1 /*
2 * Copyright (c) 1997, 2018, 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_VM_OPTO_COMPILE_HPP
26 #define SHARE_VM_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 InlineTree;
56 class Int_Array;
57 class LoadBarrierNode;
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 typedef unsigned int node_idx_t;
94 class NodeCloneInfo {
95 private:
96 uint64_t _idx_clone_orig;
97 public:
98
99 void set_idx(node_idx_t idx) {
100 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
101 }
102 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
103
104 void set_gen(int generation) {
105 uint64_t g = (uint64_t)generation << 32;
106 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
107 }
108 int gen() const { return (int)(_idx_clone_orig >> 32); }
109
110 void set(uint64_t x) { _idx_clone_orig = x; }
111 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
112 uint64_t get() const { return _idx_clone_orig; }
113
114 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
115 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
116
117 void dump() const;
118 };
119
120 class CloneMap {
121 friend class Compile;
122 private:
123 bool _debug;
124 Dict* _dict;
125 int _clone_idx; // current cloning iteration/generation in loop unroll
126 public:
127 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
128 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
129 Dict* dict() const { return _dict; }
130 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
131 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
132 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
133 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
134 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
135 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
136 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
137 int max_gen() const;
138 void clone(Node* old, Node* nnn, int gen);
139 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
140 void dump(node_idx_t key) const;
141
142 int clone_idx() const { return _clone_idx; }
143 void set_clone_idx(int x) { _clone_idx = x; }
144 bool is_debug() const { return _debug; }
145 void set_debug(bool debug) { _debug = debug; }
146 static const char* debug_option_name;
147
148 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
149 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
150 };
151
152 //------------------------------Compile----------------------------------------
153 // This class defines a top-level Compiler invocation.
154
155 class Compile : public Phase {
156 friend class VMStructs;
157
158 public:
159 // Fixed alias indexes. (See also MergeMemNode.)
160 enum {
161 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
162 AliasIdxBot = 2, // pseudo-index, aliases to everything
163 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
164 };
165
166 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
167 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
168 // then brackets are put into the log, with time stamps and node counts.
169 // (The time collection itself is always conditionalized on CITime.)
170 class TracePhase : public TraceTime {
171 private:
172 Compile* C;
173 CompileLog* _log;
174 const char* _phase_name;
175 bool _dolog;
176 public:
177 TracePhase(const char* name, elapsedTimer* accumulator);
178 ~TracePhase();
179 };
180
181 // Information per category of alias (memory slice)
182 class AliasType {
183 private:
184 friend class Compile;
185
186 int _index; // unique index, used with MergeMemNode
187 const TypePtr* _adr_type; // normalized address type
188 ciField* _field; // relevant instance field, or null if none
189 const Type* _element; // relevant array element type, or null if none
190 bool _is_rewritable; // false if the memory is write-once only
191 int _general_index; // if this is type is an instance, the general
192 // type that this is an instance of
193
194 void Init(int i, const TypePtr* at);
195
196 public:
197 int index() const { return _index; }
198 const TypePtr* adr_type() const { return _adr_type; }
199 ciField* field() const { return _field; }
200 const Type* element() const { return _element; }
201 bool is_rewritable() const { return _is_rewritable; }
202 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
203 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
204
205 void set_rewritable(bool z) { _is_rewritable = z; }
206 void set_field(ciField* f) {
207 assert(!_field,"");
208 _field = f;
209 if (f->is_final() || f->is_stable()) {
210 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
211 _is_rewritable = false;
212 }
213 }
214 void set_element(const Type* e) {
215 assert(_element == NULL, "");
216 _element = e;
217 }
218
219 BasicType basic_type() const;
220
221 void print_on(outputStream* st) PRODUCT_RETURN;
222 };
223
224 enum {
225 logAliasCacheSize = 6,
226 AliasCacheSize = (1<<logAliasCacheSize)
227 };
228 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
229 enum {
230 trapHistLength = MethodData::_trap_hist_limit
231 };
232
233 // Constant entry of the constant table.
234 class Constant {
235 private:
236 BasicType _type;
237 union {
238 jvalue _value;
239 Metadata* _metadata;
240 } _v;
241 int _offset; // offset of this constant (in bytes) relative to the constant table base.
242 float _freq;
243 bool _can_be_reused; // true (default) if the value can be shared with other users.
244
245 public:
246 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; }
247 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) :
248 _type(type),
249 _offset(-1),
250 _freq(freq),
251 _can_be_reused(can_be_reused)
252 {
253 assert(type != T_METADATA, "wrong constructor");
254 _v._value = value;
255 }
256 Constant(Metadata* metadata, bool can_be_reused = true) :
257 _type(T_METADATA),
258 _offset(-1),
259 _freq(0.0f),
260 _can_be_reused(can_be_reused)
261 {
262 _v._metadata = metadata;
263 }
264
265 bool operator==(const Constant& other);
266
267 BasicType type() const { return _type; }
268
269 jint get_jint() const { return _v._value.i; }
270 jlong get_jlong() const { return _v._value.j; }
271 jfloat get_jfloat() const { return _v._value.f; }
272 jdouble get_jdouble() const { return _v._value.d; }
273 jobject get_jobject() const { return _v._value.l; }
274
275 Metadata* get_metadata() const { return _v._metadata; }
276
277 int offset() const { return _offset; }
278 void set_offset(int offset) { _offset = offset; }
279
280 float freq() const { return _freq; }
281 void inc_freq(float freq) { _freq += freq; }
282
283 bool can_be_reused() const { return _can_be_reused; }
284 };
285
286 // Constant table.
287 class ConstantTable {
288 private:
289 GrowableArray<Constant> _constants; // Constants of this table.
290 int _size; // Size in bytes the emitted constant table takes (including padding).
291 int _table_base_offset; // Offset of the table base that gets added to the constant offsets.
292 int _nof_jump_tables; // Number of jump-tables in this constant table.
293
294 static int qsort_comparator(Constant* a, Constant* b);
295
296 // We use negative frequencies to keep the order of the
297 // jump-tables in which they were added. Otherwise we get into
298 // trouble with relocation.
299 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); }
300
301 public:
302 ConstantTable() :
303 _size(-1),
304 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit).
305 _nof_jump_tables(0)
306 {}
307
308 int size() const { assert(_size != -1, "not calculated yet"); return _size; }
309
310 int calculate_table_base_offset() const; // AD specific
311 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; }
312 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; }
313
314 void emit(CodeBuffer& cb);
315
316 // Returns the offset of the last entry (the top) of the constant table.
317 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); }
318
319 void calculate_offsets_and_size();
320 int find_offset(Constant& con) const;
321
322 void add(Constant& con);
323 Constant add(MachConstantNode* n, BasicType type, jvalue value);
324 Constant add(Metadata* metadata);
325 Constant add(MachConstantNode* n, MachOper* oper);
326 Constant add(MachConstantNode* n, jint i) {
327 jvalue value; value.i = i;
328 return add(n, T_INT, value);
329 }
330 Constant add(MachConstantNode* n, jlong j) {
331 jvalue value; value.j = j;
332 return add(n, T_LONG, value);
333 }
334 Constant add(MachConstantNode* n, jfloat f) {
335 jvalue value; value.f = f;
336 return add(n, T_FLOAT, value);
337 }
338 Constant add(MachConstantNode* n, jdouble d) {
339 jvalue value; value.d = d;
340 return add(n, T_DOUBLE, value);
341 }
342
343 // Jump-table
344 Constant add_jump_table(MachConstantNode* n);
345 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const;
346 };
347
348 private:
349 // Fixed parameters to this compilation.
350 const int _compile_id;
351 const bool _save_argument_registers; // save/restore arg regs for trampolines
352 const bool _subsume_loads; // Load can be matched as part of a larger op.
353 const bool _do_escape_analysis; // Do escape analysis.
354 const bool _eliminate_boxing; // Do boxing elimination.
355 ciMethod* _method; // The method being compiled.
356 int _entry_bci; // entry bci for osr methods.
357 const TypeFunc* _tf; // My kind of signature
358 InlineTree* _ilt; // Ditto (temporary).
359 address _stub_function; // VM entry for stub being compiled, or NULL
360 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
361 address _stub_entry_point; // Compile code entry for generated stub, or NULL
362
363 // Control of this compilation.
364 int _num_loop_opts; // Number of iterations for doing loop optimiztions
365 int _max_inline_size; // Max inline size for this compilation
366 int _freq_inline_size; // Max hot method inline size for this compilation
367 int _fixed_slots; // count of frame slots not allocated by the register
368 // allocator i.e. locks, original deopt pc, etc.
369 uintx _max_node_limit; // Max unique node count during a single compilation.
370 // For deopt
371 int _orig_pc_slot;
372 int _orig_pc_slot_offset_in_bytes;
373
374 int _major_progress; // Count of something big happening
375 bool _inlining_progress; // progress doing incremental inlining?
376 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
377 bool _has_loops; // True if the method _may_ have some loops
378 bool _has_split_ifs; // True if the method _may_ have some split-if
379 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
380 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
381 bool _has_boxed_value; // True if a boxed object is allocated
382 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
383 uint _max_vector_size; // Maximum size of generated vectors
384 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
385 uint _trap_hist[trapHistLength]; // Cumulative traps
386 bool _trap_can_recompile; // Have we emitted a recompiling trap?
387 uint _decompile_count; // Cumulative decompilation counts.
388 bool _do_inlining; // True if we intend to do inlining
389 bool _do_scheduling; // True if we intend to do scheduling
390 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
391 bool _do_count_invocations; // True if we generate code to count invocations
392 bool _do_method_data_update; // True if we generate code to update MethodData*s
393 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
394 bool _use_cmove; // True if CMove should be used without profitability analysis
395 bool _age_code; // True if we need to profile code age (decrement the aging counter)
396 int _AliasLevel; // Locally-adjusted version of AliasLevel flag.
397 bool _print_assembly; // True if we should dump assembly code for this compilation
398 bool _print_inlining; // True if we should print inlining for this compilation
399 bool _print_intrinsics; // True if we should print intrinsics for this compilation
400 #ifndef PRODUCT
401 bool _trace_opto_output;
402 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
403 #endif
404 bool _has_irreducible_loop; // Found irreducible loops
405 // JSR 292
406 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
407 RTMState _rtm_state; // State of Restricted Transactional Memory usage
408
409 // Compilation environment.
410 Arena _comp_arena; // Arena with lifetime equivalent to Compile
411 void* _barrier_set_state; // Potential GC barrier state for Compile
412 ciEnv* _env; // CI interface
413 DirectiveSet* _directive; // Compiler directive
414 CompileLog* _log; // from CompilerThread
415 const char* _failure_reason; // for record_failure/failing pattern
416 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics.
417 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching.
418 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
419 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
420 GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency
421 GrowableArray<Node*>* _opaque4_nodes; // List of Opaque4 nodes that have a default value
422 ConnectionGraph* _congraph;
423 #ifndef PRODUCT
424 IdealGraphPrinter* _printer;
425 #endif
426
427
428 // Node management
429 uint _unique; // Counter for unique Node indices
430 VectorSet _dead_node_list; // Set of dead nodes
431 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
432 // So use this to keep count and make the call O(1).
433 DEBUG_ONLY( Unique_Node_List* _modified_nodes; ) // List of nodes which inputs were modified
434
435 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
436 Arena _node_arena; // Arena for new-space Nodes
437 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
438 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
439 Node* _top; // Unique top node. (Reset by various phases.)
440
441 Node* _immutable_memory; // Initial memory state
442
443 Node* _recent_alloc_obj;
444 Node* _recent_alloc_ctl;
445
446 // Constant table
447 ConstantTable _constant_table; // The constant table for this compile.
448 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
449
450
451 // Blocked array of debugging and profiling information,
452 // tracked per node.
453 enum { _log2_node_notes_block_size = 8,
454 _node_notes_block_size = (1<<_log2_node_notes_block_size)
455 };
456 GrowableArray<Node_Notes*>* _node_note_array;
457 Node_Notes* _default_node_notes; // default notes for new nodes
458
459 // After parsing and every bulk phase we hang onto the Root instruction.
460 // The RootNode instruction is where the whole program begins. It produces
461 // the initial Control and BOTTOM for everybody else.
462
463 // Type management
464 Arena _Compile_types; // Arena for all types
465 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
466 Dict* _type_dict; // Intern table
467 CloneMap _clone_map; // used for recording history of cloned nodes
468 void* _type_hwm; // Last allocation (see Type::operator new/delete)
469 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
470 ciMethod* _last_tf_m; // Cache for
471 const TypeFunc* _last_tf; // TypeFunc::make
472 AliasType** _alias_types; // List of alias types seen so far.
473 int _num_alias_types; // Logical length of _alias_types
474 int _max_alias_types; // Physical length of _alias_types
475 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
476
477 // Parsing, optimization
478 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
479 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
480 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining.
481
482 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after
483 // main parsing has finished.
484 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
485
486 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
487
488 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
489 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
490
491
492 // Inlining may not happen in parse order which would make
493 // PrintInlining output confusing. Keep track of PrintInlining
494 // pieces in order.
495 class PrintInliningBuffer : public ResourceObj {
496 private:
497 CallGenerator* _cg;
498 stringStream* _ss;
499
500 public:
501 PrintInliningBuffer()
502 : _cg(NULL) { _ss = new stringStream(); }
503
504 stringStream* ss() const { return _ss; }
505 CallGenerator* cg() const { return _cg; }
506 void set_cg(CallGenerator* cg) { _cg = cg; }
507 };
508
509 stringStream* _print_inlining_stream;
510 GrowableArray<PrintInliningBuffer>* _print_inlining_list;
511 int _print_inlining_idx;
512 char* _print_inlining_output;
513
514 // Only keep nodes in the expensive node list that need to be optimized
515 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
516 // Use for sorting expensive nodes to bring similar nodes together
517 static int cmp_expensive_nodes(Node** n1, Node** n2);
518 // Expensive nodes list already sorted?
519 bool expensive_nodes_sorted() const;
520 // Remove the speculative part of types and clean up the graph
521 void remove_speculative_types(PhaseIterGVN &igvn);
522
523 void* _replay_inline_data; // Pointer to data loaded from file
524
525 void print_inlining_init();
526 void print_inlining_reinit();
527 void print_inlining_commit();
528 void print_inlining_push();
529 PrintInliningBuffer& print_inlining_current();
530
531 void log_late_inline_failure(CallGenerator* cg, const char* msg);
532
533 public:
534
535 void* barrier_set_state() const { return _barrier_set_state; }
536
537 outputStream* print_inlining_stream() const {
538 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
539 return _print_inlining_stream;
540 }
541
542 void print_inlining_update(CallGenerator* cg);
543 void print_inlining_update_delayed(CallGenerator* cg);
544 void print_inlining_move_to(CallGenerator* cg);
545 void print_inlining_assert_ready();
546 void print_inlining_reset();
547
548 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
549 stringStream ss;
550 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
551 print_inlining_stream()->print("%s", ss.as_string());
552 }
553
554 #ifndef PRODUCT
555 IdealGraphPrinter* printer() { return _printer; }
556 #endif
557
558 void log_late_inline(CallGenerator* cg);
559 void log_inline_id(CallGenerator* cg);
560 void log_inline_failure(const char* msg);
561
562 void* replay_inline_data() const { return _replay_inline_data; }
563
564 // Dump inlining replay data to the stream.
565 void dump_inline_data(outputStream* out);
566
567 private:
568 // Matching, CFG layout, allocation, code generation
569 PhaseCFG* _cfg; // Results of CFG finding
570 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
571 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
572 int _java_calls; // Number of java calls in the method
573 int _inner_loops; // Number of inner loops in the method
574 Matcher* _matcher; // Engine to map ideal to machine instructions
575 PhaseRegAlloc* _regalloc; // Results of register allocation.
576 int _frame_slots; // Size of total frame in stack slots
577 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries
578 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
579 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
580 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
581 int _interpreter_frame_size;
582
583 uint _node_bundling_limit;
584 Bundle* _node_bundling_base; // Information for instruction bundling
585
586 // Instruction bits passed off to the VM
587 int _method_size; // Size of nmethod code segment in bytes
588 CodeBuffer _code_buffer; // Where the code is assembled
589 int _first_block_size; // Size of unvalidated entry point code / OSR poison code
590 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers
591 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code
592 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location)
593 static int _CompiledZap_count; // counter compared against CompileZap[First/Last]
594 BufferBlob* _scratch_buffer_blob; // For temporary code buffers.
595 relocInfo* _scratch_locs_memory; // For temporary code buffers.
596 int _scratch_const_size; // For temporary code buffers.
597 bool _in_scratch_emit_size; // true when in scratch_emit_size.
598
599 void reshape_address(AddPNode* n);
600
601 public:
602 // Accessors
603
604 // The Compile instance currently active in this (compiler) thread.
605 static Compile* current() {
606 return (Compile*) ciEnv::current()->compiler_data();
607 }
608
609 // ID for this compilation. Useful for setting breakpoints in the debugger.
610 int compile_id() const { return _compile_id; }
611 DirectiveSet* directive() const { return _directive; }
612
613 // Does this compilation allow instructions to subsume loads? User
614 // instructions that subsume a load may result in an unschedulable
615 // instruction sequence.
616 bool subsume_loads() const { return _subsume_loads; }
617 /** Do escape analysis. */
618 bool do_escape_analysis() const { return _do_escape_analysis; }
619 /** Do boxing elimination. */
620 bool eliminate_boxing() const { return _eliminate_boxing; }
621 /** Do aggressive boxing elimination. */
622 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
623 bool save_argument_registers() const { return _save_argument_registers; }
624
625
626 // Other fixed compilation parameters.
627 ciMethod* method() const { return _method; }
628 int entry_bci() const { return _entry_bci; }
629 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
630 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
631 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
632 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
633 InlineTree* ilt() const { return _ilt; }
634 address stub_function() const { return _stub_function; }
635 const char* stub_name() const { return _stub_name; }
636 address stub_entry_point() const { return _stub_entry_point; }
637
638 // Control of this compilation.
639 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
640 void set_fixed_slots(int n) { _fixed_slots = n; }
641 int major_progress() const { return _major_progress; }
642 void set_inlining_progress(bool z) { _inlining_progress = z; }
643 int inlining_progress() const { return _inlining_progress; }
644 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
645 int inlining_incrementally() const { return _inlining_incrementally; }
646 void set_major_progress() { _major_progress++; }
647 void clear_major_progress() { _major_progress = 0; }
648 int num_loop_opts() const { return _num_loop_opts; }
649 void set_num_loop_opts(int n) { _num_loop_opts = n; }
650 int max_inline_size() const { return _max_inline_size; }
651 void set_freq_inline_size(int n) { _freq_inline_size = n; }
652 int freq_inline_size() const { return _freq_inline_size; }
653 void set_max_inline_size(int n) { _max_inline_size = n; }
654 bool has_loops() const { return _has_loops; }
655 void set_has_loops(bool z) { _has_loops = z; }
656 bool has_split_ifs() const { return _has_split_ifs; }
657 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
658 bool has_unsafe_access() const { return _has_unsafe_access; }
659 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
660 bool has_stringbuilder() const { return _has_stringbuilder; }
661 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
662 bool has_boxed_value() const { return _has_boxed_value; }
663 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
664 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
665 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
666 uint max_vector_size() const { return _max_vector_size; }
667 void set_max_vector_size(uint s) { _max_vector_size = s; }
668 bool clear_upper_avx() const { return _clear_upper_avx; }
669 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
670 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
671 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
672 bool trap_can_recompile() const { return _trap_can_recompile; }
673 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
674 uint decompile_count() const { return _decompile_count; }
675 void set_decompile_count(uint c) { _decompile_count = c; }
676 bool allow_range_check_smearing() const;
677 bool do_inlining() const { return _do_inlining; }
678 void set_do_inlining(bool z) { _do_inlining = z; }
679 bool do_scheduling() const { return _do_scheduling; }
680 void set_do_scheduling(bool z) { _do_scheduling = z; }
681 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
682 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
683 bool do_count_invocations() const{ return _do_count_invocations; }
684 void set_do_count_invocations(bool z){ _do_count_invocations = z; }
685 bool do_method_data_update() const { return _do_method_data_update; }
686 void set_do_method_data_update(bool z) { _do_method_data_update = z; }
687 bool do_vector_loop() const { return _do_vector_loop; }
688 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
689 bool use_cmove() const { return _use_cmove; }
690 void set_use_cmove(bool z) { _use_cmove = z; }
691 bool age_code() const { return _age_code; }
692 void set_age_code(bool z) { _age_code = z; }
693 int AliasLevel() const { return _AliasLevel; }
694 bool print_assembly() const { return _print_assembly; }
695 void set_print_assembly(bool z) { _print_assembly = z; }
696 bool print_inlining() const { return _print_inlining; }
697 void set_print_inlining(bool z) { _print_inlining = z; }
698 bool print_intrinsics() const { return _print_intrinsics; }
699 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
700 RTMState rtm_state() const { return _rtm_state; }
701 void set_rtm_state(RTMState s) { _rtm_state = s; }
702 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
703 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
704 uint max_node_limit() const { return (uint)_max_node_limit; }
705 void set_max_node_limit(uint n) { _max_node_limit = n; }
706
707 // check the CompilerOracle for special behaviours for this compile
708 bool method_has_option(const char * option) {
709 return method() != NULL && method()->has_option(option);
710 }
711
712 #ifndef PRODUCT
713 bool trace_opto_output() const { return _trace_opto_output; }
714 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
715 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
716 int _in_dump_cnt; // Required for dumping ir nodes.
717 #endif
718 bool has_irreducible_loop() const { return _has_irreducible_loop; }
719 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
720
721 // JSR 292
722 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
723 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
724
725 Ticks _latest_stage_start_counter;
726
727 void begin_method() {
728 #ifndef PRODUCT
729 if (_printer && _printer->should_print(1)) {
730 _printer->begin_method();
731 }
732 #endif
733 C->_latest_stage_start_counter.stamp();
734 }
735
736 void print_method(CompilerPhaseType cpt, int level = 1) {
737 EventCompilerPhase event;
738 if (event.should_commit()) {
739 event.set_starttime(C->_latest_stage_start_counter);
740 event.set_phase((u1) cpt);
741 event.set_compileId(C->_compile_id);
742 event.set_phaseLevel(level);
743 event.commit();
744 }
745
746
747 #ifndef PRODUCT
748 if (_printer && _printer->should_print(level)) {
749 _printer->print_method(CompilerPhaseTypeHelper::to_string(cpt), level);
750 }
751 #endif
752 C->_latest_stage_start_counter.stamp();
753 }
754
755 void end_method(int level = 1) {
756 EventCompilerPhase event;
757 if (event.should_commit()) {
758 event.set_starttime(C->_latest_stage_start_counter);
759 event.set_phase((u1) PHASE_END);
760 event.set_compileId(C->_compile_id);
761 event.set_phaseLevel(level);
762 event.commit();
763 }
764 #ifndef PRODUCT
765 if (_printer && _printer->should_print(level)) {
766 _printer->end_method();
767 }
768 #endif
769 }
770
771 int macro_count() const { return _macro_nodes->length(); }
772 int predicate_count() const { return _predicate_opaqs->length();}
773 int expensive_count() const { return _expensive_nodes->length(); }
774 Node* macro_node(int idx) const { return _macro_nodes->at(idx); }
775 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
776 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); }
777 ConnectionGraph* congraph() { return _congraph;}
778 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
779 void add_macro_node(Node * n) {
780 //assert(n->is_macro(), "must be a macro node");
781 assert(!_macro_nodes->contains(n), "duplicate entry in expand list");
782 _macro_nodes->append(n);
783 }
784 void remove_macro_node(Node * n) {
785 // this function may be called twice for a node so check
786 // that the node is in the array before attempting to remove it
787 if (_macro_nodes->contains(n))
788 _macro_nodes->remove(n);
789 // remove from _predicate_opaqs list also if it is there
790 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){
791 _predicate_opaqs->remove(n);
792 }
793 }
794 void add_expensive_node(Node * n);
795 void remove_expensive_node(Node * n) {
796 if (_expensive_nodes->contains(n)) {
797 _expensive_nodes->remove(n);
798 }
799 }
800 void add_predicate_opaq(Node * n) {
801 assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1");
802 assert(_macro_nodes->contains(n), "should have already been in macro list");
803 _predicate_opaqs->append(n);
804 }
805
806 // Range check dependent CastII nodes that can be removed after loop optimizations
807 void add_range_check_cast(Node* n);
808 void remove_range_check_cast(Node* n) {
809 if (_range_check_casts->contains(n)) {
810 _range_check_casts->remove(n);
811 }
812 }
813 Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx); }
814 int range_check_cast_count() const { return _range_check_casts->length(); }
815 // Remove all range check dependent CastIINodes.
816 void remove_range_check_casts(PhaseIterGVN &igvn);
817
818 void add_opaque4_node(Node* n);
819 void remove_opaque4_node(Node* n) {
820 if (_opaque4_nodes->contains(n)) {
821 _opaque4_nodes->remove(n);
822 }
823 }
824 Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx); }
825 int opaque4_count() const { return _opaque4_nodes->length(); }
826 void remove_opaque4_nodes(PhaseIterGVN &igvn);
827
828 // remove the opaque nodes that protect the predicates so that the unused checks and
829 // uncommon traps will be eliminated from the graph.
830 void cleanup_loop_predicates(PhaseIterGVN &igvn);
831 bool is_predicate_opaq(Node * n) {
832 return _predicate_opaqs->contains(n);
833 }
834
835 // Are there candidate expensive nodes for optimization?
836 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
837 // Check whether n1 and n2 are similar
838 static int cmp_expensive_nodes(Node* n1, Node* n2);
839 // Sort expensive nodes to locate similar expensive nodes
840 void sort_expensive_nodes();
841
842 // Compilation environment.
843 Arena* comp_arena() { return &_comp_arena; }
844 ciEnv* env() const { return _env; }
845 CompileLog* log() const { return _log; }
846 bool failing() const { return _env->failing() || _failure_reason != NULL; }
847 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
848
849 bool failure_reason_is(const char* r) const {
850 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
851 }
852
853 void record_failure(const char* reason);
854 void record_method_not_compilable(const char* reason) {
855 // Bailouts cover "all_tiers" when TieredCompilation is off.
856 env()->record_method_not_compilable(reason, !TieredCompilation);
857 // Record failure reason.
858 record_failure(reason);
859 }
860 bool check_node_count(uint margin, const char* reason) {
861 if (live_nodes() + margin > max_node_limit()) {
862 record_method_not_compilable(reason);
863 return true;
864 } else {
865 return false;
866 }
867 }
868
869 // Node management
870 uint unique() const { return _unique; }
871 uint next_unique() { return _unique++; }
872 void set_unique(uint i) { _unique = i; }
873 static int debug_idx() { return debug_only(_debug_idx)+0; }
874 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
875 Arena* node_arena() { return &_node_arena; }
876 Arena* old_arena() { return &_old_arena; }
877 RootNode* root() const { return _root; }
878 void set_root(RootNode* r) { _root = r; }
879 StartNode* start() const; // (Derived from root.)
880 void init_start(StartNode* s);
881 Node* immutable_memory();
882
883 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
884 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
885 void set_recent_alloc(Node* ctl, Node* obj) {
886 _recent_alloc_ctl = ctl;
887 _recent_alloc_obj = obj;
888 }
889 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
890 _dead_node_count++;
891 }
892 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; }
893 uint dead_node_count() { return _dead_node_count; }
894 void reset_dead_node_list() { _dead_node_list.Reset();
895 _dead_node_count = 0;
896 }
897 uint live_nodes() const {
898 int val = _unique - _dead_node_count;
899 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
900 return (uint) val;
901 }
902 #ifdef ASSERT
903 uint count_live_nodes_by_graph_walk();
904 void print_missing_nodes();
905 #endif
906
907 // Record modified nodes to check that they are put on IGVN worklist
908 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
909 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
910 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
911
912 // Constant table
913 ConstantTable& constant_table() { return _constant_table; }
914
915 MachConstantBaseNode* mach_constant_base_node();
916 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
917 // Generated by adlc, true if CallNode requires MachConstantBase.
918 bool needs_clone_jvms();
919
920 // Handy undefined Node
921 Node* top() const { return _top; }
922
923 // these are used by guys who need to know about creation and transformation of top:
924 Node* cached_top_node() { return _top; }
925 void set_cached_top_node(Node* tn);
926
927 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
928 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
929 Node_Notes* default_node_notes() const { return _default_node_notes; }
930 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
931
932 Node_Notes* node_notes_at(int idx) {
933 return locate_node_notes(_node_note_array, idx, false);
934 }
935 inline bool set_node_notes_at(int idx, Node_Notes* value);
936
937 // Copy notes from source to dest, if they exist.
938 // Overwrite dest only if source provides something.
939 // Return true if information was moved.
940 bool copy_node_notes_to(Node* dest, Node* source);
941
942 // Workhorse function to sort out the blocked Node_Notes array:
943 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
944 int idx, bool can_grow = false);
945
946 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
947
948 // Type management
949 Arena* type_arena() { return _type_arena; }
950 Dict* type_dict() { return _type_dict; }
951 void* type_hwm() { return _type_hwm; }
952 size_t type_last_size() { return _type_last_size; }
953 int num_alias_types() { return _num_alias_types; }
954
955 void init_type_arena() { _type_arena = &_Compile_types; }
956 void set_type_arena(Arena* a) { _type_arena = a; }
957 void set_type_dict(Dict* d) { _type_dict = d; }
958 void set_type_hwm(void* p) { _type_hwm = p; }
959 void set_type_last_size(size_t sz) { _type_last_size = sz; }
960
961 const TypeFunc* last_tf(ciMethod* m) {
962 return (m == _last_tf_m) ? _last_tf : NULL;
963 }
964 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
965 assert(m != NULL || tf == NULL, "");
966 _last_tf_m = m;
967 _last_tf = tf;
968 }
969
970 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
971 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); }
972 bool have_alias_type(const TypePtr* adr_type);
973 AliasType* alias_type(ciField* field);
974
975 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
976 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
977 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
978
979 // Building nodes
980 void rethrow_exceptions(JVMState* jvms);
981 void return_values(JVMState* jvms);
982 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
983
984 // Decide how to build a call.
985 // The profile factor is a discount to apply to this site's interp. profile.
986 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
987 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
988 bool allow_intrinsics = true, bool delayed_forbidden = false);
989 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
990 return should_delay_string_inlining(call_method, jvms) ||
991 should_delay_boxing_inlining(call_method, jvms);
992 }
993 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
994 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
995
996 // Helper functions to identify inlining potential at call-site
997 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
998 ciKlass* holder, ciMethod* callee,
999 const TypeOopPtr* receiver_type, bool is_virtual,
1000 bool &call_does_dispatch, int &vtable_index,
1001 bool check_access = true);
1002 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
1003 ciMethod* callee, const TypeOopPtr* receiver_type,
1004 bool check_access = true);
1005
1006 // Report if there were too many traps at a current method and bci.
1007 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1008 // If there is no MDO at all, report no trap unless told to assume it.
1009 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1010 // This version, unspecific to a particular bci, asks if
1011 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1012 bool too_many_traps(Deoptimization::DeoptReason reason,
1013 // Privately used parameter for logging:
1014 ciMethodData* logmd = NULL);
1015 // Report if there were too many recompiles at a method and bci.
1016 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1017 // Report if there were too many traps or recompiles at a method and bci.
1018 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1019 return too_many_traps(method, bci, reason) ||
1020 too_many_recompiles(method, bci, reason);
1021 }
1022 // Return a bitset with the reasons where deoptimization is allowed,
1023 // i.e., where there were not too many uncommon traps.
1024 int _allowed_reasons;
1025 int allowed_deopt_reasons() { return _allowed_reasons; }
1026 void set_allowed_deopt_reasons();
1027
1028 // Parsing, optimization
1029 PhaseGVN* initial_gvn() { return _initial_gvn; }
1030 Unique_Node_List* for_igvn() { return _for_igvn; }
1031 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
1032 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1033 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
1034
1035 // Replace n by nn using initial_gvn, calling hash_delete and
1036 // record_for_igvn as needed.
1037 void gvn_replace_by(Node* n, Node* nn);
1038
1039
1040 void identify_useful_nodes(Unique_Node_List &useful);
1041 void update_dead_node_list(Unique_Node_List &useful);
1042 void remove_useless_nodes (Unique_Node_List &useful);
1043
1044 WarmCallInfo* warm_calls() const { return _warm_calls; }
1045 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
1046 WarmCallInfo* pop_warm_call();
1047
1048 // Record this CallGenerator for inlining at the end of parsing.
1049 void add_late_inline(CallGenerator* cg) {
1050 _late_inlines.insert_before(_late_inlines_pos, cg);
1051 _late_inlines_pos++;
1052 }
1053
1054 void prepend_late_inline(CallGenerator* cg) {
1055 _late_inlines.insert_before(0, cg);
1056 }
1057
1058 void add_string_late_inline(CallGenerator* cg) {
1059 _string_late_inlines.push(cg);
1060 }
1061
1062 void add_boxing_late_inline(CallGenerator* cg) {
1063 _boxing_late_inlines.push(cg);
1064 }
1065
1066 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1067
1068 void process_print_inlining();
1069 void dump_print_inlining();
1070
1071 bool over_inlining_cutoff() const {
1072 if (!inlining_incrementally()) {
1073 return unique() > (uint)NodeCountInliningCutoff;
1074 } else {
1075 return live_nodes() > (uint)LiveNodeCountInliningCutoff;
1076 }
1077 }
1078
1079 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1080 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--; }
1081 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1082
1083 void inline_incrementally_one(PhaseIterGVN& igvn);
1084 void inline_incrementally(PhaseIterGVN& igvn);
1085 void inline_string_calls(bool parse_time);
1086 void inline_boxing_calls(PhaseIterGVN& igvn);
1087
1088 // Matching, CFG layout, allocation, code generation
1089 PhaseCFG* cfg() { return _cfg; }
1090 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1091 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1092 bool has_java_calls() const { return _java_calls > 0; }
1093 int java_calls() const { return _java_calls; }
1094 int inner_loops() const { return _inner_loops; }
1095 Matcher* matcher() { return _matcher; }
1096 PhaseRegAlloc* regalloc() { return _regalloc; }
1097 int frame_slots() const { return _frame_slots; }
1098 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words'
1099 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; }
1100 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1101 Arena* indexSet_arena() { return _indexSet_arena; }
1102 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1103 uint node_bundling_limit() { return _node_bundling_limit; }
1104 Bundle* node_bundling_base() { return _node_bundling_base; }
1105 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; }
1106 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; }
1107 bool starts_bundle(const Node *n) const;
1108 bool need_stack_bang(int frame_size_in_bytes) const;
1109 bool need_register_stack_bang() const;
1110
1111 void update_interpreter_frame_size(int size) {
1112 if (_interpreter_frame_size < size) {
1113 _interpreter_frame_size = size;
1114 }
1115 }
1116 int bang_size_in_bytes() const;
1117
1118 void set_matcher(Matcher* m) { _matcher = m; }
1119 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1120 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1121 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1122
1123 // Remember if this compilation changes hardware mode to 24-bit precision
1124 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1125 _select_24_bit_instr = selection;
1126 _in_24_bit_fp_mode = mode;
1127 }
1128
1129 void set_java_calls(int z) { _java_calls = z; }
1130 void set_inner_loops(int z) { _inner_loops = z; }
1131
1132 // Instruction bits passed off to the VM
1133 int code_size() { return _method_size; }
1134 CodeBuffer* code_buffer() { return &_code_buffer; }
1135 int first_block_size() { return _first_block_size; }
1136 void set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } }
1137 ExceptionHandlerTable* handler_table() { return &_handler_table; }
1138 ImplicitExceptionTable* inc_table() { return &_inc_table; }
1139 OopMapSet* oop_map_set() { return _oop_map_set; }
1140 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1141 Dependencies* dependencies() { return env()->dependencies(); }
1142 static int CompiledZap_count() { return _CompiledZap_count; }
1143 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; }
1144 void init_scratch_buffer_blob(int const_size);
1145 void clear_scratch_buffer_blob();
1146 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; }
1147 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; }
1148 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; }
1149
1150 // emit to scratch blob, report resulting size
1151 uint scratch_emit_size(const Node* n);
1152 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; }
1153 bool in_scratch_emit_size() const { return _in_scratch_emit_size; }
1154
1155 enum ScratchBufferBlob {
1156 #if defined(PPC64)
1157 MAX_inst_size = 2048,
1158 #else
1159 MAX_inst_size = 1024,
1160 #endif
1161 MAX_locs_size = 128, // number of relocInfo elements
1162 MAX_const_size = 128,
1163 MAX_stubs_size = 128
1164 };
1165
1166 // Major entry point. Given a Scope, compile the associated method.
1167 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1168 // replacement, entry_bci indicates the bytecode for which to compile a
1169 // continuation.
1170 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
1171 int entry_bci, bool subsume_loads, bool do_escape_analysis,
1172 bool eliminate_boxing, DirectiveSet* directive);
1173
1174 // Second major entry point. From the TypeFunc signature, generate code
1175 // to pass arguments from the Java calling convention to the C calling
1176 // convention.
1177 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1178 address stub_function, const char *stub_name,
1179 int is_fancy_jump, bool pass_tls,
1180 bool save_arg_registers, bool return_pc, DirectiveSet* directive);
1181
1182 // From the TypeFunc signature, generate code to pass arguments
1183 // from Compiled calling convention to Interpreter's calling convention
1184 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry);
1185
1186 // From the TypeFunc signature, generate code to pass arguments
1187 // from Interpreter's calling convention to Compiler's calling convention
1188 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf);
1189
1190 // Are we compiling a method?
1191 bool has_method() { return method() != NULL; }
1192
1193 // Maybe print some information about this compile.
1194 void print_compile_messages();
1195
1196 // Final graph reshaping, a post-pass after the regular optimizer is done.
1197 bool final_graph_reshaping();
1198
1199 // returns true if adr is completely contained in the given alias category
1200 bool must_alias(const TypePtr* adr, int alias_idx);
1201
1202 // returns true if adr overlaps with the given alias category
1203 bool can_alias(const TypePtr* adr, int alias_idx);
1204
1205 // Driver for converting compiler's IR into machine code bits
1206 void Output();
1207
1208 // Accessors for node bundling info.
1209 Bundle* node_bundling(const Node *n);
1210 bool valid_bundle_info(const Node *n);
1211
1212 // Schedule and Bundle the instructions
1213 void ScheduleAndBundle();
1214
1215 // Build OopMaps for each GC point
1216 void BuildOopMaps();
1217
1218 // Append debug info for the node "local" at safepoint node "sfpt" to the
1219 // "array", May also consult and add to "objs", which describes the
1220 // scalar-replaced objects.
1221 void FillLocArray( int idx, MachSafePointNode* sfpt,
1222 Node *local, GrowableArray<ScopeValue*> *array,
1223 GrowableArray<ScopeValue*> *objs );
1224
1225 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
1226 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
1227 // Requres that "objs" does not contains an ObjectValue whose id matches
1228 // that of "sv. Appends "sv".
1229 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
1230 ObjectValue* sv );
1231
1232 // Process an OopMap Element while emitting nodes
1233 void Process_OopMap_Node(MachNode *mach, int code_offset);
1234
1235 // Initialize code buffer
1236 CodeBuffer* init_buffer(uint* blk_starts);
1237
1238 // Write out basic block data to code buffer
1239 void fill_buffer(CodeBuffer* cb, uint* blk_starts);
1240
1241 // Determine which variable sized branches can be shortened
1242 void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size);
1243
1244 // Compute the size of first NumberOfLoopInstrToAlign instructions
1245 // at the head of a loop.
1246 void compute_loop_first_inst_sizes();
1247
1248 // Compute the information for the exception tables
1249 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels);
1250
1251 // Stack slots that may be unused by the calling convention but must
1252 // otherwise be preserved. On Intel this includes the return address.
1253 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1254 uint in_preserve_stack_slots();
1255
1256 // "Top of Stack" slots that may be unused by the calling convention but must
1257 // otherwise be preserved.
1258 // On Intel these are not necessary and the value can be zero.
1259 // On Sparc this describes the words reserved for storing a register window
1260 // when an interrupt occurs.
1261 static uint out_preserve_stack_slots();
1262
1263 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1264 // for calls to C. Supports the var-args backing area for register parms.
1265 uint varargs_C_out_slots_killed() const;
1266
1267 // Number of Stack Slots consumed by a synchronization entry
1268 int sync_stack_slots() const;
1269
1270 // Compute the name of old_SP. See <arch>.ad for frame layout.
1271 OptoReg::Name compute_old_SP();
1272
1273 private:
1274 // Phase control:
1275 void Init(int aliaslevel); // Prepare for a single compilation
1276 int Inline_Warm(); // Find more inlining work.
1277 void Finish_Warm(); // Give up on further inlines.
1278 void Optimize(); // Given a graph, optimize it
1279 void Code_Gen(); // Generate code from a graph
1280
1281 // Management of the AliasType table.
1282 void grow_alias_types();
1283 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1284 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1285 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1286
1287 void verify_top(Node*) const PRODUCT_RETURN;
1288
1289 // Intrinsic setup.
1290 void register_library_intrinsics(); // initializer
1291 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1292 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1293 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1294 void register_intrinsic(CallGenerator* cg); // update fn
1295
1296 #ifndef PRODUCT
1297 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT];
1298 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT];
1299 #endif
1300 // Function calls made by the public function final_graph_reshaping.
1301 // No need to be made public as they are not called elsewhere.
1302 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1303 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1304 void eliminate_redundant_card_marks(Node* n);
1305
1306 public:
1307
1308 // Note: Histogram array size is about 1 Kb.
1309 enum { // flag bits:
1310 _intrinsic_worked = 1, // succeeded at least once
1311 _intrinsic_failed = 2, // tried it but it failed
1312 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1313 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1314 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1315 };
1316 // Update histogram. Return boolean if this is a first-time occurrence.
1317 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1318 bool is_virtual, int flags) PRODUCT_RETURN0;
1319 static void print_intrinsic_statistics() PRODUCT_RETURN;
1320
1321 // Graph verification code
1322 // Walk the node list, verifying that there is a one-to-one
1323 // correspondence between Use-Def edges and Def-Use edges
1324 // The option no_dead_code enables stronger checks that the
1325 // graph is strongly connected from root in both directions.
1326 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1327
1328 // Verify GC barrier patterns
1329 void verify_barriers() PRODUCT_RETURN;
1330
1331 // End-of-run dumps.
1332 static void print_statistics() PRODUCT_RETURN;
1333
1334 // Dump formatted assembly
1335 void dump_asm(int *pcs = NULL, uint pc_limit = 0) PRODUCT_RETURN;
1336 void dump_pc(int *pcs, int pc_limit, Node *n);
1337
1338 // Verify ADLC assumptions during startup
1339 static void adlc_verification() PRODUCT_RETURN;
1340
1341 // Definitions of pd methods
1342 static void pd_compiler2_init();
1343
1344 // Static parse-time type checking logic for gen_subtype_check:
1345 enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1346 int static_subtype_check(ciKlass* superk, ciKlass* subk);
1347
1348 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1349 // Optional control dependency (for example, on range check)
1350 Node* ctrl = NULL);
1351
1352 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1353 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl);
1354
1355 // Auxiliary method for randomized fuzzing/stressing
1356 static bool randomized_select(int count);
1357
1358 // supporting clone_map
1359 CloneMap& clone_map();
1360 void set_clone_map(Dict* d);
1361
1362 bool is_compiling_clinit_for(ciKlass* k);
1363 };
1364
1365 #endif // SHARE_VM_OPTO_COMPILE_HPP