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