1 /*
2 * Copyright (c) 1997, 2025, 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 #ifndef PRODUCT
364 uint _phase_counter; // Counter for the number of already printed phases
365 uint _igv_idx; // Counter for IGV node identifiers
366 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
367 bool _trace_opto_output;
368 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
369 #endif
370 bool _has_irreducible_loop; // Found irreducible loops
371 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
372 bool _has_scoped_access; // For shared scope closure
373 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
374 int _loop_opts_cnt; // loop opts round
375 bool _has_flat_accesses; // Any known flat array accesses?
376 bool _flat_accesses_share_alias; // Initially all flat array share a single slice
377 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
378 uint _stress_seed; // Seed for stress testing
379
380 // Compilation environment.
381 Arena _comp_arena; // Arena with lifetime equivalent to Compile
382 void* _barrier_set_state; // Potential GC barrier state for Compile
383 ciEnv* _env; // CI interface
384 DirectiveSet* _directive; // Compiler directive
385 CompileLog* _log; // from CompilerThread
386 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
387 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
388 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
389 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
390 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
391 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates which can be seen as list
392 // of Template Assertion Predicates themselves.
393 GrowableArray<OpaqueTemplateAssertionPredicateNode*> _template_assertion_predicate_opaques;
394 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
395 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
396 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
397 GrowableArray<Node*> _flat_access_nodes; // List of LoadFlat and StoreFlat nodes
398 GrowableArray<Node*> _for_merge_stores_igvn; // List of nodes for IGVN merge stores
399 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
400 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
401 ConnectionGraph* _congraph;
402 #ifndef PRODUCT
403 IdealGraphPrinter* _igv_printer;
404 static IdealGraphPrinter* _debug_file_printer;
405 static IdealGraphPrinter* _debug_network_printer;
406 #endif
407
408
409 // Node management
410 uint _unique; // Counter for unique Node indices
411 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
412 // So use this to keep count and make the call O(1).
413 VectorSet _dead_node_list; // Set of dead nodes
414 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
415 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
416
417 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
418
419 // Arenas for new-space and old-space nodes.
420 // Swapped between using _node_arena.
421 // The lifetime of the old-space nodes is during xform.
422 Arena _node_arena_one;
423 Arena _node_arena_two;
424 Arena* _node_arena;
425 public:
426 Arena* swap_old_and_new() {
427 Arena* filled_arena_ptr = _node_arena;
428 Arena* old_arena_ptr = old_arena();
429 old_arena_ptr->destruct_contents();
430 _node_arena = old_arena_ptr;
431 return filled_arena_ptr;
432 }
433 private:
434 RootNode* _root; // Unique root of compilation, or null after bail-out.
435 Node* _top; // Unique top node. (Reset by various phases.)
436
437 Node* _immutable_memory; // Initial memory state
438
439 Node* _recent_alloc_obj;
440 Node* _recent_alloc_ctl;
441
442 // Constant table
443 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
444
445
446 // Blocked array of debugging and profiling information,
447 // tracked per node.
448 enum { _log2_node_notes_block_size = 8,
449 _node_notes_block_size = (1<<_log2_node_notes_block_size)
450 };
451 GrowableArray<Node_Notes*>* _node_note_array;
452 Node_Notes* _default_node_notes; // default notes for new nodes
453
454 // After parsing and every bulk phase we hang onto the Root instruction.
455 // The RootNode instruction is where the whole program begins. It produces
456 // the initial Control and BOTTOM for everybody else.
457
458 // Type management
459 Arena _Compile_types; // Arena for all types
460 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
461 Dict* _type_dict; // Intern table
462 CloneMap _clone_map; // used for recording history of cloned nodes
463 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
464 ciMethod* _last_tf_m; // Cache for
465 const TypeFunc* _last_tf; // TypeFunc::make
466 AliasType** _alias_types; // List of alias types seen so far.
467 int _num_alias_types; // Logical length of _alias_types
468 int _max_alias_types; // Physical length of _alias_types
469 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
470
471 // Parsing, optimization
472 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
473
474 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
475 // If pushed outside IGVN, the Node is processed in the next IGVN round.
476 Unique_Node_List* _igvn_worklist;
477
478 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
479 Type_Array* _types;
480
481 // Shared node hash table for GVN, IGVN and CCP.
482 NodeHash* _node_hash;
483
484 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
485 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
486 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
487
488 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
489
490 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
491 bool _has_mh_late_inlines; // Can there still be a method handle late inlining pending?
492 // false: there can't be one
493 // true: we've enqueued one at some point so there may still be one
494
495 // "MemLimit" directive was specified and the memory limit was hit during compilation
496 bool _oom;
497
498 // Only keep nodes in the expensive node list that need to be optimized
499 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
500 // Use for sorting expensive nodes to bring similar nodes together
501 static int cmp_expensive_nodes(Node** n1, Node** n2);
502 // Expensive nodes list already sorted?
503 bool expensive_nodes_sorted() const;
504 // Remove the speculative part of types and clean up the graph
505 void remove_speculative_types(PhaseIterGVN &igvn);
506
507 void* _replay_inline_data; // Pointer to data loaded from file
508
509 void log_late_inline_failure(CallGenerator* cg, const char* msg);
510 DEBUG_ONLY(bool _exception_backedge;)
511
512 void record_method_not_compilable_oom();
513
514 InlinePrinter _inline_printer;
515
516 public:
517 void* barrier_set_state() const { return _barrier_set_state; }
518
519 InlinePrinter* inline_printer() { return &_inline_printer; }
520
521 #ifndef PRODUCT
522 IdealGraphPrinter* igv_printer() { return _igv_printer; }
523 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
524 #endif
525
526 void log_late_inline(CallGenerator* cg);
527 void log_inline_id(CallGenerator* cg);
528 void log_inline_failure(const char* msg);
529
530 void* replay_inline_data() const { return _replay_inline_data; }
531
532 // Dump inlining replay data to the stream.
533 void dump_inline_data(outputStream* out);
534 void dump_inline_data_reduced(outputStream* out);
535
536 private:
537 // Matching, CFG layout, allocation, code generation
538 PhaseCFG* _cfg; // Results of CFG finding
539 int _java_calls; // Number of java calls in the method
540 int _inner_loops; // Number of inner loops in the method
541 Matcher* _matcher; // Engine to map ideal to machine instructions
542 PhaseRegAlloc* _regalloc; // Results of register allocation.
543 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
544 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
545 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
546 int _interpreter_frame_size;
547
548 // Holds dynamically allocated extensions of short-lived register masks. Such
549 // extensions are potentially quite large and need tight resource marks which
550 // may conflict with other allocations in the default resource area.
551 // Therefore, we use a dedicated resource area for register masks.
552 ResourceArea _regmask_arena;
553
554 PhaseOutput* _output;
555
556 public:
557 // Accessors
558
559 // The Compile instance currently active in this (compiler) thread.
560 static Compile* current() {
561 return (Compile*) ciEnv::current()->compiler_data();
562 }
563
564 int interpreter_frame_size() const { return _interpreter_frame_size; }
565
566 PhaseOutput* output() const { return _output; }
567 void set_output(PhaseOutput* o) { _output = o; }
568
569 // ID for this compilation. Useful for setting breakpoints in the debugger.
570 int compile_id() const { return _compile_id; }
571 DirectiveSet* directive() const { return _directive; }
572
573 // Does this compilation allow instructions to subsume loads? User
574 // instructions that subsume a load may result in an unschedulable
575 // instruction sequence.
576 bool subsume_loads() const { return _options._subsume_loads; }
577 /** Do escape analysis. */
578 bool do_escape_analysis() const { return _options._do_escape_analysis; }
579 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
580 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
581 /** Do boxing elimination. */
582 bool eliminate_boxing() const { return _options._eliminate_boxing; }
583 /** Do aggressive boxing elimination. */
584 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
585 bool should_install_code() const { return _options._install_code; }
586 /** Do locks coarsening. */
587 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
588 bool do_superword() const { return _options._do_superword; }
589
590 // Other fixed compilation parameters.
591 ciMethod* method() const { return _method; }
592 int entry_bci() const { return _entry_bci; }
593 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
594 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
595 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
596 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
597 InlineTree* ilt() const { return _ilt; }
598 address stub_function() const { return _stub_function; }
599 const char* stub_name() const { return _stub_name; }
600 StubId stub_id() const { return _stub_id; }
601 address stub_entry_point() const { return _stub_entry_point; }
602 void set_stub_entry_point(address z) { _stub_entry_point = z; }
603
604 // Control of this compilation.
605 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
606 void set_fixed_slots(int n) { _fixed_slots = n; }
607 void set_inlining_progress(bool z) { _inlining_progress = z; }
608 int inlining_progress() const { return _inlining_progress; }
609 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
610 int inlining_incrementally() const { return _inlining_incrementally; }
611 void set_strength_reduction(bool z) { _strength_reduction = z; }
612 bool strength_reduction() const { return _strength_reduction; }
613 void set_do_cleanup(bool z) { _do_cleanup = z; }
614 int do_cleanup() const { return _do_cleanup; }
615 bool major_progress() const { return _major_progress; }
616 void set_major_progress() { _major_progress = true; }
617 void restore_major_progress(bool progress) { _major_progress = _major_progress || progress; }
618 void clear_major_progress() { _major_progress = false; }
619 int max_inline_size() const { return _max_inline_size; }
620 void set_freq_inline_size(int n) { _freq_inline_size = n; }
621 int freq_inline_size() const { return _freq_inline_size; }
622 void set_max_inline_size(int n) { _max_inline_size = n; }
623 bool has_loops() const { return _has_loops; }
624 void set_has_loops(bool z) { _has_loops = z; }
625 bool has_split_ifs() const { return _has_split_ifs; }
626 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
627 bool has_unsafe_access() const { return _has_unsafe_access; }
628 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
629 bool has_stringbuilder() const { return _has_stringbuilder; }
630 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
631 bool has_boxed_value() const { return _has_boxed_value; }
632 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
633 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
634 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
635 bool has_circular_inline_type() const { return _has_circular_inline_type; }
636 void set_has_circular_inline_type(bool z) { _has_circular_inline_type = z; }
637 uint max_vector_size() const { return _max_vector_size; }
638 void set_max_vector_size(uint s) { _max_vector_size = s; }
639 bool clear_upper_avx() const { return _clear_upper_avx; }
640 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
641 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
642 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
643 bool trap_can_recompile() const { return _trap_can_recompile; }
644 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
645 uint decompile_count() const { return _decompile_count; }
646 void set_decompile_count(uint c) { _decompile_count = c; }
647 bool allow_range_check_smearing() const;
648 bool do_inlining() const { return _do_inlining; }
649 void set_do_inlining(bool z) { _do_inlining = z; }
650 bool do_scheduling() const { return _do_scheduling; }
651 void set_do_scheduling(bool z) { _do_scheduling = z; }
652 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
653 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
654 bool do_vector_loop() const { return _do_vector_loop; }
655 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
656 bool use_cmove() const { return _use_cmove; }
657 void set_use_cmove(bool z) { _use_cmove = z; }
658 bool do_aliasing() const { return _do_aliasing; }
659 bool print_assembly() const { return _print_assembly; }
660 void set_print_assembly(bool z) { _print_assembly = z; }
661 bool print_inlining() const { return _print_inlining; }
662 void set_print_inlining(bool z) { _print_inlining = z; }
663 bool print_intrinsics() const { return _print_intrinsics; }
664 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
665 uint max_node_limit() const { return (uint)_max_node_limit; }
666 void set_max_node_limit(uint n) { _max_node_limit = n; }
667 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
668 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
669 void set_flat_accesses() { _has_flat_accesses = true; }
670 bool flat_accesses_share_alias() const { return _flat_accesses_share_alias; }
671 void set_flat_accesses_share_alias(bool z) { _flat_accesses_share_alias = z; }
672 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
673 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
674
675 // Support for scalarized inline type calling convention
676 bool has_scalarized_args() const { return _method != nullptr && _method->has_scalarized_args(); }
677 bool needs_stack_repair() const { return _method != nullptr && _method->get_Method()->c2_needs_stack_repair(); }
678
679 bool has_monitors() const { return _has_monitors; }
680 void set_has_monitors(bool v) { _has_monitors = v; }
681 bool has_scoped_access() const { return _has_scoped_access; }
682 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
683
684 // check the CompilerOracle for special behaviours for this compile
685 bool method_has_option(CompileCommandEnum option) const {
686 return method() != nullptr && method()->has_option(option);
687 }
688
689 #ifndef PRODUCT
690 uint next_igv_idx() { return _igv_idx++; }
691 bool trace_opto_output() const { return _trace_opto_output; }
692 void print_phase(const char* phase_name);
693 void print_ideal_ir(const char* phase_name);
694 bool should_print_ideal() const { return _directive->PrintIdealOption; }
695 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
696 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
697 int _in_dump_cnt; // Required for dumping ir nodes.
698 #endif
699 bool has_irreducible_loop() const { return _has_irreducible_loop; }
700 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
701
702 Ticks _latest_stage_start_counter;
703
704 void begin_method();
705 void end_method();
706
707 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
708
709 #ifndef PRODUCT
710 bool should_print_igv(int level);
711 bool should_print_phase(int level) const;
712 bool should_print_ideal_phase(CompilerPhaseType cpt) const;
713 void init_igv();
714 void dump_igv(const char* graph_name, int level = 3) {
715 if (should_print_igv(level)) {
716 _igv_printer->print_graph(graph_name, nullptr);
717 }
718 }
719
720 void igv_print_method_to_file(const char* phase_name = nullptr, bool append = false, const frame* fr = nullptr);
721 void igv_print_method_to_network(const char* phase_name = nullptr, const frame* fr = nullptr);
722 void igv_print_graph_to_network(const char* name, GrowableArray<const Node*>& visible_nodes, const frame* fr);
723 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
724 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
725 #endif
726
727 const GrowableArray<ParsePredicateNode*>& parse_predicates() const {
728 return _parse_predicates;
729 }
730
731 const GrowableArray<OpaqueTemplateAssertionPredicateNode*>& template_assertion_predicate_opaques() const {
732 return _template_assertion_predicate_opaques;
733 }
734
735 int macro_count() const { return _macro_nodes.length(); }
736 int parse_predicate_count() const { return _parse_predicates.length(); }
737 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaques.length(); }
738 int expensive_count() const { return _expensive_nodes.length(); }
739 int coarsened_count() const { return _coarsened_locks.length(); }
740
741 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
742
743 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
744
745 ConnectionGraph* congraph() { return _congraph;}
746 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
747 void add_macro_node(Node * n) {
748 //assert(n->is_macro(), "must be a macro node");
749 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
750 _macro_nodes.append(n);
751 }
752 void remove_macro_node(Node* n) {
753 // this function may be called twice for a node so we can only remove it
754 // if it's still existing.
755 _macro_nodes.remove_if_existing(n);
756 // Remove from coarsened locks list if present
757 if (coarsened_count() > 0) {
758 remove_coarsened_lock(n);
759 }
760 }
761 void add_expensive_node(Node* n);
762 void remove_expensive_node(Node* n) {
763 _expensive_nodes.remove_if_existing(n);
764 }
765
766 void add_parse_predicate(ParsePredicateNode* n) {
767 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
768 _parse_predicates.append(n);
769 }
770
771 void remove_parse_predicate(ParsePredicateNode* n) {
772 if (parse_predicate_count() > 0) {
773 _parse_predicates.remove_if_existing(n);
774 }
775 }
776
777 void add_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
778 assert(!_template_assertion_predicate_opaques.contains(n),
779 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
780 _template_assertion_predicate_opaques.append(n);
781 }
782
783 void remove_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
784 if (template_assertion_predicate_count() > 0) {
785 _template_assertion_predicate_opaques.remove_if_existing(n);
786 }
787 }
788 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
789 void remove_coarsened_lock(Node* n);
790 bool coarsened_locks_consistent();
791 void mark_unbalanced_boxes() const;
792
793 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
794 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
795 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
796
797 #ifdef ASSERT
798 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
799 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
800 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
801 #endif
802
803 bool allow_macro_nodes() { return _allow_macro_nodes; }
804 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
805
806 void record_for_post_loop_opts_igvn(Node* n);
807 void remove_from_post_loop_opts_igvn(Node* n);
808 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
809
810 // Keep track of inline type nodes for later processing
811 void add_inline_type(Node* n);
812 void remove_inline_type(Node* n);
813 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
814
815 void add_flat_access(Node* n);
816 void remove_flat_access(Node* n);
817 void process_flat_accesses(PhaseIterGVN& igvn);
818
819 template <class F>
820 void for_each_flat_access(F consumer) {
821 for (int i = _flat_access_nodes.length() - 1; i >= 0; i--) {
822 consumer(_flat_access_nodes.at(i));
823 }
824 }
825
826 void adjust_flat_array_access_aliases(PhaseIterGVN& igvn);
827
828 void record_unstable_if_trap(UnstableIfTrap* trap);
829 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
830 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
831 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
832
833 bool merge_stores_phase() { return _merge_stores_phase; }
834 void set_merge_stores_phase() { _merge_stores_phase = true; }
835 void record_for_merge_stores_igvn(Node* n);
836 void remove_from_merge_stores_igvn(Node* n);
837 void process_for_merge_stores_igvn(PhaseIterGVN& igvn);
838
839 void shuffle_macro_nodes();
840 void sort_macro_nodes();
841
842 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
843
844 // Are there candidate expensive nodes for optimization?
845 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
846 // Check whether n1 and n2 are similar
847 static int cmp_expensive_nodes(Node* n1, Node* n2);
848 // Sort expensive nodes to locate similar expensive nodes
849 void sort_expensive_nodes();
850
851 // Compilation environment.
852 Arena* comp_arena() { return &_comp_arena; }
853 ciEnv* env() const { return _env; }
854 CompileLog* log() const { return _log; }
855
856 bool failing_internal() const {
857 return _env->failing() ||
858 _failure_reason.get() != nullptr;
859 }
860
861 const char* failure_reason() const {
862 return _env->failing() ? _env->failure_reason()
863 : _failure_reason.get();
864 }
865
866 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
867
868 bool failing() {
869 if (failing_internal()) {
870 return true;
871 }
872 #ifdef ASSERT
873 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
874 if (phase_verify_ideal_loop()) {
875 return false;
876 }
877 if (StressBailout) {
878 return fail_randomly();
879 }
880 #endif
881 return false;
882 }
883
884 #ifdef ASSERT
885 bool fail_randomly();
886 bool failure_is_artificial();
887 #endif
888
889 bool failure_reason_is(const char* r) const {
890 return (r == _failure_reason.get()) ||
891 (r != nullptr &&
892 _failure_reason.get() != nullptr &&
893 strcmp(r, _failure_reason.get()) == 0);
894 }
895
896 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
897 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
898 env()->record_method_not_compilable(reason);
899 // Record failure reason.
900 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
901 }
902 bool check_node_count(uint margin, const char* reason) {
903 if (oom()) {
904 record_method_not_compilable_oom();
905 return true;
906 }
907 if (live_nodes() + margin > max_node_limit()) {
908 record_method_not_compilable(reason);
909 return true;
910 } else {
911 return false;
912 }
913 }
914 bool oom() const { return _oom; }
915 void set_oom() { _oom = true; }
916
917 // Node management
918 uint unique() const { return _unique; }
919 uint next_unique() { return _unique++; }
920 void set_unique(uint i) { _unique = i; }
921 Arena* node_arena() { return _node_arena; }
922 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
923 RootNode* root() const { return _root; }
924 void set_root(RootNode* r) { _root = r; }
925 StartNode* start() const; // (Derived from root.)
926 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
927 Node* immutable_memory();
928
929 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
930 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
931 void set_recent_alloc(Node* ctl, Node* obj) {
932 _recent_alloc_ctl = ctl;
933 _recent_alloc_obj = obj;
934 }
935 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
936 _dead_node_count++;
937 }
938 void reset_dead_node_list() { _dead_node_list.reset();
939 _dead_node_count = 0;
940 }
941 uint live_nodes() const {
942 int val = _unique - _dead_node_count;
943 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
944 return (uint) val;
945 }
946 #ifdef ASSERT
947 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
948 bool phase_optimize_finished() const { return _phase_optimize_finished; }
949 uint count_live_nodes_by_graph_walk();
950 void print_missing_nodes();
951 #endif
952
953 // Record modified nodes to check that they are put on IGVN worklist
954 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
955 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
956 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
957
958 MachConstantBaseNode* mach_constant_base_node();
959 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
960 // Generated by adlc, true if CallNode requires MachConstantBase.
961 bool needs_deep_clone_jvms();
962
963 // Handy undefined Node
964 Node* top() const { return _top; }
965
966 // these are used by guys who need to know about creation and transformation of top:
967 Node* cached_top_node() { return _top; }
968 void set_cached_top_node(Node* tn);
969
970 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
971 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
972 Node_Notes* default_node_notes() const { return _default_node_notes; }
973 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
974
975 Node_Notes* node_notes_at(int idx);
976
977 inline bool set_node_notes_at(int idx, Node_Notes* value);
978 // Copy notes from source to dest, if they exist.
979 // Overwrite dest only if source provides something.
980 // Return true if information was moved.
981 bool copy_node_notes_to(Node* dest, Node* source);
982
983 // Workhorse function to sort out the blocked Node_Notes array:
984 Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
985 int idx, bool can_grow = false);
986
987 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
988
989 // Type management
990 Arena* type_arena() { return _type_arena; }
991 Dict* type_dict() { return _type_dict; }
992 size_t type_last_size() { return _type_last_size; }
993 int num_alias_types() { return _num_alias_types; }
994
995 void init_type_arena() { _type_arena = &_Compile_types; }
996 void set_type_arena(Arena* a) { _type_arena = a; }
997 void set_type_dict(Dict* d) { _type_dict = d; }
998 void set_type_last_size(size_t sz) { _type_last_size = sz; }
999
1000 const TypeFunc* last_tf(ciMethod* m) {
1001 return (m == _last_tf_m) ? _last_tf : nullptr;
1002 }
1003 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
1004 assert(m != nullptr || tf == nullptr, "");
1005 _last_tf_m = m;
1006 _last_tf = tf;
1007 }
1008
1009 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
1010 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
1011 bool have_alias_type(const TypePtr* adr_type);
1012 AliasType* alias_type(ciField* field);
1013
1014 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, nullptr, uncached)->index(); }
1015 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
1016 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
1017
1018 // Building nodes
1019 void rethrow_exceptions(JVMState* jvms);
1020 void return_values(JVMState* jvms);
1021 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1022
1023 // Decide how to build a call.
1024 // The profile factor is a discount to apply to this site's interp. profile.
1025 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1026 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
1027 bool allow_intrinsics = true);
1028 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1029 return C->directive()->should_delay_inline(call_method) ||
1030 should_delay_string_inlining(call_method, jvms) ||
1031 should_delay_boxing_inlining(call_method, jvms) ||
1032 should_delay_vector_inlining(call_method, jvms);
1033 }
1034 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1035 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1036 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
1037 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
1038
1039 // Helper functions to identify inlining potential at call-site
1040 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
1041 ciKlass* holder, ciMethod* callee,
1042 const TypeOopPtr* receiver_type, bool is_virtual,
1043 bool &call_does_dispatch, int &vtable_index,
1044 bool check_access = true);
1045 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1046 ciMethod* callee, const TypeOopPtr* receiver_type,
1047 bool check_access = true);
1048
1049 // Report if there were too many traps at a current method and bci.
1050 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1051 // If there is no MDO at all, report no trap unless told to assume it.
1052 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1053 // This version, unspecific to a particular bci, asks if
1054 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1055 bool too_many_traps(Deoptimization::DeoptReason reason,
1056 // Privately used parameter for logging:
1057 ciMethodData* logmd = nullptr);
1058 // Report if there were too many recompiles at a method and bci.
1059 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1060 // Report if there were too many traps or recompiles at a method and bci.
1061 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1062 return too_many_traps(method, bci, reason) ||
1063 too_many_recompiles(method, bci, reason);
1064 }
1065 // Return a bitset with the reasons where deoptimization is allowed,
1066 // i.e., where there were not too many uncommon traps.
1067 int _allowed_reasons;
1068 int allowed_deopt_reasons() { return _allowed_reasons; }
1069 void set_allowed_deopt_reasons();
1070
1071 // Parsing, optimization
1072 PhaseGVN* initial_gvn() { return _initial_gvn; }
1073 Unique_Node_List* igvn_worklist() {
1074 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1075 return _igvn_worklist;
1076 }
1077 Type_Array* types() {
1078 assert(_types != nullptr, "must be created in Compile::Compile");
1079 return _types;
1080 }
1081 NodeHash* node_hash() {
1082 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1083 return _node_hash;
1084 }
1085 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1086 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1087 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1088
1089 // Replace n by nn using initial_gvn, calling hash_delete and
1090 // record_for_igvn as needed.
1091 void gvn_replace_by(Node* n, Node* nn);
1092
1093
1094 void identify_useful_nodes(Unique_Node_List &useful);
1095 void update_dead_node_list(Unique_Node_List &useful);
1096 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist, const Unique_Node_List* root_and_safepoints = nullptr);
1097
1098 void remove_useless_node(Node* dead);
1099
1100 // Record this CallGenerator for inlining at the end of parsing.
1101 void add_late_inline(CallGenerator* cg) {
1102 _late_inlines.insert_before(_late_inlines_pos, cg);
1103 _late_inlines_pos++;
1104 }
1105
1106 void prepend_late_inline(CallGenerator* cg) {
1107 _late_inlines.insert_before(0, cg);
1108 }
1109
1110 void add_string_late_inline(CallGenerator* cg) {
1111 _string_late_inlines.push(cg);
1112 }
1113
1114 void add_boxing_late_inline(CallGenerator* cg) {
1115 _boxing_late_inlines.push(cg);
1116 }
1117
1118 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1119 _vector_reboxing_late_inlines.push(cg);
1120 }
1121
1122 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1123 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1124
1125 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1126 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1127
1128 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1129
1130 void dump_print_inlining();
1131
1132 bool over_inlining_cutoff() const {
1133 if (!inlining_incrementally()) {
1134 return unique() > (uint)NodeCountInliningCutoff;
1135 } else {
1136 // Give some room for incremental inlining algorithm to "breathe"
1137 // and avoid thrashing when live node count is close to the limit.
1138 // Keep in mind that live_nodes() isn't accurate during inlining until
1139 // dead node elimination step happens (see Compile::inline_incrementally).
1140 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1141 }
1142 }
1143
1144 void mark_has_mh_late_inlines() { _has_mh_late_inlines = true; }
1145 bool has_mh_late_inlines() const { return _has_mh_late_inlines; }
1146
1147 bool inline_incrementally_one();
1148 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1149 void inline_incrementally(PhaseIterGVN& igvn);
1150 bool should_stress_inlining() { return StressIncrementalInlining && (random() % 2) == 0; }
1151 bool should_delay_inlining() { return AlwaysIncrementalInline || should_stress_inlining(); }
1152 void inline_string_calls(bool parse_time);
1153 void inline_boxing_calls(PhaseIterGVN& igvn);
1154 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1155 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1156
1157 void inline_vector_reboxing_calls();
1158 bool has_vbox_nodes();
1159
1160 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1161
1162 // Matching, CFG layout, allocation, code generation
1163 PhaseCFG* cfg() { return _cfg; }
1164 bool has_java_calls() const { return _java_calls > 0; }
1165 int java_calls() const { return _java_calls; }
1166 int inner_loops() const { return _inner_loops; }
1167 Matcher* matcher() { return _matcher; }
1168 PhaseRegAlloc* regalloc() { return _regalloc; }
1169 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1170 ResourceArea* regmask_arena() { return &_regmask_arena; }
1171 Arena* indexSet_arena() { return _indexSet_arena; }
1172 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1173 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1174
1175 void update_interpreter_frame_size(int size) {
1176 if (_interpreter_frame_size < size) {
1177 _interpreter_frame_size = size;
1178 }
1179 }
1180
1181 void set_matcher(Matcher* m) { _matcher = m; }
1182 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1183 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1184 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1185
1186 void set_java_calls(int z) { _java_calls = z; }
1187 void set_inner_loops(int z) { _inner_loops = z; }
1188
1189 Dependencies* dependencies() { return env()->dependencies(); }
1190
1191 // Major entry point. Given a Scope, compile the associated method.
1192 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1193 // replacement, entry_bci indicates the bytecode for which to compile a
1194 // continuation.
1195 Compile(ciEnv* ci_env, ciMethod* target,
1196 int entry_bci, Options options, DirectiveSet* directive);
1197
1198 // Second major entry point. From the TypeFunc signature, generate code
1199 // to pass arguments from the Java calling convention to the C calling
1200 // convention.
1201 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1202 address stub_function, const char *stub_name,
1203 StubId stub_id, int is_fancy_jump, bool pass_tls,
1204 bool return_pc, DirectiveSet* directive);
1205
1206 ~Compile();
1207
1208 // Are we compiling a method?
1209 bool has_method() { return method() != nullptr; }
1210
1211 // Maybe print some information about this compile.
1212 void print_compile_messages();
1213
1214 // Final graph reshaping, a post-pass after the regular optimizer is done.
1215 bool final_graph_reshaping();
1216
1217 // returns true if adr is completely contained in the given alias category
1218 bool must_alias(const TypePtr* adr, int alias_idx);
1219
1220 // returns true if adr overlaps with the given alias category
1221 bool can_alias(const TypePtr* adr, int alias_idx);
1222
1223 // Stack slots that may be unused by the calling convention but must
1224 // otherwise be preserved. On Intel this includes the return address.
1225 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1226 uint in_preserve_stack_slots() {
1227 return SharedRuntime::in_preserve_stack_slots();
1228 }
1229
1230 // "Top of Stack" slots that may be unused by the calling convention but must
1231 // otherwise be preserved.
1232 // On Intel these are not necessary and the value can be zero.
1233 static uint out_preserve_stack_slots() {
1234 return SharedRuntime::out_preserve_stack_slots();
1235 }
1236
1237 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1238 // for calls to C. Supports the var-args backing area for register parms.
1239 uint varargs_C_out_slots_killed() const;
1240
1241 // Number of Stack Slots consumed by a synchronization entry
1242 int sync_stack_slots() const;
1243
1244 // Compute the name of old_SP. See <arch>.ad for frame layout.
1245 OptoReg::Name compute_old_SP();
1246
1247 private:
1248 // Phase control:
1249 void Init(bool aliasing); // Prepare for a single compilation
1250 void Optimize(); // Given a graph, optimize it
1251 void Code_Gen(); // Generate code from a graph
1252
1253 // Management of the AliasType table.
1254 void grow_alias_types();
1255 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1256 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1257 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1258
1259 void verify_top(Node*) const PRODUCT_RETURN;
1260
1261 // Intrinsic setup.
1262 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1263 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1264 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1265 void register_intrinsic(CallGenerator* cg); // update fn
1266
1267 #ifndef PRODUCT
1268 static juint _intrinsic_hist_count[];
1269 static jubyte _intrinsic_hist_flags[];
1270 #endif
1271 // Function calls made by the public function final_graph_reshaping.
1272 // No need to be made public as they are not called elsewhere.
1273 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1274 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1275 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1276 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1277
1278 // Logic cone optimization.
1279 void optimize_logic_cones(PhaseIterGVN &igvn);
1280 void collect_logic_cone_roots(Unique_Node_List& list);
1281 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1282 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1283 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1284 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1285 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1286 void check_no_dead_use() const NOT_DEBUG_RETURN;
1287
1288 public:
1289
1290 // Note: Histogram array size is about 1 Kb.
1291 enum { // flag bits:
1292 _intrinsic_worked = 1, // succeeded at least once
1293 _intrinsic_failed = 2, // tried it but it failed
1294 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1295 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1296 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1297 };
1298 // Update histogram. Return boolean if this is a first-time occurrence.
1299 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1300 bool is_virtual, int flags) PRODUCT_RETURN0;
1301 static void print_intrinsic_statistics() PRODUCT_RETURN;
1302
1303 // Graph verification code
1304 // Walk the node list, verifying that there is a one-to-one correspondence
1305 // between Use-Def edges and Def-Use edges. The option no_dead_code enables
1306 // stronger checks that the graph is strongly connected from starting points
1307 // in both directions.
1308 // root_and_safepoints is used to give the starting points for the traversal.
1309 // If not supplied, only root is used. When this check is called after CCP,
1310 // we need to start traversal from Root and safepoints, just like CCP does its
1311 // own traversal (see PhaseCCP::transform for reasons).
1312 //
1313 // To call this function, there are 2 ways to go:
1314 // - give root_and_safepoints to start traversal everywhere needed (like after CCP)
1315 // - if the whole graph is assumed to be reachable from Root's input,
1316 // root_and_safepoints is not needed (like in PhaseRemoveUseless).
1317 //
1318 // Failure to specify root_and_safepoints in case the graph is not fully
1319 // reachable from Root's input make this check unsound (can miss inconsistencies)
1320 // and even incomplete (can make up non-existing problems) if no_dead_code is
1321 // true.
1322 void verify_graph_edges(bool no_dead_code = false, const Unique_Node_List* root_and_safepoints = nullptr) const PRODUCT_RETURN;
1323
1324 // Verify bi-directional correspondence of edges
1325 void verify_bidirectional_edges(Unique_Node_List& visited, const Unique_Node_List* root_and_safepoints = nullptr) const;
1326
1327 // End-of-run dumps.
1328 static void print_statistics() PRODUCT_RETURN;
1329
1330 // Verify ADLC assumptions during startup
1331 static void adlc_verification() PRODUCT_RETURN;
1332
1333 // Definitions of pd methods
1334 static void pd_compiler2_init();
1335
1336 // Static parse-time type checking logic for gen_subtype_check:
1337 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1338 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1339
1340 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1341 // Optional control dependency (for example, on range check)
1342 Node* ctrl = nullptr);
1343
1344 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1345 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1346
1347 // Auxiliary method for randomized fuzzing/stressing
1348 int random();
1349 bool randomized_select(int count);
1350
1351 // seed random number generation and log the seed for repeatability.
1352 void initialize_stress_seed(const DirectiveSet* directive);
1353
1354 // supporting clone_map
1355 CloneMap& clone_map();
1356 void set_clone_map(Dict* d);
1357
1358 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1359 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1360 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1361
1362 #ifdef ASSERT
1363 VerifyMeetResult* _type_verify;
1364 void set_exception_backedge() { _exception_backedge = true; }
1365 bool has_exception_backedge() const { return _exception_backedge; }
1366 #endif
1367
1368 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1369 BasicType out_bt, BasicType in_bt);
1370
1371 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1372
1373 #ifndef PRODUCT
1374 private:
1375 // getting rid of the template makes things easier
1376 Node* make_debug_print_call(const char* str, address call_addr, PhaseGVN* gvn,
1377 Node* parm0 = nullptr, Node* parm1 = nullptr,
1378 Node* parm2 = nullptr, Node* parm3 = nullptr,
1379 Node* parm4 = nullptr, Node* parm5 = nullptr,
1380 Node* parm6 = nullptr) const;
1381
1382 public:
1383 // Creates a CallLeafNode for a runtime call that prints a static string and the values of the
1384 // nodes passed as arguments.
1385 // This function also takes care of doing the necessary wiring, including finding a suitable control
1386 // based on the nodes that need to be printed. Note that passing nodes that have incompatible controls
1387 // is undefined behavior.
1388 template <typename... TT, typename... NN>
1389 Node* make_debug_print(const char* str, PhaseGVN* gvn, NN... in) {
1390 address call_addr = CAST_FROM_FN_PTR(address, SharedRuntime::debug_print<TT...>);
1391 return make_debug_print_call(str, call_addr, gvn, in...);
1392 }
1393 #endif
1394 };
1395
1396 #endif // SHARE_OPTO_COMPILE_HPP