1 /*
2 * Copyright (c) 1998, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "ci/ciMethodData.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileLog.hpp"
30 #include "interpreter/linkResolver.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "memory/universe.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "opto/addnode.hpp"
35 #include "opto/castnode.hpp"
36 #include "opto/convertnode.hpp"
37 #include "opto/divnode.hpp"
38 #include "opto/idealGraphPrinter.hpp"
39 #include "opto/idealKit.hpp"
40 #include "opto/matcher.hpp"
41 #include "opto/memnode.hpp"
42 #include "opto/mulnode.hpp"
43 #include "opto/opaquenode.hpp"
44 #include "opto/parse.hpp"
45 #include "opto/runtime.hpp"
46 #include "opto/valuetypenode.hpp"
47 #include "runtime/deoptimization.hpp"
48 #include "runtime/sharedRuntime.hpp"
49
50 #ifndef PRODUCT
51 extern int explicit_null_checks_inserted,
52 explicit_null_checks_elided;
53 #endif
54
55 //---------------------------------array_load----------------------------------
56 void Parse::array_load(BasicType bt) {
57 const Type* elemtype = Type::TOP;
58 Node* adr = array_addressing(bt, 0, &elemtype);
59 if (stopped()) return; // guaranteed null or range check
60
61 Node* idx = pop();
62 Node* ary = pop();
63
64 // Handle value type arrays
65 const TypeOopPtr* elemptr = elemtype->make_oopptr();
66 const TypeAryPtr* ary_t = _gvn.type(ary)->is_aryptr();
67 if (elemtype->isa_valuetype() != NULL) {
68 C->set_flattened_accesses();
69 // Load from flattened value type array
70 Node* vt = ValueTypeNode::make_from_flattened(this, elemtype->value_klass(), ary, adr);
71 push(vt);
72 return;
73 } else if (elemptr != NULL && elemptr->is_valuetypeptr() && !elemptr->maybe_null()) {
74 // Load from non-flattened but flattenable value type array (elements can never be null)
75 bt = T_VALUETYPE;
76 } else if (!ary_t->is_not_flat() && !ary_t->klass_is_exact()) {
77 // Cannot statically determine if array is flattened, emit runtime check
78 assert(ValueArrayFlatten && elemptr != NULL && elemptr->can_be_value_type() &&
79 (!elemptr->is_valuetypeptr() || elemptr->value_klass()->flatten_array()), "array can't be flattened");
80 Node* ctl = control();
81 IdealKit ideal(this);
82 IdealVariable res(ideal);
83 ideal.declarations_done();
84 Node* kls = load_object_klass(ary);
85 Node* tag = load_lh_array_tag(kls);
86 ideal.if_then(tag, BoolTest::ne, intcon(Klass::_lh_array_tag_vt_value)); {
87 // non-flattened
88 sync_kit(ideal);
89 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
90 Node* ld = access_load_at(ary, adr, adr_type, elemptr, bt,
91 IN_HEAP | IS_ARRAY | C2_CONTROL_DEPENDENT_LOAD, ctl);
92 ideal.sync_kit(this);
93 ideal.set(res, ld);
94 } ideal.else_(); {
95 // flattened
96 sync_kit(ideal);
97 if (elemptr->is_valuetypeptr()) {
98 // Element type is known, cast and load from flattened representation
99 assert(elemptr->maybe_null(), "must be nullable");
100 ciValueKlass* vk = elemptr->value_klass();
101 assert(vk->flatten_array(), "must be flattenable");
102 ciArrayKlass* array_klass = ciArrayKlass::make(vk, /* never_null */ true);
103 const TypeAryPtr* arytype = TypeOopPtr::make_from_klass(array_klass)->isa_aryptr();
104 Node* cast = _gvn.transform(new CheckCastPPNode(control(), ary, arytype));
105 adr = array_element_address(cast, idx, T_VALUETYPE, ary_t->size(), control());
106 Node* vt = ValueTypeNode::make_from_flattened(this, vk, cast, adr)->allocate(this, false, false)->get_oop();
107 ideal.set(res, vt);
108 ideal.sync_kit(this);
109 } else {
110 // Element type is unknown, emit runtime call
111 Node* k_adr = basic_plus_adr(kls, in_bytes(ArrayKlass::element_klass_offset()));
112 Node* elem_klass = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), k_adr, TypeInstPtr::KLASS));
113 Node* obj_size = NULL;
114 kill_dead_locals();
115 inc_sp(2);
116 Node* alloc_obj = new_instance(elem_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
117 dec_sp(2);
118
119 AllocateNode* alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
120 assert(alloc->maybe_set_complete(&_gvn), "");
121 alloc->initialization()->set_complete_with_arraycopy();
122
123 // This membar keeps this access to an unknown flattened array
124 // correctly ordered with other unknown and known flattened
125 // array accesses.
126 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES));
127
128 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
129 // Unknown value type might contain reference fields
130 if (!bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing)) {
131 int base_off = sizeof(instanceOopDesc);
132 Node* dst_base = basic_plus_adr(alloc_obj, base_off);
133 Node* countx = obj_size;
134 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
135 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong)));
136
137 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
138 Node* lhp = basic_plus_adr(kls, in_bytes(Klass::layout_helper_offset()));
139 Node* elem_shift = make_load(NULL, lhp, TypeInt::INT, T_INT, MemNode::unordered);
140 uint header = arrayOopDesc::base_offset_in_bytes(T_VALUETYPE);
141 Node* base = basic_plus_adr(ary, header);
142 idx = Compile::conv_I2X_index(&_gvn, idx, TypeInt::POS, control());
143 Node* scale = _gvn.transform(new LShiftXNode(idx, elem_shift));
144 Node* adr = basic_plus_adr(ary, base, scale);
145
146 access_clone(adr, dst_base, countx, false);
147 } else {
148 ideal.sync_kit(this);
149 ideal.make_leaf_call(OptoRuntime::load_unknown_value_Type(),
150 CAST_FROM_FN_PTR(address, OptoRuntime::load_unknown_value),
151 "load_unknown_value",
152 ary, idx, alloc_obj);
153 sync_kit(ideal);
154 }
155
156 // This makes sure no other thread sees a partially initialized buffered value
157 insert_mem_bar_volatile(Op_MemBarStoreStore, Compile::AliasIdxRaw, alloc->proj_out_or_null(AllocateNode::RawAddress));
158
159 // Same as MemBarCPUOrder above: keep this unknown flattened
160 // array access correctly ordered with other flattened array
161 // access
162 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES));
163
164 // Prevent any use of the newly allocated value before it is
165 // fully initialized
166 alloc_obj = new CastPPNode(alloc_obj, _gvn.type(alloc_obj), true);
167 alloc_obj->set_req(0, control());
168 alloc_obj = _gvn.transform(alloc_obj);
169
170 ideal.sync_kit(this);
171
172 ideal.set(res, alloc_obj);
173 }
174 } ideal.end_if();
175 sync_kit(ideal);
176 push_node(bt, _gvn.transform(ideal.value(res)));
177 return;
178 }
179
180 if (elemtype == TypeInt::BOOL) {
181 bt = T_BOOLEAN;
182 } else if (bt == T_OBJECT) {
183 elemtype = ary_t->elem()->make_oopptr();
184 }
185
186 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
187 Node* ld = access_load_at(ary, adr, adr_type, elemtype, bt,
188 IN_HEAP | IS_ARRAY | C2_CONTROL_DEPENDENT_LOAD);
189 if (bt == T_VALUETYPE) {
190 // Loading a non-flattened (but flattenable) value type from an array
191 assert(!gvn().type(ld)->maybe_null(), "value type array elements should never be null");
192 if (elemptr->value_klass()->is_scalarizable()) {
193 ld = ValueTypeNode::make_from_oop(this, ld, elemptr->value_klass());
194 }
195 }
196
197 push_node(bt, ld);
198 }
199
200
201 //--------------------------------array_store----------------------------------
202 void Parse::array_store(BasicType bt) {
203 const Type* elemtype = Type::TOP;
204 Node* adr = array_addressing(bt, type2size[bt], &elemtype);
205 if (stopped()) return; // guaranteed null or range check
206 Node* cast_val = NULL;
207 if (bt == T_OBJECT) {
208 cast_val = array_store_check();
209 if (stopped()) return;
210 }
211 Node* val = pop_node(bt); // Value to store
212 Node* idx = pop(); // Index in the array
213 Node* ary = pop(); // The array itself
214
215 const TypeAryPtr* ary_t = _gvn.type(ary)->is_aryptr();
216 if (bt == T_OBJECT) {
217 const TypeOopPtr* elemptr = elemtype->make_oopptr();
218 const Type* val_t = _gvn.type(cast_val);
219 if (elemtype->isa_valuetype() != NULL) {
220 C->set_flattened_accesses();
221 // Store to flattened value type array
222 if (!cast_val->is_ValueType()) {
223 inc_sp(3);
224 cast_val = null_check(cast_val);
225 if (stopped()) return;
226 dec_sp(3);
227 cast_val = ValueTypeNode::make_from_oop(this, cast_val, elemtype->value_klass());
228 }
229 cast_val->as_ValueType()->store_flattened(this, ary, adr);
230 return;
231 } else if (elemptr->is_valuetypeptr() && !elemptr->maybe_null()) {
232 // Store to non-flattened but flattenable value type array (elements can never be null)
233 if (!cast_val->is_ValueType() && val_t->maybe_null()) {
234 inc_sp(3);
235 cast_val = null_check(cast_val);
236 if (stopped()) return;
237 dec_sp(3);
238 }
239 } else if (elemptr->can_be_value_type() && !ary_t->klass_is_exact() &&
240 (cast_val->is_ValueType() || val_t == TypePtr::NULL_PTR || val_t->is_oopptr()->can_be_value_type())) {
241 // Cannot statically determine if array is flattened or null-free, emit runtime checks
242 ciValueKlass* vk = NULL;
243 // Try to determine the value klass
244 if (cast_val->is_ValueType()) {
245 vk = val_t->value_klass();
246 } else if (elemptr->is_valuetypeptr()) {
247 vk = elemptr->value_klass();
248 }
249 if (!ary_t->is_not_flat() && (vk == NULL || vk->flatten_array())) {
250 // Array might be flattened
251 assert(ValueArrayFlatten && !ary_t->is_not_null_free(), "a null-ok array can't be flattened");
252 IdealKit ideal(this);
253 Node* kls = load_object_klass(ary);
254 Node* layout_val = load_lh_array_tag(kls);
255 ideal.if_then(layout_val, BoolTest::ne, intcon(Klass::_lh_array_tag_vt_value));
256 {
257 // non-flattened
258 sync_kit(ideal);
259 gen_value_array_null_guard(ary, cast_val, 3);
260 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
261 elemtype = ary_t->elem()->make_oopptr();
262 access_store_at(ary, adr, adr_type, cast_val, elemtype, bt, MO_UNORDERED | IN_HEAP | IS_ARRAY, false, false);
263 ideal.sync_kit(this);
264 }
265 ideal.else_();
266 {
267 // flattened
268 if (!cast_val->is_ValueType() && val_t->maybe_null()) {
269 // Add null check
270 sync_kit(ideal);
271 Node* null_ctl = top();
272 cast_val = null_check_oop(cast_val, &null_ctl);
273 if (null_ctl != top()) {
274 PreserveJVMState pjvms(this);
275 inc_sp(3);
276 set_control(null_ctl);
277 uncommon_trap(Deoptimization::Reason_null_check, Deoptimization::Action_none);
278 dec_sp(3);
279 }
280 ideal.sync_kit(this);
281 }
282 if (vk != NULL && !stopped()) {
283 // Element type is known, cast and store to flattened representation
284 sync_kit(ideal);
285 assert(vk->flatten_array(), "must be flattenable");
286 assert(elemptr->maybe_null(), "must be nullable");
287 ciArrayKlass* array_klass = ciArrayKlass::make(vk, /* never_null */ true);
288 const TypeAryPtr* arytype = TypeOopPtr::make_from_klass(array_klass)->isa_aryptr();
289 ary = _gvn.transform(new CheckCastPPNode(control(), ary, arytype));
290 adr = array_element_address(ary, idx, T_OBJECT, arytype->size(), control());
291 if (!cast_val->is_ValueType()) {
292 assert(!gvn().type(cast_val)->maybe_null(), "value type array elements should never be null");
293 cast_val = ValueTypeNode::make_from_oop(this, cast_val, vk);
294 }
295 cast_val->as_ValueType()->store_flattened(this, ary, adr);
296 ideal.sync_kit(this);
297 } else if (!ideal.ctrl()->is_top()) {
298 // Element type is unknown, emit runtime call
299 sync_kit(ideal);
300
301 // This membar keeps this access to an unknown flattened
302 // array correctly ordered with other unknown and known
303 // flattened array accesses.
304 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES));
305 ideal.sync_kit(this);
306
307 ideal.make_leaf_call(OptoRuntime::store_unknown_value_Type(),
308 CAST_FROM_FN_PTR(address, OptoRuntime::store_unknown_value),
309 "store_unknown_value",
310 cast_val, ary, idx);
311
312 sync_kit(ideal);
313 // Same as MemBarCPUOrder above: keep this unknown
314 // flattened array access correctly ordered with other
315 // flattened array access
316 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES));
317 ideal.sync_kit(this);
318
319 }
320 }
321 ideal.end_if();
322 sync_kit(ideal);
323 return;
324 } else if (!ary_t->is_not_null_free()) {
325 // Array is never flattened
326 gen_value_array_null_guard(ary, cast_val, 3);
327 }
328 }
329 }
330
331 if (elemtype == TypeInt::BOOL) {
332 bt = T_BOOLEAN;
333 } else if (bt == T_OBJECT) {
334 elemtype = ary_t->elem()->make_oopptr();
335 }
336
337 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
338
339 access_store_at(ary, adr, adr_type, val, elemtype, bt, MO_UNORDERED | IN_HEAP | IS_ARRAY);
340 }
341
342
343 //------------------------------array_addressing-------------------------------
344 // Pull array and index from the stack. Compute pointer-to-element.
345 Node* Parse::array_addressing(BasicType type, int vals, const Type* *result2) {
346 Node *idx = peek(0+vals); // Get from stack without popping
347 Node *ary = peek(1+vals); // in case of exception
348
349 // Null check the array base, with correct stack contents
350 ary = null_check(ary, T_ARRAY);
351 // Compile-time detect of null-exception?
352 if (stopped()) return top();
353
354 const TypeAryPtr* arytype = _gvn.type(ary)->is_aryptr();
355 const TypeInt* sizetype = arytype->size();
356 const Type* elemtype = arytype->elem();
357
358 if (UseUniqueSubclasses && result2 != NULL) {
359 const Type* el = elemtype->make_ptr();
360 if (el && el->isa_instptr()) {
361 const TypeInstPtr* toop = el->is_instptr();
362 if (toop->klass()->as_instance_klass()->unique_concrete_subklass()) {
363 // If we load from "AbstractClass[]" we must see "ConcreteSubClass".
364 const Type* subklass = Type::get_const_type(toop->klass());
365 elemtype = subklass->join_speculative(el);
366 }
367 }
368 }
369
370 // Check for big class initializers with all constant offsets
371 // feeding into a known-size array.
372 const TypeInt* idxtype = _gvn.type(idx)->is_int();
373 // See if the highest idx value is less than the lowest array bound,
374 // and if the idx value cannot be negative:
375 bool need_range_check = true;
376 if (idxtype->_hi < sizetype->_lo && idxtype->_lo >= 0) {
377 need_range_check = false;
378 if (C->log() != NULL) C->log()->elem("observe that='!need_range_check'");
379 }
380
381 ciKlass * arytype_klass = arytype->klass();
382 if ((arytype_klass != NULL) && (!arytype_klass->is_loaded())) {
383 // Only fails for some -Xcomp runs
384 // The class is unloaded. We have to run this bytecode in the interpreter.
385 uncommon_trap(Deoptimization::Reason_unloaded,
386 Deoptimization::Action_reinterpret,
387 arytype->klass(), "!loaded array");
388 return top();
389 }
390
391 // Do the range check
392 if (GenerateRangeChecks && need_range_check) {
393 Node* tst;
394 if (sizetype->_hi <= 0) {
395 // The greatest array bound is negative, so we can conclude that we're
396 // compiling unreachable code, but the unsigned compare trick used below
397 // only works with non-negative lengths. Instead, hack "tst" to be zero so
398 // the uncommon_trap path will always be taken.
399 tst = _gvn.intcon(0);
400 } else {
401 // Range is constant in array-oop, so we can use the original state of mem
402 Node* len = load_array_length(ary);
403
404 // Test length vs index (standard trick using unsigned compare)
405 Node* chk = _gvn.transform( new CmpUNode(idx, len) );
406 BoolTest::mask btest = BoolTest::lt;
407 tst = _gvn.transform( new BoolNode(chk, btest) );
408 }
409 RangeCheckNode* rc = new RangeCheckNode(control(), tst, PROB_MAX, COUNT_UNKNOWN);
410 _gvn.set_type(rc, rc->Value(&_gvn));
411 if (!tst->is_Con()) {
412 record_for_igvn(rc);
413 }
414 set_control(_gvn.transform(new IfTrueNode(rc)));
415 // Branch to failure if out of bounds
416 {
417 PreserveJVMState pjvms(this);
418 set_control(_gvn.transform(new IfFalseNode(rc)));
419 if (C->allow_range_check_smearing()) {
420 // Do not use builtin_throw, since range checks are sometimes
421 // made more stringent by an optimistic transformation.
422 // This creates "tentative" range checks at this point,
423 // which are not guaranteed to throw exceptions.
424 // See IfNode::Ideal, is_range_check, adjust_check.
425 uncommon_trap(Deoptimization::Reason_range_check,
426 Deoptimization::Action_make_not_entrant,
427 NULL, "range_check");
428 } else {
429 // If we have already recompiled with the range-check-widening
430 // heroic optimization turned off, then we must really be throwing
431 // range check exceptions.
432 builtin_throw(Deoptimization::Reason_range_check, idx);
433 }
434 }
435 }
436 // Check for always knowing you are throwing a range-check exception
437 if (stopped()) return top();
438
439 // Speculate on the array not being null-free
440 if (!arytype->is_not_null_free() && arytype->speculative() != NULL && arytype->speculative()->isa_aryptr() != NULL &&
441 arytype->speculative()->is_aryptr()->is_not_null_free() &&
442 !too_many_traps_or_recompiles(Deoptimization::Reason_speculate_class_check)) {
443 Node* tst = gen_null_free_array_check(ary);
444 {
445 BuildCutout unless(this, tst, PROB_ALWAYS);
446 uncommon_trap(Deoptimization::Reason_speculate_class_check,
447 Deoptimization::Action_maybe_recompile);
448 }
449 Node* cast = new CheckCastPPNode(control(), ary, arytype->cast_to_not_null_free());
450 replace_in_map(ary, _gvn.transform(cast));
451 }
452
453 // Make array address computation control dependent to prevent it
454 // from floating above the range check during loop optimizations.
455 Node* ptr = array_element_address(ary, idx, type, sizetype, control());
456
457 if (result2 != NULL) *result2 = elemtype;
458
459 assert(ptr != top(), "top should go hand-in-hand with stopped");
460
461 return ptr;
462 }
463
464
465 // returns IfNode
466 IfNode* Parse::jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask, float prob, float cnt) {
467 Node *cmp = _gvn.transform(new CmpINode(a, b)); // two cases: shiftcount > 32 and shiftcount <= 32
468 Node *tst = _gvn.transform(new BoolNode(cmp, mask));
469 IfNode *iff = create_and_map_if(control(), tst, prob, cnt);
470 return iff;
471 }
472
473 // return Region node
474 Node* Parse::jump_if_join(Node* iffalse, Node* iftrue) {
475 Node *region = new RegionNode(3); // 2 results
476 record_for_igvn(region);
477 region->init_req(1, iffalse);
478 region->init_req(2, iftrue );
479 _gvn.set_type(region, Type::CONTROL);
480 region = _gvn.transform(region);
481 set_control (region);
482 return region;
483 }
484
485 // sentinel value for the target bci to mark never taken branches
486 // (according to profiling)
487 static const int never_reached = INT_MAX;
488
489 //------------------------------helper for tableswitch-------------------------
490 void Parse::jump_if_true_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index, bool unc) {
491 // True branch, use existing map info
492 { PreserveJVMState pjvms(this);
493 Node *iftrue = _gvn.transform( new IfTrueNode (iff) );
494 set_control( iftrue );
495 if (unc) {
496 repush_if_args();
497 uncommon_trap(Deoptimization::Reason_unstable_if,
498 Deoptimization::Action_reinterpret,
499 NULL,
500 "taken always");
501 } else {
502 assert(dest_bci_if_true != never_reached, "inconsistent dest");
503 profile_switch_case(prof_table_index);
504 merge_new_path(dest_bci_if_true);
505 }
506 }
507
508 // False branch
509 Node *iffalse = _gvn.transform( new IfFalseNode(iff) );
510 set_control( iffalse );
511 }
512
513 void Parse::jump_if_false_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index, bool unc) {
514 // True branch, use existing map info
515 { PreserveJVMState pjvms(this);
516 Node *iffalse = _gvn.transform( new IfFalseNode (iff) );
517 set_control( iffalse );
518 if (unc) {
519 repush_if_args();
520 uncommon_trap(Deoptimization::Reason_unstable_if,
521 Deoptimization::Action_reinterpret,
522 NULL,
523 "taken never");
524 } else {
525 assert(dest_bci_if_true != never_reached, "inconsistent dest");
526 profile_switch_case(prof_table_index);
527 merge_new_path(dest_bci_if_true);
528 }
529 }
530
531 // False branch
532 Node *iftrue = _gvn.transform( new IfTrueNode(iff) );
533 set_control( iftrue );
534 }
535
536 void Parse::jump_if_always_fork(int dest_bci, int prof_table_index, bool unc) {
537 // False branch, use existing map and control()
538 if (unc) {
539 repush_if_args();
540 uncommon_trap(Deoptimization::Reason_unstable_if,
541 Deoptimization::Action_reinterpret,
542 NULL,
543 "taken never");
544 } else {
545 assert(dest_bci != never_reached, "inconsistent dest");
546 profile_switch_case(prof_table_index);
547 merge_new_path(dest_bci);
548 }
549 }
550
551
552 extern "C" {
553 static int jint_cmp(const void *i, const void *j) {
554 int a = *(jint *)i;
555 int b = *(jint *)j;
556 return a > b ? 1 : a < b ? -1 : 0;
557 }
558 }
559
560
561 // Default value for methodData switch indexing. Must be a negative value to avoid
562 // conflict with any legal switch index.
563 #define NullTableIndex -1
564
565 class SwitchRange : public StackObj {
566 // a range of integers coupled with a bci destination
567 jint _lo; // inclusive lower limit
568 jint _hi; // inclusive upper limit
569 int _dest;
570 int _table_index; // index into method data table
571 float _cnt; // how many times this range was hit according to profiling
572
573 public:
574 jint lo() const { return _lo; }
575 jint hi() const { return _hi; }
576 int dest() const { return _dest; }
577 int table_index() const { return _table_index; }
578 bool is_singleton() const { return _lo == _hi; }
579 float cnt() const { return _cnt; }
580
581 void setRange(jint lo, jint hi, int dest, int table_index, float cnt) {
582 assert(lo <= hi, "must be a non-empty range");
583 _lo = lo, _hi = hi; _dest = dest; _table_index = table_index; _cnt = cnt;
584 assert(_cnt >= 0, "");
585 }
586 bool adjoinRange(jint lo, jint hi, int dest, int table_index, float cnt, bool trim_ranges) {
587 assert(lo <= hi, "must be a non-empty range");
588 if (lo == _hi+1 && table_index == _table_index) {
589 // see merge_ranges() comment below
590 if (trim_ranges) {
591 if (cnt == 0) {
592 if (_cnt != 0) {
593 return false;
594 }
595 if (dest != _dest) {
596 _dest = never_reached;
597 }
598 } else {
599 if (_cnt == 0) {
600 return false;
601 }
602 if (dest != _dest) {
603 return false;
604 }
605 }
606 } else {
607 if (dest != _dest) {
608 return false;
609 }
610 }
611 _hi = hi;
612 _cnt += cnt;
613 return true;
614 }
615 return false;
616 }
617
618 void set (jint value, int dest, int table_index, float cnt) {
619 setRange(value, value, dest, table_index, cnt);
620 }
621 bool adjoin(jint value, int dest, int table_index, float cnt, bool trim_ranges) {
622 return adjoinRange(value, value, dest, table_index, cnt, trim_ranges);
623 }
624 bool adjoin(SwitchRange& other) {
625 return adjoinRange(other._lo, other._hi, other._dest, other._table_index, other._cnt, false);
626 }
627
628 void print() {
629 if (is_singleton())
630 tty->print(" {%d}=>%d (cnt=%f)", lo(), dest(), cnt());
631 else if (lo() == min_jint)
632 tty->print(" {..%d}=>%d (cnt=%f)", hi(), dest(), cnt());
633 else if (hi() == max_jint)
634 tty->print(" {%d..}=>%d (cnt=%f)", lo(), dest(), cnt());
635 else
636 tty->print(" {%d..%d}=>%d (cnt=%f)", lo(), hi(), dest(), cnt());
637 }
638 };
639
640 // We try to minimize the number of ranges and the size of the taken
641 // ones using profiling data. When ranges are created,
642 // SwitchRange::adjoinRange() only allows 2 adjoining ranges to merge
643 // if both were never hit or both were hit to build longer unreached
644 // ranges. Here, we now merge adjoining ranges with the same
645 // destination and finally set destination of unreached ranges to the
646 // special value never_reached because it can help minimize the number
647 // of tests that are necessary.
648 //
649 // For instance:
650 // [0, 1] to target1 sometimes taken
651 // [1, 2] to target1 never taken
652 // [2, 3] to target2 never taken
653 // would lead to:
654 // [0, 1] to target1 sometimes taken
655 // [1, 3] never taken
656 //
657 // (first 2 ranges to target1 are not merged)
658 static void merge_ranges(SwitchRange* ranges, int& rp) {
659 if (rp == 0) {
660 return;
661 }
662 int shift = 0;
663 for (int j = 0; j < rp; j++) {
664 SwitchRange& r1 = ranges[j-shift];
665 SwitchRange& r2 = ranges[j+1];
666 if (r1.adjoin(r2)) {
667 shift++;
668 } else if (shift > 0) {
669 ranges[j+1-shift] = r2;
670 }
671 }
672 rp -= shift;
673 for (int j = 0; j <= rp; j++) {
674 SwitchRange& r = ranges[j];
675 if (r.cnt() == 0 && r.dest() != never_reached) {
676 r.setRange(r.lo(), r.hi(), never_reached, r.table_index(), r.cnt());
677 }
678 }
679 }
680
681 //-------------------------------do_tableswitch--------------------------------
682 void Parse::do_tableswitch() {
683 Node* lookup = pop();
684 // Get information about tableswitch
685 int default_dest = iter().get_dest_table(0);
686 int lo_index = iter().get_int_table(1);
687 int hi_index = iter().get_int_table(2);
688 int len = hi_index - lo_index + 1;
689
690 if (len < 1) {
691 // If this is a backward branch, add safepoint
692 maybe_add_safepoint(default_dest);
693 merge(default_dest);
694 return;
695 }
696
697 ciMethodData* methodData = method()->method_data();
698 ciMultiBranchData* profile = NULL;
699 if (methodData->is_mature() && UseSwitchProfiling) {
700 ciProfileData* data = methodData->bci_to_data(bci());
701 if (data != NULL && data->is_MultiBranchData()) {
702 profile = (ciMultiBranchData*)data;
703 }
704 }
705 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
706
707 // generate decision tree, using trichotomy when possible
708 int rnum = len+2;
709 bool makes_backward_branch = false;
710 SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
711 int rp = -1;
712 if (lo_index != min_jint) {
713 uint cnt = 1;
714 if (profile != NULL) {
715 cnt = profile->default_count() / (hi_index != max_jint ? 2 : 1);
716 }
717 ranges[++rp].setRange(min_jint, lo_index-1, default_dest, NullTableIndex, cnt);
718 }
719 for (int j = 0; j < len; j++) {
720 jint match_int = lo_index+j;
721 int dest = iter().get_dest_table(j+3);
722 makes_backward_branch |= (dest <= bci());
723 int table_index = method_data_update() ? j : NullTableIndex;
724 uint cnt = 1;
725 if (profile != NULL) {
726 cnt = profile->count_at(j);
727 }
728 if (rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index, cnt, trim_ranges)) {
729 ranges[++rp].set(match_int, dest, table_index, cnt);
730 }
731 }
732 jint highest = lo_index+(len-1);
733 assert(ranges[rp].hi() == highest, "");
734 if (highest != max_jint) {
735 uint cnt = 1;
736 if (profile != NULL) {
737 cnt = profile->default_count() / (lo_index != min_jint ? 2 : 1);
738 }
739 if (!ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex, cnt, trim_ranges)) {
740 ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex, cnt);
741 }
742 }
743 assert(rp < len+2, "not too many ranges");
744
745 if (trim_ranges) {
746 merge_ranges(ranges, rp);
747 }
748
749 // Safepoint in case if backward branch observed
750 if( makes_backward_branch && UseLoopSafepoints )
751 add_safepoint();
752
753 jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
754 }
755
756
757 //------------------------------do_lookupswitch--------------------------------
758 void Parse::do_lookupswitch() {
759 Node *lookup = pop(); // lookup value
760 // Get information about lookupswitch
761 int default_dest = iter().get_dest_table(0);
762 int len = iter().get_int_table(1);
763
764 if (len < 1) { // If this is a backward branch, add safepoint
765 maybe_add_safepoint(default_dest);
766 merge(default_dest);
767 return;
768 }
769
770 ciMethodData* methodData = method()->method_data();
771 ciMultiBranchData* profile = NULL;
772 if (methodData->is_mature() && UseSwitchProfiling) {
773 ciProfileData* data = methodData->bci_to_data(bci());
774 if (data != NULL && data->is_MultiBranchData()) {
775 profile = (ciMultiBranchData*)data;
776 }
777 }
778 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
779
780 // generate decision tree, using trichotomy when possible
781 jint* table = NEW_RESOURCE_ARRAY(jint, len*3);
782 {
783 for (int j = 0; j < len; j++) {
784 table[3*j+0] = iter().get_int_table(2+2*j);
785 table[3*j+1] = iter().get_dest_table(2+2*j+1);
786 table[3*j+2] = profile == NULL ? 1 : profile->count_at(j);
787 }
788 qsort(table, len, 3*sizeof(table[0]), jint_cmp);
789 }
790
791 float defaults = 0;
792 jint prev = min_jint;
793 for (int j = 0; j < len; j++) {
794 jint match_int = table[3*j+0];
795 if (match_int != prev) {
796 defaults += (float)match_int - prev;
797 }
798 prev = match_int+1;
799 }
800 if (prev-1 != max_jint) {
801 defaults += (float)max_jint - prev + 1;
802 }
803 float default_cnt = 1;
804 if (profile != NULL) {
805 default_cnt = profile->default_count()/defaults;
806 }
807
808 int rnum = len*2+1;
809 bool makes_backward_branch = false;
810 SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
811 int rp = -1;
812 for (int j = 0; j < len; j++) {
813 jint match_int = table[3*j+0];
814 int dest = table[3*j+1];
815 int cnt = table[3*j+2];
816 int next_lo = rp < 0 ? min_jint : ranges[rp].hi()+1;
817 int table_index = method_data_update() ? j : NullTableIndex;
818 makes_backward_branch |= (dest <= bci());
819 float c = default_cnt * ((float)match_int - next_lo);
820 if (match_int != next_lo && (rp < 0 || !ranges[rp].adjoinRange(next_lo, match_int-1, default_dest, NullTableIndex, c, trim_ranges))) {
821 assert(default_dest != never_reached, "sentinel value for dead destinations");
822 ranges[++rp].setRange(next_lo, match_int-1, default_dest, NullTableIndex, c);
823 }
824 if (rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index, cnt, trim_ranges)) {
825 assert(dest != never_reached, "sentinel value for dead destinations");
826 ranges[++rp].set(match_int, dest, table_index, cnt);
827 }
828 }
829 jint highest = table[3*(len-1)];
830 assert(ranges[rp].hi() == highest, "");
831 if (highest != max_jint &&
832 !ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex, default_cnt * ((float)max_jint - highest), trim_ranges)) {
833 ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex, default_cnt * ((float)max_jint - highest));
834 }
835 assert(rp < rnum, "not too many ranges");
836
837 if (trim_ranges) {
838 merge_ranges(ranges, rp);
839 }
840
841 // Safepoint in case backward branch observed
842 if (makes_backward_branch && UseLoopSafepoints)
843 add_safepoint();
844
845 jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
846 }
847
848 static float if_prob(float taken_cnt, float total_cnt) {
849 assert(taken_cnt <= total_cnt, "");
850 if (total_cnt == 0) {
851 return PROB_FAIR;
852 }
853 float p = taken_cnt / total_cnt;
854 return MIN2(MAX2(p, PROB_MIN), PROB_MAX);
855 }
856
857 static float if_cnt(float cnt) {
858 if (cnt == 0) {
859 return COUNT_UNKNOWN;
860 }
861 return cnt;
862 }
863
864 static float sum_of_cnts(SwitchRange *lo, SwitchRange *hi) {
865 float total_cnt = 0;
866 for (SwitchRange* sr = lo; sr <= hi; sr++) {
867 total_cnt += sr->cnt();
868 }
869 return total_cnt;
870 }
871
872 class SwitchRanges : public ResourceObj {
873 public:
874 SwitchRange* _lo;
875 SwitchRange* _hi;
876 SwitchRange* _mid;
877 float _cost;
878
879 enum {
880 Start,
881 LeftDone,
882 RightDone,
883 Done
884 } _state;
885
886 SwitchRanges(SwitchRange *lo, SwitchRange *hi)
887 : _lo(lo), _hi(hi), _mid(NULL),
888 _cost(0), _state(Start) {
889 }
890
891 SwitchRanges()
892 : _lo(NULL), _hi(NULL), _mid(NULL),
893 _cost(0), _state(Start) {}
894 };
895
896 // Estimate cost of performing a binary search on lo..hi
897 static float compute_tree_cost(SwitchRange *lo, SwitchRange *hi, float total_cnt) {
898 GrowableArray<SwitchRanges> tree;
899 SwitchRanges root(lo, hi);
900 tree.push(root);
901
902 float cost = 0;
903 do {
904 SwitchRanges& r = *tree.adr_at(tree.length()-1);
905 if (r._hi != r._lo) {
906 if (r._mid == NULL) {
907 float r_cnt = sum_of_cnts(r._lo, r._hi);
908
909 if (r_cnt == 0) {
910 tree.pop();
911 cost = 0;
912 continue;
913 }
914
915 SwitchRange* mid = NULL;
916 mid = r._lo;
917 for (float cnt = 0; ; ) {
918 assert(mid <= r._hi, "out of bounds");
919 cnt += mid->cnt();
920 if (cnt > r_cnt / 2) {
921 break;
922 }
923 mid++;
924 }
925 assert(mid <= r._hi, "out of bounds");
926 r._mid = mid;
927 r._cost = r_cnt / total_cnt;
928 }
929 r._cost += cost;
930 if (r._state < SwitchRanges::LeftDone && r._mid > r._lo) {
931 cost = 0;
932 r._state = SwitchRanges::LeftDone;
933 tree.push(SwitchRanges(r._lo, r._mid-1));
934 } else if (r._state < SwitchRanges::RightDone) {
935 cost = 0;
936 r._state = SwitchRanges::RightDone;
937 tree.push(SwitchRanges(r._mid == r._lo ? r._mid+1 : r._mid, r._hi));
938 } else {
939 tree.pop();
940 cost = r._cost;
941 }
942 } else {
943 tree.pop();
944 cost = r._cost;
945 }
946 } while (tree.length() > 0);
947
948
949 return cost;
950 }
951
952 // It sometimes pays off to test most common ranges before the binary search
953 void Parse::linear_search_switch_ranges(Node* key_val, SwitchRange*& lo, SwitchRange*& hi) {
954 uint nr = hi - lo + 1;
955 float total_cnt = sum_of_cnts(lo, hi);
956
957 float min = compute_tree_cost(lo, hi, total_cnt);
958 float extra = 1;
959 float sub = 0;
960
961 SwitchRange* array1 = lo;
962 SwitchRange* array2 = NEW_RESOURCE_ARRAY(SwitchRange, nr);
963
964 SwitchRange* ranges = NULL;
965
966 while (nr >= 2) {
967 assert(lo == array1 || lo == array2, "one the 2 already allocated arrays");
968 ranges = (lo == array1) ? array2 : array1;
969
970 // Find highest frequency range
971 SwitchRange* candidate = lo;
972 for (SwitchRange* sr = lo+1; sr <= hi; sr++) {
973 if (sr->cnt() > candidate->cnt()) {
974 candidate = sr;
975 }
976 }
977 SwitchRange most_freq = *candidate;
978 if (most_freq.cnt() == 0) {
979 break;
980 }
981
982 // Copy remaining ranges into another array
983 int shift = 0;
984 for (uint i = 0; i < nr; i++) {
985 SwitchRange* sr = &lo[i];
986 if (sr != candidate) {
987 ranges[i-shift] = *sr;
988 } else {
989 shift++;
990 if (i > 0 && i < nr-1) {
991 SwitchRange prev = lo[i-1];
992 prev.setRange(prev.lo(), sr->hi(), prev.dest(), prev.table_index(), prev.cnt());
993 if (prev.adjoin(lo[i+1])) {
994 shift++;
995 i++;
996 }
997 ranges[i-shift] = prev;
998 }
999 }
1000 }
1001 nr -= shift;
1002
1003 // Evaluate cost of testing the most common range and performing a
1004 // binary search on the other ranges
1005 float cost = extra + compute_tree_cost(&ranges[0], &ranges[nr-1], total_cnt);
1006 if (cost >= min) {
1007 break;
1008 }
1009 // swap arrays
1010 lo = &ranges[0];
1011 hi = &ranges[nr-1];
1012
1013 // It pays off: emit the test for the most common range
1014 assert(most_freq.cnt() > 0, "must be taken");
1015 Node* val = _gvn.transform(new SubINode(key_val, _gvn.intcon(most_freq.lo())));
1016 Node* cmp = _gvn.transform(new CmpUNode(val, _gvn.intcon(most_freq.hi() - most_freq.lo())));
1017 Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1018 IfNode* iff = create_and_map_if(control(), tst, if_prob(most_freq.cnt(), total_cnt), if_cnt(most_freq.cnt()));
1019 jump_if_true_fork(iff, most_freq.dest(), most_freq.table_index(), false);
1020
1021 sub += most_freq.cnt() / total_cnt;
1022 extra += 1 - sub;
1023 min = cost;
1024 }
1025 }
1026
1027 //----------------------------create_jump_tables-------------------------------
1028 bool Parse::create_jump_tables(Node* key_val, SwitchRange* lo, SwitchRange* hi) {
1029 // Are jumptables enabled
1030 if (!UseJumpTables) return false;
1031
1032 // Are jumptables supported
1033 if (!Matcher::has_match_rule(Op_Jump)) return false;
1034
1035 // Don't make jump table if profiling
1036 if (method_data_update()) return false;
1037
1038 bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
1039
1040 // Decide if a guard is needed to lop off big ranges at either (or
1041 // both) end(s) of the input set. We'll call this the default target
1042 // even though we can't be sure that it is the true "default".
1043
1044 bool needs_guard = false;
1045 int default_dest;
1046 int64_t total_outlier_size = 0;
1047 int64_t hi_size = ((int64_t)hi->hi()) - ((int64_t)hi->lo()) + 1;
1048 int64_t lo_size = ((int64_t)lo->hi()) - ((int64_t)lo->lo()) + 1;
1049
1050 if (lo->dest() == hi->dest()) {
1051 total_outlier_size = hi_size + lo_size;
1052 default_dest = lo->dest();
1053 } else if (lo_size > hi_size) {
1054 total_outlier_size = lo_size;
1055 default_dest = lo->dest();
1056 } else {
1057 total_outlier_size = hi_size;
1058 default_dest = hi->dest();
1059 }
1060
1061 float total = sum_of_cnts(lo, hi);
1062 float cost = compute_tree_cost(lo, hi, total);
1063
1064 // If a guard test will eliminate very sparse end ranges, then
1065 // it is worth the cost of an extra jump.
1066 float trimmed_cnt = 0;
1067 if (total_outlier_size > (MaxJumpTableSparseness * 4)) {
1068 needs_guard = true;
1069 if (default_dest == lo->dest()) {
1070 trimmed_cnt += lo->cnt();
1071 lo++;
1072 }
1073 if (default_dest == hi->dest()) {
1074 trimmed_cnt += hi->cnt();
1075 hi--;
1076 }
1077 }
1078
1079 // Find the total number of cases and ranges
1080 int64_t num_cases = ((int64_t)hi->hi()) - ((int64_t)lo->lo()) + 1;
1081 int num_range = hi - lo + 1;
1082
1083 // Don't create table if: too large, too small, or too sparse.
1084 if (num_cases > MaxJumpTableSize)
1085 return false;
1086 if (UseSwitchProfiling) {
1087 // MinJumpTableSize is set so with a well balanced binary tree,
1088 // when the number of ranges is MinJumpTableSize, it's cheaper to
1089 // go through a JumpNode that a tree of IfNodes. Average cost of a
1090 // tree of IfNodes with MinJumpTableSize is
1091 // log2f(MinJumpTableSize) comparisons. So if the cost computed
1092 // from profile data is less than log2f(MinJumpTableSize) then
1093 // going with the binary search is cheaper.
1094 if (cost < log2f(MinJumpTableSize)) {
1095 return false;
1096 }
1097 } else {
1098 if (num_cases < MinJumpTableSize)
1099 return false;
1100 }
1101 if (num_cases > (MaxJumpTableSparseness * num_range))
1102 return false;
1103
1104 // Normalize table lookups to zero
1105 int lowval = lo->lo();
1106 key_val = _gvn.transform( new SubINode(key_val, _gvn.intcon(lowval)) );
1107
1108 // Generate a guard to protect against input keyvals that aren't
1109 // in the switch domain.
1110 if (needs_guard) {
1111 Node* size = _gvn.intcon(num_cases);
1112 Node* cmp = _gvn.transform(new CmpUNode(key_val, size));
1113 Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::ge));
1114 IfNode* iff = create_and_map_if(control(), tst, if_prob(trimmed_cnt, total), if_cnt(trimmed_cnt));
1115 jump_if_true_fork(iff, default_dest, NullTableIndex, trim_ranges && trimmed_cnt == 0);
1116
1117 total -= trimmed_cnt;
1118 }
1119
1120 // Create an ideal node JumpTable that has projections
1121 // of all possible ranges for a switch statement
1122 // The key_val input must be converted to a pointer offset and scaled.
1123 // Compare Parse::array_addressing above.
1124
1125 // Clean the 32-bit int into a real 64-bit offset.
1126 // Otherwise, the jint value 0 might turn into an offset of 0x0800000000.
1127 const TypeInt* ikeytype = TypeInt::make(0, num_cases, Type::WidenMin);
1128 // Make I2L conversion control dependent to prevent it from
1129 // floating above the range check during loop optimizations.
1130 key_val = C->conv_I2X_index(&_gvn, key_val, ikeytype, control());
1131
1132 // Shift the value by wordsize so we have an index into the table, rather
1133 // than a switch value
1134 Node *shiftWord = _gvn.MakeConX(wordSize);
1135 key_val = _gvn.transform( new MulXNode( key_val, shiftWord));
1136
1137 // Create the JumpNode
1138 Arena* arena = C->comp_arena();
1139 float* probs = (float*)arena->Amalloc(sizeof(float)*num_cases);
1140 int i = 0;
1141 if (total == 0) {
1142 for (SwitchRange* r = lo; r <= hi; r++) {
1143 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
1144 probs[i] = 1.0F / num_cases;
1145 }
1146 }
1147 } else {
1148 for (SwitchRange* r = lo; r <= hi; r++) {
1149 float prob = r->cnt()/total;
1150 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
1151 probs[i] = prob / (r->hi() - r->lo() + 1);
1152 }
1153 }
1154 }
1155
1156 ciMethodData* methodData = method()->method_data();
1157 ciMultiBranchData* profile = NULL;
1158 if (methodData->is_mature()) {
1159 ciProfileData* data = methodData->bci_to_data(bci());
1160 if (data != NULL && data->is_MultiBranchData()) {
1161 profile = (ciMultiBranchData*)data;
1162 }
1163 }
1164
1165 Node* jtn = _gvn.transform(new JumpNode(control(), key_val, num_cases, probs, profile == NULL ? COUNT_UNKNOWN : total));
1166
1167 // These are the switch destinations hanging off the jumpnode
1168 i = 0;
1169 for (SwitchRange* r = lo; r <= hi; r++) {
1170 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
1171 Node* input = _gvn.transform(new JumpProjNode(jtn, i, r->dest(), (int)(j - lowval)));
1172 {
1173 PreserveJVMState pjvms(this);
1174 set_control(input);
1175 jump_if_always_fork(r->dest(), r->table_index(), trim_ranges && r->cnt() == 0);
1176 }
1177 }
1178 }
1179 assert(i == num_cases, "miscount of cases");
1180 stop_and_kill_map(); // no more uses for this JVMS
1181 return true;
1182 }
1183
1184 //----------------------------jump_switch_ranges-------------------------------
1185 void Parse::jump_switch_ranges(Node* key_val, SwitchRange *lo, SwitchRange *hi, int switch_depth) {
1186 Block* switch_block = block();
1187 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
1188
1189 if (switch_depth == 0) {
1190 // Do special processing for the top-level call.
1191 assert(lo->lo() == min_jint, "initial range must exhaust Type::INT");
1192 assert(hi->hi() == max_jint, "initial range must exhaust Type::INT");
1193
1194 // Decrement pred-numbers for the unique set of nodes.
1195 #ifdef ASSERT
1196 if (!trim_ranges) {
1197 // Ensure that the block's successors are a (duplicate-free) set.
1198 int successors_counted = 0; // block occurrences in [hi..lo]
1199 int unique_successors = switch_block->num_successors();
1200 for (int i = 0; i < unique_successors; i++) {
1201 Block* target = switch_block->successor_at(i);
1202
1203 // Check that the set of successors is the same in both places.
1204 int successors_found = 0;
1205 for (SwitchRange* p = lo; p <= hi; p++) {
1206 if (p->dest() == target->start()) successors_found++;
1207 }
1208 assert(successors_found > 0, "successor must be known");
1209 successors_counted += successors_found;
1210 }
1211 assert(successors_counted == (hi-lo)+1, "no unexpected successors");
1212 }
1213 #endif
1214
1215 // Maybe prune the inputs, based on the type of key_val.
1216 jint min_val = min_jint;
1217 jint max_val = max_jint;
1218 const TypeInt* ti = key_val->bottom_type()->isa_int();
1219 if (ti != NULL) {
1220 min_val = ti->_lo;
1221 max_val = ti->_hi;
1222 assert(min_val <= max_val, "invalid int type");
1223 }
1224 while (lo->hi() < min_val) {
1225 lo++;
1226 }
1227 if (lo->lo() < min_val) {
1228 lo->setRange(min_val, lo->hi(), lo->dest(), lo->table_index(), lo->cnt());
1229 }
1230 while (hi->lo() > max_val) {
1231 hi--;
1232 }
1233 if (hi->hi() > max_val) {
1234 hi->setRange(hi->lo(), max_val, hi->dest(), hi->table_index(), hi->cnt());
1235 }
1236
1237 linear_search_switch_ranges(key_val, lo, hi);
1238 }
1239
1240 #ifndef PRODUCT
1241 if (switch_depth == 0) {
1242 _max_switch_depth = 0;
1243 _est_switch_depth = log2_intptr((hi-lo+1)-1)+1;
1244 }
1245 #endif
1246
1247 assert(lo <= hi, "must be a non-empty set of ranges");
1248 if (lo == hi) {
1249 jump_if_always_fork(lo->dest(), lo->table_index(), trim_ranges && lo->cnt() == 0);
1250 } else {
1251 assert(lo->hi() == (lo+1)->lo()-1, "contiguous ranges");
1252 assert(hi->lo() == (hi-1)->hi()+1, "contiguous ranges");
1253
1254 if (create_jump_tables(key_val, lo, hi)) return;
1255
1256 SwitchRange* mid = NULL;
1257 float total_cnt = sum_of_cnts(lo, hi);
1258
1259 int nr = hi - lo + 1;
1260 if (UseSwitchProfiling) {
1261 // Don't keep the binary search tree balanced: pick up mid point
1262 // that split frequencies in half.
1263 float cnt = 0;
1264 for (SwitchRange* sr = lo; sr <= hi; sr++) {
1265 cnt += sr->cnt();
1266 if (cnt >= total_cnt / 2) {
1267 mid = sr;
1268 break;
1269 }
1270 }
1271 } else {
1272 mid = lo + nr/2;
1273
1274 // if there is an easy choice, pivot at a singleton:
1275 if (nr > 3 && !mid->is_singleton() && (mid-1)->is_singleton()) mid--;
1276
1277 assert(lo < mid && mid <= hi, "good pivot choice");
1278 assert(nr != 2 || mid == hi, "should pick higher of 2");
1279 assert(nr != 3 || mid == hi-1, "should pick middle of 3");
1280 }
1281
1282
1283 Node *test_val = _gvn.intcon(mid == lo ? mid->hi() : mid->lo());
1284
1285 if (mid->is_singleton()) {
1286 IfNode *iff_ne = jump_if_fork_int(key_val, test_val, BoolTest::ne, 1-if_prob(mid->cnt(), total_cnt), if_cnt(mid->cnt()));
1287 jump_if_false_fork(iff_ne, mid->dest(), mid->table_index(), trim_ranges && mid->cnt() == 0);
1288
1289 // Special Case: If there are exactly three ranges, and the high
1290 // and low range each go to the same place, omit the "gt" test,
1291 // since it will not discriminate anything.
1292 bool eq_test_only = (hi == lo+2 && hi->dest() == lo->dest() && mid == hi-1) || mid == lo;
1293
1294 // if there is a higher range, test for it and process it:
1295 if (mid < hi && !eq_test_only) {
1296 // two comparisons of same values--should enable 1 test for 2 branches
1297 // Use BoolTest::le instead of BoolTest::gt
1298 float cnt = sum_of_cnts(lo, mid-1);
1299 IfNode *iff_le = jump_if_fork_int(key_val, test_val, BoolTest::le, if_prob(cnt, total_cnt), if_cnt(cnt));
1300 Node *iftrue = _gvn.transform( new IfTrueNode(iff_le) );
1301 Node *iffalse = _gvn.transform( new IfFalseNode(iff_le) );
1302 { PreserveJVMState pjvms(this);
1303 set_control(iffalse);
1304 jump_switch_ranges(key_val, mid+1, hi, switch_depth+1);
1305 }
1306 set_control(iftrue);
1307 }
1308
1309 } else {
1310 // mid is a range, not a singleton, so treat mid..hi as a unit
1311 float cnt = sum_of_cnts(mid == lo ? mid+1 : mid, hi);
1312 IfNode *iff_ge = jump_if_fork_int(key_val, test_val, mid == lo ? BoolTest::gt : BoolTest::ge, if_prob(cnt, total_cnt), if_cnt(cnt));
1313
1314 // if there is a higher range, test for it and process it:
1315 if (mid == hi) {
1316 jump_if_true_fork(iff_ge, mid->dest(), mid->table_index(), trim_ranges && cnt == 0);
1317 } else {
1318 Node *iftrue = _gvn.transform( new IfTrueNode(iff_ge) );
1319 Node *iffalse = _gvn.transform( new IfFalseNode(iff_ge) );
1320 { PreserveJVMState pjvms(this);
1321 set_control(iftrue);
1322 jump_switch_ranges(key_val, mid == lo ? mid+1 : mid, hi, switch_depth+1);
1323 }
1324 set_control(iffalse);
1325 }
1326 }
1327
1328 // in any case, process the lower range
1329 if (mid == lo) {
1330 if (mid->is_singleton()) {
1331 jump_switch_ranges(key_val, lo+1, hi, switch_depth+1);
1332 } else {
1333 jump_if_always_fork(lo->dest(), lo->table_index(), trim_ranges && lo->cnt() == 0);
1334 }
1335 } else {
1336 jump_switch_ranges(key_val, lo, mid-1, switch_depth+1);
1337 }
1338 }
1339
1340 // Decrease pred_count for each successor after all is done.
1341 if (switch_depth == 0) {
1342 int unique_successors = switch_block->num_successors();
1343 for (int i = 0; i < unique_successors; i++) {
1344 Block* target = switch_block->successor_at(i);
1345 // Throw away the pre-allocated path for each unique successor.
1346 target->next_path_num();
1347 }
1348 }
1349
1350 #ifndef PRODUCT
1351 _max_switch_depth = MAX2(switch_depth, _max_switch_depth);
1352 if (TraceOptoParse && Verbose && WizardMode && switch_depth == 0) {
1353 SwitchRange* r;
1354 int nsing = 0;
1355 for( r = lo; r <= hi; r++ ) {
1356 if( r->is_singleton() ) nsing++;
1357 }
1358 tty->print(">>> ");
1359 _method->print_short_name();
1360 tty->print_cr(" switch decision tree");
1361 tty->print_cr(" %d ranges (%d singletons), max_depth=%d, est_depth=%d",
1362 (int) (hi-lo+1), nsing, _max_switch_depth, _est_switch_depth);
1363 if (_max_switch_depth > _est_switch_depth) {
1364 tty->print_cr("******** BAD SWITCH DEPTH ********");
1365 }
1366 tty->print(" ");
1367 for( r = lo; r <= hi; r++ ) {
1368 r->print();
1369 }
1370 tty->cr();
1371 }
1372 #endif
1373 }
1374
1375 void Parse::modf() {
1376 Node *f2 = pop();
1377 Node *f1 = pop();
1378 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::modf_Type(),
1379 CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1380 "frem", NULL, //no memory effects
1381 f1, f2);
1382 Node* res = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0));
1383
1384 push(res);
1385 }
1386
1387 void Parse::modd() {
1388 Node *d2 = pop_pair();
1389 Node *d1 = pop_pair();
1390 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(),
1391 CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1392 "drem", NULL, //no memory effects
1393 d1, top(), d2, top());
1394 Node* res_d = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0));
1395
1396 #ifdef ASSERT
1397 Node* res_top = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 1));
1398 assert(res_top == top(), "second value must be top");
1399 #endif
1400
1401 push_pair(res_d);
1402 }
1403
1404 void Parse::l2f() {
1405 Node* f2 = pop();
1406 Node* f1 = pop();
1407 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::l2f_Type(),
1408 CAST_FROM_FN_PTR(address, SharedRuntime::l2f),
1409 "l2f", NULL, //no memory effects
1410 f1, f2);
1411 Node* res = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0));
1412
1413 push(res);
1414 }
1415
1416 void Parse::do_irem() {
1417 // Must keep both values on the expression-stack during null-check
1418 zero_check_int(peek());
1419 // Compile-time detect of null-exception?
1420 if (stopped()) return;
1421
1422 Node* b = pop();
1423 Node* a = pop();
1424
1425 const Type *t = _gvn.type(b);
1426 if (t != Type::TOP) {
1427 const TypeInt *ti = t->is_int();
1428 if (ti->is_con()) {
1429 int divisor = ti->get_con();
1430 // check for positive power of 2
1431 if (divisor > 0 &&
1432 (divisor & ~(divisor-1)) == divisor) {
1433 // yes !
1434 Node *mask = _gvn.intcon((divisor - 1));
1435 // Sigh, must handle negative dividends
1436 Node *zero = _gvn.intcon(0);
1437 IfNode *ifff = jump_if_fork_int(a, zero, BoolTest::lt, PROB_FAIR, COUNT_UNKNOWN);
1438 Node *iff = _gvn.transform( new IfFalseNode(ifff) );
1439 Node *ift = _gvn.transform( new IfTrueNode (ifff) );
1440 Node *reg = jump_if_join(ift, iff);
1441 Node *phi = PhiNode::make(reg, NULL, TypeInt::INT);
1442 // Negative path; negate/and/negate
1443 Node *neg = _gvn.transform( new SubINode(zero, a) );
1444 Node *andn= _gvn.transform( new AndINode(neg, mask) );
1445 Node *negn= _gvn.transform( new SubINode(zero, andn) );
1446 phi->init_req(1, negn);
1447 // Fast positive case
1448 Node *andx = _gvn.transform( new AndINode(a, mask) );
1449 phi->init_req(2, andx);
1450 // Push the merge
1451 push( _gvn.transform(phi) );
1452 return;
1453 }
1454 }
1455 }
1456 // Default case
1457 push( _gvn.transform( new ModINode(control(),a,b) ) );
1458 }
1459
1460 // Handle jsr and jsr_w bytecode
1461 void Parse::do_jsr() {
1462 assert(bc() == Bytecodes::_jsr || bc() == Bytecodes::_jsr_w, "wrong bytecode");
1463
1464 // Store information about current state, tagged with new _jsr_bci
1465 int return_bci = iter().next_bci();
1466 int jsr_bci = (bc() == Bytecodes::_jsr) ? iter().get_dest() : iter().get_far_dest();
1467
1468 // Update method data
1469 profile_taken_branch(jsr_bci);
1470
1471 // The way we do things now, there is only one successor block
1472 // for the jsr, because the target code is cloned by ciTypeFlow.
1473 Block* target = successor_for_bci(jsr_bci);
1474
1475 // What got pushed?
1476 const Type* ret_addr = target->peek();
1477 assert(ret_addr->singleton(), "must be a constant (cloned jsr body)");
1478
1479 // Effect on jsr on stack
1480 push(_gvn.makecon(ret_addr));
1481
1482 // Flow to the jsr.
1483 merge(jsr_bci);
1484 }
1485
1486 // Handle ret bytecode
1487 void Parse::do_ret() {
1488 // Find to whom we return.
1489 assert(block()->num_successors() == 1, "a ret can only go one place now");
1490 Block* target = block()->successor_at(0);
1491 assert(!target->is_ready(), "our arrival must be expected");
1492 profile_ret(target->flow()->start());
1493 int pnum = target->next_path_num();
1494 merge_common(target, pnum);
1495 }
1496
1497 static bool has_injected_profile(BoolTest::mask btest, Node* test, int& taken, int& not_taken) {
1498 if (btest != BoolTest::eq && btest != BoolTest::ne) {
1499 // Only ::eq and ::ne are supported for profile injection.
1500 return false;
1501 }
1502 if (test->is_Cmp() &&
1503 test->in(1)->Opcode() == Op_ProfileBoolean) {
1504 ProfileBooleanNode* profile = (ProfileBooleanNode*)test->in(1);
1505 int false_cnt = profile->false_count();
1506 int true_cnt = profile->true_count();
1507
1508 // Counts matching depends on the actual test operation (::eq or ::ne).
1509 // No need to scale the counts because profile injection was designed
1510 // to feed exact counts into VM.
1511 taken = (btest == BoolTest::eq) ? false_cnt : true_cnt;
1512 not_taken = (btest == BoolTest::eq) ? true_cnt : false_cnt;
1513
1514 profile->consume();
1515 return true;
1516 }
1517 return false;
1518 }
1519 //--------------------------dynamic_branch_prediction--------------------------
1520 // Try to gather dynamic branch prediction behavior. Return a probability
1521 // of the branch being taken and set the "cnt" field. Returns a -1.0
1522 // if we need to use static prediction for some reason.
1523 float Parse::dynamic_branch_prediction(float &cnt, BoolTest::mask btest, Node* test) {
1524 ResourceMark rm;
1525
1526 cnt = COUNT_UNKNOWN;
1527
1528 int taken = 0;
1529 int not_taken = 0;
1530
1531 bool use_mdo = !has_injected_profile(btest, test, taken, not_taken);
1532
1533 if (use_mdo) {
1534 // Use MethodData information if it is available
1535 // FIXME: free the ProfileData structure
1536 ciMethodData* methodData = method()->method_data();
1537 if (!methodData->is_mature()) return PROB_UNKNOWN;
1538 ciProfileData* data = methodData->bci_to_data(bci());
1539 if (data == NULL) {
1540 return PROB_UNKNOWN;
1541 }
1542 if (!data->is_JumpData()) return PROB_UNKNOWN;
1543
1544 // get taken and not taken values
1545 taken = data->as_JumpData()->taken();
1546 not_taken = 0;
1547 if (data->is_BranchData()) {
1548 not_taken = data->as_BranchData()->not_taken();
1549 }
1550
1551 // scale the counts to be commensurate with invocation counts:
1552 taken = method()->scale_count(taken);
1553 not_taken = method()->scale_count(not_taken);
1554 }
1555
1556 // Give up if too few (or too many, in which case the sum will overflow) counts to be meaningful.
1557 // We also check that individual counters are positive first, otherwise the sum can become positive.
1558 if (taken < 0 || not_taken < 0 || taken + not_taken < 40) {
1559 if (C->log() != NULL) {
1560 C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d'", iter().get_dest(), taken, not_taken);
1561 }
1562 return PROB_UNKNOWN;
1563 }
1564
1565 // Compute frequency that we arrive here
1566 float sum = taken + not_taken;
1567 // Adjust, if this block is a cloned private block but the
1568 // Jump counts are shared. Taken the private counts for
1569 // just this path instead of the shared counts.
1570 if( block()->count() > 0 )
1571 sum = block()->count();
1572 cnt = sum / FreqCountInvocations;
1573
1574 // Pin probability to sane limits
1575 float prob;
1576 if( !taken )
1577 prob = (0+PROB_MIN) / 2;
1578 else if( !not_taken )
1579 prob = (1+PROB_MAX) / 2;
1580 else { // Compute probability of true path
1581 prob = (float)taken / (float)(taken + not_taken);
1582 if (prob > PROB_MAX) prob = PROB_MAX;
1583 if (prob < PROB_MIN) prob = PROB_MIN;
1584 }
1585
1586 assert((cnt > 0.0f) && (prob > 0.0f),
1587 "Bad frequency assignment in if");
1588
1589 if (C->log() != NULL) {
1590 const char* prob_str = NULL;
1591 if (prob >= PROB_MAX) prob_str = (prob == PROB_MAX) ? "max" : "always";
1592 if (prob <= PROB_MIN) prob_str = (prob == PROB_MIN) ? "min" : "never";
1593 char prob_str_buf[30];
1594 if (prob_str == NULL) {
1595 jio_snprintf(prob_str_buf, sizeof(prob_str_buf), "%20.2f", prob);
1596 prob_str = prob_str_buf;
1597 }
1598 C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d' cnt='%f' prob='%s'",
1599 iter().get_dest(), taken, not_taken, cnt, prob_str);
1600 }
1601 return prob;
1602 }
1603
1604 //-----------------------------branch_prediction-------------------------------
1605 float Parse::branch_prediction(float& cnt,
1606 BoolTest::mask btest,
1607 int target_bci,
1608 Node* test) {
1609 float prob = dynamic_branch_prediction(cnt, btest, test);
1610 // If prob is unknown, switch to static prediction
1611 if (prob != PROB_UNKNOWN) return prob;
1612
1613 prob = PROB_FAIR; // Set default value
1614 if (btest == BoolTest::eq) // Exactly equal test?
1615 prob = PROB_STATIC_INFREQUENT; // Assume its relatively infrequent
1616 else if (btest == BoolTest::ne)
1617 prob = PROB_STATIC_FREQUENT; // Assume its relatively frequent
1618
1619 // If this is a conditional test guarding a backwards branch,
1620 // assume its a loop-back edge. Make it a likely taken branch.
1621 if (target_bci < bci()) {
1622 if (is_osr_parse()) { // Could be a hot OSR'd loop; force deopt
1623 // Since it's an OSR, we probably have profile data, but since
1624 // branch_prediction returned PROB_UNKNOWN, the counts are too small.
1625 // Let's make a special check here for completely zero counts.
1626 ciMethodData* methodData = method()->method_data();
1627 if (!methodData->is_empty()) {
1628 ciProfileData* data = methodData->bci_to_data(bci());
1629 // Only stop for truly zero counts, which mean an unknown part
1630 // of the OSR-ed method, and we want to deopt to gather more stats.
1631 // If you have ANY counts, then this loop is simply 'cold' relative
1632 // to the OSR loop.
1633 if (data == NULL ||
1634 (data->as_BranchData()->taken() + data->as_BranchData()->not_taken() == 0)) {
1635 // This is the only way to return PROB_UNKNOWN:
1636 return PROB_UNKNOWN;
1637 }
1638 }
1639 }
1640 prob = PROB_STATIC_FREQUENT; // Likely to take backwards branch
1641 }
1642
1643 assert(prob != PROB_UNKNOWN, "must have some guess at this point");
1644 return prob;
1645 }
1646
1647 // The magic constants are chosen so as to match the output of
1648 // branch_prediction() when the profile reports a zero taken count.
1649 // It is important to distinguish zero counts unambiguously, because
1650 // some branches (e.g., _213_javac.Assembler.eliminate) validly produce
1651 // very small but nonzero probabilities, which if confused with zero
1652 // counts would keep the program recompiling indefinitely.
1653 bool Parse::seems_never_taken(float prob) const {
1654 return prob < PROB_MIN;
1655 }
1656
1657 // True if the comparison seems to be the kind that will not change its
1658 // statistics from true to false. See comments in adjust_map_after_if.
1659 // This question is only asked along paths which are already
1660 // classifed as untaken (by seems_never_taken), so really,
1661 // if a path is never taken, its controlling comparison is
1662 // already acting in a stable fashion. If the comparison
1663 // seems stable, we will put an expensive uncommon trap
1664 // on the untaken path.
1665 bool Parse::seems_stable_comparison() const {
1666 if (C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if)) {
1667 return false;
1668 }
1669 return true;
1670 }
1671
1672 //-------------------------------repush_if_args--------------------------------
1673 // Push arguments of an "if" bytecode back onto the stack by adjusting _sp.
1674 inline int Parse::repush_if_args() {
1675 if (PrintOpto && WizardMode) {
1676 tty->print("defending against excessive implicit null exceptions on %s @%d in ",
1677 Bytecodes::name(iter().cur_bc()), iter().cur_bci());
1678 method()->print_name(); tty->cr();
1679 }
1680 int bc_depth = - Bytecodes::depth(iter().cur_bc());
1681 assert(bc_depth == 1 || bc_depth == 2, "only two kinds of branches");
1682 DEBUG_ONLY(sync_jvms()); // argument(n) requires a synced jvms
1683 assert(argument(0) != NULL, "must exist");
1684 assert(bc_depth == 1 || argument(1) != NULL, "two must exist");
1685 inc_sp(bc_depth);
1686 return bc_depth;
1687 }
1688
1689 //----------------------------------do_ifnull----------------------------------
1690 void Parse::do_ifnull(BoolTest::mask btest, Node *c) {
1691 int target_bci = iter().get_dest();
1692
1693 Block* branch_block = successor_for_bci(target_bci);
1694 Block* next_block = successor_for_bci(iter().next_bci());
1695
1696 float cnt;
1697 float prob = branch_prediction(cnt, btest, target_bci, c);
1698 if (prob == PROB_UNKNOWN) {
1699 // (An earlier version of do_ifnull omitted this trap for OSR methods.)
1700 if (PrintOpto && Verbose) {
1701 tty->print_cr("Never-taken edge stops compilation at bci %d", bci());
1702 }
1703 repush_if_args(); // to gather stats on loop
1704 // We need to mark this branch as taken so that if we recompile we will
1705 // see that it is possible. In the tiered system the interpreter doesn't
1706 // do profiling and by the time we get to the lower tier from the interpreter
1707 // the path may be cold again. Make sure it doesn't look untaken
1708 profile_taken_branch(target_bci, !ProfileInterpreter);
1709 uncommon_trap(Deoptimization::Reason_unreached,
1710 Deoptimization::Action_reinterpret,
1711 NULL, "cold");
1712 if (C->eliminate_boxing()) {
1713 // Mark the successor blocks as parsed
1714 branch_block->next_path_num();
1715 next_block->next_path_num();
1716 }
1717 return;
1718 }
1719
1720 NOT_PRODUCT(explicit_null_checks_inserted++);
1721
1722 // Generate real control flow
1723 Node *tst = _gvn.transform( new BoolNode( c, btest ) );
1724
1725 // Sanity check the probability value
1726 assert(prob > 0.0f,"Bad probability in Parser");
1727 // Need xform to put node in hash table
1728 IfNode *iff = create_and_xform_if( control(), tst, prob, cnt );
1729 assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser");
1730 // True branch
1731 { PreserveJVMState pjvms(this);
1732 Node* iftrue = _gvn.transform( new IfTrueNode (iff) );
1733 set_control(iftrue);
1734
1735 if (stopped()) { // Path is dead?
1736 NOT_PRODUCT(explicit_null_checks_elided++);
1737 if (C->eliminate_boxing()) {
1738 // Mark the successor block as parsed
1739 branch_block->next_path_num();
1740 }
1741 } else { // Path is live.
1742 // Update method data
1743 profile_taken_branch(target_bci);
1744 adjust_map_after_if(btest, c, prob, branch_block);
1745 if (!stopped()) {
1746 merge(target_bci);
1747 }
1748 }
1749 }
1750
1751 // False branch
1752 Node* iffalse = _gvn.transform( new IfFalseNode(iff) );
1753 set_control(iffalse);
1754
1755 if (stopped()) { // Path is dead?
1756 NOT_PRODUCT(explicit_null_checks_elided++);
1757 if (C->eliminate_boxing()) {
1758 // Mark the successor block as parsed
1759 next_block->next_path_num();
1760 }
1761 } else { // Path is live.
1762 // Update method data
1763 profile_not_taken_branch();
1764 adjust_map_after_if(BoolTest(btest).negate(), c, 1.0-prob, next_block);
1765 }
1766 }
1767
1768 //------------------------------------do_if------------------------------------
1769 void Parse::do_if(BoolTest::mask btest, Node* c, bool new_path, Node** ctrl_taken) {
1770 int target_bci = iter().get_dest();
1771
1772 Block* branch_block = successor_for_bci(target_bci);
1773 Block* next_block = successor_for_bci(iter().next_bci());
1774
1775 float cnt;
1776 float prob = branch_prediction(cnt, btest, target_bci, c);
1777 float untaken_prob = 1.0 - prob;
1778
1779 if (prob == PROB_UNKNOWN) {
1780 if (PrintOpto && Verbose) {
1781 tty->print_cr("Never-taken edge stops compilation at bci %d", bci());
1782 }
1783 repush_if_args(); // to gather stats on loop
1784 // We need to mark this branch as taken so that if we recompile we will
1785 // see that it is possible. In the tiered system the interpreter doesn't
1786 // do profiling and by the time we get to the lower tier from the interpreter
1787 // the path may be cold again. Make sure it doesn't look untaken
1788 profile_taken_branch(target_bci, !ProfileInterpreter);
1789 uncommon_trap(Deoptimization::Reason_unreached,
1790 Deoptimization::Action_reinterpret,
1791 NULL, "cold");
1792 if (C->eliminate_boxing()) {
1793 // Mark the successor blocks as parsed
1794 branch_block->next_path_num();
1795 next_block->next_path_num();
1796 }
1797 return;
1798 }
1799
1800 // Sanity check the probability value
1801 assert(0.0f < prob && prob < 1.0f,"Bad probability in Parser");
1802
1803 bool taken_if_true = true;
1804 // Convert BoolTest to canonical form:
1805 if (!BoolTest(btest).is_canonical()) {
1806 btest = BoolTest(btest).negate();
1807 taken_if_true = false;
1808 // prob is NOT updated here; it remains the probability of the taken
1809 // path (as opposed to the prob of the path guarded by an 'IfTrueNode').
1810 }
1811 assert(btest != BoolTest::eq, "!= is the only canonical exact test");
1812
1813 Node* tst0 = new BoolNode(c, btest);
1814 Node* tst = _gvn.transform(tst0);
1815 BoolTest::mask taken_btest = BoolTest::illegal;
1816 BoolTest::mask untaken_btest = BoolTest::illegal;
1817
1818 if (tst->is_Bool()) {
1819 // Refresh c from the transformed bool node, since it may be
1820 // simpler than the original c. Also re-canonicalize btest.
1821 // This wins when (Bool ne (Conv2B p) 0) => (Bool ne (CmpP p NULL)).
1822 // That can arise from statements like: if (x instanceof C) ...
1823 if (tst != tst0) {
1824 // Canonicalize one more time since transform can change it.
1825 btest = tst->as_Bool()->_test._test;
1826 if (!BoolTest(btest).is_canonical()) {
1827 // Reverse edges one more time...
1828 tst = _gvn.transform( tst->as_Bool()->negate(&_gvn) );
1829 btest = tst->as_Bool()->_test._test;
1830 assert(BoolTest(btest).is_canonical(), "sanity");
1831 taken_if_true = !taken_if_true;
1832 }
1833 c = tst->in(1);
1834 }
1835 BoolTest::mask neg_btest = BoolTest(btest).negate();
1836 taken_btest = taken_if_true ? btest : neg_btest;
1837 untaken_btest = taken_if_true ? neg_btest : btest;
1838 }
1839
1840 // Generate real control flow
1841 float true_prob = (taken_if_true ? prob : untaken_prob);
1842 IfNode* iff = create_and_map_if(control(), tst, true_prob, cnt);
1843 assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser");
1844 Node* taken_branch = new IfTrueNode(iff);
1845 Node* untaken_branch = new IfFalseNode(iff);
1846 if (!taken_if_true) { // Finish conversion to canonical form
1847 Node* tmp = taken_branch;
1848 taken_branch = untaken_branch;
1849 untaken_branch = tmp;
1850 }
1851
1852 // Branch is taken:
1853 { PreserveJVMState pjvms(this);
1854 taken_branch = _gvn.transform(taken_branch);
1855 set_control(taken_branch);
1856
1857 if (stopped()) {
1858 if (C->eliminate_boxing() && !new_path) {
1859 // Mark the successor block as parsed (if we haven't created a new path)
1860 branch_block->next_path_num();
1861 }
1862 } else {
1863 // Update method data
1864 profile_taken_branch(target_bci);
1865 adjust_map_after_if(taken_btest, c, prob, branch_block);
1866 if (!stopped()) {
1867 if (new_path) {
1868 // Merge by using a new path
1869 merge_new_path(target_bci);
1870 } else if (ctrl_taken != NULL) {
1871 // Don't merge but save taken branch to be wired by caller
1872 *ctrl_taken = control();
1873 } else {
1874 merge(target_bci);
1875 }
1876 }
1877 }
1878 }
1879
1880 untaken_branch = _gvn.transform(untaken_branch);
1881 set_control(untaken_branch);
1882
1883 // Branch not taken.
1884 if (stopped() && ctrl_taken == NULL) {
1885 if (C->eliminate_boxing()) {
1886 // Mark the successor block as parsed (if caller does not re-wire control flow)
1887 next_block->next_path_num();
1888 }
1889 } else {
1890 // Update method data
1891 profile_not_taken_branch();
1892 adjust_map_after_if(untaken_btest, c, untaken_prob, next_block);
1893 }
1894 }
1895
1896 void Parse::do_acmp(BoolTest::mask btest, Node* a, Node* b) {
1897 ciMethod* subst_method = ciEnv::current()->ValueBootstrapMethods_klass()->find_method(ciSymbol::isSubstitutable_name(), ciSymbol::object_object_boolean_signature());
1898 // If current method is ValueBootstrapMethods::isSubstitutable(),
1899 // compile the acmp as a regular pointer comparison otherwise we
1900 // could call ValueBootstrapMethods::isSubstitutable() back
1901 if (!EnableValhalla || (method() == subst_method)) {
1902 Node* cmp = CmpP(a, b);
1903 cmp = optimize_cmp_with_klass(cmp);
1904 do_if(btest, cmp);
1905 return;
1906 }
1907
1908 // Substitutability test
1909 if (a->is_ValueType()) {
1910 inc_sp(2);
1911 a = a->as_ValueType()->allocate(this, true)->get_oop();
1912 dec_sp(2);
1913 }
1914 if (b->is_ValueType()) {
1915 inc_sp(2);
1916 b = b->as_ValueType()->allocate(this, true)->get_oop();
1917 dec_sp(2);
1918 }
1919
1920 const TypeOopPtr* ta = _gvn.type(a)->isa_oopptr();
1921 const TypeOopPtr* tb = _gvn.type(b)->isa_oopptr();
1922
1923 if (ta == NULL || !ta->can_be_value_type_raw() ||
1924 tb == NULL || !tb->can_be_value_type_raw()) {
1925 Node* cmp = CmpP(a, b);
1926 cmp = optimize_cmp_with_klass(cmp);
1927 do_if(btest, cmp);
1928 return;
1929 }
1930
1931 Node* cmp = CmpP(a, b);
1932 cmp = optimize_cmp_with_klass(cmp);
1933 Node* eq_region = NULL;
1934 if (btest == BoolTest::eq) {
1935 do_if(btest, cmp, true);
1936 if (stopped()) {
1937 return;
1938 }
1939 } else {
1940 assert(btest == BoolTest::ne, "only eq or ne");
1941 Node* is_not_equal = NULL;
1942 eq_region = new RegionNode(3);
1943 {
1944 PreserveJVMState pjvms(this);
1945 do_if(btest, cmp, false, &is_not_equal);
1946 if (!stopped()) {
1947 eq_region->init_req(1, control());
1948 }
1949 }
1950 if (is_not_equal == NULL || is_not_equal->is_top()) {
1951 record_for_igvn(eq_region);
1952 set_control(_gvn.transform(eq_region));
1953 return;
1954 }
1955 set_control(is_not_equal);
1956 }
1957 // Pointers not equal, check for values
1958 Node* ne_region = new RegionNode(6);
1959 inc_sp(2);
1960 Node* null_ctl = top();
1961 Node* not_null_a = null_check_oop(a, &null_ctl, !too_many_traps(Deoptimization::Reason_null_check), false, false);
1962 dec_sp(2);
1963 ne_region->init_req(1, null_ctl);
1964 if (stopped()) {
1965 record_for_igvn(ne_region);
1966 set_control(_gvn.transform(ne_region));
1967 if (btest == BoolTest::ne) {
1968 {
1969 PreserveJVMState pjvms(this);
1970 int target_bci = iter().get_dest();
1971 merge(target_bci);
1972 }
1973 record_for_igvn(eq_region);
1974 set_control(_gvn.transform(eq_region));
1975 }
1976 return;
1977 }
1978
1979 Node* is_value = is_always_locked(not_null_a);
1980 Node* value_mask = _gvn.MakeConX(markOopDesc::always_locked_pattern);
1981 Node* is_value_cmp = _gvn.transform(new CmpXNode(is_value, value_mask));
1982 Node* is_value_bol = _gvn.transform(new BoolNode(is_value_cmp, BoolTest::ne));
1983 IfNode* is_value_iff = create_and_map_if(control(), is_value_bol, PROB_FAIR, COUNT_UNKNOWN);
1984 Node* not_value = _gvn.transform(new IfTrueNode(is_value_iff));
1985 set_control(_gvn.transform(new IfFalseNode(is_value_iff)));
1986 ne_region->init_req(2, not_value);
1987
1988 // One of the 2 pointers refers to a value, check if both are of
1989 // the same class
1990 inc_sp(2);
1991 null_ctl = top();
1992 Node* not_null_b = null_check_oop(b, &null_ctl, !too_many_traps(Deoptimization::Reason_null_check), false, false);
1993 dec_sp(2);
1994 ne_region->init_req(3, null_ctl);
1995 if (stopped()) {
1996 record_for_igvn(ne_region);
1997 set_control(_gvn.transform(ne_region));
1998 if (btest == BoolTest::ne) {
1999 {
2000 PreserveJVMState pjvms(this);
2001 int target_bci = iter().get_dest();
2002 merge(target_bci);
2003 }
2004 record_for_igvn(eq_region);
2005 set_control(_gvn.transform(eq_region));
2006 }
2007 return;
2008 }
2009 Node* kls_a = load_object_klass(not_null_a);
2010 Node* kls_b = load_object_klass(not_null_b);
2011 Node* kls_cmp = CmpP(kls_a, kls_b);
2012 Node* kls_bol = _gvn.transform(new BoolNode(kls_cmp, BoolTest::ne));
2013 IfNode* kls_iff = create_and_map_if(control(), kls_bol, PROB_FAIR, COUNT_UNKNOWN);
2014 Node* kls_ne = _gvn.transform(new IfTrueNode(kls_iff));
2015 set_control(_gvn.transform(new IfFalseNode(kls_iff)));
2016 ne_region->init_req(4, kls_ne);
2017
2018 if (stopped()) {
2019 record_for_igvn(ne_region);
2020 set_control(_gvn.transform(ne_region));
2021 if (btest == BoolTest::ne) {
2022 {
2023 PreserveJVMState pjvms(this);
2024 int target_bci = iter().get_dest();
2025 merge(target_bci);
2026 }
2027 record_for_igvn(eq_region);
2028 set_control(_gvn.transform(eq_region));
2029 }
2030 return;
2031 }
2032 // Both are values of the same class, we need to perform a
2033 // substitutability test. Delegate to
2034 // ValueBootstrapMethods::isSubstitutable().
2035
2036 Node* ne_io_phi = PhiNode::make(ne_region, i_o());
2037 Node* mem = reset_memory();
2038 Node* ne_mem_phi = PhiNode::make(ne_region, mem);
2039
2040 Node* eq_io_phi = NULL;
2041 Node* eq_mem_phi = NULL;
2042 if (eq_region != NULL) {
2043 eq_io_phi = PhiNode::make(eq_region, i_o());
2044 eq_mem_phi = PhiNode::make(eq_region, mem);
2045 }
2046
2047 set_all_memory(mem);
2048
2049 kill_dead_locals();
2050 CallStaticJavaNode *call = new CallStaticJavaNode(C, TypeFunc::make(subst_method), SharedRuntime::get_resolve_static_call_stub(), subst_method, bci());
2051 call->set_override_symbolic_info(true);
2052 call->init_req(TypeFunc::Parms, not_null_a);
2053 call->init_req(TypeFunc::Parms+1, not_null_b);
2054 inc_sp(2);
2055 set_edges_for_java_call(call, false, false);
2056 Node* ret = set_results_for_java_call(call, false, true);
2057 dec_sp(2);
2058
2059 // Test the return value of ValueBootstrapMethods::isSubstitutable()
2060 Node* subst_cmp = _gvn.transform(new CmpINode(ret, intcon(1)));
2061 Node* ctl = C->top();
2062 if (btest == BoolTest::eq) {
2063 PreserveJVMState pjvms(this);
2064 do_if(btest, subst_cmp);
2065 if (!stopped()) {
2066 ctl = control();
2067 }
2068 } else {
2069 assert(btest == BoolTest::ne, "only eq or ne");
2070 PreserveJVMState pjvms(this);
2071 do_if(btest, subst_cmp, false, &ctl);
2072 if (!stopped()) {
2073 eq_region->init_req(2, control());
2074 eq_io_phi->init_req(2, i_o());
2075 eq_mem_phi->init_req(2, reset_memory());
2076 }
2077 }
2078 ne_region->init_req(5, ctl);
2079 ne_io_phi->init_req(5, i_o());
2080 ne_mem_phi->init_req(5, reset_memory());
2081
2082 record_for_igvn(ne_region);
2083 set_control(_gvn.transform(ne_region));
2084 set_i_o(_gvn.transform(ne_io_phi));
2085 set_all_memory(_gvn.transform(ne_mem_phi));
2086
2087 if (btest == BoolTest::ne) {
2088 {
2089 PreserveJVMState pjvms(this);
2090 int target_bci = iter().get_dest();
2091 merge(target_bci);
2092 }
2093
2094 record_for_igvn(eq_region);
2095 set_control(_gvn.transform(eq_region));
2096 set_i_o(_gvn.transform(eq_io_phi));
2097 set_all_memory(_gvn.transform(eq_mem_phi));
2098 }
2099 }
2100
2101 bool Parse::path_is_suitable_for_uncommon_trap(float prob) const {
2102 // Don't want to speculate on uncommon traps when running with -Xcomp
2103 if (!UseInterpreter) {
2104 return false;
2105 }
2106 return (seems_never_taken(prob) && seems_stable_comparison());
2107 }
2108
2109 void Parse::maybe_add_predicate_after_if(Block* path) {
2110 if (path->is_SEL_head() && path->preds_parsed() == 0) {
2111 // Add predicates at bci of if dominating the loop so traps can be
2112 // recorded on the if's profile data
2113 int bc_depth = repush_if_args();
2114 add_predicate();
2115 dec_sp(bc_depth);
2116 path->set_has_predicates();
2117 }
2118 }
2119
2120
2121 //----------------------------adjust_map_after_if------------------------------
2122 // Adjust the JVM state to reflect the result of taking this path.
2123 // Basically, it means inspecting the CmpNode controlling this
2124 // branch, seeing how it constrains a tested value, and then
2125 // deciding if it's worth our while to encode this constraint
2126 // as graph nodes in the current abstract interpretation map.
2127 void Parse::adjust_map_after_if(BoolTest::mask btest, Node* c, float prob, Block* path) {
2128 if (!c->is_Cmp()) {
2129 maybe_add_predicate_after_if(path);
2130 return;
2131 }
2132
2133 if (stopped() || btest == BoolTest::illegal) {
2134 return; // nothing to do
2135 }
2136
2137 bool is_fallthrough = (path == successor_for_bci(iter().next_bci()));
2138
2139 if (path_is_suitable_for_uncommon_trap(prob)) {
2140 repush_if_args();
2141 uncommon_trap(Deoptimization::Reason_unstable_if,
2142 Deoptimization::Action_reinterpret,
2143 NULL,
2144 (is_fallthrough ? "taken always" : "taken never"));
2145 return;
2146 }
2147
2148 Node* val = c->in(1);
2149 Node* con = c->in(2);
2150 const Type* tcon = _gvn.type(con);
2151 const Type* tval = _gvn.type(val);
2152 bool have_con = tcon->singleton();
2153 if (tval->singleton()) {
2154 if (!have_con) {
2155 // Swap, so constant is in con.
2156 con = val;
2157 tcon = tval;
2158 val = c->in(2);
2159 tval = _gvn.type(val);
2160 btest = BoolTest(btest).commute();
2161 have_con = true;
2162 } else {
2163 // Do we have two constants? Then leave well enough alone.
2164 have_con = false;
2165 }
2166 }
2167 if (!have_con) { // remaining adjustments need a con
2168 maybe_add_predicate_after_if(path);
2169 return;
2170 }
2171
2172 sharpen_type_after_if(btest, con, tcon, val, tval);
2173 maybe_add_predicate_after_if(path);
2174 }
2175
2176
2177 static Node* extract_obj_from_klass_load(PhaseGVN* gvn, Node* n) {
2178 Node* ldk;
2179 if (n->is_DecodeNKlass()) {
2180 if (n->in(1)->Opcode() != Op_LoadNKlass) {
2181 return NULL;
2182 } else {
2183 ldk = n->in(1);
2184 }
2185 } else if (n->Opcode() != Op_LoadKlass) {
2186 return NULL;
2187 } else {
2188 ldk = n;
2189 }
2190 assert(ldk != NULL && ldk->is_Load(), "should have found a LoadKlass or LoadNKlass node");
2191
2192 Node* adr = ldk->in(MemNode::Address);
2193 intptr_t off = 0;
2194 Node* obj = AddPNode::Ideal_base_and_offset(adr, gvn, off);
2195 if (obj == NULL || off != oopDesc::klass_offset_in_bytes()) // loading oopDesc::_klass?
2196 return NULL;
2197 const TypePtr* tp = gvn->type(obj)->is_ptr();
2198 if (tp == NULL || !(tp->isa_instptr() || tp->isa_aryptr())) // is obj a Java object ptr?
2199 return NULL;
2200
2201 return obj;
2202 }
2203
2204 void Parse::sharpen_type_after_if(BoolTest::mask btest,
2205 Node* con, const Type* tcon,
2206 Node* val, const Type* tval) {
2207 // Look for opportunities to sharpen the type of a node
2208 // whose klass is compared with a constant klass.
2209 if (btest == BoolTest::eq && tcon->isa_klassptr()) {
2210 Node* obj = extract_obj_from_klass_load(&_gvn, val);
2211 const TypeOopPtr* con_type = tcon->isa_klassptr()->as_instance_type();
2212 if (obj != NULL && (con_type->isa_instptr() || con_type->isa_aryptr())) {
2213 // Found:
2214 // Bool(CmpP(LoadKlass(obj._klass), ConP(Foo.klass)), [eq])
2215 // or the narrowOop equivalent.
2216 const Type* obj_type = _gvn.type(obj);
2217 const TypeOopPtr* tboth = obj_type->join_speculative(con_type)->isa_oopptr();
2218 if (tboth != NULL && tboth->klass_is_exact() && tboth != obj_type &&
2219 tboth->higher_equal(obj_type)) {
2220 // obj has to be of the exact type Foo if the CmpP succeeds.
2221 int obj_in_map = map()->find_edge(obj);
2222 JVMState* jvms = this->jvms();
2223 if (obj_in_map >= 0 &&
2224 (jvms->is_loc(obj_in_map) || jvms->is_stk(obj_in_map))) {
2225 TypeNode* ccast = new CheckCastPPNode(control(), obj, tboth);
2226 const Type* tcc = ccast->as_Type()->type();
2227 assert(tcc != obj_type && tcc->higher_equal(obj_type), "must improve");
2228 // Delay transform() call to allow recovery of pre-cast value
2229 // at the control merge.
2230 _gvn.set_type_bottom(ccast);
2231 record_for_igvn(ccast);
2232 // Here's the payoff.
2233 replace_in_map(obj, ccast);
2234 }
2235 }
2236 }
2237 }
2238
2239 int val_in_map = map()->find_edge(val);
2240 if (val_in_map < 0) return; // replace_in_map would be useless
2241 {
2242 JVMState* jvms = this->jvms();
2243 if (!(jvms->is_loc(val_in_map) ||
2244 jvms->is_stk(val_in_map)))
2245 return; // again, it would be useless
2246 }
2247
2248 // Check for a comparison to a constant, and "know" that the compared
2249 // value is constrained on this path.
2250 assert(tcon->singleton(), "");
2251 ConstraintCastNode* ccast = NULL;
2252 Node* cast = NULL;
2253
2254 switch (btest) {
2255 case BoolTest::eq: // Constant test?
2256 {
2257 const Type* tboth = tcon->join_speculative(tval);
2258 if (tboth == tval) break; // Nothing to gain.
2259 if (tcon->isa_int()) {
2260 ccast = new CastIINode(val, tboth);
2261 } else if (tcon == TypePtr::NULL_PTR) {
2262 // Cast to null, but keep the pointer identity temporarily live.
2263 ccast = new CastPPNode(val, tboth);
2264 } else {
2265 const TypeF* tf = tcon->isa_float_constant();
2266 const TypeD* td = tcon->isa_double_constant();
2267 // Exclude tests vs float/double 0 as these could be
2268 // either +0 or -0. Just because you are equal to +0
2269 // doesn't mean you ARE +0!
2270 // Note, following code also replaces Long and Oop values.
2271 if ((!tf || tf->_f != 0.0) &&
2272 (!td || td->_d != 0.0))
2273 cast = con; // Replace non-constant val by con.
2274 }
2275 }
2276 break;
2277
2278 case BoolTest::ne:
2279 if (tcon == TypePtr::NULL_PTR) {
2280 cast = cast_not_null(val, false);
2281 }
2282 break;
2283
2284 default:
2285 // (At this point we could record int range types with CastII.)
2286 break;
2287 }
2288
2289 if (ccast != NULL) {
2290 const Type* tcc = ccast->as_Type()->type();
2291 assert(tcc != tval && tcc->higher_equal(tval), "must improve");
2292 // Delay transform() call to allow recovery of pre-cast value
2293 // at the control merge.
2294 ccast->set_req(0, control());
2295 _gvn.set_type_bottom(ccast);
2296 record_for_igvn(ccast);
2297 cast = ccast;
2298 }
2299
2300 if (cast != NULL) { // Here's the payoff.
2301 replace_in_map(val, cast);
2302 }
2303 }
2304
2305 /**
2306 * Use speculative type to optimize CmpP node: if comparison is
2307 * against the low level class, cast the object to the speculative
2308 * type if any. CmpP should then go away.
2309 *
2310 * @param c expected CmpP node
2311 * @return result of CmpP on object casted to speculative type
2312 *
2313 */
2314 Node* Parse::optimize_cmp_with_klass(Node* c) {
2315 // If this is transformed by the _gvn to a comparison with the low
2316 // level klass then we may be able to use speculation
2317 if (c->Opcode() == Op_CmpP &&
2318 (c->in(1)->Opcode() == Op_LoadKlass || c->in(1)->Opcode() == Op_DecodeNKlass) &&
2319 c->in(2)->is_Con()) {
2320 Node* load_klass = NULL;
2321 Node* decode = NULL;
2322 if (c->in(1)->Opcode() == Op_DecodeNKlass) {
2323 decode = c->in(1);
2324 load_klass = c->in(1)->in(1);
2325 } else {
2326 load_klass = c->in(1);
2327 }
2328 if (load_klass->in(2)->is_AddP()) {
2329 Node* addp = load_klass->in(2);
2330 Node* obj = addp->in(AddPNode::Address);
2331 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
2332 if (obj_type->speculative_type_not_null() != NULL) {
2333 ciKlass* k = obj_type->speculative_type();
2334 inc_sp(2);
2335 obj = maybe_cast_profiled_obj(obj, k);
2336 dec_sp(2);
2337 if (obj->is_ValueType()) {
2338 assert(obj->as_ValueType()->is_allocated(&_gvn), "must be allocated");
2339 obj = obj->as_ValueType()->get_oop();
2340 }
2341 // Make the CmpP use the casted obj
2342 addp = basic_plus_adr(obj, addp->in(AddPNode::Offset));
2343 load_klass = load_klass->clone();
2344 load_klass->set_req(2, addp);
2345 load_klass = _gvn.transform(load_klass);
2346 if (decode != NULL) {
2347 decode = decode->clone();
2348 decode->set_req(1, load_klass);
2349 load_klass = _gvn.transform(decode);
2350 }
2351 c = c->clone();
2352 c->set_req(1, load_klass);
2353 c = _gvn.transform(c);
2354 }
2355 }
2356 }
2357 return c;
2358 }
2359
2360 //------------------------------do_one_bytecode--------------------------------
2361 // Parse this bytecode, and alter the Parsers JVM->Node mapping
2362 void Parse::do_one_bytecode() {
2363 Node *a, *b, *c, *d; // Handy temps
2364 BoolTest::mask btest;
2365 int i;
2366
2367 assert(!has_exceptions(), "bytecode entry state must be clear of throws");
2368
2369 if (C->check_node_count(NodeLimitFudgeFactor * 5,
2370 "out of nodes parsing method")) {
2371 return;
2372 }
2373
2374 #ifdef ASSERT
2375 // for setting breakpoints
2376 if (TraceOptoParse) {
2377 tty->print(" @");
2378 dump_bci(bci());
2379 tty->cr();
2380 }
2381 #endif
2382
2383 switch (bc()) {
2384 case Bytecodes::_nop:
2385 // do nothing
2386 break;
2387 case Bytecodes::_lconst_0:
2388 push_pair(longcon(0));
2389 break;
2390
2391 case Bytecodes::_lconst_1:
2392 push_pair(longcon(1));
2393 break;
2394
2395 case Bytecodes::_fconst_0:
2396 push(zerocon(T_FLOAT));
2397 break;
2398
2399 case Bytecodes::_fconst_1:
2400 push(makecon(TypeF::ONE));
2401 break;
2402
2403 case Bytecodes::_fconst_2:
2404 push(makecon(TypeF::make(2.0f)));
2405 break;
2406
2407 case Bytecodes::_dconst_0:
2408 push_pair(zerocon(T_DOUBLE));
2409 break;
2410
2411 case Bytecodes::_dconst_1:
2412 push_pair(makecon(TypeD::ONE));
2413 break;
2414
2415 case Bytecodes::_iconst_m1:push(intcon(-1)); break;
2416 case Bytecodes::_iconst_0: push(intcon( 0)); break;
2417 case Bytecodes::_iconst_1: push(intcon( 1)); break;
2418 case Bytecodes::_iconst_2: push(intcon( 2)); break;
2419 case Bytecodes::_iconst_3: push(intcon( 3)); break;
2420 case Bytecodes::_iconst_4: push(intcon( 4)); break;
2421 case Bytecodes::_iconst_5: push(intcon( 5)); break;
2422 case Bytecodes::_bipush: push(intcon(iter().get_constant_u1())); break;
2423 case Bytecodes::_sipush: push(intcon(iter().get_constant_u2())); break;
2424 case Bytecodes::_aconst_null: push(null()); break;
2425 case Bytecodes::_ldc:
2426 case Bytecodes::_ldc_w:
2427 case Bytecodes::_ldc2_w:
2428 // If the constant is unresolved, run this BC once in the interpreter.
2429 {
2430 ciConstant constant = iter().get_constant();
2431 if (!constant.is_valid() ||
2432 (constant.basic_type() == T_OBJECT &&
2433 !constant.as_object()->is_loaded())) {
2434 int index = iter().get_constant_pool_index();
2435 constantTag tag = iter().get_constant_pool_tag(index);
2436 uncommon_trap(Deoptimization::make_trap_request
2437 (Deoptimization::Reason_unloaded,
2438 Deoptimization::Action_reinterpret,
2439 index),
2440 NULL, tag.internal_name());
2441 break;
2442 }
2443 assert(constant.basic_type() != T_OBJECT || constant.as_object()->is_instance(),
2444 "must be java_mirror of klass");
2445 const Type* con_type = Type::make_from_constant(constant);
2446 if (con_type != NULL) {
2447 push_node(con_type->basic_type(), makecon(con_type));
2448 }
2449 }
2450
2451 break;
2452
2453 case Bytecodes::_aload_0:
2454 push( local(0) );
2455 break;
2456 case Bytecodes::_aload_1:
2457 push( local(1) );
2458 break;
2459 case Bytecodes::_aload_2:
2460 push( local(2) );
2461 break;
2462 case Bytecodes::_aload_3:
2463 push( local(3) );
2464 break;
2465 case Bytecodes::_aload:
2466 push( local(iter().get_index()) );
2467 break;
2468
2469 case Bytecodes::_fload_0:
2470 case Bytecodes::_iload_0:
2471 push( local(0) );
2472 break;
2473 case Bytecodes::_fload_1:
2474 case Bytecodes::_iload_1:
2475 push( local(1) );
2476 break;
2477 case Bytecodes::_fload_2:
2478 case Bytecodes::_iload_2:
2479 push( local(2) );
2480 break;
2481 case Bytecodes::_fload_3:
2482 case Bytecodes::_iload_3:
2483 push( local(3) );
2484 break;
2485 case Bytecodes::_fload:
2486 case Bytecodes::_iload:
2487 push( local(iter().get_index()) );
2488 break;
2489 case Bytecodes::_lload_0:
2490 push_pair_local( 0 );
2491 break;
2492 case Bytecodes::_lload_1:
2493 push_pair_local( 1 );
2494 break;
2495 case Bytecodes::_lload_2:
2496 push_pair_local( 2 );
2497 break;
2498 case Bytecodes::_lload_3:
2499 push_pair_local( 3 );
2500 break;
2501 case Bytecodes::_lload:
2502 push_pair_local( iter().get_index() );
2503 break;
2504
2505 case Bytecodes::_dload_0:
2506 push_pair_local(0);
2507 break;
2508 case Bytecodes::_dload_1:
2509 push_pair_local(1);
2510 break;
2511 case Bytecodes::_dload_2:
2512 push_pair_local(2);
2513 break;
2514 case Bytecodes::_dload_3:
2515 push_pair_local(3);
2516 break;
2517 case Bytecodes::_dload:
2518 push_pair_local(iter().get_index());
2519 break;
2520 case Bytecodes::_fstore_0:
2521 case Bytecodes::_istore_0:
2522 case Bytecodes::_astore_0:
2523 set_local( 0, pop() );
2524 break;
2525 case Bytecodes::_fstore_1:
2526 case Bytecodes::_istore_1:
2527 case Bytecodes::_astore_1:
2528 set_local( 1, pop() );
2529 break;
2530 case Bytecodes::_fstore_2:
2531 case Bytecodes::_istore_2:
2532 case Bytecodes::_astore_2:
2533 set_local( 2, pop() );
2534 break;
2535 case Bytecodes::_fstore_3:
2536 case Bytecodes::_istore_3:
2537 case Bytecodes::_astore_3:
2538 set_local( 3, pop() );
2539 break;
2540 case Bytecodes::_fstore:
2541 case Bytecodes::_istore:
2542 case Bytecodes::_astore:
2543 set_local( iter().get_index(), pop() );
2544 break;
2545 // long stores
2546 case Bytecodes::_lstore_0:
2547 set_pair_local( 0, pop_pair() );
2548 break;
2549 case Bytecodes::_lstore_1:
2550 set_pair_local( 1, pop_pair() );
2551 break;
2552 case Bytecodes::_lstore_2:
2553 set_pair_local( 2, pop_pair() );
2554 break;
2555 case Bytecodes::_lstore_3:
2556 set_pair_local( 3, pop_pair() );
2557 break;
2558 case Bytecodes::_lstore:
2559 set_pair_local( iter().get_index(), pop_pair() );
2560 break;
2561
2562 // double stores
2563 case Bytecodes::_dstore_0:
2564 set_pair_local( 0, dstore_rounding(pop_pair()) );
2565 break;
2566 case Bytecodes::_dstore_1:
2567 set_pair_local( 1, dstore_rounding(pop_pair()) );
2568 break;
2569 case Bytecodes::_dstore_2:
2570 set_pair_local( 2, dstore_rounding(pop_pair()) );
2571 break;
2572 case Bytecodes::_dstore_3:
2573 set_pair_local( 3, dstore_rounding(pop_pair()) );
2574 break;
2575 case Bytecodes::_dstore:
2576 set_pair_local( iter().get_index(), dstore_rounding(pop_pair()) );
2577 break;
2578
2579 case Bytecodes::_pop: dec_sp(1); break;
2580 case Bytecodes::_pop2: dec_sp(2); break;
2581 case Bytecodes::_swap:
2582 a = pop();
2583 b = pop();
2584 push(a);
2585 push(b);
2586 break;
2587 case Bytecodes::_dup:
2588 a = pop();
2589 push(a);
2590 push(a);
2591 break;
2592 case Bytecodes::_dup_x1:
2593 a = pop();
2594 b = pop();
2595 push( a );
2596 push( b );
2597 push( a );
2598 break;
2599 case Bytecodes::_dup_x2:
2600 a = pop();
2601 b = pop();
2602 c = pop();
2603 push( a );
2604 push( c );
2605 push( b );
2606 push( a );
2607 break;
2608 case Bytecodes::_dup2:
2609 a = pop();
2610 b = pop();
2611 push( b );
2612 push( a );
2613 push( b );
2614 push( a );
2615 break;
2616
2617 case Bytecodes::_dup2_x1:
2618 // before: .. c, b, a
2619 // after: .. b, a, c, b, a
2620 // not tested
2621 a = pop();
2622 b = pop();
2623 c = pop();
2624 push( b );
2625 push( a );
2626 push( c );
2627 push( b );
2628 push( a );
2629 break;
2630 case Bytecodes::_dup2_x2:
2631 // before: .. d, c, b, a
2632 // after: .. b, a, d, c, b, a
2633 // not tested
2634 a = pop();
2635 b = pop();
2636 c = pop();
2637 d = pop();
2638 push( b );
2639 push( a );
2640 push( d );
2641 push( c );
2642 push( b );
2643 push( a );
2644 break;
2645
2646 case Bytecodes::_arraylength: {
2647 // Must do null-check with value on expression stack
2648 Node *ary = null_check(peek(), T_ARRAY);
2649 // Compile-time detect of null-exception?
2650 if (stopped()) return;
2651 a = pop();
2652 push(load_array_length(a));
2653 break;
2654 }
2655
2656 case Bytecodes::_baload: array_load(T_BYTE); break;
2657 case Bytecodes::_caload: array_load(T_CHAR); break;
2658 case Bytecodes::_iaload: array_load(T_INT); break;
2659 case Bytecodes::_saload: array_load(T_SHORT); break;
2660 case Bytecodes::_faload: array_load(T_FLOAT); break;
2661 case Bytecodes::_aaload: array_load(T_OBJECT); break;
2662 case Bytecodes::_laload: array_load(T_LONG); break;
2663 case Bytecodes::_daload: array_load(T_DOUBLE); break;
2664 case Bytecodes::_bastore: array_store(T_BYTE); break;
2665 case Bytecodes::_castore: array_store(T_CHAR); break;
2666 case Bytecodes::_iastore: array_store(T_INT); break;
2667 case Bytecodes::_sastore: array_store(T_SHORT); break;
2668 case Bytecodes::_fastore: array_store(T_FLOAT); break;
2669 case Bytecodes::_aastore: array_store(T_OBJECT); break;
2670 case Bytecodes::_lastore: array_store(T_LONG); break;
2671 case Bytecodes::_dastore: array_store(T_DOUBLE); break;
2672
2673 case Bytecodes::_getfield:
2674 do_getfield();
2675 break;
2676
2677 case Bytecodes::_getstatic:
2678 do_getstatic();
2679 break;
2680
2681 case Bytecodes::_putfield:
2682 do_putfield();
2683 break;
2684
2685 case Bytecodes::_putstatic:
2686 do_putstatic();
2687 break;
2688
2689 case Bytecodes::_irem:
2690 do_irem();
2691 break;
2692 case Bytecodes::_idiv:
2693 // Must keep both values on the expression-stack during null-check
2694 zero_check_int(peek());
2695 // Compile-time detect of null-exception?
2696 if (stopped()) return;
2697 b = pop();
2698 a = pop();
2699 push( _gvn.transform( new DivINode(control(),a,b) ) );
2700 break;
2701 case Bytecodes::_imul:
2702 b = pop(); a = pop();
2703 push( _gvn.transform( new MulINode(a,b) ) );
2704 break;
2705 case Bytecodes::_iadd:
2706 b = pop(); a = pop();
2707 push( _gvn.transform( new AddINode(a,b) ) );
2708 break;
2709 case Bytecodes::_ineg:
2710 a = pop();
2711 push( _gvn.transform( new SubINode(_gvn.intcon(0),a)) );
2712 break;
2713 case Bytecodes::_isub:
2714 b = pop(); a = pop();
2715 push( _gvn.transform( new SubINode(a,b) ) );
2716 break;
2717 case Bytecodes::_iand:
2718 b = pop(); a = pop();
2719 push( _gvn.transform( new AndINode(a,b) ) );
2720 break;
2721 case Bytecodes::_ior:
2722 b = pop(); a = pop();
2723 push( _gvn.transform( new OrINode(a,b) ) );
2724 break;
2725 case Bytecodes::_ixor:
2726 b = pop(); a = pop();
2727 push( _gvn.transform( new XorINode(a,b) ) );
2728 break;
2729 case Bytecodes::_ishl:
2730 b = pop(); a = pop();
2731 push( _gvn.transform( new LShiftINode(a,b) ) );
2732 break;
2733 case Bytecodes::_ishr:
2734 b = pop(); a = pop();
2735 push( _gvn.transform( new RShiftINode(a,b) ) );
2736 break;
2737 case Bytecodes::_iushr:
2738 b = pop(); a = pop();
2739 push( _gvn.transform( new URShiftINode(a,b) ) );
2740 break;
2741
2742 case Bytecodes::_fneg:
2743 a = pop();
2744 b = _gvn.transform(new NegFNode (a));
2745 push(b);
2746 break;
2747
2748 case Bytecodes::_fsub:
2749 b = pop();
2750 a = pop();
2751 c = _gvn.transform( new SubFNode(a,b) );
2752 d = precision_rounding(c);
2753 push( d );
2754 break;
2755
2756 case Bytecodes::_fadd:
2757 b = pop();
2758 a = pop();
2759 c = _gvn.transform( new AddFNode(a,b) );
2760 d = precision_rounding(c);
2761 push( d );
2762 break;
2763
2764 case Bytecodes::_fmul:
2765 b = pop();
2766 a = pop();
2767 c = _gvn.transform( new MulFNode(a,b) );
2768 d = precision_rounding(c);
2769 push( d );
2770 break;
2771
2772 case Bytecodes::_fdiv:
2773 b = pop();
2774 a = pop();
2775 c = _gvn.transform( new DivFNode(0,a,b) );
2776 d = precision_rounding(c);
2777 push( d );
2778 break;
2779
2780 case Bytecodes::_frem:
2781 if (Matcher::has_match_rule(Op_ModF)) {
2782 // Generate a ModF node.
2783 b = pop();
2784 a = pop();
2785 c = _gvn.transform( new ModFNode(0,a,b) );
2786 d = precision_rounding(c);
2787 push( d );
2788 }
2789 else {
2790 // Generate a call.
2791 modf();
2792 }
2793 break;
2794
2795 case Bytecodes::_fcmpl:
2796 b = pop();
2797 a = pop();
2798 c = _gvn.transform( new CmpF3Node( a, b));
2799 push(c);
2800 break;
2801 case Bytecodes::_fcmpg:
2802 b = pop();
2803 a = pop();
2804
2805 // Same as fcmpl but need to flip the unordered case. Swap the inputs,
2806 // which negates the result sign except for unordered. Flip the unordered
2807 // as well by using CmpF3 which implements unordered-lesser instead of
2808 // unordered-greater semantics. Finally, commute the result bits. Result
2809 // is same as using a CmpF3Greater except we did it with CmpF3 alone.
2810 c = _gvn.transform( new CmpF3Node( b, a));
2811 c = _gvn.transform( new SubINode(_gvn.intcon(0),c) );
2812 push(c);
2813 break;
2814
2815 case Bytecodes::_f2i:
2816 a = pop();
2817 push(_gvn.transform(new ConvF2INode(a)));
2818 break;
2819
2820 case Bytecodes::_d2i:
2821 a = pop_pair();
2822 b = _gvn.transform(new ConvD2INode(a));
2823 push( b );
2824 break;
2825
2826 case Bytecodes::_f2d:
2827 a = pop();
2828 b = _gvn.transform( new ConvF2DNode(a));
2829 push_pair( b );
2830 break;
2831
2832 case Bytecodes::_d2f:
2833 a = pop_pair();
2834 b = _gvn.transform( new ConvD2FNode(a));
2835 // This breaks _227_mtrt (speed & correctness) and _222_mpegaudio (speed)
2836 //b = _gvn.transform(new RoundFloatNode(0, b) );
2837 push( b );
2838 break;
2839
2840 case Bytecodes::_l2f:
2841 if (Matcher::convL2FSupported()) {
2842 a = pop_pair();
2843 b = _gvn.transform( new ConvL2FNode(a));
2844 // For i486.ad, FILD doesn't restrict precision to 24 or 53 bits.
2845 // Rather than storing the result into an FP register then pushing
2846 // out to memory to round, the machine instruction that implements
2847 // ConvL2D is responsible for rounding.
2848 // c = precision_rounding(b);
2849 c = _gvn.transform(b);
2850 push(c);
2851 } else {
2852 l2f();
2853 }
2854 break;
2855
2856 case Bytecodes::_l2d:
2857 a = pop_pair();
2858 b = _gvn.transform( new ConvL2DNode(a));
2859 // For i486.ad, rounding is always necessary (see _l2f above).
2860 // c = dprecision_rounding(b);
2861 c = _gvn.transform(b);
2862 push_pair(c);
2863 break;
2864
2865 case Bytecodes::_f2l:
2866 a = pop();
2867 b = _gvn.transform( new ConvF2LNode(a));
2868 push_pair(b);
2869 break;
2870
2871 case Bytecodes::_d2l:
2872 a = pop_pair();
2873 b = _gvn.transform( new ConvD2LNode(a));
2874 push_pair(b);
2875 break;
2876
2877 case Bytecodes::_dsub:
2878 b = pop_pair();
2879 a = pop_pair();
2880 c = _gvn.transform( new SubDNode(a,b) );
2881 d = dprecision_rounding(c);
2882 push_pair( d );
2883 break;
2884
2885 case Bytecodes::_dadd:
2886 b = pop_pair();
2887 a = pop_pair();
2888 c = _gvn.transform( new AddDNode(a,b) );
2889 d = dprecision_rounding(c);
2890 push_pair( d );
2891 break;
2892
2893 case Bytecodes::_dmul:
2894 b = pop_pair();
2895 a = pop_pair();
2896 c = _gvn.transform( new MulDNode(a,b) );
2897 d = dprecision_rounding(c);
2898 push_pair( d );
2899 break;
2900
2901 case Bytecodes::_ddiv:
2902 b = pop_pair();
2903 a = pop_pair();
2904 c = _gvn.transform( new DivDNode(0,a,b) );
2905 d = dprecision_rounding(c);
2906 push_pair( d );
2907 break;
2908
2909 case Bytecodes::_dneg:
2910 a = pop_pair();
2911 b = _gvn.transform(new NegDNode (a));
2912 push_pair(b);
2913 break;
2914
2915 case Bytecodes::_drem:
2916 if (Matcher::has_match_rule(Op_ModD)) {
2917 // Generate a ModD node.
2918 b = pop_pair();
2919 a = pop_pair();
2920 // a % b
2921
2922 c = _gvn.transform( new ModDNode(0,a,b) );
2923 d = dprecision_rounding(c);
2924 push_pair( d );
2925 }
2926 else {
2927 // Generate a call.
2928 modd();
2929 }
2930 break;
2931
2932 case Bytecodes::_dcmpl:
2933 b = pop_pair();
2934 a = pop_pair();
2935 c = _gvn.transform( new CmpD3Node( a, b));
2936 push(c);
2937 break;
2938
2939 case Bytecodes::_dcmpg:
2940 b = pop_pair();
2941 a = pop_pair();
2942 // Same as dcmpl but need to flip the unordered case.
2943 // Commute the inputs, which negates the result sign except for unordered.
2944 // Flip the unordered as well by using CmpD3 which implements
2945 // unordered-lesser instead of unordered-greater semantics.
2946 // Finally, negate the result bits. Result is same as using a
2947 // CmpD3Greater except we did it with CmpD3 alone.
2948 c = _gvn.transform( new CmpD3Node( b, a));
2949 c = _gvn.transform( new SubINode(_gvn.intcon(0),c) );
2950 push(c);
2951 break;
2952
2953
2954 // Note for longs -> lo word is on TOS, hi word is on TOS - 1
2955 case Bytecodes::_land:
2956 b = pop_pair();
2957 a = pop_pair();
2958 c = _gvn.transform( new AndLNode(a,b) );
2959 push_pair(c);
2960 break;
2961 case Bytecodes::_lor:
2962 b = pop_pair();
2963 a = pop_pair();
2964 c = _gvn.transform( new OrLNode(a,b) );
2965 push_pair(c);
2966 break;
2967 case Bytecodes::_lxor:
2968 b = pop_pair();
2969 a = pop_pair();
2970 c = _gvn.transform( new XorLNode(a,b) );
2971 push_pair(c);
2972 break;
2973
2974 case Bytecodes::_lshl:
2975 b = pop(); // the shift count
2976 a = pop_pair(); // value to be shifted
2977 c = _gvn.transform( new LShiftLNode(a,b) );
2978 push_pair(c);
2979 break;
2980 case Bytecodes::_lshr:
2981 b = pop(); // the shift count
2982 a = pop_pair(); // value to be shifted
2983 c = _gvn.transform( new RShiftLNode(a,b) );
2984 push_pair(c);
2985 break;
2986 case Bytecodes::_lushr:
2987 b = pop(); // the shift count
2988 a = pop_pair(); // value to be shifted
2989 c = _gvn.transform( new URShiftLNode(a,b) );
2990 push_pair(c);
2991 break;
2992 case Bytecodes::_lmul:
2993 b = pop_pair();
2994 a = pop_pair();
2995 c = _gvn.transform( new MulLNode(a,b) );
2996 push_pair(c);
2997 break;
2998
2999 case Bytecodes::_lrem:
3000 // Must keep both values on the expression-stack during null-check
3001 assert(peek(0) == top(), "long word order");
3002 zero_check_long(peek(1));
3003 // Compile-time detect of null-exception?
3004 if (stopped()) return;
3005 b = pop_pair();
3006 a = pop_pair();
3007 c = _gvn.transform( new ModLNode(control(),a,b) );
3008 push_pair(c);
3009 break;
3010
3011 case Bytecodes::_ldiv:
3012 // Must keep both values on the expression-stack during null-check
3013 assert(peek(0) == top(), "long word order");
3014 zero_check_long(peek(1));
3015 // Compile-time detect of null-exception?
3016 if (stopped()) return;
3017 b = pop_pair();
3018 a = pop_pair();
3019 c = _gvn.transform( new DivLNode(control(),a,b) );
3020 push_pair(c);
3021 break;
3022
3023 case Bytecodes::_ladd:
3024 b = pop_pair();
3025 a = pop_pair();
3026 c = _gvn.transform( new AddLNode(a,b) );
3027 push_pair(c);
3028 break;
3029 case Bytecodes::_lsub:
3030 b = pop_pair();
3031 a = pop_pair();
3032 c = _gvn.transform( new SubLNode(a,b) );
3033 push_pair(c);
3034 break;
3035 case Bytecodes::_lcmp:
3036 // Safepoints are now inserted _before_ branches. The long-compare
3037 // bytecode painfully produces a 3-way value (-1,0,+1) which requires a
3038 // slew of control flow. These are usually followed by a CmpI vs zero and
3039 // a branch; this pattern then optimizes to the obvious long-compare and
3040 // branch. However, if the branch is backwards there's a Safepoint
3041 // inserted. The inserted Safepoint captures the JVM state at the
3042 // pre-branch point, i.e. it captures the 3-way value. Thus if a
3043 // long-compare is used to control a loop the debug info will force
3044 // computation of the 3-way value, even though the generated code uses a
3045 // long-compare and branch. We try to rectify the situation by inserting
3046 // a SafePoint here and have it dominate and kill the safepoint added at a
3047 // following backwards branch. At this point the JVM state merely holds 2
3048 // longs but not the 3-way value.
3049 if( UseLoopSafepoints ) {
3050 switch( iter().next_bc() ) {
3051 case Bytecodes::_ifgt:
3052 case Bytecodes::_iflt:
3053 case Bytecodes::_ifge:
3054 case Bytecodes::_ifle:
3055 case Bytecodes::_ifne:
3056 case Bytecodes::_ifeq:
3057 // If this is a backwards branch in the bytecodes, add Safepoint
3058 maybe_add_safepoint(iter().next_get_dest());
3059 default:
3060 break;
3061 }
3062 }
3063 b = pop_pair();
3064 a = pop_pair();
3065 c = _gvn.transform( new CmpL3Node( a, b ));
3066 push(c);
3067 break;
3068
3069 case Bytecodes::_lneg:
3070 a = pop_pair();
3071 b = _gvn.transform( new SubLNode(longcon(0),a));
3072 push_pair(b);
3073 break;
3074 case Bytecodes::_l2i:
3075 a = pop_pair();
3076 push( _gvn.transform( new ConvL2INode(a)));
3077 break;
3078 case Bytecodes::_i2l:
3079 a = pop();
3080 b = _gvn.transform( new ConvI2LNode(a));
3081 push_pair(b);
3082 break;
3083 case Bytecodes::_i2b:
3084 // Sign extend
3085 a = pop();
3086 a = _gvn.transform( new LShiftINode(a,_gvn.intcon(24)) );
3087 a = _gvn.transform( new RShiftINode(a,_gvn.intcon(24)) );
3088 push( a );
3089 break;
3090 case Bytecodes::_i2s:
3091 a = pop();
3092 a = _gvn.transform( new LShiftINode(a,_gvn.intcon(16)) );
3093 a = _gvn.transform( new RShiftINode(a,_gvn.intcon(16)) );
3094 push( a );
3095 break;
3096 case Bytecodes::_i2c:
3097 a = pop();
3098 push( _gvn.transform( new AndINode(a,_gvn.intcon(0xFFFF)) ) );
3099 break;
3100
3101 case Bytecodes::_i2f:
3102 a = pop();
3103 b = _gvn.transform( new ConvI2FNode(a) ) ;
3104 c = precision_rounding(b);
3105 push (b);
3106 break;
3107
3108 case Bytecodes::_i2d:
3109 a = pop();
3110 b = _gvn.transform( new ConvI2DNode(a));
3111 push_pair(b);
3112 break;
3113
3114 case Bytecodes::_iinc: // Increment local
3115 i = iter().get_index(); // Get local index
3116 set_local( i, _gvn.transform( new AddINode( _gvn.intcon(iter().get_iinc_con()), local(i) ) ) );
3117 break;
3118
3119 // Exit points of synchronized methods must have an unlock node
3120 case Bytecodes::_return:
3121 return_current(NULL);
3122 break;
3123
3124 case Bytecodes::_ireturn:
3125 case Bytecodes::_areturn:
3126 case Bytecodes::_freturn:
3127 return_current(pop());
3128 break;
3129 case Bytecodes::_lreturn:
3130 return_current(pop_pair());
3131 break;
3132 case Bytecodes::_dreturn:
3133 return_current(pop_pair());
3134 break;
3135
3136 case Bytecodes::_athrow:
3137 // null exception oop throws NULL pointer exception
3138 null_check(peek());
3139 if (stopped()) return;
3140 // Hook the thrown exception directly to subsequent handlers.
3141 if (BailoutToInterpreterForThrows) {
3142 // Keep method interpreted from now on.
3143 uncommon_trap(Deoptimization::Reason_unhandled,
3144 Deoptimization::Action_make_not_compilable);
3145 return;
3146 }
3147 if (env()->jvmti_can_post_on_exceptions()) {
3148 // check if we must post exception events, take uncommon trap if so (with must_throw = false)
3149 uncommon_trap_if_should_post_on_exceptions(Deoptimization::Reason_unhandled, false);
3150 }
3151 // Here if either can_post_on_exceptions or should_post_on_exceptions is false
3152 add_exception_state(make_exception_state(peek()));
3153 break;
3154
3155 case Bytecodes::_goto: // fall through
3156 case Bytecodes::_goto_w: {
3157 int target_bci = (bc() == Bytecodes::_goto) ? iter().get_dest() : iter().get_far_dest();
3158
3159 // If this is a backwards branch in the bytecodes, add Safepoint
3160 maybe_add_safepoint(target_bci);
3161
3162 // Update method data
3163 profile_taken_branch(target_bci);
3164
3165 // Merge the current control into the target basic block
3166 merge(target_bci);
3167
3168 // See if we can get some profile data and hand it off to the next block
3169 Block *target_block = block()->successor_for_bci(target_bci);
3170 if (target_block->pred_count() != 1) break;
3171 ciMethodData* methodData = method()->method_data();
3172 if (!methodData->is_mature()) break;
3173 ciProfileData* data = methodData->bci_to_data(bci());
3174 assert(data != NULL && data->is_JumpData(), "need JumpData for taken branch");
3175 int taken = ((ciJumpData*)data)->taken();
3176 taken = method()->scale_count(taken);
3177 target_block->set_count(taken);
3178 break;
3179 }
3180
3181 case Bytecodes::_ifnull: btest = BoolTest::eq; goto handle_if_null;
3182 case Bytecodes::_ifnonnull: btest = BoolTest::ne; goto handle_if_null;
3183 handle_if_null:
3184 // If this is a backwards branch in the bytecodes, add Safepoint
3185 maybe_add_safepoint(iter().get_dest());
3186 a = null();
3187 b = pop();
3188 if (b->is_ValueType()) {
3189 // Return constant false because 'b' is always non-null
3190 c = _gvn.makecon(TypeInt::CC_GT);
3191 } else {
3192 if (!_gvn.type(b)->speculative_maybe_null() &&
3193 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
3194 inc_sp(1);
3195 Node* null_ctl = top();
3196 b = null_check_oop(b, &null_ctl, true, true, true);
3197 assert(null_ctl->is_top(), "no null control here");
3198 dec_sp(1);
3199 } else if (_gvn.type(b)->speculative_always_null() &&
3200 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
3201 inc_sp(1);
3202 b = null_assert(b);
3203 dec_sp(1);
3204 }
3205 c = _gvn.transform( new CmpPNode(b, a) );
3206 }
3207 do_ifnull(btest, c);
3208 break;
3209
3210 case Bytecodes::_if_acmpeq: btest = BoolTest::eq; goto handle_if_acmp;
3211 case Bytecodes::_if_acmpne: btest = BoolTest::ne; goto handle_if_acmp;
3212 handle_if_acmp:
3213 // If this is a backwards branch in the bytecodes, add Safepoint
3214 maybe_add_safepoint(iter().get_dest());
3215 a = access_resolve(pop(), 0);
3216 b = access_resolve(pop(), 0);
3217 do_acmp(btest, a, b);
3218 break;
3219
3220 case Bytecodes::_ifeq: btest = BoolTest::eq; goto handle_ifxx;
3221 case Bytecodes::_ifne: btest = BoolTest::ne; goto handle_ifxx;
3222 case Bytecodes::_iflt: btest = BoolTest::lt; goto handle_ifxx;
3223 case Bytecodes::_ifle: btest = BoolTest::le; goto handle_ifxx;
3224 case Bytecodes::_ifgt: btest = BoolTest::gt; goto handle_ifxx;
3225 case Bytecodes::_ifge: btest = BoolTest::ge; goto handle_ifxx;
3226 handle_ifxx:
3227 // If this is a backwards branch in the bytecodes, add Safepoint
3228 maybe_add_safepoint(iter().get_dest());
3229 a = _gvn.intcon(0);
3230 b = pop();
3231 c = _gvn.transform( new CmpINode(b, a) );
3232 do_if(btest, c);
3233 break;
3234
3235 case Bytecodes::_if_icmpeq: btest = BoolTest::eq; goto handle_if_icmp;
3236 case Bytecodes::_if_icmpne: btest = BoolTest::ne; goto handle_if_icmp;
3237 case Bytecodes::_if_icmplt: btest = BoolTest::lt; goto handle_if_icmp;
3238 case Bytecodes::_if_icmple: btest = BoolTest::le; goto handle_if_icmp;
3239 case Bytecodes::_if_icmpgt: btest = BoolTest::gt; goto handle_if_icmp;
3240 case Bytecodes::_if_icmpge: btest = BoolTest::ge; goto handle_if_icmp;
3241 handle_if_icmp:
3242 // If this is a backwards branch in the bytecodes, add Safepoint
3243 maybe_add_safepoint(iter().get_dest());
3244 a = pop();
3245 b = pop();
3246 c = _gvn.transform( new CmpINode( b, a ) );
3247 do_if(btest, c);
3248 break;
3249
3250 case Bytecodes::_tableswitch:
3251 do_tableswitch();
3252 break;
3253
3254 case Bytecodes::_lookupswitch:
3255 do_lookupswitch();
3256 break;
3257
3258 case Bytecodes::_invokestatic:
3259 case Bytecodes::_invokedynamic:
3260 case Bytecodes::_invokespecial:
3261 case Bytecodes::_invokevirtual:
3262 case Bytecodes::_invokeinterface:
3263 do_call();
3264 break;
3265 case Bytecodes::_checkcast:
3266 do_checkcast();
3267 break;
3268 case Bytecodes::_instanceof:
3269 do_instanceof();
3270 break;
3271 case Bytecodes::_anewarray:
3272 do_newarray();
3273 break;
3274 case Bytecodes::_newarray:
3275 do_newarray((BasicType)iter().get_index());
3276 break;
3277 case Bytecodes::_multianewarray:
3278 do_multianewarray();
3279 break;
3280 case Bytecodes::_new:
3281 do_new();
3282 break;
3283 case Bytecodes::_defaultvalue:
3284 do_defaultvalue();
3285 break;
3286 case Bytecodes::_withfield:
3287 do_withfield();
3288 break;
3289
3290 case Bytecodes::_jsr:
3291 case Bytecodes::_jsr_w:
3292 do_jsr();
3293 break;
3294
3295 case Bytecodes::_ret:
3296 do_ret();
3297 break;
3298
3299
3300 case Bytecodes::_monitorenter:
3301 do_monitor_enter();
3302 break;
3303
3304 case Bytecodes::_monitorexit:
3305 do_monitor_exit();
3306 break;
3307
3308 case Bytecodes::_breakpoint:
3309 // Breakpoint set concurrently to compile
3310 // %%% use an uncommon trap?
3311 C->record_failure("breakpoint in method");
3312 return;
3313
3314 default:
3315 #ifndef PRODUCT
3316 map()->dump(99);
3317 #endif
3318 tty->print("\nUnhandled bytecode %s\n", Bytecodes::name(bc()) );
3319 ShouldNotReachHere();
3320 }
3321
3322 #ifndef PRODUCT
3323 IdealGraphPrinter *printer = C->printer();
3324 if (printer && printer->should_print(1)) {
3325 char buffer[256];
3326 jio_snprintf(buffer, sizeof(buffer), "Bytecode %d: %s", bci(), Bytecodes::name(bc()));
3327 bool old = printer->traverse_outs();
3328 printer->set_traverse_outs(true);
3329 printer->print_method(buffer, 4);
3330 printer->set_traverse_outs(old);
3331 }
3332 #endif
3333 }