 Node **_in; // Array of use-def references to Nodes
325 Node **_out; // Array of def-use references to Nodes
326
327 // Input edges are split into two categories. Required edges are required
328 // for semantic correctness; order is important and nulls are allowed.
329 // Precedence edges are used to help determine execution order and are
330 // added, e.g., for scheduling purposes. They are unordered and not
331 // duplicated; they have no embedded nulls. Edges from 0 to _cnt-1
332 // are required, from _cnt to _max-1 are precedence edges.
333 node_idx_t _cnt; // Total number of required Node inputs.
334
335 node_idx_t _max; // Actual length of input array.
336
337 // Output edges are an unordered list of def-use edges which exactly
338 // correspond to required input edges which point from other nodes
339 // to this one. Thus the count of the output edges is the number of
340 // users of this node.
341 node_idx_t _outcnt; // Total number of Node outputs.
342
343 node_idx_t _outmax; // Actual length of output array.
344
345 // Grow the actual input array to the next larger power-of-2 bigger than len.
346 void grow( uint len );
347 // Grow the output array to the next larger power-of-2 bigger than len.
348 void out_grow( uint len );
349 // Resize input or output array to grow it to the next larger power-of-2
350 // bigger than len.
351 void resize_array(Node**& array, node_idx_t& max_size, uint len, bool needs_clearing);
352
353 public:
354 // Each Node is assigned a unique small/dense number. This number is used
355 // to index into auxiliary arrays of data and bit vectors.
356 // The value of _idx can be changed using the set_idx() method.
357 //
358 // The PhaseRenumberLive phase renumbers nodes based on liveness information.
359 // Therefore, it updates the value of the _idx field. The parse-time _idx is
360 // preserved in _parse_idx.
361 node_idx_t _idx;
362 DEBUG_ONLY(const node_idx_t _parse_idx;)
363 // IGV node identifier. Two nodes, possibly in different compilation phases,
364 // have the same IGV identifier if (and only if) they are the very same node
365 // (same memory address) or one is "derived" from the other (by e.g.
366 // renumbering or matching). This identifier makes it possible to follow the
367 // entire lifetime of a node in IGV even if its C2 identifier (_idx) changes.
368 NOT_PRODUCT(node_idx_t _igv_idx;)
369
370 // Get the (read-only) number of input edges
371 uint req() const { return _cnt; }
372 uint len() const { return _max; }
373 // Get the (read-only) number of output edges
374 uint outcnt() const { return _outcnt; }
375
376 #if OPTO_DU_ITERATOR_ASSERT
377 // Iterate over the out-edges of this node. Deletions are illegal.
378 inline DUIterator outs() const;
379 // Use this when the out array might have changed to suppress asserts.
380 inline DUIterator& refresh_out_pos(DUIterator& i) const;
381 // Does the node have an out at this position? (Used for iteration.)
382 inline bool has_out(DUIterator& i) const;
383 inline Node* out(DUIterator& i) const;
384 // Iterate over the out-edges of this node. All changes are illegal.
385 inline DUIterator_Fast fast_outs(DUIterator_Fast& max) const;
386 inline Node* fast_out(DUIterator_Fast& i) const;
387 // Iterate over the out-edges of this node, deleting one at a time.
388 inline DUIterator_Last last_outs(DUIterator_Last& min) const;
389 inline Node* last_out(DUIterator_Last& i) const;
390 // The inline bodies of all these methods are after the iterator definitions.
391 #else
392 // Iterate over the out-edges of this node. Deletions are illegal.
393 // This iteration uses integral indexes, to decouple from array reallocations.
394 DUIterator outs() const { return 0; }
395 // Use this when the out array might have changed to suppress asserts.
396 DUIterator refresh_out_pos(DUIterator i) const { return i; }
397
398 // Reference to the i'th output Node. Error if out of bounds.
399 Node* out(DUIterator i) const { assert(i < _outcnt, "oob"); return _out[i]; }
400 // Does the node have an out at this position? (Used for iteration.)
401 bool has_out(DUIterator i) const { return i < _outcnt; }
402
403 // Iterate over the out-edges of this node. All changes are illegal.
404 // This iteration uses a pointer internal to the out array.
405 DUIterator_Fast fast_outs(DUIterator_Fast& max) const {
406 Node** out = _out;
407 // Assign a limit pointer to the reference argument:
408 max = out + (ptrdiff_t)_outcnt;
409 // Return the base pointer:
410 return out;
411 }
412 Node* fast_out(DUIterator_Fast i) const { return *i; }
413 // Iterate over the out-edges of this node, deleting one at a time.
414 // This iteration uses a pointer internal to the out array.
415 DUIterator_Last last_outs(DUIterator_Last& min) const {
416 Node** out = _out;
417 // Assign a limit pointer to the reference argument:
418 min = out;
419 // Return the pointer to the start of the iteration:
420 return out + (ptrdiff_t)_outcnt - 1;
421 }
422 Node* last_out(DUIterator_Last i) const { return *i; }
423 #endif
424
425 // Reference to the i'th input Node. Error if out of bounds.
426 Node* in(uint i) const { assert(i < _max, "oob: i=%d, _max=%d", i, _max); return _in[i]; }
427 // Reference to the i'th input Node. null if out of bounds.
428 Node* lookup(uint i) const { return ((i < _max) ? _in[i] : nullptr); }
429 // Reference to the i'th output Node. Error if out of bounds.
430 // Use this accessor sparingly. We are going trying to use iterators instead.
431 Node* raw_out(uint i) const { assert(i < _outcnt,"oob"); return _out[i]; }
432 // Return the unique out edge.
433 Node* unique_out() const { assert(_outcnt==1,"not unique"); return _out[0]; }
434
435 // In some cases, a node n is only used by a single use, but the use may use
436 // n once or multiple times:
437 // use = ConvF2I(this)
438 // use = AddI(this, this)
439 Node* unique_multiple_edges_out_or_null() const;
440
441 // Delete out edge at position 'i' by moving last out edge to position 'i'
442 void raw_del_out(uint i) {
443 assert(i < _outcnt,"oob");
444 assert(_outcnt > 0,"oob");
445 #if OPTO_DU_ITERATOR_ASSERT
446 // Record that a change happened here.
447 DEBUG_ONLY(_last_del = _out[i]; ++_del_tick);
448 #endif
449 _out[i] = _out[--_outcnt];
450 // Smash the old edge so it can't be used accidentally.
451 DEBUG_ONLY(_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef);
452 }
453
454 #ifdef ASSERT
455 bool is_dead() const;
456 static bool is_not_dead(const Node* n);
457 bool is_reachable_from_root() const;
458 #endif
459 // Check whether node has become unreachable
460 bool is_unreachable(PhaseIterGVN &igvn) const;
461
462 // Does the node have any immediate non-debug uses?
463 bool has_non_debug_uses() const;
464
465 // Set a required input edge, also updates corresponding output edge
466 void add_req( Node *n ); // Append a NEW required input
467 void add_req( Node *n0, Node *n1 ) {
468 add_req(n0); add_req(n1); }
469 void add_req( Node *n0, Node *n1, Node *n2 ) {
470 add_req(n0); add_req(n1); add_req(n2); }
471 void add_req_batch( Node* n, uint m ); // Append m NEW required inputs (all n).
472 void del_req( uint idx ); // Delete required edge & compact
473 void del_req_ordered( uint idx ); // Delete required edge & compact with preserved order
474 void ins_req( uint i, Node *n ); // Insert a NEW required input
475 void set_req( uint i, Node *n ) {
476 assert( is_not_dead(n), "can not use dead node");
477 assert( i < _cnt, "oob: i=%d, _cnt=%d", i, _cnt);
478 assert( !VerifyHashTableKeys || _hash_lock == 0,
479 "remove node from hash table before modifying it");
480 Node** p = &_in[i]; // cache this._in, across the del_out call
481 if (*p != nullptr) (*p)->del_out((Node *)this);
482 (*p) = n;
483 if (n != nullptr) n->add_out((Node *)this);
484 Compile::current()->record_modified_node(this);
485 }
486 // Light version of set_req() to init inputs after node creation.
487 void init_req( uint i, Node *n ) {
488 assert( (i == 0 && this == n) ||
489 is_not_dead(n), "can not use dead node");
490 assert( i < _cnt, "oob");
491 assert( !VerifyHashTableKeys || _hash_lock == 0,
492 "remove node from hash table before modifying it");
493 assert( _in[i] == nullptr, "sanity");
494 _in[i] = n;
495 if (n != nullptr) n->add_out((Node *)this);
496 Compile::current()->record_modified_node(this);
497 }
498 // Find first occurrence of n among my edges:
499 int find_edge(Node* n);
500 int find_prec_edge(Node* n) {
501 for (uint i = req(); i < len(); i++) {
502 if (_in[i] == n) return i;
503 if (_in[i] == nullptr) {
504 DEBUG_ONLY( while ((++i) < len()) assert(_in[i] == nullptr, "Gap in prec edges!"); )
505 break;
506 }
507 }
508 return -1;
509 }
510 int replace_edge(Node* old, Node* neww, PhaseGVN* gvn = nullptr);
511 int replace_edges_in_range(Node* old, Node* neww, int start, int end, PhaseGVN* gvn);
512 // null out all inputs to eliminate incoming Def-Use edges.
513 void disconnect_inputs(Compile* C);
514
515 // Quickly, return true if and only if I am Compile::current()->top().
516 bool is_top() const {
517 assert((this == (Node*) Compile::current()->top()) == (_out == nullptr), "");
518 return (_out == nullptr);
519 }
520 // Reaffirm invariants for is_top. (Only from Compile::set_cached_top_node.)
521 void setup_is_top();
522
523 // Strip away casting. (It is depth-limited.)
524 Node* uncast(bool keep_deps = false) const;
525 // Return whether two Nodes are equivalent, after stripping casting.
526 bool eqv_uncast(const Node* n, bool keep_deps = false) const {
527 return (this->uncast(keep_deps) == n->uncast(keep_deps));
528 }
529
530 // Find out of current node that matches opcode.
531 Node* find_out_with(int opcode);
532 // Return true if the current node has an out that matches opcode.
533 bool has_out_with(int opcode);
534 // Return true if the current node has an out that matches any of the opcodes.
535 bool has_out_with(int opcode1, int opcode2, int opcode3, int opcode4);
536
537 private:
538 static Node* uncast_helper(const Node* n, bool keep_deps);
539
540 // Add an output edge to the end of the list
541 void add_out( Node *n ) {
542 if (is_top()) return;
543 if( _outcnt == _outmax ) out_grow(_outcnt);
544 _out[_outcnt++] = n;
545 }
546 // Delete an output edge
547 void del_out( Node *n ) {
548 if (is_top()) return;
549 Node** outp = &_out[_outcnt];
550 // Find and remove n
551 do {
552 assert(outp > _out, "Missing Def-Use edge");
553 } while (*--outp != n);
554 *outp = _out[--_outcnt];
555 // Smash the old edge so it can't be used accidentally.
556 DEBUG_ONLY(_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef);
557 // Record that a change happened here.
558 #if OPTO_DU_ITERATOR_ASSERT
559 DEBUG_ONLY(_last_del = n; ++_del_tick);
560 #endif
561 }
562 // Close gap after removing edge.
563 void close_prec_gap_at(uint gap) {
564 assert(_cnt <= gap && gap < _max, "no valid prec edge");
565 uint i = gap;
566 Node *last = nullptr;
567 for (; i < _max-1; ++i) {
568 Node *next = _in[i+1];
569 if (next == nullptr) break;
570 last = next;
571 }
572 _in[gap] = last; // Move last slot to empty one.
573 _in[i] = nullptr; // null out last slot.
574 }
575
576 public:
577 // Globally replace this node by a given new node, updating all uses.
578 void replace_by(Node* new_node);
579 // Globally replace this node by a given new node, updating all uses
580 // and cutting input edges of old node.
581 void subsume_by(Node* new_node, Compile* c) {
582 replace_by(new_node);
583 disconnect_inputs(c);
584 }
585 void set_req_X(uint i, Node *n, PhaseIterGVN *igvn);
586 void set_req_X(uint i, Node *n, PhaseGVN *gvn);
587 // Find the one non-null required input. RegionNode only
588 Node *nonnull_req() const;
589 // Add or remove precedence edges
590 void add_prec( Node *n );
591 void rm_prec( uint i );
592
593 // Note: prec(i) will not necessarily point to n if edge already exists.
594 void set_prec( uint i, Node *n ) {
595 assert(i < _max, "oob: i=%d, _max=%d", i, _max);
596 assert(is_not_dead(n), "can not use dead node");
597 assert(i >= _cnt, "not a precedence edge");
598 // Avoid spec violation: duplicated prec edge.
599 if (_in[i] == n) return;
600 if (n == nullptr || find_prec_edge(n) != -1) {
601 rm_prec(i);
602 return;
603 }
604 if (_in[i] != nullptr) _in[i]->del_out((Node *)this);
605 _in[i] = n;
606 n->add_out((Node *)this);
607 Compile::current()->record_modified_node(this);
608 }
609
610 // Set this node's index, used by cisc_version to replace current node
611 void set_idx(uint new_idx) {
612 _idx = new_idx;
613 }
614 // Swap input edge order. (Edge indexes i1 and i2 are usually 1 and 2.)
615 void swap_edges(uint i1, uint i2) {
616 DEBUG_ONLY(uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH);
617 // Def-Use info is unchanged
618 Node* n1 = in(i1);
619 Node* n2 = in(i2);
620 _in[i1] = n2;
621 _in[i2] = n1;
622 // If this node is in the hash table, make sure it doesn't need a rehash.
623 assert(check_hash == NO_HASH || check_hash == hash(), "edge swap must preserve hash code");
624 // Flip swapped edges flag.
625 if (has_swapped_edges()) {
626 remove_flag(Node::Flag_has_swapped_edges);
627 } else {
628 add_flag(Node::Flag_has_swapped_edges);
629 }
630 }
631
632 // Iterators over input Nodes for a Node X are written as:
633 // for( i = 0; i < X.req(); i++ ) ... X[i] ...
634 // NOTE: Required edges can contain embedded null pointers.
635
636 //----------------- Other Node Properties
637
638 // Generate class IDs for (some) ideal nodes so that it is possible to determine
639 // the type of a node using a non-virtual method call (the method is_<Node>() below).
640 //
641 // A class ID of an ideal node is a set of bits. In a class ID, a single bit determines
642 // the type of the node the ID represents; another subset of an ID's bits are reserved
643 // for the superclasses of the node represented by the ID.
644 //
645 // By design, if A is a supertype of B, A.is_B() returns true and B.is_A()
646 // returns false. A.is_A() returns true.
647 //
648 // If two classes, A and B, have the same superclass, a different bit of A's class id
649 // is reserved for A's type than for B's type. That bit is specified by the third
650 // parameter in the macro DEFINE_CLASS_ID.
651 //
652 // By convention, classes with deeper hierarchy are declared first. Moreover,
653 // classes with the same hierarchy depth are sorted by usage frequency.
654 //
655 // The query method masks the bits to cut off bits of subclasses and then compares
656 // the result with the class id (see the macro DEFINE_CLASS_QUERY below).
657 //
658 // Class_MachCall=30, ClassMask_MachCall=31
659 // 12 8 4 0
660 // 0 0 0 0 0 0 0 0 1 1 1 1 0
661 // | | | |
662 // | | | Bit_Mach=2
663 // | | Bit_MachReturn=4
664 // | Bit_MachSafePoint=8
665 // Bit_MachCall=16
666 //
667 // Class_CountedLoop=56, ClassMask_CountedLoop=63
668 // 12 8 4 0
669 // 0 0 0 0 0 0 0 1 1 1 0 0 0
670 // | | |
671 // | | Bit_Region=8
672 // | Bit_Loop=16
673 // Bit_CountedLoop=32
674
675 #define DEFINE_CLASS_ID(cl, supcl, subn) \
676 Bit_##cl = (Class_##supcl == 0) ? 1 << subn : (Bit_##supcl) << (1 + subn) , \
677 Class_##cl = Class_##supcl + Bit_##cl , \
678 ClassMask_##cl = ((Bit_##cl << 1) - 1) ,
679
680 // This enum is used only for C2 ideal and mach nodes with is_<node>() methods
681 // so that its values fit into 32 bits.
682 enum NodeClasses {
683 Bit_Node = 0x00000000,
684 Class_Node = 0x00000000,
685 ClassMask_Node = 0xFFFFFFFF,
686
687 DEFINE_CLASS_ID(Multi, Node, 0)
688 DEFINE_CLASS_ID(SafePoint, Multi, 0)
689 DEFINE_CLASS_ID(Call, SafePoint, 0)
690 DEFINE_CLASS_ID(CallJava, Call, 0)
691 DEFINE_CLASS_ID(CallStaticJava, CallJava, 0)
692 DEFINE_CLASS_ID(CallDynamicJava, CallJava, 1)
693 DEFINE_CLASS_ID(CallRuntime, Call, 1)
694 DEFINE_CLASS_ID(CallLeaf, CallRuntime, 0)
695 DEFINE_CLASS_ID(CallLeafNoFP, CallLeaf, 0)
696 DEFINE_CLASS_ID(CallLeafPure, CallLeaf, 1)
697 DEFINE_CLASS_ID(Allocate, Call, 2)
698 DEFINE_CLASS_ID(AllocateArray, Allocate, 0)
699 DEFINE_CLASS_ID(AbstractLock, Call, 3)
700 DEFINE_CLASS_ID(Lock, AbstractLock, 0)
701 DEFINE_CLASS_ID(Unlock, AbstractLock, 1)
702 DEFINE_CLASS_ID(ArrayCopy, Call, 4)
703 DEFINE_CLASS_ID(LoadFlat, SafePoint, 1)
704 DEFINE_CLASS_ID(StoreFlat, SafePoint, 2)
705 DEFINE_CLASS_ID(MultiBranch, Multi, 1)
706 DEFINE_CLASS_ID(PCTable, MultiBranch, 0)
707 DEFINE_CLASS_ID(Catch, PCTable, 0)
708 DEFINE_CLASS_ID(Jump, PCTable, 1)
709 DEFINE_CLASS_ID(If, MultiBranch, 1)
710 DEFINE_CLASS_ID(BaseCountedLoopEnd, If, 0)
711 DEFINE_CLASS_ID(CountedLoopEnd, BaseCountedLoopEnd, 0)
712 DEFINE_CLASS_ID(LongCountedLoopEnd, BaseCountedLoopEnd, 1)
713 DEFINE_CLASS_ID(RangeCheck, If, 1)
714 DEFINE_CLASS_ID(OuterStripMinedLoopEnd, If, 2)
715 DEFINE_CLASS_ID(ParsePredicate, If, 3)
716 DEFINE_CLASS_ID(NeverBranch, MultiBranch, 2)
717 DEFINE_CLASS_ID(Start, Multi, 2)
718 DEFINE_CLASS_ID(MemBar, Multi, 3)
719 DEFINE_CLASS_ID(Initialize, MemBar, 0)
720 DEFINE_CLASS_ID(MemBarStoreStore, MemBar, 1)
721 DEFINE_CLASS_ID(Blackhole, Multi, 4)
722
723 DEFINE_CLASS_ID(Mach, Node, 1)
724 DEFINE_CLASS_ID(MachReturn, Mach, 0)
725 DEFINE_CLASS_ID(MachSafePoint, MachReturn, 0)
726 DEFINE_CLASS_ID(MachCall, MachSafePoint, 0)
727 DEFINE_CLASS_ID(MachCallJava, MachCall, 0)
728 DEFINE_CLASS_ID(MachCallStaticJava, MachCallJava, 0)
729 DEFINE_CLASS_ID(MachCallDynamicJava, MachCallJava, 1)
730 DEFINE_CLASS_ID(MachCallRuntime, MachCall, 1)
731 DEFINE_CLASS_ID(MachCallLeaf, MachCallRuntime, 0)
732 DEFINE_CLASS_ID(MachBranch, Mach, 1)
733 DEFINE_CLASS_ID(MachIf, MachBranch, 0)
734 DEFINE_CLASS_ID(MachGoto, MachBranch, 1)
735 DEFINE_CLASS_ID(MachNullCheck, MachBranch, 2)
736 DEFINE_CLASS_ID(MachSpillCopy, Mach, 2)
737 DEFINE_CLASS_ID(MachTemp, Mach, 3)
738 DEFINE_CLASS_ID(MachConstantBase, Mach, 4)
739 DEFINE_CLASS_ID(MachConstant, Mach, 5)
740 DEFINE_CLASS_ID(MachJump, MachConstant, 0)
741 DEFINE_CLASS_ID(MachMerge, Mach, 6)
742 DEFINE_CLASS_ID(MachMemBar, Mach, 7)
743 DEFINE_CLASS_ID(MachProlog, Mach, 8)
744 DEFINE_CLASS_ID(MachVEP, Mach, 9)
745
746 DEFINE_CLASS_ID(Type, Node, 2)
747 DEFINE_CLASS_ID(Phi, Type, 0)
748 DEFINE_CLASS_ID(ConstraintCast, Type, 1)
749 DEFINE_CLASS_ID(CastII, ConstraintCast, 0)
750 DEFINE_CLASS_ID(CheckCastPP, ConstraintCast, 1)
751 DEFINE_CLASS_ID(CastLL, ConstraintCast, 2)
752 DEFINE_CLASS_ID(CastFF, ConstraintCast, 3)
753 DEFINE_CLASS_ID(CastDD, ConstraintCast, 4)
754 DEFINE_CLASS_ID(CastVV, ConstraintCast, 5)
755 DEFINE_CLASS_ID(CastPP, ConstraintCast, 6)
756 DEFINE_CLASS_ID(CastHH, ConstraintCast, 7)
757 DEFINE_CLASS_ID(CMove, Type, 3)
758 DEFINE_CLASS_ID(SafePointScalarObject, Type, 4)
759 DEFINE_CLASS_ID(DecodeNarrowPtr, Type, 5)
760 DEFINE_CLASS_ID(DecodeN, DecodeNarrowPtr, 0)
761 DEFINE_CLASS_ID(DecodeNKlass, DecodeNarrowPtr, 1)
762 DEFINE_CLASS_ID(EncodeNarrowPtr, Type, 6)
763 DEFINE_CLASS_ID(EncodeP, EncodeNarrowPtr, 0)
764 DEFINE_CLASS_ID(EncodePKlass, EncodeNarrowPtr, 1)
765 DEFINE_CLASS_ID(Vector, Type, 7)
766 DEFINE_CLASS_ID(VectorMaskCmp, Vector, 0)
767 DEFINE_CLASS_ID(VectorUnbox, Vector, 1)
768 DEFINE_CLASS_ID(VectorReinterpret, Vector, 2)
769 DEFINE_CLASS_ID(ShiftV, Vector, 3)
770 DEFINE_CLASS_ID(CompressV, Vector, 4)
771 DEFINE_CLASS_ID(ExpandV, Vector, 5)
772 DEFINE_CLASS_ID(CompressM, Vector, 6)
773 DEFINE_CLASS_ID(Reduction, Vector, 7)
774 DEFINE_CLASS_ID(NegV, Vector, 8)
775 DEFINE_CLASS_ID(SaturatingVector, Vector, 9)
776 DEFINE_CLASS_ID(MulVL, Vector, 10)
777 DEFINE_CLASS_ID(InlineType, Type, 8)
778 DEFINE_CLASS_ID(Con, Type, 9)
779 DEFINE_CLASS_ID(ConI, Con, 0)
780 DEFINE_CLASS_ID(SafePointScalarMerge, Type, 10)
781 DEFINE_CLASS_ID(Convert, Type, 11)
782
783
784 DEFINE_CLASS_ID(Proj, Node, 3)
785 DEFINE_CLASS_ID(CatchProj, Proj, 0)
786 DEFINE_CLASS_ID(JumpProj, Proj, 1)
787 DEFINE_CLASS_ID(IfProj, Proj, 2)
788 DEFINE_CLASS_ID(IfTrue, IfProj, 0)
789 DEFINE_CLASS_ID(IfFalse, IfProj, 1)
790 DEFINE_CLASS_ID(Parm, Proj, 4)
791 DEFINE_CLASS_ID(MachProj, Proj, 5)
792 DEFINE_CLASS_ID(NarrowMemProj, Proj, 6)
793
794 DEFINE_CLASS_ID(Mem, Node, 4)
795 DEFINE_CLASS_ID(Load, Mem, 0)
796 DEFINE_CLASS_ID(LoadVector, Load, 0)
797 DEFINE_CLASS_ID(LoadVectorGather, LoadVector, 0)
798 DEFINE_CLASS_ID(LoadVectorGatherMasked, LoadVector, 1)
799 DEFINE_CLASS_ID(LoadVectorMasked, LoadVector, 2)
800 DEFINE_CLASS_ID(Store, Mem, 1)
801 DEFINE_CLASS_ID(StoreVector, Store, 0)
802 DEFINE_CLASS_ID(StoreVectorScatter, StoreVector, 0)
803 DEFINE_CLASS_ID(StoreVectorScatterMasked, StoreVector, 1)
804 DEFINE_CLASS_ID(StoreVectorMasked, StoreVector, 2)
805 DEFINE_CLASS_ID(LoadStore, Mem, 2)
806 DEFINE_CLASS_ID(LoadStoreConditional, LoadStore, 0)
807 DEFINE_CLASS_ID(CompareAndSwap, LoadStoreConditional, 0)
808 DEFINE_CLASS_ID(CompareAndExchangeNode, LoadStore, 1)
809
810 DEFINE_CLASS_ID(Region, Node, 5)
811 DEFINE_CLASS_ID(Loop, Region, 0)
812 DEFINE_CLASS_ID(Root, Loop, 0)
813 DEFINE_CLASS_ID(BaseCountedLoop, Loop, 1)
814 DEFINE_CLASS_ID(CountedLoop, BaseCountedLoop, 0)
815 DEFINE_CLASS_ID(LongCountedLoop, BaseCountedLoop, 1)
816 DEFINE_CLASS_ID(OuterStripMinedLoop, Loop, 2)
817
818 DEFINE_CLASS_ID(Sub, Node, 6)
819 DEFINE_CLASS_ID(Cmp, Sub, 0)
820 DEFINE_CLASS_ID(FastLock, Cmp, 0)
821 DEFINE_CLASS_ID(FastUnlock, Cmp, 1)
822 DEFINE_CLASS_ID(SubTypeCheck, Cmp, 2)
823 DEFINE_CLASS_ID(FlatArrayCheck, Cmp, 3)
824
825 DEFINE_CLASS_ID(MergeMem, Node, 7)
826 DEFINE_CLASS_ID(Bool, Node, 8)
827 DEFINE_CLASS_ID(AddP, Node, 9)
828 DEFINE_CLASS_ID(BoxLock, Node, 10)
829 DEFINE_CLASS_ID(Add, Node, 11)
830 DEFINE_CLASS_ID(MinMax, Add, 0)
831 DEFINE_CLASS_ID(Mul, Node, 12)
832 DEFINE_CLASS_ID(ClearArray, Node, 14)
833 DEFINE_CLASS_ID(Halt, Node, 15)
834 DEFINE_CLASS_ID(Opaque1, Node, 16)
835 DEFINE_CLASS_ID(OpaqueLoopInit, Opaque1, 0)
836 DEFINE_CLASS_ID(OpaqueLoopStride, Opaque1, 1)
837 DEFINE_CLASS_ID(OpaqueMultiversioning, Opaque1, 2)
838 DEFINE_CLASS_ID(OpaqueConstantBool, Node, 17)
839 DEFINE_CLASS_ID(OpaqueInitializedAssertionPredicate, Node, 18)
840 DEFINE_CLASS_ID(OpaqueTemplateAssertionPredicate, Node, 19)
841 DEFINE_CLASS_ID(Move, Node, 20)
842 DEFINE_CLASS_ID(LShift, Node, 21)
843 DEFINE_CLASS_ID(Neg, Node, 22)
844 DEFINE_CLASS_ID(ReachabilityFence, Node, 23)
845
846 _max_classes = ClassMask_Neg
847 };
848 #undef DEFINE_CLASS_ID
849
850 // Flags are sorted by usage frequency.
851 enum NodeFlags : uint64_t {
852 Flag_is_Copy = 1ULL << 0, // should be first bit to avoid shift
853 Flag_rematerialize = 1ULL << 1,
854 Flag_needs_anti_dependence_check = 1ULL << 2,
855 Flag_is_macro = 1ULL << 3,
856 Flag_is_Con = 1ULL << 4,
857 Flag_is_cisc_alternate = 1ULL << 5,
858 Flag_is_dead_loop_safe = 1ULL << 6,
859 Flag_may_be_short_branch = 1ULL << 7,
860 Flag_avoid_back_to_back_before = 1ULL << 8,
861 Flag_avoid_back_to_back_after = 1ULL << 9,
862 Flag_has_call = 1ULL << 10,
863 Flag_has_swapped_edges = 1ULL << 11,
864 Flag_is_scheduled = 1ULL << 12,
865 Flag_is_expensive = 1ULL << 13,
866 Flag_is_predicated_vector = 1ULL << 14, // Marked on a vector node that has an additional
867 // mask input controlling the lane operations.
868 Flag_for_post_loop_opts_igvn = 1ULL << 15,
869 Flag_for_merge_stores_igvn = 1ULL << 16,
870 Flag_is_removed_by_peephole = 1ULL << 17,
871 Flag_is_predicated_using_blend = 1ULL << 18,
872 _last_flag = Flag_is_predicated_using_blend
873 };
874
875 class PD;
876
877 private:
878 juint _class_id;
879 juint _flags;
880
881 #ifdef ASSERT
882 static juint max_flags();
883 #endif
884
885 protected:
886 // These methods should be called from constructors only.
887 void init_class_id(juint c) {
888 _class_id = c; // cast out const
889 }
890 void init_flags(uint fl) {
891 assert(fl <= max_flags(), "invalid node flag");
892 _flags |= fl;
893 }
894 void clear_flag(uint fl) {
895 assert(fl <= max_flags(), "invalid node flag");
896 _flags &= ~fl;
897 }
898
899 public:
900 juint class_id() const { return _class_id; }
901
902 juint flags() const { return _flags; }
903
904 void add_flag(juint fl) { init_flags(fl); }
905
906 void remove_flag(juint fl) { clear_flag(fl); }
907
908 // Return a dense integer opcode number
909 virtual int Opcode() const;
910
911 // Virtual inherited Node size
912 virtual uint size_of() const;
913
914 // Other interesting Node properties
915 #define DEFINE_CLASS_QUERY(type) \
916 bool is_##type() const { \
917 return ((_class_id & ClassMask_##type) == Class_##type); \
918 } \
919 type##Node *as_##type() const { \
920 assert(is_##type(), "invalid node class: %s", Name()); \
921 return (type##Node*)this; \
922 } \
923 type##Node* isa_##type() const { \
924 return (is_##type()) ? as_##type() : nullptr; \
925 }
926
927 DEFINE_CLASS_QUERY(AbstractLock)
928 DEFINE_CLASS_QUERY(Add)
929 DEFINE_CLASS_QUERY(AddP)
930 DEFINE_CLASS_QUERY(Allocate)
931 DEFINE_CLASS_QUERY(AllocateArray)
932 DEFINE_CLASS_QUERY(ArrayCopy)
933 DEFINE_CLASS_QUERY(BaseCountedLoop)
934 DEFINE_CLASS_QUERY(BaseCountedLoopEnd)
935 DEFINE_CLASS_QUERY(Blackhole)
936 DEFINE_CLASS_QUERY(Bool)
937 DEFINE_CLASS_QUERY(BoxLock)
938 DEFINE_CLASS_QUERY(Call)
939 DEFINE_CLASS_QUERY(CallDynamicJava)
940 DEFINE_CLASS_QUERY(CallJava)
941 DEFINE_CLASS_QUERY(CallLeaf)
942 DEFINE_CLASS_QUERY(CallLeafNoFP)
943 DEFINE_CLASS_QUERY(CallLeafPure)
944 DEFINE_CLASS_QUERY(CallRuntime)
945 DEFINE_CLASS_QUERY(CallStaticJava)
946 DEFINE_CLASS_QUERY(Catch)
947 DEFINE_CLASS_QUERY(CatchProj)
948 DEFINE_CLASS_QUERY(CheckCastPP)
949 DEFINE_CLASS_QUERY(CastII)
950 DEFINE_CLASS_QUERY(CastLL)
951 DEFINE_CLASS_QUERY(CastFF)
952 DEFINE_CLASS_QUERY(ConI)
953 DEFINE_CLASS_QUERY(CastPP)
954 DEFINE_CLASS_QUERY(ConstraintCast)
955 DEFINE_CLASS_QUERY(ClearArray)
956 DEFINE_CLASS_QUERY(CMove)
957 DEFINE_CLASS_QUERY(Cmp)
958 DEFINE_CLASS_QUERY(Convert)
959 DEFINE_CLASS_QUERY(CountedLoop)
960 DEFINE_CLASS_QUERY(CountedLoopEnd)
961 DEFINE_CLASS_QUERY(DecodeNarrowPtr)
962 DEFINE_CLASS_QUERY(DecodeN)
963 DEFINE_CLASS_QUERY(DecodeNKlass)
964 DEFINE_CLASS_QUERY(EncodeNarrowPtr)
965 DEFINE_CLASS_QUERY(EncodeP)
966 DEFINE_CLASS_QUERY(EncodePKlass)
967 DEFINE_CLASS_QUERY(FastLock)
968 DEFINE_CLASS_QUERY(FastUnlock)
969 DEFINE_CLASS_QUERY(FlatArrayCheck)
970 DEFINE_CLASS_QUERY(Halt)
971 DEFINE_CLASS_QUERY(If)
972 DEFINE_CLASS_QUERY(RangeCheck)
973 DEFINE_CLASS_QUERY(IfProj)
974 DEFINE_CLASS_QUERY(IfFalse)
975 DEFINE_CLASS_QUERY(IfTrue)
976 DEFINE_CLASS_QUERY(Initialize)
977 DEFINE_CLASS_QUERY(Jump)
978 DEFINE_CLASS_QUERY(JumpProj)
979 DEFINE_CLASS_QUERY(LongCountedLoop)
980 DEFINE_CLASS_QUERY(LongCountedLoopEnd)
981 DEFINE_CLASS_QUERY(Load)
982 DEFINE_CLASS_QUERY(LoadStore)
983 DEFINE_CLASS_QUERY(LoadStoreConditional)
984 DEFINE_CLASS_QUERY(Lock)
985 DEFINE_CLASS_QUERY(Loop)
986 DEFINE_CLASS_QUERY(LShift)
987 DEFINE_CLASS_QUERY(Mach)
988 DEFINE_CLASS_QUERY(MachBranch)
989 DEFINE_CLASS_QUERY(MachCall)
990 DEFINE_CLASS_QUERY(MachCallDynamicJava)
991 DEFINE_CLASS_QUERY(MachCallJava)
992 DEFINE_CLASS_QUERY(MachCallLeaf)
993 DEFINE_CLASS_QUERY(MachCallRuntime)
994 DEFINE_CLASS_QUERY(MachCallStaticJava)
995 DEFINE_CLASS_QUERY(MachConstantBase)
996 DEFINE_CLASS_QUERY(MachConstant)
997 DEFINE_CLASS_QUERY(MachGoto)
998 DEFINE_CLASS_QUERY(MachIf)
999 DEFINE_CLASS_QUERY(MachJump)
1000 DEFINE_CLASS_QUERY(MachNullCheck)
1001 DEFINE_CLASS_QUERY(MachProj)
1002 DEFINE_CLASS_QUERY(MachProlog)
1003 DEFINE_CLASS_QUERY(MachReturn)
1004 DEFINE_CLASS_QUERY(MachSafePoint)
1005 DEFINE_CLASS_QUERY(MachSpillCopy)
1006 DEFINE_CLASS_QUERY(MachTemp)
1007 DEFINE_CLASS_QUERY(MachMemBar)
1008 DEFINE_CLASS_QUERY(MachMerge)
1009 DEFINE_CLASS_QUERY(MachVEP)
1010 DEFINE_CLASS_QUERY(Mem)
1011 DEFINE_CLASS_QUERY(MemBar)
1012 DEFINE_CLASS_QUERY(MemBarStoreStore)
1013 DEFINE_CLASS_QUERY(MergeMem)
1014 DEFINE_CLASS_QUERY(MinMax)
1015 DEFINE_CLASS_QUERY(Move)
1016 DEFINE_CLASS_QUERY(Mul)
1017 DEFINE_CLASS_QUERY(Multi)
1018 DEFINE_CLASS_QUERY(MultiBranch)
1019 DEFINE_CLASS_QUERY(MulVL)
1020 DEFINE_CLASS_QUERY(NarrowMemProj)
1021 DEFINE_CLASS_QUERY(Neg)
1022 DEFINE_CLASS_QUERY(NegV)
1023 DEFINE_CLASS_QUERY(NeverBranch)
1024 DEFINE_CLASS_QUERY(Opaque1)
1025 DEFINE_CLASS_QUERY(OpaqueConstantBool)
1026 DEFINE_CLASS_QUERY(OpaqueInitializedAssertionPredicate)
1027 DEFINE_CLASS_QUERY(OpaqueTemplateAssertionPredicate)
1028 DEFINE_CLASS_QUERY(OpaqueLoopInit)
1029 DEFINE_CLASS_QUERY(OpaqueLoopStride)
1030 DEFINE_CLASS_QUERY(OpaqueMultiversioning)
1031 DEFINE_CLASS_QUERY(OuterStripMinedLoop)
1032 DEFINE_CLASS_QUERY(OuterStripMinedLoopEnd)
1033 DEFINE_CLASS_QUERY(Parm)
1034 DEFINE_CLASS_QUERY(ParsePredicate)
1035 DEFINE_CLASS_QUERY(PCTable)
1036 DEFINE_CLASS_QUERY(Phi)
1037 DEFINE_CLASS_QUERY(Proj)
1038 DEFINE_CLASS_QUERY(ReachabilityFence)
1039 DEFINE_CLASS_QUERY(Reduction)
1040 DEFINE_CLASS_QUERY(Region)
1041 DEFINE_CLASS_QUERY(Root)
1042 DEFINE_CLASS_QUERY(SafePoint)
1043 DEFINE_CLASS_QUERY(SafePointScalarObject)
1044 DEFINE_CLASS_QUERY(SafePointScalarMerge)
1045 DEFINE_CLASS_QUERY(Start)
1046 DEFINE_CLASS_QUERY(Store)
1047 DEFINE_CLASS_QUERY(Sub)
1048 DEFINE_CLASS_QUERY(SubTypeCheck)
1049 DEFINE_CLASS_QUERY(Type)
1050 DEFINE_CLASS_QUERY(InlineType)
1051 DEFINE_CLASS_QUERY(LoadFlat)
1052 DEFINE_CLASS_QUERY(StoreFlat)
1053 DEFINE_CLASS_QUERY(Vector)
1054 DEFINE_CLASS_QUERY(VectorMaskCmp)
1055 DEFINE_CLASS_QUERY(VectorUnbox)
1056 DEFINE_CLASS_QUERY(VectorReinterpret)
1057 DEFINE_CLASS_QUERY(CompressV)
1058 DEFINE_CLASS_QUERY(ExpandV)
1059 DEFINE_CLASS_QUERY(CompressM)
1060 DEFINE_CLASS_QUERY(LoadVector)
1061 DEFINE_CLASS_QUERY(LoadVectorGather)
1062 DEFINE_CLASS_QUERY(LoadVectorMasked)
1063 DEFINE_CLASS_QUERY(LoadVectorGatherMasked)
1064 DEFINE_CLASS_QUERY(StoreVector)
1065 DEFINE_CLASS_QUERY(StoreVectorScatter)
1066 DEFINE_CLASS_QUERY(StoreVectorMasked)
1067 DEFINE_CLASS_QUERY(StoreVectorScatterMasked)
1068 DEFINE_CLASS_QUERY(SaturatingVector)
1069 DEFINE_CLASS_QUERY(ShiftV)
1070 DEFINE_CLASS_QUERY(Unlock)
1071
1072 #undef DEFINE_CLASS_QUERY
1073
1074 // duplicate of is_MachSpillCopy()
1075 bool is_SpillCopy () const {
1076 return ((_class_id & ClassMask_MachSpillCopy) == Class_MachSpillCopy);
1077 }
1078
1079 bool is_Con () const { return (_flags & Flag_is_Con) != 0; }
1080 // The data node which is safe to leave in dead loop during IGVN optimization.
1081 bool is_dead_loop_safe() const;
1082
1083 // is_Copy() returns copied edge index (0 or 1)
1084 uint is_Copy() const { return (_flags & Flag_is_Copy); }
1085
1086 virtual bool is_CFG() const { return false; }
1087
1088 // If this node is control-dependent on a test, can it be rerouted to a dominating equivalent
1089 // test? This means that the node can be executed safely as long as it happens after the test
1090 // that is its control input without worrying about the whole control flow. On the contrary, if
1091 // the node depends on a test that is not its control input, or if it depends on more than one
1092 // tests, then this method must return false.
1093 //
1094 // Pseudocode examples:
1095 // 1. if (y != 0) {
1096 // x / y;
1097 // }
1098 // The division depends only on the test y != 0 and can be executed anywhere y != 0 holds true.
1099 // As a result, depends_only_on_test returns true.
1100 // 2. if (y != 0) {
1101 // if (x > 1) {
1102 // x / y;
1103 // }
1104 // }
1105 // If the division x / y has its control input being the IfTrueNode of the test y != 0, then
1106 // depends_only_on_test returns true. Otherwise, if the division has its control input being the
1107 // IfTrueNode of the test x > 1, then depends_only_on_test returns false.
1108 // 3. if (y > z) {
1109 // if (z > 0) {
1110 // x / y
1111 // }
1112 // }
1113 // The division depends on both tests y > z and z > 0. As a result, depends_only_on_test returns
1114 // false.
1115 //
1116 // This method allows more freedom in certain nodes with regards to scheduling, for example it
1117 // allows nodes to float out of loops together with its test.
1118 //
1119 // This method is pessimistic, this means that it may return false even if the node satisfy the
1120 // requirements. However, it must return false if the node does not satisfy the requirements.
1121 // When a test is decomposed into multiple tests, all nodes that depend on the decomposed test
1122 // must be pinned at the lowest dominating test of those. For example, when a zero check of a
1123 // division is split through a region but the division itself is not, it must be pinned at the
1124 // merge point by returning false when calling this method.
1125 bool depends_only_on_test() const {
1126 if (is_CFG() || pinned()) {
1127 return false;
1128 }
1129 assert(in(0) != nullptr, "must have a control input");
1130 return depends_only_on_test_impl();
1131 }
1132
1133 // Return a clone of the current node that's pinned. The current node must return true for
1134 // depends_only_on_test, and the retuned node must return false. This method is called when the
1135 // node is disconnected from its test.
1136 //
1137 // Examples:
1138 // 1. for (int i = start; i <= limit; i++) {
1139 // if (!rangecheck(i, a)) {
1140 // trap;
1141 // }
1142 // a[i];
1143 // }
1144 // Loop predication can then hoist the range check out of the loop:
1145 // if (!rangecheck(start, a)) {
1146 // trap;
1147 // }
1148 // if (!rangecheck(limit, a)) {
1149 // trap;
1150 // }
1151 // for (int i = start; i <= limit; i++) {
1152 // a[i];
1153 // }
1154 // As the load a[i] now depends on both tests rangecheck(start, a) and rangecheck(limit, a), it
1155 // must be pinned at the lowest dominating test of those.
1156 //
1157 // 2. if (y > x) {
1158 // if (x >= 0) {
1159 // if (y != 0) {
1160 // x / y;
1161 // }
1162 // }
1163 // }
1164 // The test (y != 0) == true can be deduced from (y > x) == true and (x >= 0) == true, so we may
1165 // choose to elide it. In such cases, the division x / y now depends on both tests
1166 // (y > x) == true and (x >= 0) == true, so it must be pinned at the lowest dominating test of
1167 // those.
1168 //
1169 // 3. if (b) {
1170 // ...
1171 // } else {
1172 // ...
1173 // }
1174 // if (y == 0) {
1175 // trap;
1176 // }
1177 // x / y;
1178 // The division x / y depends only on the test (y == 0) == false, but if we split the test
1179 // through the merge point but not the division:
1180 // if (b) {
1181 // ...
1182 // if (y == 0) {
1183 // trap;
1184 // }
1185 // } else {
1186 // ...
1187 // if (y == 0) {
1188 // trap;
1189 // }
1190 // }
1191 // x / y;
1192 // The division now has the control input being the RegionNode merge the branches of if(b)
1193 // instead of a test that proves y != 0. As a result, it must be pinned at that node.
1194 //
1195 // There are cases where the node does not actually have a dependency on its control input. For
1196 // example, when we try to sink a LoadNode out of a loop in PhaseIdealLoop::try_sink_out_of_loop,
1197 // we clone the node so that all of the clones can be scheduled out of the loop. To prevent the
1198 // clones from being GVN-ed again, we add a control input for the node at the loop exit. For the
1199 // cases when the node does provably not depend on its control input, this method can return
1200 // nullptr.
1201 Node* pin_node_under_control() const {
1202 assert(depends_only_on_test(), "must be a depends_only_on_test node");
1203 Node* res = pin_node_under_control_impl();
1204 if (res == nullptr) {
1205 assert(is_Load(), "unexpected failure to pin for %s", Name());
1206 return nullptr;
1207 }
1208 assert(!res->depends_only_on_test(), "the result must not depends_only_on_test");
1209 assert(Opcode() == res->Opcode(), "pinning must result in the same kind of node %s - %s", Name(), res->Name());
1210 return res;
1211 }
1212
1213 private:
1214 virtual bool depends_only_on_test_impl() const { assert(false, "%s", Name()); return false; }
1215 virtual Node* pin_node_under_control_impl() const { assert(false, "%s", Name()); return nullptr; }
1216
1217 public:
1218 // When building basic blocks, I need to have a notion of block beginning
1219 // Nodes, next block selector Nodes (block enders), and next block
1220 // projections. These calls need to work on their machine equivalents. The
1221 // Ideal beginning Nodes are RootNode, RegionNode and StartNode.
1222 bool is_block_start() const {
1223 if ( is_Region() )
1224 return this == (const Node*)in(0);
1225 else
1226 return is_Start();
1227 }
1228
1229 // The Ideal control projection Nodes are IfTrue/IfFalse, JumpProjNode, Root,
1230 // Goto and Return. This call also returns the block ending Node.
1231 virtual const Node *is_block_proj() const;
1232
1233 // The node is a "macro" node which needs to be expanded before matching
1234 bool is_macro() const { return (_flags & Flag_is_macro) != 0; }
1235 // The node is expensive: the best control is set during loop opts
1236 bool is_expensive() const { return (_flags & Flag_is_expensive) != 0 && in(0) != nullptr; }
1237 // The node's original edge position is swapped.
1238 bool has_swapped_edges() const { return (_flags & Flag_has_swapped_edges) != 0; }
1239
1240 bool is_predicated_vector() const { return (_flags & Flag_is_predicated_vector) != 0; }
1241
1242 bool is_predicated_using_blend() const { return (_flags & Flag_is_predicated_using_blend) != 0; }
1243
1244 // Used in lcm to mark nodes that have scheduled
1245 bool is_scheduled() const { return (_flags & Flag_is_scheduled) != 0; }
1246
1247 bool for_post_loop_opts_igvn() const { return (_flags & Flag_for_post_loop_opts_igvn) != 0; }
1248 bool for_merge_stores_igvn() const { return (_flags & Flag_for_merge_stores_igvn) != 0; }
1249
1250 // Is 'n' possibly a loop entry (i.e. a Parse Predicate projection)?
1251 static bool may_be_loop_entry(Node* n) {
1252 return n != nullptr && n->is_IfProj() && n->in(0)->is_ParsePredicate();
1253 }
1254
1255 //----------------- Optimization
1256
1257 // Get the worst-case Type output for this Node.
1258 virtual const class Type *bottom_type() const;
1259
1260 // If we find a better type for a node, try to record it permanently.
1261 // Return true if this node actually changed.
1262 // Be sure to do the hash_delete game in the "rehash" variant.
1263 void raise_bottom_type(const Type* new_type);
1264
1265 // Get the address type with which this node uses and/or defs memory,
1266 // or null if none. The address type is conservatively wide.
1267 // Returns non-null for calls, membars, loads, stores, etc.
1268 // Returns TypePtr::BOTTOM if the node touches memory "broadly".
1269 virtual const class TypePtr *adr_type() const { return nullptr; }
1270
1271 // Return an existing node which computes the same function as this node.
1272 // The optimistic combined algorithm requires this to return a Node which
1273 // is a small number of steps away (e.g., one of my inputs).
1274 virtual Node* Identity(PhaseGVN* phase);
1275
1276 // Return the set of values this Node can take on at runtime.
1277 virtual const Type* Value(PhaseGVN* phase) const;
1278
1279 // Return a node which is more "ideal" than the current node.
1280 // The invariants on this call are subtle. If in doubt, read the
1281 // treatise in node.cpp above the default implementation AND TEST WITH
1282 // -XX:VerifyIterativeGVN=1
1283 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
1284
1285 // Some nodes have specific Ideal subgraph transformations only if they are
1286 // unique users of specific nodes. Such nodes should be put on IGVN worklist
1287 // for the transformations to happen.
1288 bool has_special_unique_user() const;
1289
1290 // Skip Proj and CatchProj nodes chains. Check for Null and Top.
1291 Node* find_exact_control(Node* ctrl);
1292
1293 // Results of the dominance analysis.
1294 enum class DomResult {
1295 NotDominate, // 'this' node does not dominate 'sub'.
1296 Dominate, // 'this' node dominates or is equal to 'sub'.
1297 EncounteredDeadCode // Result is undefined due to encountering dead code.
1298 };
1299 // Check if 'this' node dominates or equal to 'sub'.
1300 DomResult dominates(Node* sub, Node_List &nlist);
1301
1302 bool remove_dead_region(PhaseGVN *phase, bool can_reshape);
1303 public:
1304
1305 // See if there is valid pipeline info
1306 static const Pipeline *pipeline_class();
1307 virtual const Pipeline *pipeline() const;
1308
1309 // Compute the latency from the def to this instruction of the ith input node
1310 uint latency(uint i);
1311
1312 // Hash & compare functions, for pessimistic value numbering
1313
1314 // If the hash function returns the special sentinel value NO_HASH,
1315 // the node is guaranteed never to compare equal to any other node.
1316 // If we accidentally generate a hash with value NO_HASH the node
1317 // won't go into the table and we'll lose a little optimization.
1318 static const uint NO_HASH = 0;
1319 virtual uint hash() const;
1320 virtual bool cmp( const Node &n ) const;
1321
1322 // Operation appears to be iteratively computed (such as an induction variable)
1323 // It is possible for this operation to return false for a loop-varying
1324 // value, if it appears (by local graph inspection) to be computed by a simple conditional.
1325 bool is_iteratively_computed();
1326
1327 // Determine if a node is a counted loop induction variable.
1328 // NOTE: The method is defined in "loopnode.cpp".
1329 bool is_cloop_ind_var() const;
1330
1331 // Return a node with opcode "opc" and same inputs as "this" if one can
1332 // be found; Otherwise return null;
1333 Node* find_similar(int opc);
1334 bool has_same_inputs_as(const Node* other) const;
1335
1336 // Return the unique control out if only one. Null if none or more than one.
1337 Node* unique_ctrl_out_or_null() const;
1338 // Return the unique control out. Asserts if none or more than one control out.
1339 Node* unique_ctrl_out() const;
1340
1341 // Set control or add control as precedence edge
1342 void ensure_control_or_add_prec(Node* c);
1343 void add_prec_from(Node* n);
1344
1345 // Visit boundary uses of the node and apply a callback function for each.
1346 // Recursively traverse uses, stopping and applying the callback when
1347 // reaching a boundary node, defined by is_boundary. If callback_on_all is true,
1348 // it applies the callback on all the nodes seen, and not only on the boundary
1349 // nodes. Note: the function definition appears after the complete type
1350 // definition of Node_List.
1351 template <typename Callback, typename Check>
1352 void visit_uses(Callback callback, Check is_boundary, bool always_callback = false) const;
1353
1354 //----------------- Code Generation
1355
1356 // Ideal register class for Matching. Zero means unmatched instruction
1357 // (these are cloned instead of converted to machine nodes).
1358 virtual uint ideal_reg() const;
1359
1360 static const uint NotAMachineReg; // must be > max. machine register
1361
1362 // Do we Match on this edge index or not? Generally false for Control
1363 // and true for everything else. Weird for calls & returns.
1364 virtual uint match_edge(uint idx) const;
1365
1366 // Register class output is returned in
1367 virtual const RegMask &out_RegMask() const;
1368 // Register class input is expected in
1369 virtual const RegMask &in_RegMask(uint) const;
1370 // Should we clone rather than spill this instruction?
1371 bool rematerialize() const;
1372
1373 // Return JVM State Object if this Node carries debug info, or null otherwise
1374 virtual JVMState* jvms() const;
1375
1376 // Print as assembly
1377 virtual void format( PhaseRegAlloc *, outputStream* st = tty ) const;
1378 // Emit bytes using C2_MacroAssembler
1379 virtual void emit(C2_MacroAssembler *masm, PhaseRegAlloc *ra_) const;
1380 // Size of instruction in bytes
1381 virtual uint size(PhaseRegAlloc *ra_) const;
1382
1383 // Convenience function to extract an integer constant from a node.
1384 // If it is not an integer constant (either Con, CastII, or Mach),
1385 // return value_if_unknown.
1386 jint find_int_con(jint value_if_unknown) const {
1387 const TypeInt* t = find_int_type();
1388 return (t != nullptr && t->is_con()) ? t->get_con() : value_if_unknown;
1389 }
1390 // Return the constant, knowing it is an integer constant already
1391 jint get_int() const {
1392 const TypeInt* t = find_int_type();
1393 guarantee(t != nullptr, "must be con");
1394 return t->get_con();
1395 }
1396 // Here's where the work is done. Can produce non-constant int types too.
1397 const TypeInt* find_int_type() const;
1398 const TypeInteger* find_integer_type(BasicType bt) const;
1399
1400 // Same thing for long (and intptr_t, via type.hpp):
1401 jlong get_long() const {
1402 const TypeLong* t = find_long_type();
1403 guarantee(t != nullptr, "must be con");
1404 return t->get_con();
1405 }
1406 jlong find_long_con(jint value_if_unknown) const {
1407 const TypeLong* t = find_long_type();
1408 return (t != nullptr && t->is_con()) ? t->get_con() : value_if_unknown;
1409 }
1410 const TypeLong* find_long_type() const;
1411
1412 jlong get_integer_as_long(BasicType bt) const {
1413 const TypeInteger* t = find_integer_type(bt);
1414 guarantee(t != nullptr && t->is_con(), "must be con");
1415 return t->get_con_as_long(bt);
1416 }
1417 jlong find_integer_as_long(BasicType bt, jlong value_if_unknown) const {
1418 const TypeInteger* t = find_integer_type(bt);
1419 if (t == nullptr || !t->is_con()) return value_if_unknown;
1420 return t->get_con_as_long(bt);
1421 }
1422 const TypePtr* get_ptr_type() const;
1423
1424 // These guys are called by code generated by ADLC:
1425 intptr_t get_ptr() const;
1426 intptr_t get_narrowcon() const;
1427 jdouble getd() const;
1428 jfloat getf() const;
1429 jshort geth() const;
1430
1431 // Nodes which are pinned into basic blocks
1432 virtual bool pinned() const { return false; }
1433
1434 // Nodes which use memory without consuming it, hence need antidependences
1435 // More specifically, needs_anti_dependence_check returns true iff the node
1436 // (a) does a load, and (b) does not perform a store (except perhaps to a
1437 // stack slot or some other unaliased location).
1438 bool needs_anti_dependence_check() const;
1439
1440 // Return which operand this instruction may cisc-spill. In other words,
1441 // return operand position that can convert from reg to memory access
1442 virtual int cisc_operand() const { return AdlcVMDeps::Not_cisc_spillable; }
1443 bool is_cisc_alternate() const { return (_flags & Flag_is_cisc_alternate) != 0; }
1444
1445 // Whether this is a memory-writing machine node.
1446 bool is_memory_writer() const { return is_Mach() && bottom_type()->has_memory(); }
1447
1448 // Whether this is a memory phi node
1449 bool is_memory_phi() const { return is_Phi() && bottom_type() == Type::MEMORY; }
1450
1451 bool is_div_or_mod(BasicType bt) const;
1452
1453 bool is_data_proj_of_pure_function(const Node* maybe_pure_function) const;
1454
1455 //----------------- Printing, etc
1456 #ifndef PRODUCT
1457 public:
1458 Node* find(int idx, bool only_ctrl = false); // Search the graph for the given idx.
1459 Node* find_ctrl(int idx); // Search control ancestors for the given idx.
1460 void dump_bfs(const int max_distance, Node* target, const char* options, outputStream* st, const frame* fr = nullptr) const;
1461 void dump_bfs(const int max_distance, Node* target, const char* options) const; // directly to tty
1462 void dump_bfs(const int max_distance) const; // dump_bfs(max_distance, nullptr, nullptr)
1463 void dump_bfs(const int max_distance, Node* target, const char* options, void* sp, void* fp, void* pc) const;
1464 class DumpConfig {
1465 public:
1466 // overridden to implement coloring of node idx
1467 virtual void pre_dump(outputStream *st, const Node* n) = 0;
1468 virtual void post_dump(outputStream *st) = 0;
1469 };
1470 void dump_idx(bool align = false, outputStream* st = tty, DumpConfig* dc = nullptr) const;
1471 void dump_name(outputStream* st = tty, DumpConfig* dc = nullptr) const;
1472 void dump() const; // print node with newline
1473 void dump(const char* suffix, bool mark = false, outputStream* st = tty, DumpConfig* dc = nullptr) const; // Print this node.
1474 void dump(int depth) const; // Print this node, recursively to depth d
1475 void dump_ctrl(int depth) const; // Print control nodes, to depth d
1476 void dump_comp() const; // Print this node in compact representation.
1477 // Print this node in compact representation.
1478 void dump_comp(const char* suffix, outputStream *st = tty) const;
1479 private:
1480 virtual void dump_req(outputStream* st = tty, DumpConfig* dc = nullptr) const; // Print required-edge info
1481 virtual void dump_prec(outputStream* st = tty, DumpConfig* dc = nullptr) const; // Print precedence-edge info
1482 virtual void dump_out(outputStream* st = tty, DumpConfig* dc = nullptr) const; // Print the output edge info
1483 public:
1484 virtual void dump_spec(outputStream *st) const {}; // Print per-node info
1485 // Print compact per-node info
1486 virtual void dump_compact_spec(outputStream *st) const { dump_spec(st); }
1487
1488 static void verify(int verify_depth, VectorSet& visited, Node_List& worklist);
1489
1490 // This call defines a class-unique string used to identify class instances
1491 virtual const char *Name() const;
1492
1493 void dump_format(PhaseRegAlloc *ra) const; // debug access to MachNode::format(...)
1494 static bool in_dump() { return Compile::current()->_in_dump_cnt > 0; } // check if we are in a dump call
1495 #endif
1496 #ifdef ASSERT
1497 void verify_construction();
1498 bool verify_jvms(const JVMState* jvms) const;
1499
1500 Node* _debug_orig; // Original version of this, if any.
1501 Node* debug_orig() const { return _debug_orig; }
1502 void set_debug_orig(Node* orig); // _debug_orig = orig
1503 void dump_orig(outputStream *st, bool print_key = true) const;
1504
1505 uint64_t _debug_idx; // Unique value assigned to every node.
1506 uint64_t debug_idx() const { return _debug_idx; }
1507 void set_debug_idx(uint64_t debug_idx) { _debug_idx = debug_idx; }
1508
1509 int _hash_lock; // Barrier to modifications of nodes in the hash table
1510 void enter_hash_lock() { ++_hash_lock; assert(_hash_lock < 99, "in too many hash tables?"); }
1511 void exit_hash_lock() { --_hash_lock; assert(_hash_lock >= 0, "mispaired hash locks"); }
1512
1513 static void init_NodeProperty();
1514
1515 #if OPTO_DU_ITERATOR_ASSERT
1516 const Node* _last_del; // The last deleted node.
1517 uint _del_tick; // Bumped when a deletion happens..
1518 #endif
1519 #endif
1520 };
1521
1522 inline bool not_a_node(const Node* n) {
1523 if (n == nullptr) return true;
1524 if (((intptr_t)n & 1) != 0) return true; // uninitialized, etc.
1525 if (*(address*)n == badAddress) return true; // kill by Node::destruct
1526 return false;
1527 }
1528
1529 //-----------------------------------------------------------------------------
1530 // Iterators over DU info, and associated Node functions.
1531
1532 #if OPTO_DU_ITERATOR_ASSERT
1533
1534 // Common code for assertion checking on DU iterators.
1535 class DUIterator_Common {
1536 #ifdef ASSERT
1537 protected:
1538 bool _vdui; // cached value of VerifyDUIterators
1539 const Node* _node; // the node containing the _out array
1540 uint _outcnt; // cached node->_outcnt
1541 uint _del_tick; // cached node->_del_tick
1542 Node* _last; // last value produced by the iterator
1543
1544 void sample(const Node* node); // used by c'tor to set up for verifies
1545 void verify(const Node* node, bool at_end_ok = false);
1546 void verify_resync();
1547 void reset(const DUIterator_Common& that);
1548
1549 // The VDUI_ONLY macro protects code conditionalized on VerifyDUIterators
1550 #define I_VDUI_ONLY(i,x) { if ((i)._vdui) { x; } }
1551 #else
1552 #define I_VDUI_ONLY(i,x) { }
1553 #endif //ASSERT
1554 };
1555
1556 #define VDUI_ONLY(x) I_VDUI_ONLY(*this, x)
1557
1558 // Default DU iterator. Allows appends onto the out array.
1559 // Allows deletion from the out array only at the current point.
1560 // Usage:
1561 // for (DUIterator i = x->outs(); x->has_out(i); i++) {
1562 // Node* y = x->out(i);
1563 // ...
1564 // }
1565 // Compiles in product mode to a unsigned integer index, which indexes
1566 // onto a repeatedly reloaded base pointer of x->_out. The loop predicate
1567 // also reloads x->_outcnt. If you delete, you must perform "--i" just
1568 // before continuing the loop. You must delete only the last-produced
1569 // edge. You must delete only a single copy of the last-produced edge,
1570 // or else you must delete all copies at once (the first time the edge
1571 // is produced by the iterator).
1572 class DUIterator : public DUIterator_Common {
1573 friend class Node;
1574
1575 // This is the index which provides the product-mode behavior.
1576 // Whatever the product-mode version of the system does to the
1577 // DUI index is done to this index. All other fields in
1578 // this class are used only for assertion checking.
1579 uint _idx;
1580
1581 #ifdef ASSERT
1582 uint _refresh_tick; // Records the refresh activity.
1583
1584 void sample(const Node* node); // Initialize _refresh_tick etc.
1585 void verify(const Node* node, bool at_end_ok = false);
1586 void verify_increment(); // Verify an increment operation.
1587 void verify_resync(); // Verify that we can back up over a deletion.
1588 void verify_finish(); // Verify that the loop terminated properly.
1589 void refresh(); // Resample verification info.
1590 void reset(const DUIterator& that); // Resample after assignment.
1591 #endif
1592
1593 DUIterator(const Node* node, int dummy_to_avoid_conversion)
1594 { _idx = 0; DEBUG_ONLY(sample(node)); }
1595
1596 public:
1597 // initialize to garbage; clear _vdui to disable asserts
1598 DUIterator()
1599 { /*initialize to garbage*/ DEBUG_ONLY(_vdui = false); }
1600
1601 DUIterator(const DUIterator& that)
1602 { _idx = that._idx; DEBUG_ONLY(_vdui = false; reset(that)); }
1603
1604 void operator++(int dummy_to_specify_postfix_op)
1605 { _idx++; VDUI_ONLY(verify_increment()); }
1606
1607 void operator--()
1608 { VDUI_ONLY(verify_resync()); --_idx; }
1609
1610 ~DUIterator()
1611 { VDUI_ONLY(verify_finish()); }
1612
1613 void operator=(const DUIterator& that)
1614 { _idx = that._idx; DEBUG_ONLY(reset(that)); }
1615 };
1616
1617 DUIterator Node::outs() const
1618 { return DUIterator(this, 0); }
1619 DUIterator& Node::refresh_out_pos(DUIterator& i) const
1620 { I_VDUI_ONLY(i, i.refresh()); return i; }
1621 bool Node::has_out(DUIterator& i) const
1622 { I_VDUI_ONLY(i, i.verify(this,true));return i._idx < _outcnt; }
1623 Node* Node::out(DUIterator& i) const
1624 { I_VDUI_ONLY(i, i.verify(this)); return DEBUG_ONLY(i._last=) _out[i._idx]; }
1625
1626
1627 // Faster DU iterator. Disallows insertions into the out array.
1628 // Allows deletion from the out array only at the current point.
1629 // Usage:
1630 // for (DUIterator_Fast imax, i = x->fast_outs(imax); i < imax; i++) {
1631 // Node* y = x->fast_out(i);
1632 // ...
1633 // }
1634 // Compiles in product mode to raw Node** pointer arithmetic, with
1635 // no reloading of pointers from the original node x. If you delete,
1636 // you must perform "--i; --imax" just before continuing the loop.
1637 // If you delete multiple copies of the same edge, you must decrement
1638 // imax, but not i, multiple times: "--i, imax -= num_edges".
1639 class DUIterator_Fast : public DUIterator_Common {
1640 friend class Node;
1641 friend class DUIterator_Last;
1642
1643 // This is the pointer which provides the product-mode behavior.
1644 // Whatever the product-mode version of the system does to the
1645 // DUI pointer is done to this pointer. All other fields in
1646 // this class are used only for assertion checking.
1647 Node** _outp;
1648
1649 #ifdef ASSERT
1650 void verify(const Node* node, bool at_end_ok = false);
1651 void verify_limit();
1652 void verify_resync();
1653 void verify_relimit(uint n);
1654 void reset(const DUIterator_Fast& that);
1655 #endif
1656
1657 // Note: offset must be signed, since -1 is sometimes passed
1658 DUIterator_Fast(const Node* node, ptrdiff_t offset)
1659 { _outp = node->_out + offset; DEBUG_ONLY(sample(node)); }
1660
1661 public:
1662 // initialize to garbage; clear _vdui to disable asserts
1663 DUIterator_Fast()
1664 { /*initialize to garbage*/ DEBUG_ONLY(_vdui = false); }
1665
1666 DUIterator_Fast(const DUIterator_Fast& that)
1667 { _outp = that._outp; DEBUG_ONLY(_vdui = false; reset(that)); }
1668
1669 void operator++(int dummy_to_specify_postfix_op)
1670 { _outp++; VDUI_ONLY(verify(_node, true)); }
1671
1672 void operator--()
1673 { VDUI_ONLY(verify_resync()); --_outp; }
1674
1675 void operator-=(uint n) // applied to the limit only
1676 { _outp -= n; VDUI_ONLY(verify_relimit(n)); }
1677
1678 bool operator<(DUIterator_Fast& limit) {
1679 I_VDUI_ONLY(*this, this->verify(_node, true));
1680 I_VDUI_ONLY(limit, limit.verify_limit());
1681 return _outp < limit._outp;
1682 }
1683
1684 void operator=(const DUIterator_Fast& that)
1685 { _outp = that._outp; DEBUG_ONLY(reset(that)); }
1686 };
1687
1688 DUIterator_Fast Node::fast_outs(DUIterator_Fast& imax) const {
1689 // Assign a limit pointer to the reference argument:
1690 imax = DUIterator_Fast(this, (ptrdiff_t)_outcnt);
1691 // Return the base pointer:
1692 return DUIterator_Fast(this, 0);
1693 }
1694 Node* Node::fast_out(DUIterator_Fast& i) const {
1695 I_VDUI_ONLY(i, i.verify(this));
1696 return DEBUG_ONLY(i._last=) *i._outp;
1697 }
1698
1699
1700 // Faster DU iterator. Requires each successive edge to be removed.
1701 // Does not allow insertion of any edges.
1702 // Usage:
1703 // for (DUIterator_Last imin, i = x->last_outs(imin); i >= imin; i -= num_edges) {
1704 // Node* y = x->last_out(i);
1705 // ...
1706 // }
1707 // Compiles in product mode to raw Node** pointer arithmetic, with
1708 // no reloading of pointers from the original node x.
1709 class DUIterator_Last : private DUIterator_Fast {
1710 friend class Node;
1711
1712 #ifdef ASSERT
1713 void verify(const Node* node, bool at_end_ok = false);
1714 void verify_limit();
1715 void verify_step(uint num_edges);
1716 #endif
1717
1718 // Note: offset must be signed, since -1 is sometimes passed
1719 DUIterator_Last(const Node* node, ptrdiff_t offset)
1720 : DUIterator_Fast(node, offset) { }
1721
1722 void operator++(int dummy_to_specify_postfix_op) {} // do not use
1723 void operator<(int) {} // do not use
1724
1725 public:
1726 DUIterator_Last() { }
1727 // initialize to garbage
1728
1729 DUIterator_Last(const DUIterator_Last& that) = default;
1730
1731 void operator--()
1732 { _outp--; VDUI_ONLY(verify_step(1)); }
1733
1734 void operator-=(uint n)
1735 { _outp -= n; VDUI_ONLY(verify_step(n)); }
1736
1737 bool operator>=(DUIterator_Last& limit) {
1738 I_VDUI_ONLY(*this, this->verify(_node, true));
1739 I_VDUI_ONLY(limit, limit.verify_limit());
1740 return _outp >= limit._outp;
1741 }
1742
1743 DUIterator_Last& operator=(const DUIterator_Last& that) = default;
1744 };
1745
1746 DUIterator_Last Node::last_outs(DUIterator_Last& imin) const {
1747 // Assign a limit pointer to the reference argument:
1748 imin = DUIterator_Last(this, 0);
1749 // Return the initial pointer:
1750 return DUIterator_Last(this, (ptrdiff_t)_outcnt - 1);
1751 }
1752 Node* Node::last_out(DUIterator_Last& i) const {
1753 I_VDUI_ONLY(i, i.verify(this));
1754 return DEBUG_ONLY(i._last=) *i._outp;
1755 }
1756
1757 #endif //OPTO_DU_ITERATOR_ASSERT
1758
1759 #undef I_VDUI_ONLY
1760 #undef VDUI_ONLY
1761
1762 // An Iterator that truly follows the iterator pattern. Doesn't
1763 // support deletion but could be made to.
1764 //
1765 // for (SimpleDUIterator i(n); i.has_next(); i.next()) {
1766 // Node* m = i.get();
1767 //
1768 class SimpleDUIterator : public StackObj {
1769 private:
1770 Node* node;
1771 DUIterator_Fast imax;
1772 DUIterator_Fast i;
1773 public:
1774 SimpleDUIterator(Node* n): node(n), i(n->fast_outs(imax)) {}
1775 bool has_next() { return i < imax; }
1776 void next() { i++; }
1777 Node* get() { return node->fast_out(i); }
1778 };
1779
1780
1781 //-----------------------------------------------------------------------------
1782 // Map dense integer indices to Nodes. Uses classic doubling-array trick.
1783 // Abstractly provides an infinite array of Node*'s, initialized to null.
1784 // Note that the constructor just zeros things, and since I use Arena
1785 // allocation I do not need a destructor to reclaim storage.
1786 class Node_Array : public AnyObj {
1787 protected:
1788 Arena* _a; // Arena to allocate in
1789 uint _max;
1790 Node** _nodes;
1791 ReallocMark _nesting; // Safety checks for arena reallocation
1792
1793 // Grow array to required capacity
1794 void maybe_grow(uint i) {
1795 _nesting.check(_a); // Check if a potential reallocation in the arena is safe
1796 if (i >= _max) {
1797 grow(i);
1798 }
1799 }
1800 void grow(uint i);
1801
1802 public:
1803 Node_Array(Arena* a, uint max = OptoNodeListSize) : _a(a), _max(max) {
1804 _nodes = NEW_ARENA_ARRAY(a, Node*, max);
1805 clear();
1806 }
1807 Node_Array() : Node_Array(Thread::current()->resource_area()) {}
1808
1809 NONCOPYABLE(Node_Array);
1810 Node_Array& operator=(Node_Array&&) = delete;
1811 // Allow move constructor for && (eg. capture return of function)
1812 Node_Array(Node_Array&&) = default;
1813
1814 Node *operator[] ( uint i ) const // Lookup, or null for not mapped
1815 { return (i<_max) ? _nodes[i] : (Node*)nullptr; }
1816 Node* at(uint i) const { assert(i<_max,"oob"); return _nodes[i]; }
1817 Node** adr() { return _nodes; }
1818 // Extend the mapping: index i maps to Node *n.
1819 void map( uint i, Node *n ) { maybe_grow(i); _nodes[i] = n; }
1820 void insert( uint i, Node *n );
1821 void remove( uint i ); // Remove, preserving order
1822 // Clear all entries in _nodes to null but keep storage
1823 void clear() {
1824 Copy::zero_to_bytes(_nodes, _max * sizeof(Node*));
1825 }
1826
1827 uint max() const { return _max; }
1828 void dump() const;
1829 };
1830
1831 class Node_List : public Node_Array {
1832 uint _cnt;
1833 public:
1834 Node_List(uint max = OptoNodeListSize) : Node_Array(Thread::current()->resource_area(), max), _cnt(0) {}
1835 Node_List(Arena *a, uint max = OptoNodeListSize) : Node_Array(a, max), _cnt(0) {}
1836
1837 NONCOPYABLE(Node_List);
1838 Node_List& operator=(Node_List&&) = delete;
1839 // Allow move constructor for && (eg. capture return of function)
1840 Node_List(Node_List&&) = default;
1841
1842 bool contains(const Node* n) const {
1843 for (uint e = 0; e < size(); e++) {
1844 if (at(e) == n) return true;
1845 }
1846 return false;
1847 }
1848 void insert( uint i, Node *n ) { Node_Array::insert(i,n); _cnt++; }
1849 void remove( uint i ) { Node_Array::remove(i); _cnt--; }
1850 void push( Node *b ) { map(_cnt++,b); }
1851 void yank( Node *n ); // Find and remove
1852 Node *pop() { return _nodes[--_cnt]; }
1853 void clear() { _cnt = 0; Node_Array::clear(); } // retain storage
1854 void copy(const Node_List& from) {
1855 if (from._max > _max) {
1856 grow(from._max);
1857 }
1858 _cnt = from._cnt;
1859 Copy::conjoint_words_to_higher((HeapWord*)&from._nodes[0], (HeapWord*)&_nodes[0], from._max * sizeof(Node*));
1860 }
1861
1862 uint size() const { return _cnt; }
1863 void dump() const;
1864 void dump_simple() const;
1865 };
1866
1867 // Definition must appear after complete type definition of Node_List
1868 template <typename Callback, typename Check>
1869 void Node::visit_uses(Callback callback, Check is_boundary, bool always_callback) const {
1870 ResourceMark rm;
1871 VectorSet visited;
1872 Node_List worklist;
1873
1874 // The initial worklist consists of the direct uses
1875 for (DUIterator_Fast kmax, k = fast_outs(kmax); k < kmax; k++) {
1876 Node* out = fast_out(k);
1877 if (!visited.test_set(out->_idx)) { worklist.push(out); }
1878 }
1879
1880 while (worklist.size() > 0) {
1881 Node* use = worklist.pop();
1882 bool boundary = is_boundary(use);
1883 if (boundary || always_callback) {
1884 callback(use);
1885 }
1886 if (!boundary) {
1887 // Not a boundary node, continue search
1888 for (DUIterator_Fast kmax, k = use->fast_outs(kmax); k < kmax; k++) {
1889 Node* out = use->fast_out(k);
1890 if (!visited.test_set(out->_idx)) { worklist.push(out); }
1891 }
1892 }
1893 }
1894 }
1895
1896
1897 //------------------------------Unique_Node_List-------------------------------
1898 class Unique_Node_List : public Node_List {
1899 VectorSet _in_worklist;
1900 uint _clock_index; // Index in list where to pop from next
1901 public:
1902 Unique_Node_List() : Node_List(), _clock_index(0) {}
1903 Unique_Node_List(Arena *a) : Node_List(a), _in_worklist(a), _clock_index(0) {}
1904
1905 NONCOPYABLE(Unique_Node_List);
1906 Unique_Node_List& operator=(Unique_Node_List&&) = delete;
1907 // Allow move constructor for && (eg. capture return of function)
1908 Unique_Node_List(Unique_Node_List&&) = default;
1909
1910 void remove( Node *n );
1911 bool member(const Node* n) const { return _in_worklist.test(n->_idx) != 0; }
1912 VectorSet& member_set(){ return _in_worklist; }
1913
1914 void push(Node* b) {
1915 if( !_in_worklist.test_set(b->_idx) )
1916 Node_List::push(b);
1917 }
1918 void push_non_cfg_inputs_of(const Node* node) {
1919 for (uint i = 1; i < node->req(); i++) {
1920 Node* input = node->in(i);
1921 if (input != nullptr && !input->is_CFG()) {
1922 push(input);
1923 }
1924 }
1925 }
1926
1927 void push_outputs_of(const Node* node) {
1928 for (DUIterator_Fast imax, i = node->fast_outs(imax); i < imax; i++) {
1929 Node* output = node->fast_out(i);
1930 push(output);
1931 }
1932 }
1933
1934 Node *pop() {
1935 if( _clock_index >= size() ) _clock_index = 0;
1936 Node *b = at(_clock_index);
1937 map( _clock_index, Node_List::pop());
1938 if (size() != 0) _clock_index++; // Always start from 0
1939 _in_worklist.remove(b->_idx);
1940 return b;
1941 }
1942 Node *remove(uint i) {
1943 Node *b = Node_List::at(i);
1944 _in_worklist.remove(b->_idx);
1945 map(i,Node_List::pop());
1946 return b;
1947 }
1948 void yank(Node *n) {
1949 _in_worklist.remove(n->_idx);
1950 Node_List::yank(n);
1951 }
1952 void clear() {
1953 _in_worklist.clear(); // Discards storage but grows automatically
1954 Node_List::clear();
1955 _clock_index = 0;
1956 }
1957 void ensure_empty() {
1958 assert(size() == 0, "must be empty");
1959 clear(); // just in case
1960 }
1961
1962 // Used after parsing to remove useless nodes before Iterative GVN
1963 void remove_useless_nodes(VectorSet& useful);
1964
1965 // If the idx of the Nodes change, we must recompute the VectorSet
1966 void recompute_idx_set() {
1967 _in_worklist.clear();
1968 for (uint i = 0; i < size(); i++) {
1969 Node* n = at(i);
1970 _in_worklist.set(n->_idx);
1971 }
1972 }
1973
1974 #ifdef ASSERT
1975 bool is_subset_of(Unique_Node_List& other) {
1976 for (uint i = 0; i < size(); i++) {
1977 Node* n = at(i);
1978 if (!other.member(n)) {
1979 return false;
1980 }
1981 }
1982 return true;
1983 }
1984 #endif
1985
1986 bool contains(const Node* n) const {
1987 fatal("use faster member() instead");
1988 return false;
1989 }
1990
1991 #ifndef PRODUCT
1992 void print_set() const { _in_worklist.print(); }
1993 #endif
1994 };
1995
1996 // Unique_Mixed_Node_List
1997 // unique: nodes are added only once
1998 // mixed: allow new and old nodes
1999 class Unique_Mixed_Node_List : public ResourceObj {
2000 public:
2001 Unique_Mixed_Node_List() : _visited_set(cmpkey, hashkey) {}
2002
2003 void add(Node* node) {
2004 if (not_a_node(node)) {
2005 return; // Gracefully handle null, -1, 0xabababab, etc.
2006 }
2007 if (_visited_set[node] == nullptr) {
2008 _visited_set.Insert(node, node);
2009 _worklist.push(node);
2010 }
2011 }
2012
2013 Node* operator[] (uint i) const {
2014 return _worklist[i];
2015 }
2016
2017 size_t size() {
2018 return _worklist.size();
2019 }
2020
2021 private:
2022 Dict _visited_set;
2023 Node_List _worklist;
2024 };
2025
2026 // Inline definition of Compile::record_for_igvn must be deferred to this point.
2027 inline void Compile::record_for_igvn(Node* n) {
2028 _igvn_worklist->push(n);
2029 }
2030
2031 // Inline definition of Compile::remove_for_igvn must be deferred to this point.
2032 inline void Compile::remove_for_igvn(Node* n) {
2033 _igvn_worklist->remove(n);
2034 }
2035
2036 //------------------------------Node_Stack-------------------------------------
2037 class Node_Stack {
2038 protected:
2039 struct INode {
2040 Node *node; // Processed node
2041 uint indx; // Index of next node's child
2042 };
2043 INode *_inode_top; // tos, stack grows up
2044 INode *_inode_max; // End of _inodes == _inodes + _max
2045 INode *_inodes; // Array storage for the stack
2046 Arena *_a; // Arena to allocate in
2047 ReallocMark _nesting; // Safety checks for arena reallocation
2048
2049 void maybe_grow() {
2050 _nesting.check(_a); // Check if a potential reallocation in the arena is safe
2051 if (_inode_top >= _inode_max) {
2052 grow();
2053 }
2054 }
2055 void grow();
2056
2057 public:
2058 Node_Stack(int size) {
2059 size_t max = (size > OptoNodeListSize) ? size : OptoNodeListSize;
2060 _a = Thread::current()->resource_area();
2061 _inodes = NEW_ARENA_ARRAY( _a, INode, max );
2062 _inode_max = _inodes + max;
2063 _inode_top = _inodes - 1; // stack is empty
2064 }
2065
2066 Node_Stack(Arena *a, int size) : _a(a) {
2067 size_t max = (size > OptoNodeListSize) ? size : OptoNodeListSize;
2068 _inodes = NEW_ARENA_ARRAY( _a, INode, max );
2069 _inode_max = _inodes + max;
2070 _inode_top = _inodes - 1; // stack is empty
2071 }
2072
2073 void pop() {
2074 assert(_inode_top >= _inodes, "node stack underflow");
2075 --_inode_top;
2076 }
2077 void push(Node *n, uint i) {
2078 ++_inode_top;
2079 maybe_grow();
2080 INode *top = _inode_top; // optimization
2081 top->node = n;
2082 top->indx = i;
2083 }
2084 Node *node() const {
2085 return _inode_top->node;
2086 }
2087 Node* node_at(uint i) const {
2088 assert(_inodes + i <= _inode_top, "in range");
2089 return _inodes[i].node;
2090 }
2091 uint index() const {
2092 return _inode_top->indx;
2093 }
2094 uint index_at(uint i) const {
2095 assert(_inodes + i <= _inode_top, "in range");
2096 return _inodes[i].indx;
2097 }
2098 void set_node(Node *n) {
2099 _inode_top->node = n;
2100 }
2101 void set_index(uint i) {
2102 _inode_top->indx = i;
2103 }
2104 uint size_max() const { return (uint)pointer_delta(_inode_max, _inodes, sizeof(INode)); } // Max size
2105 uint size() const { return (uint)pointer_delta((_inode_top+1), _inodes, sizeof(INode)); } // Current size
2106 bool is_nonempty() const { return (_inode_top >= _inodes); }
2107 bool is_empty() const { return (_inode_top < _inodes); }
2108 void clear() { _inode_top = _inodes - 1; } // retain storage
2109
2110 // Node_Stack is used to map nodes.
2111 Node* find(uint idx) const;
2112
2113 NONCOPYABLE(Node_Stack);
2114 };
2115
2116
2117 //-----------------------------Node_Notes--------------------------------------
2118 // Debugging or profiling annotations loosely and sparsely associated
2119 // with some nodes. See Compile::node_notes_at for the accessor.
2120 class Node_Notes {
2121 JVMState* _jvms;
2122
2123 public:
2124 Node_Notes(JVMState* jvms = nullptr) {
2125 _jvms = jvms;
2126 }
2127
2128 JVMState* jvms() { return _jvms; }
2129 void set_jvms(JVMState* x) { _jvms = x; }
2130
2131 // True if there is nothing here.
2132 bool is_clear() {
2133 return (_jvms == nullptr);
2134 }
2135
2136 // Make there be nothing here.
2137 void clear() {
2138 _jvms = nullptr;
2139 }
2140
2141 // Make a new, clean node notes.
2142 static Node_Notes* make(Compile* C) {
2143 Node_Notes* nn = NEW_ARENA_ARRAY(C->comp_arena(), Node_Notes, 1);
2144 nn->clear();
2145 return nn;
2146 }
2147
2148 Node_Notes* clone(Compile* C) {
2149 Node_Notes* nn = NEW_ARENA_ARRAY(C->comp_arena(), Node_Notes, 1);
2150 (*nn) = (*this);
2151 return nn;
2152 }
2153
2154 // Absorb any information from source.
2155 bool update_from(Node_Notes* source) {
2156 bool changed = false;
2157 if (source != nullptr) {
2158 if (source->jvms() != nullptr) {
2159 set_jvms(source->jvms());
2160 changed = true;
2161 }
2162 }
2163 return changed;
2164 }
2165 };
2166
2167 // Inlined accessors for Compile::node_nodes that require the preceding class:
2168 inline Node_Notes*
2169 Compile::locate_node_notes(GrowableArray<Node_Notes*>* arr,
2170 int idx, bool can_grow) {
2171 assert(idx >= 0, "oob");
2172 int block_idx = (idx >> _log2_node_notes_block_size);
2173 int grow_by = (block_idx - (arr == nullptr? 0: arr->length()));
2174 if (grow_by >= 0) {
2175 if (!can_grow) return nullptr;
2176 grow_node_notes(arr, grow_by + 1);
2177 }
2178 if (arr == nullptr) return nullptr;
2179 // (Every element of arr is a sub-array of length _node_notes_block_size.)
2180 return arr->at(block_idx) + (idx & (_node_notes_block_size-1));
2181 }
2182
2183 inline Node_Notes* Compile::node_notes_at(int idx) {
2184 return locate_node_notes(_node_note_array, idx, false);
2185 }
2186
2187 inline bool
2188 Compile::set_node_notes_at(int idx, Node_Notes* value) {
2189 if (value == nullptr || value->is_clear())
2190 return false; // nothing to write => write nothing
2191 Node_Notes* loc = locate_node_notes(_node_note_array, idx, true);
2192 assert(loc != nullptr, "");
2193 return loc->update_from(value);
2194 }
2195
2196
2197 //------------------------------TypeNode---------------------------------------
2198 // Node with a Type constant.
2199 class TypeNode : public Node {
2200 protected:
2201 virtual uint hash() const; // Check the type
2202 virtual bool cmp( const Node &n ) const;
2203 virtual uint size_of() const; // Size is bigger
2204 const Type* const _type;
2205 public:
2206 void set_type(const Type* t) {
2207 assert(t != nullptr, "sanity");
2208 DEBUG_ONLY(uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH);
2209 *(const Type**)&_type = t; // cast away const-ness
2210 // If this node is in the hash table, make sure it doesn't need a rehash.
2211 assert(check_hash == NO_HASH || check_hash == hash(), "type change must preserve hash code");
2212 }
2213 const Type* type() const { assert(_type != nullptr, "sanity"); return _type; };
2214 TypeNode( const Type *t, uint required ) : Node(required), _type(t) {
2215 init_class_id(Class_Type);
2216 }
2217 virtual const Type* Value(PhaseGVN* phase) const;
2218 virtual Node* Ideal(PhaseGVN* phase, bool can_reshape);
2219 virtual const Type *bottom_type() const;
2220 virtual uint ideal_reg() const;
2221
2222 void make_path_dead(PhaseIterGVN* igvn, PhaseIdealLoop* loop, Node* ctrl_use, uint j, const char* phase_str);
2223 #ifndef PRODUCT
2224 virtual void dump_spec(outputStream *st) const;
2225 virtual void dump_compact_spec(outputStream *st) const;
2226 #endif
2227 void make_paths_from_here_dead(PhaseIterGVN* igvn, PhaseIdealLoop* loop, const char* phase_str);
2228 void create_halt_path(PhaseIterGVN* igvn, Node* c, PhaseIdealLoop* loop, const char* phase_str) const;
2229 };
2230
2231 #include "opto/opcodes.hpp"
2232
2233 #define Op_IL(op) \
2234 inline int Op_ ## op(BasicType bt) { \
2235 assert(bt == T_INT || bt == T_LONG, "only for int or longs"); \
2236 if (bt == T_INT) { \
2237 return Op_## op ## I; \
2238 } \
2239 return Op_## op ## L; \
2240 }
2241
2242 Op_IL(Add)
2243 Op_IL(And)
2244 Op_IL(Sub)
2245 Op_IL(Mul)
2246 Op_IL(URShift)
2247 Op_IL(LShift)
2248 Op_IL(RShift)
2249 Op_IL(Xor)
2250 Op_IL(Cmp)
2251 Op_IL(Div)
2252 Op_IL(Mod)
2253 Op_IL(UDiv)
2254 Op_IL(UMod)
2255
2256 inline int Op_ConIL(BasicType bt) {
2257 assert(bt == T_INT || bt == T_LONG, "only for int or longs");
2258 if (bt == T_INT) {
2259 return Op_ConI;
2260 }
2261 return Op_ConL;
2262 }
2263
2264 inline int Op_Cmp_unsigned(BasicType bt) {
2265 assert(bt == T_INT || bt == T_LONG, "only for int or longs");
2266 if (bt == T_INT) {
2267 return Op_CmpU;
2268 }
2269 return Op_CmpUL;
2270 }
2271
2272 inline int Op_Cast(BasicType bt) {
2273 assert(bt == T_INT || bt == T_LONG, "only for int or longs");
2274 if (bt == T_INT) {
2275 return Op_CastII;
2276 }
2277 return Op_CastLL;
2278 }
2279
2280 inline int Op_DivIL(BasicType bt, bool is_unsigned) {
2281 assert(bt == T_INT || bt == T_LONG, "only for int or longs");
2282 if (bt == T_INT) {
2283 if (is_unsigned) {
2284 return Op_UDivI;
2285 } else {
2286 return Op_DivI;
2287 }
2288 }
2289 if (is_unsigned) {
2290 return Op_UDivL;
2291 } else {
2292 return Op_DivL;
2293 }
2294 }
2295
2296 inline int Op_DivModIL(BasicType bt, bool is_unsigned) {
2297 assert(bt == T_INT || bt == T_LONG, "only for int or longs");
2298 if (bt == T_INT) {
2299 if (is_unsigned) {
2300 return Op_UDivModI;
2301 } else {
2302 return Op_DivModI;
2303 }
2304 }
2305 if (is_unsigned) {
2306 return Op_UDivModL;
2307 } else {
2308 return Op_DivModL;
2309 }
2310 }
2311
2312 // Interface to define actions that should be taken when running DataNodeBFS. Each use can extend this class to specify
2313 // a customized BFS.
2314 class BFSActions : public StackObj {
2315 public:
2316 // Should a node's inputs further be visited in the BFS traversal? By default, we visit all data inputs. Override this
2317 // method to provide a custom filter.
2318 virtual bool should_visit(Node* node) const {
2319 // By default, visit all inputs.
2320 return true;
2321 };
2322
2323 // Is the visited node a target node that we are looking for in the BFS traversal? We do not visit its inputs further
2324 // but the BFS will continue to visit all unvisited nodes in the queue.
2325 virtual bool is_target_node(Node* node) const = 0;
2326
2327 // Defines an action that should be taken when we visit a target node in the BFS traversal.
2328 // To give more freedom, we pass the direct child node to the target node such that
2329 // child->in(i) == target node. This allows to also directly replace the target node instead
2330 // of only updating its inputs.
2331 virtual void target_node_action(Node* child, uint i) = 0;
2332 };
2333
2334 // Class to perform a BFS traversal on the data nodes from a given start node. The provided BFSActions guide which
2335 // data node's inputs should be further visited, which data nodes are target nodes and what to do with the target nodes.
2336 class DataNodeBFS : public StackObj {
2337 BFSActions& _bfs_actions;
2338
2339 public:
2340 explicit DataNodeBFS(BFSActions& bfs_action) : _bfs_actions(bfs_action) {}
2341
2342 // Run the BFS starting from 'start_node' and apply the actions provided to this class.
2343 void run(Node* start_node) {
2344 ResourceMark rm;
2345 Unique_Node_List _nodes_to_visit;
2346 _nodes_to_visit.push(start_node);
2347 for (uint i = 0; i < _nodes_to_visit.size(); i++) {
2348 Node* next = _nodes_to_visit[i];
2349 for (uint j = 1; j < next->req(); j++) {
2350 Node* input = next->in(j);
2351 if (_bfs_actions.is_target_node(input)) {
2352 assert(_bfs_actions.should_visit(input), "must also pass node filter");
2353 _bfs_actions.target_node_action(next, j);
2354 } else if (_bfs_actions.should_visit(input)) {
2355 _nodes_to_visit.push(input);
2356 }
2357 }
2358 }
2359 }
2360 };
2361
2362 #endif // SHARE_OPTO_NODE_HPP