1 /*
2 * Copyright (c) 1997, 2024, 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/compiler_globals.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compilerEvent.hpp"
34 #include "compiler/cHeapStringHolder.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/phasetype.hpp"
41 #include "opto/phase.hpp"
42 #include "opto/regmask.hpp"
43 #include "runtime/deoptimization.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/timerTrace.hpp"
46 #include "runtime/vmThread.hpp"
47 #include "utilities/ticks.hpp"
48 #include "utilities/vmEnums.hpp"
49
50 class AbstractLockNode;
51 class AddPNode;
52 class Block;
53 class Bundle;
54 class CallGenerator;
55 class CallNode;
56 class CallStaticJavaNode;
57 class CloneMap;
58 class CompilationFailureInfo;
59 class ConnectionGraph;
60 class IdealGraphPrinter;
61 class InlineTree;
62 class Matcher;
63 class MachConstantNode;
64 class MachConstantBaseNode;
65 class MachNode;
66 class MachOper;
67 class MachSafePointNode;
68 class Node;
69 class Node_Array;
70 class Node_List;
71 class Node_Notes;
72 class NodeHash;
73 class NodeCloneInfo;
74 class OptoReg;
75 class ParsePredicateNode;
76 class PhaseCFG;
77 class PhaseGVN;
78 class PhaseIterGVN;
79 class PhaseRegAlloc;
80 class PhaseCCP;
81 class PhaseOutput;
82 class RootNode;
83 class relocInfo;
84 class StartNode;
85 class SafePointNode;
86 class JVMState;
87 class Type;
88 class TypeInt;
89 class TypeInteger;
90 class TypeKlassPtr;
91 class TypePtr;
92 class TypeOopPtr;
93 class TypeFunc;
94 class TypeVect;
95 class Type_Array;
96 class Unique_Node_List;
97 class UnstableIfTrap;
98 class InlineTypeNode;
99 class nmethod;
100 class Node_Stack;
101 struct Final_Reshape_Counts;
102 class VerifyMeetResult;
103
104 enum LoopOptsMode {
105 LoopOptsDefault,
106 LoopOptsNone,
107 LoopOptsMaxUnroll,
108 LoopOptsShenandoahExpand,
109 LoopOptsShenandoahPostExpand,
110 LoopOptsSkipSplitIf,
111 LoopOptsVerify
112 };
113
114 // The type of all node counts and indexes.
115 // It must hold at least 16 bits, but must also be fast to load and store.
116 // This type, if less than 32 bits, could limit the number of possible nodes.
117 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
118 typedef unsigned int node_idx_t;
119
120 class NodeCloneInfo {
121 private:
122 uint64_t _idx_clone_orig;
123 public:
124
125 void set_idx(node_idx_t idx) {
126 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
127 }
128 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
129
130 void set_gen(int generation) {
131 uint64_t g = (uint64_t)generation << 32;
132 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
133 }
134 int gen() const { return (int)(_idx_clone_orig >> 32); }
135
136 void set(uint64_t x) { _idx_clone_orig = x; }
137 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
138 uint64_t get() const { return _idx_clone_orig; }
139
140 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
141 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
142
143 void dump_on(outputStream* st) const;
144 };
145
146 class CloneMap {
147 friend class Compile;
148 private:
149 bool _debug;
150 Dict* _dict;
151 int _clone_idx; // current cloning iteration/generation in loop unroll
152 public:
153 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
154 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
155 Dict* dict() const { return _dict; }
156 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
157 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
158 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
159 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
160 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
161 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
162 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
163 int max_gen() const;
164 void clone(Node* old, Node* nnn, int gen);
165 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
166 void dump(node_idx_t key, outputStream* st) const;
167
168 int clone_idx() const { return _clone_idx; }
169 void set_clone_idx(int x) { _clone_idx = x; }
170 bool is_debug() const { return _debug; }
171 void set_debug(bool debug) { _debug = debug; }
172
173 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
174 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
175 };
176
177 class Options {
178 friend class Compile;
179 friend class VMStructs;
180 private:
181 const bool _subsume_loads; // Load can be matched as part of a larger op.
182 const bool _do_escape_analysis; // Do escape analysis.
183 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
184 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges.
185 const bool _eliminate_boxing; // Do boxing elimination.
186 const bool _do_locks_coarsening; // Do locks coarsening
187 const bool _do_superword; // Do SuperWord
188 const bool _install_code; // Install the code that was compiled
189 public:
190 Options(bool subsume_loads,
191 bool do_escape_analysis,
192 bool do_iterative_escape_analysis,
193 bool do_reduce_allocation_merges,
194 bool eliminate_boxing,
195 bool do_locks_coarsening,
196 bool do_superword,
197 bool install_code) :
198 _subsume_loads(subsume_loads),
199 _do_escape_analysis(do_escape_analysis),
200 _do_iterative_escape_analysis(do_iterative_escape_analysis),
201 _do_reduce_allocation_merges(do_reduce_allocation_merges),
202 _eliminate_boxing(eliminate_boxing),
203 _do_locks_coarsening(do_locks_coarsening),
204 _do_superword(do_superword),
205 _install_code(install_code) {
206 }
207
208 static Options for_runtime_stub() {
209 return Options(
210 /* subsume_loads = */ true,
211 /* do_escape_analysis = */ false,
212 /* do_iterative_escape_analysis = */ false,
213 /* do_reduce_allocation_merges = */ false,
214 /* eliminate_boxing = */ false,
215 /* do_lock_coarsening = */ false,
216 /* do_superword = */ true,
217 /* install_code = */ true
218 );
219 }
220 };
221
222 //------------------------------Compile----------------------------------------
223 // This class defines a top-level Compiler invocation.
224
225 class Compile : public Phase {
226 friend class VMStructs;
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* _compile;
243 CompileLog* _log;
244 const char* _phase_name;
245 bool _dolog;
246 public:
247 TracePhase(const char* name, elapsedTimer* accumulator);
248 ~TracePhase();
249 };
250
251 // Information per category of alias (memory slice)
252 class AliasType {
253 private:
254 friend class Compile;
255
256 int _index; // unique index, used with MergeMemNode
257 const TypePtr* _adr_type; // normalized address type
258 ciField* _field; // relevant instance field, or null if none
259 const Type* _element; // relevant array element type, or null if none
260 bool _is_rewritable; // false if the memory is write-once only
261 int _general_index; // if this is type is an instance, the general
262 // type that this is an instance of
263
264 void Init(int i, const TypePtr* at);
265
266 public:
267 int index() const { return _index; }
268 const TypePtr* adr_type() const { return _adr_type; }
269 ciField* field() const { return _field; }
270 const Type* element() const { return _element; }
271 bool is_rewritable() const { return _is_rewritable; }
272 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
273 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
274
275 void set_rewritable(bool z) { _is_rewritable = z; }
276 void set_field(ciField* f) {
277 assert(!_field,"");
278 _field = f;
279 if (f->is_final() || f->is_stable()) {
280 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
281 _is_rewritable = false;
282 }
283 }
284 void set_element(const Type* e) {
285 assert(_element == nullptr, "");
286 _element = e;
287 }
288
289 BasicType basic_type() const;
290
291 void print_on(outputStream* st) PRODUCT_RETURN;
292 };
293
294 enum {
295 logAliasCacheSize = 6,
296 AliasCacheSize = (1<<logAliasCacheSize)
297 };
298 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
299 enum {
300 trapHistLength = MethodData::_trap_hist_limit
301 };
302
303 private:
304 // Fixed parameters to this compilation.
305 const int _compile_id;
306 const Options _options; // Compilation options
307 ciMethod* _method; // The method being compiled.
308 int _entry_bci; // entry bci for osr methods.
309 const TypeFunc* _tf; // My kind of signature
310 InlineTree* _ilt; // Ditto (temporary).
311 address _stub_function; // VM entry for stub being compiled, or null
312 const char* _stub_name; // Name of stub or adapter being compiled, or null
313 address _stub_entry_point; // Compile code entry for generated stub, or null
314
315 // Control of this compilation.
316 int _max_inline_size; // Max inline size for this compilation
317 int _freq_inline_size; // Max hot method inline size for this compilation
318 int _fixed_slots; // count of frame slots not allocated by the register
319 // allocator i.e. locks, original deopt pc, etc.
320 uintx _max_node_limit; // Max unique node count during a single compilation.
321
322 bool _post_loop_opts_phase; // Loop opts are finished.
323 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
324
325 int _major_progress; // Count of something big happening
326 bool _inlining_progress; // progress doing incremental inlining?
327 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
328 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
329 bool _has_loops; // True if the method _may_ have some loops
330 bool _has_split_ifs; // True if the method _may_ have some split-if
331 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
332 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
333 bool _has_boxed_value; // True if a boxed object is allocated
334 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
335 bool _has_circular_inline_type; // True if method loads an inline type with a circular, non-flat field
336 uint _max_vector_size; // Maximum size of generated vectors
337 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
338 uint _trap_hist[trapHistLength]; // Cumulative traps
339 bool _trap_can_recompile; // Have we emitted a recompiling trap?
340 uint _decompile_count; // Cumulative decompilation counts.
341 bool _do_inlining; // True if we intend to do inlining
342 bool _do_scheduling; // True if we intend to do scheduling
343 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
344 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
345 bool _use_cmove; // True if CMove should be used without profitability analysis
346 bool _do_aliasing; // True if we intend to do aliasing
347 bool _print_assembly; // True if we should dump assembly code for this compilation
348 bool _print_inlining; // True if we should print inlining for this compilation
349 bool _print_intrinsics; // True if we should print intrinsics for this compilation
350 #ifndef PRODUCT
351 uint _igv_idx; // Counter for IGV node identifiers
352 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
353 bool _trace_opto_output;
354 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
355 #endif
356 bool _has_irreducible_loop; // Found irreducible loops
357 // JSR 292
358 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
359 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
360 bool _has_scoped_access; // For shared scope closure
361 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
362 int _loop_opts_cnt; // loop opts round
363 bool _has_flat_accesses; // Any known flat array accesses?
364 bool _flat_accesses_share_alias; // Initially all flat array share a single slice
365 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
366 uint _stress_seed; // Seed for stress testing
367
368 // Compilation environment.
369 Arena _comp_arena; // Arena with lifetime equivalent to Compile
370 void* _barrier_set_state; // Potential GC barrier state for Compile
371 ciEnv* _env; // CI interface
372 DirectiveSet* _directive; // Compiler directive
373 CompileLog* _log; // from CompilerThread
374 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
375 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
376 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
377 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
378 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
379 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates.
380 GrowableArray<Node*> _template_assertion_predicate_opaqs;
381 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
382 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
383 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
384 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
385 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
386 ConnectionGraph* _congraph;
387 #ifndef PRODUCT
388 IdealGraphPrinter* _igv_printer;
389 static IdealGraphPrinter* _debug_file_printer;
390 static IdealGraphPrinter* _debug_network_printer;
391 #endif
392
393
394 // Node management
395 uint _unique; // Counter for unique Node indices
396 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
397 // So use this to keep count and make the call O(1).
398 VectorSet _dead_node_list; // Set of dead nodes
399 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
400 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
401
402 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
403
404 // Arenas for new-space and old-space nodes.
405 // Swapped between using _node_arena.
406 // The lifetime of the old-space nodes is during xform.
407 Arena _node_arena_one;
408 Arena _node_arena_two;
409 Arena* _node_arena;
410 public:
411 Arena* swap_old_and_new() {
412 Arena* filled_arena_ptr = _node_arena;
413 Arena* old_arena_ptr = old_arena();
414 old_arena_ptr->destruct_contents();
415 _node_arena = old_arena_ptr;
416 return filled_arena_ptr;
417 }
418 private:
419 RootNode* _root; // Unique root of compilation, or null after bail-out.
420 Node* _top; // Unique top node. (Reset by various phases.)
421
422 Node* _immutable_memory; // Initial memory state
423
424 Node* _recent_alloc_obj;
425 Node* _recent_alloc_ctl;
426
427 // Constant table
428 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
429
430
431 // Blocked array of debugging and profiling information,
432 // tracked per node.
433 enum { _log2_node_notes_block_size = 8,
434 _node_notes_block_size = (1<<_log2_node_notes_block_size)
435 };
436 GrowableArray<Node_Notes*>* _node_note_array;
437 Node_Notes* _default_node_notes; // default notes for new nodes
438
439 // After parsing and every bulk phase we hang onto the Root instruction.
440 // The RootNode instruction is where the whole program begins. It produces
441 // the initial Control and BOTTOM for everybody else.
442
443 // Type management
444 Arena _Compile_types; // Arena for all types
445 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
446 Dict* _type_dict; // Intern table
447 CloneMap _clone_map; // used for recording history of cloned nodes
448 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
449 ciMethod* _last_tf_m; // Cache for
450 const TypeFunc* _last_tf; // TypeFunc::make
451 AliasType** _alias_types; // List of alias types seen so far.
452 int _num_alias_types; // Logical length of _alias_types
453 int _max_alias_types; // Physical length of _alias_types
454 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
455
456 // Parsing, optimization
457 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
458
459 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
460 // If pushed outside IGVN, the Node is processed in the next IGVN round.
461 Unique_Node_List* _igvn_worklist;
462
463 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
464 Type_Array* _types;
465
466 // Shared node hash table for GVN, IGVN and CCP.
467 NodeHash* _node_hash;
468
469 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
470 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
471 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
472
473 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
474
475 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
476 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
477
478 // "MemLimit" directive was specified and the memory limit was hit during compilation
479 bool _oom;
480
481 // Inlining may not happen in parse order which would make
482 // PrintInlining output confusing. Keep track of PrintInlining
483 // pieces in order.
484 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
485 private:
486 CallGenerator* _cg;
487 stringStream _ss;
488 static const size_t default_stream_buffer_size = 128;
489
490 public:
491 PrintInliningBuffer()
492 : _cg(nullptr), _ss(default_stream_buffer_size) {}
493
494 stringStream* ss() { return &_ss; }
495 CallGenerator* cg() { return _cg; }
496 void set_cg(CallGenerator* cg) { _cg = cg; }
497 };
498
499 stringStream* _print_inlining_stream;
500 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
501 int _print_inlining_idx;
502 char* _print_inlining_output;
503
504 // Only keep nodes in the expensive node list that need to be optimized
505 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
506 // Use for sorting expensive nodes to bring similar nodes together
507 static int cmp_expensive_nodes(Node** n1, Node** n2);
508 // Expensive nodes list already sorted?
509 bool expensive_nodes_sorted() const;
510 // Remove the speculative part of types and clean up the graph
511 void remove_speculative_types(PhaseIterGVN &igvn);
512
513 void* _replay_inline_data; // Pointer to data loaded from file
514
515 void print_inlining_init();
516 void print_inlining_reinit();
517 void print_inlining_commit();
518 void print_inlining_push();
519 PrintInliningBuffer* print_inlining_current();
520
521 void log_late_inline_failure(CallGenerator* cg, const char* msg);
522 DEBUG_ONLY(bool _exception_backedge;)
523
524 void record_method_not_compilable_oom();
525
526 public:
527
528 void* barrier_set_state() const { return _barrier_set_state; }
529
530 stringStream* print_inlining_stream() {
531 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
532 return _print_inlining_stream;
533 }
534
535 void print_inlining_update(CallGenerator* cg);
536 void print_inlining_update_delayed(CallGenerator* cg);
537 void print_inlining_move_to(CallGenerator* cg);
538 void print_inlining_assert_ready();
539 void print_inlining_reset();
540
541 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) {
542 stringStream ss;
543 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg);
544 print_inlining_stream()->print("%s", ss.freeze());
545 }
546
547 #ifndef PRODUCT
548 IdealGraphPrinter* igv_printer() { return _igv_printer; }
549 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
550 #endif
551
552 void log_late_inline(CallGenerator* cg);
553 void log_inline_id(CallGenerator* cg);
554 void log_inline_failure(const char* msg);
555
556 void* replay_inline_data() const { return _replay_inline_data; }
557
558 // Dump inlining replay data to the stream.
559 void dump_inline_data(outputStream* out);
560 void dump_inline_data_reduced(outputStream* out);
561
562 private:
563 // Matching, CFG layout, allocation, code generation
564 PhaseCFG* _cfg; // Results of CFG finding
565 int _java_calls; // Number of java calls in the method
566 int _inner_loops; // Number of inner loops in the method
567 Matcher* _matcher; // Engine to map ideal to machine instructions
568 PhaseRegAlloc* _regalloc; // Results of register allocation.
569 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
570 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
571 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
572 int _interpreter_frame_size;
573
574 PhaseOutput* _output;
575
576 public:
577 // Accessors
578
579 // The Compile instance currently active in this (compiler) thread.
580 static Compile* current() {
581 return (Compile*) ciEnv::current()->compiler_data();
582 }
583
584 int interpreter_frame_size() const { return _interpreter_frame_size; }
585
586 PhaseOutput* output() const { return _output; }
587 void set_output(PhaseOutput* o) { _output = o; }
588
589 // ID for this compilation. Useful for setting breakpoints in the debugger.
590 int compile_id() const { return _compile_id; }
591 DirectiveSet* directive() const { return _directive; }
592
593 // Does this compilation allow instructions to subsume loads? User
594 // instructions that subsume a load may result in an unschedulable
595 // instruction sequence.
596 bool subsume_loads() const { return _options._subsume_loads; }
597 /** Do escape analysis. */
598 bool do_escape_analysis() const { return _options._do_escape_analysis; }
599 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
600 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
601 /** Do boxing elimination. */
602 bool eliminate_boxing() const { return _options._eliminate_boxing; }
603 /** Do aggressive boxing elimination. */
604 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
605 bool should_install_code() const { return _options._install_code; }
606 /** Do locks coarsening. */
607 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
608 bool do_superword() const { return _options._do_superword; }
609
610 // Other fixed compilation parameters.
611 ciMethod* method() const { return _method; }
612 int entry_bci() const { return _entry_bci; }
613 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
614 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
615 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
616 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
617 InlineTree* ilt() const { return _ilt; }
618 address stub_function() const { return _stub_function; }
619 const char* stub_name() const { return _stub_name; }
620 address stub_entry_point() const { return _stub_entry_point; }
621 void set_stub_entry_point(address z) { _stub_entry_point = z; }
622
623 // Control of this compilation.
624 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
625 void set_fixed_slots(int n) { _fixed_slots = n; }
626 int major_progress() const { return _major_progress; }
627 void set_inlining_progress(bool z) { _inlining_progress = z; }
628 int inlining_progress() const { return _inlining_progress; }
629 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
630 int inlining_incrementally() const { return _inlining_incrementally; }
631 void set_do_cleanup(bool z) { _do_cleanup = z; }
632 int do_cleanup() const { return _do_cleanup; }
633 void set_major_progress() { _major_progress++; }
634 void restore_major_progress(int progress) { _major_progress += progress; }
635 void clear_major_progress() { _major_progress = 0; }
636 int max_inline_size() const { return _max_inline_size; }
637 void set_freq_inline_size(int n) { _freq_inline_size = n; }
638 int freq_inline_size() const { return _freq_inline_size; }
639 void set_max_inline_size(int n) { _max_inline_size = n; }
640 bool has_loops() const { return _has_loops; }
641 void set_has_loops(bool z) { _has_loops = z; }
642 bool has_split_ifs() const { return _has_split_ifs; }
643 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
644 bool has_unsafe_access() const { return _has_unsafe_access; }
645 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
646 bool has_stringbuilder() const { return _has_stringbuilder; }
647 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
648 bool has_boxed_value() const { return _has_boxed_value; }
649 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
650 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
651 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
652 bool has_circular_inline_type() const { return _has_circular_inline_type; }
653 void set_has_circular_inline_type(bool z) { _has_circular_inline_type = z; }
654 uint max_vector_size() const { return _max_vector_size; }
655 void set_max_vector_size(uint s) { _max_vector_size = s; }
656 bool clear_upper_avx() const { return _clear_upper_avx; }
657 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
658 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
659 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
660 bool trap_can_recompile() const { return _trap_can_recompile; }
661 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
662 uint decompile_count() const { return _decompile_count; }
663 void set_decompile_count(uint c) { _decompile_count = c; }
664 bool allow_range_check_smearing() const;
665 bool do_inlining() const { return _do_inlining; }
666 void set_do_inlining(bool z) { _do_inlining = z; }
667 bool do_scheduling() const { return _do_scheduling; }
668 void set_do_scheduling(bool z) { _do_scheduling = z; }
669 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
670 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
671 bool do_vector_loop() const { return _do_vector_loop; }
672 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
673 bool use_cmove() const { return _use_cmove; }
674 void set_use_cmove(bool z) { _use_cmove = z; }
675 bool do_aliasing() const { return _do_aliasing; }
676 bool print_assembly() const { return _print_assembly; }
677 void set_print_assembly(bool z) { _print_assembly = z; }
678 bool print_inlining() const { return _print_inlining; }
679 void set_print_inlining(bool z) { _print_inlining = z; }
680 bool print_intrinsics() const { return _print_intrinsics; }
681 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
682 uint max_node_limit() const { return (uint)_max_node_limit; }
683 void set_max_node_limit(uint n) { _max_node_limit = n; }
684 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
685 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
686 void set_flat_accesses() { _has_flat_accesses = true; }
687 bool flat_accesses_share_alias() const { return _flat_accesses_share_alias; }
688 void set_flat_accesses_share_alias(bool z) { _flat_accesses_share_alias = z; }
689 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
690 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
691
692 // Support for scalarized inline type calling convention
693 bool has_scalarized_args() const { return _method != nullptr && _method->has_scalarized_args(); }
694 bool needs_stack_repair() const { return _method != nullptr && _method->get_Method()->c2_needs_stack_repair(); }
695
696 bool has_monitors() const { return _has_monitors; }
697 void set_has_monitors(bool v) { _has_monitors = v; }
698 bool has_scoped_access() const { return _has_scoped_access; }
699 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
700
701 // check the CompilerOracle for special behaviours for this compile
702 bool method_has_option(CompileCommandEnum option) {
703 return method() != nullptr && method()->has_option(option);
704 }
705
706 #ifndef PRODUCT
707 uint next_igv_idx() { return _igv_idx++; }
708 bool trace_opto_output() const { return _trace_opto_output; }
709 void print_ideal_ir(const char* phase_name);
710 bool should_print_ideal() const { return _directive->PrintIdealOption; }
711 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
712 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
713 int _in_dump_cnt; // Required for dumping ir nodes.
714 #endif
715 bool has_irreducible_loop() const { return _has_irreducible_loop; }
716 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
717
718 // JSR 292
719 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
720 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
721
722 Ticks _latest_stage_start_counter;
723
724 void begin_method();
725 void end_method();
726 bool should_print_igv(int level);
727 bool should_print_phase(CompilerPhaseType cpt);
728
729 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
730
731 #ifndef PRODUCT
732 void dump_igv(const char* graph_name, int level = 3) {
733 if (should_print_igv(level)) {
734 _igv_printer->print_method(graph_name, level);
735 }
736 }
737
738 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
739 void igv_print_method_to_network(const char* phase_name = "Debug");
740 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
741 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
742 #endif
743
744 int macro_count() const { return _macro_nodes.length(); }
745 int parse_predicate_count() const { return _parse_predicates.length(); }
746 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); }
747 int expensive_count() const { return _expensive_nodes.length(); }
748 int coarsened_count() const { return _coarsened_locks.length(); }
749
750 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
751 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); }
752
753 Node* template_assertion_predicate_opaq_node(int idx) const {
754 return _template_assertion_predicate_opaqs.at(idx);
755 }
756
757 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
758
759 ConnectionGraph* congraph() { return _congraph;}
760 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
761 void add_macro_node(Node * n) {
762 //assert(n->is_macro(), "must be a macro node");
763 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
764 _macro_nodes.append(n);
765 }
766 void remove_macro_node(Node* n) {
767 // this function may be called twice for a node so we can only remove it
768 // if it's still existing.
769 _macro_nodes.remove_if_existing(n);
770 // Remove from coarsened locks list if present
771 if (coarsened_count() > 0) {
772 remove_coarsened_lock(n);
773 }
774 }
775 void add_expensive_node(Node* n);
776 void remove_expensive_node(Node* n) {
777 _expensive_nodes.remove_if_existing(n);
778 }
779
780 void add_parse_predicate(ParsePredicateNode* n) {
781 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
782 _parse_predicates.append(n);
783 }
784
785 void remove_parse_predicate(ParsePredicateNode* n) {
786 if (parse_predicate_count() > 0) {
787 _parse_predicates.remove_if_existing(n);
788 }
789 }
790
791 void add_template_assertion_predicate_opaq(Node* n) {
792 assert(!_template_assertion_predicate_opaqs.contains(n),
793 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
794 _template_assertion_predicate_opaqs.append(n);
795 }
796
797 void remove_template_assertion_predicate_opaq(Node* n) {
798 if (template_assertion_predicate_count() > 0) {
799 _template_assertion_predicate_opaqs.remove_if_existing(n);
800 }
801 }
802 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
803 void remove_coarsened_lock(Node* n);
804 bool coarsened_locks_consistent();
805 void mark_unbalanced_boxes() const;
806
807 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
808 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
809 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
810
811 #ifdef ASSERT
812 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
813 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
814 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
815 #endif
816
817 bool allow_macro_nodes() { return _allow_macro_nodes; }
818 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
819
820 void record_for_post_loop_opts_igvn(Node* n);
821 void remove_from_post_loop_opts_igvn(Node* n);
822 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
823
824 // Keep track of inline type nodes for later processing
825 void add_inline_type(Node* n);
826 void remove_inline_type(Node* n);
827 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
828
829 void adjust_flat_array_access_aliases(PhaseIterGVN& igvn);
830
831 void record_unstable_if_trap(UnstableIfTrap* trap);
832 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
833 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
834 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
835
836 void shuffle_macro_nodes();
837 void sort_macro_nodes();
838
839 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
840
841 // Are there candidate expensive nodes for optimization?
842 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
843 // Check whether n1 and n2 are similar
844 static int cmp_expensive_nodes(Node* n1, Node* n2);
845 // Sort expensive nodes to locate similar expensive nodes
846 void sort_expensive_nodes();
847
848 // Compilation environment.
849 Arena* comp_arena() { return &_comp_arena; }
850 ciEnv* env() const { return _env; }
851 CompileLog* log() const { return _log; }
852
853 bool failing_internal() const {
854 return _env->failing() ||
855 _failure_reason.get() != nullptr;
856 }
857
858 const char* failure_reason() const {
859 return _env->failing() ? _env->failure_reason()
860 : _failure_reason.get();
861 }
862
863 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
864
865 bool failing() {
866 if (failing_internal()) {
867 return true;
868 }
869 #ifdef ASSERT
870 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
871 if (phase_verify_ideal_loop()) {
872 return false;
873 }
874 if (StressBailout) {
875 return fail_randomly();
876 }
877 #endif
878 return false;
879 }
880
881 #ifdef ASSERT
882 bool fail_randomly();
883 bool failure_is_artificial();
884 #endif
885
886 bool failure_reason_is(const char* r) const {
887 return (r == _failure_reason.get()) ||
888 (r != nullptr &&
889 _failure_reason.get() != nullptr &&
890 strcmp(r, _failure_reason.get()) == 0);
891 }
892
893 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
894 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
895 env()->record_method_not_compilable(reason);
896 // Record failure reason.
897 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
898 }
899 bool check_node_count(uint margin, const char* reason) {
900 if (oom()) {
901 record_method_not_compilable_oom();
902 return true;
903 }
904 if (live_nodes() + margin > max_node_limit()) {
905 record_method_not_compilable(reason);
906 return true;
907 } else {
908 return false;
909 }
910 }
911 bool oom() const { return _oom; }
912 void set_oom() { _oom = true; }
913
914 // Node management
915 uint unique() const { return _unique; }
916 uint next_unique() { return _unique++; }
917 void set_unique(uint i) { _unique = i; }
918 Arena* node_arena() { return _node_arena; }
919 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
920 RootNode* root() const { return _root; }
921 void set_root(RootNode* r) { _root = r; }
922 StartNode* start() const; // (Derived from root.)
923 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
924 Node* immutable_memory();
925
926 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
927 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
928 void set_recent_alloc(Node* ctl, Node* obj) {
929 _recent_alloc_ctl = ctl;
930 _recent_alloc_obj = obj;
931 }
932 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
933 _dead_node_count++;
934 }
935 void reset_dead_node_list() { _dead_node_list.reset();
936 _dead_node_count = 0;
937 }
938 uint live_nodes() const {
939 int val = _unique - _dead_node_count;
940 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
941 return (uint) val;
942 }
943 #ifdef ASSERT
944 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
945 bool phase_optimize_finished() const { return _phase_optimize_finished; }
946 uint count_live_nodes_by_graph_walk();
947 void print_missing_nodes();
948 #endif
949
950 // Record modified nodes to check that they are put on IGVN worklist
951 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
952 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
953 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
954
955 MachConstantBaseNode* mach_constant_base_node();
956 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
957 // Generated by adlc, true if CallNode requires MachConstantBase.
958 bool needs_deep_clone_jvms();
959
960 // Handy undefined Node
961 Node* top() const { return _top; }
962
963 // these are used by guys who need to know about creation and transformation of top:
964 Node* cached_top_node() { return _top; }
965 void set_cached_top_node(Node* tn);
966
967 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
968 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
969 Node_Notes* default_node_notes() const { return _default_node_notes; }
970 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
971
972 Node_Notes* node_notes_at(int idx) {
973 return locate_node_notes(_node_note_array, idx, false);
974 }
975 inline bool set_node_notes_at(int idx, Node_Notes* value);
976
977 // Copy notes from source to dest, if they exist.
978 // Overwrite dest only if source provides something.
979 // Return true if information was moved.
980 bool copy_node_notes_to(Node* dest, Node* source);
981
982 // Workhorse function to sort out the blocked Node_Notes array:
983 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
984 int idx, bool can_grow = false);
985
986 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
987
988 // Type management
989 Arena* type_arena() { return _type_arena; }
990 Dict* type_dict() { return _type_dict; }
991 size_t type_last_size() { return _type_last_size; }
992 int num_alias_types() { return _num_alias_types; }
993
994 void init_type_arena() { _type_arena = &_Compile_types; }
995 void set_type_arena(Arena* a) { _type_arena = a; }
996 void set_type_dict(Dict* d) { _type_dict = d; }
997 void set_type_last_size(size_t sz) { _type_last_size = sz; }
998
999 const TypeFunc* last_tf(ciMethod* m) {
1000 return (m == _last_tf_m) ? _last_tf : nullptr;
1001 }
1002 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
1003 assert(m != nullptr || tf == nullptr, "");
1004 _last_tf_m = m;
1005 _last_tf = tf;
1006 }
1007
1008 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
1009 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
1010 bool have_alias_type(const TypePtr* adr_type);
1011 AliasType* alias_type(ciField* field);
1012
1013 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, nullptr, uncached)->index(); }
1014 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
1015 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
1016
1017 // Building nodes
1018 void rethrow_exceptions(JVMState* jvms);
1019 void return_values(JVMState* jvms);
1020 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1021
1022 // Decide how to build a call.
1023 // The profile factor is a discount to apply to this site's interp. profile.
1024 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1025 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
1026 bool allow_intrinsics = true);
1027 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1028 return should_delay_string_inlining(call_method, jvms) ||
1029 should_delay_boxing_inlining(call_method, jvms) ||
1030 should_delay_vector_inlining(call_method, jvms);
1031 }
1032 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1033 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1034 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
1035 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
1036
1037 // Helper functions to identify inlining potential at call-site
1038 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
1039 ciKlass* holder, ciMethod* callee,
1040 const TypeOopPtr* receiver_type, bool is_virtual,
1041 bool &call_does_dispatch, int &vtable_index,
1042 bool check_access = true);
1043 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1044 ciMethod* callee, const TypeOopPtr* receiver_type,
1045 bool check_access = true);
1046
1047 // Report if there were too many traps at a current method and bci.
1048 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1049 // If there is no MDO at all, report no trap unless told to assume it.
1050 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1051 // This version, unspecific to a particular bci, asks if
1052 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1053 bool too_many_traps(Deoptimization::DeoptReason reason,
1054 // Privately used parameter for logging:
1055 ciMethodData* logmd = nullptr);
1056 // Report if there were too many recompiles at a method and bci.
1057 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1058 // Report if there were too many traps or recompiles at a method and bci.
1059 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1060 return too_many_traps(method, bci, reason) ||
1061 too_many_recompiles(method, bci, reason);
1062 }
1063 // Return a bitset with the reasons where deoptimization is allowed,
1064 // i.e., where there were not too many uncommon traps.
1065 int _allowed_reasons;
1066 int allowed_deopt_reasons() { return _allowed_reasons; }
1067 void set_allowed_deopt_reasons();
1068
1069 // Parsing, optimization
1070 PhaseGVN* initial_gvn() { return _initial_gvn; }
1071 Unique_Node_List* igvn_worklist() {
1072 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1073 return _igvn_worklist;
1074 }
1075 Type_Array* types() {
1076 assert(_types != nullptr, "must be created in Compile::Compile");
1077 return _types;
1078 }
1079 NodeHash* node_hash() {
1080 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1081 return _node_hash;
1082 }
1083 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1084 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1085 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1086
1087 // Replace n by nn using initial_gvn, calling hash_delete and
1088 // record_for_igvn as needed.
1089 void gvn_replace_by(Node* n, Node* nn);
1090
1091
1092 void identify_useful_nodes(Unique_Node_List &useful);
1093 void update_dead_node_list(Unique_Node_List &useful);
1094 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist);
1095
1096 void remove_useless_node(Node* dead);
1097
1098 // Record this CallGenerator for inlining at the end of parsing.
1099 void add_late_inline(CallGenerator* cg) {
1100 _late_inlines.insert_before(_late_inlines_pos, cg);
1101 _late_inlines_pos++;
1102 }
1103
1104 void prepend_late_inline(CallGenerator* cg) {
1105 _late_inlines.insert_before(0, cg);
1106 }
1107
1108 void add_string_late_inline(CallGenerator* cg) {
1109 _string_late_inlines.push(cg);
1110 }
1111
1112 void add_boxing_late_inline(CallGenerator* cg) {
1113 _boxing_late_inlines.push(cg);
1114 }
1115
1116 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1117 _vector_reboxing_late_inlines.push(cg);
1118 }
1119
1120 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1121 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1122
1123 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1124 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1125
1126 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1127
1128 void process_print_inlining();
1129 void dump_print_inlining();
1130
1131 bool over_inlining_cutoff() const {
1132 if (!inlining_incrementally()) {
1133 return unique() > (uint)NodeCountInliningCutoff;
1134 } else {
1135 // Give some room for incremental inlining algorithm to "breathe"
1136 // and avoid thrashing when live node count is close to the limit.
1137 // Keep in mind that live_nodes() isn't accurate during inlining until
1138 // dead node elimination step happens (see Compile::inline_incrementally).
1139 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1140 }
1141 }
1142
1143 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1144 void dec_number_of_mh_late_inlines() { assert(_number_of_mh_late_inlines > 0, "_number_of_mh_late_inlines < 0 !"); _number_of_mh_late_inlines--; }
1145 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1146
1147 bool inline_incrementally_one();
1148 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1149 void inline_incrementally(PhaseIterGVN& igvn);
1150 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); }
1151 void inline_string_calls(bool parse_time);
1152 void inline_boxing_calls(PhaseIterGVN& igvn);
1153 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1154 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1155
1156 void inline_vector_reboxing_calls();
1157 bool has_vbox_nodes();
1158
1159 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1160
1161 // Matching, CFG layout, allocation, code generation
1162 PhaseCFG* cfg() { return _cfg; }
1163 bool has_java_calls() const { return _java_calls > 0; }
1164 int java_calls() const { return _java_calls; }
1165 int inner_loops() const { return _inner_loops; }
1166 Matcher* matcher() { return _matcher; }
1167 PhaseRegAlloc* regalloc() { return _regalloc; }
1168 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1169 Arena* indexSet_arena() { return _indexSet_arena; }
1170 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1171 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1172
1173 void update_interpreter_frame_size(int size) {
1174 if (_interpreter_frame_size < size) {
1175 _interpreter_frame_size = size;
1176 }
1177 }
1178
1179 void set_matcher(Matcher* m) { _matcher = m; }
1180 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1181 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1182 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1183
1184 void set_java_calls(int z) { _java_calls = z; }
1185 void set_inner_loops(int z) { _inner_loops = z; }
1186
1187 Dependencies* dependencies() { return env()->dependencies(); }
1188
1189 // Major entry point. Given a Scope, compile the associated method.
1190 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1191 // replacement, entry_bci indicates the bytecode for which to compile a
1192 // continuation.
1193 Compile(ciEnv* ci_env, ciMethod* target,
1194 int entry_bci, Options options, DirectiveSet* directive);
1195
1196 // Second major entry point. From the TypeFunc signature, generate code
1197 // to pass arguments from the Java calling convention to the C calling
1198 // convention.
1199 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1200 address stub_function, const char *stub_name,
1201 int is_fancy_jump, bool pass_tls,
1202 bool return_pc, DirectiveSet* directive);
1203
1204 ~Compile();
1205
1206 // Are we compiling a method?
1207 bool has_method() { return method() != nullptr; }
1208
1209 // Maybe print some information about this compile.
1210 void print_compile_messages();
1211
1212 // Final graph reshaping, a post-pass after the regular optimizer is done.
1213 bool final_graph_reshaping();
1214
1215 // returns true if adr is completely contained in the given alias category
1216 bool must_alias(const TypePtr* adr, int alias_idx);
1217
1218 // returns true if adr overlaps with the given alias category
1219 bool can_alias(const TypePtr* adr, int alias_idx);
1220
1221 // Stack slots that may be unused by the calling convention but must
1222 // otherwise be preserved. On Intel this includes the return address.
1223 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1224 uint in_preserve_stack_slots() {
1225 return SharedRuntime::in_preserve_stack_slots();
1226 }
1227
1228 // "Top of Stack" slots that may be unused by the calling convention but must
1229 // otherwise be preserved.
1230 // On Intel these are not necessary and the value can be zero.
1231 static uint out_preserve_stack_slots() {
1232 return SharedRuntime::out_preserve_stack_slots();
1233 }
1234
1235 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1236 // for calls to C. Supports the var-args backing area for register parms.
1237 uint varargs_C_out_slots_killed() const;
1238
1239 // Number of Stack Slots consumed by a synchronization entry
1240 int sync_stack_slots() const;
1241
1242 // Compute the name of old_SP. See <arch>.ad for frame layout.
1243 OptoReg::Name compute_old_SP();
1244
1245 private:
1246 // Phase control:
1247 void Init(bool aliasing); // Prepare for a single compilation
1248 void Optimize(); // Given a graph, optimize it
1249 void Code_Gen(); // Generate code from a graph
1250
1251 // Management of the AliasType table.
1252 void grow_alias_types();
1253 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1254 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1255 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1256
1257 void verify_top(Node*) const PRODUCT_RETURN;
1258
1259 // Intrinsic setup.
1260 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1261 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1262 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1263 void register_intrinsic(CallGenerator* cg); // update fn
1264
1265 #ifndef PRODUCT
1266 static juint _intrinsic_hist_count[];
1267 static jubyte _intrinsic_hist_flags[];
1268 #endif
1269 // Function calls made by the public function final_graph_reshaping.
1270 // No need to be made public as they are not called elsewhere.
1271 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1272 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1273 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1274 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1275
1276 // Logic cone optimization.
1277 void optimize_logic_cones(PhaseIterGVN &igvn);
1278 void collect_logic_cone_roots(Unique_Node_List& list);
1279 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1280 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1281 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1282 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1283 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1284 void check_no_dead_use() const NOT_DEBUG_RETURN;
1285
1286 public:
1287
1288 // Note: Histogram array size is about 1 Kb.
1289 enum { // flag bits:
1290 _intrinsic_worked = 1, // succeeded at least once
1291 _intrinsic_failed = 2, // tried it but it failed
1292 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1293 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1294 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1295 };
1296 // Update histogram. Return boolean if this is a first-time occurrence.
1297 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1298 bool is_virtual, int flags) PRODUCT_RETURN0;
1299 static void print_intrinsic_statistics() PRODUCT_RETURN;
1300
1301 // Graph verification code
1302 // Walk the node list, verifying that there is a one-to-one
1303 // correspondence between Use-Def edges and Def-Use edges
1304 // The option no_dead_code enables stronger checks that the
1305 // graph is strongly connected from root in both directions.
1306 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1307
1308 // Verify bi-directional correspondence of edges
1309 void verify_bidirectional_edges(Unique_Node_List &visited);
1310
1311 // End-of-run dumps.
1312 static void print_statistics() PRODUCT_RETURN;
1313
1314 // Verify ADLC assumptions during startup
1315 static void adlc_verification() PRODUCT_RETURN;
1316
1317 // Definitions of pd methods
1318 static void pd_compiler2_init();
1319
1320 // Static parse-time type checking logic for gen_subtype_check:
1321 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1322 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1323
1324 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1325 // Optional control dependency (for example, on range check)
1326 Node* ctrl = nullptr);
1327
1328 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1329 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1330
1331 Node* optimize_acmp(PhaseGVN* phase, Node* a, Node* b);
1332
1333 // Auxiliary method for randomized fuzzing/stressing
1334 int random();
1335 bool randomized_select(int count);
1336
1337 // seed random number generation and log the seed for repeatability.
1338 void initialize_stress_seed(const DirectiveSet* directive);
1339
1340 // supporting clone_map
1341 CloneMap& clone_map();
1342 void set_clone_map(Dict* d);
1343
1344 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1345 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1346 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1347
1348 #ifdef IA32
1349 private:
1350 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1351 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1352
1353 // Remember if this compilation changes hardware mode to 24-bit precision.
1354 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1355 _select_24_bit_instr = selection;
1356 _in_24_bit_fp_mode = mode;
1357 }
1358
1359 public:
1360 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1361 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1362 #endif // IA32
1363 #ifdef ASSERT
1364 VerifyMeetResult* _type_verify;
1365 void set_exception_backedge() { _exception_backedge = true; }
1366 bool has_exception_backedge() const { return _exception_backedge; }
1367 #endif
1368
1369 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1370 BasicType out_bt, BasicType in_bt);
1371
1372 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1373 };
1374
1375 #endif // SHARE_OPTO_COMPILE_HPP