1 /*
2 * Copyright (c) 2016, 2025, 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 "ci/ciFlatArrayKlass.hpp"
26 #include "gc/shared/barrierSet.hpp"
27 #include "gc/shared/c2/barrierSetC2.hpp"
28 #include "gc/shared/c2/cardTableBarrierSetC2.hpp"
29 #include "gc/shared/gc_globals.hpp"
30 #include "opto/arraycopynode.hpp"
31 #include "opto/graphKit.hpp"
32 #include "opto/inlinetypenode.hpp"
33 #include "runtime/sharedRuntime.hpp"
34 #include "utilities/macros.hpp"
35 #include "utilities/powerOfTwo.hpp"
36
37 const TypeFunc* ArrayCopyNode::_arraycopy_type_Type = nullptr;
38
39 ArrayCopyNode::ArrayCopyNode(Compile* C, bool alloc_tightly_coupled, bool has_negative_length_guard)
40 : CallNode(arraycopy_type(), nullptr, TypePtr::BOTTOM),
41 _kind(None),
42 _alloc_tightly_coupled(alloc_tightly_coupled),
43 _has_negative_length_guard(has_negative_length_guard),
44 _arguments_validated(false),
45 _src_type(TypeOopPtr::BOTTOM),
46 _dest_type(TypeOopPtr::BOTTOM) {
47 init_class_id(Class_ArrayCopy);
48 init_flags(Flag_is_macro);
49 C->add_macro_node(this);
50 }
51
52 uint ArrayCopyNode::size_of() const { return sizeof(*this); }
53
54 ArrayCopyNode* ArrayCopyNode::make(GraphKit* kit, bool may_throw,
55 Node* src, Node* src_offset,
56 Node* dest, Node* dest_offset,
57 Node* length,
58 bool alloc_tightly_coupled,
59 bool has_negative_length_guard,
60 Node* src_klass, Node* dest_klass,
61 Node* src_length, Node* dest_length) {
62
63 ArrayCopyNode* ac = new ArrayCopyNode(kit->C, alloc_tightly_coupled, has_negative_length_guard);
64 kit->set_predefined_input_for_runtime_call(ac);
65
66 ac->init_req(ArrayCopyNode::Src, src);
67 ac->init_req(ArrayCopyNode::SrcPos, src_offset);
68 ac->init_req(ArrayCopyNode::Dest, dest);
69 ac->init_req(ArrayCopyNode::DestPos, dest_offset);
70 ac->init_req(ArrayCopyNode::Length, length);
71 ac->init_req(ArrayCopyNode::SrcLen, src_length);
72 ac->init_req(ArrayCopyNode::DestLen, dest_length);
73 ac->init_req(ArrayCopyNode::SrcKlass, src_klass);
74 ac->init_req(ArrayCopyNode::DestKlass, dest_klass);
75
76 if (may_throw) {
77 ac->set_req(TypeFunc::I_O , kit->i_o());
78 kit->add_safepoint_edges(ac, false);
79 }
80
81 return ac;
82 }
83
84 void ArrayCopyNode::connect_outputs(GraphKit* kit, bool deoptimize_on_exception) {
85 kit->set_all_memory_call(this, true);
86 kit->set_control(kit->gvn().transform(new ProjNode(this,TypeFunc::Control)));
87 kit->set_i_o(kit->gvn().transform(new ProjNode(this, TypeFunc::I_O)));
88 kit->make_slow_call_ex(this, kit->env()->Throwable_klass(), true, deoptimize_on_exception);
89 kit->set_all_memory_call(this);
90 }
91
92 #ifndef PRODUCT
93 const char* ArrayCopyNode::_kind_names[] = {"arraycopy", "arraycopy, validated arguments", "clone", "oop array clone", "CopyOf", "CopyOfRange"};
94
95 void ArrayCopyNode::dump_spec(outputStream *st) const {
96 CallNode::dump_spec(st);
97 st->print(" (%s%s)", _kind_names[_kind], _alloc_tightly_coupled ? ", tightly coupled allocation" : "");
98 }
99
100 void ArrayCopyNode::dump_compact_spec(outputStream* st) const {
101 st->print("%s%s", _kind_names[_kind], _alloc_tightly_coupled ? ",tight" : "");
102 }
103 #endif
104
105 intptr_t ArrayCopyNode::get_length_if_constant(PhaseGVN *phase) const {
106 // check that length is constant
107 Node* length = in(ArrayCopyNode::Length);
108 const Type* length_type = phase->type(length);
109
110 if (length_type == Type::TOP) {
111 return -1;
112 }
113
114 assert(is_clonebasic() || is_arraycopy() || is_copyof() || is_copyofrange(), "unexpected array copy type");
115
116 return is_clonebasic() ? length->find_intptr_t_con(-1) : length->find_int_con(-1);
117 }
118
119 int ArrayCopyNode::get_count(PhaseGVN *phase) const {
120 if (is_clonebasic()) {
121 Node* src = in(ArrayCopyNode::Src);
122 const Type* src_type = phase->type(src);
123
124 if (src_type == Type::TOP) {
125 return -1;
126 }
127
128 if (src_type->isa_instptr()) {
129 const TypeInstPtr* inst_src = src_type->is_instptr();
130 ciInstanceKlass* ik = inst_src->instance_klass();
131 // ciInstanceKlass::nof_nonstatic_fields() doesn't take injected
132 // fields into account. They are rare anyway so easier to simply
133 // skip instances with injected fields.
134 if ((!inst_src->klass_is_exact() && (ik->is_interface() || ik->has_subklass())) || ik->has_injected_fields()) {
135 return -1;
136 }
137 int nb_fields = ik->nof_nonstatic_fields();
138 return nb_fields;
139 } else {
140 const TypeAryPtr* ary_src = src_type->isa_aryptr();
141 assert (ary_src != nullptr, "not an array or instance?");
142 // clone passes a length as a rounded number of longs. If we're
143 // cloning an array we'll do it element by element. If the
144 // length of the input array is constant, ArrayCopyNode::Length
145 // must be too. Note that the opposite does not need to hold,
146 // because different input array lengths (e.g. int arrays with
147 // 3 or 4 elements) might lead to the same length input
148 // (e.g. 2 double-words).
149 assert(!ary_src->size()->is_con() || (get_length_if_constant(phase) >= 0) ||
150 (UseArrayFlattening && ary_src->elem()->make_oopptr() != nullptr && ary_src->elem()->make_oopptr()->can_be_inline_type()) ||
151 phase->is_IterGVN() || phase->C->inlining_incrementally() || StressReflectiveCode, "inconsistent");
152 if (ary_src->size()->is_con()) {
153 return ary_src->size()->get_con();
154 }
155 return -1;
156 }
157 }
158
159 return get_length_if_constant(phase);
160 }
161
162 Node* ArrayCopyNode::load(BarrierSetC2* bs, PhaseGVN *phase, Node*& ctl, MergeMemNode* mem, Node* adr, const TypePtr* adr_type, const Type *type, BasicType bt) {
163 // Pin the load: if this is an array load, it's going to be dependent on a condition that's not a range check for that
164 // access. If that condition is replaced by an identical dominating one, then an unpinned load would risk floating
165 // above runtime checks that guarantee it is within bounds.
166 DecoratorSet decorators = C2_READ_ACCESS | C2_CONTROL_DEPENDENT_LOAD | IN_HEAP | C2_ARRAY_COPY | C2_UNKNOWN_CONTROL_LOAD;
167 C2AccessValuePtr addr(adr, adr_type);
168 C2OptAccess access(*phase, ctl, mem, decorators, bt, adr->in(AddPNode::Base), addr);
169 Node* res = bs->load_at(access, type);
170 ctl = access.ctl();
171 return res;
172 }
173
174 void ArrayCopyNode::store(BarrierSetC2* bs, PhaseGVN *phase, Node*& ctl, MergeMemNode* mem, Node* adr, const TypePtr* adr_type, Node* val, const Type *type, BasicType bt) {
175 DecoratorSet decorators = C2_WRITE_ACCESS | IN_HEAP | C2_ARRAY_COPY;
176 if (is_alloc_tightly_coupled()) {
177 decorators |= C2_TIGHTLY_COUPLED_ALLOC;
178 }
179 C2AccessValuePtr addr(adr, adr_type);
180 C2AccessValue value(val, type);
181 C2OptAccess access(*phase, ctl, mem, decorators, bt, adr->in(AddPNode::Base), addr);
182 bs->store_at(access, value);
183 ctl = access.ctl();
184 }
185
186
187 Node* ArrayCopyNode::try_clone_instance(PhaseGVN *phase, bool can_reshape, int count) {
188 if (!is_clonebasic()) {
189 return nullptr;
190 }
191
192 Node* base_src = in(ArrayCopyNode::Src);
193 Node* base_dest = in(ArrayCopyNode::Dest);
194 Node* ctl = in(TypeFunc::Control);
195 Node* in_mem = in(TypeFunc::Memory);
196
197 const Type* src_type = phase->type(base_src);
198 const TypeInstPtr* inst_src = src_type->isa_instptr();
199 if (inst_src == nullptr) {
200 return nullptr;
201 }
202
203 MergeMemNode* mem = phase->transform(MergeMemNode::make(in_mem))->as_MergeMem();
204 phase->record_for_igvn(mem);
205 if (can_reshape) {
206 phase->is_IterGVN()->_worklist.push(mem);
207 }
208
209
210 ciInstanceKlass* ik = inst_src->instance_klass();
211
212 if (!inst_src->klass_is_exact()) {
213 assert(!ik->is_interface(), "inconsistent klass hierarchy");
214 if (ik->has_subklass()) {
215 // Concurrent class loading.
216 // Fail fast and return NodeSentinel to indicate that the transform failed.
217 return NodeSentinel;
218 } else {
219 phase->C->dependencies()->assert_leaf_type(ik);
220 }
221 }
222
223 assert(ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem, "too many fields");
224
225 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
226 for (int i = 0; i < count; i++) {
227 ciField* field = ik->nonstatic_field_at(i);
228 const TypePtr* adr_type = phase->C->alias_type(field)->adr_type();
229 Node* off = phase->MakeConX(field->offset_in_bytes());
230 Node* next_src = phase->transform(new AddPNode(base_src,base_src,off));
231 Node* next_dest = phase->transform(new AddPNode(base_dest,base_dest,off));
232 assert(phase->C->get_alias_index(adr_type) == phase->C->get_alias_index(phase->type(next_src)->isa_ptr()),
233 "slice of address and input slice don't match");
234 assert(phase->C->get_alias_index(adr_type) == phase->C->get_alias_index(phase->type(next_dest)->isa_ptr()),
235 "slice of address and input slice don't match");
236 BasicType bt = field->layout_type();
237
238 const Type *type;
239 if (bt == T_OBJECT) {
240 if (!field->type()->is_loaded()) {
241 type = TypeInstPtr::BOTTOM;
242 } else {
243 ciType* field_klass = field->type();
244 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
245 }
246 } else {
247 type = Type::get_const_basic_type(bt);
248 }
249
250 Node* v = load(bs, phase, ctl, mem, next_src, adr_type, type, bt);
251 store(bs, phase, ctl, mem, next_dest, adr_type, v, type, bt);
252 }
253
254 if (!finish_transform(phase, can_reshape, ctl, mem)) {
255 // Return NodeSentinel to indicate that the transform failed
256 return NodeSentinel;
257 }
258
259 return mem;
260 }
261
262 bool ArrayCopyNode::prepare_array_copy(PhaseGVN *phase, bool can_reshape,
263 Node*& adr_src,
264 Node*& base_src,
265 Node*& adr_dest,
266 Node*& base_dest,
267 BasicType& copy_type,
268 const Type*& value_type,
269 bool& disjoint_bases) {
270 base_src = in(ArrayCopyNode::Src);
271 base_dest = in(ArrayCopyNode::Dest);
272 const Type* src_type = phase->type(base_src);
273 const TypeAryPtr* ary_src = src_type->isa_aryptr();
274
275 Node* src_offset = in(ArrayCopyNode::SrcPos);
276 Node* dest_offset = in(ArrayCopyNode::DestPos);
277
278 if (is_arraycopy() || is_copyofrange() || is_copyof()) {
279 const Type* dest_type = phase->type(base_dest);
280 const TypeAryPtr* ary_dest = dest_type->isa_aryptr();
281
282 // newly allocated object is guaranteed to not overlap with source object
283 disjoint_bases = is_alloc_tightly_coupled();
284 if (ary_src == nullptr || ary_src->elem() == Type::BOTTOM ||
285 ary_dest == nullptr || ary_dest->elem() == Type::BOTTOM) {
286 // We don't know if arguments are arrays
287 return false;
288 }
289
290 BasicType src_elem = ary_src->elem()->array_element_basic_type();
291 BasicType dest_elem = ary_dest->elem()->array_element_basic_type();
292 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
293 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
294
295 // TODO 8350865 What about atomicity?
296 if (src_elem != dest_elem || ary_src->is_null_free() != ary_dest->is_null_free() || ary_src->is_flat() != ary_dest->is_flat() || dest_elem == T_VOID) {
297 // We don't know if arguments are arrays of the same type
298 return false;
299 }
300
301 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
302 if ((!ary_dest->is_flat() && bs->array_copy_requires_gc_barriers(is_alloc_tightly_coupled(), dest_elem, false, false, BarrierSetC2::Optimization)) ||
303 (ary_dest->is_flat() && ary_src->elem()->inline_klass()->contains_oops() &&
304 bs->array_copy_requires_gc_barriers(is_alloc_tightly_coupled(), T_OBJECT, false, false, BarrierSetC2::Optimization))) {
305 // It's an object array copy but we can't emit the card marking that is needed
306 return false;
307 }
308
309 value_type = ary_src->elem();
310
311 uint shift = exact_log2(type2aelembytes(dest_elem));
312 if (ary_dest->is_flat()) {
313 shift = ary_src->flat_log_elem_size();
314 }
315 uint header = arrayOopDesc::base_offset_in_bytes(dest_elem);
316
317 src_offset = Compile::conv_I2X_index(phase, src_offset, ary_src->size());
318 if (src_offset->is_top()) {
319 // Offset is out of bounds (the ArrayCopyNode will be removed)
320 return false;
321 }
322 dest_offset = Compile::conv_I2X_index(phase, dest_offset, ary_dest->size());
323 if (dest_offset->is_top()) {
324 // Offset is out of bounds (the ArrayCopyNode will be removed)
325 if (can_reshape) {
326 // record src_offset, so it can be deleted later (if it is dead)
327 phase->is_IterGVN()->_worklist.push(src_offset);
328 }
329 return false;
330 }
331
332 Node* hook = new Node(1);
333 hook->init_req(0, dest_offset);
334
335 Node* src_scale = phase->transform(new LShiftXNode(src_offset, phase->intcon(shift)));
336
337 hook->destruct(phase);
338
339 Node* dest_scale = phase->transform(new LShiftXNode(dest_offset, phase->intcon(shift)));
340
341 adr_src = phase->transform(new AddPNode(base_src, base_src, src_scale));
342 adr_dest = phase->transform(new AddPNode(base_dest, base_dest, dest_scale));
343
344 adr_src = phase->transform(new AddPNode(base_src, adr_src, phase->MakeConX(header)));
345 adr_dest = phase->transform(new AddPNode(base_dest, adr_dest, phase->MakeConX(header)));
346
347 copy_type = dest_elem;
348 } else {
349 assert(ary_src != nullptr, "should be a clone");
350 assert(is_clonebasic(), "should be");
351
352 disjoint_bases = true;
353
354 if (ary_src->elem()->make_oopptr() != nullptr &&
355 ary_src->elem()->make_oopptr()->can_be_inline_type()) {
356 return false;
357 }
358
359 BasicType elem = ary_src->isa_aryptr()->elem()->array_element_basic_type();
360 if (is_reference_type(elem, true)) {
361 elem = T_OBJECT;
362 }
363
364 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
365 if ((!ary_src->is_flat() && bs->array_copy_requires_gc_barriers(true, elem, true, is_clone_inst(), BarrierSetC2::Optimization)) ||
366 (ary_src->is_flat() && ary_src->elem()->inline_klass()->contains_oops() &&
367 bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, is_clone_inst(), BarrierSetC2::Optimization))) {
368 // It's an object array copy but we can't emit the card marking that is needed
369 return false;
370 }
371
372 adr_src = phase->transform(new AddPNode(base_src, base_src, src_offset));
373 adr_dest = phase->transform(new AddPNode(base_dest, base_dest, dest_offset));
374
375 // The address is offsetted to an aligned address where a raw copy would start.
376 // If the clone copy is decomposed into load-stores - the address is adjusted to
377 // point at where the array starts.
378 const Type* toff = phase->type(src_offset);
379 int offset = toff->isa_long() ? (int) toff->is_long()->get_con() : (int) toff->is_int()->get_con();
380 int diff = arrayOopDesc::base_offset_in_bytes(elem) - offset;
381 assert(diff >= 0, "clone should not start after 1st array element");
382 if (diff > 0) {
383 adr_src = phase->transform(new AddPNode(base_src, adr_src, phase->MakeConX(diff)));
384 adr_dest = phase->transform(new AddPNode(base_dest, adr_dest, phase->MakeConX(diff)));
385 }
386 copy_type = elem;
387 value_type = ary_src->elem();
388 }
389 return true;
390 }
391
392 const TypeAryPtr* ArrayCopyNode::get_address_type(PhaseGVN* phase, const TypePtr* atp, Node* n) {
393 if (atp == TypeOopPtr::BOTTOM) {
394 atp = phase->type(n)->isa_ptr();
395 }
396 // adjust atp to be the correct array element address type
397 return atp->add_offset(Type::OffsetBot)->is_aryptr();
398 }
399
400 void ArrayCopyNode::array_copy_test_overlap(GraphKit& kit, bool disjoint_bases, int count, Node*& backward_ctl) {
401 Node* ctl = kit.control();
402 if (!disjoint_bases && count > 1) {
403 PhaseGVN& gvn = kit.gvn();
404 Node* src_offset = in(ArrayCopyNode::SrcPos);
405 Node* dest_offset = in(ArrayCopyNode::DestPos);
406 assert(src_offset != nullptr && dest_offset != nullptr, "should be");
407 Node* cmp = gvn.transform(new CmpINode(src_offset, dest_offset));
408 Node *bol = gvn.transform(new BoolNode(cmp, BoolTest::lt));
409 IfNode *iff = new IfNode(ctl, bol, PROB_FAIR, COUNT_UNKNOWN);
410
411 gvn.transform(iff);
412
413 kit.set_control(gvn.transform(new IfFalseNode(iff)));
414 backward_ctl = gvn.transform(new IfTrueNode(iff));
415 }
416 }
417
418 void ArrayCopyNode::copy(GraphKit& kit,
419 const TypeAryPtr* atp_src,
420 const TypeAryPtr* atp_dest,
421 int i,
422 Node* base_src,
423 Node* base_dest,
424 Node* adr_src,
425 Node* adr_dest,
426 BasicType copy_type,
427 const Type* value_type) {
428 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
429 Node* ctl = kit.control();
430 if (atp_dest->is_flat()) {
431 ciInlineKlass* vk = atp_src->elem()->inline_klass();
432 for (int j = 0; j < vk->nof_nonstatic_fields(); j++) {
433 ciField* field = vk->nonstatic_field_at(j);
434 int off_in_vt = field->offset_in_bytes() - vk->payload_offset();
435 Node* off = kit.MakeConX(off_in_vt + i * atp_src->flat_elem_size());
436 ciType* ft = field->type();
437 BasicType bt = type2field[ft->basic_type()];
438 assert(!field->is_flat(), "flat field encountered");
439 const Type* rt = Type::get_const_type(ft);
440 const TypePtr* adr_type = atp_src->with_field_offset(off_in_vt)->add_offset(Type::OffsetBot);
441 assert(!bs->array_copy_requires_gc_barriers(is_alloc_tightly_coupled(), bt, false, false, BarrierSetC2::Optimization), "GC barriers required");
442 Node* next_src = kit.gvn().transform(new AddPNode(base_src, adr_src, off));
443 Node* next_dest = kit.gvn().transform(new AddPNode(base_dest, adr_dest, off));
444 Node* v = load(bs, &kit.gvn(), ctl, kit.merged_memory(), next_src, adr_type, rt, bt);
445 store(bs, &kit.gvn(), ctl, kit.merged_memory(), next_dest, adr_type, v, rt, bt);
446 }
447 } else {
448 Node* off = kit.MakeConX(type2aelembytes(copy_type) * i);
449 Node* next_src = kit.gvn().transform(new AddPNode(base_src, adr_src, off));
450 Node* next_dest = kit.gvn().transform(new AddPNode(base_dest, adr_dest, off));
451 Node* v = load(bs, &kit.gvn(), ctl, kit.merged_memory(), next_src, atp_src, value_type, copy_type);
452 store(bs, &kit.gvn(), ctl, kit.merged_memory(), next_dest, atp_dest, v, value_type, copy_type);
453 }
454 kit.set_control(ctl);
455 }
456
457
458 void ArrayCopyNode::array_copy_forward(GraphKit& kit,
459 bool can_reshape,
460 const TypeAryPtr* atp_src,
461 const TypeAryPtr* atp_dest,
462 Node* adr_src,
463 Node* base_src,
464 Node* adr_dest,
465 Node* base_dest,
466 BasicType copy_type,
467 const Type* value_type,
468 int count) {
469 if (!kit.stopped()) {
470 // copy forward
471 if (count > 0) {
472 for (int i = 0; i < count; i++) {
473 copy(kit, atp_src, atp_dest, i, base_src, base_dest, adr_src, adr_dest, copy_type, value_type);
474 }
475 } else if (can_reshape) {
476 PhaseGVN& gvn = kit.gvn();
477 assert(gvn.is_IterGVN(), "");
478 gvn.record_for_igvn(adr_src);
479 gvn.record_for_igvn(adr_dest);
480 }
481 }
482 }
483
484 void ArrayCopyNode::array_copy_backward(GraphKit& kit,
485 bool can_reshape,
486 const TypeAryPtr* atp_src,
487 const TypeAryPtr* atp_dest,
488 Node* adr_src,
489 Node* base_src,
490 Node* adr_dest,
491 Node* base_dest,
492 BasicType copy_type,
493 const Type* value_type,
494 int count) {
495 if (!kit.stopped()) {
496 // copy backward
497 PhaseGVN& gvn = kit.gvn();
498
499 if (count > 0) {
500 for (int i = count-1; i >= 0; i--) {
501 copy(kit, atp_src, atp_dest, i, base_src, base_dest, adr_src, adr_dest, copy_type, value_type);
502 }
503 } else if(can_reshape) {
504 PhaseGVN& gvn = kit.gvn();
505 assert(gvn.is_IterGVN(), "");
506 gvn.record_for_igvn(adr_src);
507 gvn.record_for_igvn(adr_dest);
508 }
509 }
510 }
511
512 bool ArrayCopyNode::finish_transform(PhaseGVN *phase, bool can_reshape,
513 Node* ctl, Node *mem) {
514 if (can_reshape) {
515 PhaseIterGVN* igvn = phase->is_IterGVN();
516 igvn->set_delay_transform(false);
517 if (is_clonebasic()) {
518 Node* out_mem = proj_out(TypeFunc::Memory);
519
520 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
521 if (out_mem->outcnt() != 1 || !out_mem->raw_out(0)->is_MergeMem() ||
522 out_mem->raw_out(0)->outcnt() != 1 || !out_mem->raw_out(0)->raw_out(0)->is_MemBar()) {
523 assert(bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, is_clone_inst(), BarrierSetC2::Optimization), "can only happen with card marking");
524 return false;
525 }
526
527 igvn->replace_node(out_mem->raw_out(0), mem);
528
529 Node* out_ctl = proj_out(TypeFunc::Control);
530 igvn->replace_node(out_ctl, ctl);
531 } else {
532 // replace fallthrough projections of the ArrayCopyNode by the
533 // new memory, control and the input IO.
534 CallProjections* callprojs = extract_projections(true, false);
535
536 if (callprojs->fallthrough_ioproj != nullptr) {
537 igvn->replace_node(callprojs->fallthrough_ioproj, in(TypeFunc::I_O));
538 }
539 if (callprojs->fallthrough_memproj != nullptr) {
540 igvn->replace_node(callprojs->fallthrough_memproj, mem);
541 }
542 if (callprojs->fallthrough_catchproj != nullptr) {
543 igvn->replace_node(callprojs->fallthrough_catchproj, ctl);
544 }
545
546 // The ArrayCopyNode is not disconnected. It still has the
547 // projections for the exception case. Replace current
548 // ArrayCopyNode with a dummy new one with a top() control so
549 // that this part of the graph stays consistent but is
550 // eventually removed.
551
552 set_req(0, phase->C->top());
553 remove_dead_region(phase, can_reshape);
554 }
555 } else {
556 if (in(TypeFunc::Control) != ctl) {
557 // we can't return new memory and control from Ideal at parse time
558 assert(!is_clonebasic() || UseShenandoahGC, "added control for clone?");
559 phase->record_for_igvn(this);
560 return false;
561 }
562 }
563 return true;
564 }
565
566
567 Node *ArrayCopyNode::Ideal(PhaseGVN *phase, bool can_reshape) {
568 // Perform any generic optimizations first
569 Node* result = SafePointNode::Ideal(phase, can_reshape);
570 if (result != nullptr) {
571 return result;
572 }
573
574 if (StressArrayCopyMacroNode && !can_reshape) {
575 phase->record_for_igvn(this);
576 return nullptr;
577 }
578
579 // See if it's a small array copy and we can inline it as
580 // loads/stores
581 // Here we can only do:
582 // - arraycopy if all arguments were validated before and we don't
583 // need card marking
584 // - clone for which we don't need to do card marking
585
586 if (!is_clonebasic() && !is_arraycopy_validated() &&
587 !is_copyofrange_validated() && !is_copyof_validated()) {
588 return nullptr;
589 }
590
591 assert(in(TypeFunc::Control) != nullptr &&
592 in(TypeFunc::Memory) != nullptr &&
593 in(ArrayCopyNode::Src) != nullptr &&
594 in(ArrayCopyNode::Dest) != nullptr &&
595 in(ArrayCopyNode::Length) != nullptr &&
596 in(ArrayCopyNode::SrcPos) != nullptr &&
597 in(ArrayCopyNode::DestPos) != nullptr, "broken inputs");
598
599 if (in(TypeFunc::Control)->is_top() ||
600 in(TypeFunc::Memory)->is_top() ||
601 phase->type(in(ArrayCopyNode::Src)) == Type::TOP ||
602 phase->type(in(ArrayCopyNode::Dest)) == Type::TOP ||
603 (in(ArrayCopyNode::SrcPos) != nullptr && in(ArrayCopyNode::SrcPos)->is_top()) ||
604 (in(ArrayCopyNode::DestPos) != nullptr && in(ArrayCopyNode::DestPos)->is_top())) {
605 return nullptr;
606 }
607
608 int count = get_count(phase);
609
610 if (count < 0 || count > ArrayCopyLoadStoreMaxElem) {
611 return nullptr;
612 }
613
614 Node* src = in(ArrayCopyNode::Src);
615 Node* dest = in(ArrayCopyNode::Dest);
616 const Type* src_type = phase->type(src);
617 const Type* dest_type = phase->type(dest);
618
619 if (src_type->isa_aryptr() && dest_type->isa_instptr()) {
620 // clone used for load of unknown inline type can't be optimized at
621 // this point
622 return nullptr;
623 }
624
625 Node* mem = try_clone_instance(phase, can_reshape, count);
626 if (mem != nullptr) {
627 return (mem == NodeSentinel) ? nullptr : mem;
628 }
629
630 Node* adr_src = nullptr;
631 Node* base_src = nullptr;
632 Node* adr_dest = nullptr;
633 Node* base_dest = nullptr;
634 BasicType copy_type = T_ILLEGAL;
635 const Type* value_type = nullptr;
636 bool disjoint_bases = false;
637
638 if (!prepare_array_copy(phase, can_reshape,
639 adr_src, base_src, adr_dest, base_dest,
640 copy_type, value_type, disjoint_bases)) {
641 assert(adr_src == nullptr, "no node can be left behind");
642 assert(adr_dest == nullptr, "no node can be left behind");
643 return nullptr;
644 }
645
646 JVMState* new_jvms = nullptr;
647 SafePointNode* new_map = nullptr;
648 if (!is_clonebasic()) {
649 new_jvms = jvms()->clone_shallow(phase->C);
650 new_map = new SafePointNode(req(), new_jvms);
651 for (uint i = TypeFunc::FramePtr; i < req(); i++) {
652 new_map->init_req(i, in(i));
653 }
654 new_jvms->set_map(new_map);
655 } else {
656 new_jvms = new (phase->C) JVMState(0);
657 new_map = new SafePointNode(TypeFunc::Parms, new_jvms);
658 new_jvms->set_map(new_map);
659 }
660 new_map->set_control(in(TypeFunc::Control));
661 new_map->set_memory(MergeMemNode::make(in(TypeFunc::Memory)));
662 new_map->set_i_o(in(TypeFunc::I_O));
663 phase->record_for_igvn(new_map);
664
665 const TypeAryPtr* atp_src = get_address_type(phase, _src_type, src);
666 const TypeAryPtr* atp_dest = get_address_type(phase, _dest_type, dest);
667
668 if (can_reshape) {
669 assert(!phase->is_IterGVN()->delay_transform(), "cannot delay transforms");
670 phase->is_IterGVN()->set_delay_transform(true);
671 }
672
673 GraphKit kit(new_jvms, phase);
674
675 SafePointNode* backward_map = nullptr;
676 SafePointNode* forward_map = nullptr;
677 Node* backward_ctl = phase->C->top();
678
679 array_copy_test_overlap(kit, disjoint_bases, count, backward_ctl);
680
681 {
682 PreserveJVMState pjvms(&kit);
683
684 array_copy_forward(kit, can_reshape,
685 atp_src, atp_dest,
686 adr_src, base_src, adr_dest, base_dest,
687 copy_type, value_type, count);
688
689 forward_map = kit.stop();
690 }
691
692 kit.set_control(backward_ctl);
693 array_copy_backward(kit, can_reshape,
694 atp_src, atp_dest,
695 adr_src, base_src, adr_dest, base_dest,
696 copy_type, value_type, count);
697
698 backward_map = kit.stop();
699
700 if (!forward_map->control()->is_top() && !backward_map->control()->is_top()) {
701 assert(forward_map->i_o() == backward_map->i_o(), "need a phi on IO?");
702 Node* ctl = new RegionNode(3);
703 Node* mem = new PhiNode(ctl, Type::MEMORY, TypePtr::BOTTOM);
704 kit.set_map(forward_map);
705 ctl->init_req(1, kit.control());
706 mem->init_req(1, kit.reset_memory());
707 kit.set_map(backward_map);
708 ctl->init_req(2, kit.control());
709 mem->init_req(2, kit.reset_memory());
710 kit.set_control(phase->transform(ctl));
711 kit.set_all_memory(phase->transform(mem));
712 } else if (!forward_map->control()->is_top()) {
713 kit.set_map(forward_map);
714 } else {
715 assert(!backward_map->control()->is_top(), "no copy?");
716 kit.set_map(backward_map);
717 }
718
719 if (can_reshape) {
720 assert(phase->is_IterGVN()->delay_transform(), "should be delaying transforms");
721 phase->is_IterGVN()->set_delay_transform(false);
722 }
723
724 mem = kit.map()->memory();
725 if (!finish_transform(phase, can_reshape, kit.control(), mem)) {
726 if (!can_reshape) {
727 phase->record_for_igvn(this);
728 } else {
729 // put in worklist, so that if it happens to be dead it is removed
730 phase->is_IterGVN()->_worklist.push(mem);
731 }
732 return nullptr;
733 }
734
735 return mem;
736 }
737
738 bool ArrayCopyNode::may_modify(const TypeOopPtr* t_oop, PhaseValues* phase) {
739 Node* dest = in(ArrayCopyNode::Dest);
740 if (dest->is_top()) {
741 return false;
742 }
743 const TypeOopPtr* dest_t = phase->type(dest)->is_oopptr();
744 assert(!dest_t->is_known_instance() || _dest_type->is_known_instance(), "result of EA not recorded");
745 assert(in(ArrayCopyNode::Src)->is_top() || !phase->type(in(ArrayCopyNode::Src))->is_oopptr()->is_known_instance() ||
746 _src_type->is_known_instance(), "result of EA not recorded");
747
748 if (_dest_type != TypeOopPtr::BOTTOM || t_oop->is_known_instance()) {
749 assert(_dest_type == TypeOopPtr::BOTTOM || _dest_type->is_known_instance(), "result of EA is known instance");
750 return t_oop->instance_id() == _dest_type->instance_id();
751 }
752
753 return CallNode::may_modify_arraycopy_helper(dest_t, t_oop, phase);
754 }
755
756 bool ArrayCopyNode::may_modify_helper(const TypeOopPtr* t_oop, Node* n, PhaseValues* phase, CallNode*& call) {
757 if (n != nullptr &&
758 n->is_Call() &&
759 n->as_Call()->may_modify(t_oop, phase) &&
760 (n->as_Call()->is_ArrayCopy() || n->as_Call()->is_call_to_arraycopystub())) {
761 call = n->as_Call();
762 return true;
763 }
764 return false;
765 }
766
767 bool ArrayCopyNode::may_modify(const TypeOopPtr* t_oop, MemBarNode* mb, PhaseValues* phase, ArrayCopyNode*& ac) {
768 Node* c = mb->in(0);
769
770 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
771 // step over g1 gc barrier if we're at e.g. a clone with ReduceInitialCardMarks off
772 c = bs->step_over_gc_barrier(c);
773
774 CallNode* call = nullptr;
775 guarantee(c != nullptr, "step_over_gc_barrier failed, there must be something to step to.");
776 if (c->is_Region()) {
777 for (uint i = 1; i < c->req(); i++) {
778 if (c->in(i) != nullptr) {
779 Node* n = c->in(i)->in(0);
780 if (may_modify_helper(t_oop, n, phase, call)) {
781 ac = call->isa_ArrayCopy();
782 assert(c == mb->in(0), "only for clone");
783 return true;
784 }
785 }
786 }
787 } else if (may_modify_helper(t_oop, c->in(0), phase, call)) {
788 ac = call->isa_ArrayCopy();
789 #ifdef ASSERT
790 bool use_ReduceInitialCardMarks = BarrierSet::barrier_set()->is_a(BarrierSet::CardTableBarrierSet) &&
791 static_cast<CardTableBarrierSetC2*>(bs)->use_ReduceInitialCardMarks();
792 assert(c == mb->in(0) || (ac != nullptr && ac->is_clonebasic() && !use_ReduceInitialCardMarks), "only for clone");
793 #endif
794 return true;
795 } else if (mb->trailing_partial_array_copy()) {
796 return true;
797 }
798
799 return false;
800 }
801
802 // Does this array copy modify offsets between offset_lo and offset_hi
803 // in the destination array
804 // if must_modify is false, return true if the copy could write
805 // between offset_lo and offset_hi
806 // if must_modify is true, return true if the copy is guaranteed to
807 // write between offset_lo and offset_hi
808 bool ArrayCopyNode::modifies(intptr_t offset_lo, intptr_t offset_hi, PhaseValues* phase, bool must_modify) const {
809 assert(_kind == ArrayCopy || _kind == CopyOf || _kind == CopyOfRange, "only for real array copies");
810
811 Node* dest = in(Dest);
812 Node* dest_pos = in(DestPos);
813 Node* len = in(Length);
814
815 const TypeInt *dest_pos_t = phase->type(dest_pos)->isa_int();
816 const TypeInt *len_t = phase->type(len)->isa_int();
817 const TypeAryPtr* ary_t = phase->type(dest)->isa_aryptr();
818
819 if (dest_pos_t == nullptr || len_t == nullptr || ary_t == nullptr) {
820 return !must_modify;
821 }
822
823 BasicType ary_elem = ary_t->isa_aryptr()->elem()->array_element_basic_type();
824 if (is_reference_type(ary_elem, true)) ary_elem = T_OBJECT;
825
826 uint header = arrayOopDesc::base_offset_in_bytes(ary_elem);
827 uint elemsize = ary_t->is_flat() ? ary_t->flat_elem_size() : type2aelembytes(ary_elem);
828
829 jlong dest_pos_plus_len_lo = (((jlong)dest_pos_t->_lo) + len_t->_lo) * elemsize + header;
830 jlong dest_pos_plus_len_hi = (((jlong)dest_pos_t->_hi) + len_t->_hi) * elemsize + header;
831 jlong dest_pos_lo = ((jlong)dest_pos_t->_lo) * elemsize + header;
832 jlong dest_pos_hi = ((jlong)dest_pos_t->_hi) * elemsize + header;
833
834 if (must_modify) {
835 if (offset_lo >= dest_pos_hi && offset_hi < dest_pos_plus_len_lo) {
836 return true;
837 }
838 } else {
839 if (offset_hi >= dest_pos_lo && offset_lo < dest_pos_plus_len_hi) {
840 return true;
841 }
842 }
843 return false;
844 }
845
846 // As an optimization, choose optimum vector size for copy length known at compile time.
847 int ArrayCopyNode::get_partial_inline_vector_lane_count(BasicType type, int const_len) {
848 int lane_count = ArrayOperationPartialInlineSize/type2aelembytes(type);
849 if (const_len > 0) {
850 int size_in_bytes = const_len * type2aelembytes(type);
851 if (size_in_bytes <= 16)
852 lane_count = 16/type2aelembytes(type);
853 else if (size_in_bytes > 16 && size_in_bytes <= 32)
854 lane_count = 32/type2aelembytes(type);
855 }
856 return lane_count;
857 }