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 _for_preload; // Generate code for preload (before Java method execution), do class init barriers
186 const bool _do_superword; // Do SuperWord
187 const bool _install_code; // Install the code that was compiled
188 public:
189 Options(bool subsume_loads,
190 bool do_escape_analysis,
191 bool do_iterative_escape_analysis,
192 bool do_reduce_allocation_merges,
193 bool eliminate_boxing,
194 bool do_locks_coarsening,
195 bool do_superword,
196 bool for_preload,
197 bool install_code) :
198 _subsume_loads(subsume_loads),
199 _do_escape_analysis(do_escape_analysis),
200 _do_iterative_escape_analysis(do_iterative_escape_analysis),
201 _do_reduce_allocation_merges(do_reduce_allocation_merges),
202 _eliminate_boxing(eliminate_boxing),
203 _do_locks_coarsening(do_locks_coarsening),
204 _for_preload(for_preload),
205 _do_superword(do_superword),
206 _install_code(install_code) {
207 }
208
209 static Options for_runtime_stub() {
210 return Options(
211 /* subsume_loads = */ true,
212 /* do_escape_analysis = */ false,
213 /* do_iterative_escape_analysis = */ false,
214 /* do_reduce_allocation_merges = */ false,
215 /* eliminate_boxing = */ false,
216 /* do_lock_coarsening = */ false,
217 /* for_preload = */ false,
218 /* do_superword = */ true,
219 /* install_code = */ true
220 );
221 }
222 };
223
224 //------------------------------Compile----------------------------------------
225 // This class defines a top-level Compiler invocation.
226
227 class Compile : public Phase {
228 friend class VMStructs;
229
230 public:
231 // Fixed alias indexes. (See also MergeMemNode.)
232 enum {
233 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
234 AliasIdxBot = 2, // pseudo-index, aliases to everything
235 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
236 };
237
238 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
239 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
240 // then brackets are put into the log, with time stamps and node counts.
241 // (The time collection itself is always conditionalized on CITime.)
242 class TracePhase : public TraceTime {
243 private:
244 Compile* _compile;
245 CompileLog* _log;
246 const char* _phase_name;
247 bool _dolog;
248 public:
249 TracePhase(const char* name, elapsedTimer* accumulator);
250 ~TracePhase();
251 };
252
253 // Information per category of alias (memory slice)
254 class AliasType {
255 private:
256 friend class Compile;
257
258 int _index; // unique index, used with MergeMemNode
259 const TypePtr* _adr_type; // normalized address type
260 ciField* _field; // relevant instance field, or null if none
261 const Type* _element; // relevant array element type, or null if none
262 bool _is_rewritable; // false if the memory is write-once only
263 int _general_index; // if this is type is an instance, the general
264 // type that this is an instance of
265
266 void Init(int i, const TypePtr* at);
267
268 public:
269 int index() const { return _index; }
270 const TypePtr* adr_type() const { return _adr_type; }
271 ciField* field() const { return _field; }
272 const Type* element() const { return _element; }
273 bool is_rewritable() const { return _is_rewritable; }
274 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
275 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
276
277 void set_rewritable(bool z) { _is_rewritable = z; }
278 void set_field(ciField* f) {
279 assert(!_field,"");
280 _field = f;
281 if (f->is_final() || f->is_stable()) {
282 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
283 _is_rewritable = false;
284 }
285 }
286 void set_element(const Type* e) {
287 assert(_element == nullptr, "");
288 _element = e;
289 }
290
291 BasicType basic_type() const;
292
293 void print_on(outputStream* st) PRODUCT_RETURN;
294 };
295
296 enum {
297 logAliasCacheSize = 6,
298 AliasCacheSize = (1<<logAliasCacheSize)
299 };
300 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
301 enum {
302 trapHistLength = MethodData::_trap_hist_limit
303 };
304
305 private:
306 // Fixed parameters to this compilation.
307 const int _compile_id;
308 const Options _options; // Compilation options
309 ciMethod* _method; // The method being compiled.
310 int _entry_bci; // entry bci for osr methods.
311 const TypeFunc* _tf; // My kind of signature
312 InlineTree* _ilt; // Ditto (temporary).
313 address _stub_function; // VM entry for stub being compiled, or null
314 const char* _stub_name; // Name of stub or adapter being compiled, or null
315 address _stub_entry_point; // Compile code entry for generated stub, or null
316
317 // Control of this compilation.
318 int _max_inline_size; // Max inline size for this compilation
319 int _freq_inline_size; // Max hot method inline size for this compilation
320 int _fixed_slots; // count of frame slots not allocated by the register
321 // allocator i.e. locks, original deopt pc, etc.
322 uintx _max_node_limit; // Max unique node count during a single compilation.
323
324 bool _post_loop_opts_phase; // Loop opts are finished.
325 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
326
327 int _major_progress; // Count of something big happening
328 bool _inlining_progress; // progress doing incremental inlining?
329 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
330 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
331 bool _has_loops; // True if the method _may_ have some loops
332 bool _has_split_ifs; // True if the method _may_ have some split-if
333 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
334 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
335 bool _has_boxed_value; // True if a boxed object is allocated
336 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
337 uint _max_vector_size; // Maximum size of generated vectors
338 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
339 uint _trap_hist[trapHistLength]; // Cumulative traps
340 bool _trap_can_recompile; // Have we emitted a recompiling trap?
341 uint _decompile_count; // Cumulative decompilation counts.
342 bool _do_inlining; // True if we intend to do inlining
343 bool _do_scheduling; // True if we intend to do scheduling
344 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
345 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
346 bool _use_cmove; // True if CMove should be used without profitability analysis
347 bool _do_aliasing; // True if we intend to do aliasing
348 bool _print_assembly; // True if we should dump assembly code for this compilation
349 bool _print_inlining; // True if we should print inlining for this compilation
350 bool _print_intrinsics; // True if we should print intrinsics for this compilation
351 #ifndef PRODUCT
352 uint _igv_idx; // Counter for IGV node identifiers
353 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
354 bool _trace_opto_output;
355 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
356 #endif
357 bool _has_irreducible_loop; // Found irreducible loops
358 // JSR 292
359 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
360 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
361 bool _has_scoped_access; // For shared scope closure
362 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
363 bool _has_clinit_barriers; // True if compiled code has clinit barriers
364 int _loop_opts_cnt; // loop opts round
365 uint _stress_seed; // Seed for stress testing
366
367 // Compilation environment.
368 Arena _comp_arena; // Arena with lifetime equivalent to Compile
369 void* _barrier_set_state; // Potential GC barrier state for Compile
370 ciEnv* _env; // CI interface
371 DirectiveSet* _directive; // Compiler directive
372 CompileLog* _log; // from CompilerThread
373 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
374 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
375 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
376 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
377 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
378 GrowableArray<Node*> _template_assertion_predicate_opaqs; // List of Opaque4 nodes for Template Assertion Predicates.
379 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
380 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
381 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
382 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
383 ConnectionGraph* _congraph;
384 #ifndef PRODUCT
385 IdealGraphPrinter* _igv_printer;
386 static IdealGraphPrinter* _debug_file_printer;
387 static IdealGraphPrinter* _debug_network_printer;
388 #endif
389
390
391 // Node management
392 uint _unique; // Counter for unique Node indices
393 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
394 // So use this to keep count and make the call O(1).
395 VectorSet _dead_node_list; // Set of dead nodes
396 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
397 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
398
399 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
400
401 // Arenas for new-space and old-space nodes.
402 // Swapped between using _node_arena.
403 // The lifetime of the old-space nodes is during xform.
404 Arena _node_arena_one;
405 Arena _node_arena_two;
406 Arena* _node_arena;
407 public:
408 Arena* swap_old_and_new() {
409 Arena* filled_arena_ptr = _node_arena;
410 Arena* old_arena_ptr = old_arena();
411 old_arena_ptr->destruct_contents();
412 _node_arena = old_arena_ptr;
413 return filled_arena_ptr;
414 }
415 private:
416 RootNode* _root; // Unique root of compilation, or null after bail-out.
417 Node* _top; // Unique top node. (Reset by various phases.)
418
419 Node* _immutable_memory; // Initial memory state
420
421 Node* _recent_alloc_obj;
422 Node* _recent_alloc_ctl;
423
424 // Constant table
425 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
426
427
428 // Blocked array of debugging and profiling information,
429 // tracked per node.
430 enum { _log2_node_notes_block_size = 8,
431 _node_notes_block_size = (1<<_log2_node_notes_block_size)
432 };
433 GrowableArray<Node_Notes*>* _node_note_array;
434 Node_Notes* _default_node_notes; // default notes for new nodes
435
436 // After parsing and every bulk phase we hang onto the Root instruction.
437 // The RootNode instruction is where the whole program begins. It produces
438 // the initial Control and BOTTOM for everybody else.
439
440 // Type management
441 Arena _Compile_types; // Arena for all types
442 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
443 Dict* _type_dict; // Intern table
444 CloneMap _clone_map; // used for recording history of cloned nodes
445 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
446 ciMethod* _last_tf_m; // Cache for
447 const TypeFunc* _last_tf; // TypeFunc::make
448 AliasType** _alias_types; // List of alias types seen so far.
449 int _num_alias_types; // Logical length of _alias_types
450 int _max_alias_types; // Physical length of _alias_types
451 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
452
453 // Parsing, optimization
454 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
455
456 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
457 // If pushed outside IGVN, the Node is processed in the next IGVN round.
458 Unique_Node_List* _igvn_worklist;
459
460 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
461 Type_Array* _types;
462
463 // Shared node hash table for GVN, IGVN and CCP.
464 NodeHash* _node_hash;
465
466 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
467 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
468 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
469
470 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
471
472 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
473 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
474
475 // "MemLimit" directive was specified and the memory limit was hit during compilation
476 bool _oom;
477
478 // Inlining may not happen in parse order which would make
479 // PrintInlining output confusing. Keep track of PrintInlining
480 // pieces in order.
481 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
482 private:
483 CallGenerator* _cg;
484 stringStream _ss;
485 static const size_t default_stream_buffer_size = 128;
486
487 public:
488 PrintInliningBuffer()
489 : _cg(nullptr), _ss(default_stream_buffer_size) {}
490
491 stringStream* ss() { return &_ss; }
492 CallGenerator* cg() { return _cg; }
493 void set_cg(CallGenerator* cg) { _cg = cg; }
494 };
495
496 stringStream* _print_inlining_stream;
497 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
498 int _print_inlining_idx;
499 char* _print_inlining_output;
500
501 // Only keep nodes in the expensive node list that need to be optimized
502 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
503 // Use for sorting expensive nodes to bring similar nodes together
504 static int cmp_expensive_nodes(Node** n1, Node** n2);
505 // Expensive nodes list already sorted?
506 bool expensive_nodes_sorted() const;
507 // Remove the speculative part of types and clean up the graph
508 void remove_speculative_types(PhaseIterGVN &igvn);
509
510 void* _replay_inline_data; // Pointer to data loaded from file
511
512 void print_inlining_init();
513 void print_inlining_reinit();
514 void print_inlining_commit();
515 void print_inlining_push();
516 PrintInliningBuffer* print_inlining_current();
517
518 void log_late_inline_failure(CallGenerator* cg, const char* msg);
519 DEBUG_ONLY(bool _exception_backedge;)
520
521 void record_method_not_compilable_oom();
522
523 public:
524
525 void* barrier_set_state() const { return _barrier_set_state; }
526
527 stringStream* print_inlining_stream() {
528 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
529 return _print_inlining_stream;
530 }
531
532 void print_inlining_update(CallGenerator* cg);
533 void print_inlining_update_delayed(CallGenerator* cg);
534 void print_inlining_move_to(CallGenerator* cg);
535 void print_inlining_assert_ready();
536 void print_inlining_reset();
537
538 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) {
539 stringStream ss;
540 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg);
541 print_inlining_stream()->print("%s", ss.freeze());
542 }
543
544 #ifndef PRODUCT
545 IdealGraphPrinter* igv_printer() { return _igv_printer; }
546 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
547 #endif
548
549 void log_late_inline(CallGenerator* cg);
550 void log_inline_id(CallGenerator* cg);
551 void log_inline_failure(const char* msg);
552
553 void* replay_inline_data() const { return _replay_inline_data; }
554
555 // Dump inlining replay data to the stream.
556 void dump_inline_data(outputStream* out);
557 void dump_inline_data_reduced(outputStream* out);
558
559 private:
560 // Matching, CFG layout, allocation, code generation
561 PhaseCFG* _cfg; // Results of CFG finding
562 int _java_calls; // Number of java calls in the method
563 int _inner_loops; // Number of inner loops in the method
564 Matcher* _matcher; // Engine to map ideal to machine instructions
565 PhaseRegAlloc* _regalloc; // Results of register allocation.
566 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
567 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
568 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
569 int _interpreter_frame_size;
570
571 PhaseOutput* _output;
572
573 public:
574 // Accessors
575
576 // The Compile instance currently active in this (compiler) thread.
577 static Compile* current() {
578 return (Compile*) ciEnv::current()->compiler_data();
579 }
580
581 int interpreter_frame_size() const { return _interpreter_frame_size; }
582
583 PhaseOutput* output() const { return _output; }
584 void set_output(PhaseOutput* o) { _output = o; }
585
586 // ID for this compilation. Useful for setting breakpoints in the debugger.
587 int compile_id() const { return _compile_id; }
588 DirectiveSet* directive() const { return _directive; }
589
590 // Does this compilation allow instructions to subsume loads? User
591 // instructions that subsume a load may result in an unschedulable
592 // instruction sequence.
593 bool subsume_loads() const { return _options._subsume_loads; }
594 /** Do escape analysis. */
595 bool do_escape_analysis() const { return _options._do_escape_analysis; }
596 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
597 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
598 /** Do boxing elimination. */
599 bool eliminate_boxing() const { return _options._eliminate_boxing; }
600 /** Do aggressive boxing elimination. */
601 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
602 bool should_install_code() const { return _options._install_code; }
603 /** Do locks coarsening. */
604 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
605 bool do_superword() const { return _options._do_superword; }
606
607 bool do_clinit_barriers() const { return _options._for_preload; }
608 bool for_preload() const { return _options._for_preload; }
609
610 // Other fixed compilation parameters.
611 ciMethod* method() const { return _method; }
612 int entry_bci() const { return _entry_bci; }
613 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
614 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
615 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
616 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
617 InlineTree* ilt() const { return _ilt; }
618 address stub_function() const { return _stub_function; }
619 const char* stub_name() const { return _stub_name; }
620 address stub_entry_point() const { return _stub_entry_point; }
621 void set_stub_entry_point(address z) { _stub_entry_point = z; }
622
623 // Control of this compilation.
624 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
625 void set_fixed_slots(int n) { _fixed_slots = n; }
626 int major_progress() const { return _major_progress; }
627 void set_inlining_progress(bool z) { _inlining_progress = z; }
628 int inlining_progress() const { return _inlining_progress; }
629 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
630 int inlining_incrementally() const { return _inlining_incrementally; }
631 void set_do_cleanup(bool z) { _do_cleanup = z; }
632 int do_cleanup() const { return _do_cleanup; }
633 void set_major_progress() { _major_progress++; }
634 void restore_major_progress(int progress) { _major_progress += progress; }
635 void clear_major_progress() { _major_progress = 0; }
636 int max_inline_size() const { return _max_inline_size; }
637 void set_freq_inline_size(int n) { _freq_inline_size = n; }
638 int freq_inline_size() const { return _freq_inline_size; }
639 void set_max_inline_size(int n) { _max_inline_size = n; }
640 bool has_loops() const { return _has_loops; }
641 void set_has_loops(bool z) { _has_loops = z; }
642 bool has_split_ifs() const { return _has_split_ifs; }
643 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
644 bool has_unsafe_access() const { return _has_unsafe_access; }
645 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
646 bool has_stringbuilder() const { return _has_stringbuilder; }
647 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
648 bool has_boxed_value() const { return _has_boxed_value; }
649 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
650 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
651 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
652 uint max_vector_size() const { return _max_vector_size; }
653 void set_max_vector_size(uint s) { _max_vector_size = s; }
654 bool clear_upper_avx() const { return _clear_upper_avx; }
655 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
656 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
657 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
658 bool trap_can_recompile() const { return _trap_can_recompile; }
659 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
660 uint decompile_count() const { return _decompile_count; }
661 void set_decompile_count(uint c) { _decompile_count = c; }
662 bool allow_range_check_smearing() const;
663 bool do_inlining() const { return _do_inlining; }
664 void set_do_inlining(bool z) { _do_inlining = z; }
665 bool do_scheduling() const { return _do_scheduling; }
666 void set_do_scheduling(bool z) { _do_scheduling = z; }
667 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
668 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
669 bool do_vector_loop() const { return _do_vector_loop; }
670 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
671 bool use_cmove() const { return _use_cmove; }
672 void set_use_cmove(bool z) { _use_cmove = z; }
673 bool do_aliasing() const { return _do_aliasing; }
674 bool print_assembly() const { return _print_assembly; }
675 void set_print_assembly(bool z) { _print_assembly = z; }
676 bool print_inlining() const { return _print_inlining; }
677 void set_print_inlining(bool z) { _print_inlining = z; }
678 bool print_intrinsics() const { return _print_intrinsics; }
679 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
680 uint max_node_limit() const { return (uint)_max_node_limit; }
681 void set_max_node_limit(uint n) { _max_node_limit = n; }
682 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
683 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
684 bool has_monitors() const { return _has_monitors; }
685 void set_has_monitors(bool v) { _has_monitors = v; }
686 bool has_scoped_access() const { return _has_scoped_access; }
687 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
688 bool has_clinit_barriers() { return _has_clinit_barriers; }
689 void set_has_clinit_barriers(bool z) { _has_clinit_barriers = z; }
690
691 // check the CompilerOracle for special behaviours for this compile
692 bool method_has_option(CompileCommandEnum option) {
693 return method() != nullptr && method()->has_option(option);
694 }
695
696 #ifndef PRODUCT
697 uint next_igv_idx() { return _igv_idx++; }
698 bool trace_opto_output() const { return _trace_opto_output; }
699 void print_ideal_ir(const char* phase_name);
700 bool should_print_ideal() const { return _directive->PrintIdealOption; }
701 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
702 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
703 int _in_dump_cnt; // Required for dumping ir nodes.
704 #endif
705 bool has_irreducible_loop() const { return _has_irreducible_loop; }
706 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
707
708 // JSR 292
709 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
710 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
711
712 Ticks _latest_stage_start_counter;
713
714 void begin_method();
715 void end_method();
716 bool should_print_igv(int level);
717 bool should_print_phase(CompilerPhaseType cpt);
718
719 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
720
721 #ifndef PRODUCT
722 void dump_igv(const char* graph_name, int level = 3) {
723 if (should_print_igv(level)) {
724 _igv_printer->print_method(graph_name, level);
725 }
726 }
727
728 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
729 void igv_print_method_to_network(const char* phase_name = "Debug");
730 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
731 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
732 #endif
733
734 int macro_count() const { return _macro_nodes.length(); }
735 int parse_predicate_count() const { return _parse_predicates.length(); }
736 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); }
737 int expensive_count() const { return _expensive_nodes.length(); }
738 int coarsened_count() const { return _coarsened_locks.length(); }
739
740 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
741 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); }
742
743 Node* template_assertion_predicate_opaq_node(int idx) const {
744 return _template_assertion_predicate_opaqs.at(idx);
745 }
746
747 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
748
749 ConnectionGraph* congraph() { return _congraph;}
750 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
751 void add_macro_node(Node * n) {
752 //assert(n->is_macro(), "must be a macro node");
753 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
754 _macro_nodes.append(n);
755 }
756 void remove_macro_node(Node* n) {
757 // this function may be called twice for a node so we can only remove it
758 // if it's still existing.
759 _macro_nodes.remove_if_existing(n);
760 // Remove from coarsened locks list if present
761 if (coarsened_count() > 0) {
762 remove_coarsened_lock(n);
763 }
764 }
765 void add_expensive_node(Node* n);
766 void remove_expensive_node(Node* n) {
767 _expensive_nodes.remove_if_existing(n);
768 }
769
770 void add_parse_predicate(ParsePredicateNode* n) {
771 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
772 _parse_predicates.append(n);
773 }
774
775 void remove_parse_predicate(ParsePredicateNode* n) {
776 if (parse_predicate_count() > 0) {
777 _parse_predicates.remove_if_existing(n);
778 }
779 }
780
781 void add_template_assertion_predicate_opaq(Node* n) {
782 assert(!_template_assertion_predicate_opaqs.contains(n),
783 "duplicate entry in template assertion predicate opaque4 list");
784 _template_assertion_predicate_opaqs.append(n);
785 }
786
787 void remove_template_assertion_predicate_opaq(Node* n) {
788 if (template_assertion_predicate_count() > 0) {
789 _template_assertion_predicate_opaqs.remove_if_existing(n);
790 }
791 }
792 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
793 void remove_coarsened_lock(Node* n);
794 bool coarsened_locks_consistent();
795 void mark_unbalanced_boxes() const;
796
797 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
798 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
799 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
800
801 #ifdef ASSERT
802 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
803 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
804 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
805 #endif
806
807 bool allow_macro_nodes() { return _allow_macro_nodes; }
808 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
809
810 void record_for_post_loop_opts_igvn(Node* n);
811 void remove_from_post_loop_opts_igvn(Node* n);
812 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
813
814 void record_unstable_if_trap(UnstableIfTrap* trap);
815 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
816 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
817 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
818
819 void shuffle_macro_nodes();
820 void sort_macro_nodes();
821
822 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
823
824 // Are there candidate expensive nodes for optimization?
825 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
826 // Check whether n1 and n2 are similar
827 static int cmp_expensive_nodes(Node* n1, Node* n2);
828 // Sort expensive nodes to locate similar expensive nodes
829 void sort_expensive_nodes();
830
831 // Compilation environment.
832 Arena* comp_arena() { return &_comp_arena; }
833 ciEnv* env() const { return _env; }
834 CompileLog* log() const { return _log; }
835
836 bool failing_internal() const {
837 return _env->failing() ||
838 _failure_reason.get() != nullptr;
839 }
840
841 const char* failure_reason() const {
842 return _env->failing() ? _env->failure_reason()
843 : _failure_reason.get();
844 }
845
846 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
847
848 bool failing() {
849 if (failing_internal()) {
850 return true;
851 }
852 #ifdef ASSERT
853 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
854 if (phase_verify_ideal_loop()) {
855 return false;
856 }
857 if (StressBailout) {
858 return fail_randomly();
859 }
860 #endif
861 return false;
862 }
863
864 #ifdef ASSERT
865 bool fail_randomly();
866 bool failure_is_artificial();
867 #endif
868
869 bool failure_reason_is(const char* r) const {
870 return (r == _failure_reason.get()) ||
871 (r != nullptr &&
872 _failure_reason.get() != nullptr &&
873 strcmp(r, _failure_reason.get()) == 0);
874 }
875
876 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
877 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
878 env()->record_method_not_compilable(reason);
879 // Record failure reason.
880 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
881 }
882 bool check_node_count(uint margin, const char* reason) {
883 if (oom()) {
884 record_method_not_compilable_oom();
885 return true;
886 }
887 if (live_nodes() + margin > max_node_limit()) {
888 record_method_not_compilable(reason);
889 return true;
890 } else {
891 return false;
892 }
893 }
894 bool oom() const { return _oom; }
895 void set_oom() { _oom = true; }
896
897 // Node management
898 uint unique() const { return _unique; }
899 uint next_unique() { return _unique++; }
900 void set_unique(uint i) { _unique = i; }
901 Arena* node_arena() { return _node_arena; }
902 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
903 RootNode* root() const { return _root; }
904 void set_root(RootNode* r) { _root = r; }
905 StartNode* start() const; // (Derived from root.)
906 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
907 Node* immutable_memory();
908
909 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
910 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
911 void set_recent_alloc(Node* ctl, Node* obj) {
912 _recent_alloc_ctl = ctl;
913 _recent_alloc_obj = obj;
914 }
915 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
916 _dead_node_count++;
917 }
918 void reset_dead_node_list() { _dead_node_list.reset();
919 _dead_node_count = 0;
920 }
921 uint live_nodes() const {
922 int val = _unique - _dead_node_count;
923 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
924 return (uint) val;
925 }
926 #ifdef ASSERT
927 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
928 bool phase_optimize_finished() const { return _phase_optimize_finished; }
929 uint count_live_nodes_by_graph_walk();
930 void print_missing_nodes();
931 #endif
932
933 // Record modified nodes to check that they are put on IGVN worklist
934 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
935 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
936 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
937
938 MachConstantBaseNode* mach_constant_base_node();
939 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
940 // Generated by adlc, true if CallNode requires MachConstantBase.
941 bool needs_deep_clone_jvms();
942
943 // Handy undefined Node
944 Node* top() const { return _top; }
945
946 // these are used by guys who need to know about creation and transformation of top:
947 Node* cached_top_node() { return _top; }
948 void set_cached_top_node(Node* tn);
949
950 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
951 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
952 Node_Notes* default_node_notes() const { return _default_node_notes; }
953 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
954
955 Node_Notes* node_notes_at(int idx) {
956 return locate_node_notes(_node_note_array, idx, false);
957 }
958 inline bool set_node_notes_at(int idx, Node_Notes* value);
959
960 // Copy notes from source to dest, if they exist.
961 // Overwrite dest only if source provides something.
962 // Return true if information was moved.
963 bool copy_node_notes_to(Node* dest, Node* source);
964
965 // Workhorse function to sort out the blocked Node_Notes array:
966 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
967 int idx, bool can_grow = false);
968
969 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
970
971 // Type management
972 Arena* type_arena() { return _type_arena; }
973 Dict* type_dict() { return _type_dict; }
974 size_t type_last_size() { return _type_last_size; }
975 int num_alias_types() { return _num_alias_types; }
976
977 void init_type_arena() { _type_arena = &_Compile_types; }
978 void set_type_arena(Arena* a) { _type_arena = a; }
979 void set_type_dict(Dict* d) { _type_dict = d; }
980 void set_type_last_size(size_t sz) { _type_last_size = sz; }
981
982 const TypeFunc* last_tf(ciMethod* m) {
983 return (m == _last_tf_m) ? _last_tf : nullptr;
984 }
985 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
986 assert(m != nullptr || tf == nullptr, "");
987 _last_tf_m = m;
988 _last_tf = tf;
989 }
990
991 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
992 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); }
993 bool have_alias_type(const TypePtr* adr_type);
994 AliasType* alias_type(ciField* field);
995
996 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
997 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
998 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
999
1000 // Building nodes
1001 void rethrow_exceptions(JVMState* jvms);
1002 void return_values(JVMState* jvms);
1003 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1004
1005 // Decide how to build a call.
1006 // The profile factor is a discount to apply to this site's interp. profile.
1007 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1008 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
1009 bool allow_intrinsics = true);
1010 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1011 return should_delay_string_inlining(call_method, jvms) ||
1012 should_delay_boxing_inlining(call_method, jvms) ||
1013 should_delay_vector_inlining(call_method, jvms);
1014 }
1015 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1016 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1017 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
1018 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
1019
1020 // Helper functions to identify inlining potential at call-site
1021 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
1022 ciKlass* holder, ciMethod* callee,
1023 const TypeOopPtr* receiver_type, bool is_virtual,
1024 bool &call_does_dispatch, int &vtable_index,
1025 bool check_access = true);
1026 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1027 ciMethod* callee, const TypeOopPtr* receiver_type,
1028 bool check_access = true);
1029
1030 // Report if there were too many traps at a current method and bci.
1031 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1032 // If there is no MDO at all, report no trap unless told to assume it.
1033 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1034 // This version, unspecific to a particular bci, asks if
1035 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1036 bool too_many_traps(Deoptimization::DeoptReason reason,
1037 // Privately used parameter for logging:
1038 ciMethodData* logmd = nullptr);
1039 // Report if there were too many recompiles at a method and bci.
1040 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1041 // Report if there were too many traps or recompiles at a method and bci.
1042 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1043 return too_many_traps(method, bci, reason) ||
1044 too_many_recompiles(method, bci, reason);
1045 }
1046 // Return a bitset with the reasons where deoptimization is allowed,
1047 // i.e., where there were not too many uncommon traps.
1048 int _allowed_reasons;
1049 int allowed_deopt_reasons() { return _allowed_reasons; }
1050 void set_allowed_deopt_reasons();
1051
1052 // Parsing, optimization
1053 PhaseGVN* initial_gvn() { return _initial_gvn; }
1054 Unique_Node_List* igvn_worklist() {
1055 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1056 return _igvn_worklist;
1057 }
1058 Type_Array* types() {
1059 assert(_types != nullptr, "must be created in Compile::Compile");
1060 return _types;
1061 }
1062 NodeHash* node_hash() {
1063 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1064 return _node_hash;
1065 }
1066 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1067 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1068 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1069
1070 // Replace n by nn using initial_gvn, calling hash_delete and
1071 // record_for_igvn as needed.
1072 void gvn_replace_by(Node* n, Node* nn);
1073
1074
1075 void identify_useful_nodes(Unique_Node_List &useful);
1076 void update_dead_node_list(Unique_Node_List &useful);
1077 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist);
1078
1079 void remove_useless_node(Node* dead);
1080
1081 // Record this CallGenerator for inlining at the end of parsing.
1082 void add_late_inline(CallGenerator* cg) {
1083 _late_inlines.insert_before(_late_inlines_pos, cg);
1084 _late_inlines_pos++;
1085 }
1086
1087 void prepend_late_inline(CallGenerator* cg) {
1088 _late_inlines.insert_before(0, cg);
1089 }
1090
1091 void add_string_late_inline(CallGenerator* cg) {
1092 _string_late_inlines.push(cg);
1093 }
1094
1095 void add_boxing_late_inline(CallGenerator* cg) {
1096 _boxing_late_inlines.push(cg);
1097 }
1098
1099 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1100 _vector_reboxing_late_inlines.push(cg);
1101 }
1102
1103 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1104 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1105
1106 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1107 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1108
1109 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1110
1111 void process_print_inlining();
1112 void dump_print_inlining();
1113
1114 bool over_inlining_cutoff() const {
1115 if (!inlining_incrementally()) {
1116 return unique() > (uint)NodeCountInliningCutoff;
1117 } else {
1118 // Give some room for incremental inlining algorithm to "breathe"
1119 // and avoid thrashing when live node count is close to the limit.
1120 // Keep in mind that live_nodes() isn't accurate during inlining until
1121 // dead node elimination step happens (see Compile::inline_incrementally).
1122 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1123 }
1124 }
1125
1126 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1127 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--; }
1128 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1129
1130 bool inline_incrementally_one();
1131 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1132 void inline_incrementally(PhaseIterGVN& igvn);
1133 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); }
1134 void inline_string_calls(bool parse_time);
1135 void inline_boxing_calls(PhaseIterGVN& igvn);
1136 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1137 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1138
1139 void inline_vector_reboxing_calls();
1140 bool has_vbox_nodes();
1141
1142 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1143
1144 // Matching, CFG layout, allocation, code generation
1145 PhaseCFG* cfg() { return _cfg; }
1146 bool has_java_calls() const { return _java_calls > 0; }
1147 int java_calls() const { return _java_calls; }
1148 int inner_loops() const { return _inner_loops; }
1149 Matcher* matcher() { return _matcher; }
1150 PhaseRegAlloc* regalloc() { return _regalloc; }
1151 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1152 Arena* indexSet_arena() { return _indexSet_arena; }
1153 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1154 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1155
1156 void update_interpreter_frame_size(int size) {
1157 if (_interpreter_frame_size < size) {
1158 _interpreter_frame_size = size;
1159 }
1160 }
1161
1162 void set_matcher(Matcher* m) { _matcher = m; }
1163 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1164 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1165 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1166
1167 void set_java_calls(int z) { _java_calls = z; }
1168 void set_inner_loops(int z) { _inner_loops = z; }
1169
1170 Dependencies* dependencies() { return env()->dependencies(); }
1171
1172 // Major entry point. Given a Scope, compile the associated method.
1173 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1174 // replacement, entry_bci indicates the bytecode for which to compile a
1175 // continuation.
1176 Compile(ciEnv* ci_env, ciMethod* target,
1177 int entry_bci, Options options, DirectiveSet* directive);
1178
1179 // Second major entry point. From the TypeFunc signature, generate code
1180 // to pass arguments from the Java calling convention to the C calling
1181 // convention.
1182 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1183 address stub_function, const char *stub_name,
1184 int is_fancy_jump, bool pass_tls,
1185 bool return_pc, DirectiveSet* directive);
1186
1187 ~Compile();
1188
1189 // Are we compiling a method?
1190 bool has_method() { return method() != nullptr; }
1191
1192 // Maybe print some information about this compile.
1193 void print_compile_messages();
1194
1195 // Final graph reshaping, a post-pass after the regular optimizer is done.
1196 bool final_graph_reshaping();
1197
1198 // returns true if adr is completely contained in the given alias category
1199 bool must_alias(const TypePtr* adr, int alias_idx);
1200
1201 // returns true if adr overlaps with the given alias category
1202 bool can_alias(const TypePtr* adr, int alias_idx);
1203
1204 // Stack slots that may be unused by the calling convention but must
1205 // otherwise be preserved. On Intel this includes the return address.
1206 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1207 uint in_preserve_stack_slots() {
1208 return SharedRuntime::in_preserve_stack_slots();
1209 }
1210
1211 // "Top of Stack" slots that may be unused by the calling convention but must
1212 // otherwise be preserved.
1213 // On Intel these are not necessary and the value can be zero.
1214 static uint out_preserve_stack_slots() {
1215 return SharedRuntime::out_preserve_stack_slots();
1216 }
1217
1218 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1219 // for calls to C. Supports the var-args backing area for register parms.
1220 uint varargs_C_out_slots_killed() const;
1221
1222 // Number of Stack Slots consumed by a synchronization entry
1223 int sync_stack_slots() const;
1224
1225 // Compute the name of old_SP. See <arch>.ad for frame layout.
1226 OptoReg::Name compute_old_SP();
1227
1228 private:
1229 // Phase control:
1230 void Init(bool aliasing); // Prepare for a single compilation
1231 void Optimize(); // Given a graph, optimize it
1232 void Code_Gen(); // Generate code from a graph
1233
1234 // Management of the AliasType table.
1235 void grow_alias_types();
1236 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1237 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1238 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1239
1240 void verify_top(Node*) const PRODUCT_RETURN;
1241
1242 // Intrinsic setup.
1243 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1244 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1245 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1246 void register_intrinsic(CallGenerator* cg); // update fn
1247
1248 #ifndef PRODUCT
1249 static juint _intrinsic_hist_count[];
1250 static jubyte _intrinsic_hist_flags[];
1251 #endif
1252 // Function calls made by the public function final_graph_reshaping.
1253 // No need to be made public as they are not called elsewhere.
1254 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1255 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1256 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1257 void eliminate_redundant_card_marks(Node* n);
1258 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1259
1260 // Logic cone optimization.
1261 void optimize_logic_cones(PhaseIterGVN &igvn);
1262 void collect_logic_cone_roots(Unique_Node_List& list);
1263 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1264 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1265 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1266 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1267 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1268 void check_no_dead_use() const NOT_DEBUG_RETURN;
1269
1270 public:
1271
1272 // Note: Histogram array size is about 1 Kb.
1273 enum { // flag bits:
1274 _intrinsic_worked = 1, // succeeded at least once
1275 _intrinsic_failed = 2, // tried it but it failed
1276 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1277 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1278 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1279 };
1280 // Update histogram. Return boolean if this is a first-time occurrence.
1281 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1282 bool is_virtual, int flags) PRODUCT_RETURN0;
1283 static void print_intrinsic_statistics() PRODUCT_RETURN;
1284
1285 // Graph verification code
1286 // Walk the node list, verifying that there is a one-to-one
1287 // correspondence between Use-Def edges and Def-Use edges
1288 // The option no_dead_code enables stronger checks that the
1289 // graph is strongly connected from root in both directions.
1290 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1291
1292 // Verify bi-directional correspondence of edges
1293 void verify_bidirectional_edges(Unique_Node_List &visited);
1294
1295 // End-of-run dumps.
1296 static void print_statistics() PRODUCT_RETURN;
1297
1298 // Verify ADLC assumptions during startup
1299 static void adlc_verification() PRODUCT_RETURN;
1300
1301 // Definitions of pd methods
1302 static void pd_compiler2_init();
1303
1304 // Static parse-time type checking logic for gen_subtype_check:
1305 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1306 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1307
1308 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1309 // Optional control dependency (for example, on range check)
1310 Node* ctrl = nullptr);
1311
1312 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1313 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1314
1315 // Auxiliary methods for randomized fuzzing/stressing
1316 int random();
1317 bool randomized_select(int count);
1318
1319 // seed random number generation and log the seed for repeatability.
1320 void initialize_stress_seed(const DirectiveSet* directive);
1321
1322 // supporting clone_map
1323 CloneMap& clone_map();
1324 void set_clone_map(Dict* d);
1325
1326 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1327 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1328 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1329
1330 #ifdef IA32
1331 private:
1332 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1333 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1334
1335 // Remember if this compilation changes hardware mode to 24-bit precision.
1336 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1337 _select_24_bit_instr = selection;
1338 _in_24_bit_fp_mode = mode;
1339 }
1340
1341 public:
1342 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1343 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1344 #endif // IA32
1345 #ifdef ASSERT
1346 VerifyMeetResult* _type_verify;
1347 void set_exception_backedge() { _exception_backedge = true; }
1348 bool has_exception_backedge() const { return _exception_backedge; }
1349 #endif
1350
1351 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1352 BasicType out_bt, BasicType in_bt);
1353
1354 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1355 };
1356
1357 #endif // SHARE_OPTO_COMPILE_HPP