1 /*
2 * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef SHARE_OPTO_COMPILE_HPP
26 #define SHARE_OPTO_COMPILE_HPP
27
28 #include "asm/codeBuffer.hpp"
29 #include "ci/compilerInterface.hpp"
30 #include "code/debugInfoRec.hpp"
31 #include "compiler/cHeapStringHolder.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compiler_globals.hpp"
34 #include "compiler/compilerEvent.hpp"
35 #include "libadt/dict.hpp"
36 #include "libadt/vectset.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/methodData.hpp"
39 #include "opto/idealGraphPrinter.hpp"
40 #include "opto/phase.hpp"
41 #include "opto/phasetype.hpp"
42 #include "opto/printinlining.hpp"
43 #include "opto/regmask.hpp"
44 #include "runtime/deoptimization.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/timerTrace.hpp"
47 #include "runtime/vmThread.hpp"
48 #include "utilities/ticks.hpp"
49 #include "utilities/vmEnums.hpp"
50
51 class AbstractLockNode;
52 class AddPNode;
53 class Block;
54 class Bundle;
55 class CallGenerator;
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 OpaqueTemplateAssertionPredicateNode;
75 class OptoReg;
76 class ParsePredicateNode;
77 class PhaseCFG;
78 class PhaseGVN;
79 class PhaseIterGVN;
80 class PhaseRegAlloc;
81 class PhaseCCP;
82 class PhaseOutput;
83 class RootNode;
84 class relocInfo;
85 class StartNode;
86 class SafePointNode;
87 class JVMState;
88 class Type;
89 class TypeInt;
90 class TypeInteger;
91 class TypeKlassPtr;
92 class TypePtr;
93 class TypeOopPtr;
94 class TypeFunc;
95 class TypeVect;
96 class Type_Array;
97 class Unique_Node_List;
98 class UnstableIfTrap;
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 LoopOptsSkipSplitIf,
110 LoopOptsVerify
111 };
112
113 // The type of all node counts and indexes.
114 // It must hold at least 16 bits, but must also be fast to load and store.
115 // This type, if less than 32 bits, could limit the number of possible nodes.
116 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
117 typedef unsigned int node_idx_t;
118
119 class NodeCloneInfo {
120 private:
121 uint64_t _idx_clone_orig;
122 public:
123
124 void set_idx(node_idx_t idx) {
125 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
126 }
127 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
128
129 void set_gen(int generation) {
130 uint64_t g = (uint64_t)generation << 32;
131 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
132 }
133 int gen() const { return (int)(_idx_clone_orig >> 32); }
134
135 void set(uint64_t x) { _idx_clone_orig = x; }
136 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
137 uint64_t get() const { return _idx_clone_orig; }
138
139 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
140 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
141
142 void dump_on(outputStream* st) const;
143 };
144
145 class CloneMap {
146 friend class Compile;
147 private:
148 bool _debug;
149 Dict* _dict;
150 int _clone_idx; // current cloning iteration/generation in loop unroll
151 public:
152 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
153 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
154 Dict* dict() const { return _dict; }
155 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
156 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
157 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
158 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
159 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
160 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
161 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
162 int max_gen() const;
163 void clone(Node* old, Node* nnn, int gen);
164 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
165 void dump(node_idx_t key, outputStream* st) const;
166
167 int clone_idx() const { return _clone_idx; }
168 void set_clone_idx(int x) { _clone_idx = x; }
169 bool is_debug() const { return _debug; }
170 void set_debug(bool debug) { _debug = debug; }
171
172 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
173 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
174 };
175
176 class Options {
177 friend class Compile;
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
225 public:
226 // Fixed alias indexes. (See also MergeMemNode.)
227 enum {
228 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
229 AliasIdxBot = 2, // pseudo-index, aliases to everything
230 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
231 };
232
233 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
234 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
235 // then brackets are put into the log, with time stamps and node counts.
236 // (The time collection itself is always conditionalized on CITime.)
237 class TracePhase : public TraceTime {
238 private:
239 Compile* const _compile;
240 CompileLog* _log;
241 const bool _dolog;
242 public:
243 TracePhase(PhaseTraceId phaseTraceId);
244 TracePhase(const char* name, PhaseTraceId phaseTraceId);
245 ~TracePhase();
246 const char* phase_name() const { return title(); }
247 };
248
249 // Information per category of alias (memory slice)
250 class AliasType {
251 private:
252 friend class Compile;
253
254 int _index; // unique index, used with MergeMemNode
255 const TypePtr* _adr_type; // normalized address type
256 ciField* _field; // relevant instance field, or null if none
257 const Type* _element; // relevant array element type, or null if none
258 bool _is_rewritable; // false if the memory is write-once only
259 int _general_index; // if this is type is an instance, the general
260 // type that this is an instance of
261
262 void Init(int i, const TypePtr* at);
263
264 public:
265 int index() const { return _index; }
266 const TypePtr* adr_type() const { return _adr_type; }
267 ciField* field() const { return _field; }
268 const Type* element() const { return _element; }
269 bool is_rewritable() const { return _is_rewritable; }
270 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
271 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
272
273 void set_rewritable(bool z) { _is_rewritable = z; }
274 void set_field(ciField* f) {
275 assert(!_field,"");
276 _field = f;
277 if (f->is_final() || f->is_stable()) {
278 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
279 _is_rewritable = false;
280 }
281 }
282 void set_element(const Type* e) {
283 assert(_element == nullptr, "");
284 _element = e;
285 }
286
287 BasicType basic_type() const;
288
289 void print_on(outputStream* st) PRODUCT_RETURN;
290 };
291
292 enum {
293 logAliasCacheSize = 6,
294 AliasCacheSize = (1<<logAliasCacheSize)
295 };
296 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
297 enum {
298 trapHistLength = MethodData::_trap_hist_limit
299 };
300
301 private:
302 // Fixed parameters to this compilation.
303 const int _compile_id;
304 const Options _options; // Compilation options
305 ciMethod* _method; // The method being compiled.
306 int _entry_bci; // entry bci for osr methods.
307 const TypeFunc* _tf; // My kind of signature
308 InlineTree* _ilt; // Ditto (temporary).
309 address _stub_function; // VM entry for stub being compiled, or null
310 const char* _stub_name; // Name of stub or adapter being compiled, or null
311 StubId _stub_id; // unique id for stub or NO_STUBID
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 _merge_stores_phase; // Phase for merging stores, after post loop opts phase.
323 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
324
325 /* If major progress is set:
326 * Marks that the loop tree information (get_ctrl, idom, get_loop, etc.) could be invalid, and we need to rebuild the loop tree.
327 * It also indicates that the graph was changed in a way that is promising to be able to apply more loop optimization.
328 * If major progress is not set:
329 * Loop tree information is valid.
330 * If major progress is not set at the end of a loop opts phase, then we can stop loop opts, because we do not expect any further progress if we did more loop opts phases.
331 *
332 * This is not 100% accurate, the semantics of major progress has become less clear over time, but this is the general idea.
333 */
334 bool _major_progress;
335 bool _inlining_progress; // progress doing incremental inlining?
336 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
337 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
338 bool _has_loops; // True if the method _may_ have some loops
339 bool _has_split_ifs; // True if the method _may_ have some split-if
340 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
341 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
342 bool _has_boxed_value; // True if a boxed object is allocated
343 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
344 uint _max_vector_size; // Maximum size of generated vectors
345 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
346 uint _trap_hist[trapHistLength]; // Cumulative traps
347 bool _trap_can_recompile; // Have we emitted a recompiling trap?
348 uint _decompile_count; // Cumulative decompilation counts.
349 bool _do_inlining; // True if we intend to do inlining
350 bool _do_scheduling; // True if we intend to do scheduling
351 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
352 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
353 bool _use_cmove; // True if CMove should be used without profitability analysis
354 bool _do_aliasing; // True if we intend to do aliasing
355 bool _print_assembly; // True if we should dump assembly code for this compilation
356 bool _print_inlining; // True if we should print inlining for this compilation
357 bool _print_intrinsics; // True if we should print intrinsics for this compilation
358 #ifndef PRODUCT
359 uint _phase_counter; // Counter for the number of already printed phases
360 uint _igv_idx; // Counter for IGV node identifiers
361 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
362 bool _trace_opto_output;
363 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
364 #endif
365 bool _has_irreducible_loop; // Found irreducible loops
366 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
367 bool _has_scoped_access; // For shared scope closure
368 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
369 int _loop_opts_cnt; // loop opts round
370 uint _stress_seed; // Seed for stress testing
371
372 // Compilation environment.
373 Arena _comp_arena; // Arena with lifetime equivalent to Compile
374 void* _barrier_set_state; // Potential GC barrier state for Compile
375 ciEnv* _env; // CI interface
376 DirectiveSet* _directive; // Compiler directive
377 CompileLog* _log; // from CompilerThread
378 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
379 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
380 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
381 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
382 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
383 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates which can be seen as list
384 // of Template Assertion Predicates themselves.
385 GrowableArray<OpaqueTemplateAssertionPredicateNode*> _template_assertion_predicate_opaques;
386 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
387 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
388 GrowableArray<Node*> _for_merge_stores_igvn; // List of nodes for IGVN merge stores
389 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
390 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
391 ConnectionGraph* _congraph;
392 #ifndef PRODUCT
393 IdealGraphPrinter* _igv_printer;
394 static IdealGraphPrinter* _debug_file_printer;
395 static IdealGraphPrinter* _debug_network_printer;
396 #endif
397
398
399 // Node management
400 uint _unique; // Counter for unique Node indices
401 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
402 // So use this to keep count and make the call O(1).
403 VectorSet _dead_node_list; // Set of dead nodes
404 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
405 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
406
407 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
408
409 // Arenas for new-space and old-space nodes.
410 // Swapped between using _node_arena.
411 // The lifetime of the old-space nodes is during xform.
412 Arena _node_arena_one;
413 Arena _node_arena_two;
414 Arena* _node_arena;
415 public:
416 Arena* swap_old_and_new() {
417 Arena* filled_arena_ptr = _node_arena;
418 Arena* old_arena_ptr = old_arena();
419 old_arena_ptr->destruct_contents();
420 _node_arena = old_arena_ptr;
421 return filled_arena_ptr;
422 }
423 private:
424 RootNode* _root; // Unique root of compilation, or null after bail-out.
425 Node* _top; // Unique top node. (Reset by various phases.)
426
427 Node* _immutable_memory; // Initial memory state
428
429 Node* _recent_alloc_obj;
430 Node* _recent_alloc_ctl;
431
432 // Constant table
433 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
434
435
436 // Blocked array of debugging and profiling information,
437 // tracked per node.
438 enum { _log2_node_notes_block_size = 8,
439 _node_notes_block_size = (1<<_log2_node_notes_block_size)
440 };
441 GrowableArray<Node_Notes*>* _node_note_array;
442 Node_Notes* _default_node_notes; // default notes for new nodes
443
444 // After parsing and every bulk phase we hang onto the Root instruction.
445 // The RootNode instruction is where the whole program begins. It produces
446 // the initial Control and BOTTOM for everybody else.
447
448 // Type management
449 Arena _Compile_types; // Arena for all types
450 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
451 Dict* _type_dict; // Intern table
452 CloneMap _clone_map; // used for recording history of cloned nodes
453 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
454 ciMethod* _last_tf_m; // Cache for
455 const TypeFunc* _last_tf; // TypeFunc::make
456 AliasType** _alias_types; // List of alias types seen so far.
457 int _num_alias_types; // Logical length of _alias_types
458 int _max_alias_types; // Physical length of _alias_types
459 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
460
461 // Parsing, optimization
462 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
463
464 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
465 // If pushed outside IGVN, the Node is processed in the next IGVN round.
466 Unique_Node_List* _igvn_worklist;
467
468 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
469 Type_Array* _types;
470
471 // Shared node hash table for GVN, IGVN and CCP.
472 NodeHash* _node_hash;
473
474 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
475 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
476 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
477
478 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
479
480 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
481 bool _has_mh_late_inlines; // Can there still be a method handle late inlining pending?
482 // false: there can't be one
483 // true: we've enqueued one at some point so there may still be one
484
485 // "MemLimit" directive was specified and the memory limit was hit during compilation
486 bool _oom;
487
488 // Only keep nodes in the expensive node list that need to be optimized
489 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
490 // Use for sorting expensive nodes to bring similar nodes together
491 static int cmp_expensive_nodes(Node** n1, Node** n2);
492 // Expensive nodes list already sorted?
493 bool expensive_nodes_sorted() const;
494 // Remove the speculative part of types and clean up the graph
495 void remove_speculative_types(PhaseIterGVN &igvn);
496
497 void* _replay_inline_data; // Pointer to data loaded from file
498
499 void log_late_inline_failure(CallGenerator* cg, const char* msg);
500 DEBUG_ONLY(bool _exception_backedge;)
501
502 void record_method_not_compilable_oom();
503
504 InlinePrinter _inline_printer;
505
506 public:
507 void* barrier_set_state() const { return _barrier_set_state; }
508
509 InlinePrinter* inline_printer() { return &_inline_printer; }
510
511 #ifndef PRODUCT
512 IdealGraphPrinter* igv_printer() { return _igv_printer; }
513 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
514 #endif
515
516 void log_late_inline(CallGenerator* cg);
517 void log_inline_id(CallGenerator* cg);
518 void log_inline_failure(const char* msg);
519
520 void* replay_inline_data() const { return _replay_inline_data; }
521
522 // Dump inlining replay data to the stream.
523 void dump_inline_data(outputStream* out);
524 void dump_inline_data_reduced(outputStream* out);
525
526 private:
527 // Matching, CFG layout, allocation, code generation
528 PhaseCFG* _cfg; // Results of CFG finding
529 int _java_calls; // Number of java calls in the method
530 int _inner_loops; // Number of inner loops in the method
531 Matcher* _matcher; // Engine to map ideal to machine instructions
532 PhaseRegAlloc* _regalloc; // Results of register allocation.
533 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
534 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
535 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
536 int _interpreter_frame_size;
537
538 // Holds dynamically allocated extensions of short-lived register masks. Such
539 // extensions are potentially quite large and need tight resource marks which
540 // may conflict with other allocations in the default resource area.
541 // Therefore, we use a dedicated resource area for register masks.
542 ResourceArea _regmask_arena;
543
544 PhaseOutput* _output;
545
546 public:
547 // Accessors
548
549 // The Compile instance currently active in this (compiler) thread.
550 static Compile* current() {
551 return (Compile*) ciEnv::current()->compiler_data();
552 }
553
554 int interpreter_frame_size() const { return _interpreter_frame_size; }
555
556 PhaseOutput* output() const { return _output; }
557 void set_output(PhaseOutput* o) { _output = o; }
558
559 // ID for this compilation. Useful for setting breakpoints in the debugger.
560 int compile_id() const { return _compile_id; }
561 DirectiveSet* directive() const { return _directive; }
562
563 // Does this compilation allow instructions to subsume loads? User
564 // instructions that subsume a load may result in an unschedulable
565 // instruction sequence.
566 bool subsume_loads() const { return _options._subsume_loads; }
567 /** Do escape analysis. */
568 bool do_escape_analysis() const { return _options._do_escape_analysis; }
569 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
570 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
571 /** Do boxing elimination. */
572 bool eliminate_boxing() const { return _options._eliminate_boxing; }
573 /** Do aggressive boxing elimination. */
574 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
575 bool should_install_code() const { return _options._install_code; }
576 /** Do locks coarsening. */
577 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
578 bool do_superword() const { return _options._do_superword; }
579
580 // Other fixed compilation parameters.
581 ciMethod* method() const { return _method; }
582 int entry_bci() const { return _entry_bci; }
583 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
584 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
585 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
586 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
587 InlineTree* ilt() const { return _ilt; }
588 address stub_function() const { return _stub_function; }
589 const char* stub_name() const { return _stub_name; }
590 StubId stub_id() const { return _stub_id; }
591 address stub_entry_point() const { return _stub_entry_point; }
592 void set_stub_entry_point(address z) { _stub_entry_point = z; }
593
594 // Control of this compilation.
595 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
596 void set_fixed_slots(int n) { _fixed_slots = n; }
597 void set_inlining_progress(bool z) { _inlining_progress = z; }
598 int inlining_progress() const { return _inlining_progress; }
599 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
600 int inlining_incrementally() const { return _inlining_incrementally; }
601 void set_do_cleanup(bool z) { _do_cleanup = z; }
602 int do_cleanup() const { return _do_cleanup; }
603 bool major_progress() const { return _major_progress; }
604 void set_major_progress() { _major_progress = true; }
605 void restore_major_progress(bool progress) { _major_progress = _major_progress || progress; }
606 void clear_major_progress() { _major_progress = false; }
607 int max_inline_size() const { return _max_inline_size; }
608 void set_freq_inline_size(int n) { _freq_inline_size = n; }
609 int freq_inline_size() const { return _freq_inline_size; }
610 void set_max_inline_size(int n) { _max_inline_size = n; }
611 bool has_loops() const { return _has_loops; }
612 void set_has_loops(bool z) { _has_loops = z; }
613 bool has_split_ifs() const { return _has_split_ifs; }
614 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
615 bool has_unsafe_access() const { return _has_unsafe_access; }
616 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
617 bool has_stringbuilder() const { return _has_stringbuilder; }
618 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
619 bool has_boxed_value() const { return _has_boxed_value; }
620 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
621 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
622 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
623 uint max_vector_size() const { return _max_vector_size; }
624 void set_max_vector_size(uint s) { _max_vector_size = s; }
625 bool clear_upper_avx() const { return _clear_upper_avx; }
626 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
627 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
628 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
629 bool trap_can_recompile() const { return _trap_can_recompile; }
630 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
631 uint decompile_count() const { return _decompile_count; }
632 void set_decompile_count(uint c) { _decompile_count = c; }
633 bool allow_range_check_smearing() const;
634 bool do_inlining() const { return _do_inlining; }
635 void set_do_inlining(bool z) { _do_inlining = z; }
636 bool do_scheduling() const { return _do_scheduling; }
637 void set_do_scheduling(bool z) { _do_scheduling = z; }
638 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
639 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
640 bool do_vector_loop() const { return _do_vector_loop; }
641 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
642 bool use_cmove() const { return _use_cmove; }
643 void set_use_cmove(bool z) { _use_cmove = z; }
644 bool do_aliasing() const { return _do_aliasing; }
645 bool print_assembly() const { return _print_assembly; }
646 void set_print_assembly(bool z) { _print_assembly = z; }
647 bool print_inlining() const { return _print_inlining; }
648 void set_print_inlining(bool z) { _print_inlining = z; }
649 bool print_intrinsics() const { return _print_intrinsics; }
650 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
651 uint max_node_limit() const { return (uint)_max_node_limit; }
652 void set_max_node_limit(uint n) { _max_node_limit = n; }
653 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
654 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
655 bool has_monitors() const { return _has_monitors; }
656 void set_has_monitors(bool v) { _has_monitors = v; }
657 bool has_scoped_access() const { return _has_scoped_access; }
658 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
659
660 // check the CompilerOracle for special behaviours for this compile
661 bool method_has_option(CompileCommandEnum option) const {
662 return method() != nullptr && method()->has_option(option);
663 }
664
665 #ifndef PRODUCT
666 uint next_igv_idx() { return _igv_idx++; }
667 bool trace_opto_output() const { return _trace_opto_output; }
668 void print_phase(const char* phase_name);
669 void print_ideal_ir(const char* phase_name);
670 bool should_print_ideal() const { return _directive->PrintIdealOption; }
671 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
672 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
673 int _in_dump_cnt; // Required for dumping ir nodes.
674 #endif
675 bool has_irreducible_loop() const { return _has_irreducible_loop; }
676 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
677
678 Ticks _latest_stage_start_counter;
679
680 void begin_method();
681 void end_method();
682
683 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
684
685 #ifndef PRODUCT
686 bool should_print_igv(int level);
687 bool should_print_phase(int level) const;
688 bool should_print_ideal_phase(CompilerPhaseType cpt) const;
689 void init_igv();
690 void dump_igv(const char* graph_name, int level = 3) {
691 if (should_print_igv(level)) {
692 _igv_printer->print_graph(graph_name, nullptr);
693 }
694 }
695
696 void igv_print_method_to_file(const char* phase_name = nullptr, bool append = false, const frame* fr = nullptr);
697 void igv_print_method_to_network(const char* phase_name = nullptr, const frame* fr = nullptr);
698 void igv_print_graph_to_network(const char* name, GrowableArray<const Node*>& visible_nodes, const frame* fr);
699 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
700 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
701 #endif
702
703 const GrowableArray<ParsePredicateNode*>& parse_predicates() const {
704 return _parse_predicates;
705 }
706
707 const GrowableArray<OpaqueTemplateAssertionPredicateNode*>& template_assertion_predicate_opaques() const {
708 return _template_assertion_predicate_opaques;
709 }
710
711 int macro_count() const { return _macro_nodes.length(); }
712 int parse_predicate_count() const { return _parse_predicates.length(); }
713 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaques.length(); }
714 int expensive_count() const { return _expensive_nodes.length(); }
715 int coarsened_count() const { return _coarsened_locks.length(); }
716
717 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
718
719 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
720
721 ConnectionGraph* congraph() { return _congraph;}
722 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
723 void add_macro_node(Node * n) {
724 //assert(n->is_macro(), "must be a macro node");
725 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
726 _macro_nodes.append(n);
727 }
728 void remove_macro_node(Node* n) {
729 // this function may be called twice for a node so we can only remove it
730 // if it's still existing.
731 _macro_nodes.remove_if_existing(n);
732 // Remove from coarsened locks list if present
733 if (coarsened_count() > 0) {
734 remove_coarsened_lock(n);
735 }
736 }
737 void add_expensive_node(Node* n);
738 void remove_expensive_node(Node* n) {
739 _expensive_nodes.remove_if_existing(n);
740 }
741
742 void add_parse_predicate(ParsePredicateNode* n) {
743 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
744 _parse_predicates.append(n);
745 }
746
747 void remove_parse_predicate(ParsePredicateNode* n) {
748 if (parse_predicate_count() > 0) {
749 _parse_predicates.remove_if_existing(n);
750 }
751 }
752
753 void add_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
754 assert(!_template_assertion_predicate_opaques.contains(n),
755 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
756 _template_assertion_predicate_opaques.append(n);
757 }
758
759 void remove_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
760 if (template_assertion_predicate_count() > 0) {
761 _template_assertion_predicate_opaques.remove_if_existing(n);
762 }
763 }
764 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
765 void remove_coarsened_lock(Node* n);
766 bool coarsened_locks_consistent();
767 void mark_unbalanced_boxes() const;
768
769 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
770 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
771 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
772
773 #ifdef ASSERT
774 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
775 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
776 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
777 #endif
778
779 bool allow_macro_nodes() { return _allow_macro_nodes; }
780 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
781
782 void record_for_post_loop_opts_igvn(Node* n);
783 void remove_from_post_loop_opts_igvn(Node* n);
784 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
785
786 void record_unstable_if_trap(UnstableIfTrap* trap);
787 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
788 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
789 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
790
791 bool merge_stores_phase() { return _merge_stores_phase; }
792 void set_merge_stores_phase() { _merge_stores_phase = true; }
793 void record_for_merge_stores_igvn(Node* n);
794 void remove_from_merge_stores_igvn(Node* n);
795 void process_for_merge_stores_igvn(PhaseIterGVN& igvn);
796
797 void shuffle_macro_nodes();
798 void sort_macro_nodes();
799
800 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
801
802 // Are there candidate expensive nodes for optimization?
803 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
804 // Check whether n1 and n2 are similar
805 static int cmp_expensive_nodes(Node* n1, Node* n2);
806 // Sort expensive nodes to locate similar expensive nodes
807 void sort_expensive_nodes();
808
809 // Compilation environment.
810 Arena* comp_arena() { return &_comp_arena; }
811 ciEnv* env() const { return _env; }
812 CompileLog* log() const { return _log; }
813
814 bool failing_internal() const {
815 return _env->failing() ||
816 _failure_reason.get() != nullptr;
817 }
818
819 const char* failure_reason() const {
820 return _env->failing() ? _env->failure_reason()
821 : _failure_reason.get();
822 }
823
824 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
825
826 bool failing() {
827 if (failing_internal()) {
828 return true;
829 }
830 #ifdef ASSERT
831 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
832 if (phase_verify_ideal_loop()) {
833 return false;
834 }
835 if (StressBailout) {
836 return fail_randomly();
837 }
838 #endif
839 return false;
840 }
841
842 #ifdef ASSERT
843 bool fail_randomly();
844 bool failure_is_artificial();
845 #endif
846
847 bool failure_reason_is(const char* r) const {
848 return (r == _failure_reason.get()) ||
849 (r != nullptr &&
850 _failure_reason.get() != nullptr &&
851 strcmp(r, _failure_reason.get()) == 0);
852 }
853
854 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
855 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
856 env()->record_method_not_compilable(reason);
857 // Record failure reason.
858 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
859 }
860 bool check_node_count(uint margin, const char* reason) {
861 if (oom()) {
862 record_method_not_compilable_oom();
863 return true;
864 }
865 if (live_nodes() + margin > max_node_limit()) {
866 record_method_not_compilable(reason);
867 return true;
868 } else {
869 return false;
870 }
871 }
872 bool oom() const { return _oom; }
873 void set_oom() { _oom = true; }
874
875 // Node management
876 uint unique() const { return _unique; }
877 uint next_unique() { return _unique++; }
878 void set_unique(uint i) { _unique = i; }
879 Arena* node_arena() { return _node_arena; }
880 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
881 RootNode* root() const { return _root; }
882 void set_root(RootNode* r) { _root = r; }
883 StartNode* start() const; // (Derived from root.)
884 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
885 Node* immutable_memory();
886
887 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
888 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
889 void set_recent_alloc(Node* ctl, Node* obj) {
890 _recent_alloc_ctl = ctl;
891 _recent_alloc_obj = obj;
892 }
893 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
894 _dead_node_count++;
895 }
896 void reset_dead_node_list() { _dead_node_list.reset();
897 _dead_node_count = 0;
898 }
899 uint live_nodes() const {
900 int val = _unique - _dead_node_count;
901 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
902 return (uint) val;
903 }
904 #ifdef ASSERT
905 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
906 bool phase_optimize_finished() const { return _phase_optimize_finished; }
907 uint count_live_nodes_by_graph_walk();
908 void print_missing_nodes();
909 #endif
910
911 // Record modified nodes to check that they are put on IGVN worklist
912 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
913 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
914 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
915
916 MachConstantBaseNode* mach_constant_base_node();
917 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
918 // Generated by adlc, true if CallNode requires MachConstantBase.
919 bool needs_deep_clone_jvms();
920
921 // Handy undefined Node
922 Node* top() const { return _top; }
923
924 // these are used by guys who need to know about creation and transformation of top:
925 Node* cached_top_node() { return _top; }
926 void set_cached_top_node(Node* tn);
927
928 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
929 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
930 Node_Notes* default_node_notes() const { return _default_node_notes; }
931 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
932
933 Node_Notes* node_notes_at(int idx);
934
935 inline bool set_node_notes_at(int idx, Node_Notes* value);
936 // Copy notes from source to dest, if they exist.
937 // Overwrite dest only if source provides something.
938 // Return true if information was moved.
939 bool copy_node_notes_to(Node* dest, Node* source);
940
941 // Workhorse function to sort out the blocked Node_Notes array:
942 Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
943 int idx, bool can_grow = false);
944
945 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
946
947 // Type management
948 Arena* type_arena() { return _type_arena; }
949 Dict* type_dict() { return _type_dict; }
950 size_t type_last_size() { return _type_last_size; }
951 int num_alias_types() { return _num_alias_types; }
952
953 void init_type_arena() { _type_arena = &_Compile_types; }
954 void set_type_arena(Arena* a) { _type_arena = a; }
955 void set_type_dict(Dict* d) { _type_dict = d; }
956 void set_type_last_size(size_t sz) { _type_last_size = sz; }
957
958 const TypeFunc* last_tf(ciMethod* m) {
959 return (m == _last_tf_m) ? _last_tf : nullptr;
960 }
961 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
962 assert(m != nullptr || tf == nullptr, "");
963 _last_tf_m = m;
964 _last_tf = tf;
965 }
966
967 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
968 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); }
969 bool have_alias_type(const TypePtr* adr_type);
970 AliasType* alias_type(ciField* field);
971
972 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
973 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
974 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
975
976 // Building nodes
977 void rethrow_exceptions(JVMState* jvms);
978 void return_values(JVMState* jvms);
979 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
980
981 // Decide how to build a call.
982 // The profile factor is a discount to apply to this site's interp. profile.
983 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
984 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
985 bool allow_intrinsics = true);
986 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
987 return C->directive()->should_delay_inline(call_method) ||
988 should_delay_string_inlining(call_method, jvms) ||
989 should_delay_boxing_inlining(call_method, jvms) ||
990 should_delay_vector_inlining(call_method, jvms);
991 }
992 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
993 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
994 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
995 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
996
997 // Helper functions to identify inlining potential at call-site
998 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
999 ciKlass* holder, ciMethod* callee,
1000 const TypeOopPtr* receiver_type, bool is_virtual,
1001 bool &call_does_dispatch, int &vtable_index,
1002 bool check_access = true);
1003 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1004 ciMethod* callee, const TypeOopPtr* receiver_type,
1005 bool check_access = true);
1006
1007 // Report if there were too many traps at a current method and bci.
1008 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1009 // If there is no MDO at all, report no trap unless told to assume it.
1010 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1011 // This version, unspecific to a particular bci, asks if
1012 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1013 bool too_many_traps(Deoptimization::DeoptReason reason,
1014 // Privately used parameter for logging:
1015 ciMethodData* logmd = nullptr);
1016 // Report if there were too many recompiles at a method and bci.
1017 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1018 // Report if there were too many traps or recompiles at a method and bci.
1019 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1020 return too_many_traps(method, bci, reason) ||
1021 too_many_recompiles(method, bci, reason);
1022 }
1023 // Return a bitset with the reasons where deoptimization is allowed,
1024 // i.e., where there were not too many uncommon traps.
1025 int _allowed_reasons;
1026 int allowed_deopt_reasons() { return _allowed_reasons; }
1027 void set_allowed_deopt_reasons();
1028
1029 // Parsing, optimization
1030 PhaseGVN* initial_gvn() { return _initial_gvn; }
1031 Unique_Node_List* igvn_worklist() {
1032 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1033 return _igvn_worklist;
1034 }
1035 Type_Array* types() {
1036 assert(_types != nullptr, "must be created in Compile::Compile");
1037 return _types;
1038 }
1039 NodeHash* node_hash() {
1040 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1041 return _node_hash;
1042 }
1043 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1044 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1045 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1046
1047 // Replace n by nn using initial_gvn, calling hash_delete and
1048 // record_for_igvn as needed.
1049 void gvn_replace_by(Node* n, Node* nn);
1050
1051
1052 void identify_useful_nodes(Unique_Node_List &useful);
1053 void update_dead_node_list(Unique_Node_List &useful);
1054 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist, const Unique_Node_List* root_and_safepoints = nullptr);
1055
1056 void remove_useless_node(Node* dead);
1057
1058 // Record this CallGenerator for inlining at the end of parsing.
1059 void add_late_inline(CallGenerator* cg) {
1060 _late_inlines.insert_before(_late_inlines_pos, cg);
1061 _late_inlines_pos++;
1062 }
1063
1064 void prepend_late_inline(CallGenerator* cg) {
1065 _late_inlines.insert_before(0, cg);
1066 }
1067
1068 void add_string_late_inline(CallGenerator* cg) {
1069 _string_late_inlines.push(cg);
1070 }
1071
1072 void add_boxing_late_inline(CallGenerator* cg) {
1073 _boxing_late_inlines.push(cg);
1074 }
1075
1076 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1077 _vector_reboxing_late_inlines.push(cg);
1078 }
1079
1080 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1081 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1082
1083 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1084 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1085
1086 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1087
1088 void dump_print_inlining();
1089
1090 bool over_inlining_cutoff() const {
1091 if (!inlining_incrementally()) {
1092 return unique() > (uint)NodeCountInliningCutoff;
1093 } else {
1094 // Give some room for incremental inlining algorithm to "breathe"
1095 // and avoid thrashing when live node count is close to the limit.
1096 // Keep in mind that live_nodes() isn't accurate during inlining until
1097 // dead node elimination step happens (see Compile::inline_incrementally).
1098 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1099 }
1100 }
1101
1102 void mark_has_mh_late_inlines() { _has_mh_late_inlines = true; }
1103 bool has_mh_late_inlines() const { return _has_mh_late_inlines; }
1104
1105 bool inline_incrementally_one();
1106 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1107 void inline_incrementally(PhaseIterGVN& igvn);
1108 bool should_stress_inlining() { return StressIncrementalInlining && (random() % 2) == 0; }
1109 bool should_delay_inlining() { return AlwaysIncrementalInline || should_stress_inlining(); }
1110 void inline_string_calls(bool parse_time);
1111 void inline_boxing_calls(PhaseIterGVN& igvn);
1112 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1113 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1114
1115 void inline_vector_reboxing_calls();
1116 bool has_vbox_nodes();
1117
1118 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1119
1120 // Matching, CFG layout, allocation, code generation
1121 PhaseCFG* cfg() { return _cfg; }
1122 bool has_java_calls() const { return _java_calls > 0; }
1123 int java_calls() const { return _java_calls; }
1124 int inner_loops() const { return _inner_loops; }
1125 Matcher* matcher() { return _matcher; }
1126 PhaseRegAlloc* regalloc() { return _regalloc; }
1127 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1128 ResourceArea* regmask_arena() { return &_regmask_arena; }
1129 Arena* indexSet_arena() { return _indexSet_arena; }
1130 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1131 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1132
1133 void update_interpreter_frame_size(int size) {
1134 if (_interpreter_frame_size < size) {
1135 _interpreter_frame_size = size;
1136 }
1137 }
1138
1139 void set_matcher(Matcher* m) { _matcher = m; }
1140 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1141 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1142 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1143
1144 void set_java_calls(int z) { _java_calls = z; }
1145 void set_inner_loops(int z) { _inner_loops = z; }
1146
1147 Dependencies* dependencies() { return env()->dependencies(); }
1148
1149 // Major entry point. Given a Scope, compile the associated method.
1150 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1151 // replacement, entry_bci indicates the bytecode for which to compile a
1152 // continuation.
1153 Compile(ciEnv* ci_env, ciMethod* target,
1154 int entry_bci, Options options, DirectiveSet* directive);
1155
1156 // Second major entry point. From the TypeFunc signature, generate code
1157 // to pass arguments from the Java calling convention to the C calling
1158 // convention.
1159 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1160 address stub_function, const char *stub_name,
1161 StubId stub_id, int is_fancy_jump, bool pass_tls,
1162 bool return_pc, DirectiveSet* directive);
1163
1164 ~Compile();
1165
1166 // Are we compiling a method?
1167 bool has_method() { return method() != nullptr; }
1168
1169 // Maybe print some information about this compile.
1170 void print_compile_messages();
1171
1172 // Final graph reshaping, a post-pass after the regular optimizer is done.
1173 bool final_graph_reshaping();
1174
1175 // returns true if adr is completely contained in the given alias category
1176 bool must_alias(const TypePtr* adr, int alias_idx);
1177
1178 // returns true if adr overlaps with the given alias category
1179 bool can_alias(const TypePtr* adr, int alias_idx);
1180
1181 // Stack slots that may be unused by the calling convention but must
1182 // otherwise be preserved. On Intel this includes the return address.
1183 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1184 uint in_preserve_stack_slots() {
1185 return SharedRuntime::in_preserve_stack_slots();
1186 }
1187
1188 // "Top of Stack" slots that may be unused by the calling convention but must
1189 // otherwise be preserved.
1190 // On Intel these are not necessary and the value can be zero.
1191 static uint out_preserve_stack_slots() {
1192 return SharedRuntime::out_preserve_stack_slots();
1193 }
1194
1195 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1196 // for calls to C. Supports the var-args backing area for register parms.
1197 uint varargs_C_out_slots_killed() const;
1198
1199 // Number of Stack Slots consumed by a synchronization entry
1200 int sync_stack_slots() const;
1201
1202 // Compute the name of old_SP. See <arch>.ad for frame layout.
1203 OptoReg::Name compute_old_SP();
1204
1205 private:
1206 // Phase control:
1207 void Init(bool aliasing); // Prepare for a single compilation
1208 void Optimize(); // Given a graph, optimize it
1209 void Code_Gen(); // Generate code from a graph
1210
1211 // Management of the AliasType table.
1212 void grow_alias_types();
1213 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1214 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1215 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1216
1217 void verify_top(Node*) const PRODUCT_RETURN;
1218
1219 // Intrinsic setup.
1220 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1221 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1222 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1223 void register_intrinsic(CallGenerator* cg); // update fn
1224
1225 #ifndef PRODUCT
1226 static juint _intrinsic_hist_count[];
1227 static jubyte _intrinsic_hist_flags[];
1228 #endif
1229 // Function calls made by the public function final_graph_reshaping.
1230 // No need to be made public as they are not called elsewhere.
1231 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1232 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1233 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1234 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1235
1236 // Logic cone optimization.
1237 void optimize_logic_cones(PhaseIterGVN &igvn);
1238 void collect_logic_cone_roots(Unique_Node_List& list);
1239 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1240 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1241 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1242 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1243 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1244 void check_no_dead_use() const NOT_DEBUG_RETURN;
1245
1246 public:
1247
1248 // Note: Histogram array size is about 1 Kb.
1249 enum { // flag bits:
1250 _intrinsic_worked = 1, // succeeded at least once
1251 _intrinsic_failed = 2, // tried it but it failed
1252 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1253 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1254 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1255 };
1256 // Update histogram. Return boolean if this is a first-time occurrence.
1257 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1258 bool is_virtual, int flags) PRODUCT_RETURN0;
1259 static void print_intrinsic_statistics() PRODUCT_RETURN;
1260
1261 // Graph verification code
1262 // Walk the node list, verifying that there is a one-to-one correspondence
1263 // between Use-Def edges and Def-Use edges. The option no_dead_code enables
1264 // stronger checks that the graph is strongly connected from starting points
1265 // in both directions.
1266 // root_and_safepoints is used to give the starting points for the traversal.
1267 // If not supplied, only root is used. When this check is called after CCP,
1268 // we need to start traversal from Root and safepoints, just like CCP does its
1269 // own traversal (see PhaseCCP::transform for reasons).
1270 //
1271 // To call this function, there are 2 ways to go:
1272 // - give root_and_safepoints to start traversal everywhere needed (like after CCP)
1273 // - if the whole graph is assumed to be reachable from Root's input,
1274 // root_and_safepoints is not needed (like in PhaseRemoveUseless).
1275 //
1276 // Failure to specify root_and_safepoints in case the graph is not fully
1277 // reachable from Root's input make this check unsound (can miss inconsistencies)
1278 // and even incomplete (can make up non-existing problems) if no_dead_code is
1279 // true.
1280 void verify_graph_edges(bool no_dead_code = false, const Unique_Node_List* root_and_safepoints = nullptr) const PRODUCT_RETURN;
1281
1282 // Verify bi-directional correspondence of edges
1283 void verify_bidirectional_edges(Unique_Node_List& visited, const Unique_Node_List* root_and_safepoints = nullptr) const;
1284
1285 // End-of-run dumps.
1286 static void print_statistics() PRODUCT_RETURN;
1287
1288 // Verify ADLC assumptions during startup
1289 static void adlc_verification() PRODUCT_RETURN;
1290
1291 // Definitions of pd methods
1292 static void pd_compiler2_init();
1293
1294 // Static parse-time type checking logic for gen_subtype_check:
1295 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1296 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1297
1298 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1299 // Optional control dependency (for example, on range check)
1300 Node* ctrl = nullptr);
1301
1302 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1303 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1304
1305 // Auxiliary methods for randomized fuzzing/stressing
1306 int random();
1307 bool randomized_select(int count);
1308
1309 // seed random number generation and log the seed for repeatability.
1310 void initialize_stress_seed(const DirectiveSet* directive);
1311
1312 // supporting clone_map
1313 CloneMap& clone_map();
1314 void set_clone_map(Dict* d);
1315
1316 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1317 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1318 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1319
1320 #ifdef ASSERT
1321 VerifyMeetResult* _type_verify;
1322 void set_exception_backedge() { _exception_backedge = true; }
1323 bool has_exception_backedge() const { return _exception_backedge; }
1324 #endif
1325
1326 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1327 BasicType out_bt, BasicType in_bt);
1328
1329 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1330
1331 #ifndef PRODUCT
1332 private:
1333 // getting rid of the template makes things easier
1334 Node* make_debug_print_call(const char* str, address call_addr, PhaseGVN* gvn,
1335 Node* parm0 = nullptr, Node* parm1 = nullptr,
1336 Node* parm2 = nullptr, Node* parm3 = nullptr,
1337 Node* parm4 = nullptr, Node* parm5 = nullptr,
1338 Node* parm6 = nullptr) const;
1339
1340 public:
1341 // Creates a CallLeafNode for a runtime call that prints a static string and the values of the
1342 // nodes passed as arguments.
1343 // This function also takes care of doing the necessary wiring, including finding a suitable control
1344 // based on the nodes that need to be printed. Note that passing nodes that have incompatible controls
1345 // is undefined behavior.
1346 template <typename... TT, typename... NN>
1347 Node* make_debug_print(const char* str, PhaseGVN* gvn, NN... in) {
1348 address call_addr = CAST_FROM_FN_PTR(address, SharedRuntime::debug_print<TT...>);
1349 return make_debug_print_call(str, call_addr, gvn, in...);
1350 }
1351 #endif
1352 };
1353
1354 #endif // SHARE_OPTO_COMPILE_HPP