|
1 /* |
|
2 * Copyright (c) 1997, 2014, 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/port.hpp" |
|
36 #include "libadt/vectset.hpp" |
|
37 #include "memory/resourceArea.hpp" |
|
38 #include "opto/idealGraphPrinter.hpp" |
|
39 #include "opto/phasetype.hpp" |
|
40 #include "opto/phase.hpp" |
|
41 #include "opto/regmask.hpp" |
|
42 #include "runtime/deoptimization.hpp" |
|
43 #include "runtime/vmThread.hpp" |
|
44 #include "trace/tracing.hpp" |
|
45 #include "utilities/ticks.hpp" |
|
46 |
|
47 class Block; |
|
48 class Bundle; |
|
49 class C2Compiler; |
|
50 class CallGenerator; |
|
51 class ConnectionGraph; |
|
52 class InlineTree; |
|
53 class Int_Array; |
|
54 class Matcher; |
|
55 class MachConstantNode; |
|
56 class MachConstantBaseNode; |
|
57 class MachNode; |
|
58 class MachOper; |
|
59 class MachSafePointNode; |
|
60 class Node; |
|
61 class Node_Array; |
|
62 class Node_Notes; |
|
63 class OptoReg; |
|
64 class PhaseCFG; |
|
65 class PhaseGVN; |
|
66 class PhaseIterGVN; |
|
67 class PhaseRegAlloc; |
|
68 class PhaseCCP; |
|
69 class PhaseCCP_DCE; |
|
70 class RootNode; |
|
71 class relocInfo; |
|
72 class Scope; |
|
73 class StartNode; |
|
74 class SafePointNode; |
|
75 class JVMState; |
|
76 class Type; |
|
77 class TypeData; |
|
78 class TypePtr; |
|
79 class TypeOopPtr; |
|
80 class TypeFunc; |
|
81 class Unique_Node_List; |
|
82 class nmethod; |
|
83 class WarmCallInfo; |
|
84 class Node_Stack; |
|
85 struct Final_Reshape_Counts; |
|
86 |
|
87 //------------------------------Compile---------------------------------------- |
|
88 // This class defines a top-level Compiler invocation. |
|
89 |
|
90 class Compile : public Phase { |
|
91 friend class VMStructs; |
|
92 |
|
93 public: |
|
94 // Fixed alias indexes. (See also MergeMemNode.) |
|
95 enum { |
|
96 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value) |
|
97 AliasIdxBot = 2, // pseudo-index, aliases to everything |
|
98 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM |
|
99 }; |
|
100 |
|
101 // Variant of TraceTime(NULL, &_t_accumulator, TimeCompiler); |
|
102 // Integrated with logging. If logging is turned on, and dolog is true, |
|
103 // then brackets are put into the log, with time stamps and node counts. |
|
104 // (The time collection itself is always conditionalized on TimeCompiler.) |
|
105 class TracePhase : public TraceTime { |
|
106 private: |
|
107 Compile* C; |
|
108 CompileLog* _log; |
|
109 const char* _phase_name; |
|
110 bool _dolog; |
|
111 public: |
|
112 TracePhase(const char* name, elapsedTimer* accumulator, bool dolog); |
|
113 ~TracePhase(); |
|
114 }; |
|
115 |
|
116 // Information per category of alias (memory slice) |
|
117 class AliasType { |
|
118 private: |
|
119 friend class Compile; |
|
120 |
|
121 int _index; // unique index, used with MergeMemNode |
|
122 const TypePtr* _adr_type; // normalized address type |
|
123 ciField* _field; // relevant instance field, or null if none |
|
124 const Type* _element; // relevant array element type, or null if none |
|
125 bool _is_rewritable; // false if the memory is write-once only |
|
126 int _general_index; // if this is type is an instance, the general |
|
127 // type that this is an instance of |
|
128 |
|
129 void Init(int i, const TypePtr* at); |
|
130 |
|
131 public: |
|
132 int index() const { return _index; } |
|
133 const TypePtr* adr_type() const { return _adr_type; } |
|
134 ciField* field() const { return _field; } |
|
135 const Type* element() const { return _element; } |
|
136 bool is_rewritable() const { return _is_rewritable; } |
|
137 bool is_volatile() const { return (_field ? _field->is_volatile() : false); } |
|
138 int general_index() const { return (_general_index != 0) ? _general_index : _index; } |
|
139 |
|
140 void set_rewritable(bool z) { _is_rewritable = z; } |
|
141 void set_field(ciField* f) { |
|
142 assert(!_field,""); |
|
143 _field = f; |
|
144 if (f->is_final() || f->is_stable()) { |
|
145 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops. |
|
146 _is_rewritable = false; |
|
147 } |
|
148 } |
|
149 void set_element(const Type* e) { |
|
150 assert(_element == NULL, ""); |
|
151 _element = e; |
|
152 } |
|
153 |
|
154 void print_on(outputStream* st) PRODUCT_RETURN; |
|
155 }; |
|
156 |
|
157 enum { |
|
158 logAliasCacheSize = 6, |
|
159 AliasCacheSize = (1<<logAliasCacheSize) |
|
160 }; |
|
161 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type |
|
162 enum { |
|
163 trapHistLength = MethodData::_trap_hist_limit |
|
164 }; |
|
165 |
|
166 // Constant entry of the constant table. |
|
167 class Constant { |
|
168 private: |
|
169 BasicType _type; |
|
170 union { |
|
171 jvalue _value; |
|
172 Metadata* _metadata; |
|
173 } _v; |
|
174 int _offset; // offset of this constant (in bytes) relative to the constant table base. |
|
175 float _freq; |
|
176 bool _can_be_reused; // true (default) if the value can be shared with other users. |
|
177 |
|
178 public: |
|
179 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; } |
|
180 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) : |
|
181 _type(type), |
|
182 _offset(-1), |
|
183 _freq(freq), |
|
184 _can_be_reused(can_be_reused) |
|
185 { |
|
186 assert(type != T_METADATA, "wrong constructor"); |
|
187 _v._value = value; |
|
188 } |
|
189 Constant(Metadata* metadata, bool can_be_reused = true) : |
|
190 _type(T_METADATA), |
|
191 _offset(-1), |
|
192 _freq(0.0f), |
|
193 _can_be_reused(can_be_reused) |
|
194 { |
|
195 _v._metadata = metadata; |
|
196 } |
|
197 |
|
198 bool operator==(const Constant& other); |
|
199 |
|
200 BasicType type() const { return _type; } |
|
201 |
|
202 jlong get_jlong() const { return _v._value.j; } |
|
203 jfloat get_jfloat() const { return _v._value.f; } |
|
204 jdouble get_jdouble() const { return _v._value.d; } |
|
205 jobject get_jobject() const { return _v._value.l; } |
|
206 |
|
207 Metadata* get_metadata() const { return _v._metadata; } |
|
208 |
|
209 int offset() const { return _offset; } |
|
210 void set_offset(int offset) { _offset = offset; } |
|
211 |
|
212 float freq() const { return _freq; } |
|
213 void inc_freq(float freq) { _freq += freq; } |
|
214 |
|
215 bool can_be_reused() const { return _can_be_reused; } |
|
216 }; |
|
217 |
|
218 // Constant table. |
|
219 class ConstantTable { |
|
220 private: |
|
221 GrowableArray<Constant> _constants; // Constants of this table. |
|
222 int _size; // Size in bytes the emitted constant table takes (including padding). |
|
223 int _table_base_offset; // Offset of the table base that gets added to the constant offsets. |
|
224 int _nof_jump_tables; // Number of jump-tables in this constant table. |
|
225 |
|
226 static int qsort_comparator(Constant* a, Constant* b); |
|
227 |
|
228 // We use negative frequencies to keep the order of the |
|
229 // jump-tables in which they were added. Otherwise we get into |
|
230 // trouble with relocation. |
|
231 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); } |
|
232 |
|
233 public: |
|
234 ConstantTable() : |
|
235 _size(-1), |
|
236 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit). |
|
237 _nof_jump_tables(0) |
|
238 {} |
|
239 |
|
240 int size() const { assert(_size != -1, "not calculated yet"); return _size; } |
|
241 |
|
242 int calculate_table_base_offset() const; // AD specific |
|
243 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; } |
|
244 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; } |
|
245 |
|
246 void emit(CodeBuffer& cb); |
|
247 |
|
248 // Returns the offset of the last entry (the top) of the constant table. |
|
249 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); } |
|
250 |
|
251 void calculate_offsets_and_size(); |
|
252 int find_offset(Constant& con) const; |
|
253 |
|
254 void add(Constant& con); |
|
255 Constant add(MachConstantNode* n, BasicType type, jvalue value); |
|
256 Constant add(Metadata* metadata); |
|
257 Constant add(MachConstantNode* n, MachOper* oper); |
|
258 Constant add(MachConstantNode* n, jfloat f) { |
|
259 jvalue value; value.f = f; |
|
260 return add(n, T_FLOAT, value); |
|
261 } |
|
262 Constant add(MachConstantNode* n, jdouble d) { |
|
263 jvalue value; value.d = d; |
|
264 return add(n, T_DOUBLE, value); |
|
265 } |
|
266 |
|
267 // Jump-table |
|
268 Constant add_jump_table(MachConstantNode* n); |
|
269 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const; |
|
270 }; |
|
271 |
|
272 private: |
|
273 // Fixed parameters to this compilation. |
|
274 const int _compile_id; |
|
275 const bool _save_argument_registers; // save/restore arg regs for trampolines |
|
276 const bool _subsume_loads; // Load can be matched as part of a larger op. |
|
277 const bool _do_escape_analysis; // Do escape analysis. |
|
278 const bool _eliminate_boxing; // Do boxing elimination. |
|
279 ciMethod* _method; // The method being compiled. |
|
280 int _entry_bci; // entry bci for osr methods. |
|
281 const TypeFunc* _tf; // My kind of signature |
|
282 InlineTree* _ilt; // Ditto (temporary). |
|
283 address _stub_function; // VM entry for stub being compiled, or NULL |
|
284 const char* _stub_name; // Name of stub or adapter being compiled, or NULL |
|
285 address _stub_entry_point; // Compile code entry for generated stub, or NULL |
|
286 |
|
287 // Control of this compilation. |
|
288 int _num_loop_opts; // Number of iterations for doing loop optimiztions |
|
289 int _max_inline_size; // Max inline size for this compilation |
|
290 int _freq_inline_size; // Max hot method inline size for this compilation |
|
291 int _fixed_slots; // count of frame slots not allocated by the register |
|
292 // allocator i.e. locks, original deopt pc, etc. |
|
293 // For deopt |
|
294 int _orig_pc_slot; |
|
295 int _orig_pc_slot_offset_in_bytes; |
|
296 |
|
297 int _major_progress; // Count of something big happening |
|
298 bool _inlining_progress; // progress doing incremental inlining? |
|
299 bool _inlining_incrementally;// Are we doing incremental inlining (post parse) |
|
300 bool _has_loops; // True if the method _may_ have some loops |
|
301 bool _has_split_ifs; // True if the method _may_ have some split-if |
|
302 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. |
|
303 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated |
|
304 bool _has_boxed_value; // True if a boxed object is allocated |
|
305 int _max_vector_size; // Maximum size of generated vectors |
|
306 uint _trap_hist[trapHistLength]; // Cumulative traps |
|
307 bool _trap_can_recompile; // Have we emitted a recompiling trap? |
|
308 uint _decompile_count; // Cumulative decompilation counts. |
|
309 bool _do_inlining; // True if we intend to do inlining |
|
310 bool _do_scheduling; // True if we intend to do scheduling |
|
311 bool _do_freq_based_layout; // True if we intend to do frequency based block layout |
|
312 bool _do_count_invocations; // True if we generate code to count invocations |
|
313 bool _do_method_data_update; // True if we generate code to update MethodData*s |
|
314 int _AliasLevel; // Locally-adjusted version of AliasLevel flag. |
|
315 bool _print_assembly; // True if we should dump assembly code for this compilation |
|
316 bool _print_inlining; // True if we should print inlining for this compilation |
|
317 bool _print_intrinsics; // True if we should print intrinsics for this compilation |
|
318 #ifndef PRODUCT |
|
319 bool _trace_opto_output; |
|
320 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing |
|
321 #endif |
|
322 bool _has_irreducible_loop; // Found irreducible loops |
|
323 // JSR 292 |
|
324 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. |
|
325 RTMState _rtm_state; // State of Restricted Transactional Memory usage |
|
326 |
|
327 // Compilation environment. |
|
328 Arena _comp_arena; // Arena with lifetime equivalent to Compile |
|
329 ciEnv* _env; // CI interface |
|
330 CompileLog* _log; // from CompilerThread |
|
331 const char* _failure_reason; // for record_failure/failing pattern |
|
332 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics. |
|
333 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching. |
|
334 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates. |
|
335 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common |
|
336 ConnectionGraph* _congraph; |
|
337 #ifndef PRODUCT |
|
338 IdealGraphPrinter* _printer; |
|
339 #endif |
|
340 |
|
341 |
|
342 // Node management |
|
343 uint _unique; // Counter for unique Node indices |
|
344 VectorSet _dead_node_list; // Set of dead nodes |
|
345 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N). |
|
346 // So use this to keep count and make the call O(1). |
|
347 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx> |
|
348 Arena _node_arena; // Arena for new-space Nodes |
|
349 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform |
|
350 RootNode* _root; // Unique root of compilation, or NULL after bail-out. |
|
351 Node* _top; // Unique top node. (Reset by various phases.) |
|
352 |
|
353 Node* _immutable_memory; // Initial memory state |
|
354 |
|
355 Node* _recent_alloc_obj; |
|
356 Node* _recent_alloc_ctl; |
|
357 |
|
358 // Constant table |
|
359 ConstantTable _constant_table; // The constant table for this compile. |
|
360 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton. |
|
361 |
|
362 |
|
363 // Blocked array of debugging and profiling information, |
|
364 // tracked per node. |
|
365 enum { _log2_node_notes_block_size = 8, |
|
366 _node_notes_block_size = (1<<_log2_node_notes_block_size) |
|
367 }; |
|
368 GrowableArray<Node_Notes*>* _node_note_array; |
|
369 Node_Notes* _default_node_notes; // default notes for new nodes |
|
370 |
|
371 // After parsing and every bulk phase we hang onto the Root instruction. |
|
372 // The RootNode instruction is where the whole program begins. It produces |
|
373 // the initial Control and BOTTOM for everybody else. |
|
374 |
|
375 // Type management |
|
376 Arena _Compile_types; // Arena for all types |
|
377 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared() |
|
378 Dict* _type_dict; // Intern table |
|
379 void* _type_hwm; // Last allocation (see Type::operator new/delete) |
|
380 size_t _type_last_size; // Last allocation size (see Type::operator new/delete) |
|
381 ciMethod* _last_tf_m; // Cache for |
|
382 const TypeFunc* _last_tf; // TypeFunc::make |
|
383 AliasType** _alias_types; // List of alias types seen so far. |
|
384 int _num_alias_types; // Logical length of _alias_types |
|
385 int _max_alias_types; // Physical length of _alias_types |
|
386 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking |
|
387 |
|
388 // Parsing, optimization |
|
389 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN |
|
390 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN |
|
391 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining. |
|
392 |
|
393 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after |
|
394 // main parsing has finished. |
|
395 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations |
|
396 |
|
397 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations |
|
398 |
|
399 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining) |
|
400 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending |
|
401 |
|
402 |
|
403 // Inlining may not happen in parse order which would make |
|
404 // PrintInlining output confusing. Keep track of PrintInlining |
|
405 // pieces in order. |
|
406 class PrintInliningBuffer : public ResourceObj { |
|
407 private: |
|
408 CallGenerator* _cg; |
|
409 stringStream* _ss; |
|
410 |
|
411 public: |
|
412 PrintInliningBuffer() |
|
413 : _cg(NULL) { _ss = new stringStream(); } |
|
414 |
|
415 stringStream* ss() const { return _ss; } |
|
416 CallGenerator* cg() const { return _cg; } |
|
417 void set_cg(CallGenerator* cg) { _cg = cg; } |
|
418 }; |
|
419 |
|
420 GrowableArray<PrintInliningBuffer>* _print_inlining_list; |
|
421 int _print_inlining_idx; |
|
422 |
|
423 // Only keep nodes in the expensive node list that need to be optimized |
|
424 void cleanup_expensive_nodes(PhaseIterGVN &igvn); |
|
425 // Use for sorting expensive nodes to bring similar nodes together |
|
426 static int cmp_expensive_nodes(Node** n1, Node** n2); |
|
427 // Expensive nodes list already sorted? |
|
428 bool expensive_nodes_sorted() const; |
|
429 // Remove the speculative part of types and clean up the graph |
|
430 void remove_speculative_types(PhaseIterGVN &igvn); |
|
431 |
|
432 // Are we within a PreserveJVMState block? |
|
433 int _preserve_jvm_state; |
|
434 |
|
435 void* _replay_inline_data; // Pointer to data loaded from file |
|
436 |
|
437 public: |
|
438 |
|
439 outputStream* print_inlining_stream() const { |
|
440 return _print_inlining_list->adr_at(_print_inlining_idx)->ss(); |
|
441 } |
|
442 |
|
443 void print_inlining_skip(CallGenerator* cg) { |
|
444 if (_print_inlining) { |
|
445 _print_inlining_list->adr_at(_print_inlining_idx)->set_cg(cg); |
|
446 _print_inlining_idx++; |
|
447 _print_inlining_list->insert_before(_print_inlining_idx, PrintInliningBuffer()); |
|
448 } |
|
449 } |
|
450 |
|
451 void print_inlining_insert(CallGenerator* cg) { |
|
452 if (_print_inlining) { |
|
453 for (int i = 0; i < _print_inlining_list->length(); i++) { |
|
454 if (_print_inlining_list->adr_at(i)->cg() == cg) { |
|
455 _print_inlining_list->insert_before(i+1, PrintInliningBuffer()); |
|
456 _print_inlining_idx = i+1; |
|
457 _print_inlining_list->adr_at(i)->set_cg(NULL); |
|
458 return; |
|
459 } |
|
460 } |
|
461 ShouldNotReachHere(); |
|
462 } |
|
463 } |
|
464 |
|
465 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) { |
|
466 stringStream ss; |
|
467 CompileTask::print_inlining(&ss, method, inline_level, bci, msg); |
|
468 print_inlining_stream()->print("%s", ss.as_string()); |
|
469 } |
|
470 |
|
471 void* replay_inline_data() const { return _replay_inline_data; } |
|
472 |
|
473 // Dump inlining replay data to the stream. |
|
474 void dump_inline_data(outputStream* out); |
|
475 |
|
476 private: |
|
477 // Matching, CFG layout, allocation, code generation |
|
478 PhaseCFG* _cfg; // Results of CFG finding |
|
479 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result |
|
480 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results |
|
481 int _java_calls; // Number of java calls in the method |
|
482 int _inner_loops; // Number of inner loops in the method |
|
483 Matcher* _matcher; // Engine to map ideal to machine instructions |
|
484 PhaseRegAlloc* _regalloc; // Results of register allocation. |
|
485 int _frame_slots; // Size of total frame in stack slots |
|
486 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries |
|
487 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout) |
|
488 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin |
|
489 void* _indexSet_free_block_list; // free list of IndexSet bit blocks |
|
490 int _interpreter_frame_size; |
|
491 |
|
492 uint _node_bundling_limit; |
|
493 Bundle* _node_bundling_base; // Information for instruction bundling |
|
494 |
|
495 // Instruction bits passed off to the VM |
|
496 int _method_size; // Size of nmethod code segment in bytes |
|
497 CodeBuffer _code_buffer; // Where the code is assembled |
|
498 int _first_block_size; // Size of unvalidated entry point code / OSR poison code |
|
499 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers |
|
500 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code |
|
501 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location) |
|
502 static int _CompiledZap_count; // counter compared against CompileZap[First/Last] |
|
503 BufferBlob* _scratch_buffer_blob; // For temporary code buffers. |
|
504 relocInfo* _scratch_locs_memory; // For temporary code buffers. |
|
505 int _scratch_const_size; // For temporary code buffers. |
|
506 bool _in_scratch_emit_size; // true when in scratch_emit_size. |
|
507 |
|
508 public: |
|
509 // Accessors |
|
510 |
|
511 // The Compile instance currently active in this (compiler) thread. |
|
512 static Compile* current() { |
|
513 return (Compile*) ciEnv::current()->compiler_data(); |
|
514 } |
|
515 |
|
516 // ID for this compilation. Useful for setting breakpoints in the debugger. |
|
517 int compile_id() const { return _compile_id; } |
|
518 |
|
519 // Does this compilation allow instructions to subsume loads? User |
|
520 // instructions that subsume a load may result in an unschedulable |
|
521 // instruction sequence. |
|
522 bool subsume_loads() const { return _subsume_loads; } |
|
523 /** Do escape analysis. */ |
|
524 bool do_escape_analysis() const { return _do_escape_analysis; } |
|
525 /** Do boxing elimination. */ |
|
526 bool eliminate_boxing() const { return _eliminate_boxing; } |
|
527 /** Do aggressive boxing elimination. */ |
|
528 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; } |
|
529 bool save_argument_registers() const { return _save_argument_registers; } |
|
530 |
|
531 |
|
532 // Other fixed compilation parameters. |
|
533 ciMethod* method() const { return _method; } |
|
534 int entry_bci() const { return _entry_bci; } |
|
535 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; } |
|
536 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); } |
|
537 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; } |
|
538 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; } |
|
539 InlineTree* ilt() const { return _ilt; } |
|
540 address stub_function() const { return _stub_function; } |
|
541 const char* stub_name() const { return _stub_name; } |
|
542 address stub_entry_point() const { return _stub_entry_point; } |
|
543 |
|
544 // Control of this compilation. |
|
545 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; } |
|
546 void set_fixed_slots(int n) { _fixed_slots = n; } |
|
547 int major_progress() const { return _major_progress; } |
|
548 void set_inlining_progress(bool z) { _inlining_progress = z; } |
|
549 int inlining_progress() const { return _inlining_progress; } |
|
550 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; } |
|
551 int inlining_incrementally() const { return _inlining_incrementally; } |
|
552 void set_major_progress() { _major_progress++; } |
|
553 void clear_major_progress() { _major_progress = 0; } |
|
554 int num_loop_opts() const { return _num_loop_opts; } |
|
555 void set_num_loop_opts(int n) { _num_loop_opts = n; } |
|
556 int max_inline_size() const { return _max_inline_size; } |
|
557 void set_freq_inline_size(int n) { _freq_inline_size = n; } |
|
558 int freq_inline_size() const { return _freq_inline_size; } |
|
559 void set_max_inline_size(int n) { _max_inline_size = n; } |
|
560 bool has_loops() const { return _has_loops; } |
|
561 void set_has_loops(bool z) { _has_loops = z; } |
|
562 bool has_split_ifs() const { return _has_split_ifs; } |
|
563 void set_has_split_ifs(bool z) { _has_split_ifs = z; } |
|
564 bool has_unsafe_access() const { return _has_unsafe_access; } |
|
565 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } |
|
566 bool has_stringbuilder() const { return _has_stringbuilder; } |
|
567 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } |
|
568 bool has_boxed_value() const { return _has_boxed_value; } |
|
569 void set_has_boxed_value(bool z) { _has_boxed_value = z; } |
|
570 int max_vector_size() const { return _max_vector_size; } |
|
571 void set_max_vector_size(int s) { _max_vector_size = s; } |
|
572 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } |
|
573 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } |
|
574 bool trap_can_recompile() const { return _trap_can_recompile; } |
|
575 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; } |
|
576 uint decompile_count() const { return _decompile_count; } |
|
577 void set_decompile_count(uint c) { _decompile_count = c; } |
|
578 bool allow_range_check_smearing() const; |
|
579 bool do_inlining() const { return _do_inlining; } |
|
580 void set_do_inlining(bool z) { _do_inlining = z; } |
|
581 bool do_scheduling() const { return _do_scheduling; } |
|
582 void set_do_scheduling(bool z) { _do_scheduling = z; } |
|
583 bool do_freq_based_layout() const{ return _do_freq_based_layout; } |
|
584 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; } |
|
585 bool do_count_invocations() const{ return _do_count_invocations; } |
|
586 void set_do_count_invocations(bool z){ _do_count_invocations = z; } |
|
587 bool do_method_data_update() const { return _do_method_data_update; } |
|
588 void set_do_method_data_update(bool z) { _do_method_data_update = z; } |
|
589 int AliasLevel() const { return _AliasLevel; } |
|
590 bool print_assembly() const { return _print_assembly; } |
|
591 void set_print_assembly(bool z) { _print_assembly = z; } |
|
592 bool print_inlining() const { return _print_inlining; } |
|
593 void set_print_inlining(bool z) { _print_inlining = z; } |
|
594 bool print_intrinsics() const { return _print_intrinsics; } |
|
595 void set_print_intrinsics(bool z) { _print_intrinsics = z; } |
|
596 RTMState rtm_state() const { return _rtm_state; } |
|
597 void set_rtm_state(RTMState s) { _rtm_state = s; } |
|
598 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; } |
|
599 bool profile_rtm() const { return _rtm_state == ProfileRTM; } |
|
600 // check the CompilerOracle for special behaviours for this compile |
|
601 bool method_has_option(const char * option) { |
|
602 return method() != NULL && method()->has_option(option); |
|
603 } |
|
604 #ifndef PRODUCT |
|
605 bool trace_opto_output() const { return _trace_opto_output; } |
|
606 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; } |
|
607 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } |
|
608 int _in_dump_cnt; // Required for dumping ir nodes. |
|
609 #endif |
|
610 bool has_irreducible_loop() const { return _has_irreducible_loop; } |
|
611 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; } |
|
612 |
|
613 // JSR 292 |
|
614 bool has_method_handle_invokes() const { return _has_method_handle_invokes; } |
|
615 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } |
|
616 |
|
617 Ticks _latest_stage_start_counter; |
|
618 |
|
619 void begin_method() { |
|
620 #ifndef PRODUCT |
|
621 if (_printer) _printer->begin_method(this); |
|
622 #endif |
|
623 C->_latest_stage_start_counter.stamp(); |
|
624 } |
|
625 |
|
626 void print_method(CompilerPhaseType cpt, int level = 1) { |
|
627 EventCompilerPhase event; |
|
628 if (event.should_commit()) { |
|
629 event.set_starttime(C->_latest_stage_start_counter); |
|
630 event.set_phase((u1) cpt); |
|
631 event.set_compileID(C->_compile_id); |
|
632 event.set_phaseLevel(level); |
|
633 event.commit(); |
|
634 } |
|
635 |
|
636 |
|
637 #ifndef PRODUCT |
|
638 if (_printer) _printer->print_method(this, CompilerPhaseTypeHelper::to_string(cpt), level); |
|
639 #endif |
|
640 C->_latest_stage_start_counter.stamp(); |
|
641 } |
|
642 |
|
643 void end_method(int level = 1) { |
|
644 EventCompilerPhase event; |
|
645 if (event.should_commit()) { |
|
646 event.set_starttime(C->_latest_stage_start_counter); |
|
647 event.set_phase((u1) PHASE_END); |
|
648 event.set_compileID(C->_compile_id); |
|
649 event.set_phaseLevel(level); |
|
650 event.commit(); |
|
651 } |
|
652 #ifndef PRODUCT |
|
653 if (_printer) _printer->end_method(); |
|
654 #endif |
|
655 } |
|
656 |
|
657 int macro_count() const { return _macro_nodes->length(); } |
|
658 int predicate_count() const { return _predicate_opaqs->length();} |
|
659 int expensive_count() const { return _expensive_nodes->length(); } |
|
660 Node* macro_node(int idx) const { return _macro_nodes->at(idx); } |
|
661 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);} |
|
662 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); } |
|
663 ConnectionGraph* congraph() { return _congraph;} |
|
664 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;} |
|
665 void add_macro_node(Node * n) { |
|
666 //assert(n->is_macro(), "must be a macro node"); |
|
667 assert(!_macro_nodes->contains(n), " duplicate entry in expand list"); |
|
668 _macro_nodes->append(n); |
|
669 } |
|
670 void remove_macro_node(Node * n) { |
|
671 // this function may be called twice for a node so check |
|
672 // that the node is in the array before attempting to remove it |
|
673 if (_macro_nodes->contains(n)) |
|
674 _macro_nodes->remove(n); |
|
675 // remove from _predicate_opaqs list also if it is there |
|
676 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){ |
|
677 _predicate_opaqs->remove(n); |
|
678 } |
|
679 } |
|
680 void add_expensive_node(Node * n); |
|
681 void remove_expensive_node(Node * n) { |
|
682 if (_expensive_nodes->contains(n)) { |
|
683 _expensive_nodes->remove(n); |
|
684 } |
|
685 } |
|
686 void add_predicate_opaq(Node * n) { |
|
687 assert(!_predicate_opaqs->contains(n), " duplicate entry in predicate opaque1"); |
|
688 assert(_macro_nodes->contains(n), "should have already been in macro list"); |
|
689 _predicate_opaqs->append(n); |
|
690 } |
|
691 // remove the opaque nodes that protect the predicates so that the unused checks and |
|
692 // uncommon traps will be eliminated from the graph. |
|
693 void cleanup_loop_predicates(PhaseIterGVN &igvn); |
|
694 bool is_predicate_opaq(Node * n) { |
|
695 return _predicate_opaqs->contains(n); |
|
696 } |
|
697 |
|
698 // Are there candidate expensive nodes for optimization? |
|
699 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn); |
|
700 // Check whether n1 and n2 are similar |
|
701 static int cmp_expensive_nodes(Node* n1, Node* n2); |
|
702 // Sort expensive nodes to locate similar expensive nodes |
|
703 void sort_expensive_nodes(); |
|
704 |
|
705 // Compilation environment. |
|
706 Arena* comp_arena() { return &_comp_arena; } |
|
707 ciEnv* env() const { return _env; } |
|
708 CompileLog* log() const { return _log; } |
|
709 bool failing() const { return _env->failing() || _failure_reason != NULL; } |
|
710 const char* failure_reason() { return _failure_reason; } |
|
711 bool failure_reason_is(const char* r) { return (r==_failure_reason) || (r!=NULL && _failure_reason!=NULL && strcmp(r, _failure_reason)==0); } |
|
712 |
|
713 void record_failure(const char* reason); |
|
714 void record_method_not_compilable(const char* reason, bool all_tiers = false) { |
|
715 // All bailouts cover "all_tiers" when TieredCompilation is off. |
|
716 if (!TieredCompilation) all_tiers = true; |
|
717 env()->record_method_not_compilable(reason, all_tiers); |
|
718 // Record failure reason. |
|
719 record_failure(reason); |
|
720 } |
|
721 void record_method_not_compilable_all_tiers(const char* reason) { |
|
722 record_method_not_compilable(reason, true); |
|
723 } |
|
724 bool check_node_count(uint margin, const char* reason) { |
|
725 if (live_nodes() + margin > (uint)MaxNodeLimit) { |
|
726 record_method_not_compilable(reason); |
|
727 return true; |
|
728 } else { |
|
729 return false; |
|
730 } |
|
731 } |
|
732 |
|
733 // Node management |
|
734 uint unique() const { return _unique; } |
|
735 uint next_unique() { return _unique++; } |
|
736 void set_unique(uint i) { _unique = i; } |
|
737 static int debug_idx() { return debug_only(_debug_idx)+0; } |
|
738 static void set_debug_idx(int i) { debug_only(_debug_idx = i); } |
|
739 Arena* node_arena() { return &_node_arena; } |
|
740 Arena* old_arena() { return &_old_arena; } |
|
741 RootNode* root() const { return _root; } |
|
742 void set_root(RootNode* r) { _root = r; } |
|
743 StartNode* start() const; // (Derived from root.) |
|
744 void init_start(StartNode* s); |
|
745 Node* immutable_memory(); |
|
746 |
|
747 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; } |
|
748 Node* recent_alloc_obj() const { return _recent_alloc_obj; } |
|
749 void set_recent_alloc(Node* ctl, Node* obj) { |
|
750 _recent_alloc_ctl = ctl; |
|
751 _recent_alloc_obj = obj; |
|
752 } |
|
753 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return; |
|
754 _dead_node_count++; |
|
755 } |
|
756 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; } |
|
757 uint dead_node_count() { return _dead_node_count; } |
|
758 void reset_dead_node_list() { _dead_node_list.Reset(); |
|
759 _dead_node_count = 0; |
|
760 } |
|
761 uint live_nodes() const { |
|
762 int val = _unique - _dead_node_count; |
|
763 assert (val >= 0, err_msg_res("number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count)); |
|
764 return (uint) val; |
|
765 } |
|
766 #ifdef ASSERT |
|
767 uint count_live_nodes_by_graph_walk(); |
|
768 void print_missing_nodes(); |
|
769 #endif |
|
770 |
|
771 // Constant table |
|
772 ConstantTable& constant_table() { return _constant_table; } |
|
773 |
|
774 MachConstantBaseNode* mach_constant_base_node(); |
|
775 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; } |
|
776 // Generated by adlc, true if CallNode requires MachConstantBase. |
|
777 bool needs_clone_jvms(); |
|
778 |
|
779 // Handy undefined Node |
|
780 Node* top() const { return _top; } |
|
781 |
|
782 // these are used by guys who need to know about creation and transformation of top: |
|
783 Node* cached_top_node() { return _top; } |
|
784 void set_cached_top_node(Node* tn); |
|
785 |
|
786 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; } |
|
787 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; } |
|
788 Node_Notes* default_node_notes() const { return _default_node_notes; } |
|
789 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; } |
|
790 |
|
791 Node_Notes* node_notes_at(int idx) { |
|
792 return locate_node_notes(_node_note_array, idx, false); |
|
793 } |
|
794 inline bool set_node_notes_at(int idx, Node_Notes* value); |
|
795 |
|
796 // Copy notes from source to dest, if they exist. |
|
797 // Overwrite dest only if source provides something. |
|
798 // Return true if information was moved. |
|
799 bool copy_node_notes_to(Node* dest, Node* source); |
|
800 |
|
801 // Workhorse function to sort out the blocked Node_Notes array: |
|
802 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr, |
|
803 int idx, bool can_grow = false); |
|
804 |
|
805 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by); |
|
806 |
|
807 // Type management |
|
808 Arena* type_arena() { return _type_arena; } |
|
809 Dict* type_dict() { return _type_dict; } |
|
810 void* type_hwm() { return _type_hwm; } |
|
811 size_t type_last_size() { return _type_last_size; } |
|
812 int num_alias_types() { return _num_alias_types; } |
|
813 |
|
814 void init_type_arena() { _type_arena = &_Compile_types; } |
|
815 void set_type_arena(Arena* a) { _type_arena = a; } |
|
816 void set_type_dict(Dict* d) { _type_dict = d; } |
|
817 void set_type_hwm(void* p) { _type_hwm = p; } |
|
818 void set_type_last_size(size_t sz) { _type_last_size = sz; } |
|
819 |
|
820 const TypeFunc* last_tf(ciMethod* m) { |
|
821 return (m == _last_tf_m) ? _last_tf : NULL; |
|
822 } |
|
823 void set_last_tf(ciMethod* m, const TypeFunc* tf) { |
|
824 assert(m != NULL || tf == NULL, ""); |
|
825 _last_tf_m = m; |
|
826 _last_tf = tf; |
|
827 } |
|
828 |
|
829 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; } |
|
830 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); } |
|
831 bool have_alias_type(const TypePtr* adr_type); |
|
832 AliasType* alias_type(ciField* field); |
|
833 |
|
834 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); } |
|
835 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); } |
|
836 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); } |
|
837 |
|
838 // Building nodes |
|
839 void rethrow_exceptions(JVMState* jvms); |
|
840 void return_values(JVMState* jvms); |
|
841 JVMState* build_start_state(StartNode* start, const TypeFunc* tf); |
|
842 |
|
843 // Decide how to build a call. |
|
844 // The profile factor is a discount to apply to this site's interp. profile. |
|
845 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, |
|
846 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL, |
|
847 bool allow_intrinsics = true, bool delayed_forbidden = false); |
|
848 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) { |
|
849 return should_delay_string_inlining(call_method, jvms) || |
|
850 should_delay_boxing_inlining(call_method, jvms); |
|
851 } |
|
852 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms); |
|
853 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms); |
|
854 |
|
855 // Helper functions to identify inlining potential at call-site |
|
856 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass, |
|
857 ciKlass* holder, ciMethod* callee, |
|
858 const TypeOopPtr* receiver_type, bool is_virtual, |
|
859 bool &call_does_dispatch, int &vtable_index); |
|
860 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass, |
|
861 ciMethod* callee, const TypeOopPtr* receiver_type); |
|
862 |
|
863 // Report if there were too many traps at a current method and bci. |
|
864 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. |
|
865 // If there is no MDO at all, report no trap unless told to assume it. |
|
866 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason); |
|
867 // This version, unspecific to a particular bci, asks if |
|
868 // PerMethodTrapLimit was exceeded for all inlined methods seen so far. |
|
869 bool too_many_traps(Deoptimization::DeoptReason reason, |
|
870 // Privately used parameter for logging: |
|
871 ciMethodData* logmd = NULL); |
|
872 // Report if there were too many recompiles at a method and bci. |
|
873 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason); |
|
874 // Return a bitset with the reasons where deoptimization is allowed, |
|
875 // i.e., where there were not too many uncommon traps. |
|
876 int _allowed_reasons; |
|
877 int allowed_deopt_reasons() { return _allowed_reasons; } |
|
878 void set_allowed_deopt_reasons(); |
|
879 |
|
880 // Parsing, optimization |
|
881 PhaseGVN* initial_gvn() { return _initial_gvn; } |
|
882 Unique_Node_List* for_igvn() { return _for_igvn; } |
|
883 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List. |
|
884 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } |
|
885 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; } |
|
886 |
|
887 // Replace n by nn using initial_gvn, calling hash_delete and |
|
888 // record_for_igvn as needed. |
|
889 void gvn_replace_by(Node* n, Node* nn); |
|
890 |
|
891 |
|
892 void identify_useful_nodes(Unique_Node_List &useful); |
|
893 void update_dead_node_list(Unique_Node_List &useful); |
|
894 void remove_useless_nodes (Unique_Node_List &useful); |
|
895 |
|
896 WarmCallInfo* warm_calls() const { return _warm_calls; } |
|
897 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; } |
|
898 WarmCallInfo* pop_warm_call(); |
|
899 |
|
900 // Record this CallGenerator for inlining at the end of parsing. |
|
901 void add_late_inline(CallGenerator* cg) { |
|
902 _late_inlines.insert_before(_late_inlines_pos, cg); |
|
903 _late_inlines_pos++; |
|
904 } |
|
905 |
|
906 void prepend_late_inline(CallGenerator* cg) { |
|
907 _late_inlines.insert_before(0, cg); |
|
908 } |
|
909 |
|
910 void add_string_late_inline(CallGenerator* cg) { |
|
911 _string_late_inlines.push(cg); |
|
912 } |
|
913 |
|
914 void add_boxing_late_inline(CallGenerator* cg) { |
|
915 _boxing_late_inlines.push(cg); |
|
916 } |
|
917 |
|
918 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful); |
|
919 |
|
920 void dump_inlining(); |
|
921 |
|
922 bool over_inlining_cutoff() const { |
|
923 if (!inlining_incrementally()) { |
|
924 return unique() > (uint)NodeCountInliningCutoff; |
|
925 } else { |
|
926 return live_nodes() > (uint)LiveNodeCountInliningCutoff; |
|
927 } |
|
928 } |
|
929 |
|
930 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; } |
|
931 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--; } |
|
932 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; } |
|
933 |
|
934 void inline_incrementally_one(PhaseIterGVN& igvn); |
|
935 void inline_incrementally(PhaseIterGVN& igvn); |
|
936 void inline_string_calls(bool parse_time); |
|
937 void inline_boxing_calls(PhaseIterGVN& igvn); |
|
938 |
|
939 // Matching, CFG layout, allocation, code generation |
|
940 PhaseCFG* cfg() { return _cfg; } |
|
941 bool select_24_bit_instr() const { return _select_24_bit_instr; } |
|
942 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } |
|
943 bool has_java_calls() const { return _java_calls > 0; } |
|
944 int java_calls() const { return _java_calls; } |
|
945 int inner_loops() const { return _inner_loops; } |
|
946 Matcher* matcher() { return _matcher; } |
|
947 PhaseRegAlloc* regalloc() { return _regalloc; } |
|
948 int frame_slots() const { return _frame_slots; } |
|
949 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words' |
|
950 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; } |
|
951 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; } |
|
952 Arena* indexSet_arena() { return _indexSet_arena; } |
|
953 void* indexSet_free_block_list() { return _indexSet_free_block_list; } |
|
954 uint node_bundling_limit() { return _node_bundling_limit; } |
|
955 Bundle* node_bundling_base() { return _node_bundling_base; } |
|
956 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; } |
|
957 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; } |
|
958 bool starts_bundle(const Node *n) const; |
|
959 bool need_stack_bang(int frame_size_in_bytes) const; |
|
960 bool need_register_stack_bang() const; |
|
961 |
|
962 void update_interpreter_frame_size(int size) { |
|
963 if (_interpreter_frame_size < size) { |
|
964 _interpreter_frame_size = size; |
|
965 } |
|
966 } |
|
967 int bang_size_in_bytes() const; |
|
968 |
|
969 void set_matcher(Matcher* m) { _matcher = m; } |
|
970 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; } |
|
971 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; } |
|
972 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; } |
|
973 |
|
974 // Remember if this compilation changes hardware mode to 24-bit precision |
|
975 void set_24_bit_selection_and_mode(bool selection, bool mode) { |
|
976 _select_24_bit_instr = selection; |
|
977 _in_24_bit_fp_mode = mode; |
|
978 } |
|
979 |
|
980 void set_java_calls(int z) { _java_calls = z; } |
|
981 void set_inner_loops(int z) { _inner_loops = z; } |
|
982 |
|
983 // Instruction bits passed off to the VM |
|
984 int code_size() { return _method_size; } |
|
985 CodeBuffer* code_buffer() { return &_code_buffer; } |
|
986 int first_block_size() { return _first_block_size; } |
|
987 void set_frame_complete(int off) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } |
|
988 ExceptionHandlerTable* handler_table() { return &_handler_table; } |
|
989 ImplicitExceptionTable* inc_table() { return &_inc_table; } |
|
990 OopMapSet* oop_map_set() { return _oop_map_set; } |
|
991 DebugInformationRecorder* debug_info() { return env()->debug_info(); } |
|
992 Dependencies* dependencies() { return env()->dependencies(); } |
|
993 static int CompiledZap_count() { return _CompiledZap_count; } |
|
994 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; } |
|
995 void init_scratch_buffer_blob(int const_size); |
|
996 void clear_scratch_buffer_blob(); |
|
997 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; } |
|
998 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; } |
|
999 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; } |
|
1000 |
|
1001 // emit to scratch blob, report resulting size |
|
1002 uint scratch_emit_size(const Node* n); |
|
1003 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; } |
|
1004 bool in_scratch_emit_size() const { return _in_scratch_emit_size; } |
|
1005 |
|
1006 enum ScratchBufferBlob { |
|
1007 MAX_inst_size = 1024, |
|
1008 MAX_locs_size = 128, // number of relocInfo elements |
|
1009 MAX_const_size = 128, |
|
1010 MAX_stubs_size = 128 |
|
1011 }; |
|
1012 |
|
1013 // Major entry point. Given a Scope, compile the associated method. |
|
1014 // For normal compilations, entry_bci is InvocationEntryBci. For on stack |
|
1015 // replacement, entry_bci indicates the bytecode for which to compile a |
|
1016 // continuation. |
|
1017 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, |
|
1018 int entry_bci, bool subsume_loads, bool do_escape_analysis, |
|
1019 bool eliminate_boxing); |
|
1020 |
|
1021 // Second major entry point. From the TypeFunc signature, generate code |
|
1022 // to pass arguments from the Java calling convention to the C calling |
|
1023 // convention. |
|
1024 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(), |
|
1025 address stub_function, const char *stub_name, |
|
1026 int is_fancy_jump, bool pass_tls, |
|
1027 bool save_arg_registers, bool return_pc); |
|
1028 |
|
1029 // From the TypeFunc signature, generate code to pass arguments |
|
1030 // from Compiled calling convention to Interpreter's calling convention |
|
1031 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry); |
|
1032 |
|
1033 // From the TypeFunc signature, generate code to pass arguments |
|
1034 // from Interpreter's calling convention to Compiler's calling convention |
|
1035 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf); |
|
1036 |
|
1037 // Are we compiling a method? |
|
1038 bool has_method() { return method() != NULL; } |
|
1039 |
|
1040 // Maybe print some information about this compile. |
|
1041 void print_compile_messages(); |
|
1042 |
|
1043 // Final graph reshaping, a post-pass after the regular optimizer is done. |
|
1044 bool final_graph_reshaping(); |
|
1045 |
|
1046 // returns true if adr is completely contained in the given alias category |
|
1047 bool must_alias(const TypePtr* adr, int alias_idx); |
|
1048 |
|
1049 // returns true if adr overlaps with the given alias category |
|
1050 bool can_alias(const TypePtr* adr, int alias_idx); |
|
1051 |
|
1052 // Driver for converting compiler's IR into machine code bits |
|
1053 void Output(); |
|
1054 |
|
1055 // Accessors for node bundling info. |
|
1056 Bundle* node_bundling(const Node *n); |
|
1057 bool valid_bundle_info(const Node *n); |
|
1058 |
|
1059 // Schedule and Bundle the instructions |
|
1060 void ScheduleAndBundle(); |
|
1061 |
|
1062 // Build OopMaps for each GC point |
|
1063 void BuildOopMaps(); |
|
1064 |
|
1065 // Append debug info for the node "local" at safepoint node "sfpt" to the |
|
1066 // "array", May also consult and add to "objs", which describes the |
|
1067 // scalar-replaced objects. |
|
1068 void FillLocArray( int idx, MachSafePointNode* sfpt, |
|
1069 Node *local, GrowableArray<ScopeValue*> *array, |
|
1070 GrowableArray<ScopeValue*> *objs ); |
|
1071 |
|
1072 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL. |
|
1073 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id); |
|
1074 // Requres that "objs" does not contains an ObjectValue whose id matches |
|
1075 // that of "sv. Appends "sv". |
|
1076 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs, |
|
1077 ObjectValue* sv ); |
|
1078 |
|
1079 // Process an OopMap Element while emitting nodes |
|
1080 void Process_OopMap_Node(MachNode *mach, int code_offset); |
|
1081 |
|
1082 // Initialize code buffer |
|
1083 CodeBuffer* init_buffer(uint* blk_starts); |
|
1084 |
|
1085 // Write out basic block data to code buffer |
|
1086 void fill_buffer(CodeBuffer* cb, uint* blk_starts); |
|
1087 |
|
1088 // Determine which variable sized branches can be shortened |
|
1089 void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size); |
|
1090 |
|
1091 // Compute the size of first NumberOfLoopInstrToAlign instructions |
|
1092 // at the head of a loop. |
|
1093 void compute_loop_first_inst_sizes(); |
|
1094 |
|
1095 // Compute the information for the exception tables |
|
1096 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels); |
|
1097 |
|
1098 // Stack slots that may be unused by the calling convention but must |
|
1099 // otherwise be preserved. On Intel this includes the return address. |
|
1100 // On PowerPC it includes the 4 words holding the old TOC & LR glue. |
|
1101 uint in_preserve_stack_slots(); |
|
1102 |
|
1103 // "Top of Stack" slots that may be unused by the calling convention but must |
|
1104 // otherwise be preserved. |
|
1105 // On Intel these are not necessary and the value can be zero. |
|
1106 // On Sparc this describes the words reserved for storing a register window |
|
1107 // when an interrupt occurs. |
|
1108 static uint out_preserve_stack_slots(); |
|
1109 |
|
1110 // Number of outgoing stack slots killed above the out_preserve_stack_slots |
|
1111 // for calls to C. Supports the var-args backing area for register parms. |
|
1112 uint varargs_C_out_slots_killed() const; |
|
1113 |
|
1114 // Number of Stack Slots consumed by a synchronization entry |
|
1115 int sync_stack_slots() const; |
|
1116 |
|
1117 // Compute the name of old_SP. See <arch>.ad for frame layout. |
|
1118 OptoReg::Name compute_old_SP(); |
|
1119 |
|
1120 #ifdef ENABLE_ZAP_DEAD_LOCALS |
|
1121 static bool is_node_getting_a_safepoint(Node*); |
|
1122 void Insert_zap_nodes(); |
|
1123 Node* call_zap_node(MachSafePointNode* n, int block_no); |
|
1124 #endif |
|
1125 |
|
1126 private: |
|
1127 // Phase control: |
|
1128 void Init(int aliaslevel); // Prepare for a single compilation |
|
1129 int Inline_Warm(); // Find more inlining work. |
|
1130 void Finish_Warm(); // Give up on further inlines. |
|
1131 void Optimize(); // Given a graph, optimize it |
|
1132 void Code_Gen(); // Generate code from a graph |
|
1133 |
|
1134 // Management of the AliasType table. |
|
1135 void grow_alias_types(); |
|
1136 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type); |
|
1137 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const; |
|
1138 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field); |
|
1139 |
|
1140 void verify_top(Node*) const PRODUCT_RETURN; |
|
1141 |
|
1142 // Intrinsic setup. |
|
1143 void register_library_intrinsics(); // initializer |
|
1144 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor |
|
1145 int intrinsic_insertion_index(ciMethod* m, bool is_virtual); // helper |
|
1146 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn |
|
1147 void register_intrinsic(CallGenerator* cg); // update fn |
|
1148 |
|
1149 #ifndef PRODUCT |
|
1150 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT]; |
|
1151 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT]; |
|
1152 #endif |
|
1153 // Function calls made by the public function final_graph_reshaping. |
|
1154 // No need to be made public as they are not called elsewhere. |
|
1155 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc); |
|
1156 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ); |
|
1157 void eliminate_redundant_card_marks(Node* n); |
|
1158 |
|
1159 public: |
|
1160 |
|
1161 // Note: Histogram array size is about 1 Kb. |
|
1162 enum { // flag bits: |
|
1163 _intrinsic_worked = 1, // succeeded at least once |
|
1164 _intrinsic_failed = 2, // tried it but it failed |
|
1165 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps) |
|
1166 _intrinsic_virtual = 8, // was seen in the virtual form (rare) |
|
1167 _intrinsic_both = 16 // was seen in the non-virtual form (usual) |
|
1168 }; |
|
1169 // Update histogram. Return boolean if this is a first-time occurrence. |
|
1170 static bool gather_intrinsic_statistics(vmIntrinsics::ID id, |
|
1171 bool is_virtual, int flags) PRODUCT_RETURN0; |
|
1172 static void print_intrinsic_statistics() PRODUCT_RETURN; |
|
1173 |
|
1174 // Graph verification code |
|
1175 // Walk the node list, verifying that there is a one-to-one |
|
1176 // correspondence between Use-Def edges and Def-Use edges |
|
1177 // The option no_dead_code enables stronger checks that the |
|
1178 // graph is strongly connected from root in both directions. |
|
1179 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN; |
|
1180 |
|
1181 // Verify GC barrier patterns |
|
1182 void verify_barriers() PRODUCT_RETURN; |
|
1183 |
|
1184 // End-of-run dumps. |
|
1185 static void print_statistics() PRODUCT_RETURN; |
|
1186 |
|
1187 // Dump formatted assembly |
|
1188 void dump_asm(int *pcs = NULL, uint pc_limit = 0) PRODUCT_RETURN; |
|
1189 void dump_pc(int *pcs, int pc_limit, Node *n); |
|
1190 |
|
1191 // Verify ADLC assumptions during startup |
|
1192 static void adlc_verification() PRODUCT_RETURN; |
|
1193 |
|
1194 // Definitions of pd methods |
|
1195 static void pd_compiler2_init(); |
|
1196 |
|
1197 // Auxiliary method for randomized fuzzing/stressing |
|
1198 static bool randomized_select(int count); |
|
1199 |
|
1200 // enter a PreserveJVMState block |
|
1201 void inc_preserve_jvm_state() { |
|
1202 _preserve_jvm_state++; |
|
1203 } |
|
1204 |
|
1205 // exit a PreserveJVMState block |
|
1206 void dec_preserve_jvm_state() { |
|
1207 _preserve_jvm_state--; |
|
1208 assert(_preserve_jvm_state >= 0, "_preserve_jvm_state shouldn't be negative"); |
|
1209 } |
|
1210 |
|
1211 bool has_preserve_jvm_state() const { |
|
1212 return _preserve_jvm_state > 0; |
|
1213 } |
|
1214 }; |
|
1215 |
|
1216 #endif // SHARE_VM_OPTO_COMPILE_HPP |