1 /*
2 * Copyright (c) 1997, 2023, 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/compilerOracle.hpp"
33 #include "compiler/compileBroker.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/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
49 class AbstractLockNode;
50 class AddPNode;
51 class Block;
52 class Bundle;
53 class CallGenerator;
54 class CallStaticJavaNode;
55 class CloneMap;
56 class ConnectionGraph;
57 class IdealGraphPrinter;
58 class InlineTree;
59 class Matcher;
60 class MachConstantNode;
61 class MachConstantBaseNode;
62 class MachNode;
63 class MachOper;
64 class MachSafePointNode;
65 class Node;
66 class Node_Array;
67 class Node_List;
68 class Node_Notes;
69 class NodeCloneInfo;
70 class OptoReg;
71 class PhaseCFG;
72 class PhaseGVN;
73 class PhaseIterGVN;
74 class PhaseRegAlloc;
75 class PhaseCCP;
76 class PhaseOutput;
77 class RootNode;
78 class relocInfo;
79 class StartNode;
80 class SafePointNode;
81 class JVMState;
82 class Type;
83 class TypeInt;
84 class TypeInteger;
85 class TypeKlassPtr;
86 class TypePtr;
87 class TypeOopPtr;
88 class TypeFunc;
89 class TypeVect;
90 class Unique_Node_List;
91 class UnstableIfTrap;
92 class nmethod;
93 class Node_Stack;
94 struct Final_Reshape_Counts;
95
96 enum LoopOptsMode {
97 LoopOptsDefault,
98 LoopOptsNone,
99 LoopOptsMaxUnroll,
100 LoopOptsShenandoahExpand,
101 LoopOptsShenandoahPostExpand,
102 LoopOptsSkipSplitIf,
103 LoopOptsVerify
104 };
105
106 // The type of all node counts and indexes.
107 // It must hold at least 16 bits, but must also be fast to load and store.
108 // This type, if less than 32 bits, could limit the number of possible nodes.
109 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
110 typedef unsigned int node_idx_t;
111
112 class NodeCloneInfo {
113 private:
114 uint64_t _idx_clone_orig;
115 public:
116
117 void set_idx(node_idx_t idx) {
118 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
119 }
120 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
121
122 void set_gen(int generation) {
123 uint64_t g = (uint64_t)generation << 32;
124 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
125 }
126 int gen() const { return (int)(_idx_clone_orig >> 32); }
127
128 void set(uint64_t x) { _idx_clone_orig = x; }
129 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
130 uint64_t get() const { return _idx_clone_orig; }
131
132 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
133 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
134
135 void dump() const;
136 };
137
138 class CloneMap {
139 friend class Compile;
140 private:
141 bool _debug;
142 Dict* _dict;
143 int _clone_idx; // current cloning iteration/generation in loop unroll
144 public:
145 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
146 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
147 Dict* dict() const { return _dict; }
148 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
149 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
150 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
151 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
152 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
153 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
154 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
155 int max_gen() const;
156 void clone(Node* old, Node* nnn, int gen);
157 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
158 void dump(node_idx_t key) const;
159
160 int clone_idx() const { return _clone_idx; }
161 void set_clone_idx(int x) { _clone_idx = x; }
162 bool is_debug() const { return _debug; }
163 void set_debug(bool debug) { _debug = debug; }
164
165 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
166 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
167 };
168
169 class Options {
170 friend class Compile;
171 friend class VMStructs;
172 private:
173 const bool _subsume_loads; // Load can be matched as part of a larger op.
174 const bool _do_escape_analysis; // Do escape analysis.
175 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
176 const bool _eliminate_boxing; // Do boxing elimination.
177 const bool _do_locks_coarsening; // Do locks coarsening
178 const bool _install_code; // Install the code that was compiled
179 public:
180 Options(bool subsume_loads, bool do_escape_analysis,
181 bool do_iterative_escape_analysis,
182 bool eliminate_boxing, bool do_locks_coarsening,
183 bool install_code) :
184 _subsume_loads(subsume_loads),
185 _do_escape_analysis(do_escape_analysis),
186 _do_iterative_escape_analysis(do_iterative_escape_analysis),
187 _eliminate_boxing(eliminate_boxing),
188 _do_locks_coarsening(do_locks_coarsening),
189 _install_code(install_code) {
190 }
191
192 static Options for_runtime_stub() {
193 return Options(
194 /* subsume_loads = */ true,
195 /* do_escape_analysis = */ false,
196 /* do_iterative_escape_analysis = */ false,
197 /* eliminate_boxing = */ false,
198 /* do_lock_coarsening = */ false,
199 /* install_code = */ true
200 );
201 }
202 };
203
204 //------------------------------Compile----------------------------------------
205 // This class defines a top-level Compiler invocation.
206
207 class Compile : public Phase {
208 friend class VMStructs;
209
210 public:
211 // Fixed alias indexes. (See also MergeMemNode.)
212 enum {
213 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
214 AliasIdxBot = 2, // pseudo-index, aliases to everything
215 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
216 };
217
218 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
219 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
220 // then brackets are put into the log, with time stamps and node counts.
221 // (The time collection itself is always conditionalized on CITime.)
222 class TracePhase : public TraceTime {
223 private:
224 Compile* C;
225 CompileLog* _log;
226 const char* _phase_name;
227 bool _dolog;
228 public:
229 TracePhase(const char* name, elapsedTimer* accumulator);
230 ~TracePhase();
231 };
232
233 // Information per category of alias (memory slice)
234 class AliasType {
235 private:
236 friend class Compile;
237
238 int _index; // unique index, used with MergeMemNode
239 const TypePtr* _adr_type; // normalized address type
240 ciField* _field; // relevant instance field, or null if none
241 const Type* _element; // relevant array element type, or null if none
242 bool _is_rewritable; // false if the memory is write-once only
243 int _general_index; // if this is type is an instance, the general
244 // type that this is an instance of
245
246 void Init(int i, const TypePtr* at);
247
248 public:
249 int index() const { return _index; }
250 const TypePtr* adr_type() const { return _adr_type; }
251 ciField* field() const { return _field; }
252 const Type* element() const { return _element; }
253 bool is_rewritable() const { return _is_rewritable; }
254 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
255 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
256
257 void set_rewritable(bool z) { _is_rewritable = z; }
258 void set_field(ciField* f) {
259 assert(!_field,"");
260 _field = f;
261 if (f->is_final() || f->is_stable()) {
262 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
263 _is_rewritable = false;
264 }
265 }
266 void set_element(const Type* e) {
267 assert(_element == NULL, "");
268 _element = e;
269 }
270
271 BasicType basic_type() const;
272
273 void print_on(outputStream* st) PRODUCT_RETURN;
274 };
275
276 enum {
277 logAliasCacheSize = 6,
278 AliasCacheSize = (1<<logAliasCacheSize)
279 };
280 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
281 enum {
282 trapHistLength = MethodData::_trap_hist_limit
283 };
284
285 private:
286 // Fixed parameters to this compilation.
287 const int _compile_id;
288 const Options _options; // Compilation options
289 ciMethod* _method; // The method being compiled.
290 int _entry_bci; // entry bci for osr methods.
291 const TypeFunc* _tf; // My kind of signature
292 InlineTree* _ilt; // Ditto (temporary).
293 address _stub_function; // VM entry for stub being compiled, or NULL
294 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
295 address _stub_entry_point; // Compile code entry for generated stub, or NULL
296
297 // Control of this compilation.
298 int _max_inline_size; // Max inline size for this compilation
299 int _freq_inline_size; // Max hot method inline size for this compilation
300 int _fixed_slots; // count of frame slots not allocated by the register
301 // allocator i.e. locks, original deopt pc, etc.
302 uintx _max_node_limit; // Max unique node count during a single compilation.
303
304 bool _post_loop_opts_phase; // Loop opts are finished.
305
306 int _major_progress; // Count of something big happening
307 bool _inlining_progress; // progress doing incremental inlining?
308 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
309 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
310 bool _has_loops; // True if the method _may_ have some loops
311 bool _has_split_ifs; // True if the method _may_ have some split-if
312 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
313 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
314 bool _has_boxed_value; // True if a boxed object is allocated
315 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
316 uint _max_vector_size; // Maximum size of generated vectors
317 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
318 uint _trap_hist[trapHistLength]; // Cumulative traps
319 bool _trap_can_recompile; // Have we emitted a recompiling trap?
320 uint _decompile_count; // Cumulative decompilation counts.
321 bool _do_inlining; // True if we intend to do inlining
322 bool _do_scheduling; // True if we intend to do scheduling
323 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
324 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
325 bool _use_cmove; // True if CMove should be used without profitability analysis
326 bool _do_aliasing; // True if we intend to do aliasing
327 bool _print_assembly; // True if we should dump assembly code for this compilation
328 bool _print_inlining; // True if we should print inlining for this compilation
329 bool _print_intrinsics; // True if we should print intrinsics for this compilation
330 #ifndef PRODUCT
331 uint _igv_idx; // Counter for IGV node identifiers
332 bool _trace_opto_output;
333 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
334 #endif
335 bool _has_irreducible_loop; // Found irreducible loops
336 // JSR 292
337 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
338 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
339 RTMState _rtm_state; // State of Restricted Transactional Memory usage
340 int _loop_opts_cnt; // loop opts round
341 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
342 uint _stress_seed; // Seed for stress testing
343
344 // Compilation environment.
345 Arena _comp_arena; // Arena with lifetime equivalent to Compile
346 void* _barrier_set_state; // Potential GC barrier state for Compile
347 ciEnv* _env; // CI interface
348 DirectiveSet* _directive; // Compiler directive
349 CompileLog* _log; // from CompilerThread
350 const char* _failure_reason; // for record_failure/failing pattern
351 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
352 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
353 GrowableArray<Node*> _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
354 GrowableArray<Node*> _skeleton_predicate_opaqs; // List of Opaque4 nodes for the loop skeleton predicates.
355 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
356 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
357 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
358 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
359 ConnectionGraph* _congraph;
360 #ifndef PRODUCT
361 IdealGraphPrinter* _igv_printer;
362 static IdealGraphPrinter* _debug_file_printer;
363 static IdealGraphPrinter* _debug_network_printer;
364 #endif
365
366
367 // Node management
368 uint _unique; // Counter for unique Node indices
369 VectorSet _dead_node_list; // Set of dead nodes
370 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
371 // So use this to keep count and make the call O(1).
372 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
373 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
374
375 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
376 Arena _node_arena; // Arena for new-space Nodes
377 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
378 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
379 Node* _top; // Unique top node. (Reset by various phases.)
380
381 Node* _immutable_memory; // Initial memory state
382
383 Node* _recent_alloc_obj;
384 Node* _recent_alloc_ctl;
385
386 // Constant table
387 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
388
389
390 // Blocked array of debugging and profiling information,
391 // tracked per node.
392 enum { _log2_node_notes_block_size = 8,
393 _node_notes_block_size = (1<<_log2_node_notes_block_size)
394 };
395 GrowableArray<Node_Notes*>* _node_note_array;
396 Node_Notes* _default_node_notes; // default notes for new nodes
397
398 // After parsing and every bulk phase we hang onto the Root instruction.
399 // The RootNode instruction is where the whole program begins. It produces
400 // the initial Control and BOTTOM for everybody else.
401
402 // Type management
403 Arena _Compile_types; // Arena for all types
404 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
405 Dict* _type_dict; // Intern table
406 CloneMap _clone_map; // used for recording history of cloned nodes
407 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
408 ciMethod* _last_tf_m; // Cache for
409 const TypeFunc* _last_tf; // TypeFunc::make
410 AliasType** _alias_types; // List of alias types seen so far.
411 int _num_alias_types; // Logical length of _alias_types
412 int _max_alias_types; // Physical length of _alias_types
413 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
414
415 // Parsing, optimization
416 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
417 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
418
419 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
420 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
421 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
422
423 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
424
425 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
426 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
427
428 // Inlining may not happen in parse order which would make
429 // PrintInlining output confusing. Keep track of PrintInlining
430 // pieces in order.
431 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
432 private:
433 CallGenerator* _cg;
434 stringStream _ss;
435 static const size_t default_stream_buffer_size = 128;
436
437 public:
438 PrintInliningBuffer()
439 : _cg(NULL), _ss(default_stream_buffer_size) {}
440
441 stringStream* ss() { return &_ss; }
442 CallGenerator* cg() { return _cg; }
443 void set_cg(CallGenerator* cg) { _cg = cg; }
444 };
445
446 stringStream* _print_inlining_stream;
447 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
448 int _print_inlining_idx;
449 char* _print_inlining_output;
450
451 // Only keep nodes in the expensive node list that need to be optimized
452 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
453 // Use for sorting expensive nodes to bring similar nodes together
454 static int cmp_expensive_nodes(Node** n1, Node** n2);
455 // Expensive nodes list already sorted?
456 bool expensive_nodes_sorted() const;
457 // Remove the speculative part of types and clean up the graph
458 void remove_speculative_types(PhaseIterGVN &igvn);
459
460 void* _replay_inline_data; // Pointer to data loaded from file
461
462 void print_inlining_init();
463 void print_inlining_reinit();
464 void print_inlining_commit();
465 void print_inlining_push();
466 PrintInliningBuffer* print_inlining_current();
467
468 void log_late_inline_failure(CallGenerator* cg, const char* msg);
469 DEBUG_ONLY(bool _exception_backedge;)
470
471 public:
472
473 void* barrier_set_state() const { return _barrier_set_state; }
474
475 stringStream* print_inlining_stream() {
476 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
477 return _print_inlining_stream;
478 }
479
480 void print_inlining_update(CallGenerator* cg);
481 void print_inlining_update_delayed(CallGenerator* cg);
482 void print_inlining_move_to(CallGenerator* cg);
483 void print_inlining_assert_ready();
484 void print_inlining_reset();
485
486 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
487 stringStream ss;
488 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
489 print_inlining_stream()->print("%s", ss.freeze());
490 }
491
492 #ifndef PRODUCT
493 IdealGraphPrinter* igv_printer() { return _igv_printer; }
494 #endif
495
496 void log_late_inline(CallGenerator* cg);
497 void log_inline_id(CallGenerator* cg);
498 void log_inline_failure(const char* msg);
499
500 void* replay_inline_data() const { return _replay_inline_data; }
501
502 // Dump inlining replay data to the stream.
503 void dump_inline_data(outputStream* out);
504 void dump_inline_data_reduced(outputStream* out);
505
506 private:
507 // Matching, CFG layout, allocation, code generation
508 PhaseCFG* _cfg; // Results of CFG finding
509 int _java_calls; // Number of java calls in the method
510 int _inner_loops; // Number of inner loops in the method
511 Matcher* _matcher; // Engine to map ideal to machine instructions
512 PhaseRegAlloc* _regalloc; // Results of register allocation.
513 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
514 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
515 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
516 int _interpreter_frame_size;
517
518 PhaseOutput* _output;
519
520 public:
521 // Accessors
522
523 // The Compile instance currently active in this (compiler) thread.
524 static Compile* current() {
525 return (Compile*) ciEnv::current()->compiler_data();
526 }
527
528 int interpreter_frame_size() const { return _interpreter_frame_size; }
529
530 PhaseOutput* output() const { return _output; }
531 void set_output(PhaseOutput* o) { _output = o; }
532
533 // ID for this compilation. Useful for setting breakpoints in the debugger.
534 int compile_id() const { return _compile_id; }
535 DirectiveSet* directive() const { return _directive; }
536
537 // Does this compilation allow instructions to subsume loads? User
538 // instructions that subsume a load may result in an unschedulable
539 // instruction sequence.
540 bool subsume_loads() const { return _options._subsume_loads; }
541 /** Do escape analysis. */
542 bool do_escape_analysis() const { return _options._do_escape_analysis; }
543 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
544 /** Do boxing elimination. */
545 bool eliminate_boxing() const { return _options._eliminate_boxing; }
546 /** Do aggressive boxing elimination. */
547 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
548 bool should_install_code() const { return _options._install_code; }
549 /** Do locks coarsening. */
550 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
551
552 // Other fixed compilation parameters.
553 ciMethod* method() const { return _method; }
554 int entry_bci() const { return _entry_bci; }
555 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
556 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
557 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
558 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
559 InlineTree* ilt() const { return _ilt; }
560 address stub_function() const { return _stub_function; }
561 const char* stub_name() const { return _stub_name; }
562 address stub_entry_point() const { return _stub_entry_point; }
563 void set_stub_entry_point(address z) { _stub_entry_point = z; }
564
565 // Control of this compilation.
566 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
567 void set_fixed_slots(int n) { _fixed_slots = n; }
568 int major_progress() const { return _major_progress; }
569 void set_inlining_progress(bool z) { _inlining_progress = z; }
570 int inlining_progress() const { return _inlining_progress; }
571 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
572 int inlining_incrementally() const { return _inlining_incrementally; }
573 void set_do_cleanup(bool z) { _do_cleanup = z; }
574 int do_cleanup() const { return _do_cleanup; }
575 void set_major_progress() { _major_progress++; }
576 void restore_major_progress(int progress) { _major_progress += progress; }
577 void clear_major_progress() { _major_progress = 0; }
578 int max_inline_size() const { return _max_inline_size; }
579 void set_freq_inline_size(int n) { _freq_inline_size = n; }
580 int freq_inline_size() const { return _freq_inline_size; }
581 void set_max_inline_size(int n) { _max_inline_size = n; }
582 bool has_loops() const { return _has_loops; }
583 void set_has_loops(bool z) { _has_loops = z; }
584 bool has_split_ifs() const { return _has_split_ifs; }
585 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
586 bool has_unsafe_access() const { return _has_unsafe_access; }
587 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
588 bool has_stringbuilder() const { return _has_stringbuilder; }
589 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
590 bool has_boxed_value() const { return _has_boxed_value; }
591 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
592 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
593 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
594 uint max_vector_size() const { return _max_vector_size; }
595 void set_max_vector_size(uint s) { _max_vector_size = s; }
596 bool clear_upper_avx() const { return _clear_upper_avx; }
597 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
598 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
599 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
600 bool trap_can_recompile() const { return _trap_can_recompile; }
601 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
602 uint decompile_count() const { return _decompile_count; }
603 void set_decompile_count(uint c) { _decompile_count = c; }
604 bool allow_range_check_smearing() const;
605 bool do_inlining() const { return _do_inlining; }
606 void set_do_inlining(bool z) { _do_inlining = z; }
607 bool do_scheduling() const { return _do_scheduling; }
608 void set_do_scheduling(bool z) { _do_scheduling = z; }
609 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
610 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
611 bool do_vector_loop() const { return _do_vector_loop; }
612 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
613 bool use_cmove() const { return _use_cmove; }
614 void set_use_cmove(bool z) { _use_cmove = z; }
615 bool do_aliasing() const { return _do_aliasing; }
616 bool print_assembly() const { return _print_assembly; }
617 void set_print_assembly(bool z) { _print_assembly = z; }
618 bool print_inlining() const { return _print_inlining; }
619 void set_print_inlining(bool z) { _print_inlining = z; }
620 bool print_intrinsics() const { return _print_intrinsics; }
621 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
622 RTMState rtm_state() const { return _rtm_state; }
623 void set_rtm_state(RTMState s) { _rtm_state = s; }
624 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
625 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
626 uint max_node_limit() const { return (uint)_max_node_limit; }
627 void set_max_node_limit(uint n) { _max_node_limit = n; }
628 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
629 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
630 bool has_monitors() const { return _has_monitors; }
631 void set_has_monitors(bool v) { _has_monitors = v; }
632
633 // check the CompilerOracle for special behaviours for this compile
634 bool method_has_option(enum CompileCommand option) {
635 return method() != NULL && method()->has_option(option);
636 }
637
638 #ifndef PRODUCT
639 uint next_igv_idx() { return _igv_idx++; }
640 bool trace_opto_output() const { return _trace_opto_output; }
641 void print_ideal_ir(const char* phase_name);
642 bool should_print_ideal() const { return _directive->PrintIdealOption; }
643 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
644 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
645 int _in_dump_cnt; // Required for dumping ir nodes.
646 #endif
647 bool has_irreducible_loop() const { return _has_irreducible_loop; }
648 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
649
650 // JSR 292
651 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
652 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
653
654 Ticks _latest_stage_start_counter;
655
656 void begin_method();
657 void end_method();
658 bool should_print_igv(int level);
659 bool should_print_phase(CompilerPhaseType cpt);
660
661 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
662
663 #ifndef PRODUCT
664 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
665 void igv_print_method_to_network(const char* phase_name = "Debug");
666 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
667 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
668 #endif
669
670 int macro_count() const { return _macro_nodes.length(); }
671 int predicate_count() const { return _predicate_opaqs.length(); }
672 int skeleton_predicate_count() const { return _skeleton_predicate_opaqs.length(); }
673 int expensive_count() const { return _expensive_nodes.length(); }
674 int coarsened_count() const { return _coarsened_locks.length(); }
675
676 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
677 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs.at(idx); }
678 Node* skeleton_predicate_opaque4_node(int idx) const { return _skeleton_predicate_opaqs.at(idx); }
679 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
680
681 ConnectionGraph* congraph() { return _congraph;}
682 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
683 void add_macro_node(Node * n) {
684 //assert(n->is_macro(), "must be a macro node");
685 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
686 _macro_nodes.append(n);
687 }
688 void remove_macro_node(Node* n) {
689 // this function may be called twice for a node so we can only remove it
690 // if it's still existing.
691 _macro_nodes.remove_if_existing(n);
692 // remove from _predicate_opaqs list also if it is there
693 if (predicate_count() > 0) {
694 _predicate_opaqs.remove_if_existing(n);
695 }
696 // Remove from coarsened locks list if present
697 if (coarsened_count() > 0) {
698 remove_coarsened_lock(n);
699 }
700 }
701 void add_expensive_node(Node* n);
702 void remove_expensive_node(Node* n) {
703 _expensive_nodes.remove_if_existing(n);
704 }
705 void add_predicate_opaq(Node* n) {
706 assert(!_predicate_opaqs.contains(n), "duplicate entry in predicate opaque1");
707 assert(_macro_nodes.contains(n), "should have already been in macro list");
708 _predicate_opaqs.append(n);
709 }
710 void add_skeleton_predicate_opaq(Node* n) {
711 assert(!_skeleton_predicate_opaqs.contains(n), "duplicate entry in skeleton predicate opaque4 list");
712 _skeleton_predicate_opaqs.append(n);
713 }
714 void remove_skeleton_predicate_opaq(Node* n) {
715 if (skeleton_predicate_count() > 0) {
716 _skeleton_predicate_opaqs.remove_if_existing(n);
717 }
718 }
719 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
720 void remove_coarsened_lock(Node* n);
721 bool coarsened_locks_consistent();
722
723 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
724 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
725 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
726
727 void record_for_post_loop_opts_igvn(Node* n);
728 void remove_from_post_loop_opts_igvn(Node* n);
729 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
730
731 void record_unstable_if_trap(UnstableIfTrap* trap);
732 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
733 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
734 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
735
736 void sort_macro_nodes();
737
738 // remove the opaque nodes that protect the predicates so that the unused checks and
739 // uncommon traps will be eliminated from the graph.
740 void cleanup_loop_predicates(PhaseIterGVN &igvn);
741 bool is_predicate_opaq(Node* n) {
742 return _predicate_opaqs.contains(n);
743 }
744
745 // Are there candidate expensive nodes for optimization?
746 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
747 // Check whether n1 and n2 are similar
748 static int cmp_expensive_nodes(Node* n1, Node* n2);
749 // Sort expensive nodes to locate similar expensive nodes
750 void sort_expensive_nodes();
751
752 // Compilation environment.
753 Arena* comp_arena() { return &_comp_arena; }
754 ciEnv* env() const { return _env; }
755 CompileLog* log() const { return _log; }
756 bool failing() const { return _env->failing() || _failure_reason != NULL; }
757 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
758
759 bool failure_reason_is(const char* r) const {
760 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
761 }
762
763 void record_failure(const char* reason);
764 void record_method_not_compilable(const char* reason) {
765 env()->record_method_not_compilable(reason);
766 // Record failure reason.
767 record_failure(reason);
768 }
769 bool check_node_count(uint margin, const char* reason) {
770 if (live_nodes() + margin > max_node_limit()) {
771 record_method_not_compilable(reason);
772 return true;
773 } else {
774 return false;
775 }
776 }
777
778 // Node management
779 uint unique() const { return _unique; }
780 uint next_unique() { return _unique++; }
781 void set_unique(uint i) { _unique = i; }
782 static int debug_idx() { return debug_only(_debug_idx)+0; }
783 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
784 Arena* node_arena() { return &_node_arena; }
785 Arena* old_arena() { return &_old_arena; }
786 RootNode* root() const { return _root; }
787 void set_root(RootNode* r) { _root = r; }
788 StartNode* start() const; // (Derived from root.)
789 void init_start(StartNode* s);
790 Node* immutable_memory();
791
792 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
793 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
794 void set_recent_alloc(Node* ctl, Node* obj) {
795 _recent_alloc_ctl = ctl;
796 _recent_alloc_obj = obj;
797 }
798 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
799 _dead_node_count++;
800 }
801 void reset_dead_node_list() { _dead_node_list.reset();
802 _dead_node_count = 0;
803 }
804 uint live_nodes() const {
805 int val = _unique - _dead_node_count;
806 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
807 return (uint) val;
808 }
809 #ifdef ASSERT
810 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
811 bool phase_optimize_finished() const { return _phase_optimize_finished; }
812 uint count_live_nodes_by_graph_walk();
813 void print_missing_nodes();
814 #endif
815
816 // Record modified nodes to check that they are put on IGVN worklist
817 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
818 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
819 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
820
821 MachConstantBaseNode* mach_constant_base_node();
822 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
823 // Generated by adlc, true if CallNode requires MachConstantBase.
824 bool needs_deep_clone_jvms();
825
826 // Handy undefined Node
827 Node* top() const { return _top; }
828
829 // these are used by guys who need to know about creation and transformation of top:
830 Node* cached_top_node() { return _top; }
831 void set_cached_top_node(Node* tn);
832
833 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
834 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
835 Node_Notes* default_node_notes() const { return _default_node_notes; }
836 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
837
838 Node_Notes* node_notes_at(int idx) {
839 return locate_node_notes(_node_note_array, idx, false);
840 }
841 inline bool set_node_notes_at(int idx, Node_Notes* value);
842
843 // Copy notes from source to dest, if they exist.
844 // Overwrite dest only if source provides something.
845 // Return true if information was moved.
846 bool copy_node_notes_to(Node* dest, Node* source);
847
848 // Workhorse function to sort out the blocked Node_Notes array:
849 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
850 int idx, bool can_grow = false);
851
852 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
853
854 // Type management
855 Arena* type_arena() { return _type_arena; }
856 Dict* type_dict() { return _type_dict; }
857 size_t type_last_size() { return _type_last_size; }
858 int num_alias_types() { return _num_alias_types; }
859
860 void init_type_arena() { _type_arena = &_Compile_types; }
861 void set_type_arena(Arena* a) { _type_arena = a; }
862 void set_type_dict(Dict* d) { _type_dict = d; }
863 void set_type_last_size(size_t sz) { _type_last_size = sz; }
864
865 const TypeFunc* last_tf(ciMethod* m) {
866 return (m == _last_tf_m) ? _last_tf : NULL;
867 }
868 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
869 assert(m != NULL || tf == NULL, "");
870 _last_tf_m = m;
871 _last_tf = tf;
872 }
873
874 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
875 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); }
876 bool have_alias_type(const TypePtr* adr_type);
877 AliasType* alias_type(ciField* field);
878
879 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
880 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
881 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
882
883 // Building nodes
884 void rethrow_exceptions(JVMState* jvms);
885 void return_values(JVMState* jvms);
886 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
887
888 // Decide how to build a call.
889 // The profile factor is a discount to apply to this site's interp. profile.
890 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
891 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
892 bool allow_intrinsics = true);
893 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
894 return should_delay_string_inlining(call_method, jvms) ||
895 should_delay_boxing_inlining(call_method, jvms) ||
896 should_delay_vector_inlining(call_method, jvms);
897 }
898 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
899 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
900 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
901 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
902
903 // Helper functions to identify inlining potential at call-site
904 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
905 ciKlass* holder, ciMethod* callee,
906 const TypeOopPtr* receiver_type, bool is_virtual,
907 bool &call_does_dispatch, int &vtable_index,
908 bool check_access = true);
909 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
910 ciMethod* callee, const TypeOopPtr* receiver_type,
911 bool check_access = true);
912
913 // Report if there were too many traps at a current method and bci.
914 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
915 // If there is no MDO at all, report no trap unless told to assume it.
916 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
917 // This version, unspecific to a particular bci, asks if
918 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
919 bool too_many_traps(Deoptimization::DeoptReason reason,
920 // Privately used parameter for logging:
921 ciMethodData* logmd = NULL);
922 // Report if there were too many recompiles at a method and bci.
923 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
924 // Report if there were too many traps or recompiles at a method and bci.
925 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
926 return too_many_traps(method, bci, reason) ||
927 too_many_recompiles(method, bci, reason);
928 }
929 // Return a bitset with the reasons where deoptimization is allowed,
930 // i.e., where there were not too many uncommon traps.
931 int _allowed_reasons;
932 int allowed_deopt_reasons() { return _allowed_reasons; }
933 void set_allowed_deopt_reasons();
934
935 // Parsing, optimization
936 PhaseGVN* initial_gvn() { return _initial_gvn; }
937 Unique_Node_List* for_igvn() { return _for_igvn; }
938 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
939 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
940 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
941
942 // Replace n by nn using initial_gvn, calling hash_delete and
943 // record_for_igvn as needed.
944 void gvn_replace_by(Node* n, Node* nn);
945
946
947 void identify_useful_nodes(Unique_Node_List &useful);
948 void update_dead_node_list(Unique_Node_List &useful);
949 void disconnect_useless_nodes(Unique_Node_List &useful, Unique_Node_List* worklist);
950
951 void remove_useless_node(Node* dead);
952
953 // Record this CallGenerator for inlining at the end of parsing.
954 void add_late_inline(CallGenerator* cg) {
955 _late_inlines.insert_before(_late_inlines_pos, cg);
956 _late_inlines_pos++;
957 }
958
959 void prepend_late_inline(CallGenerator* cg) {
960 _late_inlines.insert_before(0, cg);
961 }
962
963 void add_string_late_inline(CallGenerator* cg) {
964 _string_late_inlines.push(cg);
965 }
966
967 void add_boxing_late_inline(CallGenerator* cg) {
968 _boxing_late_inlines.push(cg);
969 }
970
971 void add_vector_reboxing_late_inline(CallGenerator* cg) {
972 _vector_reboxing_late_inlines.push(cg);
973 }
974
975 void remove_useless_nodes (GrowableArray<Node*>& node_list, Unique_Node_List &useful);
976
977 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
978 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
979
980 void remove_useless_coarsened_locks(Unique_Node_List& useful);
981
982 void process_print_inlining();
983 void dump_print_inlining();
984
985 bool over_inlining_cutoff() const {
986 if (!inlining_incrementally()) {
987 return unique() > (uint)NodeCountInliningCutoff;
988 } else {
989 // Give some room for incremental inlining algorithm to "breathe"
990 // and avoid thrashing when live node count is close to the limit.
991 // Keep in mind that live_nodes() isn't accurate during inlining until
992 // dead node elimination step happens (see Compile::inline_incrementally).
993 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
994 }
995 }
996
997 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
998 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--; }
999 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1000
1001 bool inline_incrementally_one();
1002 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1003 void inline_incrementally(PhaseIterGVN& igvn);
1004 void inline_string_calls(bool parse_time);
1005 void inline_boxing_calls(PhaseIterGVN& igvn);
1006 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1007 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1008
1009 void inline_vector_reboxing_calls();
1010 bool has_vbox_nodes();
1011
1012 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1013
1014 // Matching, CFG layout, allocation, code generation
1015 PhaseCFG* cfg() { return _cfg; }
1016 bool has_java_calls() const { return _java_calls > 0; }
1017 int java_calls() const { return _java_calls; }
1018 int inner_loops() const { return _inner_loops; }
1019 Matcher* matcher() { return _matcher; }
1020 PhaseRegAlloc* regalloc() { return _regalloc; }
1021 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1022 Arena* indexSet_arena() { return _indexSet_arena; }
1023 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1024 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1025
1026 void update_interpreter_frame_size(int size) {
1027 if (_interpreter_frame_size < size) {
1028 _interpreter_frame_size = size;
1029 }
1030 }
1031
1032 void set_matcher(Matcher* m) { _matcher = m; }
1033 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1034 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1035 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1036
1037 void set_java_calls(int z) { _java_calls = z; }
1038 void set_inner_loops(int z) { _inner_loops = z; }
1039
1040 Dependencies* dependencies() { return env()->dependencies(); }
1041
1042 // Major entry point. Given a Scope, compile the associated method.
1043 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1044 // replacement, entry_bci indicates the bytecode for which to compile a
1045 // continuation.
1046 Compile(ciEnv* ci_env, ciMethod* target,
1047 int entry_bci, Options options, DirectiveSet* directive);
1048
1049 // Second major entry point. From the TypeFunc signature, generate code
1050 // to pass arguments from the Java calling convention to the C calling
1051 // convention.
1052 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1053 address stub_function, const char *stub_name,
1054 int is_fancy_jump, bool pass_tls,
1055 bool return_pc, DirectiveSet* directive);
1056
1057 ~Compile() {
1058 delete _print_inlining_stream;
1059 };
1060
1061 // Are we compiling a method?
1062 bool has_method() { return method() != NULL; }
1063
1064 // Maybe print some information about this compile.
1065 void print_compile_messages();
1066
1067 // Final graph reshaping, a post-pass after the regular optimizer is done.
1068 bool final_graph_reshaping();
1069
1070 // returns true if adr is completely contained in the given alias category
1071 bool must_alias(const TypePtr* adr, int alias_idx);
1072
1073 // returns true if adr overlaps with the given alias category
1074 bool can_alias(const TypePtr* adr, int alias_idx);
1075
1076 // Stack slots that may be unused by the calling convention but must
1077 // otherwise be preserved. On Intel this includes the return address.
1078 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1079 uint in_preserve_stack_slots() {
1080 return SharedRuntime::in_preserve_stack_slots();
1081 }
1082
1083 // "Top of Stack" slots that may be unused by the calling convention but must
1084 // otherwise be preserved.
1085 // On Intel these are not necessary and the value can be zero.
1086 static uint out_preserve_stack_slots() {
1087 return SharedRuntime::out_preserve_stack_slots();
1088 }
1089
1090 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1091 // for calls to C. Supports the var-args backing area for register parms.
1092 uint varargs_C_out_slots_killed() const;
1093
1094 // Number of Stack Slots consumed by a synchronization entry
1095 int sync_stack_slots() const;
1096
1097 // Compute the name of old_SP. See <arch>.ad for frame layout.
1098 OptoReg::Name compute_old_SP();
1099
1100 private:
1101 // Phase control:
1102 void Init(bool aliasing); // Prepare for a single compilation
1103 void Optimize(); // Given a graph, optimize it
1104 void Code_Gen(); // Generate code from a graph
1105
1106 // Management of the AliasType table.
1107 void grow_alias_types();
1108 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1109 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1110 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1111
1112 void verify_top(Node*) const PRODUCT_RETURN;
1113
1114 // Intrinsic setup.
1115 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1116 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1117 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1118 void register_intrinsic(CallGenerator* cg); // update fn
1119
1120 #ifndef PRODUCT
1121 static juint _intrinsic_hist_count[];
1122 static jubyte _intrinsic_hist_flags[];
1123 #endif
1124 // Function calls made by the public function final_graph_reshaping.
1125 // No need to be made public as they are not called elsewhere.
1126 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1127 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1128 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1129 void eliminate_redundant_card_marks(Node* n);
1130
1131 // Logic cone optimization.
1132 void optimize_logic_cones(PhaseIterGVN &igvn);
1133 void collect_logic_cone_roots(Unique_Node_List& list);
1134 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1135 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1136 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1137 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1138 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1139 void check_no_dead_use() const NOT_DEBUG_RETURN;
1140
1141 public:
1142
1143 // Note: Histogram array size is about 1 Kb.
1144 enum { // flag bits:
1145 _intrinsic_worked = 1, // succeeded at least once
1146 _intrinsic_failed = 2, // tried it but it failed
1147 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1148 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1149 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1150 };
1151 // Update histogram. Return boolean if this is a first-time occurrence.
1152 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1153 bool is_virtual, int flags) PRODUCT_RETURN0;
1154 static void print_intrinsic_statistics() PRODUCT_RETURN;
1155
1156 // Graph verification code
1157 // Walk the node list, verifying that there is a one-to-one
1158 // correspondence between Use-Def edges and Def-Use edges
1159 // The option no_dead_code enables stronger checks that the
1160 // graph is strongly connected from root in both directions.
1161 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1162
1163 // Verify bi-directional correspondence of edges
1164 void verify_bidirectional_edges(Unique_Node_List &visited);
1165
1166 // End-of-run dumps.
1167 static void print_statistics() PRODUCT_RETURN;
1168
1169 // Verify ADLC assumptions during startup
1170 static void adlc_verification() PRODUCT_RETURN;
1171
1172 // Definitions of pd methods
1173 static void pd_compiler2_init();
1174
1175 // Static parse-time type checking logic for gen_subtype_check:
1176 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1177 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1178
1179 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1180 // Optional control dependency (for example, on range check)
1181 Node* ctrl = NULL);
1182
1183 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1184 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1185
1186 // Auxiliary methods for randomized fuzzing/stressing
1187 int random();
1188 bool randomized_select(int count);
1189
1190 // supporting clone_map
1191 CloneMap& clone_map();
1192 void set_clone_map(Dict* d);
1193
1194 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1195 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1196 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1197
1198 #ifdef IA32
1199 private:
1200 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1201 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1202
1203 // Remember if this compilation changes hardware mode to 24-bit precision.
1204 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1205 _select_24_bit_instr = selection;
1206 _in_24_bit_fp_mode = mode;
1207 }
1208
1209 public:
1210 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1211 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1212 #endif // IA32
1213 #ifdef ASSERT
1214 bool _type_verify_symmetry;
1215 void set_exception_backedge() { _exception_backedge = true; }
1216 bool has_exception_backedge() const { return _exception_backedge; }
1217 #endif
1218
1219 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1220 BasicType out_bt, BasicType in_bt);
1221
1222 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1223 };
1224
1225 #endif // SHARE_OPTO_COMPILE_HPP