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