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