1 /*
2 * Copyright (c) 1997, 2015, 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 "classfile/systemDictionary.hpp"
27 #include "compiler/compileLog.hpp"
28 #include "memory/allocation.inline.hpp"
29 #include "oops/objArrayKlass.hpp"
30 #include "opto/addnode.hpp"
31 #include "opto/cfgnode.hpp"
32 #include "opto/compile.hpp"
33 #include "opto/connode.hpp"
34 #include "opto/convertnode.hpp"
35 #include "opto/loopnode.hpp"
36 #include "opto/machnode.hpp"
37 #include "opto/matcher.hpp"
38 #include "opto/memnode.hpp"
39 #include "opto/mulnode.hpp"
40 #include "opto/narrowptrnode.hpp"
41 #include "opto/phaseX.hpp"
42 #include "opto/regmask.hpp"
43 #include "utilities/copy.hpp"
44
45 // Portions of code courtesy of Clifford Click
46
47 // Optimization - Graph Style
48
49 static Node *step_through_mergemem(PhaseGVN *phase, MergeMemNode *mmem, const TypePtr *tp, const TypePtr *adr_check, outputStream *st);
50
51 //=============================================================================
52 uint MemNode::size_of() const { return sizeof(*this); }
53
54 const TypePtr *MemNode::adr_type() const {
55 Node* adr = in(Address);
56 if (adr == NULL) return NULL; // node is dead
57 const TypePtr* cross_check = NULL;
58 DEBUG_ONLY(cross_check = _adr_type);
59 return calculate_adr_type(adr->bottom_type(), cross_check);
60 }
61
62 #ifndef PRODUCT
63 void MemNode::dump_spec(outputStream *st) const {
64 if (in(Address) == NULL) return; // node is dead
65 #ifndef ASSERT
66 // fake the missing field
67 const TypePtr* _adr_type = NULL;
68 if (in(Address) != NULL)
69 _adr_type = in(Address)->bottom_type()->isa_ptr();
70 #endif
71 dump_adr_type(this, _adr_type, st);
72
73 Compile* C = Compile::current();
74 if( C->alias_type(_adr_type)->is_volatile() )
75 st->print(" Volatile!");
76 }
77
78 void MemNode::dump_adr_type(const Node* mem, const TypePtr* adr_type, outputStream *st) {
79 st->print(" @");
80 if (adr_type == NULL) {
81 st->print("NULL");
82 } else {
83 adr_type->dump_on(st);
84 Compile* C = Compile::current();
85 Compile::AliasType* atp = NULL;
86 if (C->have_alias_type(adr_type)) atp = C->alias_type(adr_type);
87 if (atp == NULL)
88 st->print(", idx=?\?;");
89 else if (atp->index() == Compile::AliasIdxBot)
90 st->print(", idx=Bot;");
91 else if (atp->index() == Compile::AliasIdxTop)
92 st->print(", idx=Top;");
93 else if (atp->index() == Compile::AliasIdxRaw)
94 st->print(", idx=Raw;");
95 else {
96 ciField* field = atp->field();
97 if (field) {
98 st->print(", name=");
99 field->print_name_on(st);
100 }
101 st->print(", idx=%d;", atp->index());
102 }
103 }
104 }
105
106 extern void print_alias_types();
107
108 #endif
109
110 Node *MemNode::optimize_simple_memory_chain(Node *mchain, const TypeOopPtr *t_oop, Node *load, PhaseGVN *phase) {
111 assert((t_oop != NULL), "sanity");
112 bool is_instance = t_oop->is_known_instance_field();
113 bool is_boxed_value_load = t_oop->is_ptr_to_boxed_value() &&
114 (load != NULL) && load->is_Load() &&
115 (phase->is_IterGVN() != NULL);
116 if (!(is_instance || is_boxed_value_load))
117 return mchain; // don't try to optimize non-instance types
118 uint instance_id = t_oop->instance_id();
119 Node *start_mem = phase->C->start()->proj_out(TypeFunc::Memory);
120 Node *prev = NULL;
121 Node *result = mchain;
122 while (prev != result) {
123 prev = result;
124 if (result == start_mem)
125 break; // hit one of our sentinels
126 // skip over a call which does not affect this memory slice
127 if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) {
128 Node *proj_in = result->in(0);
129 if (proj_in->is_Allocate() && proj_in->_idx == instance_id) {
130 break; // hit one of our sentinels
131 } else if (proj_in->is_Call()) {
132 CallNode *call = proj_in->as_Call();
133 if (!call->may_modify(t_oop, phase)) { // returns false for instances
134 result = call->in(TypeFunc::Memory);
135 }
136 } else if (proj_in->is_Initialize()) {
137 AllocateNode* alloc = proj_in->as_Initialize()->allocation();
138 // Stop if this is the initialization for the object instance which
139 // which contains this memory slice, otherwise skip over it.
140 if ((alloc == NULL) || (alloc->_idx == instance_id)) {
141 break;
142 }
143 if (is_instance) {
144 result = proj_in->in(TypeFunc::Memory);
145 } else if (is_boxed_value_load) {
146 Node* klass = alloc->in(AllocateNode::KlassNode);
147 const TypeKlassPtr* tklass = phase->type(klass)->is_klassptr();
148 if (tklass->klass_is_exact() && !tklass->klass()->equals(t_oop->klass())) {
149 result = proj_in->in(TypeFunc::Memory); // not related allocation
150 }
151 }
152 } else if (proj_in->is_MemBar()) {
153 result = proj_in->in(TypeFunc::Memory);
154 } else {
155 assert(false, "unexpected projection");
156 }
157 } else if (result->is_ClearArray()) {
158 if (!is_instance || !ClearArrayNode::step_through(&result, instance_id, phase)) {
159 // Can not bypass initialization of the instance
160 // we are looking for.
161 break;
162 }
163 // Otherwise skip it (the call updated 'result' value).
164 } else if (result->is_MergeMem()) {
165 result = step_through_mergemem(phase, result->as_MergeMem(), t_oop, NULL, tty);
166 }
167 }
168 return result;
169 }
170
171 Node *MemNode::optimize_memory_chain(Node *mchain, const TypePtr *t_adr, Node *load, PhaseGVN *phase) {
172 const TypeOopPtr* t_oop = t_adr->isa_oopptr();
173 if (t_oop == NULL)
174 return mchain; // don't try to optimize non-oop types
175 Node* result = optimize_simple_memory_chain(mchain, t_oop, load, phase);
176 bool is_instance = t_oop->is_known_instance_field();
177 PhaseIterGVN *igvn = phase->is_IterGVN();
178 if (is_instance && igvn != NULL && result->is_Phi()) {
179 PhiNode *mphi = result->as_Phi();
180 assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
181 const TypePtr *t = mphi->adr_type();
182 if (t == TypePtr::BOTTOM || t == TypeRawPtr::BOTTOM ||
183 t->isa_oopptr() && !t->is_oopptr()->is_known_instance() &&
184 t->is_oopptr()->cast_to_exactness(true)
185 ->is_oopptr()->cast_to_ptr_type(t_oop->ptr())
186 ->is_oopptr()->cast_to_instance_id(t_oop->instance_id()) == t_oop) {
187 // clone the Phi with our address type
188 result = mphi->split_out_instance(t_adr, igvn);
189 } else {
190 assert(phase->C->get_alias_index(t) == phase->C->get_alias_index(t_adr), "correct memory chain");
191 }
192 }
193 return result;
194 }
195
196 static Node *step_through_mergemem(PhaseGVN *phase, MergeMemNode *mmem, const TypePtr *tp, const TypePtr *adr_check, outputStream *st) {
197 uint alias_idx = phase->C->get_alias_index(tp);
198 Node *mem = mmem;
199 #ifdef ASSERT
200 {
201 // Check that current type is consistent with the alias index used during graph construction
202 assert(alias_idx >= Compile::AliasIdxRaw, "must not be a bad alias_idx");
203 bool consistent = adr_check == NULL || adr_check->empty() ||
204 phase->C->must_alias(adr_check, alias_idx );
205 // Sometimes dead array references collapse to a[-1], a[-2], or a[-3]
206 if( !consistent && adr_check != NULL && !adr_check->empty() &&
207 tp->isa_aryptr() && tp->offset() == Type::OffsetBot &&
208 adr_check->isa_aryptr() && adr_check->offset() != Type::OffsetBot &&
209 ( adr_check->offset() == arrayOopDesc::length_offset_in_bytes() ||
210 adr_check->offset() == oopDesc::klass_offset_in_bytes() ||
211 adr_check->offset() == oopDesc::mark_offset_in_bytes() ) ) {
212 // don't assert if it is dead code.
213 consistent = true;
214 }
215 if( !consistent ) {
216 st->print("alias_idx==%d, adr_check==", alias_idx);
217 if( adr_check == NULL ) {
218 st->print("NULL");
219 } else {
220 adr_check->dump();
221 }
222 st->cr();
223 print_alias_types();
224 assert(consistent, "adr_check must match alias idx");
225 }
226 }
227 #endif
228 // TypeOopPtr::NOTNULL+any is an OOP with unknown offset - generally
229 // means an array I have not precisely typed yet. Do not do any
230 // alias stuff with it any time soon.
231 const TypeOopPtr *toop = tp->isa_oopptr();
232 if( tp->base() != Type::AnyPtr &&
233 !(toop &&
234 toop->klass() != NULL &&
235 toop->klass()->is_java_lang_Object() &&
236 toop->offset() == Type::OffsetBot) ) {
237 // compress paths and change unreachable cycles to TOP
238 // If not, we can update the input infinitely along a MergeMem cycle
239 // Equivalent code in PhiNode::Ideal
240 Node* m = phase->transform(mmem);
241 // If transformed to a MergeMem, get the desired slice
242 // Otherwise the returned node represents memory for every slice
243 mem = (m->is_MergeMem())? m->as_MergeMem()->memory_at(alias_idx) : m;
244 // Update input if it is progress over what we have now
245 }
246 return mem;
247 }
248
249 //--------------------------Ideal_common---------------------------------------
250 // Look for degenerate control and memory inputs. Bypass MergeMem inputs.
251 // Unhook non-raw memories from complete (macro-expanded) initializations.
252 Node *MemNode::Ideal_common(PhaseGVN *phase, bool can_reshape) {
253 // If our control input is a dead region, kill all below the region
254 Node *ctl = in(MemNode::Control);
255 if (ctl && remove_dead_region(phase, can_reshape))
256 return this;
257 ctl = in(MemNode::Control);
258 // Don't bother trying to transform a dead node
259 if (ctl && ctl->is_top()) return NodeSentinel;
260
261 PhaseIterGVN *igvn = phase->is_IterGVN();
262 // Wait if control on the worklist.
263 if (ctl && can_reshape && igvn != NULL) {
264 Node* bol = NULL;
265 Node* cmp = NULL;
266 if (ctl->in(0)->is_If()) {
267 assert(ctl->is_IfTrue() || ctl->is_IfFalse(), "sanity");
268 bol = ctl->in(0)->in(1);
269 if (bol->is_Bool())
270 cmp = ctl->in(0)->in(1)->in(1);
271 }
272 if (igvn->_worklist.member(ctl) ||
273 (bol != NULL && igvn->_worklist.member(bol)) ||
274 (cmp != NULL && igvn->_worklist.member(cmp)) ) {
275 // This control path may be dead.
276 // Delay this memory node transformation until the control is processed.
277 phase->is_IterGVN()->_worklist.push(this);
278 return NodeSentinel; // caller will return NULL
279 }
280 }
281 // Ignore if memory is dead, or self-loop
282 Node *mem = in(MemNode::Memory);
283 if (phase->type( mem ) == Type::TOP) return NodeSentinel; // caller will return NULL
284 assert(mem != this, "dead loop in MemNode::Ideal");
285
286 if (can_reshape && igvn != NULL && igvn->_worklist.member(mem)) {
287 // This memory slice may be dead.
288 // Delay this mem node transformation until the memory is processed.
289 phase->is_IterGVN()->_worklist.push(this);
290 return NodeSentinel; // caller will return NULL
291 }
292
293 Node *address = in(MemNode::Address);
294 const Type *t_adr = phase->type(address);
295 if (t_adr == Type::TOP) return NodeSentinel; // caller will return NULL
296
297 if (can_reshape && igvn != NULL &&
298 (igvn->_worklist.member(address) ||
299 igvn->_worklist.size() > 0 && (t_adr != adr_type())) ) {
300 // The address's base and type may change when the address is processed.
301 // Delay this mem node transformation until the address is processed.
302 phase->is_IterGVN()->_worklist.push(this);
303 return NodeSentinel; // caller will return NULL
304 }
305
306 // Do NOT remove or optimize the next lines: ensure a new alias index
307 // is allocated for an oop pointer type before Escape Analysis.
308 // Note: C++ will not remove it since the call has side effect.
309 if (t_adr->isa_oopptr()) {
310 int alias_idx = phase->C->get_alias_index(t_adr->is_ptr());
311 }
312
313 Node* base = NULL;
314 if (address->is_AddP()) {
315 base = address->in(AddPNode::Base);
316 }
317 if (base != NULL && phase->type(base)->higher_equal(TypePtr::NULL_PTR) &&
318 !t_adr->isa_rawptr()) {
319 // Note: raw address has TOP base and top->higher_equal(TypePtr::NULL_PTR) is true.
320 // Skip this node optimization if its address has TOP base.
321 return NodeSentinel; // caller will return NULL
322 }
323
324 // Avoid independent memory operations
325 Node* old_mem = mem;
326
327 // The code which unhooks non-raw memories from complete (macro-expanded)
328 // initializations was removed. After macro-expansion all stores catched
329 // by Initialize node became raw stores and there is no information
330 // which memory slices they modify. So it is unsafe to move any memory
331 // operation above these stores. Also in most cases hooked non-raw memories
332 // were already unhooked by using information from detect_ptr_independence()
333 // and find_previous_store().
334
335 if (mem->is_MergeMem()) {
336 MergeMemNode* mmem = mem->as_MergeMem();
337 const TypePtr *tp = t_adr->is_ptr();
338
339 mem = step_through_mergemem(phase, mmem, tp, adr_type(), tty);
340 }
341
342 if (mem != old_mem) {
343 set_req(MemNode::Memory, mem);
344 if (can_reshape && old_mem->outcnt() == 0) {
345 igvn->_worklist.push(old_mem);
346 }
347 if (phase->type( mem ) == Type::TOP) return NodeSentinel;
348 return this;
349 }
350
351 // let the subclass continue analyzing...
352 return NULL;
353 }
354
355 // Helper function for proving some simple control dominations.
356 // Attempt to prove that all control inputs of 'dom' dominate 'sub'.
357 // Already assumes that 'dom' is available at 'sub', and that 'sub'
358 // is not a constant (dominated by the method's StartNode).
359 // Used by MemNode::find_previous_store to prove that the
360 // control input of a memory operation predates (dominates)
361 // an allocation it wants to look past.
362 bool MemNode::all_controls_dominate(Node* dom, Node* sub) {
363 if (dom == NULL || dom->is_top() || sub == NULL || sub->is_top())
364 return false; // Conservative answer for dead code
365
366 // Check 'dom'. Skip Proj and CatchProj nodes.
367 dom = dom->find_exact_control(dom);
368 if (dom == NULL || dom->is_top())
369 return false; // Conservative answer for dead code
370
371 if (dom == sub) {
372 // For the case when, for example, 'sub' is Initialize and the original
373 // 'dom' is Proj node of the 'sub'.
374 return false;
375 }
376
377 if (dom->is_Con() || dom->is_Start() || dom->is_Root() || dom == sub)
378 return true;
379
380 // 'dom' dominates 'sub' if its control edge and control edges
381 // of all its inputs dominate or equal to sub's control edge.
382
383 // Currently 'sub' is either Allocate, Initialize or Start nodes.
384 // Or Region for the check in LoadNode::Ideal();
385 // 'sub' should have sub->in(0) != NULL.
386 assert(sub->is_Allocate() || sub->is_Initialize() || sub->is_Start() ||
387 sub->is_Region() || sub->is_Call(), "expecting only these nodes");
388
389 // Get control edge of 'sub'.
390 Node* orig_sub = sub;
391 sub = sub->find_exact_control(sub->in(0));
392 if (sub == NULL || sub->is_top())
393 return false; // Conservative answer for dead code
394
395 assert(sub->is_CFG(), "expecting control");
396
397 if (sub == dom)
398 return true;
399
400 if (sub->is_Start() || sub->is_Root())
401 return false;
402
403 {
404 // Check all control edges of 'dom'.
405
406 ResourceMark rm;
407 Arena* arena = Thread::current()->resource_area();
408 Node_List nlist(arena);
409 Unique_Node_List dom_list(arena);
410
411 dom_list.push(dom);
412 bool only_dominating_controls = false;
413
414 for (uint next = 0; next < dom_list.size(); next++) {
415 Node* n = dom_list.at(next);
416 if (n == orig_sub)
417 return false; // One of dom's inputs dominated by sub.
418 if (!n->is_CFG() && n->pinned()) {
419 // Check only own control edge for pinned non-control nodes.
420 n = n->find_exact_control(n->in(0));
421 if (n == NULL || n->is_top())
422 return false; // Conservative answer for dead code
423 assert(n->is_CFG(), "expecting control");
424 dom_list.push(n);
425 } else if (n->is_Con() || n->is_Start() || n->is_Root()) {
426 only_dominating_controls = true;
427 } else if (n->is_CFG()) {
428 if (n->dominates(sub, nlist))
429 only_dominating_controls = true;
430 else
431 return false;
432 } else {
433 // First, own control edge.
434 Node* m = n->find_exact_control(n->in(0));
435 if (m != NULL) {
436 if (m->is_top())
437 return false; // Conservative answer for dead code
438 dom_list.push(m);
439 }
440 // Now, the rest of edges.
441 uint cnt = n->req();
442 for (uint i = 1; i < cnt; i++) {
443 m = n->find_exact_control(n->in(i));
444 if (m == NULL || m->is_top())
445 continue;
446 dom_list.push(m);
447 }
448 }
449 }
450 return only_dominating_controls;
451 }
452 }
453
454 //---------------------detect_ptr_independence---------------------------------
455 // Used by MemNode::find_previous_store to prove that two base
456 // pointers are never equal.
457 // The pointers are accompanied by their associated allocations,
458 // if any, which have been previously discovered by the caller.
459 bool MemNode::detect_ptr_independence(Node* p1, AllocateNode* a1,
460 Node* p2, AllocateNode* a2,
461 PhaseTransform* phase) {
462 // Attempt to prove that these two pointers cannot be aliased.
463 // They may both manifestly be allocations, and they should differ.
464 // Or, if they are not both allocations, they can be distinct constants.
465 // Otherwise, one is an allocation and the other a pre-existing value.
466 if (a1 == NULL && a2 == NULL) { // neither an allocation
467 return (p1 != p2) && p1->is_Con() && p2->is_Con();
468 } else if (a1 != NULL && a2 != NULL) { // both allocations
469 return (a1 != a2);
470 } else if (a1 != NULL) { // one allocation a1
471 // (Note: p2->is_Con implies p2->in(0)->is_Root, which dominates.)
472 return all_controls_dominate(p2, a1);
473 } else { //(a2 != NULL) // one allocation a2
474 return all_controls_dominate(p1, a2);
475 }
476 return false;
477 }
478
479
480 // The logic for reordering loads and stores uses four steps:
481 // (a) Walk carefully past stores and initializations which we
482 // can prove are independent of this load.
483 // (b) Observe that the next memory state makes an exact match
484 // with self (load or store), and locate the relevant store.
485 // (c) Ensure that, if we were to wire self directly to the store,
486 // the optimizer would fold it up somehow.
487 // (d) Do the rewiring, and return, depending on some other part of
488 // the optimizer to fold up the load.
489 // This routine handles steps (a) and (b). Steps (c) and (d) are
490 // specific to loads and stores, so they are handled by the callers.
491 // (Currently, only LoadNode::Ideal has steps (c), (d). More later.)
492 //
493 Node* MemNode::find_previous_store(PhaseTransform* phase) {
494 Node* ctrl = in(MemNode::Control);
495 Node* adr = in(MemNode::Address);
496 intptr_t offset = 0;
497 Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
498 AllocateNode* alloc = AllocateNode::Ideal_allocation(base, phase);
499
500 if (offset == Type::OffsetBot)
501 return NULL; // cannot unalias unless there are precise offsets
502
503 const TypeOopPtr *addr_t = adr->bottom_type()->isa_oopptr();
504
505 intptr_t size_in_bytes = memory_size();
506
507 Node* mem = in(MemNode::Memory); // start searching here...
508
509 int cnt = 50; // Cycle limiter
510 for (;;) { // While we can dance past unrelated stores...
511 if (--cnt < 0) break; // Caught in cycle or a complicated dance?
512
513 if (mem->is_Store()) {
514 Node* st_adr = mem->in(MemNode::Address);
515 intptr_t st_offset = 0;
516 Node* st_base = AddPNode::Ideal_base_and_offset(st_adr, phase, st_offset);
517 if (st_base == NULL)
518 break; // inscrutable pointer
519 if (st_offset != offset && st_offset != Type::OffsetBot) {
520 const int MAX_STORE = BytesPerLong;
521 if (st_offset >= offset + size_in_bytes ||
522 st_offset <= offset - MAX_STORE ||
523 st_offset <= offset - mem->as_Store()->memory_size()) {
524 // Success: The offsets are provably independent.
525 // (You may ask, why not just test st_offset != offset and be done?
526 // The answer is that stores of different sizes can co-exist
527 // in the same sequence of RawMem effects. We sometimes initialize
528 // a whole 'tile' of array elements with a single jint or jlong.)
529 mem = mem->in(MemNode::Memory);
530 continue; // (a) advance through independent store memory
531 }
532 }
533 if (st_base != base &&
534 detect_ptr_independence(base, alloc,
535 st_base,
536 AllocateNode::Ideal_allocation(st_base, phase),
537 phase)) {
538 // Success: The bases are provably independent.
539 mem = mem->in(MemNode::Memory);
540 continue; // (a) advance through independent store memory
541 }
542
543 // (b) At this point, if the bases or offsets do not agree, we lose,
544 // since we have not managed to prove 'this' and 'mem' independent.
545 if (st_base == base && st_offset == offset) {
546 return mem; // let caller handle steps (c), (d)
547 }
548
549 } else if (mem->is_Proj() && mem->in(0)->is_Initialize()) {
550 InitializeNode* st_init = mem->in(0)->as_Initialize();
551 AllocateNode* st_alloc = st_init->allocation();
552 if (st_alloc == NULL)
553 break; // something degenerated
554 bool known_identical = false;
555 bool known_independent = false;
556 if (alloc == st_alloc)
557 known_identical = true;
558 else if (alloc != NULL)
559 known_independent = true;
560 else if (all_controls_dominate(this, st_alloc))
561 known_independent = true;
562
563 if (known_independent) {
564 // The bases are provably independent: Either they are
565 // manifestly distinct allocations, or else the control
566 // of this load dominates the store's allocation.
567 int alias_idx = phase->C->get_alias_index(adr_type());
568 if (alias_idx == Compile::AliasIdxRaw) {
569 mem = st_alloc->in(TypeFunc::Memory);
570 } else {
571 mem = st_init->memory(alias_idx);
572 }
573 continue; // (a) advance through independent store memory
574 }
575
576 // (b) at this point, if we are not looking at a store initializing
577 // the same allocation we are loading from, we lose.
578 if (known_identical) {
579 // From caller, can_see_stored_value will consult find_captured_store.
580 return mem; // let caller handle steps (c), (d)
581 }
582
583 } else if (addr_t != NULL && addr_t->is_known_instance_field()) {
584 // Can't use optimize_simple_memory_chain() since it needs PhaseGVN.
585 if (mem->is_Proj() && mem->in(0)->is_Call()) {
586 CallNode *call = mem->in(0)->as_Call();
587 if (!call->may_modify(addr_t, phase)) {
588 mem = call->in(TypeFunc::Memory);
589 continue; // (a) advance through independent call memory
590 }
591 } else if (mem->is_Proj() && mem->in(0)->is_MemBar()) {
592 mem = mem->in(0)->in(TypeFunc::Memory);
593 continue; // (a) advance through independent MemBar memory
594 } else if (mem->is_ClearArray()) {
595 if (ClearArrayNode::step_through(&mem, (uint)addr_t->instance_id(), phase)) {
596 // (the call updated 'mem' value)
597 continue; // (a) advance through independent allocation memory
598 } else {
599 // Can not bypass initialization of the instance
600 // we are looking for.
601 return mem;
602 }
603 } else if (mem->is_MergeMem()) {
604 int alias_idx = phase->C->get_alias_index(adr_type());
605 mem = mem->as_MergeMem()->memory_at(alias_idx);
606 continue; // (a) advance through independent MergeMem memory
607 }
608 }
609
610 // Unless there is an explicit 'continue', we must bail out here,
611 // because 'mem' is an inscrutable memory state (e.g., a call).
612 break;
613 }
614
615 return NULL; // bail out
616 }
617
618 //----------------------calculate_adr_type-------------------------------------
619 // Helper function. Notices when the given type of address hits top or bottom.
620 // Also, asserts a cross-check of the type against the expected address type.
621 const TypePtr* MemNode::calculate_adr_type(const Type* t, const TypePtr* cross_check) {
622 if (t == Type::TOP) return NULL; // does not touch memory any more?
623 #ifdef PRODUCT
624 cross_check = NULL;
625 #else
626 if (!VerifyAliases || is_error_reported() || Node::in_dump()) cross_check = NULL;
627 #endif
628 const TypePtr* tp = t->isa_ptr();
629 if (tp == NULL) {
630 assert(cross_check == NULL || cross_check == TypePtr::BOTTOM, "expected memory type must be wide");
631 return TypePtr::BOTTOM; // touches lots of memory
632 } else {
633 #ifdef ASSERT
634 // %%%% [phh] We don't check the alias index if cross_check is
635 // TypeRawPtr::BOTTOM. Needs to be investigated.
636 if (cross_check != NULL &&
637 cross_check != TypePtr::BOTTOM &&
638 cross_check != TypeRawPtr::BOTTOM) {
639 // Recheck the alias index, to see if it has changed (due to a bug).
640 Compile* C = Compile::current();
641 assert(C->get_alias_index(cross_check) == C->get_alias_index(tp),
642 "must stay in the original alias category");
643 // The type of the address must be contained in the adr_type,
644 // disregarding "null"-ness.
645 // (We make an exception for TypeRawPtr::BOTTOM, which is a bit bucket.)
646 const TypePtr* tp_notnull = tp->join(TypePtr::NOTNULL)->is_ptr();
647 assert(cross_check->meet(tp_notnull) == cross_check->remove_speculative(),
648 "real address must not escape from expected memory type");
649 }
650 #endif
651 return tp;
652 }
653 }
654
655 //=============================================================================
656 // Should LoadNode::Ideal() attempt to remove control edges?
657 bool LoadNode::can_remove_control() const {
658 return true;
659 }
660 uint LoadNode::size_of() const { return sizeof(*this); }
661 uint LoadNode::cmp( const Node &n ) const
662 { return !Type::cmp( _type, ((LoadNode&)n)._type ); }
663 const Type *LoadNode::bottom_type() const { return _type; }
664 uint LoadNode::ideal_reg() const {
665 return _type->ideal_reg();
666 }
667
668 #ifndef PRODUCT
669 void LoadNode::dump_spec(outputStream *st) const {
670 MemNode::dump_spec(st);
671 if( !Verbose && !WizardMode ) {
672 // standard dump does this in Verbose and WizardMode
673 st->print(" #"); _type->dump_on(st);
674 }
675 }
676 #endif
677
678 #ifdef ASSERT
679 //----------------------------is_immutable_value-------------------------------
680 // Helper function to allow a raw load without control edge for some cases
681 bool LoadNode::is_immutable_value(Node* adr) {
682 return (adr->is_AddP() && adr->in(AddPNode::Base)->is_top() &&
683 adr->in(AddPNode::Address)->Opcode() == Op_ThreadLocal &&
684 (adr->in(AddPNode::Offset)->find_intptr_t_con(-1) ==
685 in_bytes(JavaThread::osthread_offset())));
686 }
687 #endif
688
689 //----------------------------LoadNode::make-----------------------------------
690 // Polymorphic factory method:
691 Node *LoadNode::make(PhaseGVN& gvn, Node *ctl, Node *mem, Node *adr, const TypePtr* adr_type, const Type *rt, BasicType bt, MemOrd mo) {
692 Compile* C = gvn.C;
693
694 // sanity check the alias category against the created node type
695 assert(!(adr_type->isa_oopptr() &&
696 adr_type->offset() == oopDesc::klass_offset_in_bytes()),
697 "use LoadKlassNode instead");
698 assert(!(adr_type->isa_aryptr() &&
699 adr_type->offset() == arrayOopDesc::length_offset_in_bytes()),
700 "use LoadRangeNode instead");
701 // Check control edge of raw loads
702 assert( ctl != NULL || C->get_alias_index(adr_type) != Compile::AliasIdxRaw ||
703 // oop will be recorded in oop map if load crosses safepoint
704 rt->isa_oopptr() || is_immutable_value(adr),
705 "raw memory operations should have control edge");
706 switch (bt) {
707 case T_BOOLEAN: return new LoadUBNode(ctl, mem, adr, adr_type, rt->is_int(), mo);
708 case T_BYTE: return new LoadBNode (ctl, mem, adr, adr_type, rt->is_int(), mo);
709 case T_INT: return new LoadINode (ctl, mem, adr, adr_type, rt->is_int(), mo);
710 case T_CHAR: return new LoadUSNode(ctl, mem, adr, adr_type, rt->is_int(), mo);
711 case T_SHORT: return new LoadSNode (ctl, mem, adr, adr_type, rt->is_int(), mo);
712 case T_LONG: return new LoadLNode (ctl, mem, adr, adr_type, rt->is_long(), mo);
713 case T_FLOAT: return new LoadFNode (ctl, mem, adr, adr_type, rt, mo);
714 case T_DOUBLE: return new LoadDNode (ctl, mem, adr, adr_type, rt, mo);
715 case T_ADDRESS: return new LoadPNode (ctl, mem, adr, adr_type, rt->is_ptr(), mo);
716 case T_OBJECT:
717 #ifdef _LP64
718 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
719 Node* load = gvn.transform(new LoadNNode(ctl, mem, adr, adr_type, rt->make_narrowoop(), mo));
720 return new DecodeNNode(load, load->bottom_type()->make_ptr());
721 } else
722 #endif
723 {
724 assert(!adr->bottom_type()->is_ptr_to_narrowoop() && !adr->bottom_type()->is_ptr_to_narrowklass(), "should have got back a narrow oop");
725 return new LoadPNode(ctl, mem, adr, adr_type, rt->is_oopptr(), mo);
726 }
727 }
728 ShouldNotReachHere();
729 return (LoadNode*)NULL;
730 }
731
732 LoadLNode* LoadLNode::make_atomic(Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, const Type* rt, MemOrd mo) {
733 bool require_atomic = true;
734 return new LoadLNode(ctl, mem, adr, adr_type, rt->is_long(), mo, require_atomic);
735 }
736
737 LoadDNode* LoadDNode::make_atomic(Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, const Type* rt, MemOrd mo) {
738 bool require_atomic = true;
739 return new LoadDNode(ctl, mem, adr, adr_type, rt, mo, require_atomic);
740 }
741
742
743
744 //------------------------------hash-------------------------------------------
745 uint LoadNode::hash() const {
746 // unroll addition of interesting fields
747 return (uintptr_t)in(Control) + (uintptr_t)in(Memory) + (uintptr_t)in(Address);
748 }
749
750 static bool skip_through_membars(Compile::AliasType* atp, const TypeInstPtr* tp, bool eliminate_boxing) {
751 if ((atp != NULL) && (atp->index() >= Compile::AliasIdxRaw)) {
752 bool non_volatile = (atp->field() != NULL) && !atp->field()->is_volatile();
753 bool is_stable_ary = FoldStableValues &&
754 (tp != NULL) && (tp->isa_aryptr() != NULL) &&
755 tp->isa_aryptr()->is_stable();
756
757 return (eliminate_boxing && non_volatile) || is_stable_ary;
758 }
759
760 return false;
761 }
762
763 //---------------------------can_see_stored_value------------------------------
764 // This routine exists to make sure this set of tests is done the same
765 // everywhere. We need to make a coordinated change: first LoadNode::Ideal
766 // will change the graph shape in a way which makes memory alive twice at the
767 // same time (uses the Oracle model of aliasing), then some
768 // LoadXNode::Identity will fold things back to the equivalence-class model
769 // of aliasing.
770 Node* MemNode::can_see_stored_value(Node* st, PhaseTransform* phase) const {
771 Node* ld_adr = in(MemNode::Address);
772 intptr_t ld_off = 0;
773 AllocateNode* ld_alloc = AllocateNode::Ideal_allocation(ld_adr, phase, ld_off);
774 const TypeInstPtr* tp = phase->type(ld_adr)->isa_instptr();
775 Compile::AliasType* atp = (tp != NULL) ? phase->C->alias_type(tp) : NULL;
776 // This is more general than load from boxing objects.
777 if (skip_through_membars(atp, tp, phase->C->eliminate_boxing())) {
778 uint alias_idx = atp->index();
779 bool final = !atp->is_rewritable();
780 Node* result = NULL;
781 Node* current = st;
782 // Skip through chains of MemBarNodes checking the MergeMems for
783 // new states for the slice of this load. Stop once any other
784 // kind of node is encountered. Loads from final memory can skip
785 // through any kind of MemBar but normal loads shouldn't skip
786 // through MemBarAcquire since the could allow them to move out of
787 // a synchronized region.
788 while (current->is_Proj()) {
789 int opc = current->in(0)->Opcode();
790 if ((final && (opc == Op_MemBarAcquire ||
791 opc == Op_MemBarAcquireLock ||
792 opc == Op_LoadFence)) ||
793 opc == Op_MemBarRelease ||
794 opc == Op_StoreFence ||
795 opc == Op_MemBarReleaseLock ||
796 opc == Op_MemBarCPUOrder) {
797 Node* mem = current->in(0)->in(TypeFunc::Memory);
798 if (mem->is_MergeMem()) {
799 MergeMemNode* merge = mem->as_MergeMem();
800 Node* new_st = merge->memory_at(alias_idx);
801 if (new_st == merge->base_memory()) {
802 // Keep searching
803 current = new_st;
804 continue;
805 }
806 // Save the new memory state for the slice and fall through
807 // to exit.
808 result = new_st;
809 }
810 }
811 break;
812 }
813 if (result != NULL) {
814 st = result;
815 }
816 }
817
818 // Loop around twice in the case Load -> Initialize -> Store.
819 // (See PhaseIterGVN::add_users_to_worklist, which knows about this case.)
820 for (int trip = 0; trip <= 1; trip++) {
821
822 if (st->is_Store()) {
823 Node* st_adr = st->in(MemNode::Address);
824 if (!phase->eqv(st_adr, ld_adr)) {
825 // Try harder before giving up... Match raw and non-raw pointers.
826 intptr_t st_off = 0;
827 AllocateNode* alloc = AllocateNode::Ideal_allocation(st_adr, phase, st_off);
828 if (alloc == NULL) return NULL;
829 if (alloc != ld_alloc) return NULL;
830 if (ld_off != st_off) return NULL;
831 // At this point we have proven something like this setup:
832 // A = Allocate(...)
833 // L = LoadQ(, AddP(CastPP(, A.Parm),, #Off))
834 // S = StoreQ(, AddP(, A.Parm , #Off), V)
835 // (Actually, we haven't yet proven the Q's are the same.)
836 // In other words, we are loading from a casted version of
837 // the same pointer-and-offset that we stored to.
838 // Thus, we are able to replace L by V.
839 }
840 // Now prove that we have a LoadQ matched to a StoreQ, for some Q.
841 if (store_Opcode() != st->Opcode())
842 return NULL;
843 return st->in(MemNode::ValueIn);
844 }
845
846 // A load from a freshly-created object always returns zero.
847 // (This can happen after LoadNode::Ideal resets the load's memory input
848 // to find_captured_store, which returned InitializeNode::zero_memory.)
849 if (st->is_Proj() && st->in(0)->is_Allocate() &&
850 (st->in(0) == ld_alloc) &&
851 (ld_off >= st->in(0)->as_Allocate()->minimum_header_size())) {
852 // return a zero value for the load's basic type
853 // (This is one of the few places where a generic PhaseTransform
854 // can create new nodes. Think of it as lazily manifesting
855 // virtually pre-existing constants.)
856 return phase->zerocon(memory_type());
857 }
858
859 // A load from an initialization barrier can match a captured store.
860 if (st->is_Proj() && st->in(0)->is_Initialize()) {
861 InitializeNode* init = st->in(0)->as_Initialize();
862 AllocateNode* alloc = init->allocation();
863 if ((alloc != NULL) && (alloc == ld_alloc)) {
864 // examine a captured store value
865 st = init->find_captured_store(ld_off, memory_size(), phase);
866 if (st != NULL)
867 continue; // take one more trip around
868 }
869 }
870
871 // Load boxed value from result of valueOf() call is input parameter.
872 if (this->is_Load() && ld_adr->is_AddP() &&
873 (tp != NULL) && tp->is_ptr_to_boxed_value()) {
874 intptr_t ignore = 0;
875 Node* base = AddPNode::Ideal_base_and_offset(ld_adr, phase, ignore);
876 if (base != NULL && base->is_Proj() &&
877 base->as_Proj()->_con == TypeFunc::Parms &&
878 base->in(0)->is_CallStaticJava() &&
879 base->in(0)->as_CallStaticJava()->is_boxing_method()) {
880 return base->in(0)->in(TypeFunc::Parms);
881 }
882 }
883
884 break;
885 }
886
887 return NULL;
888 }
889
890 //----------------------is_instance_field_load_with_local_phi------------------
891 bool LoadNode::is_instance_field_load_with_local_phi(Node* ctrl) {
892 if( in(Memory)->is_Phi() && in(Memory)->in(0) == ctrl &&
893 in(Address)->is_AddP() ) {
894 const TypeOopPtr* t_oop = in(Address)->bottom_type()->isa_oopptr();
895 // Only instances and boxed values.
896 if( t_oop != NULL &&
897 (t_oop->is_ptr_to_boxed_value() ||
898 t_oop->is_known_instance_field()) &&
899 t_oop->offset() != Type::OffsetBot &&
900 t_oop->offset() != Type::OffsetTop) {
901 return true;
902 }
903 }
904 return false;
905 }
906
907 //------------------------------Identity---------------------------------------
908 // Loads are identity if previous store is to same address
909 Node *LoadNode::Identity( PhaseTransform *phase ) {
910 // If the previous store-maker is the right kind of Store, and the store is
911 // to the same address, then we are equal to the value stored.
912 Node* mem = in(Memory);
913 Node* value = can_see_stored_value(mem, phase);
914 if( value ) {
915 // byte, short & char stores truncate naturally.
916 // A load has to load the truncated value which requires
917 // some sort of masking operation and that requires an
918 // Ideal call instead of an Identity call.
919 if (memory_size() < BytesPerInt) {
920 // If the input to the store does not fit with the load's result type,
921 // it must be truncated via an Ideal call.
922 if (!phase->type(value)->higher_equal(phase->type(this)))
923 return this;
924 }
925 // (This works even when value is a Con, but LoadNode::Value
926 // usually runs first, producing the singleton type of the Con.)
927 return value;
928 }
929
930 // Search for an existing data phi which was generated before for the same
931 // instance's field to avoid infinite generation of phis in a loop.
932 Node *region = mem->in(0);
933 if (is_instance_field_load_with_local_phi(region)) {
934 const TypeOopPtr *addr_t = in(Address)->bottom_type()->isa_oopptr();
935 int this_index = phase->C->get_alias_index(addr_t);
936 int this_offset = addr_t->offset();
937 int this_iid = addr_t->instance_id();
938 if (!addr_t->is_known_instance() &&
939 addr_t->is_ptr_to_boxed_value()) {
940 // Use _idx of address base (could be Phi node) for boxed values.
941 intptr_t ignore = 0;
942 Node* base = AddPNode::Ideal_base_and_offset(in(Address), phase, ignore);
943 this_iid = base->_idx;
944 }
945 const Type* this_type = bottom_type();
946 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
947 Node* phi = region->fast_out(i);
948 if (phi->is_Phi() && phi != mem &&
949 phi->as_Phi()->is_same_inst_field(this_type, this_iid, this_index, this_offset)) {
950 return phi;
951 }
952 }
953 }
954
955 return this;
956 }
957
958 // We're loading from an object which has autobox behaviour.
959 // If this object is result of a valueOf call we'll have a phi
960 // merging a newly allocated object and a load from the cache.
961 // We want to replace this load with the original incoming
962 // argument to the valueOf call.
963 Node* LoadNode::eliminate_autobox(PhaseGVN* phase) {
964 assert(phase->C->eliminate_boxing(), "sanity");
965 intptr_t ignore = 0;
966 Node* base = AddPNode::Ideal_base_and_offset(in(Address), phase, ignore);
967 if ((base == NULL) || base->is_Phi()) {
968 // Push the loads from the phi that comes from valueOf up
969 // through it to allow elimination of the loads and the recovery
970 // of the original value. It is done in split_through_phi().
971 return NULL;
972 } else if (base->is_Load() ||
973 base->is_DecodeN() && base->in(1)->is_Load()) {
974 // Eliminate the load of boxed value for integer types from the cache
975 // array by deriving the value from the index into the array.
976 // Capture the offset of the load and then reverse the computation.
977
978 // Get LoadN node which loads a boxing object from 'cache' array.
979 if (base->is_DecodeN()) {
980 base = base->in(1);
981 }
982 if (!base->in(Address)->is_AddP()) {
983 return NULL; // Complex address
984 }
985 AddPNode* address = base->in(Address)->as_AddP();
986 Node* cache_base = address->in(AddPNode::Base);
987 if ((cache_base != NULL) && cache_base->is_DecodeN()) {
988 // Get ConP node which is static 'cache' field.
989 cache_base = cache_base->in(1);
990 }
991 if ((cache_base != NULL) && cache_base->is_Con()) {
992 const TypeAryPtr* base_type = cache_base->bottom_type()->isa_aryptr();
993 if ((base_type != NULL) && base_type->is_autobox_cache()) {
994 Node* elements[4];
995 int shift = exact_log2(type2aelembytes(T_OBJECT));
996 int count = address->unpack_offsets(elements, ARRAY_SIZE(elements));
997 if ((count > 0) && elements[0]->is_Con() &&
998 ((count == 1) ||
999 (count == 2) && elements[1]->Opcode() == Op_LShiftX &&
1000 elements[1]->in(2) == phase->intcon(shift))) {
1001 ciObjArray* array = base_type->const_oop()->as_obj_array();
1002 // Fetch the box object cache[0] at the base of the array and get its value
1003 ciInstance* box = array->obj_at(0)->as_instance();
1004 ciInstanceKlass* ik = box->klass()->as_instance_klass();
1005 assert(ik->is_box_klass(), "sanity");
1006 assert(ik->nof_nonstatic_fields() == 1, "change following code");
1007 if (ik->nof_nonstatic_fields() == 1) {
1008 // This should be true nonstatic_field_at requires calling
1009 // nof_nonstatic_fields so check it anyway
1010 ciConstant c = box->field_value(ik->nonstatic_field_at(0));
1011 BasicType bt = c.basic_type();
1012 // Only integer types have boxing cache.
1013 assert(bt == T_BOOLEAN || bt == T_CHAR ||
1014 bt == T_BYTE || bt == T_SHORT ||
1015 bt == T_INT || bt == T_LONG, err_msg_res("wrong type = %s", type2name(bt)));
1016 jlong cache_low = (bt == T_LONG) ? c.as_long() : c.as_int();
1017 if (cache_low != (int)cache_low) {
1018 return NULL; // should not happen since cache is array indexed by value
1019 }
1020 jlong offset = arrayOopDesc::base_offset_in_bytes(T_OBJECT) - (cache_low << shift);
1021 if (offset != (int)offset) {
1022 return NULL; // should not happen since cache is array indexed by value
1023 }
1024 // Add up all the offsets making of the address of the load
1025 Node* result = elements[0];
1026 for (int i = 1; i < count; i++) {
1027 result = phase->transform(new AddXNode(result, elements[i]));
1028 }
1029 // Remove the constant offset from the address and then
1030 result = phase->transform(new AddXNode(result, phase->MakeConX(-(int)offset)));
1031 // remove the scaling of the offset to recover the original index.
1032 if (result->Opcode() == Op_LShiftX && result->in(2) == phase->intcon(shift)) {
1033 // Peel the shift off directly but wrap it in a dummy node
1034 // since Ideal can't return existing nodes
1035 result = new RShiftXNode(result->in(1), phase->intcon(0));
1036 } else if (result->is_Add() && result->in(2)->is_Con() &&
1037 result->in(1)->Opcode() == Op_LShiftX &&
1038 result->in(1)->in(2) == phase->intcon(shift)) {
1039 // We can't do general optimization: ((X<<Z) + Y) >> Z ==> X + (Y>>Z)
1040 // but for boxing cache access we know that X<<Z will not overflow
1041 // (there is range check) so we do this optimizatrion by hand here.
1042 Node* add_con = new RShiftXNode(result->in(2), phase->intcon(shift));
1043 result = new AddXNode(result->in(1)->in(1), phase->transform(add_con));
1044 } else {
1045 result = new RShiftXNode(result, phase->intcon(shift));
1046 }
1047 #ifdef _LP64
1048 if (bt != T_LONG) {
1049 result = new ConvL2INode(phase->transform(result));
1050 }
1051 #else
1052 if (bt == T_LONG) {
1053 result = new ConvI2LNode(phase->transform(result));
1054 }
1055 #endif
1056 // Boxing/unboxing can be done from signed & unsigned loads (e.g. LoadUB -> ... -> LoadB pair).
1057 // Need to preserve unboxing load type if it is unsigned.
1058 switch(this->Opcode()) {
1059 case Op_LoadUB:
1060 result = new AndINode(phase->transform(result), phase->intcon(0xFF));
1061 break;
1062 case Op_LoadUS:
1063 result = new AndINode(phase->transform(result), phase->intcon(0xFFFF));
1064 break;
1065 }
1066 return result;
1067 }
1068 }
1069 }
1070 }
1071 }
1072 return NULL;
1073 }
1074
1075 static bool stable_phi(PhiNode* phi, PhaseGVN *phase) {
1076 Node* region = phi->in(0);
1077 if (region == NULL) {
1078 return false; // Wait stable graph
1079 }
1080 uint cnt = phi->req();
1081 for (uint i = 1; i < cnt; i++) {
1082 Node* rc = region->in(i);
1083 if (rc == NULL || phase->type(rc) == Type::TOP)
1084 return false; // Wait stable graph
1085 Node* in = phi->in(i);
1086 if (in == NULL || phase->type(in) == Type::TOP)
1087 return false; // Wait stable graph
1088 }
1089 return true;
1090 }
1091 //------------------------------split_through_phi------------------------------
1092 // Split instance or boxed field load through Phi.
1093 Node *LoadNode::split_through_phi(PhaseGVN *phase) {
1094 Node* mem = in(Memory);
1095 Node* address = in(Address);
1096 const TypeOopPtr *t_oop = phase->type(address)->isa_oopptr();
1097
1098 assert((t_oop != NULL) &&
1099 (t_oop->is_known_instance_field() ||
1100 t_oop->is_ptr_to_boxed_value()), "invalide conditions");
1101
1102 Compile* C = phase->C;
1103 intptr_t ignore = 0;
1104 Node* base = AddPNode::Ideal_base_and_offset(address, phase, ignore);
1105 bool base_is_phi = (base != NULL) && base->is_Phi();
1106 bool load_boxed_values = t_oop->is_ptr_to_boxed_value() && C->aggressive_unboxing() &&
1107 (base != NULL) && (base == address->in(AddPNode::Base)) &&
1108 phase->type(base)->higher_equal(TypePtr::NOTNULL);
1109
1110 if (!((mem->is_Phi() || base_is_phi) &&
1111 (load_boxed_values || t_oop->is_known_instance_field()))) {
1112 return NULL; // memory is not Phi
1113 }
1114
1115 if (mem->is_Phi()) {
1116 if (!stable_phi(mem->as_Phi(), phase)) {
1117 return NULL; // Wait stable graph
1118 }
1119 uint cnt = mem->req();
1120 // Check for loop invariant memory.
1121 if (cnt == 3) {
1122 for (uint i = 1; i < cnt; i++) {
1123 Node* in = mem->in(i);
1124 Node* m = optimize_memory_chain(in, t_oop, this, phase);
1125 if (m == mem) {
1126 set_req(Memory, mem->in(cnt - i));
1127 return this; // made change
1128 }
1129 }
1130 }
1131 }
1132 if (base_is_phi) {
1133 if (!stable_phi(base->as_Phi(), phase)) {
1134 return NULL; // Wait stable graph
1135 }
1136 uint cnt = base->req();
1137 // Check for loop invariant memory.
1138 if (cnt == 3) {
1139 for (uint i = 1; i < cnt; i++) {
1140 if (base->in(i) == base) {
1141 return NULL; // Wait stable graph
1142 }
1143 }
1144 }
1145 }
1146
1147 bool load_boxed_phi = load_boxed_values && base_is_phi && (base->in(0) == mem->in(0));
1148
1149 // Split through Phi (see original code in loopopts.cpp).
1150 assert(C->have_alias_type(t_oop), "instance should have alias type");
1151
1152 // Do nothing here if Identity will find a value
1153 // (to avoid infinite chain of value phis generation).
1154 if (!phase->eqv(this, this->Identity(phase)))
1155 return NULL;
1156
1157 // Select Region to split through.
1158 Node* region;
1159 if (!base_is_phi) {
1160 assert(mem->is_Phi(), "sanity");
1161 region = mem->in(0);
1162 // Skip if the region dominates some control edge of the address.
1163 if (!MemNode::all_controls_dominate(address, region))
1164 return NULL;
1165 } else if (!mem->is_Phi()) {
1166 assert(base_is_phi, "sanity");
1167 region = base->in(0);
1168 // Skip if the region dominates some control edge of the memory.
1169 if (!MemNode::all_controls_dominate(mem, region))
1170 return NULL;
1171 } else if (base->in(0) != mem->in(0)) {
1172 assert(base_is_phi && mem->is_Phi(), "sanity");
1173 if (MemNode::all_controls_dominate(mem, base->in(0))) {
1174 region = base->in(0);
1175 } else if (MemNode::all_controls_dominate(address, mem->in(0))) {
1176 region = mem->in(0);
1177 } else {
1178 return NULL; // complex graph
1179 }
1180 } else {
1181 assert(base->in(0) == mem->in(0), "sanity");
1182 region = mem->in(0);
1183 }
1184
1185 const Type* this_type = this->bottom_type();
1186 int this_index = C->get_alias_index(t_oop);
1187 int this_offset = t_oop->offset();
1188 int this_iid = t_oop->instance_id();
1189 if (!t_oop->is_known_instance() && load_boxed_values) {
1190 // Use _idx of address base for boxed values.
1191 this_iid = base->_idx;
1192 }
1193 PhaseIterGVN* igvn = phase->is_IterGVN();
1194 Node* phi = new PhiNode(region, this_type, NULL, this_iid, this_index, this_offset);
1195 for (uint i = 1; i < region->req(); i++) {
1196 Node* x;
1197 Node* the_clone = NULL;
1198 if (region->in(i) == C->top()) {
1199 x = C->top(); // Dead path? Use a dead data op
1200 } else {
1201 x = this->clone(); // Else clone up the data op
1202 the_clone = x; // Remember for possible deletion.
1203 // Alter data node to use pre-phi inputs
1204 if (this->in(0) == region) {
1205 x->set_req(0, region->in(i));
1206 } else {
1207 x->set_req(0, NULL);
1208 }
1209 if (mem->is_Phi() && (mem->in(0) == region)) {
1210 x->set_req(Memory, mem->in(i)); // Use pre-Phi input for the clone.
1211 }
1212 if (address->is_Phi() && address->in(0) == region) {
1213 x->set_req(Address, address->in(i)); // Use pre-Phi input for the clone
1214 }
1215 if (base_is_phi && (base->in(0) == region)) {
1216 Node* base_x = base->in(i); // Clone address for loads from boxed objects.
1217 Node* adr_x = phase->transform(new AddPNode(base_x,base_x,address->in(AddPNode::Offset)));
1218 x->set_req(Address, adr_x);
1219 }
1220 }
1221 // Check for a 'win' on some paths
1222 const Type *t = x->Value(igvn);
1223
1224 bool singleton = t->singleton();
1225
1226 // See comments in PhaseIdealLoop::split_thru_phi().
1227 if (singleton && t == Type::TOP) {
1228 singleton &= region->is_Loop() && (i != LoopNode::EntryControl);
1229 }
1230
1231 if (singleton) {
1232 x = igvn->makecon(t);
1233 } else {
1234 // We now call Identity to try to simplify the cloned node.
1235 // Note that some Identity methods call phase->type(this).
1236 // Make sure that the type array is big enough for
1237 // our new node, even though we may throw the node away.
1238 // (This tweaking with igvn only works because x is a new node.)
1239 igvn->set_type(x, t);
1240 // If x is a TypeNode, capture any more-precise type permanently into Node
1241 // otherwise it will be not updated during igvn->transform since
1242 // igvn->type(x) is set to x->Value() already.
1243 x->raise_bottom_type(t);
1244 Node *y = x->Identity(igvn);
1245 if (y != x) {
1246 x = y;
1247 } else {
1248 y = igvn->hash_find_insert(x);
1249 if (y) {
1250 x = y;
1251 } else {
1252 // Else x is a new node we are keeping
1253 // We do not need register_new_node_with_optimizer
1254 // because set_type has already been called.
1255 igvn->_worklist.push(x);
1256 }
1257 }
1258 }
1259 if (x != the_clone && the_clone != NULL) {
1260 igvn->remove_dead_node(the_clone);
1261 }
1262 phi->set_req(i, x);
1263 }
1264 // Record Phi
1265 igvn->register_new_node_with_optimizer(phi);
1266 return phi;
1267 }
1268
1269 //------------------------------Ideal------------------------------------------
1270 // If the load is from Field memory and the pointer is non-null, it might be possible to
1271 // zero out the control input.
1272 // If the offset is constant and the base is an object allocation,
1273 // try to hook me up to the exact initializing store.
1274 Node *LoadNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1275 Node* p = MemNode::Ideal_common(phase, can_reshape);
1276 if (p) return (p == NodeSentinel) ? NULL : p;
1277
1278 Node* ctrl = in(MemNode::Control);
1279 Node* address = in(MemNode::Address);
1280 bool progress = false;
1281
1282 // Skip up past a SafePoint control. Cannot do this for Stores because
1283 // pointer stores & cardmarks must stay on the same side of a SafePoint.
1284 if( ctrl != NULL && ctrl->Opcode() == Op_SafePoint &&
1285 phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw ) {
1286 ctrl = ctrl->in(0);
1287 set_req(MemNode::Control,ctrl);
1288 progress = true;
1289 }
1290
1291 intptr_t ignore = 0;
1292 Node* base = AddPNode::Ideal_base_and_offset(address, phase, ignore);
1293 if (base != NULL
1294 && phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw) {
1295 // Check for useless control edge in some common special cases
1296 if (in(MemNode::Control) != NULL
1297 && can_remove_control()
1298 && phase->type(base)->higher_equal(TypePtr::NOTNULL)
1299 && all_controls_dominate(base, phase->C->start())) {
1300 // A method-invariant, non-null address (constant or 'this' argument).
1301 set_req(MemNode::Control, NULL);
1302 progress = true;
1303 }
1304 }
1305
1306 Node* mem = in(MemNode::Memory);
1307 const TypePtr *addr_t = phase->type(address)->isa_ptr();
1308
1309 if (can_reshape && (addr_t != NULL)) {
1310 // try to optimize our memory input
1311 Node* opt_mem = MemNode::optimize_memory_chain(mem, addr_t, this, phase);
1312 if (opt_mem != mem) {
1313 set_req(MemNode::Memory, opt_mem);
1314 if (phase->type( opt_mem ) == Type::TOP) return NULL;
1315 return this;
1316 }
1317 const TypeOopPtr *t_oop = addr_t->isa_oopptr();
1318 if ((t_oop != NULL) &&
1319 (t_oop->is_known_instance_field() ||
1320 t_oop->is_ptr_to_boxed_value())) {
1321 PhaseIterGVN *igvn = phase->is_IterGVN();
1322 if (igvn != NULL && igvn->_worklist.member(opt_mem)) {
1323 // Delay this transformation until memory Phi is processed.
1324 phase->is_IterGVN()->_worklist.push(this);
1325 return NULL;
1326 }
1327 // Split instance field load through Phi.
1328 Node* result = split_through_phi(phase);
1329 if (result != NULL) return result;
1330
1331 if (t_oop->is_ptr_to_boxed_value()) {
1332 Node* result = eliminate_autobox(phase);
1333 if (result != NULL) return result;
1334 }
1335 }
1336 }
1337
1338 // Check for prior store with a different base or offset; make Load
1339 // independent. Skip through any number of them. Bail out if the stores
1340 // are in an endless dead cycle and report no progress. This is a key
1341 // transform for Reflection. However, if after skipping through the Stores
1342 // we can't then fold up against a prior store do NOT do the transform as
1343 // this amounts to using the 'Oracle' model of aliasing. It leaves the same
1344 // array memory alive twice: once for the hoisted Load and again after the
1345 // bypassed Store. This situation only works if EVERYBODY who does
1346 // anti-dependence work knows how to bypass. I.e. we need all
1347 // anti-dependence checks to ask the same Oracle. Right now, that Oracle is
1348 // the alias index stuff. So instead, peek through Stores and IFF we can
1349 // fold up, do so.
1350 Node* prev_mem = find_previous_store(phase);
1351 // Steps (a), (b): Walk past independent stores to find an exact match.
1352 if (prev_mem != NULL && prev_mem != in(MemNode::Memory)) {
1353 // (c) See if we can fold up on the spot, but don't fold up here.
1354 // Fold-up might require truncation (for LoadB/LoadS/LoadUS) or
1355 // just return a prior value, which is done by Identity calls.
1356 if (can_see_stored_value(prev_mem, phase)) {
1357 // Make ready for step (d):
1358 set_req(MemNode::Memory, prev_mem);
1359 return this;
1360 }
1361 }
1362
1363 return progress ? this : NULL;
1364 }
1365
1366 // Helper to recognize certain Klass fields which are invariant across
1367 // some group of array types (e.g., int[] or all T[] where T < Object).
1368 const Type*
1369 LoadNode::load_array_final_field(const TypeKlassPtr *tkls,
1370 ciKlass* klass) const {
1371 if (tkls->offset() == in_bytes(Klass::modifier_flags_offset())) {
1372 // The field is Klass::_modifier_flags. Return its (constant) value.
1373 // (Folds up the 2nd indirection in aClassConstant.getModifiers().)
1374 assert(this->Opcode() == Op_LoadI, "must load an int from _modifier_flags");
1375 return TypeInt::make(klass->modifier_flags());
1376 }
1377 if (tkls->offset() == in_bytes(Klass::access_flags_offset())) {
1378 // The field is Klass::_access_flags. Return its (constant) value.
1379 // (Folds up the 2nd indirection in Reflection.getClassAccessFlags(aClassConstant).)
1380 assert(this->Opcode() == Op_LoadI, "must load an int from _access_flags");
1381 return TypeInt::make(klass->access_flags());
1382 }
1383 if (tkls->offset() == in_bytes(Klass::layout_helper_offset())) {
1384 // The field is Klass::_layout_helper. Return its constant value if known.
1385 assert(this->Opcode() == Op_LoadI, "must load an int from _layout_helper");
1386 return TypeInt::make(klass->layout_helper());
1387 }
1388
1389 // No match.
1390 return NULL;
1391 }
1392
1393 // Try to constant-fold a stable array element.
1394 static const Type* fold_stable_ary_elem(const TypeAryPtr* ary, int off, BasicType loadbt) {
1395 assert(ary->const_oop(), "array should be constant");
1396 assert(ary->is_stable(), "array should be stable");
1397
1398 // Decode the results of GraphKit::array_element_address.
1399 ciArray* aobj = ary->const_oop()->as_array();
1400 ciConstant con = aobj->element_value_by_offset(off);
1401
1402 if (con.basic_type() != T_ILLEGAL && !con.is_null_or_zero()) {
1403 const Type* con_type = Type::make_from_constant(con);
1404 if (con_type != NULL) {
1405 if (con_type->isa_aryptr()) {
1406 // Join with the array element type, in case it is also stable.
1407 int dim = ary->stable_dimension();
1408 con_type = con_type->is_aryptr()->cast_to_stable(true, dim-1);
1409 }
1410 if (loadbt == T_NARROWOOP && con_type->isa_oopptr()) {
1411 con_type = con_type->make_narrowoop();
1412 }
1413 #ifndef PRODUCT
1414 if (TraceIterativeGVN) {
1415 tty->print("FoldStableValues: array element [off=%d]: con_type=", off);
1416 con_type->dump(); tty->cr();
1417 }
1418 #endif //PRODUCT
1419 return con_type;
1420 }
1421 }
1422 return NULL;
1423 }
1424
1425 //------------------------------Value-----------------------------------------
1426 const Type *LoadNode::Value( PhaseTransform *phase ) const {
1427 // Either input is TOP ==> the result is TOP
1428 Node* mem = in(MemNode::Memory);
1429 const Type *t1 = phase->type(mem);
1430 if (t1 == Type::TOP) return Type::TOP;
1431 Node* adr = in(MemNode::Address);
1432 const TypePtr* tp = phase->type(adr)->isa_ptr();
1433 if (tp == NULL || tp->empty()) return Type::TOP;
1434 int off = tp->offset();
1435 assert(off != Type::OffsetTop, "case covered by TypePtr::empty");
1436 Compile* C = phase->C;
1437
1438 // Try to guess loaded type from pointer type
1439 if (tp->isa_aryptr()) {
1440 const TypeAryPtr* ary = tp->is_aryptr();
1441 const Type* t = ary->elem();
1442
1443 // Determine whether the reference is beyond the header or not, by comparing
1444 // the offset against the offset of the start of the array's data.
1445 // Different array types begin at slightly different offsets (12 vs. 16).
1446 // We choose T_BYTE as an example base type that is least restrictive
1447 // as to alignment, which will therefore produce the smallest
1448 // possible base offset.
1449 const int min_base_off = arrayOopDesc::base_offset_in_bytes(T_BYTE);
1450 const bool off_beyond_header = ((uint)off >= (uint)min_base_off);
1451
1452 // Try to constant-fold a stable array element.
1453 if (FoldStableValues && ary->is_stable() && ary->const_oop() != NULL) {
1454 // Make sure the reference is not into the header and the offset is constant
1455 if (off_beyond_header && adr->is_AddP() && off != Type::OffsetBot) {
1456 const Type* con_type = fold_stable_ary_elem(ary, off, memory_type());
1457 if (con_type != NULL) {
1458 return con_type;
1459 }
1460 }
1461 }
1462
1463 // Don't do this for integer types. There is only potential profit if
1464 // the element type t is lower than _type; that is, for int types, if _type is
1465 // more restrictive than t. This only happens here if one is short and the other
1466 // char (both 16 bits), and in those cases we've made an intentional decision
1467 // to use one kind of load over the other. See AndINode::Ideal and 4965907.
1468 // Also, do not try to narrow the type for a LoadKlass, regardless of offset.
1469 //
1470 // Yes, it is possible to encounter an expression like (LoadKlass p1:(AddP x x 8))
1471 // where the _gvn.type of the AddP is wider than 8. This occurs when an earlier
1472 // copy p0 of (AddP x x 8) has been proven equal to p1, and the p0 has been
1473 // subsumed by p1. If p1 is on the worklist but has not yet been re-transformed,
1474 // it is possible that p1 will have a type like Foo*[int+]:NotNull*+any.
1475 // In fact, that could have been the original type of p1, and p1 could have
1476 // had an original form like p1:(AddP x x (LShiftL quux 3)), where the
1477 // expression (LShiftL quux 3) independently optimized to the constant 8.
1478 if ((t->isa_int() == NULL) && (t->isa_long() == NULL)
1479 && (_type->isa_vect() == NULL)
1480 && Opcode() != Op_LoadKlass && Opcode() != Op_LoadNKlass) {
1481 // t might actually be lower than _type, if _type is a unique
1482 // concrete subclass of abstract class t.
1483 if (off_beyond_header) { // is the offset beyond the header?
1484 const Type* jt = t->join_speculative(_type);
1485 // In any case, do not allow the join, per se, to empty out the type.
1486 if (jt->empty() && !t->empty()) {
1487 // This can happen if a interface-typed array narrows to a class type.
1488 jt = _type;
1489 }
1490 #ifdef ASSERT
1491 if (phase->C->eliminate_boxing() && adr->is_AddP()) {
1492 // The pointers in the autobox arrays are always non-null
1493 Node* base = adr->in(AddPNode::Base);
1494 if ((base != NULL) && base->is_DecodeN()) {
1495 // Get LoadN node which loads IntegerCache.cache field
1496 base = base->in(1);
1497 }
1498 if ((base != NULL) && base->is_Con()) {
1499 const TypeAryPtr* base_type = base->bottom_type()->isa_aryptr();
1500 if ((base_type != NULL) && base_type->is_autobox_cache()) {
1501 // It could be narrow oop
1502 assert(jt->make_ptr()->ptr() == TypePtr::NotNull,"sanity");
1503 }
1504 }
1505 }
1506 #endif
1507 return jt;
1508 }
1509 }
1510 } else if (tp->base() == Type::InstPtr) {
1511 ciEnv* env = C->env();
1512 const TypeInstPtr* tinst = tp->is_instptr();
1513 ciKlass* klass = tinst->klass();
1514 assert( off != Type::OffsetBot ||
1515 // arrays can be cast to Objects
1516 tp->is_oopptr()->klass()->is_java_lang_Object() ||
1517 // unsafe field access may not have a constant offset
1518 C->has_unsafe_access(),
1519 "Field accesses must be precise" );
1520 // For oop loads, we expect the _type to be precise
1521 if (klass == env->String_klass() &&
1522 adr->is_AddP() && off != Type::OffsetBot) {
1523 // For constant Strings treat the final fields as compile time constants.
1524 Node* base = adr->in(AddPNode::Base);
1525 const TypeOopPtr* t = phase->type(base)->isa_oopptr();
1526 if (t != NULL && t->singleton()) {
1527 ciField* field = env->String_klass()->get_field_by_offset(off, false);
1528 if (field != NULL && field->is_final()) {
1529 ciObject* string = t->const_oop();
1530 ciConstant constant = string->as_instance()->field_value(field);
1531 if (constant.basic_type() == T_INT) {
1532 return TypeInt::make(constant.as_int());
1533 } else if (constant.basic_type() == T_ARRAY) {
1534 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
1535 return TypeNarrowOop::make_from_constant(constant.as_object(), true);
1536 } else {
1537 return TypeOopPtr::make_from_constant(constant.as_object(), true);
1538 }
1539 }
1540 }
1541 }
1542 }
1543 // Optimizations for constant objects
1544 ciObject* const_oop = tinst->const_oop();
1545 if (const_oop != NULL) {
1546 // For constant Boxed value treat the target field as a compile time constant.
1547 if (tinst->is_ptr_to_boxed_value()) {
1548 return tinst->get_const_boxed_value();
1549 } else
1550 // For constant CallSites treat the target field as a compile time constant.
1551 if (const_oop->is_call_site()) {
1552 ciCallSite* call_site = const_oop->as_call_site();
1553 ciField* field = call_site->klass()->as_instance_klass()->get_field_by_offset(off, /*is_static=*/ false);
1554 if (field != NULL && field->is_call_site_target()) {
1555 ciMethodHandle* target = call_site->get_target();
1556 if (target != NULL) { // just in case
1557 ciConstant constant(T_OBJECT, target);
1558 const Type* t;
1559 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
1560 t = TypeNarrowOop::make_from_constant(constant.as_object(), true);
1561 } else {
1562 t = TypeOopPtr::make_from_constant(constant.as_object(), true);
1563 }
1564 // Add a dependence for invalidation of the optimization.
1565 if (!call_site->is_constant_call_site()) {
1566 C->dependencies()->assert_call_site_target_value(call_site, target);
1567 }
1568 return t;
1569 }
1570 }
1571 }
1572 }
1573 } else if (tp->base() == Type::KlassPtr) {
1574 assert( off != Type::OffsetBot ||
1575 // arrays can be cast to Objects
1576 tp->is_klassptr()->klass()->is_java_lang_Object() ||
1577 // also allow array-loading from the primary supertype
1578 // array during subtype checks
1579 Opcode() == Op_LoadKlass,
1580 "Field accesses must be precise" );
1581 // For klass/static loads, we expect the _type to be precise
1582 }
1583
1584 const TypeKlassPtr *tkls = tp->isa_klassptr();
1585 if (tkls != NULL && !StressReflectiveCode) {
1586 ciKlass* klass = tkls->klass();
1587 if (klass->is_loaded() && tkls->klass_is_exact()) {
1588 // We are loading a field from a Klass metaobject whose identity
1589 // is known at compile time (the type is "exact" or "precise").
1590 // Check for fields we know are maintained as constants by the VM.
1591 if (tkls->offset() == in_bytes(Klass::super_check_offset_offset())) {
1592 // The field is Klass::_super_check_offset. Return its (constant) value.
1593 // (Folds up type checking code.)
1594 assert(Opcode() == Op_LoadI, "must load an int from _super_check_offset");
1595 return TypeInt::make(klass->super_check_offset());
1596 }
1597 // Compute index into primary_supers array
1598 juint depth = (tkls->offset() - in_bytes(Klass::primary_supers_offset())) / sizeof(Klass*);
1599 // Check for overflowing; use unsigned compare to handle the negative case.
1600 if( depth < ciKlass::primary_super_limit() ) {
1601 // The field is an element of Klass::_primary_supers. Return its (constant) value.
1602 // (Folds up type checking code.)
1603 assert(Opcode() == Op_LoadKlass, "must load a klass from _primary_supers");
1604 ciKlass *ss = klass->super_of_depth(depth);
1605 return ss ? TypeKlassPtr::make(ss) : TypePtr::NULL_PTR;
1606 }
1607 const Type* aift = load_array_final_field(tkls, klass);
1608 if (aift != NULL) return aift;
1609 if (tkls->offset() == in_bytes(Klass::java_mirror_offset())) {
1610 // The field is Klass::_java_mirror. Return its (constant) value.
1611 // (Folds up the 2nd indirection in anObjConstant.getClass().)
1612 assert(Opcode() == Op_LoadP, "must load an oop from _java_mirror");
1613 return TypeInstPtr::make(klass->java_mirror());
1614 }
1615 }
1616
1617 // We can still check if we are loading from the primary_supers array at a
1618 // shallow enough depth. Even though the klass is not exact, entries less
1619 // than or equal to its super depth are correct.
1620 if (klass->is_loaded() ) {
1621 ciType *inner = klass;
1622 while( inner->is_obj_array_klass() )
1623 inner = inner->as_obj_array_klass()->base_element_type();
1624 if( inner->is_instance_klass() &&
1625 !inner->as_instance_klass()->flags().is_interface() ) {
1626 // Compute index into primary_supers array
1627 juint depth = (tkls->offset() - in_bytes(Klass::primary_supers_offset())) / sizeof(Klass*);
1628 // Check for overflowing; use unsigned compare to handle the negative case.
1629 if( depth < ciKlass::primary_super_limit() &&
1630 depth <= klass->super_depth() ) { // allow self-depth checks to handle self-check case
1631 // The field is an element of Klass::_primary_supers. Return its (constant) value.
1632 // (Folds up type checking code.)
1633 assert(Opcode() == Op_LoadKlass, "must load a klass from _primary_supers");
1634 ciKlass *ss = klass->super_of_depth(depth);
1635 return ss ? TypeKlassPtr::make(ss) : TypePtr::NULL_PTR;
1636 }
1637 }
1638 }
1639
1640 // If the type is enough to determine that the thing is not an array,
1641 // we can give the layout_helper a positive interval type.
1642 // This will help short-circuit some reflective code.
1643 if (tkls->offset() == in_bytes(Klass::layout_helper_offset())
1644 && !klass->is_array_klass() // not directly typed as an array
1645 && !klass->is_interface() // specifically not Serializable & Cloneable
1646 && !klass->is_java_lang_Object() // not the supertype of all T[]
1647 ) {
1648 // Note: When interfaces are reliable, we can narrow the interface
1649 // test to (klass != Serializable && klass != Cloneable).
1650 assert(Opcode() == Op_LoadI, "must load an int from _layout_helper");
1651 jint min_size = Klass::instance_layout_helper(oopDesc::header_size(), false);
1652 // The key property of this type is that it folds up tests
1653 // for array-ness, since it proves that the layout_helper is positive.
1654 // Thus, a generic value like the basic object layout helper works fine.
1655 return TypeInt::make(min_size, max_jint, Type::WidenMin);
1656 }
1657 }
1658
1659 // If we are loading from a freshly-allocated object, produce a zero,
1660 // if the load is provably beyond the header of the object.
1661 // (Also allow a variable load from a fresh array to produce zero.)
1662 const TypeOopPtr *tinst = tp->isa_oopptr();
1663 bool is_instance = (tinst != NULL) && tinst->is_known_instance_field();
1664 bool is_boxed_value = (tinst != NULL) && tinst->is_ptr_to_boxed_value();
1665 if (ReduceFieldZeroing || is_instance || is_boxed_value) {
1666 Node* value = can_see_stored_value(mem,phase);
1667 if (value != NULL && value->is_Con()) {
1668 assert(value->bottom_type()->higher_equal(_type),"sanity");
1669 return value->bottom_type();
1670 }
1671 }
1672
1673 if (is_instance) {
1674 // If we have an instance type and our memory input is the
1675 // programs's initial memory state, there is no matching store,
1676 // so just return a zero of the appropriate type
1677 Node *mem = in(MemNode::Memory);
1678 if (mem->is_Parm() && mem->in(0)->is_Start()) {
1679 assert(mem->as_Parm()->_con == TypeFunc::Memory, "must be memory Parm");
1680 return Type::get_zero_type(_type->basic_type());
1681 }
1682 }
1683 return _type;
1684 }
1685
1686 //------------------------------match_edge-------------------------------------
1687 // Do we Match on this edge index or not? Match only the address.
1688 uint LoadNode::match_edge(uint idx) const {
1689 return idx == MemNode::Address;
1690 }
1691
1692 //--------------------------LoadBNode::Ideal--------------------------------------
1693 //
1694 // If the previous store is to the same address as this load,
1695 // and the value stored was larger than a byte, replace this load
1696 // with the value stored truncated to a byte. If no truncation is
1697 // needed, the replacement is done in LoadNode::Identity().
1698 //
1699 Node *LoadBNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1700 Node* mem = in(MemNode::Memory);
1701 Node* value = can_see_stored_value(mem,phase);
1702 if( value && !phase->type(value)->higher_equal( _type ) ) {
1703 Node *result = phase->transform( new LShiftINode(value, phase->intcon(24)) );
1704 return new RShiftINode(result, phase->intcon(24));
1705 }
1706 // Identity call will handle the case where truncation is not needed.
1707 return LoadNode::Ideal(phase, can_reshape);
1708 }
1709
1710 const Type* LoadBNode::Value(PhaseTransform *phase) const {
1711 Node* mem = in(MemNode::Memory);
1712 Node* value = can_see_stored_value(mem,phase);
1713 if (value != NULL && value->is_Con() &&
1714 !value->bottom_type()->higher_equal(_type)) {
1715 // If the input to the store does not fit with the load's result type,
1716 // it must be truncated. We can't delay until Ideal call since
1717 // a singleton Value is needed for split_thru_phi optimization.
1718 int con = value->get_int();
1719 return TypeInt::make((con << 24) >> 24);
1720 }
1721 return LoadNode::Value(phase);
1722 }
1723
1724 //--------------------------LoadUBNode::Ideal-------------------------------------
1725 //
1726 // If the previous store is to the same address as this load,
1727 // and the value stored was larger than a byte, replace this load
1728 // with the value stored truncated to a byte. If no truncation is
1729 // needed, the replacement is done in LoadNode::Identity().
1730 //
1731 Node* LoadUBNode::Ideal(PhaseGVN* phase, bool can_reshape) {
1732 Node* mem = in(MemNode::Memory);
1733 Node* value = can_see_stored_value(mem, phase);
1734 if (value && !phase->type(value)->higher_equal(_type))
1735 return new AndINode(value, phase->intcon(0xFF));
1736 // Identity call will handle the case where truncation is not needed.
1737 return LoadNode::Ideal(phase, can_reshape);
1738 }
1739
1740 const Type* LoadUBNode::Value(PhaseTransform *phase) const {
1741 Node* mem = in(MemNode::Memory);
1742 Node* value = can_see_stored_value(mem,phase);
1743 if (value != NULL && value->is_Con() &&
1744 !value->bottom_type()->higher_equal(_type)) {
1745 // If the input to the store does not fit with the load's result type,
1746 // it must be truncated. We can't delay until Ideal call since
1747 // a singleton Value is needed for split_thru_phi optimization.
1748 int con = value->get_int();
1749 return TypeInt::make(con & 0xFF);
1750 }
1751 return LoadNode::Value(phase);
1752 }
1753
1754 //--------------------------LoadUSNode::Ideal-------------------------------------
1755 //
1756 // If the previous store is to the same address as this load,
1757 // and the value stored was larger than a char, replace this load
1758 // with the value stored truncated to a char. If no truncation is
1759 // needed, the replacement is done in LoadNode::Identity().
1760 //
1761 Node *LoadUSNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1762 Node* mem = in(MemNode::Memory);
1763 Node* value = can_see_stored_value(mem,phase);
1764 if( value && !phase->type(value)->higher_equal( _type ) )
1765 return new AndINode(value,phase->intcon(0xFFFF));
1766 // Identity call will handle the case where truncation is not needed.
1767 return LoadNode::Ideal(phase, can_reshape);
1768 }
1769
1770 const Type* LoadUSNode::Value(PhaseTransform *phase) const {
1771 Node* mem = in(MemNode::Memory);
1772 Node* value = can_see_stored_value(mem,phase);
1773 if (value != NULL && value->is_Con() &&
1774 !value->bottom_type()->higher_equal(_type)) {
1775 // If the input to the store does not fit with the load's result type,
1776 // it must be truncated. We can't delay until Ideal call since
1777 // a singleton Value is needed for split_thru_phi optimization.
1778 int con = value->get_int();
1779 return TypeInt::make(con & 0xFFFF);
1780 }
1781 return LoadNode::Value(phase);
1782 }
1783
1784 //--------------------------LoadSNode::Ideal--------------------------------------
1785 //
1786 // If the previous store is to the same address as this load,
1787 // and the value stored was larger than a short, replace this load
1788 // with the value stored truncated to a short. If no truncation is
1789 // needed, the replacement is done in LoadNode::Identity().
1790 //
1791 Node *LoadSNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1792 Node* mem = in(MemNode::Memory);
1793 Node* value = can_see_stored_value(mem,phase);
1794 if( value && !phase->type(value)->higher_equal( _type ) ) {
1795 Node *result = phase->transform( new LShiftINode(value, phase->intcon(16)) );
1796 return new RShiftINode(result, phase->intcon(16));
1797 }
1798 // Identity call will handle the case where truncation is not needed.
1799 return LoadNode::Ideal(phase, can_reshape);
1800 }
1801
1802 const Type* LoadSNode::Value(PhaseTransform *phase) const {
1803 Node* mem = in(MemNode::Memory);
1804 Node* value = can_see_stored_value(mem,phase);
1805 if (value != NULL && value->is_Con() &&
1806 !value->bottom_type()->higher_equal(_type)) {
1807 // If the input to the store does not fit with the load's result type,
1808 // it must be truncated. We can't delay until Ideal call since
1809 // a singleton Value is needed for split_thru_phi optimization.
1810 int con = value->get_int();
1811 return TypeInt::make((con << 16) >> 16);
1812 }
1813 return LoadNode::Value(phase);
1814 }
1815
1816 //=============================================================================
1817 //----------------------------LoadKlassNode::make------------------------------
1818 // Polymorphic factory method:
1819 Node* LoadKlassNode::make(PhaseGVN& gvn, Node* ctl, Node* mem, Node* adr, const TypePtr* at, const TypeKlassPtr* tk) {
1820 // sanity check the alias category against the created node type
1821 const TypePtr *adr_type = adr->bottom_type()->isa_ptr();
1822 assert(adr_type != NULL, "expecting TypeKlassPtr");
1823 #ifdef _LP64
1824 if (adr_type->is_ptr_to_narrowklass()) {
1825 assert(UseCompressedClassPointers, "no compressed klasses");
1826 Node* load_klass = gvn.transform(new LoadNKlassNode(ctl, mem, adr, at, tk->make_narrowklass(), MemNode::unordered));
1827 return new DecodeNKlassNode(load_klass, load_klass->bottom_type()->make_ptr());
1828 }
1829 #endif
1830 assert(!adr_type->is_ptr_to_narrowklass() && !adr_type->is_ptr_to_narrowoop(), "should have got back a narrow oop");
1831 return new LoadKlassNode(ctl, mem, adr, at, tk, MemNode::unordered);
1832 }
1833
1834 //------------------------------Value------------------------------------------
1835 const Type *LoadKlassNode::Value( PhaseTransform *phase ) const {
1836 return klass_value_common(phase);
1837 }
1838
1839 // In most cases, LoadKlassNode does not have the control input set. If the control
1840 // input is set, it must not be removed (by LoadNode::Ideal()).
1841 bool LoadKlassNode::can_remove_control() const {
1842 return false;
1843 }
1844
1845 const Type *LoadNode::klass_value_common( PhaseTransform *phase ) const {
1846 // Either input is TOP ==> the result is TOP
1847 const Type *t1 = phase->type( in(MemNode::Memory) );
1848 if (t1 == Type::TOP) return Type::TOP;
1849 Node *adr = in(MemNode::Address);
1850 const Type *t2 = phase->type( adr );
1851 if (t2 == Type::TOP) return Type::TOP;
1852 const TypePtr *tp = t2->is_ptr();
1853 if (TypePtr::above_centerline(tp->ptr()) ||
1854 tp->ptr() == TypePtr::Null) return Type::TOP;
1855
1856 // Return a more precise klass, if possible
1857 const TypeInstPtr *tinst = tp->isa_instptr();
1858 if (tinst != NULL) {
1859 ciInstanceKlass* ik = tinst->klass()->as_instance_klass();
1860 int offset = tinst->offset();
1861 if (ik == phase->C->env()->Class_klass()
1862 && (offset == java_lang_Class::klass_offset_in_bytes() ||
1863 offset == java_lang_Class::array_klass_offset_in_bytes())) {
1864 // We are loading a special hidden field from a Class mirror object,
1865 // the field which points to the VM's Klass metaobject.
1866 ciType* t = tinst->java_mirror_type();
1867 // java_mirror_type returns non-null for compile-time Class constants.
1868 if (t != NULL) {
1869 // constant oop => constant klass
1870 if (offset == java_lang_Class::array_klass_offset_in_bytes()) {
1871 if (t->is_void()) {
1872 // We cannot create a void array. Since void is a primitive type return null
1873 // klass. Users of this result need to do a null check on the returned klass.
1874 return TypePtr::NULL_PTR;
1875 }
1876 return TypeKlassPtr::make(ciArrayKlass::make(t));
1877 }
1878 if (!t->is_klass()) {
1879 // a primitive Class (e.g., int.class) has NULL for a klass field
1880 return TypePtr::NULL_PTR;
1881 }
1882 // (Folds up the 1st indirection in aClassConstant.getModifiers().)
1883 return TypeKlassPtr::make(t->as_klass());
1884 }
1885 // non-constant mirror, so we can't tell what's going on
1886 }
1887 if( !ik->is_loaded() )
1888 return _type; // Bail out if not loaded
1889 if (offset == oopDesc::klass_offset_in_bytes()) {
1890 if (tinst->klass_is_exact()) {
1891 return TypeKlassPtr::make(ik);
1892 }
1893 // See if we can become precise: no subklasses and no interface
1894 // (Note: We need to support verified interfaces.)
1895 if (!ik->is_interface() && !ik->has_subklass()) {
1896 //assert(!UseExactTypes, "this code should be useless with exact types");
1897 // Add a dependence; if any subclass added we need to recompile
1898 if (!ik->is_final()) {
1899 // %%% should use stronger assert_unique_concrete_subtype instead
1900 phase->C->dependencies()->assert_leaf_type(ik);
1901 }
1902 // Return precise klass
1903 return TypeKlassPtr::make(ik);
1904 }
1905
1906 // Return root of possible klass
1907 return TypeKlassPtr::make(TypePtr::NotNull, ik, 0/*offset*/);
1908 }
1909 }
1910
1911 // Check for loading klass from an array
1912 const TypeAryPtr *tary = tp->isa_aryptr();
1913 if( tary != NULL ) {
1914 ciKlass *tary_klass = tary->klass();
1915 if (tary_klass != NULL // can be NULL when at BOTTOM or TOP
1916 && tary->offset() == oopDesc::klass_offset_in_bytes()) {
1917 if (tary->klass_is_exact()) {
1918 return TypeKlassPtr::make(tary_klass);
1919 }
1920 ciArrayKlass *ak = tary->klass()->as_array_klass();
1921 // If the klass is an object array, we defer the question to the
1922 // array component klass.
1923 if( ak->is_obj_array_klass() ) {
1924 assert( ak->is_loaded(), "" );
1925 ciKlass *base_k = ak->as_obj_array_klass()->base_element_klass();
1926 if( base_k->is_loaded() && base_k->is_instance_klass() ) {
1927 ciInstanceKlass* ik = base_k->as_instance_klass();
1928 // See if we can become precise: no subklasses and no interface
1929 if (!ik->is_interface() && !ik->has_subklass()) {
1930 //assert(!UseExactTypes, "this code should be useless with exact types");
1931 // Add a dependence; if any subclass added we need to recompile
1932 if (!ik->is_final()) {
1933 phase->C->dependencies()->assert_leaf_type(ik);
1934 }
1935 // Return precise array klass
1936 return TypeKlassPtr::make(ak);
1937 }
1938 }
1939 return TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
1940 } else { // Found a type-array?
1941 //assert(!UseExactTypes, "this code should be useless with exact types");
1942 assert( ak->is_type_array_klass(), "" );
1943 return TypeKlassPtr::make(ak); // These are always precise
1944 }
1945 }
1946 }
1947
1948 // Check for loading klass from an array klass
1949 const TypeKlassPtr *tkls = tp->isa_klassptr();
1950 if (tkls != NULL && !StressReflectiveCode) {
1951 ciKlass* klass = tkls->klass();
1952 if( !klass->is_loaded() )
1953 return _type; // Bail out if not loaded
1954 if( klass->is_obj_array_klass() &&
1955 tkls->offset() == in_bytes(ObjArrayKlass::element_klass_offset())) {
1956 ciKlass* elem = klass->as_obj_array_klass()->element_klass();
1957 // // Always returning precise element type is incorrect,
1958 // // e.g., element type could be object and array may contain strings
1959 // return TypeKlassPtr::make(TypePtr::Constant, elem, 0);
1960
1961 // The array's TypeKlassPtr was declared 'precise' or 'not precise'
1962 // according to the element type's subclassing.
1963 return TypeKlassPtr::make(tkls->ptr(), elem, 0/*offset*/);
1964 }
1965 if( klass->is_instance_klass() && tkls->klass_is_exact() &&
1966 tkls->offset() == in_bytes(Klass::super_offset())) {
1967 ciKlass* sup = klass->as_instance_klass()->super();
1968 // The field is Klass::_super. Return its (constant) value.
1969 // (Folds up the 2nd indirection in aClassConstant.getSuperClass().)
1970 return sup ? TypeKlassPtr::make(sup) : TypePtr::NULL_PTR;
1971 }
1972 }
1973
1974 // Bailout case
1975 return LoadNode::Value(phase);
1976 }
1977
1978 //------------------------------Identity---------------------------------------
1979 // To clean up reflective code, simplify k.java_mirror.as_klass to plain k.
1980 // Also feed through the klass in Allocate(...klass...)._klass.
1981 Node* LoadKlassNode::Identity( PhaseTransform *phase ) {
1982 return klass_identity_common(phase);
1983 }
1984
1985 Node* LoadNode::klass_identity_common(PhaseTransform *phase ) {
1986 Node* x = LoadNode::Identity(phase);
1987 if (x != this) return x;
1988
1989 // Take apart the address into an oop and and offset.
1990 // Return 'this' if we cannot.
1991 Node* adr = in(MemNode::Address);
1992 intptr_t offset = 0;
1993 Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
1994 if (base == NULL) return this;
1995 const TypeOopPtr* toop = phase->type(adr)->isa_oopptr();
1996 if (toop == NULL) return this;
1997
1998 // We can fetch the klass directly through an AllocateNode.
1999 // This works even if the klass is not constant (clone or newArray).
2000 if (offset == oopDesc::klass_offset_in_bytes()) {
2001 Node* allocated_klass = AllocateNode::Ideal_klass(base, phase);
2002 if (allocated_klass != NULL) {
2003 return allocated_klass;
2004 }
2005 }
2006
2007 // Simplify k.java_mirror.as_klass to plain k, where k is a Klass*.
2008 // See inline_native_Class_query for occurrences of these patterns.
2009 // Java Example: x.getClass().isAssignableFrom(y)
2010 //
2011 // This improves reflective code, often making the Class
2012 // mirror go completely dead. (Current exception: Class
2013 // mirrors may appear in debug info, but we could clean them out by
2014 // introducing a new debug info operator for Klass*.java_mirror).
2015 if (toop->isa_instptr() && toop->klass() == phase->C->env()->Class_klass()
2016 && offset == java_lang_Class::klass_offset_in_bytes()) {
2017 // We are loading a special hidden field from a Class mirror,
2018 // the field which points to its Klass or ArrayKlass metaobject.
2019 if (base->is_Load()) {
2020 Node* adr2 = base->in(MemNode::Address);
2021 const TypeKlassPtr* tkls = phase->type(adr2)->isa_klassptr();
2022 if (tkls != NULL && !tkls->empty()
2023 && (tkls->klass()->is_instance_klass() ||
2024 tkls->klass()->is_array_klass())
2025 && adr2->is_AddP()
2026 ) {
2027 int mirror_field = in_bytes(Klass::java_mirror_offset());
2028 if (tkls->offset() == mirror_field) {
2029 return adr2->in(AddPNode::Base);
2030 }
2031 }
2032 }
2033 }
2034
2035 return this;
2036 }
2037
2038
2039 //------------------------------Value------------------------------------------
2040 const Type *LoadNKlassNode::Value( PhaseTransform *phase ) const {
2041 const Type *t = klass_value_common(phase);
2042 if (t == Type::TOP)
2043 return t;
2044
2045 return t->make_narrowklass();
2046 }
2047
2048 //------------------------------Identity---------------------------------------
2049 // To clean up reflective code, simplify k.java_mirror.as_klass to narrow k.
2050 // Also feed through the klass in Allocate(...klass...)._klass.
2051 Node* LoadNKlassNode::Identity( PhaseTransform *phase ) {
2052 Node *x = klass_identity_common(phase);
2053
2054 const Type *t = phase->type( x );
2055 if( t == Type::TOP ) return x;
2056 if( t->isa_narrowklass()) return x;
2057 assert (!t->isa_narrowoop(), "no narrow oop here");
2058
2059 return phase->transform(new EncodePKlassNode(x, t->make_narrowklass()));
2060 }
2061
2062 //------------------------------Value-----------------------------------------
2063 const Type *LoadRangeNode::Value( PhaseTransform *phase ) const {
2064 // Either input is TOP ==> the result is TOP
2065 const Type *t1 = phase->type( in(MemNode::Memory) );
2066 if( t1 == Type::TOP ) return Type::TOP;
2067 Node *adr = in(MemNode::Address);
2068 const Type *t2 = phase->type( adr );
2069 if( t2 == Type::TOP ) return Type::TOP;
2070 const TypePtr *tp = t2->is_ptr();
2071 if (TypePtr::above_centerline(tp->ptr())) return Type::TOP;
2072 const TypeAryPtr *tap = tp->isa_aryptr();
2073 if( !tap ) return _type;
2074 return tap->size();
2075 }
2076
2077 //-------------------------------Ideal---------------------------------------
2078 // Feed through the length in AllocateArray(...length...)._length.
2079 Node *LoadRangeNode::Ideal(PhaseGVN *phase, bool can_reshape) {
2080 Node* p = MemNode::Ideal_common(phase, can_reshape);
2081 if (p) return (p == NodeSentinel) ? NULL : p;
2082
2083 // Take apart the address into an oop and and offset.
2084 // Return 'this' if we cannot.
2085 Node* adr = in(MemNode::Address);
2086 intptr_t offset = 0;
2087 Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
2088 if (base == NULL) return NULL;
2089 const TypeAryPtr* tary = phase->type(adr)->isa_aryptr();
2090 if (tary == NULL) return NULL;
2091
2092 // We can fetch the length directly through an AllocateArrayNode.
2093 // This works even if the length is not constant (clone or newArray).
2094 if (offset == arrayOopDesc::length_offset_in_bytes()) {
2095 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(base, phase);
2096 if (alloc != NULL) {
2097 Node* allocated_length = alloc->Ideal_length();
2098 Node* len = alloc->make_ideal_length(tary, phase);
2099 if (allocated_length != len) {
2100 // New CastII improves on this.
2101 return len;
2102 }
2103 }
2104 }
2105
2106 return NULL;
2107 }
2108
2109 //------------------------------Identity---------------------------------------
2110 // Feed through the length in AllocateArray(...length...)._length.
2111 Node* LoadRangeNode::Identity( PhaseTransform *phase ) {
2112 Node* x = LoadINode::Identity(phase);
2113 if (x != this) return x;
2114
2115 // Take apart the address into an oop and and offset.
2116 // Return 'this' if we cannot.
2117 Node* adr = in(MemNode::Address);
2118 intptr_t offset = 0;
2119 Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
2120 if (base == NULL) return this;
2121 const TypeAryPtr* tary = phase->type(adr)->isa_aryptr();
2122 if (tary == NULL) return this;
2123
2124 // We can fetch the length directly through an AllocateArrayNode.
2125 // This works even if the length is not constant (clone or newArray).
2126 if (offset == arrayOopDesc::length_offset_in_bytes()) {
2127 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(base, phase);
2128 if (alloc != NULL) {
2129 Node* allocated_length = alloc->Ideal_length();
2130 // Do not allow make_ideal_length to allocate a CastII node.
2131 Node* len = alloc->make_ideal_length(tary, phase, false);
2132 if (allocated_length == len) {
2133 // Return allocated_length only if it would not be improved by a CastII.
2134 return allocated_length;
2135 }
2136 }
2137 }
2138
2139 return this;
2140
2141 }
2142
2143 //=============================================================================
2144 //---------------------------StoreNode::make-----------------------------------
2145 // Polymorphic factory method:
2146 StoreNode* StoreNode::make(PhaseGVN& gvn, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val, BasicType bt, MemOrd mo) {
2147 assert((mo == unordered || mo == release), "unexpected");
2148 Compile* C = gvn.C;
2149 assert(C->get_alias_index(adr_type) != Compile::AliasIdxRaw ||
2150 ctl != NULL, "raw memory operations should have control edge");
2151
2152 switch (bt) {
2153 case T_BOOLEAN:
2154 case T_BYTE: return new StoreBNode(ctl, mem, adr, adr_type, val, mo);
2155 case T_INT: return new StoreINode(ctl, mem, adr, adr_type, val, mo);
2156 case T_CHAR:
2157 case T_SHORT: return new StoreCNode(ctl, mem, adr, adr_type, val, mo);
2158 case T_LONG: return new StoreLNode(ctl, mem, adr, adr_type, val, mo);
2159 case T_FLOAT: return new StoreFNode(ctl, mem, adr, adr_type, val, mo);
2160 case T_DOUBLE: return new StoreDNode(ctl, mem, adr, adr_type, val, mo);
2161 case T_METADATA:
2162 case T_ADDRESS:
2163 case T_OBJECT:
2164 #ifdef _LP64
2165 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2166 val = gvn.transform(new EncodePNode(val, val->bottom_type()->make_narrowoop()));
2167 return new StoreNNode(ctl, mem, adr, adr_type, val, mo);
2168 } else if (adr->bottom_type()->is_ptr_to_narrowklass() ||
2169 (UseCompressedClassPointers && val->bottom_type()->isa_klassptr() &&
2170 adr->bottom_type()->isa_rawptr())) {
2171 val = gvn.transform(new EncodePKlassNode(val, val->bottom_type()->make_narrowklass()));
2172 return new StoreNKlassNode(ctl, mem, adr, adr_type, val, mo);
2173 }
2174 #endif
2175 {
2176 return new StorePNode(ctl, mem, adr, adr_type, val, mo);
2177 }
2178 }
2179 ShouldNotReachHere();
2180 return (StoreNode*)NULL;
2181 }
2182
2183 StoreLNode* StoreLNode::make_atomic(Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val, MemOrd mo) {
2184 bool require_atomic = true;
2185 return new StoreLNode(ctl, mem, adr, adr_type, val, mo, require_atomic);
2186 }
2187
2188 StoreDNode* StoreDNode::make_atomic(Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val, MemOrd mo) {
2189 bool require_atomic = true;
2190 return new StoreDNode(ctl, mem, adr, adr_type, val, mo, require_atomic);
2191 }
2192
2193
2194 //--------------------------bottom_type----------------------------------------
2195 const Type *StoreNode::bottom_type() const {
2196 return Type::MEMORY;
2197 }
2198
2199 //------------------------------hash-------------------------------------------
2200 uint StoreNode::hash() const {
2201 // unroll addition of interesting fields
2202 //return (uintptr_t)in(Control) + (uintptr_t)in(Memory) + (uintptr_t)in(Address) + (uintptr_t)in(ValueIn);
2203
2204 // Since they are not commoned, do not hash them:
2205 return NO_HASH;
2206 }
2207
2208 //------------------------------Ideal------------------------------------------
2209 // Change back-to-back Store(, p, x) -> Store(m, p, y) to Store(m, p, x).
2210 // When a store immediately follows a relevant allocation/initialization,
2211 // try to capture it into the initialization, or hoist it above.
2212 Node *StoreNode::Ideal(PhaseGVN *phase, bool can_reshape) {
2213 Node* p = MemNode::Ideal_common(phase, can_reshape);
2214 if (p) return (p == NodeSentinel) ? NULL : p;
2215
2216 Node* mem = in(MemNode::Memory);
2217 Node* address = in(MemNode::Address);
2218
2219 // Back-to-back stores to same address? Fold em up. Generally
2220 // unsafe if I have intervening uses... Also disallowed for StoreCM
2221 // since they must follow each StoreP operation. Redundant StoreCMs
2222 // are eliminated just before matching in final_graph_reshape.
2223 if (mem->is_Store() && mem->in(MemNode::Address)->eqv_uncast(address) &&
2224 mem->Opcode() != Op_StoreCM) {
2225 // Looking at a dead closed cycle of memory?
2226 assert(mem != mem->in(MemNode::Memory), "dead loop in StoreNode::Ideal");
2227
2228 assert(Opcode() == mem->Opcode() ||
2229 phase->C->get_alias_index(adr_type()) == Compile::AliasIdxRaw,
2230 "no mismatched stores, except on raw memory");
2231
2232 if (mem->outcnt() == 1 && // check for intervening uses
2233 mem->as_Store()->memory_size() <= this->memory_size()) {
2234 // If anybody other than 'this' uses 'mem', we cannot fold 'mem' away.
2235 // For example, 'mem' might be the final state at a conditional return.
2236 // Or, 'mem' might be used by some node which is live at the same time
2237 // 'this' is live, which might be unschedulable. So, require exactly
2238 // ONE user, the 'this' store, until such time as we clone 'mem' for
2239 // each of 'mem's uses (thus making the exactly-1-user-rule hold true).
2240 if (can_reshape) { // (%%% is this an anachronism?)
2241 set_req_X(MemNode::Memory, mem->in(MemNode::Memory),
2242 phase->is_IterGVN());
2243 } else {
2244 // It's OK to do this in the parser, since DU info is always accurate,
2245 // and the parser always refers to nodes via SafePointNode maps.
2246 set_req(MemNode::Memory, mem->in(MemNode::Memory));
2247 }
2248 return this;
2249 }
2250 }
2251
2252 // Capture an unaliased, unconditional, simple store into an initializer.
2253 // Or, if it is independent of the allocation, hoist it above the allocation.
2254 if (ReduceFieldZeroing && /*can_reshape &&*/
2255 mem->is_Proj() && mem->in(0)->is_Initialize()) {
2256 InitializeNode* init = mem->in(0)->as_Initialize();
2257 intptr_t offset = init->can_capture_store(this, phase, can_reshape);
2258 if (offset > 0) {
2259 Node* moved = init->capture_store(this, offset, phase, can_reshape);
2260 // If the InitializeNode captured me, it made a raw copy of me,
2261 // and I need to disappear.
2262 if (moved != NULL) {
2263 // %%% hack to ensure that Ideal returns a new node:
2264 mem = MergeMemNode::make(mem);
2265 return mem; // fold me away
2266 }
2267 }
2268 }
2269
2270 return NULL; // No further progress
2271 }
2272
2273 //------------------------------Value-----------------------------------------
2274 const Type *StoreNode::Value( PhaseTransform *phase ) const {
2275 // Either input is TOP ==> the result is TOP
2276 const Type *t1 = phase->type( in(MemNode::Memory) );
2277 if( t1 == Type::TOP ) return Type::TOP;
2278 const Type *t2 = phase->type( in(MemNode::Address) );
2279 if( t2 == Type::TOP ) return Type::TOP;
2280 const Type *t3 = phase->type( in(MemNode::ValueIn) );
2281 if( t3 == Type::TOP ) return Type::TOP;
2282 return Type::MEMORY;
2283 }
2284
2285 //------------------------------Identity---------------------------------------
2286 // Remove redundant stores:
2287 // Store(m, p, Load(m, p)) changes to m.
2288 // Store(, p, x) -> Store(m, p, x) changes to Store(m, p, x).
2289 Node *StoreNode::Identity( PhaseTransform *phase ) {
2290 Node* mem = in(MemNode::Memory);
2291 Node* adr = in(MemNode::Address);
2292 Node* val = in(MemNode::ValueIn);
2293
2294 // Load then Store? Then the Store is useless
2295 if (val->is_Load() &&
2296 val->in(MemNode::Address)->eqv_uncast(adr) &&
2297 val->in(MemNode::Memory )->eqv_uncast(mem) &&
2298 val->as_Load()->store_Opcode() == Opcode()) {
2299 return mem;
2300 }
2301
2302 // Two stores in a row of the same value?
2303 if (mem->is_Store() &&
2304 mem->in(MemNode::Address)->eqv_uncast(adr) &&
2305 mem->in(MemNode::ValueIn)->eqv_uncast(val) &&
2306 mem->Opcode() == Opcode()) {
2307 return mem;
2308 }
2309
2310 // Store of zero anywhere into a freshly-allocated object?
2311 // Then the store is useless.
2312 // (It must already have been captured by the InitializeNode.)
2313 if (ReduceFieldZeroing && phase->type(val)->is_zero_type()) {
2314 // a newly allocated object is already all-zeroes everywhere
2315 if (mem->is_Proj() && mem->in(0)->is_Allocate()) {
2316 return mem;
2317 }
2318
2319 // the store may also apply to zero-bits in an earlier object
2320 Node* prev_mem = find_previous_store(phase);
2321 // Steps (a), (b): Walk past independent stores to find an exact match.
2322 if (prev_mem != NULL) {
2323 Node* prev_val = can_see_stored_value(prev_mem, phase);
2324 if (prev_val != NULL && phase->eqv(prev_val, val)) {
2325 // prev_val and val might differ by a cast; it would be good
2326 // to keep the more informative of the two.
2327 return mem;
2328 }
2329 }
2330 }
2331
2332 return this;
2333 }
2334
2335 //------------------------------match_edge-------------------------------------
2336 // Do we Match on this edge index or not? Match only memory & value
2337 uint StoreNode::match_edge(uint idx) const {
2338 return idx == MemNode::Address || idx == MemNode::ValueIn;
2339 }
2340
2341 //------------------------------cmp--------------------------------------------
2342 // Do not common stores up together. They generally have to be split
2343 // back up anyways, so do not bother.
2344 uint StoreNode::cmp( const Node &n ) const {
2345 return (&n == this); // Always fail except on self
2346 }
2347
2348 //------------------------------Ideal_masked_input-----------------------------
2349 // Check for a useless mask before a partial-word store
2350 // (StoreB ... (AndI valIn conIa) )
2351 // If (conIa & mask == mask) this simplifies to
2352 // (StoreB ... (valIn) )
2353 Node *StoreNode::Ideal_masked_input(PhaseGVN *phase, uint mask) {
2354 Node *val = in(MemNode::ValueIn);
2355 if( val->Opcode() == Op_AndI ) {
2356 const TypeInt *t = phase->type( val->in(2) )->isa_int();
2357 if( t && t->is_con() && (t->get_con() & mask) == mask ) {
2358 set_req(MemNode::ValueIn, val->in(1));
2359 return this;
2360 }
2361 }
2362 return NULL;
2363 }
2364
2365
2366 //------------------------------Ideal_sign_extended_input----------------------
2367 // Check for useless sign-extension before a partial-word store
2368 // (StoreB ... (RShiftI _ (LShiftI _ valIn conIL ) conIR) )
2369 // If (conIL == conIR && conIR <= num_bits) this simplifies to
2370 // (StoreB ... (valIn) )
2371 Node *StoreNode::Ideal_sign_extended_input(PhaseGVN *phase, int num_bits) {
2372 Node *val = in(MemNode::ValueIn);
2373 if( val->Opcode() == Op_RShiftI ) {
2374 const TypeInt *t = phase->type( val->in(2) )->isa_int();
2375 if( t && t->is_con() && (t->get_con() <= num_bits) ) {
2376 Node *shl = val->in(1);
2377 if( shl->Opcode() == Op_LShiftI ) {
2378 const TypeInt *t2 = phase->type( shl->in(2) )->isa_int();
2379 if( t2 && t2->is_con() && (t2->get_con() == t->get_con()) ) {
2380 set_req(MemNode::ValueIn, shl->in(1));
2381 return this;
2382 }
2383 }
2384 }
2385 }
2386 return NULL;
2387 }
2388
2389 //------------------------------value_never_loaded-----------------------------------
2390 // Determine whether there are any possible loads of the value stored.
2391 // For simplicity, we actually check if there are any loads from the
2392 // address stored to, not just for loads of the value stored by this node.
2393 //
2394 bool StoreNode::value_never_loaded( PhaseTransform *phase) const {
2395 Node *adr = in(Address);
2396 const TypeOopPtr *adr_oop = phase->type(adr)->isa_oopptr();
2397 if (adr_oop == NULL)
2398 return false;
2399 if (!adr_oop->is_known_instance_field())
2400 return false; // if not a distinct instance, there may be aliases of the address
2401 for (DUIterator_Fast imax, i = adr->fast_outs(imax); i < imax; i++) {
2402 Node *use = adr->fast_out(i);
2403 if (use->is_Load() || use->is_LoadStore()) {
2404 return false;
2405 }
2406 }
2407 return true;
2408 }
2409
2410 //=============================================================================
2411 //------------------------------Ideal------------------------------------------
2412 // If the store is from an AND mask that leaves the low bits untouched, then
2413 // we can skip the AND operation. If the store is from a sign-extension
2414 // (a left shift, then right shift) we can skip both.
2415 Node *StoreBNode::Ideal(PhaseGVN *phase, bool can_reshape){
2416 Node *progress = StoreNode::Ideal_masked_input(phase, 0xFF);
2417 if( progress != NULL ) return progress;
2418
2419 progress = StoreNode::Ideal_sign_extended_input(phase, 24);
2420 if( progress != NULL ) return progress;
2421
2422 // Finally check the default case
2423 return StoreNode::Ideal(phase, can_reshape);
2424 }
2425
2426 //=============================================================================
2427 //------------------------------Ideal------------------------------------------
2428 // If the store is from an AND mask that leaves the low bits untouched, then
2429 // we can skip the AND operation
2430 Node *StoreCNode::Ideal(PhaseGVN *phase, bool can_reshape){
2431 Node *progress = StoreNode::Ideal_masked_input(phase, 0xFFFF);
2432 if( progress != NULL ) return progress;
2433
2434 progress = StoreNode::Ideal_sign_extended_input(phase, 16);
2435 if( progress != NULL ) return progress;
2436
2437 // Finally check the default case
2438 return StoreNode::Ideal(phase, can_reshape);
2439 }
2440
2441 //=============================================================================
2442 //------------------------------Identity---------------------------------------
2443 Node *StoreCMNode::Identity( PhaseTransform *phase ) {
2444 // No need to card mark when storing a null ptr
2445 Node* my_store = in(MemNode::OopStore);
2446 if (my_store->is_Store()) {
2447 const Type *t1 = phase->type( my_store->in(MemNode::ValueIn) );
2448 if( t1 == TypePtr::NULL_PTR ) {
2449 return in(MemNode::Memory);
2450 }
2451 }
2452 return this;
2453 }
2454
2455 //=============================================================================
2456 //------------------------------Ideal---------------------------------------
2457 Node *StoreCMNode::Ideal(PhaseGVN *phase, bool can_reshape){
2458 Node* progress = StoreNode::Ideal(phase, can_reshape);
2459 if (progress != NULL) return progress;
2460
2461 Node* my_store = in(MemNode::OopStore);
2462 if (my_store->is_MergeMem()) {
2463 Node* mem = my_store->as_MergeMem()->memory_at(oop_alias_idx());
2464 set_req(MemNode::OopStore, mem);
2465 return this;
2466 }
2467
2468 return NULL;
2469 }
2470
2471 //------------------------------Value-----------------------------------------
2472 const Type *StoreCMNode::Value( PhaseTransform *phase ) const {
2473 // Either input is TOP ==> the result is TOP
2474 const Type *t = phase->type( in(MemNode::Memory) );
2475 if( t == Type::TOP ) return Type::TOP;
2476 t = phase->type( in(MemNode::Address) );
2477 if( t == Type::TOP ) return Type::TOP;
2478 t = phase->type( in(MemNode::ValueIn) );
2479 if( t == Type::TOP ) return Type::TOP;
2480 // If extra input is TOP ==> the result is TOP
2481 t = phase->type( in(MemNode::OopStore) );
2482 if( t == Type::TOP ) return Type::TOP;
2483
2484 return StoreNode::Value( phase );
2485 }
2486
2487
2488 //=============================================================================
2489 //----------------------------------SCMemProjNode------------------------------
2490 const Type * SCMemProjNode::Value( PhaseTransform *phase ) const
2491 {
2492 return bottom_type();
2493 }
2494
2495 //=============================================================================
2496 //----------------------------------LoadStoreNode------------------------------
2497 LoadStoreNode::LoadStoreNode( Node *c, Node *mem, Node *adr, Node *val, const TypePtr* at, const Type* rt, uint required )
2498 : Node(required),
2499 _type(rt),
2500 _adr_type(at)
2501 {
2502 init_req(MemNode::Control, c );
2503 init_req(MemNode::Memory , mem);
2504 init_req(MemNode::Address, adr);
2505 init_req(MemNode::ValueIn, val);
2506 init_class_id(Class_LoadStore);
2507 }
2508
2509 uint LoadStoreNode::ideal_reg() const {
2510 return _type->ideal_reg();
2511 }
2512
2513 bool LoadStoreNode::result_not_used() const {
2514 for( DUIterator_Fast imax, i = fast_outs(imax); i < imax; i++ ) {
2515 Node *x = fast_out(i);
2516 if (x->Opcode() == Op_SCMemProj) continue;
2517 return false;
2518 }
2519 return true;
2520 }
2521
2522 uint LoadStoreNode::size_of() const { return sizeof(*this); }
2523
2524 //=============================================================================
2525 //----------------------------------LoadStoreConditionalNode--------------------
2526 LoadStoreConditionalNode::LoadStoreConditionalNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex ) : LoadStoreNode(c, mem, adr, val, NULL, TypeInt::BOOL, 5) {
2527 init_req(ExpectedIn, ex );
2528 }
2529
2530 //=============================================================================
2531 //-------------------------------adr_type--------------------------------------
2532 // Do we Match on this edge index or not? Do not match memory
2533 const TypePtr* ClearArrayNode::adr_type() const {
2534 Node *adr = in(3);
2535 if (adr == NULL) return NULL; // node is dead
2536 return MemNode::calculate_adr_type(adr->bottom_type());
2537 }
2538
2539 //------------------------------match_edge-------------------------------------
2540 // Do we Match on this edge index or not? Do not match memory
2541 uint ClearArrayNode::match_edge(uint idx) const {
2542 return idx > 1;
2543 }
2544
2545 //------------------------------Identity---------------------------------------
2546 // Clearing a zero length array does nothing
2547 Node *ClearArrayNode::Identity( PhaseTransform *phase ) {
2548 return phase->type(in(2))->higher_equal(TypeX::ZERO) ? in(1) : this;
2549 }
2550
2551 //------------------------------Idealize---------------------------------------
2552 // Clearing a short array is faster with stores
2553 Node *ClearArrayNode::Ideal(PhaseGVN *phase, bool can_reshape){
2554 const int unit = BytesPerLong;
2555 const TypeX* t = phase->type(in(2))->isa_intptr_t();
2556 if (!t) return NULL;
2557 if (!t->is_con()) return NULL;
2558 intptr_t raw_count = t->get_con();
2559 intptr_t size = raw_count;
2560 if (!Matcher::init_array_count_is_in_bytes) size *= unit;
2561 // Clearing nothing uses the Identity call.
2562 // Negative clears are possible on dead ClearArrays
2563 // (see jck test stmt114.stmt11402.val).
2564 if (size <= 0 || size % unit != 0) return NULL;
2565 intptr_t count = size / unit;
2566 // Length too long; use fast hardware clear
2567 if (size > Matcher::init_array_short_size) return NULL;
2568 Node *mem = in(1);
2569 if( phase->type(mem)==Type::TOP ) return NULL;
2570 Node *adr = in(3);
2571 const Type* at = phase->type(adr);
2572 if( at==Type::TOP ) return NULL;
2573 const TypePtr* atp = at->isa_ptr();
2574 // adjust atp to be the correct array element address type
2575 if (atp == NULL) atp = TypePtr::BOTTOM;
2576 else atp = atp->add_offset(Type::OffsetBot);
2577 // Get base for derived pointer purposes
2578 if( adr->Opcode() != Op_AddP ) Unimplemented();
2579 Node *base = adr->in(1);
2580
2581 Node *zero = phase->makecon(TypeLong::ZERO);
2582 Node *off = phase->MakeConX(BytesPerLong);
2583 mem = new StoreLNode(in(0),mem,adr,atp,zero,MemNode::unordered,false);
2584 count--;
2585 while( count-- ) {
2586 mem = phase->transform(mem);
2587 adr = phase->transform(new AddPNode(base,adr,off));
2588 mem = new StoreLNode(in(0),mem,adr,atp,zero,MemNode::unordered,false);
2589 }
2590 return mem;
2591 }
2592
2593 //----------------------------step_through----------------------------------
2594 // Return allocation input memory edge if it is different instance
2595 // or itself if it is the one we are looking for.
2596 bool ClearArrayNode::step_through(Node** np, uint instance_id, PhaseTransform* phase) {
2597 Node* n = *np;
2598 assert(n->is_ClearArray(), "sanity");
2599 intptr_t offset;
2600 AllocateNode* alloc = AllocateNode::Ideal_allocation(n->in(3), phase, offset);
2601 // This method is called only before Allocate nodes are expanded
2602 // during macro nodes expansion. Before that ClearArray nodes are
2603 // only generated in PhaseMacroExpand::generate_arraycopy() (before
2604 // Allocate nodes are expanded) which follows allocations.
2605 assert(alloc != NULL, "should have allocation");
2606 if (alloc->_idx == instance_id) {
2607 // Can not bypass initialization of the instance we are looking for.
2608 return false;
2609 }
2610 // Otherwise skip it.
2611 InitializeNode* init = alloc->initialization();
2612 if (init != NULL)
2613 *np = init->in(TypeFunc::Memory);
2614 else
2615 *np = alloc->in(TypeFunc::Memory);
2616 return true;
2617 }
2618
2619 //----------------------------clear_memory-------------------------------------
2620 // Generate code to initialize object storage to zero.
2621 Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
2622 intptr_t start_offset,
2623 Node* end_offset,
2624 PhaseGVN* phase) {
2625 intptr_t offset = start_offset;
2626
2627 int unit = BytesPerLong;
2628 if ((offset % unit) != 0) {
2629 Node* adr = new AddPNode(dest, dest, phase->MakeConX(offset));
2630 adr = phase->transform(adr);
2631 const TypePtr* atp = TypeRawPtr::BOTTOM;
2632 mem = StoreNode::make(*phase, ctl, mem, adr, atp, phase->zerocon(T_INT), T_INT, MemNode::unordered);
2633 mem = phase->transform(mem);
2634 offset += BytesPerInt;
2635 }
2636 assert((offset % unit) == 0, "");
2637
2638 // Initialize the remaining stuff, if any, with a ClearArray.
2639 return clear_memory(ctl, mem, dest, phase->MakeConX(offset), end_offset, phase);
2640 }
2641
2642 Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
2643 Node* start_offset,
2644 Node* end_offset,
2645 PhaseGVN* phase) {
2646 if (start_offset == end_offset) {
2647 // nothing to do
2648 return mem;
2649 }
2650
2651 int unit = BytesPerLong;
2652 Node* zbase = start_offset;
2653 Node* zend = end_offset;
2654
2655 // Scale to the unit required by the CPU:
2656 if (!Matcher::init_array_count_is_in_bytes) {
2657 Node* shift = phase->intcon(exact_log2(unit));
2658 zbase = phase->transform(new URShiftXNode(zbase, shift) );
2659 zend = phase->transform(new URShiftXNode(zend, shift) );
2660 }
2661
2662 // Bulk clear double-words
2663 Node* zsize = phase->transform(new SubXNode(zend, zbase) );
2664 Node* adr = phase->transform(new AddPNode(dest, dest, start_offset) );
2665 mem = new ClearArrayNode(ctl, mem, zsize, adr);
2666 return phase->transform(mem);
2667 }
2668
2669 Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
2670 intptr_t start_offset,
2671 intptr_t end_offset,
2672 PhaseGVN* phase) {
2673 if (start_offset == end_offset) {
2674 // nothing to do
2675 return mem;
2676 }
2677
2678 assert((end_offset % BytesPerInt) == 0, "odd end offset");
2679 intptr_t done_offset = end_offset;
2680 if ((done_offset % BytesPerLong) != 0) {
2681 done_offset -= BytesPerInt;
2682 }
2683 if (done_offset > start_offset) {
2684 mem = clear_memory(ctl, mem, dest,
2685 start_offset, phase->MakeConX(done_offset), phase);
2686 }
2687 if (done_offset < end_offset) { // emit the final 32-bit store
2688 Node* adr = new AddPNode(dest, dest, phase->MakeConX(done_offset));
2689 adr = phase->transform(adr);
2690 const TypePtr* atp = TypeRawPtr::BOTTOM;
2691 mem = StoreNode::make(*phase, ctl, mem, adr, atp, phase->zerocon(T_INT), T_INT, MemNode::unordered);
2692 mem = phase->transform(mem);
2693 done_offset += BytesPerInt;
2694 }
2695 assert(done_offset == end_offset, "");
2696 return mem;
2697 }
2698
2699 //=============================================================================
2700 MemBarNode::MemBarNode(Compile* C, int alias_idx, Node* precedent)
2701 : MultiNode(TypeFunc::Parms + (precedent == NULL? 0: 1)),
2702 _adr_type(C->get_adr_type(alias_idx))
2703 {
2704 init_class_id(Class_MemBar);
2705 Node* top = C->top();
2706 init_req(TypeFunc::I_O,top);
2707 init_req(TypeFunc::FramePtr,top);
2708 init_req(TypeFunc::ReturnAdr,top);
2709 if (precedent != NULL)
2710 init_req(TypeFunc::Parms, precedent);
2711 }
2712
2713 //------------------------------cmp--------------------------------------------
2714 uint MemBarNode::hash() const { return NO_HASH; }
2715 uint MemBarNode::cmp( const Node &n ) const {
2716 return (&n == this); // Always fail except on self
2717 }
2718
2719 //------------------------------make-------------------------------------------
2720 MemBarNode* MemBarNode::make(Compile* C, int opcode, int atp, Node* pn) {
2721 switch (opcode) {
2722 case Op_MemBarAcquire: return new MemBarAcquireNode(C, atp, pn);
2723 case Op_LoadFence: return new LoadFenceNode(C, atp, pn);
2724 case Op_MemBarRelease: return new MemBarReleaseNode(C, atp, pn);
2725 case Op_StoreFence: return new StoreFenceNode(C, atp, pn);
2726 case Op_MemBarAcquireLock: return new MemBarAcquireLockNode(C, atp, pn);
2727 case Op_MemBarReleaseLock: return new MemBarReleaseLockNode(C, atp, pn);
2728 case Op_MemBarVolatile: return new MemBarVolatileNode(C, atp, pn);
2729 case Op_MemBarCPUOrder: return new MemBarCPUOrderNode(C, atp, pn);
2730 case Op_Initialize: return new InitializeNode(C, atp, pn);
2731 case Op_MemBarStoreStore: return new MemBarStoreStoreNode(C, atp, pn);
2732 default: ShouldNotReachHere(); return NULL;
2733 }
2734 }
2735
2736 //------------------------------Ideal------------------------------------------
2737 // Return a node which is more "ideal" than the current node. Strip out
2738 // control copies
2739 Node *MemBarNode::Ideal(PhaseGVN *phase, bool can_reshape) {
2740 if (remove_dead_region(phase, can_reshape)) return this;
2741 // Don't bother trying to transform a dead node
2742 if (in(0) && in(0)->is_top()) {
2743 return NULL;
2744 }
2745
2746 bool progress = false;
2747 // Eliminate volatile MemBars for scalar replaced objects.
2748 if (can_reshape && req() == (Precedent+1)) {
2749 bool eliminate = false;
2750 int opc = Opcode();
2751 if ((opc == Op_MemBarAcquire || opc == Op_MemBarVolatile)) {
2752 // Volatile field loads and stores.
2753 Node* my_mem = in(MemBarNode::Precedent);
2754 // The MembarAquire may keep an unused LoadNode alive through the Precedent edge
2755 if ((my_mem != NULL) && (opc == Op_MemBarAcquire) && (my_mem->outcnt() == 1)) {
2756 // if the Precedent is a decodeN and its input (a Load) is used at more than one place,
2757 // replace this Precedent (decodeN) with the Load instead.
2758 if ((my_mem->Opcode() == Op_DecodeN) && (my_mem->in(1)->outcnt() > 1)) {
2759 Node* load_node = my_mem->in(1);
2760 set_req(MemBarNode::Precedent, load_node);
2761 phase->is_IterGVN()->_worklist.push(my_mem);
2762 my_mem = load_node;
2763 } else {
2764 assert(my_mem->unique_out() == this, "sanity");
2765 del_req(Precedent);
2766 phase->is_IterGVN()->_worklist.push(my_mem); // remove dead node later
2767 my_mem = NULL;
2768 }
2769 progress = true;
2770 }
2771 if (my_mem != NULL && my_mem->is_Mem()) {
2772 const TypeOopPtr* t_oop = my_mem->in(MemNode::Address)->bottom_type()->isa_oopptr();
2773 // Check for scalar replaced object reference.
2774 if( t_oop != NULL && t_oop->is_known_instance_field() &&
2775 t_oop->offset() != Type::OffsetBot &&
2776 t_oop->offset() != Type::OffsetTop) {
2777 eliminate = true;
2778 }
2779 }
2780 } else if (opc == Op_MemBarRelease) {
2781 // Final field stores.
2782 Node* alloc = AllocateNode::Ideal_allocation(in(MemBarNode::Precedent), phase);
2783 if ((alloc != NULL) && alloc->is_Allocate() &&
2784 alloc->as_Allocate()->_is_non_escaping) {
2785 // The allocated object does not escape.
2786 eliminate = true;
2787 }
2788 }
2789 if (eliminate) {
2790 // Replace MemBar projections by its inputs.
2791 PhaseIterGVN* igvn = phase->is_IterGVN();
2792 igvn->replace_node(proj_out(TypeFunc::Memory), in(TypeFunc::Memory));
2793 igvn->replace_node(proj_out(TypeFunc::Control), in(TypeFunc::Control));
2794 // Must return either the original node (now dead) or a new node
2795 // (Do not return a top here, since that would break the uniqueness of top.)
2796 return new ConINode(TypeInt::ZERO);
2797 }
2798 }
2799 return progress ? this : NULL;
2800 }
2801
2802 //------------------------------Value------------------------------------------
2803 const Type *MemBarNode::Value( PhaseTransform *phase ) const {
2804 if( !in(0) ) return Type::TOP;
2805 if( phase->type(in(0)) == Type::TOP )
2806 return Type::TOP;
2807 return TypeTuple::MEMBAR;
2808 }
2809
2810 //------------------------------match------------------------------------------
2811 // Construct projections for memory.
2812 Node *MemBarNode::match( const ProjNode *proj, const Matcher *m ) {
2813 switch (proj->_con) {
2814 case TypeFunc::Control:
2815 case TypeFunc::Memory:
2816 return new MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
2817 }
2818 ShouldNotReachHere();
2819 return NULL;
2820 }
2821
2822 //===========================InitializeNode====================================
2823 // SUMMARY:
2824 // This node acts as a memory barrier on raw memory, after some raw stores.
2825 // The 'cooked' oop value feeds from the Initialize, not the Allocation.
2826 // The Initialize can 'capture' suitably constrained stores as raw inits.
2827 // It can coalesce related raw stores into larger units (called 'tiles').
2828 // It can avoid zeroing new storage for memory units which have raw inits.
2829 // At macro-expansion, it is marked 'complete', and does not optimize further.
2830 //
2831 // EXAMPLE:
2832 // The object 'new short[2]' occupies 16 bytes in a 32-bit machine.
2833 // ctl = incoming control; mem* = incoming memory
2834 // (Note: A star * on a memory edge denotes I/O and other standard edges.)
2835 // First allocate uninitialized memory and fill in the header:
2836 // alloc = (Allocate ctl mem* 16 #short[].klass ...)
2837 // ctl := alloc.Control; mem* := alloc.Memory*
2838 // rawmem = alloc.Memory; rawoop = alloc.RawAddress
2839 // Then initialize to zero the non-header parts of the raw memory block:
2840 // init = (Initialize alloc.Control alloc.Memory* alloc.RawAddress)
2841 // ctl := init.Control; mem.SLICE(#short[*]) := init.Memory
2842 // After the initialize node executes, the object is ready for service:
2843 // oop := (CheckCastPP init.Control alloc.RawAddress #short[])
2844 // Suppose its body is immediately initialized as {1,2}:
2845 // store1 = (StoreC init.Control init.Memory (+ oop 12) 1)
2846 // store2 = (StoreC init.Control store1 (+ oop 14) 2)
2847 // mem.SLICE(#short[*]) := store2
2848 //
2849 // DETAILS:
2850 // An InitializeNode collects and isolates object initialization after
2851 // an AllocateNode and before the next possible safepoint. As a
2852 // memory barrier (MemBarNode), it keeps critical stores from drifting
2853 // down past any safepoint or any publication of the allocation.
2854 // Before this barrier, a newly-allocated object may have uninitialized bits.
2855 // After this barrier, it may be treated as a real oop, and GC is allowed.
2856 //
2857 // The semantics of the InitializeNode include an implicit zeroing of
2858 // the new object from object header to the end of the object.
2859 // (The object header and end are determined by the AllocateNode.)
2860 //
2861 // Certain stores may be added as direct inputs to the InitializeNode.
2862 // These stores must update raw memory, and they must be to addresses
2863 // derived from the raw address produced by AllocateNode, and with
2864 // a constant offset. They must be ordered by increasing offset.
2865 // The first one is at in(RawStores), the last at in(req()-1).
2866 // Unlike most memory operations, they are not linked in a chain,
2867 // but are displayed in parallel as users of the rawmem output of
2868 // the allocation.
2869 //
2870 // (See comments in InitializeNode::capture_store, which continue
2871 // the example given above.)
2872 //
2873 // When the associated Allocate is macro-expanded, the InitializeNode
2874 // may be rewritten to optimize collected stores. A ClearArrayNode
2875 // may also be created at that point to represent any required zeroing.
2876 // The InitializeNode is then marked 'complete', prohibiting further
2877 // capturing of nearby memory operations.
2878 //
2879 // During macro-expansion, all captured initializations which store
2880 // constant values of 32 bits or smaller are coalesced (if advantageous)
2881 // into larger 'tiles' 32 or 64 bits. This allows an object to be
2882 // initialized in fewer memory operations. Memory words which are
2883 // covered by neither tiles nor non-constant stores are pre-zeroed
2884 // by explicit stores of zero. (The code shape happens to do all
2885 // zeroing first, then all other stores, with both sequences occurring
2886 // in order of ascending offsets.)
2887 //
2888 // Alternatively, code may be inserted between an AllocateNode and its
2889 // InitializeNode, to perform arbitrary initialization of the new object.
2890 // E.g., the object copying intrinsics insert complex data transfers here.
2891 // The initialization must then be marked as 'complete' disable the
2892 // built-in zeroing semantics and the collection of initializing stores.
2893 //
2894 // While an InitializeNode is incomplete, reads from the memory state
2895 // produced by it are optimizable if they match the control edge and
2896 // new oop address associated with the allocation/initialization.
2897 // They return a stored value (if the offset matches) or else zero.
2898 // A write to the memory state, if it matches control and address,
2899 // and if it is to a constant offset, may be 'captured' by the
2900 // InitializeNode. It is cloned as a raw memory operation and rewired
2901 // inside the initialization, to the raw oop produced by the allocation.
2902 // Operations on addresses which are provably distinct (e.g., to
2903 // other AllocateNodes) are allowed to bypass the initialization.
2904 //
2905 // The effect of all this is to consolidate object initialization
2906 // (both arrays and non-arrays, both piecewise and bulk) into a
2907 // single location, where it can be optimized as a unit.
2908 //
2909 // Only stores with an offset less than TrackedInitializationLimit words
2910 // will be considered for capture by an InitializeNode. This puts a
2911 // reasonable limit on the complexity of optimized initializations.
2912
2913 //---------------------------InitializeNode------------------------------------
2914 InitializeNode::InitializeNode(Compile* C, int adr_type, Node* rawoop)
2915 : _is_complete(Incomplete), _does_not_escape(false),
2916 MemBarNode(C, adr_type, rawoop)
2917 {
2918 init_class_id(Class_Initialize);
2919
2920 assert(adr_type == Compile::AliasIdxRaw, "only valid atp");
2921 assert(in(RawAddress) == rawoop, "proper init");
2922 // Note: allocation() can be NULL, for secondary initialization barriers
2923 }
2924
2925 // Since this node is not matched, it will be processed by the
2926 // register allocator. Declare that there are no constraints
2927 // on the allocation of the RawAddress edge.
2928 const RegMask &InitializeNode::in_RegMask(uint idx) const {
2929 // This edge should be set to top, by the set_complete. But be conservative.
2930 if (idx == InitializeNode::RawAddress)
2931 return *(Compile::current()->matcher()->idealreg2spillmask[in(idx)->ideal_reg()]);
2932 return RegMask::Empty;
2933 }
2934
2935 Node* InitializeNode::memory(uint alias_idx) {
2936 Node* mem = in(Memory);
2937 if (mem->is_MergeMem()) {
2938 return mem->as_MergeMem()->memory_at(alias_idx);
2939 } else {
2940 // incoming raw memory is not split
2941 return mem;
2942 }
2943 }
2944
2945 bool InitializeNode::is_non_zero() {
2946 if (is_complete()) return false;
2947 remove_extra_zeroes();
2948 return (req() > RawStores);
2949 }
2950
2951 void InitializeNode::set_complete(PhaseGVN* phase) {
2952 assert(!is_complete(), "caller responsibility");
2953 _is_complete = Complete;
2954
2955 // After this node is complete, it contains a bunch of
2956 // raw-memory initializations. There is no need for
2957 // it to have anything to do with non-raw memory effects.
2958 // Therefore, tell all non-raw users to re-optimize themselves,
2959 // after skipping the memory effects of this initialization.
2960 PhaseIterGVN* igvn = phase->is_IterGVN();
2961 if (igvn) igvn->add_users_to_worklist(this);
2962 }
2963
2964 // convenience function
2965 // return false if the init contains any stores already
2966 bool AllocateNode::maybe_set_complete(PhaseGVN* phase) {
2967 InitializeNode* init = initialization();
2968 if (init == NULL || init->is_complete()) return false;
2969 init->remove_extra_zeroes();
2970 // for now, if this allocation has already collected any inits, bail:
2971 if (init->is_non_zero()) return false;
2972 init->set_complete(phase);
2973 return true;
2974 }
2975
2976 void InitializeNode::remove_extra_zeroes() {
2977 if (req() == RawStores) return;
2978 Node* zmem = zero_memory();
2979 uint fill = RawStores;
2980 for (uint i = fill; i < req(); i++) {
2981 Node* n = in(i);
2982 if (n->is_top() || n == zmem) continue; // skip
2983 if (fill < i) set_req(fill, n); // compact
2984 ++fill;
2985 }
2986 // delete any empty spaces created:
2987 while (fill < req()) {
2988 del_req(fill);
2989 }
2990 }
2991
2992 // Helper for remembering which stores go with which offsets.
2993 intptr_t InitializeNode::get_store_offset(Node* st, PhaseTransform* phase) {
2994 if (!st->is_Store()) return -1; // can happen to dead code via subsume_node
2995 intptr_t offset = -1;
2996 Node* base = AddPNode::Ideal_base_and_offset(st->in(MemNode::Address),
2997 phase, offset);
2998 if (base == NULL) return -1; // something is dead,
2999 if (offset < 0) return -1; // dead, dead
3000 return offset;
3001 }
3002
3003 // Helper for proving that an initialization expression is
3004 // "simple enough" to be folded into an object initialization.
3005 // Attempts to prove that a store's initial value 'n' can be captured
3006 // within the initialization without creating a vicious cycle, such as:
3007 // { Foo p = new Foo(); p.next = p; }
3008 // True for constants and parameters and small combinations thereof.
3009 bool InitializeNode::detect_init_independence(Node* n, int& count) {
3010 if (n == NULL) return true; // (can this really happen?)
3011 if (n->is_Proj()) n = n->in(0);
3012 if (n == this) return false; // found a cycle
3013 if (n->is_Con()) return true;
3014 if (n->is_Start()) return true; // params, etc., are OK
3015 if (n->is_Root()) return true; // even better
3016
3017 Node* ctl = n->in(0);
3018 if (ctl != NULL && !ctl->is_top()) {
3019 if (ctl->is_Proj()) ctl = ctl->in(0);
3020 if (ctl == this) return false;
3021
3022 // If we already know that the enclosing memory op is pinned right after
3023 // the init, then any control flow that the store has picked up
3024 // must have preceded the init, or else be equal to the init.
3025 // Even after loop optimizations (which might change control edges)
3026 // a store is never pinned *before* the availability of its inputs.
3027 if (!MemNode::all_controls_dominate(n, this))
3028 return false; // failed to prove a good control
3029 }
3030
3031 // Check data edges for possible dependencies on 'this'.
3032 if ((count += 1) > 20) return false; // complexity limit
3033 for (uint i = 1; i < n->req(); i++) {
3034 Node* m = n->in(i);
3035 if (m == NULL || m == n || m->is_top()) continue;
3036 uint first_i = n->find_edge(m);
3037 if (i != first_i) continue; // process duplicate edge just once
3038 if (!detect_init_independence(m, count)) {
3039 return false;
3040 }
3041 }
3042
3043 return true;
3044 }
3045
3046 // Here are all the checks a Store must pass before it can be moved into
3047 // an initialization. Returns zero if a check fails.
3048 // On success, returns the (constant) offset to which the store applies,
3049 // within the initialized memory.
3050 intptr_t InitializeNode::can_capture_store(StoreNode* st, PhaseTransform* phase, bool can_reshape) {
3051 const int FAIL = 0;
3052 if (st->req() != MemNode::ValueIn + 1)
3053 return FAIL; // an inscrutable StoreNode (card mark?)
3054 Node* ctl = st->in(MemNode::Control);
3055 if (!(ctl != NULL && ctl->is_Proj() && ctl->in(0) == this))
3056 return FAIL; // must be unconditional after the initialization
3057 Node* mem = st->in(MemNode::Memory);
3058 if (!(mem->is_Proj() && mem->in(0) == this))
3059 return FAIL; // must not be preceded by other stores
3060 Node* adr = st->in(MemNode::Address);
3061 intptr_t offset;
3062 AllocateNode* alloc = AllocateNode::Ideal_allocation(adr, phase, offset);
3063 if (alloc == NULL)
3064 return FAIL; // inscrutable address
3065 if (alloc != allocation())
3066 return FAIL; // wrong allocation! (store needs to float up)
3067 Node* val = st->in(MemNode::ValueIn);
3068 int complexity_count = 0;
3069 if (!detect_init_independence(val, complexity_count))
3070 return FAIL; // stored value must be 'simple enough'
3071
3072 // The Store can be captured only if nothing after the allocation
3073 // and before the Store is using the memory location that the store
3074 // overwrites.
3075 bool failed = false;
3076 // If is_complete_with_arraycopy() is true the shape of the graph is
3077 // well defined and is safe so no need for extra checks.
3078 if (!is_complete_with_arraycopy()) {
3079 // We are going to look at each use of the memory state following
3080 // the allocation to make sure nothing reads the memory that the
3081 // Store writes.
3082 const TypePtr* t_adr = phase->type(adr)->isa_ptr();
3083 int alias_idx = phase->C->get_alias_index(t_adr);
3084 ResourceMark rm;
3085 Unique_Node_List mems;
3086 mems.push(mem);
3087 Node* unique_merge = NULL;
3088 for (uint next = 0; next < mems.size(); ++next) {
3089 Node *m = mems.at(next);
3090 for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) {
3091 Node *n = m->fast_out(j);
3092 if (n->outcnt() == 0) {
3093 continue;
3094 }
3095 if (n == st) {
3096 continue;
3097 } else if (n->in(0) != NULL && n->in(0) != ctl) {
3098 // If the control of this use is different from the control
3099 // of the Store which is right after the InitializeNode then
3100 // this node cannot be between the InitializeNode and the
3101 // Store.
3102 continue;
3103 } else if (n->is_MergeMem()) {
3104 if (n->as_MergeMem()->memory_at(alias_idx) == m) {
3105 // We can hit a MergeMemNode (that will likely go away
3106 // later) that is a direct use of the memory state
3107 // following the InitializeNode on the same slice as the
3108 // store node that we'd like to capture. We need to check
3109 // the uses of the MergeMemNode.
3110 mems.push(n);
3111 }
3112 } else if (n->is_Mem()) {
3113 Node* other_adr = n->in(MemNode::Address);
3114 if (other_adr == adr) {
3115 failed = true;
3116 break;
3117 } else {
3118 const TypePtr* other_t_adr = phase->type(other_adr)->isa_ptr();
3119 if (other_t_adr != NULL) {
3120 int other_alias_idx = phase->C->get_alias_index(other_t_adr);
3121 if (other_alias_idx == alias_idx) {
3122 // A load from the same memory slice as the store right
3123 // after the InitializeNode. We check the control of the
3124 // object/array that is loaded from. If it's the same as
3125 // the store control then we cannot capture the store.
3126 assert(!n->is_Store(), "2 stores to same slice on same control?");
3127 Node* base = other_adr;
3128 assert(base->is_AddP(), err_msg_res("should be addp but is %s", base->Name()));
3129 base = base->in(AddPNode::Base);
3130 if (base != NULL) {
3131 base = base->uncast();
3132 if (base->is_Proj() && base->in(0) == alloc) {
3133 failed = true;
3134 break;
3135 }
3136 }
3137 }
3138 }
3139 }
3140 } else {
3141 failed = true;
3142 break;
3143 }
3144 }
3145 }
3146 }
3147 if (failed) {
3148 if (!can_reshape) {
3149 // We decided we couldn't capture the store during parsing. We
3150 // should try again during the next IGVN once the graph is
3151 // cleaner.
3152 phase->C->record_for_igvn(st);
3153 }
3154 return FAIL;
3155 }
3156
3157 return offset; // success
3158 }
3159
3160 // Find the captured store in(i) which corresponds to the range
3161 // [start..start+size) in the initialized object.
3162 // If there is one, return its index i. If there isn't, return the
3163 // negative of the index where it should be inserted.
3164 // Return 0 if the queried range overlaps an initialization boundary
3165 // or if dead code is encountered.
3166 // If size_in_bytes is zero, do not bother with overlap checks.
3167 int InitializeNode::captured_store_insertion_point(intptr_t start,
3168 int size_in_bytes,
3169 PhaseTransform* phase) {
3170 const int FAIL = 0, MAX_STORE = BytesPerLong;
3171
3172 if (is_complete())
3173 return FAIL; // arraycopy got here first; punt
3174
3175 assert(allocation() != NULL, "must be present");
3176
3177 // no negatives, no header fields:
3178 if (start < (intptr_t) allocation()->minimum_header_size()) return FAIL;
3179
3180 // after a certain size, we bail out on tracking all the stores:
3181 intptr_t ti_limit = (TrackedInitializationLimit * HeapWordSize);
3182 if (start >= ti_limit) return FAIL;
3183
3184 for (uint i = InitializeNode::RawStores, limit = req(); ; ) {
3185 if (i >= limit) return -(int)i; // not found; here is where to put it
3186
3187 Node* st = in(i);
3188 intptr_t st_off = get_store_offset(st, phase);
3189 if (st_off < 0) {
3190 if (st != zero_memory()) {
3191 return FAIL; // bail out if there is dead garbage
3192 }
3193 } else if (st_off > start) {
3194 // ...we are done, since stores are ordered
3195 if (st_off < start + size_in_bytes) {
3196 return FAIL; // the next store overlaps
3197 }
3198 return -(int)i; // not found; here is where to put it
3199 } else if (st_off < start) {
3200 if (size_in_bytes != 0 &&
3201 start < st_off + MAX_STORE &&
3202 start < st_off + st->as_Store()->memory_size()) {
3203 return FAIL; // the previous store overlaps
3204 }
3205 } else {
3206 if (size_in_bytes != 0 &&
3207 st->as_Store()->memory_size() != size_in_bytes) {
3208 return FAIL; // mismatched store size
3209 }
3210 return i;
3211 }
3212
3213 ++i;
3214 }
3215 }
3216
3217 // Look for a captured store which initializes at the offset 'start'
3218 // with the given size. If there is no such store, and no other
3219 // initialization interferes, then return zero_memory (the memory
3220 // projection of the AllocateNode).
3221 Node* InitializeNode::find_captured_store(intptr_t start, int size_in_bytes,
3222 PhaseTransform* phase) {
3223 assert(stores_are_sane(phase), "");
3224 int i = captured_store_insertion_point(start, size_in_bytes, phase);
3225 if (i == 0) {
3226 return NULL; // something is dead
3227 } else if (i < 0) {
3228 return zero_memory(); // just primordial zero bits here
3229 } else {
3230 Node* st = in(i); // here is the store at this position
3231 assert(get_store_offset(st->as_Store(), phase) == start, "sanity");
3232 return st;
3233 }
3234 }
3235
3236 // Create, as a raw pointer, an address within my new object at 'offset'.
3237 Node* InitializeNode::make_raw_address(intptr_t offset,
3238 PhaseTransform* phase) {
3239 Node* addr = in(RawAddress);
3240 if (offset != 0) {
3241 Compile* C = phase->C;
3242 addr = phase->transform( new AddPNode(C->top(), addr,
3243 phase->MakeConX(offset)) );
3244 }
3245 return addr;
3246 }
3247
3248 // Clone the given store, converting it into a raw store
3249 // initializing a field or element of my new object.
3250 // Caller is responsible for retiring the original store,
3251 // with subsume_node or the like.
3252 //
3253 // From the example above InitializeNode::InitializeNode,
3254 // here are the old stores to be captured:
3255 // store1 = (StoreC init.Control init.Memory (+ oop 12) 1)
3256 // store2 = (StoreC init.Control store1 (+ oop 14) 2)
3257 //
3258 // Here is the changed code; note the extra edges on init:
3259 // alloc = (Allocate ...)
3260 // rawoop = alloc.RawAddress
3261 // rawstore1 = (StoreC alloc.Control alloc.Memory (+ rawoop 12) 1)
3262 // rawstore2 = (StoreC alloc.Control alloc.Memory (+ rawoop 14) 2)
3263 // init = (Initialize alloc.Control alloc.Memory rawoop
3264 // rawstore1 rawstore2)
3265 //
3266 Node* InitializeNode::capture_store(StoreNode* st, intptr_t start,
3267 PhaseTransform* phase, bool can_reshape) {
3268 assert(stores_are_sane(phase), "");
3269
3270 if (start < 0) return NULL;
3271 assert(can_capture_store(st, phase, can_reshape) == start, "sanity");
3272
3273 Compile* C = phase->C;
3274 int size_in_bytes = st->memory_size();
3275 int i = captured_store_insertion_point(start, size_in_bytes, phase);
3276 if (i == 0) return NULL; // bail out
3277 Node* prev_mem = NULL; // raw memory for the captured store
3278 if (i > 0) {
3279 prev_mem = in(i); // there is a pre-existing store under this one
3280 set_req(i, C->top()); // temporarily disconnect it
3281 // See StoreNode::Ideal 'st->outcnt() == 1' for the reason to disconnect.
3282 } else {
3283 i = -i; // no pre-existing store
3284 prev_mem = zero_memory(); // a slice of the newly allocated object
3285 if (i > InitializeNode::RawStores && in(i-1) == prev_mem)
3286 set_req(--i, C->top()); // reuse this edge; it has been folded away
3287 else
3288 ins_req(i, C->top()); // build a new edge
3289 }
3290 Node* new_st = st->clone();
3291 new_st->set_req(MemNode::Control, in(Control));
3292 new_st->set_req(MemNode::Memory, prev_mem);
3293 new_st->set_req(MemNode::Address, make_raw_address(start, phase));
3294 new_st = phase->transform(new_st);
3295
3296 // At this point, new_st might have swallowed a pre-existing store
3297 // at the same offset, or perhaps new_st might have disappeared,
3298 // if it redundantly stored the same value (or zero to fresh memory).
3299
3300 // In any case, wire it in:
3301 phase->igvn_rehash_node_delayed(this);
3302 set_req(i, new_st);
3303
3304 // The caller may now kill the old guy.
3305 DEBUG_ONLY(Node* check_st = find_captured_store(start, size_in_bytes, phase));
3306 assert(check_st == new_st || check_st == NULL, "must be findable");
3307 assert(!is_complete(), "");
3308 return new_st;
3309 }
3310
3311 static bool store_constant(jlong* tiles, int num_tiles,
3312 intptr_t st_off, int st_size,
3313 jlong con) {
3314 if ((st_off & (st_size-1)) != 0)
3315 return false; // strange store offset (assume size==2**N)
3316 address addr = (address)tiles + st_off;
3317 assert(st_off >= 0 && addr+st_size <= (address)&tiles[num_tiles], "oob");
3318 switch (st_size) {
3319 case sizeof(jbyte): *(jbyte*) addr = (jbyte) con; break;
3320 case sizeof(jchar): *(jchar*) addr = (jchar) con; break;
3321 case sizeof(jint): *(jint*) addr = (jint) con; break;
3322 case sizeof(jlong): *(jlong*) addr = (jlong) con; break;
3323 default: return false; // strange store size (detect size!=2**N here)
3324 }
3325 return true; // return success to caller
3326 }
3327
3328 // Coalesce subword constants into int constants and possibly
3329 // into long constants. The goal, if the CPU permits,
3330 // is to initialize the object with a small number of 64-bit tiles.
3331 // Also, convert floating-point constants to bit patterns.
3332 // Non-constants are not relevant to this pass.
3333 //
3334 // In terms of the running example on InitializeNode::InitializeNode
3335 // and InitializeNode::capture_store, here is the transformation
3336 // of rawstore1 and rawstore2 into rawstore12:
3337 // alloc = (Allocate ...)
3338 // rawoop = alloc.RawAddress
3339 // tile12 = 0x00010002
3340 // rawstore12 = (StoreI alloc.Control alloc.Memory (+ rawoop 12) tile12)
3341 // init = (Initialize alloc.Control alloc.Memory rawoop rawstore12)
3342 //
3343 void
3344 InitializeNode::coalesce_subword_stores(intptr_t header_size,
3345 Node* size_in_bytes,
3346 PhaseGVN* phase) {
3347 Compile* C = phase->C;
3348
3349 assert(stores_are_sane(phase), "");
3350 // Note: After this pass, they are not completely sane,
3351 // since there may be some overlaps.
3352
3353 int old_subword = 0, old_long = 0, new_int = 0, new_long = 0;
3354
3355 intptr_t ti_limit = (TrackedInitializationLimit * HeapWordSize);
3356 intptr_t size_limit = phase->find_intptr_t_con(size_in_bytes, ti_limit);
3357 size_limit = MIN2(size_limit, ti_limit);
3358 size_limit = align_size_up(size_limit, BytesPerLong);
3359 int num_tiles = size_limit / BytesPerLong;
3360
3361 // allocate space for the tile map:
3362 const int small_len = DEBUG_ONLY(true ? 3 :) 30; // keep stack frames small
3363 jlong tiles_buf[small_len];
3364 Node* nodes_buf[small_len];
3365 jlong inits_buf[small_len];
3366 jlong* tiles = ((num_tiles <= small_len) ? &tiles_buf[0]
3367 : NEW_RESOURCE_ARRAY(jlong, num_tiles));
3368 Node** nodes = ((num_tiles <= small_len) ? &nodes_buf[0]
3369 : NEW_RESOURCE_ARRAY(Node*, num_tiles));
3370 jlong* inits = ((num_tiles <= small_len) ? &inits_buf[0]
3371 : NEW_RESOURCE_ARRAY(jlong, num_tiles));
3372 // tiles: exact bitwise model of all primitive constants
3373 // nodes: last constant-storing node subsumed into the tiles model
3374 // inits: which bytes (in each tile) are touched by any initializations
3375
3376 //// Pass A: Fill in the tile model with any relevant stores.
3377
3378 Copy::zero_to_bytes(tiles, sizeof(tiles[0]) * num_tiles);
3379 Copy::zero_to_bytes(nodes, sizeof(nodes[0]) * num_tiles);
3380 Copy::zero_to_bytes(inits, sizeof(inits[0]) * num_tiles);
3381 Node* zmem = zero_memory(); // initially zero memory state
3382 for (uint i = InitializeNode::RawStores, limit = req(); i < limit; i++) {
3383 Node* st = in(i);
3384 intptr_t st_off = get_store_offset(st, phase);
3385
3386 // Figure out the store's offset and constant value:
3387 if (st_off < header_size) continue; //skip (ignore header)
3388 if (st->in(MemNode::Memory) != zmem) continue; //skip (odd store chain)
3389 int st_size = st->as_Store()->memory_size();
3390 if (st_off + st_size > size_limit) break;
3391
3392 // Record which bytes are touched, whether by constant or not.
3393 if (!store_constant(inits, num_tiles, st_off, st_size, (jlong) -1))
3394 continue; // skip (strange store size)
3395
3396 const Type* val = phase->type(st->in(MemNode::ValueIn));
3397 if (!val->singleton()) continue; //skip (non-con store)
3398 BasicType type = val->basic_type();
3399
3400 jlong con = 0;
3401 switch (type) {
3402 case T_INT: con = val->is_int()->get_con(); break;
3403 case T_LONG: con = val->is_long()->get_con(); break;
3404 case T_FLOAT: con = jint_cast(val->getf()); break;
3405 case T_DOUBLE: con = jlong_cast(val->getd()); break;
3406 default: continue; //skip (odd store type)
3407 }
3408
3409 if (type == T_LONG && Matcher::isSimpleConstant64(con) &&
3410 st->Opcode() == Op_StoreL) {
3411 continue; // This StoreL is already optimal.
3412 }
3413
3414 // Store down the constant.
3415 store_constant(tiles, num_tiles, st_off, st_size, con);
3416
3417 intptr_t j = st_off >> LogBytesPerLong;
3418
3419 if (type == T_INT && st_size == BytesPerInt
3420 && (st_off & BytesPerInt) == BytesPerInt) {
3421 jlong lcon = tiles[j];
3422 if (!Matcher::isSimpleConstant64(lcon) &&
3423 st->Opcode() == Op_StoreI) {
3424 // This StoreI is already optimal by itself.
3425 jint* intcon = (jint*) &tiles[j];
3426 intcon[1] = 0; // undo the store_constant()
3427
3428 // If the previous store is also optimal by itself, back up and
3429 // undo the action of the previous loop iteration... if we can.
3430 // But if we can't, just let the previous half take care of itself.
3431 st = nodes[j];
3432 st_off -= BytesPerInt;
3433 con = intcon[0];
3434 if (con != 0 && st != NULL && st->Opcode() == Op_StoreI) {
3435 assert(st_off >= header_size, "still ignoring header");
3436 assert(get_store_offset(st, phase) == st_off, "must be");
3437 assert(in(i-1) == zmem, "must be");
3438 DEBUG_ONLY(const Type* tcon = phase->type(st->in(MemNode::ValueIn)));
3439 assert(con == tcon->is_int()->get_con(), "must be");
3440 // Undo the effects of the previous loop trip, which swallowed st:
3441 intcon[0] = 0; // undo store_constant()
3442 set_req(i-1, st); // undo set_req(i, zmem)
3443 nodes[j] = NULL; // undo nodes[j] = st
3444 --old_subword; // undo ++old_subword
3445 }
3446 continue; // This StoreI is already optimal.
3447 }
3448 }
3449
3450 // This store is not needed.
3451 set_req(i, zmem);
3452 nodes[j] = st; // record for the moment
3453 if (st_size < BytesPerLong) // something has changed
3454 ++old_subword; // includes int/float, but who's counting...
3455 else ++old_long;
3456 }
3457
3458 if ((old_subword + old_long) == 0)
3459 return; // nothing more to do
3460
3461 //// Pass B: Convert any non-zero tiles into optimal constant stores.
3462 // Be sure to insert them before overlapping non-constant stores.
3463 // (E.g., byte[] x = { 1,2,y,4 } => x[int 0] = 0x01020004, x[2]=y.)
3464 for (int j = 0; j < num_tiles; j++) {
3465 jlong con = tiles[j];
3466 jlong init = inits[j];
3467 if (con == 0) continue;
3468 jint con0, con1; // split the constant, address-wise
3469 jint init0, init1; // split the init map, address-wise
3470 { union { jlong con; jint intcon[2]; } u;
3471 u.con = con;
3472 con0 = u.intcon[0];
3473 con1 = u.intcon[1];
3474 u.con = init;
3475 init0 = u.intcon[0];
3476 init1 = u.intcon[1];
3477 }
3478
3479 Node* old = nodes[j];
3480 assert(old != NULL, "need the prior store");
3481 intptr_t offset = (j * BytesPerLong);
3482
3483 bool split = !Matcher::isSimpleConstant64(con);
3484
3485 if (offset < header_size) {
3486 assert(offset + BytesPerInt >= header_size, "second int counts");
3487 assert(*(jint*)&tiles[j] == 0, "junk in header");
3488 split = true; // only the second word counts
3489 // Example: int a[] = { 42 ... }
3490 } else if (con0 == 0 && init0 == -1) {
3491 split = true; // first word is covered by full inits
3492 // Example: int a[] = { ... foo(), 42 ... }
3493 } else if (con1 == 0 && init1 == -1) {
3494 split = true; // second word is covered by full inits
3495 // Example: int a[] = { ... 42, foo() ... }
3496 }
3497
3498 // Here's a case where init0 is neither 0 nor -1:
3499 // byte a[] = { ... 0,0,foo(),0, 0,0,0,42 ... }
3500 // Assuming big-endian memory, init0, init1 are 0x0000FF00, 0x000000FF.
3501 // In this case the tile is not split; it is (jlong)42.
3502 // The big tile is stored down, and then the foo() value is inserted.
3503 // (If there were foo(),foo() instead of foo(),0, init0 would be -1.)
3504
3505 Node* ctl = old->in(MemNode::Control);
3506 Node* adr = make_raw_address(offset, phase);
3507 const TypePtr* atp = TypeRawPtr::BOTTOM;
3508
3509 // One or two coalesced stores to plop down.
3510 Node* st[2];
3511 intptr_t off[2];
3512 int nst = 0;
3513 if (!split) {
3514 ++new_long;
3515 off[nst] = offset;
3516 st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
3517 phase->longcon(con), T_LONG, MemNode::unordered);
3518 } else {
3519 // Omit either if it is a zero.
3520 if (con0 != 0) {
3521 ++new_int;
3522 off[nst] = offset;
3523 st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
3524 phase->intcon(con0), T_INT, MemNode::unordered);
3525 }
3526 if (con1 != 0) {
3527 ++new_int;
3528 offset += BytesPerInt;
3529 adr = make_raw_address(offset, phase);
3530 off[nst] = offset;
3531 st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
3532 phase->intcon(con1), T_INT, MemNode::unordered);
3533 }
3534 }
3535
3536 // Insert second store first, then the first before the second.
3537 // Insert each one just before any overlapping non-constant stores.
3538 while (nst > 0) {
3539 Node* st1 = st[--nst];
3540 C->copy_node_notes_to(st1, old);
3541 st1 = phase->transform(st1);
3542 offset = off[nst];
3543 assert(offset >= header_size, "do not smash header");
3544 int ins_idx = captured_store_insertion_point(offset, /*size:*/0, phase);
3545 guarantee(ins_idx != 0, "must re-insert constant store");
3546 if (ins_idx < 0) ins_idx = -ins_idx; // never overlap
3547 if (ins_idx > InitializeNode::RawStores && in(ins_idx-1) == zmem)
3548 set_req(--ins_idx, st1);
3549 else
3550 ins_req(ins_idx, st1);
3551 }
3552 }
3553
3554 if (PrintCompilation && WizardMode)
3555 tty->print_cr("Changed %d/%d subword/long constants into %d/%d int/long",
3556 old_subword, old_long, new_int, new_long);
3557 if (C->log() != NULL)
3558 C->log()->elem("comment that='%d/%d subword/long to %d/%d int/long'",
3559 old_subword, old_long, new_int, new_long);
3560
3561 // Clean up any remaining occurrences of zmem:
3562 remove_extra_zeroes();
3563 }
3564
3565 // Explore forward from in(start) to find the first fully initialized
3566 // word, and return its offset. Skip groups of subword stores which
3567 // together initialize full words. If in(start) is itself part of a
3568 // fully initialized word, return the offset of in(start). If there
3569 // are no following full-word stores, or if something is fishy, return
3570 // a negative value.
3571 intptr_t InitializeNode::find_next_fullword_store(uint start, PhaseGVN* phase) {
3572 int int_map = 0;
3573 intptr_t int_map_off = 0;
3574 const int FULL_MAP = right_n_bits(BytesPerInt); // the int_map we hope for
3575
3576 for (uint i = start, limit = req(); i < limit; i++) {
3577 Node* st = in(i);
3578
3579 intptr_t st_off = get_store_offset(st, phase);
3580 if (st_off < 0) break; // return conservative answer
3581
3582 int st_size = st->as_Store()->memory_size();
3583 if (st_size >= BytesPerInt && (st_off % BytesPerInt) == 0) {
3584 return st_off; // we found a complete word init
3585 }
3586
3587 // update the map:
3588
3589 intptr_t this_int_off = align_size_down(st_off, BytesPerInt);
3590 if (this_int_off != int_map_off) {
3591 // reset the map:
3592 int_map = 0;
3593 int_map_off = this_int_off;
3594 }
3595
3596 int subword_off = st_off - this_int_off;
3597 int_map |= right_n_bits(st_size) << subword_off;
3598 if ((int_map & FULL_MAP) == FULL_MAP) {
3599 return this_int_off; // we found a complete word init
3600 }
3601
3602 // Did this store hit or cross the word boundary?
3603 intptr_t next_int_off = align_size_down(st_off + st_size, BytesPerInt);
3604 if (next_int_off == this_int_off + BytesPerInt) {
3605 // We passed the current int, without fully initializing it.
3606 int_map_off = next_int_off;
3607 int_map >>= BytesPerInt;
3608 } else if (next_int_off > this_int_off + BytesPerInt) {
3609 // We passed the current and next int.
3610 return this_int_off + BytesPerInt;
3611 }
3612 }
3613
3614 return -1;
3615 }
3616
3617
3618 // Called when the associated AllocateNode is expanded into CFG.
3619 // At this point, we may perform additional optimizations.
3620 // Linearize the stores by ascending offset, to make memory
3621 // activity as coherent as possible.
3622 Node* InitializeNode::complete_stores(Node* rawctl, Node* rawmem, Node* rawptr,
3623 intptr_t header_size,
3624 Node* size_in_bytes,
3625 PhaseGVN* phase) {
3626 assert(!is_complete(), "not already complete");
3627 assert(stores_are_sane(phase), "");
3628 assert(allocation() != NULL, "must be present");
3629
3630 remove_extra_zeroes();
3631
3632 if (ReduceFieldZeroing || ReduceBulkZeroing)
3633 // reduce instruction count for common initialization patterns
3634 coalesce_subword_stores(header_size, size_in_bytes, phase);
3635
3636 Node* zmem = zero_memory(); // initially zero memory state
3637 Node* inits = zmem; // accumulating a linearized chain of inits
3638 #ifdef ASSERT
3639 intptr_t first_offset = allocation()->minimum_header_size();
3640 intptr_t last_init_off = first_offset; // previous init offset
3641 intptr_t last_init_end = first_offset; // previous init offset+size
3642 intptr_t last_tile_end = first_offset; // previous tile offset+size
3643 #endif
3644 intptr_t zeroes_done = header_size;
3645
3646 bool do_zeroing = true; // we might give up if inits are very sparse
3647 int big_init_gaps = 0; // how many large gaps have we seen?
3648
3649 if (ZeroTLAB) do_zeroing = false;
3650 if (!ReduceFieldZeroing && !ReduceBulkZeroing) do_zeroing = false;
3651
3652 for (uint i = InitializeNode::RawStores, limit = req(); i < limit; i++) {
3653 Node* st = in(i);
3654 intptr_t st_off = get_store_offset(st, phase);
3655 if (st_off < 0)
3656 break; // unknown junk in the inits
3657 if (st->in(MemNode::Memory) != zmem)
3658 break; // complicated store chains somehow in list
3659
3660 int st_size = st->as_Store()->memory_size();
3661 intptr_t next_init_off = st_off + st_size;
3662
3663 if (do_zeroing && zeroes_done < next_init_off) {
3664 // See if this store needs a zero before it or under it.
3665 intptr_t zeroes_needed = st_off;
3666
3667 if (st_size < BytesPerInt) {
3668 // Look for subword stores which only partially initialize words.
3669 // If we find some, we must lay down some word-level zeroes first,
3670 // underneath the subword stores.
3671 //
3672 // Examples:
3673 // byte[] a = { p,q,r,s } => a[0]=p,a[1]=q,a[2]=r,a[3]=s
3674 // byte[] a = { x,y,0,0 } => a[0..3] = 0, a[0]=x,a[1]=y
3675 // byte[] a = { 0,0,z,0 } => a[0..3] = 0, a[2]=z
3676 //
3677 // Note: coalesce_subword_stores may have already done this,
3678 // if it was prompted by constant non-zero subword initializers.
3679 // But this case can still arise with non-constant stores.
3680
3681 intptr_t next_full_store = find_next_fullword_store(i, phase);
3682
3683 // In the examples above:
3684 // in(i) p q r s x y z
3685 // st_off 12 13 14 15 12 13 14
3686 // st_size 1 1 1 1 1 1 1
3687 // next_full_s. 12 16 16 16 16 16 16
3688 // z's_done 12 16 16 16 12 16 12
3689 // z's_needed 12 16 16 16 16 16 16
3690 // zsize 0 0 0 0 4 0 4
3691 if (next_full_store < 0) {
3692 // Conservative tack: Zero to end of current word.
3693 zeroes_needed = align_size_up(zeroes_needed, BytesPerInt);
3694 } else {
3695 // Zero to beginning of next fully initialized word.
3696 // Or, don't zero at all, if we are already in that word.
3697 assert(next_full_store >= zeroes_needed, "must go forward");
3698 assert((next_full_store & (BytesPerInt-1)) == 0, "even boundary");
3699 zeroes_needed = next_full_store;
3700 }
3701 }
3702
3703 if (zeroes_needed > zeroes_done) {
3704 intptr_t zsize = zeroes_needed - zeroes_done;
3705 // Do some incremental zeroing on rawmem, in parallel with inits.
3706 zeroes_done = align_size_down(zeroes_done, BytesPerInt);
3707 rawmem = ClearArrayNode::clear_memory(rawctl, rawmem, rawptr,
3708 zeroes_done, zeroes_needed,
3709 phase);
3710 zeroes_done = zeroes_needed;
3711 if (zsize > Matcher::init_array_short_size && ++big_init_gaps > 2)
3712 do_zeroing = false; // leave the hole, next time
3713 }
3714 }
3715
3716 // Collect the store and move on:
3717 st->set_req(MemNode::Memory, inits);
3718 inits = st; // put it on the linearized chain
3719 set_req(i, zmem); // unhook from previous position
3720
3721 if (zeroes_done == st_off)
3722 zeroes_done = next_init_off;
3723
3724 assert(!do_zeroing || zeroes_done >= next_init_off, "don't miss any");
3725
3726 #ifdef ASSERT
3727 // Various order invariants. Weaker than stores_are_sane because
3728 // a large constant tile can be filled in by smaller non-constant stores.
3729 assert(st_off >= last_init_off, "inits do not reverse");
3730 last_init_off = st_off;
3731 const Type* val = NULL;
3732 if (st_size >= BytesPerInt &&
3733 (val = phase->type(st->in(MemNode::ValueIn)))->singleton() &&
3734 (int)val->basic_type() < (int)T_OBJECT) {
3735 assert(st_off >= last_tile_end, "tiles do not overlap");
3736 assert(st_off >= last_init_end, "tiles do not overwrite inits");
3737 last_tile_end = MAX2(last_tile_end, next_init_off);
3738 } else {
3739 intptr_t st_tile_end = align_size_up(next_init_off, BytesPerLong);
3740 assert(st_tile_end >= last_tile_end, "inits stay with tiles");
3741 assert(st_off >= last_init_end, "inits do not overlap");
3742 last_init_end = next_init_off; // it's a non-tile
3743 }
3744 #endif //ASSERT
3745 }
3746
3747 remove_extra_zeroes(); // clear out all the zmems left over
3748 add_req(inits);
3749
3750 if (!ZeroTLAB) {
3751 // If anything remains to be zeroed, zero it all now.
3752 zeroes_done = align_size_down(zeroes_done, BytesPerInt);
3753 // if it is the last unused 4 bytes of an instance, forget about it
3754 intptr_t size_limit = phase->find_intptr_t_con(size_in_bytes, max_jint);
3755 if (zeroes_done + BytesPerLong >= size_limit) {
3756 assert(allocation() != NULL, "");
3757 if (allocation()->Opcode() == Op_Allocate) {
3758 Node* klass_node = allocation()->in(AllocateNode::KlassNode);
3759 ciKlass* k = phase->type(klass_node)->is_klassptr()->klass();
3760 if (zeroes_done == k->layout_helper())
3761 zeroes_done = size_limit;
3762 }
3763 }
3764 if (zeroes_done < size_limit) {
3765 rawmem = ClearArrayNode::clear_memory(rawctl, rawmem, rawptr,
3766 zeroes_done, size_in_bytes, phase);
3767 }
3768 }
3769
3770 set_complete(phase);
3771 return rawmem;
3772 }
3773
3774
3775 #ifdef ASSERT
3776 bool InitializeNode::stores_are_sane(PhaseTransform* phase) {
3777 if (is_complete())
3778 return true; // stores could be anything at this point
3779 assert(allocation() != NULL, "must be present");
3780 intptr_t last_off = allocation()->minimum_header_size();
3781 for (uint i = InitializeNode::RawStores; i < req(); i++) {
3782 Node* st = in(i);
3783 intptr_t st_off = get_store_offset(st, phase);
3784 if (st_off < 0) continue; // ignore dead garbage
3785 if (last_off > st_off) {
3786 tty->print_cr("*** bad store offset at %d: " INTX_FORMAT " > " INTX_FORMAT, i, last_off, st_off);
3787 this->dump(2);
3788 assert(false, "ascending store offsets");
3789 return false;
3790 }
3791 last_off = st_off + st->as_Store()->memory_size();
3792 }
3793 return true;
3794 }
3795 #endif //ASSERT
3796
3797
3798
3799
3800 //============================MergeMemNode=====================================
3801 //
3802 // SEMANTICS OF MEMORY MERGES: A MergeMem is a memory state assembled from several
3803 // contributing store or call operations. Each contributor provides the memory
3804 // state for a particular "alias type" (see Compile::alias_type). For example,
3805 // if a MergeMem has an input X for alias category #6, then any memory reference
3806 // to alias category #6 may use X as its memory state input, as an exact equivalent
3807 // to using the MergeMem as a whole.
3808 // Load<6>( MergeMem(<6>: X, ...), p ) <==> Load<6>(X,p)
3809 //
3810 // (Here, the <N> notation gives the index of the relevant adr_type.)
3811 //
3812 // In one special case (and more cases in the future), alias categories overlap.
3813 // The special alias category "Bot" (Compile::AliasIdxBot) includes all memory
3814 // states. Therefore, if a MergeMem has only one contributing input W for Bot,
3815 // it is exactly equivalent to that state W:
3816 // MergeMem(<Bot>: W) <==> W
3817 //
3818 // Usually, the merge has more than one input. In that case, where inputs
3819 // overlap (i.e., one is Bot), the narrower alias type determines the memory
3820 // state for that type, and the wider alias type (Bot) fills in everywhere else:
3821 // Load<5>( MergeMem(<Bot>: W, <6>: X), p ) <==> Load<5>(W,p)
3822 // Load<6>( MergeMem(<Bot>: W, <6>: X), p ) <==> Load<6>(X,p)
3823 //
3824 // A merge can take a "wide" memory state as one of its narrow inputs.
3825 // This simply means that the merge observes out only the relevant parts of
3826 // the wide input. That is, wide memory states arriving at narrow merge inputs
3827 // are implicitly "filtered" or "sliced" as necessary. (This is rare.)
3828 //
3829 // These rules imply that MergeMem nodes may cascade (via their <Bot> links),
3830 // and that memory slices "leak through":
3831 // MergeMem(<Bot>: MergeMem(<Bot>: W, <7>: Y)) <==> MergeMem(<Bot>: W, <7>: Y)
3832 //
3833 // But, in such a cascade, repeated memory slices can "block the leak":
3834 // MergeMem(<Bot>: MergeMem(<Bot>: W, <7>: Y), <7>: Y') <==> MergeMem(<Bot>: W, <7>: Y')
3835 //
3836 // In the last example, Y is not part of the combined memory state of the
3837 // outermost MergeMem. The system must, of course, prevent unschedulable
3838 // memory states from arising, so you can be sure that the state Y is somehow
3839 // a precursor to state Y'.
3840 //
3841 //
3842 // REPRESENTATION OF MEMORY MERGES: The indexes used to address the Node::in array
3843 // of each MergeMemNode array are exactly the numerical alias indexes, including
3844 // but not limited to AliasIdxTop, AliasIdxBot, and AliasIdxRaw. The functions
3845 // Compile::alias_type (and kin) produce and manage these indexes.
3846 //
3847 // By convention, the value of in(AliasIdxTop) (i.e., in(1)) is always the top node.
3848 // (Note that this provides quick access to the top node inside MergeMem methods,
3849 // without the need to reach out via TLS to Compile::current.)
3850 //
3851 // As a consequence of what was just described, a MergeMem that represents a full
3852 // memory state has an edge in(AliasIdxBot) which is a "wide" memory state,
3853 // containing all alias categories.
3854 //
3855 // MergeMem nodes never (?) have control inputs, so in(0) is NULL.
3856 //
3857 // All other edges in(N) (including in(AliasIdxRaw), which is in(3)) are either
3858 // a memory state for the alias type <N>, or else the top node, meaning that
3859 // there is no particular input for that alias type. Note that the length of
3860 // a MergeMem is variable, and may be extended at any time to accommodate new
3861 // memory states at larger alias indexes. When merges grow, they are of course
3862 // filled with "top" in the unused in() positions.
3863 //
3864 // This use of top is named "empty_memory()", or "empty_mem" (no-memory) as a variable.
3865 // (Top was chosen because it works smoothly with passes like GCM.)
3866 //
3867 // For convenience, we hardwire the alias index for TypeRawPtr::BOTTOM. (It is
3868 // the type of random VM bits like TLS references.) Since it is always the
3869 // first non-Bot memory slice, some low-level loops use it to initialize an
3870 // index variable: for (i = AliasIdxRaw; i < req(); i++).
3871 //
3872 //
3873 // ACCESSORS: There is a special accessor MergeMemNode::base_memory which returns
3874 // the distinguished "wide" state. The accessor MergeMemNode::memory_at(N) returns
3875 // the memory state for alias type <N>, or (if there is no particular slice at <N>,
3876 // it returns the base memory. To prevent bugs, memory_at does not accept <Top>
3877 // or <Bot> indexes. The iterator MergeMemStream provides robust iteration over
3878 // MergeMem nodes or pairs of such nodes, ensuring that the non-top edges are visited.
3879 //
3880 // %%%% We may get rid of base_memory as a separate accessor at some point; it isn't
3881 // really that different from the other memory inputs. An abbreviation called
3882 // "bot_memory()" for "memory_at(AliasIdxBot)" would keep code tidy.
3883 //
3884 //
3885 // PARTIAL MEMORY STATES: During optimization, MergeMem nodes may arise that represent
3886 // partial memory states. When a Phi splits through a MergeMem, the copy of the Phi
3887 // that "emerges though" the base memory will be marked as excluding the alias types
3888 // of the other (narrow-memory) copies which "emerged through" the narrow edges:
3889 //
3890 // Phi<Bot>(U, MergeMem(<Bot>: W, <8>: Y))
3891 // ==Ideal=> MergeMem(<Bot>: Phi<Bot-8>(U, W), Phi<8>(U, Y))
3892 //
3893 // This strange "subtraction" effect is necessary to ensure IGVN convergence.
3894 // (It is currently unimplemented.) As you can see, the resulting merge is
3895 // actually a disjoint union of memory states, rather than an overlay.
3896 //
3897
3898 //------------------------------MergeMemNode-----------------------------------
3899 Node* MergeMemNode::make_empty_memory() {
3900 Node* empty_memory = (Node*) Compile::current()->top();
3901 assert(empty_memory->is_top(), "correct sentinel identity");
3902 return empty_memory;
3903 }
3904
3905 MergeMemNode::MergeMemNode(Node *new_base) : Node(1+Compile::AliasIdxRaw) {
3906 init_class_id(Class_MergeMem);
3907 // all inputs are nullified in Node::Node(int)
3908 // set_input(0, NULL); // no control input
3909
3910 // Initialize the edges uniformly to top, for starters.
3911 Node* empty_mem = make_empty_memory();
3912 for (uint i = Compile::AliasIdxTop; i < req(); i++) {
3913 init_req(i,empty_mem);
3914 }
3915 assert(empty_memory() == empty_mem, "");
3916
3917 if( new_base != NULL && new_base->is_MergeMem() ) {
3918 MergeMemNode* mdef = new_base->as_MergeMem();
3919 assert(mdef->empty_memory() == empty_mem, "consistent sentinels");
3920 for (MergeMemStream mms(this, mdef); mms.next_non_empty2(); ) {
3921 mms.set_memory(mms.memory2());
3922 }
3923 assert(base_memory() == mdef->base_memory(), "");
3924 } else {
3925 set_base_memory(new_base);
3926 }
3927 }
3928
3929 // Make a new, untransformed MergeMem with the same base as 'mem'.
3930 // If mem is itself a MergeMem, populate the result with the same edges.
3931 MergeMemNode* MergeMemNode::make(Node* mem) {
3932 return new MergeMemNode(mem);
3933 }
3934
3935 //------------------------------cmp--------------------------------------------
3936 uint MergeMemNode::hash() const { return NO_HASH; }
3937 uint MergeMemNode::cmp( const Node &n ) const {
3938 return (&n == this); // Always fail except on self
3939 }
3940
3941 //------------------------------Identity---------------------------------------
3942 Node* MergeMemNode::Identity(PhaseTransform *phase) {
3943 // Identity if this merge point does not record any interesting memory
3944 // disambiguations.
3945 Node* base_mem = base_memory();
3946 Node* empty_mem = empty_memory();
3947 if (base_mem != empty_mem) { // Memory path is not dead?
3948 for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
3949 Node* mem = in(i);
3950 if (mem != empty_mem && mem != base_mem) {
3951 return this; // Many memory splits; no change
3952 }
3953 }
3954 }
3955 return base_mem; // No memory splits; ID on the one true input
3956 }
3957
3958 //------------------------------Ideal------------------------------------------
3959 // This method is invoked recursively on chains of MergeMem nodes
3960 Node *MergeMemNode::Ideal(PhaseGVN *phase, bool can_reshape) {
3961 // Remove chain'd MergeMems
3962 //
3963 // This is delicate, because the each "in(i)" (i >= Raw) is interpreted
3964 // relative to the "in(Bot)". Since we are patching both at the same time,
3965 // we have to be careful to read each "in(i)" relative to the old "in(Bot)",
3966 // but rewrite each "in(i)" relative to the new "in(Bot)".
3967 Node *progress = NULL;
3968
3969
3970 Node* old_base = base_memory();
3971 Node* empty_mem = empty_memory();
3972 if (old_base == empty_mem)
3973 return NULL; // Dead memory path.
3974
3975 MergeMemNode* old_mbase;
3976 if (old_base != NULL && old_base->is_MergeMem())
3977 old_mbase = old_base->as_MergeMem();
3978 else
3979 old_mbase = NULL;
3980 Node* new_base = old_base;
3981
3982 // simplify stacked MergeMems in base memory
3983 if (old_mbase) new_base = old_mbase->base_memory();
3984
3985 // the base memory might contribute new slices beyond my req()
3986 if (old_mbase) grow_to_match(old_mbase);
3987
3988 // Look carefully at the base node if it is a phi.
3989 PhiNode* phi_base;
3990 if (new_base != NULL && new_base->is_Phi())
3991 phi_base = new_base->as_Phi();
3992 else
3993 phi_base = NULL;
3994
3995 Node* phi_reg = NULL;
3996 uint phi_len = (uint)-1;
3997 if (phi_base != NULL && !phi_base->is_copy()) {
3998 // do not examine phi if degraded to a copy
3999 phi_reg = phi_base->region();
4000 phi_len = phi_base->req();
4001 // see if the phi is unfinished
4002 for (uint i = 1; i < phi_len; i++) {
4003 if (phi_base->in(i) == NULL) {
4004 // incomplete phi; do not look at it yet!
4005 phi_reg = NULL;
4006 phi_len = (uint)-1;
4007 break;
4008 }
4009 }
4010 }
4011
4012 // Note: We do not call verify_sparse on entry, because inputs
4013 // can normalize to the base_memory via subsume_node or similar
4014 // mechanisms. This method repairs that damage.
4015
4016 assert(!old_mbase || old_mbase->is_empty_memory(empty_mem), "consistent sentinels");
4017
4018 // Look at each slice.
4019 for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
4020 Node* old_in = in(i);
4021 // calculate the old memory value
4022 Node* old_mem = old_in;
4023 if (old_mem == empty_mem) old_mem = old_base;
4024 assert(old_mem == memory_at(i), "");
4025
4026 // maybe update (reslice) the old memory value
4027
4028 // simplify stacked MergeMems
4029 Node* new_mem = old_mem;
4030 MergeMemNode* old_mmem;
4031 if (old_mem != NULL && old_mem->is_MergeMem())
4032 old_mmem = old_mem->as_MergeMem();
4033 else
4034 old_mmem = NULL;
4035 if (old_mmem == this) {
4036 // This can happen if loops break up and safepoints disappear.
4037 // A merge of BotPtr (default) with a RawPtr memory derived from a
4038 // safepoint can be rewritten to a merge of the same BotPtr with
4039 // the BotPtr phi coming into the loop. If that phi disappears
4040 // also, we can end up with a self-loop of the mergemem.
4041 // In general, if loops degenerate and memory effects disappear,
4042 // a mergemem can be left looking at itself. This simply means
4043 // that the mergemem's default should be used, since there is
4044 // no longer any apparent effect on this slice.
4045 // Note: If a memory slice is a MergeMem cycle, it is unreachable
4046 // from start. Update the input to TOP.
4047 new_mem = (new_base == this || new_base == empty_mem)? empty_mem : new_base;
4048 }
4049 else if (old_mmem != NULL) {
4050 new_mem = old_mmem->memory_at(i);
4051 }
4052 // else preceding memory was not a MergeMem
4053
4054 // replace equivalent phis (unfortunately, they do not GVN together)
4055 if (new_mem != NULL && new_mem != new_base &&
4056 new_mem->req() == phi_len && new_mem->in(0) == phi_reg) {
4057 if (new_mem->is_Phi()) {
4058 PhiNode* phi_mem = new_mem->as_Phi();
4059 for (uint i = 1; i < phi_len; i++) {
4060 if (phi_base->in(i) != phi_mem->in(i)) {
4061 phi_mem = NULL;
4062 break;
4063 }
4064 }
4065 if (phi_mem != NULL) {
4066 // equivalent phi nodes; revert to the def
4067 new_mem = new_base;
4068 }
4069 }
4070 }
4071
4072 // maybe store down a new value
4073 Node* new_in = new_mem;
4074 if (new_in == new_base) new_in = empty_mem;
4075
4076 if (new_in != old_in) {
4077 // Warning: Do not combine this "if" with the previous "if"
4078 // A memory slice might have be be rewritten even if it is semantically
4079 // unchanged, if the base_memory value has changed.
4080 set_req(i, new_in);
4081 progress = this; // Report progress
4082 }
4083 }
4084
4085 if (new_base != old_base) {
4086 set_req(Compile::AliasIdxBot, new_base);
4087 // Don't use set_base_memory(new_base), because we need to update du.
4088 assert(base_memory() == new_base, "");
4089 progress = this;
4090 }
4091
4092 if( base_memory() == this ) {
4093 // a self cycle indicates this memory path is dead
4094 set_req(Compile::AliasIdxBot, empty_mem);
4095 }
4096
4097 // Resolve external cycles by calling Ideal on a MergeMem base_memory
4098 // Recursion must occur after the self cycle check above
4099 if( base_memory()->is_MergeMem() ) {
4100 MergeMemNode *new_mbase = base_memory()->as_MergeMem();
4101 Node *m = phase->transform(new_mbase); // Rollup any cycles
4102 if( m != NULL && (m->is_top() ||
4103 m->is_MergeMem() && m->as_MergeMem()->base_memory() == empty_mem) ) {
4104 // propagate rollup of dead cycle to self
4105 set_req(Compile::AliasIdxBot, empty_mem);
4106 }
4107 }
4108
4109 if( base_memory() == empty_mem ) {
4110 progress = this;
4111 // Cut inputs during Parse phase only.
4112 // During Optimize phase a dead MergeMem node will be subsumed by Top.
4113 if( !can_reshape ) {
4114 for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
4115 if( in(i) != empty_mem ) { set_req(i, empty_mem); }
4116 }
4117 }
4118 }
4119
4120 if( !progress && base_memory()->is_Phi() && can_reshape ) {
4121 // Check if PhiNode::Ideal's "Split phis through memory merges"
4122 // transform should be attempted. Look for this->phi->this cycle.
4123 uint merge_width = req();
4124 if (merge_width > Compile::AliasIdxRaw) {
4125 PhiNode* phi = base_memory()->as_Phi();
4126 for( uint i = 1; i < phi->req(); ++i ) {// For all paths in
4127 if (phi->in(i) == this) {
4128 phase->is_IterGVN()->_worklist.push(phi);
4129 break;
4130 }
4131 }
4132 }
4133 }
4134
4135 assert(progress || verify_sparse(), "please, no dups of base");
4136 return progress;
4137 }
4138
4139 //-------------------------set_base_memory-------------------------------------
4140 void MergeMemNode::set_base_memory(Node *new_base) {
4141 Node* empty_mem = empty_memory();
4142 set_req(Compile::AliasIdxBot, new_base);
4143 assert(memory_at(req()) == new_base, "must set default memory");
4144 // Clear out other occurrences of new_base:
4145 if (new_base != empty_mem) {
4146 for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
4147 if (in(i) == new_base) set_req(i, empty_mem);
4148 }
4149 }
4150 }
4151
4152 //------------------------------out_RegMask------------------------------------
4153 const RegMask &MergeMemNode::out_RegMask() const {
4154 return RegMask::Empty;
4155 }
4156
4157 //------------------------------dump_spec--------------------------------------
4158 #ifndef PRODUCT
4159 void MergeMemNode::dump_spec(outputStream *st) const {
4160 st->print(" {");
4161 Node* base_mem = base_memory();
4162 for( uint i = Compile::AliasIdxRaw; i < req(); i++ ) {
4163 Node* mem = (in(i) != NULL) ? memory_at(i) : base_mem;
4164 if (mem == base_mem) { st->print(" -"); continue; }
4165 st->print( " N%d:", mem->_idx );
4166 Compile::current()->get_adr_type(i)->dump_on(st);
4167 }
4168 st->print(" }");
4169 }
4170 #endif // !PRODUCT
4171
4172
4173 #ifdef ASSERT
4174 static bool might_be_same(Node* a, Node* b) {
4175 if (a == b) return true;
4176 if (!(a->is_Phi() || b->is_Phi())) return false;
4177 // phis shift around during optimization
4178 return true; // pretty stupid...
4179 }
4180
4181 // verify a narrow slice (either incoming or outgoing)
4182 static void verify_memory_slice(const MergeMemNode* m, int alias_idx, Node* n) {
4183 if (!VerifyAliases) return; // don't bother to verify unless requested
4184 if (is_error_reported()) return; // muzzle asserts when debugging an error
4185 if (Node::in_dump()) return; // muzzle asserts when printing
4186 assert(alias_idx >= Compile::AliasIdxRaw, "must not disturb base_memory or sentinel");
4187 assert(n != NULL, "");
4188 // Elide intervening MergeMem's
4189 while (n->is_MergeMem()) {
4190 n = n->as_MergeMem()->memory_at(alias_idx);
4191 }
4192 Compile* C = Compile::current();
4193 const TypePtr* n_adr_type = n->adr_type();
4194 if (n == m->empty_memory()) {
4195 // Implicit copy of base_memory()
4196 } else if (n_adr_type != TypePtr::BOTTOM) {
4197 assert(n_adr_type != NULL, "new memory must have a well-defined adr_type");
4198 assert(C->must_alias(n_adr_type, alias_idx), "new memory must match selected slice");
4199 } else {
4200 // A few places like make_runtime_call "know" that VM calls are narrow,
4201 // and can be used to update only the VM bits stored as TypeRawPtr::BOTTOM.
4202 bool expected_wide_mem = false;
4203 if (n == m->base_memory()) {
4204 expected_wide_mem = true;
4205 } else if (alias_idx == Compile::AliasIdxRaw ||
4206 n == m->memory_at(Compile::AliasIdxRaw)) {
4207 expected_wide_mem = true;
4208 } else if (!C->alias_type(alias_idx)->is_rewritable()) {
4209 // memory can "leak through" calls on channels that
4210 // are write-once. Allow this also.
4211 expected_wide_mem = true;
4212 }
4213 assert(expected_wide_mem, "expected narrow slice replacement");
4214 }
4215 }
4216 #else // !ASSERT
4217 #define verify_memory_slice(m,i,n) (void)(0) // PRODUCT version is no-op
4218 #endif
4219
4220
4221 //-----------------------------memory_at---------------------------------------
4222 Node* MergeMemNode::memory_at(uint alias_idx) const {
4223 assert(alias_idx >= Compile::AliasIdxRaw ||
4224 alias_idx == Compile::AliasIdxBot && Compile::current()->AliasLevel() == 0,
4225 "must avoid base_memory and AliasIdxTop");
4226
4227 // Otherwise, it is a narrow slice.
4228 Node* n = alias_idx < req() ? in(alias_idx) : empty_memory();
4229 Compile *C = Compile::current();
4230 if (is_empty_memory(n)) {
4231 // the array is sparse; empty slots are the "top" node
4232 n = base_memory();
4233 assert(Node::in_dump()
4234 || n == NULL || n->bottom_type() == Type::TOP
4235 || n->adr_type() == NULL // address is TOP
4236 || n->adr_type() == TypePtr::BOTTOM
4237 || n->adr_type() == TypeRawPtr::BOTTOM
4238 || Compile::current()->AliasLevel() == 0,
4239 "must be a wide memory");
4240 // AliasLevel == 0 if we are organizing the memory states manually.
4241 // See verify_memory_slice for comments on TypeRawPtr::BOTTOM.
4242 } else {
4243 // make sure the stored slice is sane
4244 #ifdef ASSERT
4245 if (is_error_reported() || Node::in_dump()) {
4246 } else if (might_be_same(n, base_memory())) {
4247 // Give it a pass: It is a mostly harmless repetition of the base.
4248 // This can arise normally from node subsumption during optimization.
4249 } else {
4250 verify_memory_slice(this, alias_idx, n);
4251 }
4252 #endif
4253 }
4254 return n;
4255 }
4256
4257 //---------------------------set_memory_at-------------------------------------
4258 void MergeMemNode::set_memory_at(uint alias_idx, Node *n) {
4259 verify_memory_slice(this, alias_idx, n);
4260 Node* empty_mem = empty_memory();
4261 if (n == base_memory()) n = empty_mem; // collapse default
4262 uint need_req = alias_idx+1;
4263 if (req() < need_req) {
4264 if (n == empty_mem) return; // already the default, so do not grow me
4265 // grow the sparse array
4266 do {
4267 add_req(empty_mem);
4268 } while (req() < need_req);
4269 }
4270 set_req( alias_idx, n );
4271 }
4272
4273
4274
4275 //--------------------------iteration_setup------------------------------------
4276 void MergeMemNode::iteration_setup(const MergeMemNode* other) {
4277 if (other != NULL) {
4278 grow_to_match(other);
4279 // invariant: the finite support of mm2 is within mm->req()
4280 #ifdef ASSERT
4281 for (uint i = req(); i < other->req(); i++) {
4282 assert(other->is_empty_memory(other->in(i)), "slice left uncovered");
4283 }
4284 #endif
4285 }
4286 // Replace spurious copies of base_memory by top.
4287 Node* base_mem = base_memory();
4288 if (base_mem != NULL && !base_mem->is_top()) {
4289 for (uint i = Compile::AliasIdxBot+1, imax = req(); i < imax; i++) {
4290 if (in(i) == base_mem)
4291 set_req(i, empty_memory());
4292 }
4293 }
4294 }
4295
4296 //---------------------------grow_to_match-------------------------------------
4297 void MergeMemNode::grow_to_match(const MergeMemNode* other) {
4298 Node* empty_mem = empty_memory();
4299 assert(other->is_empty_memory(empty_mem), "consistent sentinels");
4300 // look for the finite support of the other memory
4301 for (uint i = other->req(); --i >= req(); ) {
4302 if (other->in(i) != empty_mem) {
4303 uint new_len = i+1;
4304 while (req() < new_len) add_req(empty_mem);
4305 break;
4306 }
4307 }
4308 }
4309
4310 //---------------------------verify_sparse-------------------------------------
4311 #ifndef PRODUCT
4312 bool MergeMemNode::verify_sparse() const {
4313 assert(is_empty_memory(make_empty_memory()), "sane sentinel");
4314 Node* base_mem = base_memory();
4315 // The following can happen in degenerate cases, since empty==top.
4316 if (is_empty_memory(base_mem)) return true;
4317 for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
4318 assert(in(i) != NULL, "sane slice");
4319 if (in(i) == base_mem) return false; // should have been the sentinel value!
4320 }
4321 return true;
4322 }
4323
4324 bool MergeMemStream::match_memory(Node* mem, const MergeMemNode* mm, int idx) {
4325 Node* n;
4326 n = mm->in(idx);
4327 if (mem == n) return true; // might be empty_memory()
4328 n = (idx == Compile::AliasIdxBot)? mm->base_memory(): mm->memory_at(idx);
4329 if (mem == n) return true;
4330 while (n->is_Phi() && (n = n->as_Phi()->is_copy()) != NULL) {
4331 if (mem == n) return true;
4332 if (n == NULL) break;
4333 }
4334 return false;
4335 }
4336 #endif // !PRODUCT