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