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 CallStaticJavaNode;
56 class CloneMap;
57 class CompilationFailureInfo;
58 class ConnectionGraph;
59 class IdealGraphPrinter;
60 class InlineTree;
61 class Matcher;
62 class MachConstantNode;
63 class MachConstantBaseNode;
64 class MachNode;
65 class MachOper;
66 class MachSafePointNode;
67 class Node;
68 class Node_Array;
69 class Node_List;
70 class Node_Notes;
71 class NodeHash;
72 class NodeCloneInfo;
73 class OptoReg;
74 class ParsePredicateNode;
75 class PhaseCFG;
76 class PhaseGVN;
77 class PhaseIterGVN;
78 class PhaseRegAlloc;
79 class PhaseCCP;
80 class PhaseOutput;
81 class RootNode;
82 class relocInfo;
83 class StartNode;
84 class SafePointNode;
85 class JVMState;
86 class Type;
87 class TypeInt;
88 class TypeInteger;
89 class TypeKlassPtr;
90 class TypePtr;
91 class TypeOopPtr;
92 class TypeFunc;
93 class TypeVect;
94 class Type_Array;
95 class Unique_Node_List;
96 class UnstableIfTrap;
97 class nmethod;
98 class Node_Stack;
99 struct Final_Reshape_Counts;
100 class VerifyMeetResult;
101
102 enum LoopOptsMode {
103 LoopOptsDefault,
104 LoopOptsNone,
105 LoopOptsMaxUnroll,
106 LoopOptsShenandoahExpand,
107 LoopOptsShenandoahPostExpand,
108 LoopOptsSkipSplitIf,
109 LoopOptsVerify
110 };
111
112 // The type of all node counts and indexes.
113 // It must hold at least 16 bits, but must also be fast to load and store.
114 // This type, if less than 32 bits, could limit the number of possible nodes.
115 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
116 typedef unsigned int node_idx_t;
117
118 class NodeCloneInfo {
119 private:
120 uint64_t _idx_clone_orig;
121 public:
122
123 void set_idx(node_idx_t idx) {
124 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
125 }
126 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
127
128 void set_gen(int generation) {
129 uint64_t g = (uint64_t)generation << 32;
130 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
131 }
132 int gen() const { return (int)(_idx_clone_orig >> 32); }
133
134 void set(uint64_t x) { _idx_clone_orig = x; }
135 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
136 uint64_t get() const { return _idx_clone_orig; }
137
138 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
139 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
140
141 void dump_on(outputStream* st) const;
142 };
143
144 class CloneMap {
145 friend class Compile;
146 private:
147 bool _debug;
148 Dict* _dict;
149 int _clone_idx; // current cloning iteration/generation in loop unroll
150 public:
151 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
152 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
153 Dict* dict() const { return _dict; }
154 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
155 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
156 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
157 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
158 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
159 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
160 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
161 int max_gen() const;
162 void clone(Node* old, Node* nnn, int gen);
163 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
164 void dump(node_idx_t key, outputStream* st) const;
165
166 int clone_idx() const { return _clone_idx; }
167 void set_clone_idx(int x) { _clone_idx = x; }
168 bool is_debug() const { return _debug; }
169 void set_debug(bool debug) { _debug = debug; }
170
171 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
172 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
173 };
174
175 class Options {
176 friend class Compile;
177 friend class VMStructs;
178 private:
179 const bool _subsume_loads; // Load can be matched as part of a larger op.
180 const bool _do_escape_analysis; // Do escape analysis.
181 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
182 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges.
183 const bool _eliminate_boxing; // Do boxing elimination.
184 const bool _do_locks_coarsening; // Do locks coarsening
185 const bool _do_superword; // Do SuperWord
186 const bool _install_code; // Install the code that was compiled
187 public:
188 Options(bool subsume_loads,
189 bool do_escape_analysis,
190 bool do_iterative_escape_analysis,
191 bool do_reduce_allocation_merges,
192 bool eliminate_boxing,
193 bool do_locks_coarsening,
194 bool do_superword,
195 bool install_code) :
196 _subsume_loads(subsume_loads),
197 _do_escape_analysis(do_escape_analysis),
198 _do_iterative_escape_analysis(do_iterative_escape_analysis),
199 _do_reduce_allocation_merges(do_reduce_allocation_merges),
200 _eliminate_boxing(eliminate_boxing),
201 _do_locks_coarsening(do_locks_coarsening),
202 _do_superword(do_superword),
203 _install_code(install_code) {
204 }
205
206 static Options for_runtime_stub() {
207 return Options(
208 /* subsume_loads = */ true,
209 /* do_escape_analysis = */ false,
210 /* do_iterative_escape_analysis = */ false,
211 /* do_reduce_allocation_merges = */ false,
212 /* eliminate_boxing = */ false,
213 /* do_lock_coarsening = */ false,
214 /* do_superword = */ true,
215 /* install_code = */ true
216 );
217 }
218 };
219
220 //------------------------------Compile----------------------------------------
221 // This class defines a top-level Compiler invocation.
222
223 class Compile : public Phase {
224 friend class VMStructs;
225
226 public:
227 // Fixed alias indexes. (See also MergeMemNode.)
228 enum {
229 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
230 AliasIdxBot = 2, // pseudo-index, aliases to everything
231 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
232 };
233
234 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
235 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
236 // then brackets are put into the log, with time stamps and node counts.
237 // (The time collection itself is always conditionalized on CITime.)
238 class TracePhase : public TraceTime {
239 private:
240 Compile* _compile;
241 CompileLog* _log;
242 bool _dolog;
243 public:
244 TracePhase(PhaseTraceId phaseTraceId);
245 TracePhase(const char* name, PhaseTraceId phaseTraceId);
246 ~TracePhase();
247 const char* phase_name() const { return title(); }
248 };
249
250 // Information per category of alias (memory slice)
251 class AliasType {
252 private:
253 friend class Compile;
254
255 int _index; // unique index, used with MergeMemNode
256 const TypePtr* _adr_type; // normalized address type
257 ciField* _field; // relevant instance field, or null if none
258 const Type* _element; // relevant array element type, or null if none
259 bool _is_rewritable; // false if the memory is write-once only
260 int _general_index; // if this is type is an instance, the general
261 // type that this is an instance of
262
263 void Init(int i, const TypePtr* at);
264
265 public:
266 int index() const { return _index; }
267 const TypePtr* adr_type() const { return _adr_type; }
268 ciField* field() const { return _field; }
269 const Type* element() const { return _element; }
270 bool is_rewritable() const { return _is_rewritable; }
271 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
272 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
273
274 void set_rewritable(bool z) { _is_rewritable = z; }
275 void set_field(ciField* f) {
276 assert(!_field,"");
277 _field = f;
278 if (f->is_final() || f->is_stable()) {
279 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
280 _is_rewritable = false;
281 }
282 }
283 void set_element(const Type* e) {
284 assert(_element == nullptr, "");
285 _element = e;
286 }
287
288 BasicType basic_type() const;
289
290 void print_on(outputStream* st) PRODUCT_RETURN;
291 };
292
293 enum {
294 logAliasCacheSize = 6,
295 AliasCacheSize = (1<<logAliasCacheSize)
296 };
297 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
298 enum {
299 trapHistLength = MethodData::_trap_hist_limit
300 };
301
302 private:
303 // Fixed parameters to this compilation.
304 const int _compile_id;
305 const Options _options; // Compilation options
306 ciMethod* _method; // The method being compiled.
307 int _entry_bci; // entry bci for osr methods.
308 const TypeFunc* _tf; // My kind of signature
309 InlineTree* _ilt; // Ditto (temporary).
310 address _stub_function; // VM entry for stub being compiled, or null
311 const char* _stub_name; // Name of stub or adapter being compiled, or null
312 address _stub_entry_point; // Compile code entry for generated stub, or null
313
314 // Control of this compilation.
315 int _max_inline_size; // Max inline size for this compilation
316 int _freq_inline_size; // Max hot method inline size for this compilation
317 int _fixed_slots; // count of frame slots not allocated by the register
318 // allocator i.e. locks, original deopt pc, etc.
319 uintx _max_node_limit; // Max unique node count during a single compilation.
320
321 bool _post_loop_opts_phase; // Loop opts are finished.
322 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
323
324 int _major_progress; // Count of something big happening
325 bool _inlining_progress; // progress doing incremental inlining?
326 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
327 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
328 bool _has_loops; // True if the method _may_ have some loops
329 bool _has_split_ifs; // True if the method _may_ have some split-if
330 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
331 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
332 bool _has_boxed_value; // True if a boxed object is allocated
333 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
334 uint _max_vector_size; // Maximum size of generated vectors
335 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
336 uint _trap_hist[trapHistLength]; // Cumulative traps
337 bool _trap_can_recompile; // Have we emitted a recompiling trap?
338 uint _decompile_count; // Cumulative decompilation counts.
339 bool _do_inlining; // True if we intend to do inlining
340 bool _do_scheduling; // True if we intend to do scheduling
341 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
342 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
343 bool _use_cmove; // True if CMove should be used without profitability analysis
344 bool _do_aliasing; // True if we intend to do aliasing
345 bool _print_assembly; // True if we should dump assembly code for this compilation
346 bool _print_inlining; // True if we should print inlining for this compilation
347 bool _print_intrinsics; // True if we should print intrinsics for this compilation
348 #ifndef PRODUCT
349 uint _igv_idx; // Counter for IGV node identifiers
350 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
351 bool _trace_opto_output;
352 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
353 #endif
354 bool _has_irreducible_loop; // Found irreducible loops
355 // JSR 292
356 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
357 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
358 bool _has_scoped_access; // For shared scope closure
359 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
360 int _loop_opts_cnt; // loop opts round
361 uint _stress_seed; // Seed for stress testing
362
363 // Compilation environment.
364 Arena _comp_arena; // Arena with lifetime equivalent to Compile
365 void* _barrier_set_state; // Potential GC barrier state for Compile
366 ciEnv* _env; // CI interface
367 DirectiveSet* _directive; // Compiler directive
368 CompileLog* _log; // from CompilerThread
369 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
370 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
371 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
372 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
373 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
374 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates.
375 GrowableArray<Node*> _template_assertion_predicate_opaqs;
376 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
377 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
378 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
379 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
380 ConnectionGraph* _congraph;
381 #ifndef PRODUCT
382 IdealGraphPrinter* _igv_printer;
383 static IdealGraphPrinter* _debug_file_printer;
384 static IdealGraphPrinter* _debug_network_printer;
385 #endif
386
387
388 // Node management
389 uint _unique; // Counter for unique Node indices
390 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
391 // So use this to keep count and make the call O(1).
392 VectorSet _dead_node_list; // Set of dead nodes
393 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
394 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
395
396 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
397
398 // Arenas for new-space and old-space nodes.
399 // Swapped between using _node_arena.
400 // The lifetime of the old-space nodes is during xform.
401 Arena _node_arena_one;
402 Arena _node_arena_two;
403 Arena* _node_arena;
404 public:
405 Arena* swap_old_and_new() {
406 Arena* filled_arena_ptr = _node_arena;
407 Arena* old_arena_ptr = old_arena();
408 old_arena_ptr->destruct_contents();
409 _node_arena = old_arena_ptr;
410 return filled_arena_ptr;
411 }
412 private:
413 RootNode* _root; // Unique root of compilation, or null after bail-out.
414 Node* _top; // Unique top node. (Reset by various phases.)
415
416 Node* _immutable_memory; // Initial memory state
417
418 Node* _recent_alloc_obj;
419 Node* _recent_alloc_ctl;
420
421 // Constant table
422 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
423
424
425 // Blocked array of debugging and profiling information,
426 // tracked per node.
427 enum { _log2_node_notes_block_size = 8,
428 _node_notes_block_size = (1<<_log2_node_notes_block_size)
429 };
430 GrowableArray<Node_Notes*>* _node_note_array;
431 Node_Notes* _default_node_notes; // default notes for new nodes
432
433 // After parsing and every bulk phase we hang onto the Root instruction.
434 // The RootNode instruction is where the whole program begins. It produces
435 // the initial Control and BOTTOM for everybody else.
436
437 // Type management
438 Arena _Compile_types; // Arena for all types
439 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
440 Dict* _type_dict; // Intern table
441 CloneMap _clone_map; // used for recording history of cloned nodes
442 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
443 ciMethod* _last_tf_m; // Cache for
444 const TypeFunc* _last_tf; // TypeFunc::make
445 AliasType** _alias_types; // List of alias types seen so far.
446 int _num_alias_types; // Logical length of _alias_types
447 int _max_alias_types; // Physical length of _alias_types
448 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
449
450 // Parsing, optimization
451 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
452
453 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
454 // If pushed outside IGVN, the Node is processed in the next IGVN round.
455 Unique_Node_List* _igvn_worklist;
456
457 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
458 Type_Array* _types;
459
460 // Shared node hash table for GVN, IGVN and CCP.
461 NodeHash* _node_hash;
462
463 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
464 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
465 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
466
467 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
468
469 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
470 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
471
472 // "MemLimit" directive was specified and the memory limit was hit during compilation
473 bool _oom;
474
475 // Inlining may not happen in parse order which would make
476 // PrintInlining output confusing. Keep track of PrintInlining
477 // pieces in order.
478 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
479 private:
480 CallGenerator* _cg;
481 stringStream _ss;
482 static const size_t default_stream_buffer_size = 128;
483
484 public:
485 PrintInliningBuffer()
486 : _cg(nullptr), _ss(default_stream_buffer_size) {}
487
488 stringStream* ss() { return &_ss; }
489 CallGenerator* cg() { return _cg; }
490 void set_cg(CallGenerator* cg) { _cg = cg; }
491 };
492
493 stringStream* _print_inlining_stream;
494 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
495 int _print_inlining_idx;
496 char* _print_inlining_output;
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 print_inlining_init();
510 void print_inlining_reinit();
511 void print_inlining_commit();
512 void print_inlining_push();
513 PrintInliningBuffer* print_inlining_current();
514
515 void log_late_inline_failure(CallGenerator* cg, const char* msg);
516 DEBUG_ONLY(bool _exception_backedge;)
517
518 void record_method_not_compilable_oom();
519
520 public:
521
522 void* barrier_set_state() const { return _barrier_set_state; }
523
524 stringStream* print_inlining_stream() {
525 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
526 return _print_inlining_stream;
527 }
528
529 void print_inlining_update(CallGenerator* cg);
530 void print_inlining_update_delayed(CallGenerator* cg);
531 void print_inlining_move_to(CallGenerator* cg);
532 void print_inlining_assert_ready();
533 void print_inlining_reset();
534
535 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) {
536 stringStream ss;
537 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg);
538 print_inlining_stream()->print("%s", ss.freeze());
539 }
540
541 #ifndef PRODUCT
542 IdealGraphPrinter* igv_printer() { return _igv_printer; }
543 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
544 #endif
545
546 void log_late_inline(CallGenerator* cg);
547 void log_inline_id(CallGenerator* cg);
548 void log_inline_failure(const char* msg);
549
550 void* replay_inline_data() const { return _replay_inline_data; }
551
552 // Dump inlining replay data to the stream.
553 void dump_inline_data(outputStream* out);
554 void dump_inline_data_reduced(outputStream* out);
555
556 private:
557 // Matching, CFG layout, allocation, code generation
558 PhaseCFG* _cfg; // Results of CFG finding
559 int _java_calls; // Number of java calls in the method
560 int _inner_loops; // Number of inner loops in the method
561 Matcher* _matcher; // Engine to map ideal to machine instructions
562 PhaseRegAlloc* _regalloc; // Results of register allocation.
563 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
564 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
565 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
566 int _interpreter_frame_size;
567
568 PhaseOutput* _output;
569
570 public:
571 // Accessors
572
573 // The Compile instance currently active in this (compiler) thread.
574 static Compile* current() {
575 return (Compile*) ciEnv::current()->compiler_data();
576 }
577
578 int interpreter_frame_size() const { return _interpreter_frame_size; }
579
580 PhaseOutput* output() const { return _output; }
581 void set_output(PhaseOutput* o) { _output = o; }
582
583 // ID for this compilation. Useful for setting breakpoints in the debugger.
584 int compile_id() const { return _compile_id; }
585 DirectiveSet* directive() const { return _directive; }
586
587 // Does this compilation allow instructions to subsume loads? User
588 // instructions that subsume a load may result in an unschedulable
589 // instruction sequence.
590 bool subsume_loads() const { return _options._subsume_loads; }
591 /** Do escape analysis. */
592 bool do_escape_analysis() const { return _options._do_escape_analysis; }
593 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
594 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
595 /** Do boxing elimination. */
596 bool eliminate_boxing() const { return _options._eliminate_boxing; }
597 /** Do aggressive boxing elimination. */
598 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
599 bool should_install_code() const { return _options._install_code; }
600 /** Do locks coarsening. */
601 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
602 bool do_superword() const { return _options._do_superword; }
603
604 // Other fixed compilation parameters.
605 ciMethod* method() const { return _method; }
606 int entry_bci() const { return _entry_bci; }
607 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
608 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
609 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
610 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
611 InlineTree* ilt() const { return _ilt; }
612 address stub_function() const { return _stub_function; }
613 const char* stub_name() const { return _stub_name; }
614 address stub_entry_point() const { return _stub_entry_point; }
615 void set_stub_entry_point(address z) { _stub_entry_point = z; }
616
617 // Control of this compilation.
618 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
619 void set_fixed_slots(int n) { _fixed_slots = n; }
620 int major_progress() const { return _major_progress; }
621 void set_inlining_progress(bool z) { _inlining_progress = z; }
622 int inlining_progress() const { return _inlining_progress; }
623 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
624 int inlining_incrementally() const { return _inlining_incrementally; }
625 void set_do_cleanup(bool z) { _do_cleanup = z; }
626 int do_cleanup() const { return _do_cleanup; }
627 void set_major_progress() { _major_progress++; }
628 void restore_major_progress(int progress) { _major_progress += progress; }
629 void clear_major_progress() { _major_progress = 0; }
630 int max_inline_size() const { return _max_inline_size; }
631 void set_freq_inline_size(int n) { _freq_inline_size = n; }
632 int freq_inline_size() const { return _freq_inline_size; }
633 void set_max_inline_size(int n) { _max_inline_size = n; }
634 bool has_loops() const { return _has_loops; }
635 void set_has_loops(bool z) { _has_loops = z; }
636 bool has_split_ifs() const { return _has_split_ifs; }
637 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
638 bool has_unsafe_access() const { return _has_unsafe_access; }
639 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
640 bool has_stringbuilder() const { return _has_stringbuilder; }
641 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
642 bool has_boxed_value() const { return _has_boxed_value; }
643 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
644 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
645 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
646 uint max_vector_size() const { return _max_vector_size; }
647 void set_max_vector_size(uint s) { _max_vector_size = s; }
648 bool clear_upper_avx() const { return _clear_upper_avx; }
649 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
650 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
651 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
652 bool trap_can_recompile() const { return _trap_can_recompile; }
653 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
654 uint decompile_count() const { return _decompile_count; }
655 void set_decompile_count(uint c) { _decompile_count = c; }
656 bool allow_range_check_smearing() const;
657 bool do_inlining() const { return _do_inlining; }
658 void set_do_inlining(bool z) { _do_inlining = z; }
659 bool do_scheduling() const { return _do_scheduling; }
660 void set_do_scheduling(bool z) { _do_scheduling = z; }
661 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
662 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
663 bool do_vector_loop() const { return _do_vector_loop; }
664 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
665 bool use_cmove() const { return _use_cmove; }
666 void set_use_cmove(bool z) { _use_cmove = z; }
667 bool do_aliasing() const { return _do_aliasing; }
668 bool print_assembly() const { return _print_assembly; }
669 void set_print_assembly(bool z) { _print_assembly = z; }
670 bool print_inlining() const { return _print_inlining; }
671 void set_print_inlining(bool z) { _print_inlining = z; }
672 bool print_intrinsics() const { return _print_intrinsics; }
673 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
674 uint max_node_limit() const { return (uint)_max_node_limit; }
675 void set_max_node_limit(uint n) { _max_node_limit = n; }
676 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
677 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
678 bool has_monitors() const { return _has_monitors; }
679 void set_has_monitors(bool v) { _has_monitors = v; }
680 bool has_scoped_access() const { return _has_scoped_access; }
681 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
682
683 // check the CompilerOracle for special behaviours for this compile
684 bool method_has_option(CompileCommandEnum option) {
685 return method() != nullptr && method()->has_option(option);
686 }
687
688 #ifndef PRODUCT
689 uint next_igv_idx() { return _igv_idx++; }
690 bool trace_opto_output() const { return _trace_opto_output; }
691 void print_ideal_ir(const char* phase_name);
692 bool should_print_ideal() const { return _directive->PrintIdealOption; }
693 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
694 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
695 int _in_dump_cnt; // Required for dumping ir nodes.
696 #endif
697 bool has_irreducible_loop() const { return _has_irreducible_loop; }
698 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
699
700 // JSR 292
701 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
702 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
703
704 Ticks _latest_stage_start_counter;
705
706 void begin_method();
707 void end_method();
708 bool should_print_igv(int level);
709 bool should_print_phase(CompilerPhaseType cpt);
710
711 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
712
713 #ifndef PRODUCT
714 void dump_igv(const char* graph_name, int level = 3) {
715 if (should_print_igv(level)) {
716 _igv_printer->print_method(graph_name, level);
717 }
718 }
719
720 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
721 void igv_print_method_to_network(const char* phase_name = "Debug");
722 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
723 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
724 #endif
725
726 int macro_count() const { return _macro_nodes.length(); }
727 int parse_predicate_count() const { return _parse_predicates.length(); }
728 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); }
729 int expensive_count() const { return _expensive_nodes.length(); }
730 int coarsened_count() const { return _coarsened_locks.length(); }
731
732 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
733 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); }
734
735 Node* template_assertion_predicate_opaq_node(int idx) const {
736 return _template_assertion_predicate_opaqs.at(idx);
737 }
738
739 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
740
741 ConnectionGraph* congraph() { return _congraph;}
742 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
743 void add_macro_node(Node * n) {
744 //assert(n->is_macro(), "must be a macro node");
745 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
746 _macro_nodes.append(n);
747 }
748 void remove_macro_node(Node* n) {
749 // this function may be called twice for a node so we can only remove it
750 // if it's still existing.
751 _macro_nodes.remove_if_existing(n);
752 // Remove from coarsened locks list if present
753 if (coarsened_count() > 0) {
754 remove_coarsened_lock(n);
755 }
756 }
757 void add_expensive_node(Node* n);
758 void remove_expensive_node(Node* n) {
759 _expensive_nodes.remove_if_existing(n);
760 }
761
762 void add_parse_predicate(ParsePredicateNode* n) {
763 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
764 _parse_predicates.append(n);
765 }
766
767 void remove_parse_predicate(ParsePredicateNode* n) {
768 if (parse_predicate_count() > 0) {
769 _parse_predicates.remove_if_existing(n);
770 }
771 }
772
773 void add_template_assertion_predicate_opaq(Node* n) {
774 assert(!_template_assertion_predicate_opaqs.contains(n),
775 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
776 _template_assertion_predicate_opaqs.append(n);
777 }
778
779 void remove_template_assertion_predicate_opaq(Node* n) {
780 if (template_assertion_predicate_count() > 0) {
781 _template_assertion_predicate_opaqs.remove_if_existing(n);
782 }
783 }
784 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
785 void remove_coarsened_lock(Node* n);
786 bool coarsened_locks_consistent();
787 void mark_unbalanced_boxes() const;
788
789 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
790 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
791 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
792
793 #ifdef ASSERT
794 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
795 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
796 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
797 #endif
798
799 bool allow_macro_nodes() { return _allow_macro_nodes; }
800 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
801
802 void record_for_post_loop_opts_igvn(Node* n);
803 void remove_from_post_loop_opts_igvn(Node* n);
804 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
805
806 void record_unstable_if_trap(UnstableIfTrap* trap);
807 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
808 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
809 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
810
811 void shuffle_macro_nodes();
812 void sort_macro_nodes();
813
814 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
815
816 // Are there candidate expensive nodes for optimization?
817 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
818 // Check whether n1 and n2 are similar
819 static int cmp_expensive_nodes(Node* n1, Node* n2);
820 // Sort expensive nodes to locate similar expensive nodes
821 void sort_expensive_nodes();
822
823 // Compilation environment.
824 Arena* comp_arena() { return &_comp_arena; }
825 ciEnv* env() const { return _env; }
826 CompileLog* log() const { return _log; }
827
828 bool failing_internal() const {
829 return _env->failing() ||
830 _failure_reason.get() != nullptr;
831 }
832
833 const char* failure_reason() const {
834 return _env->failing() ? _env->failure_reason()
835 : _failure_reason.get();
836 }
837
838 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
839
840 bool failing() {
841 if (failing_internal()) {
842 return true;
843 }
844 #ifdef ASSERT
845 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
846 if (phase_verify_ideal_loop()) {
847 return false;
848 }
849 if (StressBailout) {
850 return fail_randomly();
851 }
852 #endif
853 return false;
854 }
855
856 #ifdef ASSERT
857 bool fail_randomly();
858 bool failure_is_artificial();
859 #endif
860
861 bool failure_reason_is(const char* r) const {
862 return (r == _failure_reason.get()) ||
863 (r != nullptr &&
864 _failure_reason.get() != nullptr &&
865 strcmp(r, _failure_reason.get()) == 0);
866 }
867
868 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
869 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
870 env()->record_method_not_compilable(reason);
871 // Record failure reason.
872 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
873 }
874 bool check_node_count(uint margin, const char* reason) {
875 if (oom()) {
876 record_method_not_compilable_oom();
877 return true;
878 }
879 if (live_nodes() + margin > max_node_limit()) {
880 record_method_not_compilable(reason);
881 return true;
882 } else {
883 return false;
884 }
885 }
886 bool oom() const { return _oom; }
887 void set_oom() { _oom = true; }
888
889 // Node management
890 uint unique() const { return _unique; }
891 uint next_unique() { return _unique++; }
892 void set_unique(uint i) { _unique = i; }
893 Arena* node_arena() { return _node_arena; }
894 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
895 RootNode* root() const { return _root; }
896 void set_root(RootNode* r) { _root = r; }
897 StartNode* start() const; // (Derived from root.)
898 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
899 Node* immutable_memory();
900
901 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
902 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
903 void set_recent_alloc(Node* ctl, Node* obj) {
904 _recent_alloc_ctl = ctl;
905 _recent_alloc_obj = obj;
906 }
907 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
908 _dead_node_count++;
909 }
910 void reset_dead_node_list() { _dead_node_list.reset();
911 _dead_node_count = 0;
912 }
913 uint live_nodes() const {
914 int val = _unique - _dead_node_count;
915 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
916 return (uint) val;
917 }
918 #ifdef ASSERT
919 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
920 bool phase_optimize_finished() const { return _phase_optimize_finished; }
921 uint count_live_nodes_by_graph_walk();
922 void print_missing_nodes();
923 #endif
924
925 // Record modified nodes to check that they are put on IGVN worklist
926 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
927 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
928 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
929
930 MachConstantBaseNode* mach_constant_base_node();
931 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
932 // Generated by adlc, true if CallNode requires MachConstantBase.
933 bool needs_deep_clone_jvms();
934
935 // Handy undefined Node
936 Node* top() const { return _top; }
937
938 // these are used by guys who need to know about creation and transformation of top:
939 Node* cached_top_node() { return _top; }
940 void set_cached_top_node(Node* tn);
941
942 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
943 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
944 Node_Notes* default_node_notes() const { return _default_node_notes; }
945 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
946
947 Node_Notes* node_notes_at(int idx) {
948 return locate_node_notes(_node_note_array, idx, false);
949 }
950 inline bool set_node_notes_at(int idx, Node_Notes* value);
951
952 // Copy notes from source to dest, if they exist.
953 // Overwrite dest only if source provides something.
954 // Return true if information was moved.
955 bool copy_node_notes_to(Node* dest, Node* source);
956
957 // Workhorse function to sort out the blocked Node_Notes array:
958 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
959 int idx, bool can_grow = false);
960
961 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
962
963 // Type management
964 Arena* type_arena() { return _type_arena; }
965 Dict* type_dict() { return _type_dict; }
966 size_t type_last_size() { return _type_last_size; }
967 int num_alias_types() { return _num_alias_types; }
968
969 void init_type_arena() { _type_arena = &_Compile_types; }
970 void set_type_arena(Arena* a) { _type_arena = a; }
971 void set_type_dict(Dict* d) { _type_dict = d; }
972 void set_type_last_size(size_t sz) { _type_last_size = sz; }
973
974 const TypeFunc* last_tf(ciMethod* m) {
975 return (m == _last_tf_m) ? _last_tf : nullptr;
976 }
977 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
978 assert(m != nullptr || tf == nullptr, "");
979 _last_tf_m = m;
980 _last_tf = tf;
981 }
982
983 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
984 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); }
985 bool have_alias_type(const TypePtr* adr_type);
986 AliasType* alias_type(ciField* field);
987
988 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
989 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
990 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
991
992 // Building nodes
993 void rethrow_exceptions(JVMState* jvms);
994 void return_values(JVMState* jvms);
995 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
996
997 // Decide how to build a call.
998 // The profile factor is a discount to apply to this site's interp. profile.
999 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1000 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
1001 bool allow_intrinsics = true);
1002 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1003 return should_delay_string_inlining(call_method, jvms) ||
1004 should_delay_boxing_inlining(call_method, jvms) ||
1005 should_delay_vector_inlining(call_method, jvms);
1006 }
1007 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1008 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1009 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
1010 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
1011
1012 // Helper functions to identify inlining potential at call-site
1013 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
1014 ciKlass* holder, ciMethod* callee,
1015 const TypeOopPtr* receiver_type, bool is_virtual,
1016 bool &call_does_dispatch, int &vtable_index,
1017 bool check_access = true);
1018 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1019 ciMethod* callee, const TypeOopPtr* receiver_type,
1020 bool check_access = true);
1021
1022 // Report if there were too many traps at a current method and bci.
1023 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1024 // If there is no MDO at all, report no trap unless told to assume it.
1025 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1026 // This version, unspecific to a particular bci, asks if
1027 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1028 bool too_many_traps(Deoptimization::DeoptReason reason,
1029 // Privately used parameter for logging:
1030 ciMethodData* logmd = nullptr);
1031 // Report if there were too many recompiles at a method and bci.
1032 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1033 // Report if there were too many traps or recompiles at a method and bci.
1034 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1035 return too_many_traps(method, bci, reason) ||
1036 too_many_recompiles(method, bci, reason);
1037 }
1038 // Return a bitset with the reasons where deoptimization is allowed,
1039 // i.e., where there were not too many uncommon traps.
1040 int _allowed_reasons;
1041 int allowed_deopt_reasons() { return _allowed_reasons; }
1042 void set_allowed_deopt_reasons();
1043
1044 // Parsing, optimization
1045 PhaseGVN* initial_gvn() { return _initial_gvn; }
1046 Unique_Node_List* igvn_worklist() {
1047 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1048 return _igvn_worklist;
1049 }
1050 Type_Array* types() {
1051 assert(_types != nullptr, "must be created in Compile::Compile");
1052 return _types;
1053 }
1054 NodeHash* node_hash() {
1055 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1056 return _node_hash;
1057 }
1058 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1059 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1060 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1061
1062 // Replace n by nn using initial_gvn, calling hash_delete and
1063 // record_for_igvn as needed.
1064 void gvn_replace_by(Node* n, Node* nn);
1065
1066
1067 void identify_useful_nodes(Unique_Node_List &useful);
1068 void update_dead_node_list(Unique_Node_List &useful);
1069 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist);
1070
1071 void remove_useless_node(Node* dead);
1072
1073 // Record this CallGenerator for inlining at the end of parsing.
1074 void add_late_inline(CallGenerator* cg) {
1075 _late_inlines.insert_before(_late_inlines_pos, cg);
1076 _late_inlines_pos++;
1077 }
1078
1079 void prepend_late_inline(CallGenerator* cg) {
1080 _late_inlines.insert_before(0, cg);
1081 }
1082
1083 void add_string_late_inline(CallGenerator* cg) {
1084 _string_late_inlines.push(cg);
1085 }
1086
1087 void add_boxing_late_inline(CallGenerator* cg) {
1088 _boxing_late_inlines.push(cg);
1089 }
1090
1091 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1092 _vector_reboxing_late_inlines.push(cg);
1093 }
1094
1095 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1096 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1097
1098 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1099 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1100
1101 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1102
1103 void process_print_inlining();
1104 void dump_print_inlining();
1105
1106 bool over_inlining_cutoff() const {
1107 if (!inlining_incrementally()) {
1108 return unique() > (uint)NodeCountInliningCutoff;
1109 } else {
1110 // Give some room for incremental inlining algorithm to "breathe"
1111 // and avoid thrashing when live node count is close to the limit.
1112 // Keep in mind that live_nodes() isn't accurate during inlining until
1113 // dead node elimination step happens (see Compile::inline_incrementally).
1114 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1115 }
1116 }
1117
1118 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1119 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--; }
1120 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1121
1122 bool inline_incrementally_one();
1123 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1124 void inline_incrementally(PhaseIterGVN& igvn);
1125 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); }
1126 void inline_string_calls(bool parse_time);
1127 void inline_boxing_calls(PhaseIterGVN& igvn);
1128 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1129 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1130
1131 void inline_vector_reboxing_calls();
1132 bool has_vbox_nodes();
1133
1134 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1135
1136 // Matching, CFG layout, allocation, code generation
1137 PhaseCFG* cfg() { return _cfg; }
1138 bool has_java_calls() const { return _java_calls > 0; }
1139 int java_calls() const { return _java_calls; }
1140 int inner_loops() const { return _inner_loops; }
1141 Matcher* matcher() { return _matcher; }
1142 PhaseRegAlloc* regalloc() { return _regalloc; }
1143 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1144 Arena* indexSet_arena() { return _indexSet_arena; }
1145 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1146 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1147
1148 void update_interpreter_frame_size(int size) {
1149 if (_interpreter_frame_size < size) {
1150 _interpreter_frame_size = size;
1151 }
1152 }
1153
1154 void set_matcher(Matcher* m) { _matcher = m; }
1155 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1156 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1157 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1158
1159 void set_java_calls(int z) { _java_calls = z; }
1160 void set_inner_loops(int z) { _inner_loops = z; }
1161
1162 Dependencies* dependencies() { return env()->dependencies(); }
1163
1164 // Major entry point. Given a Scope, compile the associated method.
1165 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1166 // replacement, entry_bci indicates the bytecode for which to compile a
1167 // continuation.
1168 Compile(ciEnv* ci_env, ciMethod* target,
1169 int entry_bci, Options options, DirectiveSet* directive);
1170
1171 // Second major entry point. From the TypeFunc signature, generate code
1172 // to pass arguments from the Java calling convention to the C calling
1173 // convention.
1174 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1175 address stub_function, const char *stub_name,
1176 int is_fancy_jump, bool pass_tls,
1177 bool return_pc, DirectiveSet* directive);
1178
1179 ~Compile();
1180
1181 // Are we compiling a method?
1182 bool has_method() { return method() != nullptr; }
1183
1184 // Maybe print some information about this compile.
1185 void print_compile_messages();
1186
1187 // Final graph reshaping, a post-pass after the regular optimizer is done.
1188 bool final_graph_reshaping();
1189
1190 // returns true if adr is completely contained in the given alias category
1191 bool must_alias(const TypePtr* adr, int alias_idx);
1192
1193 // returns true if adr overlaps with the given alias category
1194 bool can_alias(const TypePtr* adr, int alias_idx);
1195
1196 // Stack slots that may be unused by the calling convention but must
1197 // otherwise be preserved. On Intel this includes the return address.
1198 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1199 uint in_preserve_stack_slots() {
1200 return SharedRuntime::in_preserve_stack_slots();
1201 }
1202
1203 // "Top of Stack" slots that may be unused by the calling convention but must
1204 // otherwise be preserved.
1205 // On Intel these are not necessary and the value can be zero.
1206 static uint out_preserve_stack_slots() {
1207 return SharedRuntime::out_preserve_stack_slots();
1208 }
1209
1210 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1211 // for calls to C. Supports the var-args backing area for register parms.
1212 uint varargs_C_out_slots_killed() const;
1213
1214 // Number of Stack Slots consumed by a synchronization entry
1215 int sync_stack_slots() const;
1216
1217 // Compute the name of old_SP. See <arch>.ad for frame layout.
1218 OptoReg::Name compute_old_SP();
1219
1220 private:
1221 // Phase control:
1222 void Init(bool aliasing); // Prepare for a single compilation
1223 void Optimize(); // Given a graph, optimize it
1224 void Code_Gen(); // Generate code from a graph
1225
1226 // Management of the AliasType table.
1227 void grow_alias_types();
1228 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1229 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1230 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1231
1232 void verify_top(Node*) const PRODUCT_RETURN;
1233
1234 // Intrinsic setup.
1235 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1236 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1237 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1238 void register_intrinsic(CallGenerator* cg); // update fn
1239
1240 #ifndef PRODUCT
1241 static juint _intrinsic_hist_count[];
1242 static jubyte _intrinsic_hist_flags[];
1243 #endif
1244 // Function calls made by the public function final_graph_reshaping.
1245 // No need to be made public as they are not called elsewhere.
1246 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1247 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1248 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1249 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1250
1251 // Logic cone optimization.
1252 void optimize_logic_cones(PhaseIterGVN &igvn);
1253 void collect_logic_cone_roots(Unique_Node_List& list);
1254 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1255 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1256 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1257 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1258 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1259 void check_no_dead_use() const NOT_DEBUG_RETURN;
1260
1261 public:
1262
1263 // Note: Histogram array size is about 1 Kb.
1264 enum { // flag bits:
1265 _intrinsic_worked = 1, // succeeded at least once
1266 _intrinsic_failed = 2, // tried it but it failed
1267 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1268 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1269 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1270 };
1271 // Update histogram. Return boolean if this is a first-time occurrence.
1272 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1273 bool is_virtual, int flags) PRODUCT_RETURN0;
1274 static void print_intrinsic_statistics() PRODUCT_RETURN;
1275
1276 // Graph verification code
1277 // Walk the node list, verifying that there is a one-to-one
1278 // correspondence between Use-Def edges and Def-Use edges
1279 // The option no_dead_code enables stronger checks that the
1280 // graph is strongly connected from root in both directions.
1281 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1282
1283 // Verify bi-directional correspondence of edges
1284 void verify_bidirectional_edges(Unique_Node_List &visited);
1285
1286 // End-of-run dumps.
1287 static void print_statistics() PRODUCT_RETURN;
1288
1289 // Verify ADLC assumptions during startup
1290 static void adlc_verification() PRODUCT_RETURN;
1291
1292 // Definitions of pd methods
1293 static void pd_compiler2_init();
1294
1295 // Static parse-time type checking logic for gen_subtype_check:
1296 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1297 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1298
1299 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1300 // Optional control dependency (for example, on range check)
1301 Node* ctrl = nullptr);
1302
1303 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1304 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1305
1306 // Auxiliary methods for randomized fuzzing/stressing
1307 int random();
1308 bool randomized_select(int count);
1309
1310 // seed random number generation and log the seed for repeatability.
1311 void initialize_stress_seed(const DirectiveSet* directive);
1312
1313 // supporting clone_map
1314 CloneMap& clone_map();
1315 void set_clone_map(Dict* d);
1316
1317 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1318 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1319 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1320
1321 #ifdef IA32
1322 private:
1323 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1324 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1325
1326 // Remember if this compilation changes hardware mode to 24-bit precision.
1327 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1328 _select_24_bit_instr = selection;
1329 _in_24_bit_fp_mode = mode;
1330 }
1331
1332 public:
1333 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1334 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1335 #endif // IA32
1336 #ifdef ASSERT
1337 VerifyMeetResult* _type_verify;
1338 void set_exception_backedge() { _exception_backedge = true; }
1339 bool has_exception_backedge() const { return _exception_backedge; }
1340 #endif
1341
1342 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1343 BasicType out_bt, BasicType in_bt);
1344
1345 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1346 };
1347
1348 #endif // SHARE_OPTO_COMPILE_HPP