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 C2Compiler;
52 class CallGenerator;
53 class CloneMap;
54 class ConnectionGraph;
55 class IdealGraphPrinter;
56 class InlineTree;
57 class Int_Array;
58 class Matcher;
59 class MachConstantNode;
60 class MachConstantBaseNode;
61 class MachNode;
62 class MachOper;
63 class MachSafePointNode;
64 class Node;
65 class Node_Array;
66 class Node_Notes;
67 class NodeCloneInfo;
68 class OptoReg;
69 class PhaseCFG;
70 class PhaseGVN;
71 class PhaseIterGVN;
72 class PhaseRegAlloc;
73 class PhaseCCP;
74 class PhaseCCP_DCE;
75 class RootNode;
76 class relocInfo;
77 class Scope;
78 class StartNode;
79 class SafePointNode;
80 class JVMState;
81 class Type;
82 class TypeData;
83 class TypeInt;
84 class TypePtr;
85 class TypeOopPtr;
86 class TypeFunc;
87 class Unique_Node_List;
88 class nmethod;
89 class WarmCallInfo;
90 class Node_Stack;
91 struct Final_Reshape_Counts;
92
93 enum LoopOptsMode {
94 LoopOptsDefault,
95 LoopOptsNone,
96 LoopOptsShenandoahExpand,
97 LoopOptsShenandoahPostExpand,
98 LoopOptsSkipSplitIf,
99 LoopOptsVerify
100 };
101
102 typedef unsigned int node_idx_t;
103 class NodeCloneInfo {
104 private:
105 uint64_t _idx_clone_orig;
106 public:
107
108 void set_idx(node_idx_t idx) {
109 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
110 }
111 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
112
113 void set_gen(int generation) {
114 uint64_t g = (uint64_t)generation << 32;
115 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
116 }
117 int gen() const { return (int)(_idx_clone_orig >> 32); }
118
119 void set(uint64_t x) { _idx_clone_orig = x; }
120 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
121 uint64_t get() const { return _idx_clone_orig; }
122
123 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
124 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
125
126 void dump() const;
127 };
128
129 class CloneMap {
130 friend class Compile;
131 private:
132 bool _debug;
133 Dict* _dict;
134 int _clone_idx; // current cloning iteration/generation in loop unroll
135 public:
136 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
137 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
138 Dict* dict() const { return _dict; }
139 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
140 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
141 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
142 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
143 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
144 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
145 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
146 int max_gen() const;
147 void clone(Node* old, Node* nnn, int gen);
148 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
149 void dump(node_idx_t key) const;
150
151 int clone_idx() const { return _clone_idx; }
152 void set_clone_idx(int x) { _clone_idx = x; }
153 bool is_debug() const { return _debug; }
154 void set_debug(bool debug) { _debug = debug; }
155 static const char* debug_option_name;
156
157 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
158 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
159 };
160
161 //------------------------------Compile----------------------------------------
162 // This class defines a top-level Compiler invocation.
163
164 class Compile : public Phase {
165 friend class VMStructs;
166
167 public:
168 // Fixed alias indexes. (See also MergeMemNode.)
169 enum {
170 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
171 AliasIdxBot = 2, // pseudo-index, aliases to everything
172 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
173 };
174
175 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
176 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
177 // then brackets are put into the log, with time stamps and node counts.
178 // (The time collection itself is always conditionalized on CITime.)
179 class TracePhase : public TraceTime {
180 private:
181 Compile* C;
182 CompileLog* _log;
183 const char* _phase_name;
184 bool _dolog;
185 public:
186 TracePhase(const char* name, elapsedTimer* accumulator);
187 ~TracePhase();
188 };
189
190 // Information per category of alias (memory slice)
191 class AliasType {
192 private:
193 friend class Compile;
194
195 int _index; // unique index, used with MergeMemNode
196 const TypePtr* _adr_type; // normalized address type
197 ciField* _field; // relevant instance field, or null if none
198 const Type* _element; // relevant array element type, or null if none
199 bool _is_rewritable; // false if the memory is write-once only
200 int _general_index; // if this is type is an instance, the general
201 // type that this is an instance of
202
203 void Init(int i, const TypePtr* at);
204
205 public:
206 int index() const { return _index; }
207 const TypePtr* adr_type() const { return _adr_type; }
208 ciField* field() const { return _field; }
209 const Type* element() const { return _element; }
210 bool is_rewritable() const { return _is_rewritable; }
211 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
212 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
213
214 void set_rewritable(bool z) { _is_rewritable = z; }
215 void set_field(ciField* f) {
216 assert(!_field,"");
217 _field = f;
218 if (f->is_final() || f->is_stable()) {
219 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
220 _is_rewritable = false;
221 }
222 }
223 void set_element(const Type* e) {
224 assert(_element == NULL, "");
225 _element = e;
226 }
227
228 BasicType basic_type() const;
229
230 void print_on(outputStream* st) PRODUCT_RETURN;
231 };
232
233 enum {
234 logAliasCacheSize = 6,
235 AliasCacheSize = (1<<logAliasCacheSize)
236 };
237 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
238 enum {
239 trapHistLength = MethodData::_trap_hist_limit
240 };
241
242 // Constant entry of the constant table.
243 class Constant {
244 private:
245 BasicType _type;
246 union {
247 jvalue _value;
248 Metadata* _metadata;
249 } _v;
250 int _offset; // offset of this constant (in bytes) relative to the constant table base.
251 float _freq;
252 bool _can_be_reused; // true (default) if the value can be shared with other users.
253
254 public:
255 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; }
256 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) :
257 _type(type),
258 _offset(-1),
259 _freq(freq),
260 _can_be_reused(can_be_reused)
261 {
262 assert(type != T_METADATA, "wrong constructor");
263 _v._value = value;
264 }
265 Constant(Metadata* metadata, bool can_be_reused = true) :
266 _type(T_METADATA),
267 _offset(-1),
268 _freq(0.0f),
269 _can_be_reused(can_be_reused)
270 {
271 _v._metadata = metadata;
272 }
273
274 bool operator==(const Constant& other);
275
276 BasicType type() const { return _type; }
277
278 jint get_jint() const { return _v._value.i; }
279 jlong get_jlong() const { return _v._value.j; }
280 jfloat get_jfloat() const { return _v._value.f; }
281 jdouble get_jdouble() const { return _v._value.d; }
282 jobject get_jobject() const { return _v._value.l; }
283
284 Metadata* get_metadata() const { return _v._metadata; }
285
286 int offset() const { return _offset; }
287 void set_offset(int offset) { _offset = offset; }
288
289 float freq() const { return _freq; }
290 void inc_freq(float freq) { _freq += freq; }
291
292 bool can_be_reused() const { return _can_be_reused; }
293 };
294
295 // Constant table.
296 class ConstantTable {
297 private:
298 GrowableArray<Constant> _constants; // Constants of this table.
299 int _size; // Size in bytes the emitted constant table takes (including padding).
300 int _table_base_offset; // Offset of the table base that gets added to the constant offsets.
301 int _nof_jump_tables; // Number of jump-tables in this constant table.
302
303 static int qsort_comparator(Constant* a, Constant* b);
304
305 // We use negative frequencies to keep the order of the
306 // jump-tables in which they were added. Otherwise we get into
307 // trouble with relocation.
308 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); }
309
310 public:
311 ConstantTable() :
312 _size(-1),
313 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit).
314 _nof_jump_tables(0)
315 {}
316
317 int size() const { assert(_size != -1, "not calculated yet"); return _size; }
318
319 int calculate_table_base_offset() const; // AD specific
320 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; }
321 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; }
322
323 void emit(CodeBuffer& cb);
324
325 // Returns the offset of the last entry (the top) of the constant table.
326 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); }
327
328 void calculate_offsets_and_size();
329 int find_offset(Constant& con) const;
330
331 void add(Constant& con);
332 Constant add(MachConstantNode* n, BasicType type, jvalue value);
333 Constant add(Metadata* metadata);
334 Constant add(MachConstantNode* n, MachOper* oper);
335 Constant add(MachConstantNode* n, jint i) {
336 jvalue value; value.i = i;
337 return add(n, T_INT, value);
338 }
339 Constant add(MachConstantNode* n, jlong j) {
340 jvalue value; value.j = j;
341 return add(n, T_LONG, value);
342 }
343 Constant add(MachConstantNode* n, jfloat f) {
344 jvalue value; value.f = f;
345 return add(n, T_FLOAT, value);
346 }
347 Constant add(MachConstantNode* n, jdouble d) {
348 jvalue value; value.d = d;
349 return add(n, T_DOUBLE, value);
350 }
351
352 // Jump-table
353 Constant add_jump_table(MachConstantNode* n);
354 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const;
355 };
356
357 private:
358 // Fixed parameters to this compilation.
359 const int _compile_id;
360 const bool _save_argument_registers; // save/restore arg regs for trampolines
361 const bool _subsume_loads; // Load can be matched as part of a larger op.
362 const bool _do_escape_analysis; // Do escape analysis.
363 const bool _eliminate_boxing; // Do boxing elimination.
364 ciMethod* _method; // The method being compiled.
365 int _entry_bci; // entry bci for osr methods.
366 const TypeFunc* _tf; // My kind of signature
367 InlineTree* _ilt; // Ditto (temporary).
368 address _stub_function; // VM entry for stub being compiled, or NULL
369 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
370 address _stub_entry_point; // Compile code entry for generated stub, or NULL
371
372 // Control of this compilation.
373 int _max_inline_size; // Max inline size for this compilation
374 int _freq_inline_size; // Max hot method inline size for this compilation
375 int _fixed_slots; // count of frame slots not allocated by the register
376 // allocator i.e. locks, original deopt pc, etc.
377 uintx _max_node_limit; // Max unique node count during a single compilation.
378 // For deopt
379 int _orig_pc_slot;
380 int _orig_pc_slot_offset_in_bytes;
381
382 int _major_progress; // Count of something big happening
383 bool _inlining_progress; // progress doing incremental inlining?
384 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
385 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
386 bool _has_loops; // True if the method _may_ have some loops
387 bool _has_split_ifs; // True if the method _may_ have some split-if
388 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
389 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
390 bool _has_boxed_value; // True if a boxed object is allocated
391 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
392 uint _max_vector_size; // Maximum size of generated vectors
393 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
394 uint _trap_hist[trapHistLength]; // Cumulative traps
395 bool _trap_can_recompile; // Have we emitted a recompiling trap?
396 uint _decompile_count; // Cumulative decompilation counts.
397 bool _do_inlining; // True if we intend to do inlining
398 bool _do_scheduling; // True if we intend to do scheduling
399 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
400 bool _do_count_invocations; // True if we generate code to count invocations
401 bool _do_method_data_update; // True if we generate code to update MethodData*s
402 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
403 bool _use_cmove; // True if CMove should be used without profitability analysis
404 bool _age_code; // True if we need to profile code age (decrement the aging counter)
405 int _AliasLevel; // Locally-adjusted version of AliasLevel flag.
406 bool _print_assembly; // True if we should dump assembly code for this compilation
407 bool _print_inlining; // True if we should print inlining for this compilation
408 bool _print_intrinsics; // True if we should print intrinsics for this compilation
409 #ifndef PRODUCT
410 bool _trace_opto_output;
411 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
412 #endif
413 bool _has_irreducible_loop; // Found irreducible loops
414 // JSR 292
415 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
416 RTMState _rtm_state; // State of Restricted Transactional Memory usage
417 int _loop_opts_cnt; // loop opts round
418 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
419
420 // Compilation environment.
421 Arena _comp_arena; // Arena with lifetime equivalent to Compile
422 void* _barrier_set_state; // Potential GC barrier state for Compile
423 ciEnv* _env; // CI interface
424 DirectiveSet* _directive; // Compiler directive
425 CompileLog* _log; // from CompilerThread
426 const char* _failure_reason; // for record_failure/failing pattern
427 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics.
428 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching.
429 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
430 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
431 GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency
432 GrowableArray<Node*>* _opaque4_nodes; // List of Opaque4 nodes that have a default value
433 ConnectionGraph* _congraph;
434 #ifndef PRODUCT
435 IdealGraphPrinter* _printer;
436 #endif
437
438
439 // Node management
440 uint _unique; // Counter for unique Node indices
441 VectorSet _dead_node_list; // Set of dead nodes
442 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
443 // So use this to keep count and make the call O(1).
444 DEBUG_ONLY( Unique_Node_List* _modified_nodes; ) // List of nodes which inputs were modified
445
446 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
447 Arena _node_arena; // Arena for new-space Nodes
448 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
449 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
450 Node* _top; // Unique top node. (Reset by various phases.)
451
452 Node* _immutable_memory; // Initial memory state
453
454 Node* _recent_alloc_obj;
455 Node* _recent_alloc_ctl;
456
457 // Constant table
458 ConstantTable _constant_table; // The constant table for this compile.
459 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
460
461
462 // Blocked array of debugging and profiling information,
463 // tracked per node.
464 enum { _log2_node_notes_block_size = 8,
465 _node_notes_block_size = (1<<_log2_node_notes_block_size)
466 };
467 GrowableArray<Node_Notes*>* _node_note_array;
468 Node_Notes* _default_node_notes; // default notes for new nodes
469
470 // After parsing and every bulk phase we hang onto the Root instruction.
471 // The RootNode instruction is where the whole program begins. It produces
472 // the initial Control and BOTTOM for everybody else.
473
474 // Type management
475 Arena _Compile_types; // Arena for all types
476 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
477 Dict* _type_dict; // Intern table
478 CloneMap _clone_map; // used for recording history of cloned nodes
479 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
480 ciMethod* _last_tf_m; // Cache for
481 const TypeFunc* _last_tf; // TypeFunc::make
482 AliasType** _alias_types; // List of alias types seen so far.
483 int _num_alias_types; // Logical length of _alias_types
484 int _max_alias_types; // Physical length of _alias_types
485 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
486
487 // Parsing, optimization
488 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
489 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
490 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining.
491
492 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after
493 // main parsing has finished.
494 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
495
496 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
497
498 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
499 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
500
501
502 // Inlining may not happen in parse order which would make
503 // PrintInlining output confusing. Keep track of PrintInlining
504 // pieces in order.
505 class PrintInliningBuffer : public ResourceObj {
506 private:
507 CallGenerator* _cg;
508 stringStream* _ss;
509
510 public:
511 PrintInliningBuffer()
512 : _cg(NULL) { _ss = new stringStream(); }
513
514 stringStream* ss() const { return _ss; }
515 CallGenerator* cg() const { return _cg; }
516 void set_cg(CallGenerator* cg) { _cg = cg; }
517 };
518
519 stringStream* _print_inlining_stream;
520 GrowableArray<PrintInliningBuffer>* _print_inlining_list;
521 int _print_inlining_idx;
522 char* _print_inlining_output;
523
524 // Only keep nodes in the expensive node list that need to be optimized
525 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
526 // Use for sorting expensive nodes to bring similar nodes together
527 static int cmp_expensive_nodes(Node** n1, Node** n2);
528 // Expensive nodes list already sorted?
529 bool expensive_nodes_sorted() const;
530 // Remove the speculative part of types and clean up the graph
531 void remove_speculative_types(PhaseIterGVN &igvn);
532
533 void* _replay_inline_data; // Pointer to data loaded from file
534
535 void print_inlining_init();
536 void print_inlining_reinit();
537 void print_inlining_commit();
538 void print_inlining_push();
539 PrintInliningBuffer& print_inlining_current();
540
541 void log_late_inline_failure(CallGenerator* cg, const char* msg);
542
543 public:
544
545 void* barrier_set_state() const { return _barrier_set_state; }
546
547 outputStream* print_inlining_stream() const {
548 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
549 return _print_inlining_stream;
550 }
551
552 void print_inlining_update(CallGenerator* cg);
553 void print_inlining_update_delayed(CallGenerator* cg);
554 void print_inlining_move_to(CallGenerator* cg);
555 void print_inlining_assert_ready();
556 void print_inlining_reset();
557
558 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
559 stringStream ss;
560 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
561 print_inlining_stream()->print("%s", ss.as_string());
562 }
563
564 #ifndef PRODUCT
565 IdealGraphPrinter* printer() { return _printer; }
566 #endif
567
568 void log_late_inline(CallGenerator* cg);
569 void log_inline_id(CallGenerator* cg);
570 void log_inline_failure(const char* msg);
571
572 void* replay_inline_data() const { return _replay_inline_data; }
573
574 // Dump inlining replay data to the stream.
575 void dump_inline_data(outputStream* out);
576
577 private:
578 // Matching, CFG layout, allocation, code generation
579 PhaseCFG* _cfg; // Results of CFG finding
580 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
581 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
582 int _java_calls; // Number of java calls in the method
583 int _inner_loops; // Number of inner loops in the method
584 Matcher* _matcher; // Engine to map ideal to machine instructions
585 PhaseRegAlloc* _regalloc; // Results of register allocation.
586 int _frame_slots; // Size of total frame in stack slots
587 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries
588 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
589 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
590 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
591 int _interpreter_frame_size;
592
593 uint _node_bundling_limit;
594 Bundle* _node_bundling_base; // Information for instruction bundling
595
596 // Instruction bits passed off to the VM
597 int _method_size; // Size of nmethod code segment in bytes
598 CodeBuffer _code_buffer; // Where the code is assembled
599 int _first_block_size; // Size of unvalidated entry point code / OSR poison code
600 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers
601 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code
602 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location)
603 static int _CompiledZap_count; // counter compared against CompileZap[First/Last]
604 BufferBlob* _scratch_buffer_blob; // For temporary code buffers.
605 relocInfo* _scratch_locs_memory; // For temporary code buffers.
606 int _scratch_const_size; // For temporary code buffers.
607 bool _in_scratch_emit_size; // true when in scratch_emit_size.
608
609 void reshape_address(AddPNode* n);
610
611 public:
612 // Accessors
613
614 // The Compile instance currently active in this (compiler) thread.
615 static Compile* current() {
616 return (Compile*) ciEnv::current()->compiler_data();
617 }
618
619 // ID for this compilation. Useful for setting breakpoints in the debugger.
620 int compile_id() const { return _compile_id; }
621 DirectiveSet* directive() const { return _directive; }
622
623 // Does this compilation allow instructions to subsume loads? User
624 // instructions that subsume a load may result in an unschedulable
625 // instruction sequence.
626 bool subsume_loads() const { return _subsume_loads; }
627 /** Do escape analysis. */
628 bool do_escape_analysis() const { return _do_escape_analysis; }
629 /** Do boxing elimination. */
630 bool eliminate_boxing() const { return _eliminate_boxing; }
631 /** Do aggressive boxing elimination. */
632 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
633 bool save_argument_registers() const { return _save_argument_registers; }
634
635
636 // Other fixed compilation parameters.
637 ciMethod* method() const { return _method; }
638 int entry_bci() const { return _entry_bci; }
639 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
640 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
641 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
642 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
643 InlineTree* ilt() const { return _ilt; }
644 address stub_function() const { return _stub_function; }
645 const char* stub_name() const { return _stub_name; }
646 address stub_entry_point() const { return _stub_entry_point; }
647
648 // Control of this compilation.
649 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
650 void set_fixed_slots(int n) { _fixed_slots = n; }
651 int major_progress() const { return _major_progress; }
652 void set_inlining_progress(bool z) { _inlining_progress = z; }
653 int inlining_progress() const { return _inlining_progress; }
654 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
655 int inlining_incrementally() const { return _inlining_incrementally; }
656 void set_do_cleanup(bool z) { _do_cleanup = z; }
657 int do_cleanup() const { return _do_cleanup; }
658 void set_major_progress() { _major_progress++; }
659 void restore_major_progress(int progress) { _major_progress += progress; }
660 void clear_major_progress() { _major_progress = 0; }
661 int max_inline_size() const { return _max_inline_size; }
662 void set_freq_inline_size(int n) { _freq_inline_size = n; }
663 int freq_inline_size() const { return _freq_inline_size; }
664 void set_max_inline_size(int n) { _max_inline_size = n; }
665 bool has_loops() const { return _has_loops; }
666 void set_has_loops(bool z) { _has_loops = z; }
667 bool has_split_ifs() const { return _has_split_ifs; }
668 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
669 bool has_unsafe_access() const { return _has_unsafe_access; }
670 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
671 bool has_stringbuilder() const { return _has_stringbuilder; }
672 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
673 bool has_boxed_value() const { return _has_boxed_value; }
674 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
675 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
676 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
677 uint max_vector_size() const { return _max_vector_size; }
678 void set_max_vector_size(uint s) { _max_vector_size = s; }
679 bool clear_upper_avx() const { return _clear_upper_avx; }
680 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
681 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
682 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
683 bool trap_can_recompile() const { return _trap_can_recompile; }
684 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
685 uint decompile_count() const { return _decompile_count; }
686 void set_decompile_count(uint c) { _decompile_count = c; }
687 bool allow_range_check_smearing() const;
688 bool do_inlining() const { return _do_inlining; }
689 void set_do_inlining(bool z) { _do_inlining = z; }
690 bool do_scheduling() const { return _do_scheduling; }
691 void set_do_scheduling(bool z) { _do_scheduling = z; }
692 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
693 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
694 bool do_count_invocations() const{ return _do_count_invocations; }
695 void set_do_count_invocations(bool z){ _do_count_invocations = z; }
696 bool do_method_data_update() const { return _do_method_data_update; }
697 void set_do_method_data_update(bool z) { _do_method_data_update = z; }
698 bool do_vector_loop() const { return _do_vector_loop; }
699 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
700 bool use_cmove() const { return _use_cmove; }
701 void set_use_cmove(bool z) { _use_cmove = z; }
702 bool age_code() const { return _age_code; }
703 void set_age_code(bool z) { _age_code = z; }
704 int AliasLevel() const { return _AliasLevel; }
705 bool print_assembly() const { return _print_assembly; }
706 void set_print_assembly(bool z) { _print_assembly = z; }
707 bool print_inlining() const { return _print_inlining; }
708 void set_print_inlining(bool z) { _print_inlining = z; }
709 bool print_intrinsics() const { return _print_intrinsics; }
710 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
711 RTMState rtm_state() const { return _rtm_state; }
712 void set_rtm_state(RTMState s) { _rtm_state = s; }
713 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
714 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
715 uint max_node_limit() const { return (uint)_max_node_limit; }
716 void set_max_node_limit(uint n) { _max_node_limit = n; }
717 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
718 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
719
720 // check the CompilerOracle for special behaviours for this compile
721 bool method_has_option(const char * option) {
722 return method() != NULL && method()->has_option(option);
723 }
724
725 #ifndef PRODUCT
726 bool trace_opto_output() const { return _trace_opto_output; }
727 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
728 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
729 int _in_dump_cnt; // Required for dumping ir nodes.
730 #endif
731 bool has_irreducible_loop() const { return _has_irreducible_loop; }
732 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
733
734 // JSR 292
735 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
736 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
737
738 Ticks _latest_stage_start_counter;
739
740 void begin_method() {
741 #ifndef PRODUCT
742 if (_printer && _printer->should_print(1)) {
743 _printer->begin_method();
744 }
745 #endif
746 C->_latest_stage_start_counter.stamp();
747 }
748
749 bool should_print(int level = 1) {
750 #ifndef PRODUCT
751 return (_printer && _printer->should_print(level));
752 #else
753 return false;
754 #endif
755 }
756
757 void print_method(CompilerPhaseType cpt, int level = 1, int idx = 0) {
758 EventCompilerPhase event;
759 if (event.should_commit()) {
760 event.set_starttime(C->_latest_stage_start_counter);
761 event.set_phase((u1) cpt);
762 event.set_compileId(C->_compile_id);
763 event.set_phaseLevel(level);
764 event.commit();
765 }
766
767 #ifndef PRODUCT
768 if (should_print(level)) {
769 char output[1024];
770 if (idx != 0) {
771 sprintf(output, "%s:%d", CompilerPhaseTypeHelper::to_string(cpt), idx);
772 } else {
773 sprintf(output, "%s", CompilerPhaseTypeHelper::to_string(cpt));
774 }
775 _printer->print_method(output, level);
776 }
777 #endif
778 C->_latest_stage_start_counter.stamp();
779 }
780
781 void end_method(int level = 1) {
782 EventCompilerPhase event;
783 if (event.should_commit()) {
784 event.set_starttime(C->_latest_stage_start_counter);
785 event.set_phase((u1) PHASE_END);
786 event.set_compileId(C->_compile_id);
787 event.set_phaseLevel(level);
788 event.commit();
789 }
790 #ifndef PRODUCT
791 if (_printer && _printer->should_print(level)) {
792 _printer->end_method();
793 }
794 #endif
795 }
796
797 int macro_count() const { return _macro_nodes->length(); }
798 int predicate_count() const { return _predicate_opaqs->length();}
799 int expensive_count() const { return _expensive_nodes->length(); }
800 Node* macro_node(int idx) const { return _macro_nodes->at(idx); }
801 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
802 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); }
803 ConnectionGraph* congraph() { return _congraph;}
804 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
805 void add_macro_node(Node * n) {
806 //assert(n->is_macro(), "must be a macro node");
807 assert(!_macro_nodes->contains(n), "duplicate entry in expand list");
808 _macro_nodes->append(n);
809 }
810 void remove_macro_node(Node * n) {
811 // this function may be called twice for a node so check
812 // that the node is in the array before attempting to remove it
813 if (_macro_nodes->contains(n))
814 _macro_nodes->remove(n);
815 // remove from _predicate_opaqs list also if it is there
816 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){
817 _predicate_opaqs->remove(n);
818 }
819 }
820 void add_expensive_node(Node * n);
821 void remove_expensive_node(Node * n) {
822 if (_expensive_nodes->contains(n)) {
823 _expensive_nodes->remove(n);
824 }
825 }
826 void add_predicate_opaq(Node * n) {
827 assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1");
828 assert(_macro_nodes->contains(n), "should have already been in macro list");
829 _predicate_opaqs->append(n);
830 }
831
832 // Range check dependent CastII nodes that can be removed after loop optimizations
833 void add_range_check_cast(Node* n);
834 void remove_range_check_cast(Node* n) {
835 if (_range_check_casts->contains(n)) {
836 _range_check_casts->remove(n);
837 }
838 }
839 Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx); }
840 int range_check_cast_count() const { return _range_check_casts->length(); }
841 // Remove all range check dependent CastIINodes.
842 void remove_range_check_casts(PhaseIterGVN &igvn);
843
844 void add_opaque4_node(Node* n);
845 void remove_opaque4_node(Node* n) {
846 if (_opaque4_nodes->contains(n)) {
847 _opaque4_nodes->remove(n);
848 }
849 }
850 Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx); }
851 int opaque4_count() const { return _opaque4_nodes->length(); }
852 void remove_opaque4_nodes(PhaseIterGVN &igvn);
853
854 // remove the opaque nodes that protect the predicates so that the unused checks and
855 // uncommon traps will be eliminated from the graph.
856 void cleanup_loop_predicates(PhaseIterGVN &igvn);
857 bool is_predicate_opaq(Node * n) {
858 return _predicate_opaqs->contains(n);
859 }
860
861 // Are there candidate expensive nodes for optimization?
862 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
863 // Check whether n1 and n2 are similar
864 static int cmp_expensive_nodes(Node* n1, Node* n2);
865 // Sort expensive nodes to locate similar expensive nodes
866 void sort_expensive_nodes();
867
868 // Compilation environment.
869 Arena* comp_arena() { return &_comp_arena; }
870 ciEnv* env() const { return _env; }
871 CompileLog* log() const { return _log; }
872 bool failing() const { return _env->failing() || _failure_reason != NULL; }
873 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
874
875 bool failure_reason_is(const char* r) const {
876 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
877 }
878
879 void record_failure(const char* reason);
880 void record_method_not_compilable(const char* reason) {
881 // Bailouts cover "all_tiers" when TieredCompilation is off.
882 env()->record_method_not_compilable(reason, !TieredCompilation);
883 // Record failure reason.
884 record_failure(reason);
885 }
886 bool check_node_count(uint margin, const char* reason) {
887 if (live_nodes() + margin > max_node_limit()) {
888 record_method_not_compilable(reason);
889 return true;
890 } else {
891 return false;
892 }
893 }
894
895 // Node management
896 uint unique() const { return _unique; }
897 uint next_unique() { return _unique++; }
898 void set_unique(uint i) { _unique = i; }
899 static int debug_idx() { return debug_only(_debug_idx)+0; }
900 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
901 Arena* node_arena() { return &_node_arena; }
902 Arena* old_arena() { return &_old_arena; }
903 RootNode* root() const { return _root; }
904 void set_root(RootNode* r) { _root = r; }
905 StartNode* start() const; // (Derived from root.)
906 void init_start(StartNode* s);
907 Node* immutable_memory();
908
909 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
910 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
911 void set_recent_alloc(Node* ctl, Node* obj) {
912 _recent_alloc_ctl = ctl;
913 _recent_alloc_obj = obj;
914 }
915 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
916 _dead_node_count++;
917 }
918 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; }
919 uint dead_node_count() { return _dead_node_count; }
920 void reset_dead_node_list() { _dead_node_list.Reset();
921 _dead_node_count = 0;
922 }
923 uint live_nodes() const {
924 int val = _unique - _dead_node_count;
925 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
926 return (uint) val;
927 }
928 #ifdef ASSERT
929 uint count_live_nodes_by_graph_walk();
930 void print_missing_nodes();
931 #endif
932
933 // Record modified nodes to check that they are put on IGVN worklist
934 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
935 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
936 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
937
938 // Constant table
939 ConstantTable& constant_table() { return _constant_table; }
940
941 MachConstantBaseNode* mach_constant_base_node();
942 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
943 // Generated by adlc, true if CallNode requires MachConstantBase.
944 bool needs_clone_jvms();
945
946 // Handy undefined Node
947 Node* top() const { return _top; }
948
949 // these are used by guys who need to know about creation and transformation of top:
950 Node* cached_top_node() { return _top; }
951 void set_cached_top_node(Node* tn);
952
953 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
954 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
955 Node_Notes* default_node_notes() const { return _default_node_notes; }
956 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
957
958 Node_Notes* node_notes_at(int idx) {
959 return locate_node_notes(_node_note_array, idx, false);
960 }
961 inline bool set_node_notes_at(int idx, Node_Notes* value);
962
963 // Copy notes from source to dest, if they exist.
964 // Overwrite dest only if source provides something.
965 // Return true if information was moved.
966 bool copy_node_notes_to(Node* dest, Node* source);
967
968 // Workhorse function to sort out the blocked Node_Notes array:
969 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
970 int idx, bool can_grow = false);
971
972 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
973
974 // Type management
975 Arena* type_arena() { return _type_arena; }
976 Dict* type_dict() { return _type_dict; }
977 size_t type_last_size() { return _type_last_size; }
978 int num_alias_types() { return _num_alias_types; }
979
980 void init_type_arena() { _type_arena = &_Compile_types; }
981 void set_type_arena(Arena* a) { _type_arena = a; }
982 void set_type_dict(Dict* d) { _type_dict = d; }
983 void set_type_last_size(size_t sz) { _type_last_size = sz; }
984
985 const TypeFunc* last_tf(ciMethod* m) {
986 return (m == _last_tf_m) ? _last_tf : NULL;
987 }
988 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
989 assert(m != NULL || tf == NULL, "");
990 _last_tf_m = m;
991 _last_tf = tf;
992 }
993
994 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
995 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); }
996 bool have_alias_type(const TypePtr* adr_type);
997 AliasType* alias_type(ciField* field);
998
999 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
1000 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
1001 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
1002
1003 // Building nodes
1004 void rethrow_exceptions(JVMState* jvms);
1005 void return_values(JVMState* jvms);
1006 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
1007
1008 // Decide how to build a call.
1009 // The profile factor is a discount to apply to this site's interp. profile.
1010 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1011 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
1012 bool allow_intrinsics = true, bool delayed_forbidden = false);
1013 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1014 return should_delay_string_inlining(call_method, jvms) ||
1015 should_delay_boxing_inlining(call_method, jvms);
1016 }
1017 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1018 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1019
1020 // Helper functions to identify inlining potential at call-site
1021 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
1022 ciKlass* holder, ciMethod* callee,
1023 const TypeOopPtr* receiver_type, bool is_virtual,
1024 bool &call_does_dispatch, int &vtable_index,
1025 bool check_access = true);
1026 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
1027 ciMethod* callee, const TypeOopPtr* receiver_type,
1028 bool check_access = true);
1029
1030 // Report if there were too many traps at a current method and bci.
1031 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1032 // If there is no MDO at all, report no trap unless told to assume it.
1033 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1034 // This version, unspecific to a particular bci, asks if
1035 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1036 bool too_many_traps(Deoptimization::DeoptReason reason,
1037 // Privately used parameter for logging:
1038 ciMethodData* logmd = NULL);
1039 // Report if there were too many recompiles at a method and bci.
1040 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1041 // Report if there were too many traps or recompiles at a method and bci.
1042 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1043 return too_many_traps(method, bci, reason) ||
1044 too_many_recompiles(method, bci, reason);
1045 }
1046 // Return a bitset with the reasons where deoptimization is allowed,
1047 // i.e., where there were not too many uncommon traps.
1048 int _allowed_reasons;
1049 int allowed_deopt_reasons() { return _allowed_reasons; }
1050 void set_allowed_deopt_reasons();
1051
1052 // Parsing, optimization
1053 PhaseGVN* initial_gvn() { return _initial_gvn; }
1054 Unique_Node_List* for_igvn() { return _for_igvn; }
1055 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
1056 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1057 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
1058
1059 // Replace n by nn using initial_gvn, calling hash_delete and
1060 // record_for_igvn as needed.
1061 void gvn_replace_by(Node* n, Node* nn);
1062
1063
1064 void identify_useful_nodes(Unique_Node_List &useful);
1065 void update_dead_node_list(Unique_Node_List &useful);
1066 void remove_useless_nodes (Unique_Node_List &useful);
1067
1068 WarmCallInfo* warm_calls() const { return _warm_calls; }
1069 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
1070 WarmCallInfo* pop_warm_call();
1071
1072 // Record this CallGenerator for inlining at the end of parsing.
1073 void add_late_inline(CallGenerator* cg) {
1074 _late_inlines.insert_before(_late_inlines_pos, cg);
1075 _late_inlines_pos++;
1076 }
1077
1078 void prepend_late_inline(CallGenerator* cg) {
1079 _late_inlines.insert_before(0, cg);
1080 }
1081
1082 void add_string_late_inline(CallGenerator* cg) {
1083 _string_late_inlines.push(cg);
1084 }
1085
1086 void add_boxing_late_inline(CallGenerator* cg) {
1087 _boxing_late_inlines.push(cg);
1088 }
1089
1090 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1091
1092 void process_print_inlining();
1093 void dump_print_inlining();
1094
1095 bool over_inlining_cutoff() const {
1096 if (!inlining_incrementally()) {
1097 return unique() > (uint)NodeCountInliningCutoff;
1098 } else {
1099 // Give some room for incremental inlining algorithm to "breathe"
1100 // and avoid thrashing when live node count is close to the limit.
1101 // Keep in mind that live_nodes() isn't accurate during inlining until
1102 // dead node elimination step happens (see Compile::inline_incrementally).
1103 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1104 }
1105 }
1106
1107 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1108 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--; }
1109 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1110
1111 bool inline_incrementally_one();
1112 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1113 void inline_incrementally(PhaseIterGVN& igvn);
1114 void inline_string_calls(bool parse_time);
1115 void inline_boxing_calls(PhaseIterGVN& igvn);
1116 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1117 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1118
1119 // Matching, CFG layout, allocation, code generation
1120 PhaseCFG* cfg() { return _cfg; }
1121 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1122 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1123 bool has_java_calls() const { return _java_calls > 0; }
1124 int java_calls() const { return _java_calls; }
1125 int inner_loops() const { return _inner_loops; }
1126 Matcher* matcher() { return _matcher; }
1127 PhaseRegAlloc* regalloc() { return _regalloc; }
1128 int frame_slots() const { return _frame_slots; }
1129 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words'
1130 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; }
1131 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1132 Arena* indexSet_arena() { return _indexSet_arena; }
1133 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1134 uint node_bundling_limit() { return _node_bundling_limit; }
1135 Bundle* node_bundling_base() { return _node_bundling_base; }
1136 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; }
1137 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; }
1138 bool starts_bundle(const Node *n) const;
1139 bool need_stack_bang(int frame_size_in_bytes) const;
1140 bool need_register_stack_bang() const;
1141
1142 void update_interpreter_frame_size(int size) {
1143 if (_interpreter_frame_size < size) {
1144 _interpreter_frame_size = size;
1145 }
1146 }
1147 int bang_size_in_bytes() const;
1148
1149 void set_matcher(Matcher* m) { _matcher = m; }
1150 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1151 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1152 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1153
1154 // Remember if this compilation changes hardware mode to 24-bit precision
1155 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1156 _select_24_bit_instr = selection;
1157 _in_24_bit_fp_mode = mode;
1158 }
1159
1160 void set_java_calls(int z) { _java_calls = z; }
1161 void set_inner_loops(int z) { _inner_loops = z; }
1162
1163 // Instruction bits passed off to the VM
1164 int code_size() { return _method_size; }
1165 CodeBuffer* code_buffer() { return &_code_buffer; }
1166 int first_block_size() { return _first_block_size; }
1167 void set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } }
1168 ExceptionHandlerTable* handler_table() { return &_handler_table; }
1169 ImplicitExceptionTable* inc_table() { return &_inc_table; }
1170 OopMapSet* oop_map_set() { return _oop_map_set; }
1171 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1172 Dependencies* dependencies() { return env()->dependencies(); }
1173 static int CompiledZap_count() { return _CompiledZap_count; }
1174 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; }
1175 void init_scratch_buffer_blob(int const_size);
1176 void clear_scratch_buffer_blob();
1177 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; }
1178 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; }
1179 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; }
1180
1181 // emit to scratch blob, report resulting size
1182 uint scratch_emit_size(const Node* n);
1183 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; }
1184 bool in_scratch_emit_size() const { return _in_scratch_emit_size; }
1185
1186 enum ScratchBufferBlob {
1187 MAX_inst_size = 2048,
1188 MAX_locs_size = 128, // number of relocInfo elements
1189 MAX_const_size = 128,
1190 MAX_stubs_size = 128
1191 };
1192
1193 // Major entry point. Given a Scope, compile the associated method.
1194 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1195 // replacement, entry_bci indicates the bytecode for which to compile a
1196 // continuation.
1197 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
1198 int entry_bci, bool subsume_loads, bool do_escape_analysis,
1199 bool eliminate_boxing, DirectiveSet* directive);
1200
1201 // Second major entry point. From the TypeFunc signature, generate code
1202 // to pass arguments from the Java calling convention to the C calling
1203 // convention.
1204 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1205 address stub_function, const char *stub_name,
1206 int is_fancy_jump, bool pass_tls,
1207 bool save_arg_registers, bool return_pc, DirectiveSet* directive);
1208
1209 // From the TypeFunc signature, generate code to pass arguments
1210 // from Compiled calling convention to Interpreter's calling convention
1211 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry);
1212
1213 // From the TypeFunc signature, generate code to pass arguments
1214 // from Interpreter's calling convention to Compiler's calling convention
1215 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf);
1216
1217 // Are we compiling a method?
1218 bool has_method() { return method() != NULL; }
1219
1220 // Maybe print some information about this compile.
1221 void print_compile_messages();
1222
1223 // Final graph reshaping, a post-pass after the regular optimizer is done.
1224 bool final_graph_reshaping();
1225
1226 // returns true if adr is completely contained in the given alias category
1227 bool must_alias(const TypePtr* adr, int alias_idx);
1228
1229 // returns true if adr overlaps with the given alias category
1230 bool can_alias(const TypePtr* adr, int alias_idx);
1231
1232 // Driver for converting compiler's IR into machine code bits
1233 void Output();
1234
1235 // Accessors for node bundling info.
1236 Bundle* node_bundling(const Node *n);
1237 bool valid_bundle_info(const Node *n);
1238
1239 // Schedule and Bundle the instructions
1240 void ScheduleAndBundle();
1241
1242 // Build OopMaps for each GC point
1243 void BuildOopMaps();
1244
1245 // Append debug info for the node "local" at safepoint node "sfpt" to the
1246 // "array", May also consult and add to "objs", which describes the
1247 // scalar-replaced objects.
1248 void FillLocArray( int idx, MachSafePointNode* sfpt,
1249 Node *local, GrowableArray<ScopeValue*> *array,
1250 GrowableArray<ScopeValue*> *objs );
1251
1252 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
1253 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
1254 // Requres that "objs" does not contains an ObjectValue whose id matches
1255 // that of "sv. Appends "sv".
1256 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
1257 ObjectValue* sv );
1258
1259 // Process an OopMap Element while emitting nodes
1260 void Process_OopMap_Node(MachNode *mach, int code_offset);
1261
1262 class BufferSizingData {
1263 public:
1264 int _stub;
1265 int _code;
1266 int _const;
1267 int _reloc;
1268
1269 BufferSizingData() :
1270 _stub(0),
1271 _code(0),
1272 _const(0),
1273 _reloc(0)
1274 { };
1275 };
1276
1277 // Initialize code buffer
1278 void init_buffer1(int& const_req);
1279 CodeBuffer* init_buffer2(BufferSizingData& buf_sizes);
1280
1281 // Write out basic block data to code buffer
1282 void fill_buffer(CodeBuffer* cb, uint* blk_starts);
1283
1284 // Determine which variable sized branches can be shortened
1285 void shorten_branches(uint* blk_starts, BufferSizingData& buf_sizes);
1286
1287 // Compute the size of first NumberOfLoopInstrToAlign instructions
1288 // at the head of a loop.
1289 void compute_loop_first_inst_sizes();
1290
1291 // Compute the information for the exception tables
1292 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels);
1293
1294 // Stack slots that may be unused by the calling convention but must
1295 // otherwise be preserved. On Intel this includes the return address.
1296 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1297 uint in_preserve_stack_slots();
1298
1299 // "Top of Stack" slots that may be unused by the calling convention but must
1300 // otherwise be preserved.
1301 // On Intel these are not necessary and the value can be zero.
1302 // On Sparc this describes the words reserved for storing a register window
1303 // when an interrupt occurs.
1304 static uint out_preserve_stack_slots();
1305
1306 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1307 // for calls to C. Supports the var-args backing area for register parms.
1308 uint varargs_C_out_slots_killed() const;
1309
1310 // Number of Stack Slots consumed by a synchronization entry
1311 int sync_stack_slots() const;
1312
1313 // Compute the name of old_SP. See <arch>.ad for frame layout.
1314 OptoReg::Name compute_old_SP();
1315
1316 private:
1317 // Phase control:
1318 void Init(int aliaslevel); // Prepare for a single compilation
1319 int Inline_Warm(); // Find more inlining work.
1320 void Finish_Warm(); // Give up on further inlines.
1321 void Optimize(); // Given a graph, optimize it
1322 void Code_Gen(); // Generate code from a graph
1323
1324 // Management of the AliasType table.
1325 void grow_alias_types();
1326 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1327 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1328 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1329
1330 void verify_top(Node*) const PRODUCT_RETURN;
1331
1332 // Intrinsic setup.
1333 void register_library_intrinsics(); // initializer
1334 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1335 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1336 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1337 void register_intrinsic(CallGenerator* cg); // update fn
1338
1339 #ifndef PRODUCT
1340 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT];
1341 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT];
1342 #endif
1343 // Function calls made by the public function final_graph_reshaping.
1344 // No need to be made public as they are not called elsewhere.
1345 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1346 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop);
1347 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1348 void eliminate_redundant_card_marks(Node* n);
1349
1350 public:
1351
1352 // Note: Histogram array size is about 1 Kb.
1353 enum { // flag bits:
1354 _intrinsic_worked = 1, // succeeded at least once
1355 _intrinsic_failed = 2, // tried it but it failed
1356 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1357 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1358 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1359 };
1360 // Update histogram. Return boolean if this is a first-time occurrence.
1361 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1362 bool is_virtual, int flags) PRODUCT_RETURN0;
1363 static void print_intrinsic_statistics() PRODUCT_RETURN;
1364
1365 // Graph verification code
1366 // Walk the node list, verifying that there is a one-to-one
1367 // correspondence between Use-Def edges and Def-Use edges
1368 // The option no_dead_code enables stronger checks that the
1369 // graph is strongly connected from root in both directions.
1370 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1371
1372 // End-of-run dumps.
1373 static void print_statistics() PRODUCT_RETURN;
1374
1375 // Dump formatted assembly
1376 #if defined(SUPPORT_OPTO_ASSEMBLY)
1377 void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit);
1378 void dump_asm(int* pcs = NULL, uint pc_limit = 0) { dump_asm_on(tty, pcs, pc_limit); }
1379 #else
1380 void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit) { return; }
1381 void dump_asm(int* pcs = NULL, uint pc_limit = 0) { return; }
1382 #endif
1383 void dump_pc(int *pcs, int pc_limit, Node *n);
1384
1385 // Verify ADLC assumptions during startup
1386 static void adlc_verification() PRODUCT_RETURN;
1387
1388 // Definitions of pd methods
1389 static void pd_compiler2_init();
1390
1391 // Static parse-time type checking logic for gen_subtype_check:
1392 enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1393 int static_subtype_check(ciKlass* superk, ciKlass* subk);
1394
1395 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1396 // Optional control dependency (for example, on range check)
1397 Node* ctrl = NULL);
1398
1399 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1400 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl);
1401
1402 // Auxiliary method for randomized fuzzing/stressing
1403 static bool randomized_select(int count);
1404
1405 // supporting clone_map
1406 CloneMap& clone_map();
1407 void set_clone_map(Dict* d);
1408
1409 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1410 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1411 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1412 };
1413
1414 #endif // SHARE_OPTO_COMPILE_HPP