1 /*
2 * Copyright (c) 1997, 2019, 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 "code/exceptionHandlerTable.hpp"
32 #include "compiler/compilerOracle.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "libadt/dict.hpp"
35 #include "libadt/vectset.hpp"
36 #include "jfr/jfrEvents.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/timerTrace.hpp"
45 #include "runtime/vmThread.hpp"
46 #include "utilities/ticks.hpp"
47
48 class AddPNode;
49 class Block;
50 class Bundle;
51 class CallNode;
52 class C2Compiler;
53 class CallGenerator;
54 class CloneMap;
55 class ConnectionGraph;
56 class IdealGraphPrinter;
57 class InlineTree;
58 class Int_Array;
59 class LoadBarrierNode;
60 class Matcher;
61 class MachConstantNode;
62 class MachConstantBaseNode;
63 class MachNode;
64 class MachOper;
65 class MachSafePointNode;
66 class Node;
67 class Node_Array;
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 PhaseCCP_DCE;
77 class RootNode;
78 class relocInfo;
79 class Scope;
80 class StartNode;
81 class SafePointNode;
82 class JVMState;
83 class Type;
84 class TypeData;
85 class TypeInt;
86 class TypePtr;
87 class TypeOopPtr;
88 class TypeFunc;
89 class ValueTypeBaseNode;
90 class Unique_Node_List;
91 class nmethod;
92 class WarmCallInfo;
93 class Node_Stack;
94 struct Final_Reshape_Counts;
95
96 enum LoopOptsMode {
97 LoopOptsDefault,
98 LoopOptsNone,
99 LoopOptsShenandoahExpand,
100 LoopOptsShenandoahPostExpand,
101 LoopOptsZBarrierInsertion,
102 LoopOptsSkipSplitIf,
103 LoopOptsVerify
104 };
105
106 typedef unsigned int node_idx_t;
107 class NodeCloneInfo {
108 private:
109 uint64_t _idx_clone_orig;
110 public:
111
112 void set_idx(node_idx_t idx) {
113 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
114 }
115 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
116
117 void set_gen(int generation) {
118 uint64_t g = (uint64_t)generation << 32;
119 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
120 }
121 int gen() const { return (int)(_idx_clone_orig >> 32); }
122
123 void set(uint64_t x) { _idx_clone_orig = x; }
124 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
125 uint64_t get() const { return _idx_clone_orig; }
126
127 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
128 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
129
130 void dump() const;
131 };
132
133 class CloneMap {
134 friend class Compile;
135 private:
136 bool _debug;
137 Dict* _dict;
138 int _clone_idx; // current cloning iteration/generation in loop unroll
139 public:
140 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
141 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
142 Dict* dict() const { return _dict; }
143 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
144 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
145 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
146 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
147 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
148 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
149 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
150 int max_gen() const;
151 void clone(Node* old, Node* nnn, int gen);
152 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
153 void dump(node_idx_t key) const;
154
155 int clone_idx() const { return _clone_idx; }
156 void set_clone_idx(int x) { _clone_idx = x; }
157 bool is_debug() const { return _debug; }
158 void set_debug(bool debug) { _debug = debug; }
159 static const char* debug_option_name;
160
161 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
162 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
163 };
164
165 //------------------------------Compile----------------------------------------
166 // This class defines a top-level Compiler invocation.
167
168 class Compile : public Phase {
169 friend class VMStructs;
170
171 public:
172 // Fixed alias indexes. (See also MergeMemNode.)
173 enum {
174 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
175 AliasIdxBot = 2, // pseudo-index, aliases to everything
176 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
177 };
178
179 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
180 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
181 // then brackets are put into the log, with time stamps and node counts.
182 // (The time collection itself is always conditionalized on CITime.)
183 class TracePhase : public TraceTime {
184 private:
185 Compile* C;
186 CompileLog* _log;
187 const char* _phase_name;
188 bool _dolog;
189 public:
190 TracePhase(const char* name, elapsedTimer* accumulator);
191 ~TracePhase();
192 };
193
194 // Information per category of alias (memory slice)
195 class AliasType {
196 private:
197 friend class Compile;
198
199 int _index; // unique index, used with MergeMemNode
200 const TypePtr* _adr_type; // normalized address type
201 ciField* _field; // relevant instance field, or null if none
202 const Type* _element; // relevant array element type, or null if none
203 bool _is_rewritable; // false if the memory is write-once only
204 int _general_index; // if this is type is an instance, the general
205 // type that this is an instance of
206
207 void Init(int i, const TypePtr* at);
208
209 public:
210 int index() const { return _index; }
211 const TypePtr* adr_type() const { return _adr_type; }
212 ciField* field() const { return _field; }
213 const Type* element() const { return _element; }
214 bool is_rewritable() const { return _is_rewritable; }
215 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
216 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
217
218 void set_rewritable(bool z) { _is_rewritable = z; }
219 void set_field(ciField* f) {
220 assert(!_field,"");
221 _field = f;
222 if (f->is_final() || f->is_stable()) {
223 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
224 _is_rewritable = false;
225 }
226 }
227 void set_element(const Type* e) {
228 assert(_element == NULL, "");
229 _element = e;
230 }
231
232 BasicType basic_type() const;
233
234 void print_on(outputStream* st) PRODUCT_RETURN;
235 };
236
237 enum {
238 logAliasCacheSize = 6,
239 AliasCacheSize = (1<<logAliasCacheSize)
240 };
241 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
242 enum {
243 trapHistLength = MethodData::_trap_hist_limit
244 };
245
246 // Constant entry of the constant table.
247 class Constant {
248 private:
249 BasicType _type;
250 union {
251 jvalue _value;
252 Metadata* _metadata;
253 } _v;
254 int _offset; // offset of this constant (in bytes) relative to the constant table base.
255 float _freq;
256 bool _can_be_reused; // true (default) if the value can be shared with other users.
257
258 public:
259 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; }
260 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) :
261 _type(type),
262 _offset(-1),
263 _freq(freq),
264 _can_be_reused(can_be_reused)
265 {
266 assert(type != T_METADATA, "wrong constructor");
267 _v._value = value;
268 }
269 Constant(Metadata* metadata, bool can_be_reused = true) :
270 _type(T_METADATA),
271 _offset(-1),
272 _freq(0.0f),
273 _can_be_reused(can_be_reused)
274 {
275 _v._metadata = metadata;
276 }
277
278 bool operator==(const Constant& other);
279
280 BasicType type() const { return _type; }
281
282 jint get_jint() const { return _v._value.i; }
283 jlong get_jlong() const { return _v._value.j; }
284 jfloat get_jfloat() const { return _v._value.f; }
285 jdouble get_jdouble() const { return _v._value.d; }
286 jobject get_jobject() const { return _v._value.l; }
287
288 Metadata* get_metadata() const { return _v._metadata; }
289
290 int offset() const { return _offset; }
291 void set_offset(int offset) { _offset = offset; }
292
293 float freq() const { return _freq; }
294 void inc_freq(float freq) { _freq += freq; }
295
296 bool can_be_reused() const { return _can_be_reused; }
297 };
298
299 // Constant table.
300 class ConstantTable {
301 private:
302 GrowableArray<Constant> _constants; // Constants of this table.
303 int _size; // Size in bytes the emitted constant table takes (including padding).
304 int _table_base_offset; // Offset of the table base that gets added to the constant offsets.
305 int _nof_jump_tables; // Number of jump-tables in this constant table.
306
307 static int qsort_comparator(Constant* a, Constant* b);
308
309 // We use negative frequencies to keep the order of the
310 // jump-tables in which they were added. Otherwise we get into
311 // trouble with relocation.
312 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); }
313
314 public:
315 ConstantTable() :
316 _size(-1),
317 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit).
318 _nof_jump_tables(0)
319 {}
320
321 int size() const { assert(_size != -1, "not calculated yet"); return _size; }
322
323 int calculate_table_base_offset() const; // AD specific
324 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; }
325 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; }
326
327 void emit(CodeBuffer& cb);
328
329 // Returns the offset of the last entry (the top) of the constant table.
330 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); }
331
332 void calculate_offsets_and_size();
333 int find_offset(Constant& con) const;
334
335 void add(Constant& con);
336 Constant add(MachConstantNode* n, BasicType type, jvalue value);
337 Constant add(Metadata* metadata);
338 Constant add(MachConstantNode* n, MachOper* oper);
339 Constant add(MachConstantNode* n, jint i) {
340 jvalue value; value.i = i;
341 return add(n, T_INT, value);
342 }
343 Constant add(MachConstantNode* n, jlong j) {
344 jvalue value; value.j = j;
345 return add(n, T_LONG, value);
346 }
347 Constant add(MachConstantNode* n, jfloat f) {
348 jvalue value; value.f = f;
349 return add(n, T_FLOAT, value);
350 }
351 Constant add(MachConstantNode* n, jdouble d) {
352 jvalue value; value.d = d;
353 return add(n, T_DOUBLE, value);
354 }
355
356 // Jump-table
357 Constant add_jump_table(MachConstantNode* n);
358 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const;
359 };
360
361 private:
362 // Fixed parameters to this compilation.
363 const int _compile_id;
364 const bool _save_argument_registers; // save/restore arg regs for trampolines
365 const bool _subsume_loads; // Load can be matched as part of a larger op.
366 const bool _do_escape_analysis; // Do escape analysis.
367 const bool _eliminate_boxing; // Do boxing elimination.
368 ciMethod* _method; // The method being compiled.
369 int _entry_bci; // entry bci for osr methods.
370 const TypeFunc* _tf; // My kind of signature
371 InlineTree* _ilt; // Ditto (temporary).
372 address _stub_function; // VM entry for stub being compiled, or NULL
373 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
374 address _stub_entry_point; // Compile code entry for generated stub, or NULL
375
376 // Control of this compilation.
377 int _max_inline_size; // Max inline size for this compilation
378 int _freq_inline_size; // Max hot method inline size for this compilation
379 int _fixed_slots; // count of frame slots not allocated by the register
380 // allocator i.e. locks, original deopt pc, etc.
381 uintx _max_node_limit; // Max unique node count during a single compilation.
382 // For deopt
383 int _orig_pc_slot;
384 int _orig_pc_slot_offset_in_bytes;
385
386 // For value type calling convention
387 int _sp_inc_slot;
388 int _sp_inc_slot_offset_in_bytes;
389
390 int _major_progress; // Count of something big happening
391 bool _inlining_progress; // progress doing incremental inlining?
392 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
393 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
394 bool _has_loops; // True if the method _may_ have some loops
395 bool _has_split_ifs; // True if the method _may_ have some split-if
396 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
397 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
398 bool _has_boxed_value; // True if a boxed object is allocated
399 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
400 uint _max_vector_size; // Maximum size of generated vectors
401 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
402 uint _trap_hist[trapHistLength]; // Cumulative traps
403 bool _trap_can_recompile; // Have we emitted a recompiling trap?
404 uint _decompile_count; // Cumulative decompilation counts.
405 bool _do_inlining; // True if we intend to do inlining
406 bool _do_scheduling; // True if we intend to do scheduling
407 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
408 bool _do_count_invocations; // True if we generate code to count invocations
409 bool _do_method_data_update; // True if we generate code to update MethodData*s
410 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
411 bool _use_cmove; // True if CMove should be used without profitability analysis
412 bool _age_code; // True if we need to profile code age (decrement the aging counter)
413 int _AliasLevel; // Locally-adjusted version of AliasLevel flag.
414 bool _print_assembly; // True if we should dump assembly code for this compilation
415 bool _print_inlining; // True if we should print inlining for this compilation
416 bool _print_intrinsics; // True if we should print intrinsics for this compilation
417 #ifndef PRODUCT
418 bool _trace_opto_output;
419 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
420 #endif
421 bool _has_irreducible_loop; // Found irreducible loops
422 // JSR 292
423 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
424 RTMState _rtm_state; // State of Restricted Transactional Memory usage
425 int _loop_opts_cnt; // loop opts round
426 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
427 bool _has_flattened_accesses; // Any known flattened array accesses?
428 bool _flattened_accesses_share_alias; // Initially all flattened array share a single slice
429
430 // Compilation environment.
431 Arena _comp_arena; // Arena with lifetime equivalent to Compile
432 void* _barrier_set_state; // Potential GC barrier state for Compile
433 ciEnv* _env; // CI interface
434 DirectiveSet* _directive; // Compiler directive
435 CompileLog* _log; // from CompilerThread
436 const char* _failure_reason; // for record_failure/failing pattern
437 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics.
438 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching.
439 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
440 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
441 GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency
442 GrowableArray<Node*>* _opaque4_nodes; // List of Opaque4 nodes that have a default value
443 Unique_Node_List* _value_type_nodes; // List of ValueType nodes
444 ConnectionGraph* _congraph;
445 #ifndef PRODUCT
446 IdealGraphPrinter* _printer;
447 #endif
448
449
450 // Node management
451 uint _unique; // Counter for unique Node indices
452 VectorSet _dead_node_list; // Set of dead nodes
453 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
454 // So use this to keep count and make the call O(1).
455 DEBUG_ONLY( Unique_Node_List* _modified_nodes; ) // List of nodes which inputs were modified
456
457 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
458 Arena _node_arena; // Arena for new-space Nodes
459 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
460 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
461 Node* _top; // Unique top node. (Reset by various phases.)
462
463 Node* _immutable_memory; // Initial memory state
464
465 Node* _recent_alloc_obj;
466 Node* _recent_alloc_ctl;
467
468 // Constant table
469 ConstantTable _constant_table; // The constant table for this compile.
470 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
471
472
473 // Blocked array of debugging and profiling information,
474 // tracked per node.
475 enum { _log2_node_notes_block_size = 8,
476 _node_notes_block_size = (1<<_log2_node_notes_block_size)
477 };
478 GrowableArray<Node_Notes*>* _node_note_array;
479 Node_Notes* _default_node_notes; // default notes for new nodes
480
481 // After parsing and every bulk phase we hang onto the Root instruction.
482 // The RootNode instruction is where the whole program begins. It produces
483 // the initial Control and BOTTOM for everybody else.
484
485 // Type management
486 Arena _Compile_types; // Arena for all types
487 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
488 Dict* _type_dict; // Intern table
489 CloneMap _clone_map; // used for recording history of cloned nodes
490 void* _type_hwm; // Last allocation (see Type::operator new/delete)
491 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
492 ciMethod* _last_tf_m; // Cache for
493 const TypeFunc* _last_tf; // TypeFunc::make
494 AliasType** _alias_types; // List of alias types seen so far.
495 int _num_alias_types; // Logical length of _alias_types
496 int _max_alias_types; // Physical length of _alias_types
497 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
498
499 // Parsing, optimization
500 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
501 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
502 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining.
503
504 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after
505 // main parsing has finished.
506 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
507
508 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
509
510 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
511 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
512
513
514 // Inlining may not happen in parse order which would make
515 // PrintInlining output confusing. Keep track of PrintInlining
516 // pieces in order.
517 class PrintInliningBuffer : public ResourceObj {
518 private:
519 CallGenerator* _cg;
520 stringStream* _ss;
521
522 public:
523 PrintInliningBuffer()
524 : _cg(NULL) { _ss = new stringStream(); }
525
526 stringStream* ss() const { return _ss; }
527 CallGenerator* cg() const { return _cg; }
528 void set_cg(CallGenerator* cg) { _cg = cg; }
529 };
530
531 stringStream* _print_inlining_stream;
532 GrowableArray<PrintInliningBuffer>* _print_inlining_list;
533 int _print_inlining_idx;
534 char* _print_inlining_output;
535
536 // Only keep nodes in the expensive node list that need to be optimized
537 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
538 // Use for sorting expensive nodes to bring similar nodes together
539 static int cmp_expensive_nodes(Node** n1, Node** n2);
540 // Expensive nodes list already sorted?
541 bool expensive_nodes_sorted() const;
542 // Remove the speculative part of types and clean up the graph
543 void remove_speculative_types(PhaseIterGVN &igvn);
544
545 void* _replay_inline_data; // Pointer to data loaded from file
546
547 void print_inlining_init();
548 void print_inlining_reinit();
549 void print_inlining_commit();
550 void print_inlining_push();
551 PrintInliningBuffer& print_inlining_current();
552
553 void log_late_inline_failure(CallGenerator* cg, const char* msg);
554
555 public:
556
557 void* barrier_set_state() const { return _barrier_set_state; }
558
559 outputStream* print_inlining_stream() const {
560 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
561 return _print_inlining_stream;
562 }
563
564 void print_inlining_update(CallGenerator* cg);
565 void print_inlining_update_delayed(CallGenerator* cg);
566 void print_inlining_move_to(CallGenerator* cg);
567 void print_inlining_assert_ready();
568 void print_inlining_reset();
569
570 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
571 stringStream ss;
572 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
573 print_inlining_stream()->print("%s", ss.as_string());
574 }
575
576 #ifndef PRODUCT
577 IdealGraphPrinter* printer() { return _printer; }
578 #endif
579
580 void log_late_inline(CallGenerator* cg);
581 void log_inline_id(CallGenerator* cg);
582 void log_inline_failure(const char* msg);
583
584 void* replay_inline_data() const { return _replay_inline_data; }
585
586 // Dump inlining replay data to the stream.
587 void dump_inline_data(outputStream* out);
588
589 private:
590 // Matching, CFG layout, allocation, code generation
591 PhaseCFG* _cfg; // Results of CFG finding
592 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
593 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
594 int _java_calls; // Number of java calls in the method
595 int _inner_loops; // Number of inner loops in the method
596 Matcher* _matcher; // Engine to map ideal to machine instructions
597 PhaseRegAlloc* _regalloc; // Results of register allocation.
598 int _frame_slots; // Size of total frame in stack slots
599 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries
600 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
601 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
602 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
603 int _interpreter_frame_size;
604
605 uint _node_bundling_limit;
606 Bundle* _node_bundling_base; // Information for instruction bundling
607
608 // Instruction bits passed off to the VM
609 int _method_size; // Size of nmethod code segment in bytes
610 CodeBuffer _code_buffer; // Where the code is assembled
611 int _first_block_size; // Size of unvalidated entry point code / OSR poison code
612 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers
613 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code
614 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location)
615 static int _CompiledZap_count; // counter compared against CompileZap[First/Last]
616 BufferBlob* _scratch_buffer_blob; // For temporary code buffers.
617 relocInfo* _scratch_locs_memory; // For temporary code buffers.
618 int _scratch_const_size; // For temporary code buffers.
619 bool _in_scratch_emit_size; // true when in scratch_emit_size.
620
621 void reshape_address(AddPNode* n);
622
623 public:
624 // Accessors
625
626 // The Compile instance currently active in this (compiler) thread.
627 static Compile* current() {
628 return (Compile*) ciEnv::current()->compiler_data();
629 }
630
631 // ID for this compilation. Useful for setting breakpoints in the debugger.
632 int compile_id() const { return _compile_id; }
633 DirectiveSet* directive() const { return _directive; }
634
635 // Does this compilation allow instructions to subsume loads? User
636 // instructions that subsume a load may result in an unschedulable
637 // instruction sequence.
638 bool subsume_loads() const { return _subsume_loads; }
639 /** Do escape analysis. */
640 bool do_escape_analysis() const { return _do_escape_analysis; }
641 /** Do boxing elimination. */
642 bool eliminate_boxing() const { return _eliminate_boxing; }
643 /** Do aggressive boxing elimination. */
644 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
645 bool save_argument_registers() const { return _save_argument_registers; }
646
647
648 // Other fixed compilation parameters.
649 ciMethod* method() const { return _method; }
650 int entry_bci() const { return _entry_bci; }
651 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
652 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
653 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
654 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
655 InlineTree* ilt() const { return _ilt; }
656 address stub_function() const { return _stub_function; }
657 const char* stub_name() const { return _stub_name; }
658 address stub_entry_point() const { return _stub_entry_point; }
659
660 // Control of this compilation.
661 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
662 void set_fixed_slots(int n) { _fixed_slots = n; }
663 int major_progress() const { return _major_progress; }
664 void set_inlining_progress(bool z) { _inlining_progress = z; }
665 int inlining_progress() const { return _inlining_progress; }
666 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
667 int inlining_incrementally() const { return _inlining_incrementally; }
668 void set_do_cleanup(bool z) { _do_cleanup = z; }
669 int do_cleanup() const { return _do_cleanup; }
670 void set_major_progress() { _major_progress++; }
671 void restore_major_progress(int progress) { _major_progress += progress; }
672 void clear_major_progress() { _major_progress = 0; }
673 int max_inline_size() const { return _max_inline_size; }
674 void set_freq_inline_size(int n) { _freq_inline_size = n; }
675 int freq_inline_size() const { return _freq_inline_size; }
676 void set_max_inline_size(int n) { _max_inline_size = n; }
677 bool has_loops() const { return _has_loops; }
678 void set_has_loops(bool z) { _has_loops = z; }
679 bool has_split_ifs() const { return _has_split_ifs; }
680 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
681 bool has_unsafe_access() const { return _has_unsafe_access; }
682 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
683 bool has_stringbuilder() const { return _has_stringbuilder; }
684 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
685 bool has_boxed_value() const { return _has_boxed_value; }
686 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
687 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
688 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
689 uint max_vector_size() const { return _max_vector_size; }
690 void set_max_vector_size(uint s) { _max_vector_size = s; }
691 bool clear_upper_avx() const { return _clear_upper_avx; }
692 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
693 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
694 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
695 bool trap_can_recompile() const { return _trap_can_recompile; }
696 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
697 uint decompile_count() const { return _decompile_count; }
698 void set_decompile_count(uint c) { _decompile_count = c; }
699 bool allow_range_check_smearing() const;
700 bool do_inlining() const { return _do_inlining; }
701 void set_do_inlining(bool z) { _do_inlining = z; }
702 bool do_scheduling() const { return _do_scheduling; }
703 void set_do_scheduling(bool z) { _do_scheduling = z; }
704 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
705 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
706 bool do_count_invocations() const{ return _do_count_invocations; }
707 void set_do_count_invocations(bool z){ _do_count_invocations = z; }
708 bool do_method_data_update() const { return _do_method_data_update; }
709 void set_do_method_data_update(bool z) { _do_method_data_update = z; }
710 bool do_vector_loop() const { return _do_vector_loop; }
711 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
712 bool use_cmove() const { return _use_cmove; }
713 void set_use_cmove(bool z) { _use_cmove = z; }
714 bool age_code() const { return _age_code; }
715 void set_age_code(bool z) { _age_code = z; }
716 int AliasLevel() const { return _AliasLevel; }
717 bool print_assembly() const { return _print_assembly; }
718 void set_print_assembly(bool z) { _print_assembly = z; }
719 bool print_inlining() const { return _print_inlining; }
720 void set_print_inlining(bool z) { _print_inlining = z; }
721 bool print_intrinsics() const { return _print_intrinsics; }
722 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
723 RTMState rtm_state() const { return _rtm_state; }
724 void set_rtm_state(RTMState s) { _rtm_state = s; }
725 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
726 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
727 uint max_node_limit() const { return (uint)_max_node_limit; }
728 void set_max_node_limit(uint n) { _max_node_limit = n; }
729 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
730 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
731 void set_flattened_accesses() { _has_flattened_accesses = true; }
732 bool flattened_accesses_share_alias() const { return _flattened_accesses_share_alias; }
733 void set_flattened_accesses_share_alias(bool z) { _flattened_accesses_share_alias = z; }
734
735 // Support for scalarized value type calling convention
736 bool has_scalarized_args() const { return _method != NULL && _method->has_scalarized_args(); }
737 bool needs_stack_repair() const { return _method != NULL && _method->get_Method()->c2_needs_stack_repair(); }
738 int sp_inc_offset() const { return _sp_inc_slot_offset_in_bytes; }
739
740 // check the CompilerOracle for special behaviours for this compile
741 bool method_has_option(const char * option) {
742 return method() != NULL && method()->has_option(option);
743 }
744
745 #ifndef PRODUCT
746 bool trace_opto_output() const { return _trace_opto_output; }
747 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
748 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
749 int _in_dump_cnt; // Required for dumping ir nodes.
750 #endif
751 bool has_irreducible_loop() const { return _has_irreducible_loop; }
752 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
753
754 // JSR 292
755 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
756 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
757
758 Ticks _latest_stage_start_counter;
759
760 void begin_method() {
761 #ifndef PRODUCT
762 if (_printer && _printer->should_print(1)) {
763 _printer->begin_method();
764 }
765 #endif
766 C->_latest_stage_start_counter.stamp();
767 }
768
769 bool should_print(int level = 1) {
770 #ifndef PRODUCT
771 return (_printer && _printer->should_print(level));
772 #else
773 return false;
774 #endif
775 }
776
777 void print_method(CompilerPhaseType cpt, int level = 1, int idx = 0) {
778 EventCompilerPhase event;
779 if (event.should_commit()) {
780 event.set_starttime(C->_latest_stage_start_counter);
781 event.set_phase((u1) cpt);
782 event.set_compileId(C->_compile_id);
783 event.set_phaseLevel(level);
784 event.commit();
785 }
786
787 #ifndef PRODUCT
788 if (should_print(level)) {
789 char output[1024];
790 if (idx != 0) {
791 sprintf(output, "%s:%d", CompilerPhaseTypeHelper::to_string(cpt), idx);
792 } else {
793 sprintf(output, "%s", CompilerPhaseTypeHelper::to_string(cpt));
794 }
795 _printer->print_method(output, level);
796 }
797 #endif
798 C->_latest_stage_start_counter.stamp();
799 }
800
801 void end_method(int level = 1) {
802 EventCompilerPhase event;
803 if (event.should_commit()) {
804 event.set_starttime(C->_latest_stage_start_counter);
805 event.set_phase((u1) PHASE_END);
806 event.set_compileId(C->_compile_id);
807 event.set_phaseLevel(level);
808 event.commit();
809 }
810 #ifndef PRODUCT
811 if (_printer && _printer->should_print(level)) {
812 _printer->end_method();
813 }
814 #endif
815 }
816
817 int macro_count() const { return _macro_nodes->length(); }
818 int predicate_count() const { return _predicate_opaqs->length();}
819 int expensive_count() const { return _expensive_nodes->length(); }
820 Node* macro_node(int idx) const { return _macro_nodes->at(idx); }
821 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
822 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); }
823 ConnectionGraph* congraph() { return _congraph;}
824 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
825 void add_macro_node(Node * n) {
826 //assert(n->is_macro(), "must be a macro node");
827 assert(!_macro_nodes->contains(n), "duplicate entry in expand list");
828 _macro_nodes->append(n);
829 }
830 void remove_macro_node(Node * n) {
831 // this function may be called twice for a node so check
832 // that the node is in the array before attempting to remove it
833 if (_macro_nodes->contains(n))
834 _macro_nodes->remove(n);
835 // remove from _predicate_opaqs list also if it is there
836 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){
837 _predicate_opaqs->remove(n);
838 }
839 }
840 void add_expensive_node(Node * n);
841 void remove_expensive_node(Node * n) {
842 if (_expensive_nodes->contains(n)) {
843 _expensive_nodes->remove(n);
844 }
845 }
846 void add_predicate_opaq(Node * n) {
847 assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1");
848 assert(_macro_nodes->contains(n), "should have already been in macro list");
849 _predicate_opaqs->append(n);
850 }
851
852 // Range check dependent CastII nodes that can be removed after loop optimizations
853 void add_range_check_cast(Node* n);
854 void remove_range_check_cast(Node* n) {
855 if (_range_check_casts->contains(n)) {
856 _range_check_casts->remove(n);
857 }
858 }
859 Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx); }
860 int range_check_cast_count() const { return _range_check_casts->length(); }
861 // Remove all range check dependent CastIINodes.
862 void remove_range_check_casts(PhaseIterGVN &igvn);
863
864 void add_opaque4_node(Node* n);
865 void remove_opaque4_node(Node* n) {
866 if (_opaque4_nodes->contains(n)) {
867 _opaque4_nodes->remove(n);
868 }
869 }
870 Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx); }
871 int opaque4_count() const { return _opaque4_nodes->length(); }
872 void remove_opaque4_nodes(PhaseIterGVN &igvn);
873
874 // Keep track of value type nodes for later processing
875 void add_value_type(Node* n);
876 void remove_value_type(Node* n);
877 void process_value_types(PhaseIterGVN &igvn);
878 bool can_add_value_type() const { return _value_type_nodes != NULL; }
879
880 void adjust_flattened_array_access_aliases(PhaseIterGVN& igvn);
881
882 // remove the opaque nodes that protect the predicates so that the unused checks and
883 // uncommon traps will be eliminated from the graph.
884 void cleanup_loop_predicates(PhaseIterGVN &igvn);
885 bool is_predicate_opaq(Node * n) {
886 return _predicate_opaqs->contains(n);
887 }
888
889 // Are there candidate expensive nodes for optimization?
890 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
891 // Check whether n1 and n2 are similar
892 static int cmp_expensive_nodes(Node* n1, Node* n2);
893 // Sort expensive nodes to locate similar expensive nodes
894 void sort_expensive_nodes();
895
896 // Compilation environment.
897 Arena* comp_arena() { return &_comp_arena; }
898 ciEnv* env() const { return _env; }
899 CompileLog* log() const { return _log; }
900 bool failing() const { return _env->failing() || _failure_reason != NULL; }
901 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
902
903 bool failure_reason_is(const char* r) const {
904 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
905 }
906
907 void record_failure(const char* reason);
908 void record_method_not_compilable(const char* reason) {
909 // Bailouts cover "all_tiers" when TieredCompilation is off.
910 env()->record_method_not_compilable(reason, !TieredCompilation);
911 // Record failure reason.
912 record_failure(reason);
913 }
914 bool check_node_count(uint margin, const char* reason) {
915 if (live_nodes() + margin > max_node_limit()) {
916 record_method_not_compilable(reason);
917 return true;
918 } else {
919 return false;
920 }
921 }
922
923 // Node management
924 uint unique() const { return _unique; }
925 uint next_unique() { return _unique++; }
926 void set_unique(uint i) { _unique = i; }
927 static int debug_idx() { return debug_only(_debug_idx)+0; }
928 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
929 Arena* node_arena() { return &_node_arena; }
930 Arena* old_arena() { return &_old_arena; }
931 RootNode* root() const { return _root; }
932 void set_root(RootNode* r) { _root = r; }
933 StartNode* start() const; // (Derived from root.)
934 void init_start(StartNode* s);
935 Node* immutable_memory();
936
937 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
938 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
939 void set_recent_alloc(Node* ctl, Node* obj) {
940 _recent_alloc_ctl = ctl;
941 _recent_alloc_obj = obj;
942 }
943 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
944 _dead_node_count++;
945 }
946 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; }
947 uint dead_node_count() { return _dead_node_count; }
948 void reset_dead_node_list() { _dead_node_list.Reset();
949 _dead_node_count = 0;
950 }
951 uint live_nodes() const {
952 int val = _unique - _dead_node_count;
953 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
954 return (uint) val;
955 }
956 #ifdef ASSERT
957 uint count_live_nodes_by_graph_walk();
958 void print_missing_nodes();
959 #endif
960
961 // Record modified nodes to check that they are put on IGVN worklist
962 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
963 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
964 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
965
966 // Constant table
967 ConstantTable& constant_table() { return _constant_table; }
968
969 MachConstantBaseNode* mach_constant_base_node();
970 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
971 // Generated by adlc, true if CallNode requires MachConstantBase.
972 bool needs_clone_jvms();
973
974 // Handy undefined Node
975 Node* top() const { return _top; }
976
977 // these are used by guys who need to know about creation and transformation of top:
978 Node* cached_top_node() { return _top; }
979 void set_cached_top_node(Node* tn);
980
981 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
982 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
983 Node_Notes* default_node_notes() const { return _default_node_notes; }
984 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
985
986 Node_Notes* node_notes_at(int idx) {
987 return locate_node_notes(_node_note_array, idx, false);
988 }
989 inline bool set_node_notes_at(int idx, Node_Notes* value);
990
991 // Copy notes from source to dest, if they exist.
992 // Overwrite dest only if source provides something.
993 // Return true if information was moved.
994 bool copy_node_notes_to(Node* dest, Node* source);
995
996 // Workhorse function to sort out the blocked Node_Notes array:
997 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
998 int idx, bool can_grow = false);
999
1000 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
1001
1002 // Type management
1003 Arena* type_arena() { return _type_arena; }
1004 Dict* type_dict() { return _type_dict; }
1005 void* type_hwm() { return _type_hwm; }
1006 size_t type_last_size() { return _type_last_size; }
1007 int num_alias_types() { return _num_alias_types; }
1008
1009 void init_type_arena() { _type_arena = &_Compile_types; }
1010 void set_type_arena(Arena* a) { _type_arena = a; }
1011 void set_type_dict(Dict* d) { _type_dict = d; }
1012 void set_type_hwm(void* p) { _type_hwm = p; }
1013 void set_type_last_size(size_t sz) { _type_last_size = sz; }
1014
1015 const TypeFunc* last_tf(ciMethod* m) {
1016 return (m == _last_tf_m) ? _last_tf : NULL;
1017 }
1018 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
1019 assert(m != NULL || tf == NULL, "");
1020 _last_tf_m = m;
1021 _last_tf = tf;
1022 }
1023
1024 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
1025 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
1026 bool have_alias_type(const TypePtr* adr_type);
1027 AliasType* alias_type(ciField* field);
1028
1029 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, NULL, uncached)->index(); }
1030 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
1031 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
1032
1033 // Building nodes
1034 void rethrow_exceptions(JVMState* jvms);
1035 void return_values(JVMState* jvms);
1036 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1037
1038 // Decide how to build a call.
1039 // The profile factor is a discount to apply to this site's interp. profile.
1040 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1041 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
1042 bool allow_intrinsics = true, bool delayed_forbidden = false);
1043 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1044 return should_delay_string_inlining(call_method, jvms) ||
1045 should_delay_boxing_inlining(call_method, jvms);
1046 }
1047 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1048 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1049
1050 // Helper functions to identify inlining potential at call-site
1051 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
1052 ciKlass* holder, ciMethod* callee,
1053 const TypeOopPtr* receiver_type, bool is_virtual,
1054 bool &call_does_dispatch, int &vtable_index,
1055 bool check_access = true);
1056 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
1057 ciMethod* callee, const TypeOopPtr* receiver_type,
1058 bool check_access = true);
1059
1060 // Report if there were too many traps at a current method and bci.
1061 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1062 // If there is no MDO at all, report no trap unless told to assume it.
1063 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1064 // This version, unspecific to a particular bci, asks if
1065 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1066 bool too_many_traps(Deoptimization::DeoptReason reason,
1067 // Privately used parameter for logging:
1068 ciMethodData* logmd = NULL);
1069 // Report if there were too many recompiles at a method and bci.
1070 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1071 // Report if there were too many traps or recompiles at a method and bci.
1072 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1073 return too_many_traps(method, bci, reason) ||
1074 too_many_recompiles(method, bci, reason);
1075 }
1076 // Return a bitset with the reasons where deoptimization is allowed,
1077 // i.e., where there were not too many uncommon traps.
1078 int _allowed_reasons;
1079 int allowed_deopt_reasons() { return _allowed_reasons; }
1080 void set_allowed_deopt_reasons();
1081
1082 // Parsing, optimization
1083 PhaseGVN* initial_gvn() { return _initial_gvn; }
1084 Unique_Node_List* for_igvn() { return _for_igvn; }
1085 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
1086 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1087 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
1088
1089 // Replace n by nn using initial_gvn, calling hash_delete and
1090 // record_for_igvn as needed.
1091 void gvn_replace_by(Node* n, Node* nn);
1092
1093
1094 void identify_useful_nodes(Unique_Node_List &useful);
1095 void update_dead_node_list(Unique_Node_List &useful);
1096 void remove_useless_nodes (Unique_Node_List &useful);
1097
1098 WarmCallInfo* warm_calls() const { return _warm_calls; }
1099 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
1100 WarmCallInfo* pop_warm_call();
1101
1102 // Record this CallGenerator for inlining at the end of parsing.
1103 void add_late_inline(CallGenerator* cg) {
1104 _late_inlines.insert_before(_late_inlines_pos, cg);
1105 _late_inlines_pos++;
1106 }
1107
1108 void prepend_late_inline(CallGenerator* cg) {
1109 _late_inlines.insert_before(0, cg);
1110 }
1111
1112 void add_string_late_inline(CallGenerator* cg) {
1113 _string_late_inlines.push(cg);
1114 }
1115
1116 void add_boxing_late_inline(CallGenerator* cg) {
1117 _boxing_late_inlines.push(cg);
1118 }
1119
1120 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1121
1122 void process_print_inlining();
1123 void dump_print_inlining();
1124
1125 bool over_inlining_cutoff() const {
1126 if (!inlining_incrementally()) {
1127 return unique() > (uint)NodeCountInliningCutoff;
1128 } else {
1129 // Give some room for incremental inlining algorithm to "breathe"
1130 // and avoid thrashing when live node count is close to the limit.
1131 // Keep in mind that live_nodes() isn't accurate during inlining until
1132 // dead node elimination step happens (see Compile::inline_incrementally).
1133 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1134 }
1135 }
1136
1137 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1138 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--; }
1139 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1140
1141 bool inline_incrementally_one();
1142 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1143 void inline_incrementally(PhaseIterGVN& igvn);
1144 void inline_string_calls(bool parse_time);
1145 void inline_boxing_calls(PhaseIterGVN& igvn);
1146 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1147 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1148
1149 // Matching, CFG layout, allocation, code generation
1150 PhaseCFG* cfg() { return _cfg; }
1151 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1152 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1153 bool has_java_calls() const { return _java_calls > 0; }
1154 int java_calls() const { return _java_calls; }
1155 int inner_loops() const { return _inner_loops; }
1156 Matcher* matcher() { return _matcher; }
1157 PhaseRegAlloc* regalloc() { return _regalloc; }
1158 int frame_slots() const { return _frame_slots; }
1159 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words'
1160 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; }
1161 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1162 Arena* indexSet_arena() { return _indexSet_arena; }
1163 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1164 uint node_bundling_limit() { return _node_bundling_limit; }
1165 Bundle* node_bundling_base() { return _node_bundling_base; }
1166 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; }
1167 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; }
1168 bool starts_bundle(const Node *n) const;
1169 bool need_stack_bang(int frame_size_in_bytes) const;
1170 bool need_register_stack_bang() const;
1171
1172 void update_interpreter_frame_size(int size) {
1173 if (_interpreter_frame_size < size) {
1174 _interpreter_frame_size = size;
1175 }
1176 }
1177 int bang_size_in_bytes() const;
1178
1179 void set_matcher(Matcher* m) { _matcher = m; }
1180 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1181 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1182 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1183
1184 // Remember if this compilation changes hardware mode to 24-bit precision
1185 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1186 _select_24_bit_instr = selection;
1187 _in_24_bit_fp_mode = mode;
1188 }
1189
1190 void set_java_calls(int z) { _java_calls = z; }
1191 void set_inner_loops(int z) { _inner_loops = z; }
1192
1193 // Instruction bits passed off to the VM
1194 int code_size() { return _method_size; }
1195 CodeBuffer* code_buffer() { return &_code_buffer; }
1196 int first_block_size() { return _first_block_size; }
1197 void set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } }
1198 ExceptionHandlerTable* handler_table() { return &_handler_table; }
1199 ImplicitExceptionTable* inc_table() { return &_inc_table; }
1200 OopMapSet* oop_map_set() { return _oop_map_set; }
1201 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1202 Dependencies* dependencies() { return env()->dependencies(); }
1203 static int CompiledZap_count() { return _CompiledZap_count; }
1204 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; }
1205 void init_scratch_buffer_blob(int const_size);
1206 void clear_scratch_buffer_blob();
1207 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; }
1208 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; }
1209 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; }
1210
1211 // emit to scratch blob, report resulting size
1212 uint scratch_emit_size(const Node* n);
1213 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; }
1214 bool in_scratch_emit_size() const { return _in_scratch_emit_size; }
1215
1216 enum ScratchBufferBlob {
1217 #if defined(PPC64)
1218 MAX_inst_size = 2048,
1219 #else
1220 MAX_inst_size = 1024,
1221 #endif
1222 MAX_locs_size = 128, // number of relocInfo elements
1223 MAX_const_size = 128,
1224 MAX_stubs_size = 128
1225 };
1226
1227 // Major entry point. Given a Scope, compile the associated method.
1228 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1229 // replacement, entry_bci indicates the bytecode for which to compile a
1230 // continuation.
1231 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
1232 int entry_bci, bool subsume_loads, bool do_escape_analysis,
1233 bool eliminate_boxing, DirectiveSet* directive);
1234
1235 // Second major entry point. From the TypeFunc signature, generate code
1236 // to pass arguments from the Java calling convention to the C calling
1237 // convention.
1238 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1239 address stub_function, const char *stub_name,
1240 int is_fancy_jump, bool pass_tls,
1241 bool save_arg_registers, bool return_pc, DirectiveSet* directive);
1242
1243 // From the TypeFunc signature, generate code to pass arguments
1244 // from Compiled calling convention to Interpreter's calling convention
1245 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry);
1246
1247 // From the TypeFunc signature, generate code to pass arguments
1248 // from Interpreter's calling convention to Compiler's calling convention
1249 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf);
1250
1251 // Are we compiling a method?
1252 bool has_method() { return method() != NULL; }
1253
1254 // Maybe print some information about this compile.
1255 void print_compile_messages();
1256
1257 // Final graph reshaping, a post-pass after the regular optimizer is done.
1258 bool final_graph_reshaping();
1259
1260 // returns true if adr is completely contained in the given alias category
1261 bool must_alias(const TypePtr* adr, int alias_idx);
1262
1263 // returns true if adr overlaps with the given alias category
1264 bool can_alias(const TypePtr* adr, int alias_idx);
1265
1266 // Driver for converting compiler's IR into machine code bits
1267 void Output();
1268
1269 // Accessors for node bundling info.
1270 Bundle* node_bundling(const Node *n);
1271 bool valid_bundle_info(const Node *n);
1272
1273 // Schedule and Bundle the instructions
1274 void ScheduleAndBundle();
1275
1276 // Build OopMaps for each GC point
1277 void BuildOopMaps();
1278
1279 // Append debug info for the node "local" at safepoint node "sfpt" to the
1280 // "array", May also consult and add to "objs", which describes the
1281 // scalar-replaced objects.
1282 void FillLocArray( int idx, MachSafePointNode* sfpt,
1283 Node *local, GrowableArray<ScopeValue*> *array,
1284 GrowableArray<ScopeValue*> *objs );
1285
1286 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
1287 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
1288 // Requres that "objs" does not contains an ObjectValue whose id matches
1289 // that of "sv. Appends "sv".
1290 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
1291 ObjectValue* sv );
1292
1293 // Process an OopMap Element while emitting nodes
1294 void Process_OopMap_Node(MachNode *mach, int code_offset);
1295
1296 // Initialize code buffer
1297 CodeBuffer* init_buffer(uint* blk_starts);
1298
1299 // Write out basic block data to code buffer
1300 void fill_buffer(CodeBuffer* cb, uint* blk_starts);
1301
1302 // Determine which variable sized branches can be shortened
1303 void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size);
1304
1305 // Compute the size of first NumberOfLoopInstrToAlign instructions
1306 // at the head of a loop.
1307 void compute_loop_first_inst_sizes();
1308
1309 // Compute the information for the exception tables
1310 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels);
1311
1312 // Stack slots that may be unused by the calling convention but must
1313 // otherwise be preserved. On Intel this includes the return address.
1314 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1315 uint in_preserve_stack_slots();
1316
1317 // "Top of Stack" slots that may be unused by the calling convention but must
1318 // otherwise be preserved.
1319 // On Intel these are not necessary and the value can be zero.
1320 // On Sparc this describes the words reserved for storing a register window
1321 // when an interrupt occurs.
1322 static uint out_preserve_stack_slots();
1323
1324 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1325 // for calls to C. Supports the var-args backing area for register parms.
1326 uint varargs_C_out_slots_killed() const;
1327
1328 // Number of Stack Slots consumed by a synchronization entry
1329 int sync_stack_slots() const;
1330
1331 // Compute the name of old_SP. See <arch>.ad for frame layout.
1332 OptoReg::Name compute_old_SP();
1333
1334 private:
1335 // Phase control:
1336 void Init(int aliaslevel); // Prepare for a single compilation
1337 int Inline_Warm(); // Find more inlining work.
1338 void Finish_Warm(); // Give up on further inlines.
1339 void Optimize(); // Given a graph, optimize it
1340 void Code_Gen(); // Generate code from a graph
1341
1342 // Management of the AliasType table.
1343 void grow_alias_types();
1344 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1345 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1346 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1347
1348 void verify_top(Node*) const PRODUCT_RETURN;
1349
1350 // Intrinsic setup.
1351 void register_library_intrinsics(); // initializer
1352 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1353 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1354 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1355 void register_intrinsic(CallGenerator* cg); // update fn
1356
1357 #ifndef PRODUCT
1358 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT];
1359 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT];
1360 #endif
1361 // Function calls made by the public function final_graph_reshaping.
1362 // No need to be made public as they are not called elsewhere.
1363 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1364 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop);
1365 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1366 void eliminate_redundant_card_marks(Node* n);
1367
1368 public:
1369
1370 // Note: Histogram array size is about 1 Kb.
1371 enum { // flag bits:
1372 _intrinsic_worked = 1, // succeeded at least once
1373 _intrinsic_failed = 2, // tried it but it failed
1374 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1375 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1376 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1377 };
1378 // Update histogram. Return boolean if this is a first-time occurrence.
1379 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1380 bool is_virtual, int flags) PRODUCT_RETURN0;
1381 static void print_intrinsic_statistics() PRODUCT_RETURN;
1382
1383 // Graph verification code
1384 // Walk the node list, verifying that there is a one-to-one
1385 // correspondence between Use-Def edges and Def-Use edges
1386 // The option no_dead_code enables stronger checks that the
1387 // graph is strongly connected from root in both directions.
1388 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1389
1390 // End-of-run dumps.
1391 static void print_statistics() PRODUCT_RETURN;
1392
1393 // Dump formatted assembly
1394 #if defined(SUPPORT_OPTO_ASSEMBLY)
1395 void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit);
1396 void dump_asm(int* pcs = NULL, uint pc_limit = 0) { dump_asm_on(tty, pcs, pc_limit); }
1397 #else
1398 void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit) { return; }
1399 void dump_asm(int* pcs = NULL, uint pc_limit = 0) { return; }
1400 #endif
1401 void dump_pc(int *pcs, int pc_limit, Node *n);
1402
1403 // Verify ADLC assumptions during startup
1404 static void adlc_verification() PRODUCT_RETURN;
1405
1406 // Definitions of pd methods
1407 static void pd_compiler2_init();
1408
1409 // Static parse-time type checking logic for gen_subtype_check:
1410 enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1411 int static_subtype_check(ciKlass* superk, ciKlass* subk);
1412
1413 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1414 // Optional control dependency (for example, on range check)
1415 Node* ctrl = NULL);
1416
1417 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1418 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl);
1419
1420 Node* optimize_acmp(PhaseGVN* phase, Node* a, Node* b);
1421
1422 // Auxiliary method for randomized fuzzing/stressing
1423 static bool randomized_select(int count);
1424
1425 // supporting clone_map
1426 CloneMap& clone_map();
1427 void set_clone_map(Dict* d);
1428
1429 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1430 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1431 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1432 };
1433
1434 #endif // SHARE_OPTO_COMPILE_HPP