1 /*
2 * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef SHARE_OPTO_COMPILE_HPP
26 #define SHARE_OPTO_COMPILE_HPP
27
28 #include "asm/codeBuffer.hpp"
29 #include "ci/compilerInterface.hpp"
30 #include "code/debugInfoRec.hpp"
31 #include "compiler/cHeapStringHolder.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compiler_globals.hpp"
34 #include "compiler/compilerEvent.hpp"
35 #include "libadt/dict.hpp"
36 #include "libadt/vectset.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/methodData.hpp"
39 #include "opto/idealGraphPrinter.hpp"
40 #include "opto/phase.hpp"
41 #include "opto/phasetype.hpp"
42 #include "opto/printinlining.hpp"
43 #include "opto/regmask.hpp"
44 #include "runtime/deoptimization.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/timerTrace.hpp"
47 #include "runtime/vmThread.hpp"
48 #include "utilities/growableArray.hpp"
49 #include "utilities/ticks.hpp"
50 #include "utilities/vmEnums.hpp"
51
52 class AbstractLockNode;
53 class AddPNode;
54 class Block;
55 class Bundle;
56 class CallGenerator;
57 class CallNode;
58 class CallStaticJavaNode;
59 class CloneMap;
60 class CompilationFailureInfo;
61 class ConnectionGraph;
62 class IdealGraphPrinter;
63 class InlineTree;
64 class Matcher;
65 class MachConstantNode;
66 class MachConstantBaseNode;
67 class MachNode;
68 class MachOper;
69 class MachSafePointNode;
70 class Node;
71 class Node_Array;
72 class Node_List;
73 class Node_Notes;
74 class NodeHash;
75 class NodeCloneInfo;
76 class OpaqueTemplateAssertionPredicateNode;
77 class OptoReg;
78 class ParsePredicateNode;
79 class PhaseCFG;
80 class PhaseGVN;
81 class PhaseIterGVN;
82 class PhaseRegAlloc;
83 class PhaseCCP;
84 class PhaseOutput;
85 class ReachabilityFenceNode;
86 class RootNode;
87 class relocInfo;
88 class StartNode;
89 class SafePointNode;
90 class JVMState;
91 class Type;
92 class TypeInt;
93 class TypeInteger;
94 class TypeKlassPtr;
95 class TypePtr;
96 class TypeOopPtr;
97 class TypeFunc;
98 class TypeVect;
99 class Type_Array;
100 class Unique_Node_List;
101 class UnstableIfTrap;
102 class InlineTypeNode;
103 class nmethod;
104 class Node_Stack;
105 struct Final_Reshape_Counts;
106 class VerifyMeetResult;
107
108 enum LoopOptsMode {
109 LoopOptsDefault,
110 LoopOptsNone,
111 LoopOptsMaxUnroll,
112 LoopOptsShenandoahExpand,
113 LoopOptsSkipSplitIf,
114 LoopOptsVerify,
115 PostLoopOptsExpandReachabilityFences
116 };
117
118 // The type of all node counts and indexes.
119 // It must hold at least 16 bits, but must also be fast to load and store.
120 // This type, if less than 32 bits, could limit the number of possible nodes.
121 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
122 typedef unsigned int node_idx_t;
123
124 class NodeCloneInfo {
125 private:
126 uint64_t _idx_clone_orig;
127 public:
128
129 void set_idx(node_idx_t idx) {
130 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
131 }
132 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
133
134 void set_gen(int generation) {
135 uint64_t g = (uint64_t)generation << 32;
136 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
137 }
138 int gen() const { return (int)(_idx_clone_orig >> 32); }
139
140 void set(uint64_t x) { _idx_clone_orig = x; }
141 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
142 uint64_t get() const { return _idx_clone_orig; }
143
144 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
145 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
146
147 void dump_on(outputStream* st) const;
148 };
149
150 class CloneMap {
151 friend class Compile;
152 private:
153 bool _debug;
154 Dict* _dict;
155 int _clone_idx; // current cloning iteration/generation in loop unroll
156 public:
157 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
158 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
159 Dict* dict() const { return _dict; }
160 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
161 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
162 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
163 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
164 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
165 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
166 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
167 int max_gen() const;
168 void clone(Node* old, Node* nnn, int gen);
169 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
170 void dump(node_idx_t key, outputStream* st) const;
171
172 int clone_idx() const { return _clone_idx; }
173 void set_clone_idx(int x) { _clone_idx = x; }
174 bool is_debug() const { return _debug; }
175 void set_debug(bool debug) { _debug = debug; }
176
177 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
178 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
179 };
180
181 class Options {
182 friend class Compile;
183 private:
184 const bool _subsume_loads; // Load can be matched as part of a larger op.
185 const bool _do_escape_analysis; // Do escape analysis.
186 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
187 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges.
188 const bool _eliminate_boxing; // Do boxing elimination.
189 const bool _do_locks_coarsening; // Do locks coarsening
190 const bool _do_superword; // Do SuperWord
191 const bool _install_code; // Install the code that was compiled
192 public:
193 Options(bool subsume_loads,
194 bool do_escape_analysis,
195 bool do_iterative_escape_analysis,
196 bool do_reduce_allocation_merges,
197 bool eliminate_boxing,
198 bool do_locks_coarsening,
199 bool do_superword,
200 bool install_code) :
201 _subsume_loads(subsume_loads),
202 _do_escape_analysis(do_escape_analysis),
203 _do_iterative_escape_analysis(do_iterative_escape_analysis),
204 _do_reduce_allocation_merges(do_reduce_allocation_merges),
205 _eliminate_boxing(eliminate_boxing),
206 _do_locks_coarsening(do_locks_coarsening),
207 _do_superword(do_superword),
208 _install_code(install_code) {
209 }
210
211 static Options for_runtime_stub() {
212 return Options(
213 /* subsume_loads = */ true,
214 /* do_escape_analysis = */ false,
215 /* do_iterative_escape_analysis = */ false,
216 /* do_reduce_allocation_merges = */ false,
217 /* eliminate_boxing = */ false,
218 /* do_lock_coarsening = */ false,
219 /* do_superword = */ true,
220 /* install_code = */ true
221 );
222 }
223 };
224
225 //------------------------------Compile----------------------------------------
226 // This class defines a top-level Compiler invocation.
227
228 class Compile : public Phase {
229
230 public:
231 // Fixed alias indexes. (See also MergeMemNode.)
232 enum {
233 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
234 AliasIdxBot = 2, // pseudo-index, aliases to everything
235 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
236 };
237
238 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
239 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
240 // then brackets are put into the log, with time stamps and node counts.
241 // (The time collection itself is always conditionalized on CITime.)
242 class TracePhase : public TraceTime {
243 private:
244 Compile* const _compile;
245 CompileLog* _log;
246 const bool _dolog;
247 public:
248 TracePhase(PhaseTraceId phaseTraceId);
249 TracePhase(const char* name, PhaseTraceId phaseTraceId);
250 ~TracePhase();
251 const char* phase_name() const { return title(); }
252 };
253
254 // Information per category of alias (memory slice)
255 class AliasType {
256 private:
257 friend class Compile;
258
259 int _index; // unique index, used with MergeMemNode
260 const TypePtr* _adr_type; // normalized address type
261 ciField* _field; // relevant instance field, or null if none
262 const Type* _element; // relevant array element type, or null if none
263 bool _is_rewritable; // false if the memory is write-once only
264 int _general_index; // if this is type is an instance, the general
265 // type that this is an instance of
266
267 void Init(int i, const TypePtr* at);
268
269 public:
270 int index() const { return _index; }
271 const TypePtr* adr_type() const { return _adr_type; }
272 ciField* field() const { return _field; }
273 const Type* element() const { return _element; }
274 bool is_rewritable() const { return _is_rewritable; }
275 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
276 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
277
278 void set_rewritable(bool z) { _is_rewritable = z; }
279 void set_field(ciField* f) {
280 assert(!_field,"");
281 _field = f;
282 if (f->is_final() || f->is_stable()) {
283 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
284 _is_rewritable = false;
285 }
286 }
287 void set_element(const Type* e) {
288 assert(_element == nullptr, "");
289 _element = e;
290 }
291
292 BasicType basic_type() const;
293
294 void print_on(outputStream* st) PRODUCT_RETURN;
295 };
296
297 enum {
298 logAliasCacheSize = 6,
299 AliasCacheSize = (1<<logAliasCacheSize)
300 };
301 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
302 enum {
303 trapHistLength = MethodData::_trap_hist_limit
304 };
305
306 private:
307 // Fixed parameters to this compilation.
308 const int _compile_id;
309 const Options _options; // Compilation options
310 ciMethod* _method; // The method being compiled.
311 int _entry_bci; // entry bci for osr methods.
312 const TypeFunc* _tf; // My kind of signature
313 InlineTree* _ilt; // Ditto (temporary).
314 address _stub_function; // VM entry for stub being compiled, or null
315 const char* _stub_name; // Name of stub or adapter being compiled, or null
316 StubId _stub_id; // unique id for stub or NO_STUBID
317 address _stub_entry_point; // Compile code entry for generated stub, or null
318
319 // Control of this compilation.
320 int _max_inline_size; // Max inline size for this compilation
321 int _freq_inline_size; // Max hot method inline size for this compilation
322 int _fixed_slots; // count of frame slots not allocated by the register
323 // allocator i.e. locks, original deopt pc, etc.
324 uintx _max_node_limit; // Max unique node count during a single compilation.
325
326 bool _post_loop_opts_phase; // Loop opts are finished.
327 bool _merge_stores_phase; // Phase for merging stores, after post loop opts phase.
328 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
329
330 /* If major progress is set:
331 * Marks that the loop tree information (get_ctrl, idom, get_loop, etc.) could be invalid, and we need to rebuild the loop tree.
332 * It also indicates that the graph was changed in a way that is promising to be able to apply more loop optimization.
333 * If major progress is not set:
334 * Loop tree information is valid.
335 * If major progress is not set at the end of a loop opts phase, then we can stop loop opts, because we do not expect any further progress if we did more loop opts phases.
336 *
337 * This is not 100% accurate, the semantics of major progress has become less clear over time, but this is the general idea.
338 */
339 bool _major_progress;
340 bool _inlining_progress; // progress doing incremental inlining?
341 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
342 bool _strength_reduction; // Are we doing strength reduction to direct call
343 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
344 bool _has_loops; // True if the method _may_ have some loops
345 bool _has_split_ifs; // True if the method _may_ have some split-if
346 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
347 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
348 bool _has_boxed_value; // True if a boxed object is allocated
349 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
350 bool _has_circular_inline_type; // True if method loads an inline type with a circular, non-flat field
351 bool _needs_nm_slot; // True if an extra stack slot is needed to hold the null marker at scalarized returns
352 uint _max_vector_size; // Maximum size of generated vectors
353 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
354 uint _trap_hist[trapHistLength]; // Cumulative traps
355 bool _trap_can_recompile; // Have we emitted a recompiling trap?
356 uint _decompile_count; // Cumulative decompilation counts.
357 bool _do_inlining; // True if we intend to do inlining
358 bool _do_scheduling; // True if we intend to do scheduling
359 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
360 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
361 bool _use_cmove; // True if CMove should be used without profitability analysis
362 bool _do_aliasing; // True if we intend to do aliasing
363 bool _print_assembly; // True if we should dump assembly code for this compilation
364 bool _print_inlining; // True if we should print inlining for this compilation
365 bool _print_intrinsics; // True if we should print intrinsics for this compilation
366 bool _print_phase_loop_opts; // True if we should print before and after loop opts phase
367 #ifndef PRODUCT
368 uint _phase_counter; // Counter for the number of already printed phases
369 uint _igv_idx; // Counter for IGV node identifiers
370 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
371 bool _trace_opto_output;
372 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
373 #endif
374 bool _has_irreducible_loop; // Found irreducible loops
375 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
376 bool _has_scoped_access; // For shared scope closure
377 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
378 int _loop_opts_cnt; // loop opts round
379 bool _has_flat_accesses; // Any known flat array accesses?
380 bool _flat_accesses_share_alias; // Initially all flat array share a single slice
381 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
382 uint _stress_seed; // Seed for stress testing
383
384 // Compilation environment.
385 Arena _comp_arena; // Arena with lifetime equivalent to Compile
386 void* _barrier_set_state; // Potential GC barrier state for Compile
387 ciEnv* _env; // CI interface
388 DirectiveSet* _directive; // Compiler directive
389 CompileLog* _log; // from CompilerThread
390 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
391 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
392 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
393 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
394 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
395 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates which can be seen as list
396 // of Template Assertion Predicates themselves.
397 GrowableArray<OpaqueTemplateAssertionPredicateNode*> _template_assertion_predicate_opaques;
398 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
399 GrowableArray<ReachabilityFenceNode*> _reachability_fences; // List of reachability fences
400 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
401 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
402 GrowableArray<Node*> _flat_access_nodes; // List of LoadFlat and StoreFlat nodes
403 GrowableArray<Node*> _for_merge_stores_igvn; // List of nodes for IGVN merge stores
404 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
405 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
406 ConnectionGraph* _congraph;
407 #ifndef PRODUCT
408 IdealGraphPrinter* _igv_printer;
409 static IdealGraphPrinter* _debug_file_printer;
410 static IdealGraphPrinter* _debug_network_printer;
411 #endif
412
413
414 // Node management
415 uint _unique; // Counter for unique Node indices
416 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
417 // So use this to keep count and make the call O(1).
418 VectorSet _dead_node_list; // Set of dead nodes
419 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
420 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
421
422 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
423
424 // Arenas for new-space and old-space nodes.
425 // Swapped between using _node_arena.
426 // The lifetime of the old-space nodes is during xform.
427 Arena _node_arena_one;
428 Arena _node_arena_two;
429 Arena* _node_arena;
430 public:
431 Arena* swap_old_and_new() {
432 Arena* filled_arena_ptr = _node_arena;
433 Arena* old_arena_ptr = old_arena();
434 old_arena_ptr->destruct_contents();
435 _node_arena = old_arena_ptr;
436 return filled_arena_ptr;
437 }
438 private:
439 RootNode* _root; // Unique root of compilation, or null after bail-out.
440 Node* _top; // Unique top node. (Reset by various phases.)
441
442 Node* _immutable_memory; // Initial memory state
443
444 Node* _recent_alloc_obj;
445 Node* _recent_alloc_ctl;
446
447 // Constant table
448 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
449
450
451 // Blocked array of debugging and profiling information,
452 // tracked per node.
453 enum { _log2_node_notes_block_size = 8,
454 _node_notes_block_size = (1<<_log2_node_notes_block_size)
455 };
456 GrowableArray<Node_Notes*>* _node_note_array;
457 Node_Notes* _default_node_notes; // default notes for new nodes
458
459 // After parsing and every bulk phase we hang onto the Root instruction.
460 // The RootNode instruction is where the whole program begins. It produces
461 // the initial Control and BOTTOM for everybody else.
462
463 // Type management
464 Arena _Compile_types; // Arena for all types
465 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
466 Dict* _type_dict; // Intern table
467 CloneMap _clone_map; // used for recording history of cloned nodes
468 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
469 ciMethod* _last_tf_m; // Cache for
470 const TypeFunc* _last_tf; // TypeFunc::make
471 AliasType** _alias_types; // List of alias types seen so far.
472 int _num_alias_types; // Logical length of _alias_types
473 int _max_alias_types; // Physical length of _alias_types
474 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
475
476 // Parsing, optimization
477 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
478
479 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
480 // If pushed outside IGVN, the Node is processed in the next IGVN round.
481 Unique_Node_List* _igvn_worklist;
482
483 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
484 Type_Array* _types;
485
486 // Shared node hash table for GVN, IGVN and CCP.
487 NodeHash* _node_hash;
488
489 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
490 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
491 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
492
493 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
494
495 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
496 bool _has_mh_late_inlines; // Can there still be a method handle late inlining pending?
497 // false: there can't be one
498 // true: we've enqueued one at some point so there may still be one
499
500 // "MemLimit" directive was specified and the memory limit was hit during compilation
501 bool _oom;
502
503 // Only keep nodes in the expensive node list that need to be optimized
504 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
505 // Use for sorting expensive nodes to bring similar nodes together
506 static int cmp_expensive_nodes(Node** n1, Node** n2);
507 // Expensive nodes list already sorted?
508 bool expensive_nodes_sorted() const;
509 // Remove the speculative part of types and clean up the graph
510 void remove_speculative_types(PhaseIterGVN &igvn);
511
512 void* _replay_inline_data; // Pointer to data loaded from file
513
514 void log_late_inline_failure(CallGenerator* cg, const char* msg);
515 DEBUG_ONLY(bool _exception_backedge;)
516
517 void record_method_not_compilable_oom();
518
519 InlinePrinter _inline_printer;
520
521 public:
522 void* barrier_set_state() const { return _barrier_set_state; }
523
524 InlinePrinter* inline_printer() { return &_inline_printer; }
525
526 #ifndef PRODUCT
527 IdealGraphPrinter* igv_printer() { return _igv_printer; }
528 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
529 #endif
530
531 void log_late_inline(CallGenerator* cg);
532 void log_inline_id(CallGenerator* cg);
533 void log_inline_failure(const char* msg);
534
535 void* replay_inline_data() const { return _replay_inline_data; }
536
537 // Dump inlining replay data to the stream.
538 void dump_inline_data(outputStream* out);
539 void dump_inline_data_reduced(outputStream* out);
540
541 private:
542 // Matching, CFG layout, allocation, code generation
543 PhaseCFG* _cfg; // Results of CFG finding
544 int _java_calls; // Number of java calls in the method
545 int _inner_loops; // Number of inner loops in the method
546 Matcher* _matcher; // Engine to map ideal to machine instructions
547 PhaseRegAlloc* _regalloc; // Results of register allocation.
548 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
549 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
550 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
551 int _interpreter_frame_size;
552
553 // Holds dynamically allocated extensions of short-lived register masks. Such
554 // extensions are potentially quite large and need tight resource marks which
555 // may conflict with other allocations in the default resource area.
556 // Therefore, we use a dedicated resource area for register masks.
557 ResourceArea _regmask_arena;
558
559 PhaseOutput* _output;
560
561 public:
562 // Accessors
563
564 // The Compile instance currently active in this (compiler) thread.
565 static Compile* current() {
566 return (Compile*) ciEnv::current()->compiler_data();
567 }
568
569 int interpreter_frame_size() const { return _interpreter_frame_size; }
570
571 PhaseOutput* output() const { return _output; }
572 void set_output(PhaseOutput* o) { _output = o; }
573
574 // ID for this compilation. Useful for setting breakpoints in the debugger.
575 int compile_id() const { return _compile_id; }
576 DirectiveSet* directive() const { return _directive; }
577
578 // Does this compilation allow instructions to subsume loads? User
579 // instructions that subsume a load may result in an unschedulable
580 // instruction sequence.
581 bool subsume_loads() const { return _options._subsume_loads; }
582 /** Do escape analysis. */
583 bool do_escape_analysis() const { return _options._do_escape_analysis; }
584 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
585 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
586 /** Do boxing elimination. */
587 bool eliminate_boxing() const { return _options._eliminate_boxing; }
588 /** Do aggressive boxing elimination. */
589 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
590 bool should_install_code() const { return _options._install_code; }
591 /** Do locks coarsening. */
592 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
593 bool do_superword() const { return _options._do_superword; }
594
595 // Other fixed compilation parameters.
596 ciMethod* method() const { return _method; }
597 int entry_bci() const { return _entry_bci; }
598 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
599 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
600 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
601 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
602 InlineTree* ilt() const { return _ilt; }
603 address stub_function() const { return _stub_function; }
604 const char* stub_name() const { return _stub_name; }
605 StubId stub_id() const { return _stub_id; }
606 address stub_entry_point() const { return _stub_entry_point; }
607 void set_stub_entry_point(address z) { _stub_entry_point = z; }
608
609 // Control of this compilation.
610 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
611 void set_fixed_slots(int n) { _fixed_slots = n; }
612 void set_inlining_progress(bool z) { _inlining_progress = z; }
613 int inlining_progress() const { return _inlining_progress; }
614 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
615 int inlining_incrementally() const { return _inlining_incrementally; }
616 void set_strength_reduction(bool z) { _strength_reduction = z; }
617 bool strength_reduction() const { return _strength_reduction; }
618 void set_do_cleanup(bool z) { _do_cleanup = z; }
619 int do_cleanup() const { return _do_cleanup; }
620 bool major_progress() const { return _major_progress; }
621 void set_major_progress() { _major_progress = true; }
622 void restore_major_progress(bool progress) { _major_progress = _major_progress || progress; }
623 void clear_major_progress() { _major_progress = false; }
624 int max_inline_size() const { return _max_inline_size; }
625 void set_freq_inline_size(int n) { _freq_inline_size = n; }
626 int freq_inline_size() const { return _freq_inline_size; }
627 void set_max_inline_size(int n) { _max_inline_size = n; }
628 bool has_loops() const { return _has_loops; }
629 void set_has_loops(bool z) { _has_loops = z; }
630 bool has_split_ifs() const { return _has_split_ifs; }
631 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
632 bool has_unsafe_access() const { return _has_unsafe_access; }
633 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
634 bool has_stringbuilder() const { return _has_stringbuilder; }
635 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
636 bool has_boxed_value() const { return _has_boxed_value; }
637 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
638 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
639 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
640 bool has_circular_inline_type() const { return _has_circular_inline_type; }
641 void set_has_circular_inline_type(bool z) { _has_circular_inline_type = z; }
642 uint max_vector_size() const { return _max_vector_size; }
643 void set_max_vector_size(uint s) { _max_vector_size = s; }
644 bool clear_upper_avx() const { return _clear_upper_avx; }
645 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
646 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
647 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
648 bool trap_can_recompile() const { return _trap_can_recompile; }
649 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
650 uint decompile_count() const { return _decompile_count; }
651 void set_decompile_count(uint c) { _decompile_count = c; }
652 bool allow_range_check_smearing() const;
653 bool do_inlining() const { return _do_inlining; }
654 void set_do_inlining(bool z) { _do_inlining = z; }
655 bool do_scheduling() const { return _do_scheduling; }
656 void set_do_scheduling(bool z) { _do_scheduling = z; }
657 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
658 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
659 bool do_vector_loop() const { return _do_vector_loop; }
660 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
661 bool use_cmove() const { return _use_cmove; }
662 void set_use_cmove(bool z) { _use_cmove = z; }
663 bool do_aliasing() const { return _do_aliasing; }
664 bool print_assembly() const { return _print_assembly; }
665 void set_print_assembly(bool z) { _print_assembly = z; }
666 bool print_inlining() const { return _print_inlining; }
667 void set_print_inlining(bool z) { _print_inlining = z; }
668 bool print_intrinsics() const { return _print_intrinsics; }
669 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
670 uint max_node_limit() const { return (uint)_max_node_limit; }
671 void set_max_node_limit(uint n) { _max_node_limit = n; }
672 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
673 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
674 void set_flat_accesses() { _has_flat_accesses = true; }
675 bool flat_accesses_share_alias() const { return _flat_accesses_share_alias; }
676 void set_flat_accesses_share_alias(bool z) { _flat_accesses_share_alias = z; }
677 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
678 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
679
680 // Support for scalarized inline type calling convention
681 bool has_scalarized_args() const { return _method != nullptr && _method->has_scalarized_args(); }
682 bool needs_stack_repair() const { return _method != nullptr && _method->c2_needs_stack_repair(); }
683 bool needs_nm_slot() const { return _needs_nm_slot; }
684 void set_needs_nm_slot(bool v) { _needs_nm_slot = v; }
685
686 bool has_monitors() const { return _has_monitors; }
687 void set_has_monitors(bool v) { _has_monitors = v; }
688 bool has_scoped_access() const { return _has_scoped_access; }
689 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
690
691 // check the CompilerOracle for special behaviours for this compile
692 bool method_has_option(CompileCommandEnum option) const {
693 return method() != nullptr && method()->has_option(option);
694 }
695
696 #ifndef PRODUCT
697 uint next_igv_idx() { return _igv_idx++; }
698 bool trace_opto_output() const { return _trace_opto_output; }
699 void print_phase(const char* phase_name);
700 void print_ideal_ir(const char* compile_phase_name) const;
701 bool should_print_ideal() const { return _directive->PrintIdealOption; }
702 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
703 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
704 int _in_dump_cnt; // Required for dumping ir nodes.
705 #endif
706 bool has_irreducible_loop() const { return _has_irreducible_loop; }
707 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
708
709 Ticks _latest_stage_start_counter;
710
711 void begin_method();
712 void end_method();
713
714 void print_method(CompilerPhaseType compile_phase, int level, Node* n = nullptr);
715
716 #ifndef PRODUCT
717 bool should_print_igv(int level);
718 bool should_print_phase(int level) const;
719 bool should_print_ideal_phase(CompilerPhaseType cpt) const;
720 void init_igv();
721 void dump_igv(const char* graph_name, int level = 3) {
722 if (should_print_igv(level)) {
723 _igv_printer->print_graph(graph_name, nullptr);
724 }
725 }
726
727 void igv_print_method_to_file(const char* phase_name = nullptr, bool append = false, const frame* fr = nullptr);
728 void igv_print_method_to_network(const char* phase_name = nullptr, const frame* fr = nullptr);
729 void igv_print_graph_to_network(const char* name, GrowableArray<const Node*>& visible_nodes, const frame* fr);
730 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
731 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
732 #endif
733
734 const GrowableArray<ParsePredicateNode*>& parse_predicates() const {
735 return _parse_predicates;
736 }
737
738 const GrowableArray<OpaqueTemplateAssertionPredicateNode*>& template_assertion_predicate_opaques() const {
739 return _template_assertion_predicate_opaques;
740 }
741
742 int macro_count() const { return _macro_nodes.length(); }
743 int parse_predicate_count() const { return _parse_predicates.length(); }
744 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaques.length(); }
745 int expensive_count() const { return _expensive_nodes.length(); }
746 int coarsened_count() const { return _coarsened_locks.length(); }
747 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
748
749 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
750
751 ReachabilityFenceNode* reachability_fence(int idx) const { return _reachability_fences.at(idx); }
752 int reachability_fences_count() const { return _reachability_fences.length(); }
753
754 ConnectionGraph* congraph() { return _congraph;}
755 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
756 void add_macro_node(Node * n) {
757 //assert(n->is_macro(), "must be a macro node");
758 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
759 _macro_nodes.append(n);
760 }
761 void remove_macro_node(Node* n) {
762 // this function may be called twice for a node so we can only remove it
763 // if it's still existing.
764 _macro_nodes.remove_if_existing(n);
765 // Remove from coarsened locks list if present
766 if (coarsened_count() > 0) {
767 remove_coarsened_lock(n);
768 }
769 }
770 void add_expensive_node(Node* n);
771 void remove_expensive_node(Node* n) {
772 _expensive_nodes.remove_if_existing(n);
773 }
774
775 void add_reachability_fence(ReachabilityFenceNode* rf) {
776 _reachability_fences.append(rf);
777 }
778
779 void remove_reachability_fence(ReachabilityFenceNode* n) {
780 _reachability_fences.remove_if_existing(n);
781 }
782
783 void add_parse_predicate(ParsePredicateNode* n) {
784 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
785 _parse_predicates.append(n);
786 }
787
788 void remove_parse_predicate(ParsePredicateNode* n) {
789 if (parse_predicate_count() > 0) {
790 _parse_predicates.remove_if_existing(n);
791 }
792 }
793
794 void add_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
795 assert(!_template_assertion_predicate_opaques.contains(n),
796 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
797 _template_assertion_predicate_opaques.append(n);
798 }
799
800 void remove_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
801 if (template_assertion_predicate_count() > 0) {
802 _template_assertion_predicate_opaques.remove_if_existing(n);
803 }
804 }
805 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
806 void remove_coarsened_lock(Node* n);
807 bool coarsened_locks_consistent();
808 void mark_unbalanced_boxes() const;
809
810 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
811 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
812 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
813
814 #ifdef ASSERT
815 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
816 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
817 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
818 #endif
819
820 bool allow_macro_nodes() { return _allow_macro_nodes; }
821 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
822
823 void record_for_post_loop_opts_igvn(Node* n);
824 void remove_from_post_loop_opts_igvn(Node* n);
825 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
826
827 // Keep track of inline type nodes for later processing
828 void add_inline_type(Node* n);
829 void remove_inline_type(Node* n);
830
831 bool clear_argument_if_only_used_as_buffer_at_calls(Node* result_cast, PhaseIterGVN& igvn);
832
833 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
834
835 void add_flat_access(Node* n);
836 void remove_flat_access(Node* n);
837 void process_flat_accesses(PhaseIterGVN& igvn);
838
839 template <class F>
840 void for_each_flat_access(F consumer) {
841 for (int i = _flat_access_nodes.length() - 1; i >= 0; i--) {
842 consumer(_flat_access_nodes.at(i));
843 }
844 }
845
846 void adjust_flat_array_access_aliases(PhaseIterGVN& igvn);
847
848 void record_unstable_if_trap(UnstableIfTrap* trap);
849 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
850 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
851 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
852
853 bool merge_stores_phase() { return _merge_stores_phase; }
854 void set_merge_stores_phase() { _merge_stores_phase = true; }
855 void record_for_merge_stores_igvn(Node* n);
856 void remove_from_merge_stores_igvn(Node* n);
857 void process_for_merge_stores_igvn(PhaseIterGVN& igvn);
858
859 void shuffle_late_inlines();
860 void shuffle_macro_nodes();
861 void sort_macro_nodes();
862
863 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
864
865 // Are there candidate expensive nodes for optimization?
866 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
867 // Check whether n1 and n2 are similar
868 static int cmp_expensive_nodes(Node* n1, Node* n2);
869 // Sort expensive nodes to locate similar expensive nodes
870 void sort_expensive_nodes();
871
872 // Compilation environment.
873 Arena* comp_arena() { return &_comp_arena; }
874 ciEnv* env() const { return _env; }
875 CompileLog* log() const { return _log; }
876
877 bool failing_internal() const {
878 return _env->failing() ||
879 _failure_reason.get() != nullptr;
880 }
881
882 const char* failure_reason() const {
883 return _env->failing() ? _env->failure_reason()
884 : _failure_reason.get();
885 }
886
887 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
888
889 bool failing() {
890 if (failing_internal()) {
891 return true;
892 }
893 #ifdef ASSERT
894 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
895 if (phase_verify_ideal_loop()) {
896 return false;
897 }
898 if (StressBailout) {
899 return fail_randomly();
900 }
901 #endif
902 return false;
903 }
904
905 #ifdef ASSERT
906 bool fail_randomly();
907 bool failure_is_artificial();
908 #endif
909
910 bool failure_reason_is(const char* r) const {
911 return (r == _failure_reason.get()) ||
912 (r != nullptr &&
913 _failure_reason.get() != nullptr &&
914 strcmp(r, _failure_reason.get()) == 0);
915 }
916
917 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
918 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
919 env()->record_method_not_compilable(reason);
920 // Record failure reason.
921 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
922 }
923 bool check_node_count(uint margin, const char* reason) {
924 if (oom()) {
925 record_method_not_compilable_oom();
926 return true;
927 }
928 if (live_nodes() + margin > max_node_limit()) {
929 record_method_not_compilable(reason);
930 return true;
931 } else {
932 return false;
933 }
934 }
935 bool oom() const { return _oom; }
936 void set_oom() { _oom = true; }
937
938 // Node management
939 uint unique() const { return _unique; }
940 uint next_unique() { return _unique++; }
941 void set_unique(uint i) { _unique = i; }
942 Arena* node_arena() { return _node_arena; }
943 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
944 RootNode* root() const { return _root; }
945 void set_root(RootNode* r) { _root = r; }
946 StartNode* start() const; // (Derived from root.)
947 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
948 Node* immutable_memory();
949
950 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
951 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
952 void set_recent_alloc(Node* ctl, Node* obj) {
953 _recent_alloc_ctl = ctl;
954 _recent_alloc_obj = obj;
955 }
956 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
957 _dead_node_count++;
958 }
959 void reset_dead_node_list() { _dead_node_list.reset();
960 _dead_node_count = 0;
961 }
962 uint live_nodes() const {
963 int val = _unique - _dead_node_count;
964 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
965 return (uint) val;
966 }
967 #ifdef ASSERT
968 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
969 bool phase_optimize_finished() const { return _phase_optimize_finished; }
970 uint count_live_nodes_by_graph_walk();
971 void print_missing_nodes();
972 #endif
973
974 // Record modified nodes to check that they are put on IGVN worklist
975 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
976 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
977 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
978
979 MachConstantBaseNode* mach_constant_base_node();
980 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
981 // Generated by adlc, true if CallNode requires MachConstantBase.
982 bool needs_deep_clone_jvms();
983
984 // Handy undefined Node
985 Node* top() const { return _top; }
986
987 // these are used by guys who need to know about creation and transformation of top:
988 Node* cached_top_node() { return _top; }
989 void set_cached_top_node(Node* tn);
990
991 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
992 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
993 Node_Notes* default_node_notes() const { return _default_node_notes; }
994 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
995
996 Node_Notes* node_notes_at(int idx);
997
998 inline bool set_node_notes_at(int idx, Node_Notes* value);
999 // Copy notes from source to dest, if they exist.
1000 // Overwrite dest only if source provides something.
1001 // Return true if information was moved.
1002 bool copy_node_notes_to(Node* dest, Node* source);
1003
1004 // Workhorse function to sort out the blocked Node_Notes array:
1005 Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
1006 int idx, bool can_grow = false);
1007
1008 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
1009
1010 // Type management
1011 Arena* type_arena() { return _type_arena; }
1012 Dict* type_dict() { return _type_dict; }
1013 size_t type_last_size() { return _type_last_size; }
1014 int num_alias_types() { return _num_alias_types; }
1015
1016 void init_type_arena() { _type_arena = &_Compile_types; }
1017 void set_type_arena(Arena* a) { _type_arena = a; }
1018 void set_type_dict(Dict* d) { _type_dict = d; }
1019 void set_type_last_size(size_t sz) { _type_last_size = sz; }
1020
1021 const TypeFunc* last_tf(ciMethod* m) {
1022 return (m == _last_tf_m) ? _last_tf : nullptr;
1023 }
1024 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
1025 assert(m != nullptr || tf == nullptr, "");
1026 _last_tf_m = m;
1027 _last_tf = tf;
1028 }
1029
1030 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
1031 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
1032 bool have_alias_type(const TypePtr* adr_type);
1033 AliasType* alias_type(ciField* field);
1034
1035 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, nullptr, uncached)->index(); }
1036 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
1037 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
1038
1039 // Building nodes
1040 void rethrow_exceptions(JVMState* jvms);
1041 void return_values(JVMState* jvms);
1042 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1043
1044 // Decide how to build a call.
1045 // The profile factor is a discount to apply to this site's interp. profile.
1046 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1047 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
1048 bool allow_intrinsics = true);
1049 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1050 return C->directive()->should_delay_inline(call_method) ||
1051 should_delay_string_inlining(call_method, jvms) ||
1052 should_delay_boxing_inlining(call_method, jvms) ||
1053 should_delay_vector_inlining(call_method, jvms);
1054 }
1055 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1056 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1057 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
1058 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
1059
1060 // Helper functions to identify inlining potential at call-site
1061 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
1062 ciKlass* holder, ciMethod* callee,
1063 const TypeOopPtr* receiver_type, bool is_virtual,
1064 bool &call_does_dispatch, int &vtable_index,
1065 bool check_access = true);
1066 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1067 ciMethod* callee, const TypeOopPtr* receiver_type,
1068 bool check_access = true);
1069
1070 // Report if there were too many traps at a current method and bci.
1071 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1072 // If there is no MDO at all, report no trap unless told to assume it.
1073 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1074 // This version, unspecific to a particular bci, asks if
1075 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1076 bool too_many_traps(Deoptimization::DeoptReason reason,
1077 // Privately used parameter for logging:
1078 ciMethodData* logmd = nullptr);
1079 // Report if there were too many recompiles at a method and bci.
1080 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1081 // Report if there were too many traps or recompiles at a method and bci.
1082 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1083 return too_many_traps(method, bci, reason) ||
1084 too_many_recompiles(method, bci, reason);
1085 }
1086 // Return a bitset with the reasons where deoptimization is allowed,
1087 // i.e., where there were not too many uncommon traps.
1088 int _allowed_reasons;
1089 int allowed_deopt_reasons() { return _allowed_reasons; }
1090 void set_allowed_deopt_reasons();
1091
1092 // Parsing, optimization
1093 PhaseGVN* initial_gvn() { return _initial_gvn; }
1094 Unique_Node_List* igvn_worklist() {
1095 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1096 return _igvn_worklist;
1097 }
1098 Type_Array* types() {
1099 assert(_types != nullptr, "must be created in Compile::Compile");
1100 return _types;
1101 }
1102 NodeHash* node_hash() {
1103 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1104 return _node_hash;
1105 }
1106 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1107 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1108 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1109
1110 // Replace n by nn using initial_gvn, calling hash_delete and
1111 // record_for_igvn as needed.
1112 void gvn_replace_by(Node* n, Node* nn);
1113
1114
1115 void identify_useful_nodes(Unique_Node_List &useful);
1116 void update_dead_node_list(Unique_Node_List &useful);
1117 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist, const Unique_Node_List* root_and_safepoints = nullptr);
1118
1119 void remove_useless_node(Node* dead);
1120
1121 // Record this CallGenerator for inlining at the end of parsing.
1122 void add_late_inline(CallGenerator* cg) {
1123 _late_inlines.insert_before(_late_inlines_pos, cg);
1124 if (StressIncrementalInlining) {
1125 assert(_late_inlines_pos < _late_inlines.length(), "unthinkable!");
1126 if (_late_inlines.length() - _late_inlines_pos >= 2) {
1127 int j = (C->random() % (_late_inlines.length() - _late_inlines_pos)) + _late_inlines_pos;
1128 swap(_late_inlines.at(_late_inlines_pos), _late_inlines.at(j));
1129 }
1130 }
1131 _late_inlines_pos++;
1132 }
1133
1134 void prepend_late_inline(CallGenerator* cg) {
1135 _late_inlines.insert_before(0, cg);
1136 }
1137
1138 void add_string_late_inline(CallGenerator* cg) {
1139 _string_late_inlines.push(cg);
1140 }
1141
1142 void add_boxing_late_inline(CallGenerator* cg) {
1143 _boxing_late_inlines.push(cg);
1144 }
1145
1146 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1147 _vector_reboxing_late_inlines.push(cg);
1148 }
1149
1150 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1151 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1152
1153 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1154 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1155
1156 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1157
1158 void dump_print_inlining();
1159
1160 bool over_inlining_cutoff() const {
1161 if (!inlining_incrementally()) {
1162 return unique() > (uint)NodeCountInliningCutoff;
1163 } else {
1164 // Give some room for incremental inlining algorithm to "breathe"
1165 // and avoid thrashing when live node count is close to the limit.
1166 // Keep in mind that live_nodes() isn't accurate during inlining until
1167 // dead node elimination step happens (see Compile::inline_incrementally).
1168 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1169 }
1170 }
1171
1172 void mark_has_mh_late_inlines() { _has_mh_late_inlines = true; }
1173 bool has_mh_late_inlines() const { return _has_mh_late_inlines; }
1174
1175 bool inline_incrementally_one();
1176 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1177 void inline_incrementally(PhaseIterGVN& igvn);
1178 bool should_stress_inlining() { return StressIncrementalInlining && (random() % 2) == 0; }
1179 bool should_delay_inlining() { return AlwaysIncrementalInline || should_stress_inlining(); }
1180 void inline_string_calls(bool parse_time);
1181 void inline_boxing_calls(PhaseIterGVN& igvn);
1182 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1183 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1184
1185 void inline_vector_reboxing_calls();
1186 bool has_vbox_nodes();
1187
1188 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1189
1190 // Matching, CFG layout, allocation, code generation
1191 PhaseCFG* cfg() { return _cfg; }
1192 bool has_java_calls() const { return _java_calls > 0; }
1193 int java_calls() const { return _java_calls; }
1194 int inner_loops() const { return _inner_loops; }
1195 Matcher* matcher() { return _matcher; }
1196 PhaseRegAlloc* regalloc() { return _regalloc; }
1197 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1198 ResourceArea* regmask_arena() { return &_regmask_arena; }
1199 Arena* indexSet_arena() { return _indexSet_arena; }
1200 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1201 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1202
1203 void update_interpreter_frame_size(int size) {
1204 if (_interpreter_frame_size < size) {
1205 _interpreter_frame_size = size;
1206 }
1207 }
1208
1209 void set_matcher(Matcher* m) { _matcher = m; }
1210 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1211 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1212 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1213
1214 void set_java_calls(int z) { _java_calls = z; }
1215 void set_inner_loops(int z) { _inner_loops = z; }
1216
1217 Dependencies* dependencies() { return env()->dependencies(); }
1218
1219 // Major entry point. Given a Scope, compile the associated method.
1220 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1221 // replacement, entry_bci indicates the bytecode for which to compile a
1222 // continuation.
1223 Compile(ciEnv* ci_env, ciMethod* target,
1224 int entry_bci, Options options, DirectiveSet* directive);
1225
1226 // Second major entry point. From the TypeFunc signature, generate code
1227 // to pass arguments from the Java calling convention to the C calling
1228 // convention.
1229 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1230 address stub_function, const char *stub_name,
1231 StubId stub_id, int is_fancy_jump, bool pass_tls,
1232 bool return_pc, DirectiveSet* directive);
1233
1234 ~Compile();
1235
1236 // Are we compiling a method?
1237 bool has_method() { return method() != nullptr; }
1238
1239 // Maybe print some information about this compile.
1240 void print_compile_messages();
1241
1242 // Final graph reshaping, a post-pass after the regular optimizer is done.
1243 bool final_graph_reshaping();
1244
1245 // returns true if adr is completely contained in the given alias category
1246 bool must_alias(const TypePtr* adr, int alias_idx);
1247
1248 // returns true if adr overlaps with the given alias category
1249 bool can_alias(const TypePtr* adr, int alias_idx);
1250
1251 // Stack slots that may be unused by the calling convention but must
1252 // otherwise be preserved. On Intel this includes the return address.
1253 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1254 uint in_preserve_stack_slots() {
1255 return SharedRuntime::in_preserve_stack_slots();
1256 }
1257
1258 // "Top of Stack" slots that may be unused by the calling convention but must
1259 // otherwise be preserved.
1260 // On Intel these are not necessary and the value can be zero.
1261 static uint out_preserve_stack_slots() {
1262 return SharedRuntime::out_preserve_stack_slots();
1263 }
1264
1265 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1266 // for calls to C. Supports the var-args backing area for register parms.
1267 uint varargs_C_out_slots_killed() const;
1268
1269 // Number of Stack Slots consumed by a synchronization entry
1270 int sync_stack_slots() const;
1271
1272 // Compute the name of old_SP. See <arch>.ad for frame layout.
1273 OptoReg::Name compute_old_SP();
1274
1275 private:
1276 // Phase control:
1277 void Init(bool aliasing); // Prepare for a single compilation
1278 void Optimize(); // Given a graph, optimize it
1279 void Code_Gen(); // Generate code from a graph
1280
1281 // Management of the AliasType table.
1282 void grow_alias_types();
1283 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1284 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1285 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1286
1287 void verify_top(Node*) const PRODUCT_RETURN;
1288
1289 // Intrinsic setup.
1290 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1291 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1292 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1293 void register_intrinsic(CallGenerator* cg); // update fn
1294
1295 #ifndef PRODUCT
1296 static juint _intrinsic_hist_count[];
1297 static jubyte _intrinsic_hist_flags[];
1298 #endif
1299 // Function calls made by the public function final_graph_reshaping.
1300 // No need to be made public as they are not called elsewhere.
1301 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1302 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1303 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1304 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1305
1306 // Logic cone optimization.
1307 void optimize_logic_cones(PhaseIterGVN &igvn);
1308 void collect_logic_cone_roots(Unique_Node_List& list);
1309 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1310 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1311 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1312 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1313 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1314 void check_no_dead_use() const NOT_DEBUG_RETURN;
1315
1316 public:
1317
1318 // Note: Histogram array size is about 1 Kb.
1319 enum { // flag bits:
1320 _intrinsic_worked = 1, // succeeded at least once
1321 _intrinsic_failed = 2, // tried it but it failed
1322 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1323 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1324 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1325 };
1326 // Update histogram. Return boolean if this is a first-time occurrence.
1327 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1328 bool is_virtual, int flags) PRODUCT_RETURN0;
1329 static void print_intrinsic_statistics() PRODUCT_RETURN;
1330
1331 // Graph verification code
1332 // Walk the node list, verifying that there is a one-to-one correspondence
1333 // between Use-Def edges and Def-Use edges. The option no_dead_code enables
1334 // stronger checks that the graph is strongly connected from starting points
1335 // in both directions.
1336 // root_and_safepoints is used to give the starting points for the traversal.
1337 // If not supplied, only root is used. When this check is called after CCP,
1338 // we need to start traversal from Root and safepoints, just like CCP does its
1339 // own traversal (see PhaseCCP::transform for reasons).
1340 //
1341 // To call this function, there are 2 ways to go:
1342 // - give root_and_safepoints to start traversal everywhere needed (like after CCP)
1343 // - if the whole graph is assumed to be reachable from Root's input,
1344 // root_and_safepoints is not needed (like in PhaseRemoveUseless).
1345 //
1346 // Failure to specify root_and_safepoints in case the graph is not fully
1347 // reachable from Root's input make this check unsound (can miss inconsistencies)
1348 // and even incomplete (can make up non-existing problems) if no_dead_code is
1349 // true.
1350 void verify_graph_edges(bool no_dead_code = false, const Unique_Node_List* root_and_safepoints = nullptr) const PRODUCT_RETURN;
1351
1352 // Verify bi-directional correspondence of edges
1353 void verify_bidirectional_edges(Unique_Node_List& visited, const Unique_Node_List* root_and_safepoints = nullptr) const;
1354
1355 // End-of-run dumps.
1356 static void print_statistics() PRODUCT_RETURN;
1357
1358 // Verify ADLC assumptions during startup
1359 static void adlc_verification() PRODUCT_RETURN;
1360
1361 // Definitions of pd methods
1362 static void pd_compiler2_init();
1363
1364 // Materialize reachability fences from reachability edges on safepoints.
1365 void expand_reachability_edges(Unique_Node_List& safepoints);
1366
1367 // Static parse-time type checking logic for gen_subtype_check:
1368 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1369 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1370
1371 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1372 // Optional control dependency (for example, on range check)
1373 Node* ctrl = nullptr);
1374
1375 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1376 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1377
1378 // Auxiliary method for randomized fuzzing/stressing
1379 int random();
1380 bool randomized_select(int count);
1381
1382 // seed random number generation and log the seed for repeatability.
1383 void initialize_stress_seed(const DirectiveSet* directive);
1384
1385 // supporting clone_map
1386 CloneMap& clone_map();
1387 void set_clone_map(Dict* d);
1388
1389 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1390 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1391 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1392
1393 #ifdef ASSERT
1394 VerifyMeetResult* _type_verify;
1395 void set_exception_backedge() { _exception_backedge = true; }
1396 bool has_exception_backedge() const { return _exception_backedge; }
1397 #endif
1398
1399 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1400 BasicType out_bt, BasicType in_bt);
1401
1402 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1403
1404 #ifndef PRODUCT
1405 private:
1406 // getting rid of the template makes things easier
1407 Node* make_debug_print_call(const char* str, address call_addr, PhaseGVN* gvn,
1408 Node* parm0 = nullptr, Node* parm1 = nullptr,
1409 Node* parm2 = nullptr, Node* parm3 = nullptr,
1410 Node* parm4 = nullptr, Node* parm5 = nullptr,
1411 Node* parm6 = nullptr) const;
1412
1413 public:
1414 // Creates a CallLeafNode for a runtime call that prints a static string and the values of the
1415 // nodes passed as arguments.
1416 // This function also takes care of doing the necessary wiring, including finding a suitable control
1417 // based on the nodes that need to be printed. Note that passing nodes that have incompatible controls
1418 // is undefined behavior.
1419 template <typename... TT, typename... NN>
1420 Node* make_debug_print(const char* str, PhaseGVN* gvn, NN... in) {
1421 address call_addr = CAST_FROM_FN_PTR(address, SharedRuntime::debug_print<TT...>);
1422 return make_debug_print_call(str, call_addr, gvn, in...);
1423 }
1424 #endif
1425 };
1426
1427 #endif // SHARE_OPTO_COMPILE_HPP