1 /*
2 * Copyright (c) 1999, 2017, 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 "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "opto/addnode.hpp"
34 #include "opto/arraycopynode.hpp"
35 #include "opto/c2compiler.hpp"
36 #include "opto/callGenerator.hpp"
37 #include "opto/castnode.hpp"
38 #include "opto/cfgnode.hpp"
39 #include "opto/convertnode.hpp"
40 #include "opto/countbitsnode.hpp"
41 #include "opto/intrinsicnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/mathexactnode.hpp"
44 #include "opto/movenode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/subnode.hpp"
52 #include "prims/nativeLookup.hpp"
53 #include "prims/unsafe.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #ifdef TRACE_HAVE_INTRINSICS
56 #include "trace/traceMacros.hpp"
57 #endif
58
59 class LibraryIntrinsic : public InlineCallGenerator {
60 // Extend the set of intrinsics known to the runtime:
61 public:
62 private:
63 bool _is_virtual;
64 bool _does_virtual_dispatch;
65 int8_t _predicates_count; // Intrinsic is predicated by several conditions
66 int8_t _last_predicate; // Last generated predicate
67 vmIntrinsics::ID _intrinsic_id;
68
69 public:
70 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
71 : InlineCallGenerator(m),
72 _is_virtual(is_virtual),
73 _does_virtual_dispatch(does_virtual_dispatch),
74 _predicates_count((int8_t)predicates_count),
75 _last_predicate((int8_t)-1),
76 _intrinsic_id(id)
77 {
78 }
79 virtual bool is_intrinsic() const { return true; }
80 virtual bool is_virtual() const { return _is_virtual; }
81 virtual bool is_predicated() const { return _predicates_count > 0; }
82 virtual int predicates_count() const { return _predicates_count; }
83 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
84 virtual JVMState* generate(JVMState* jvms);
85 virtual Node* generate_predicate(JVMState* jvms, int predicate);
86 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
87 };
88
89
90 // Local helper class for LibraryIntrinsic:
91 class LibraryCallKit : public GraphKit {
92 private:
93 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
94 Node* _result; // the result node, if any
95 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
96
97 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
98
99 public:
100 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
101 : GraphKit(jvms),
102 _intrinsic(intrinsic),
103 _result(NULL)
104 {
105 // Check if this is a root compile. In that case we don't have a caller.
106 if (!jvms->has_method()) {
107 _reexecute_sp = sp();
108 } else {
109 // Find out how many arguments the interpreter needs when deoptimizing
110 // and save the stack pointer value so it can used by uncommon_trap.
111 // We find the argument count by looking at the declared signature.
112 bool ignored_will_link;
113 ciSignature* declared_signature = NULL;
114 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
115 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
116 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
117 }
118 }
119
120 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
121
122 ciMethod* caller() const { return jvms()->method(); }
123 int bci() const { return jvms()->bci(); }
124 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
125 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
126 ciMethod* callee() const { return _intrinsic->method(); }
127
128 bool try_to_inline(int predicate);
129 Node* try_to_predicate(int predicate);
130
131 void push_result() {
132 // Push the result onto the stack.
133 if (!stopped() && result() != NULL) {
134 BasicType bt = result()->bottom_type()->basic_type();
135 push_node(bt, result());
136 }
137 }
138
139 private:
140 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
141 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
142 }
143
144 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
145 void set_result(RegionNode* region, PhiNode* value);
146 Node* result() { return _result; }
147
148 virtual int reexecute_sp() { return _reexecute_sp; }
149
150 // Helper functions to inline natives
151 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
152 Node* generate_slow_guard(Node* test, RegionNode* region);
153 Node* generate_fair_guard(Node* test, RegionNode* region);
154 Node* generate_negative_guard(Node* index, RegionNode* region,
155 // resulting CastII of index:
156 Node* *pos_index = NULL);
157 Node* generate_limit_guard(Node* offset, Node* subseq_length,
158 Node* array_length,
159 RegionNode* region);
160 void generate_string_range_check(Node* array, Node* offset,
161 Node* length, bool char_count);
162 Node* generate_current_thread(Node* &tls_output);
163 Node* load_mirror_from_klass(Node* klass);
164 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
165 RegionNode* region, int null_path,
166 int offset);
167 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path) {
169 int offset = java_lang_Class::klass_offset_in_bytes();
170 return load_klass_from_mirror_common(mirror, never_see_null,
171 region, null_path,
172 offset);
173 }
174 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
175 RegionNode* region, int null_path) {
176 int offset = java_lang_Class::array_klass_offset_in_bytes();
177 return load_klass_from_mirror_common(mirror, never_see_null,
178 region, null_path,
179 offset);
180 }
181 Node* generate_access_flags_guard(Node* kls,
182 int modifier_mask, int modifier_bits,
183 RegionNode* region);
184 Node* generate_interface_guard(Node* kls, RegionNode* region);
185 Node* generate_array_guard(Node* kls, RegionNode* region) {
186 return generate_array_guard_common(kls, region, false, false);
187 }
188 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, true);
190 }
191 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, true, false);
193 }
194 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, true);
196 }
197 Node* generate_array_guard_common(Node* kls, RegionNode* region,
198 bool obj_array, bool not_array);
199 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
200 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
201 bool is_virtual = false, bool is_static = false);
202 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
203 return generate_method_call(method_id, false, true);
204 }
205 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, true, false);
207 }
208 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
209 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
210
211 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
213 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
214 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
215 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
216 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfChar();
218 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_toBytesU();
220 bool inline_string_getCharsU();
221 bool inline_string_copy(bool compress);
222 bool inline_string_char_access(bool is_store);
223 Node* round_double_node(Node* n);
224 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
225 bool inline_math_native(vmIntrinsics::ID id);
226 bool inline_math(vmIntrinsics::ID id);
227 template <typename OverflowOp>
228 bool inline_math_overflow(Node* arg1, Node* arg2);
229 void inline_math_mathExact(Node* math, Node* test);
230 bool inline_math_addExactI(bool is_increment);
231 bool inline_math_addExactL(bool is_increment);
232 bool inline_math_multiplyExactI();
233 bool inline_math_multiplyExactL();
234 bool inline_math_negateExactI();
235 bool inline_math_negateExactL();
236 bool inline_math_subtractExactI(bool is_decrement);
237 bool inline_math_subtractExactL(bool is_decrement);
238 bool inline_min_max(vmIntrinsics::ID id);
239 bool inline_notify(vmIntrinsics::ID id);
240 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
241 // This returns Type::AnyPtr, RawPtr, or OopPtr.
242 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
243 Node* make_unsafe_address(Node*& base, Node* offset, BasicType type = T_ILLEGAL);
244 // Helper for inline_unsafe_access.
245 // Generates the guards that check whether the result of
246 // Unsafe.getObject should be recorded in an SATB log buffer.
247 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
248
249 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
250 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
251 static bool klass_needs_init_guard(Node* kls);
252 bool inline_unsafe_allocate();
253 bool inline_unsafe_newArray(bool uninitialized);
254 bool inline_unsafe_copyMemory();
255 bool inline_native_currentThread();
256
257 bool inline_native_time_funcs(address method, const char* funcName);
258 #ifdef TRACE_HAVE_INTRINSICS
259 bool inline_native_classID();
260 bool inline_native_getBufferWriter();
261 #endif
262 bool inline_native_isInterrupted();
263 bool inline_native_Class_query(vmIntrinsics::ID id);
264 bool inline_native_subtype_check();
265 bool inline_native_getLength();
266 bool inline_array_copyOf(bool is_copyOfRange);
267 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
268 bool inline_preconditions_checkIndex();
269 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
270 bool inline_native_clone(bool is_virtual);
271 bool inline_native_Reflection_getCallerClass();
272 // Helper function for inlining native object hash method
273 bool inline_native_hashcode(bool is_virtual, bool is_static);
274 bool inline_native_getClass();
275
276 // Helper functions for inlining arraycopy
277 bool inline_arraycopy();
278 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
279 RegionNode* slow_region);
280 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
281 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
282 uint new_idx);
283
284 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
285 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
286 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
287 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
288 bool inline_unsafe_fence(vmIntrinsics::ID id);
289 bool inline_onspinwait();
290 bool inline_fp_conversions(vmIntrinsics::ID id);
291 bool inline_number_methods(vmIntrinsics::ID id);
292 bool inline_reference_get();
293 bool inline_Class_cast();
294 bool inline_aescrypt_Block(vmIntrinsics::ID id);
295 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
296 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
297 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
298 Node* inline_counterMode_AESCrypt_predicate();
299 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
300 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
301 bool inline_ghash_processBlocks();
302 bool inline_sha_implCompress(vmIntrinsics::ID id);
303 bool inline_digestBase_implCompressMB(int predicate);
304 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
305 bool long_state, address stubAddr, const char *stubName,
306 Node* src_start, Node* ofs, Node* limit);
307 Node* get_state_from_sha_object(Node *sha_object);
308 Node* get_state_from_sha5_object(Node *sha_object);
309 Node* inline_digestBase_implCompressMB_predicate(int predicate);
310 bool inline_encodeISOArray();
311 bool inline_updateCRC32();
312 bool inline_updateBytesCRC32();
313 bool inline_updateByteBufferCRC32();
314 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
315 bool inline_updateBytesCRC32C();
316 bool inline_updateDirectByteBufferCRC32C();
317 bool inline_updateBytesAdler32();
318 bool inline_updateByteBufferAdler32();
319 bool inline_multiplyToLen();
320 bool inline_hasNegatives();
321 bool inline_squareToLen();
322 bool inline_mulAdd();
323 bool inline_montgomeryMultiply();
324 bool inline_montgomerySquare();
325 bool inline_vectorizedMismatch();
326 bool inline_fma(vmIntrinsics::ID id);
327
328 bool inline_profileBoolean();
329 bool inline_isCompileConstant();
330 };
331
332 //---------------------------make_vm_intrinsic----------------------------
333 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
334 vmIntrinsics::ID id = m->intrinsic_id();
335 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
336
337 if (!m->is_loaded()) {
338 // Do not attempt to inline unloaded methods.
339 return NULL;
340 }
341
342 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
343 bool is_available = false;
344
345 {
346 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
347 // the compiler must transition to '_thread_in_vm' state because both
348 // methods access VM-internal data.
349 VM_ENTRY_MARK;
350 methodHandle mh(THREAD, m->get_Method());
351 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
352 !C->directive()->is_intrinsic_disabled(mh) &&
353 !vmIntrinsics::is_disabled_by_flags(mh);
354
355 }
356
357 if (is_available) {
358 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
359 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
360 return new LibraryIntrinsic(m, is_virtual,
361 vmIntrinsics::predicates_needed(id),
362 vmIntrinsics::does_virtual_dispatch(id),
363 (vmIntrinsics::ID) id);
364 } else {
365 return NULL;
366 }
367 }
368
369 //----------------------register_library_intrinsics-----------------------
370 // Initialize this file's data structures, for each Compile instance.
371 void Compile::register_library_intrinsics() {
372 // Nothing to do here.
373 }
374
375 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
376 LibraryCallKit kit(jvms, this);
377 Compile* C = kit.C;
378 int nodes = C->unique();
379 #ifndef PRODUCT
380 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
381 char buf[1000];
382 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
383 tty->print_cr("Intrinsic %s", str);
384 }
385 #endif
386 ciMethod* callee = kit.callee();
387 const int bci = kit.bci();
388
389 // Try to inline the intrinsic.
390 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
391 kit.try_to_inline(_last_predicate)) {
392 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
393 : "(intrinsic)";
394 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
395 if (C->print_intrinsics() || C->print_inlining()) {
396 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
397 }
398 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
399 if (C->log()) {
400 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
401 vmIntrinsics::name_at(intrinsic_id()),
402 (is_virtual() ? " virtual='1'" : ""),
403 C->unique() - nodes);
404 }
405 // Push the result from the inlined method onto the stack.
406 kit.push_result();
407 C->print_inlining_update(this);
408 return kit.transfer_exceptions_into_jvms();
409 }
410
411 // The intrinsic bailed out
412 if (jvms->has_method()) {
413 // Not a root compile.
414 const char* msg;
415 if (callee->intrinsic_candidate()) {
416 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
417 } else {
418 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
419 : "failed to inline (intrinsic), method not annotated";
420 }
421 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
422 if (C->print_intrinsics() || C->print_inlining()) {
423 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
424 }
425 } else {
426 // Root compile
427 ResourceMark rm;
428 stringStream msg_stream;
429 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
430 vmIntrinsics::name_at(intrinsic_id()),
431 is_virtual() ? " (virtual)" : "", bci);
432 const char *msg = msg_stream.as_string();
433 log_debug(jit, inlining)("%s", msg);
434 if (C->print_intrinsics() || C->print_inlining()) {
435 tty->print("%s", msg);
436 }
437 }
438 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
439 C->print_inlining_update(this);
440 return NULL;
441 }
442
443 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
444 LibraryCallKit kit(jvms, this);
445 Compile* C = kit.C;
446 int nodes = C->unique();
447 _last_predicate = predicate;
448 #ifndef PRODUCT
449 assert(is_predicated() && predicate < predicates_count(), "sanity");
450 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
451 char buf[1000];
452 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
453 tty->print_cr("Predicate for intrinsic %s", str);
454 }
455 #endif
456 ciMethod* callee = kit.callee();
457 const int bci = kit.bci();
458
459 Node* slow_ctl = kit.try_to_predicate(predicate);
460 if (!kit.failing()) {
461 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
462 : "(intrinsic, predicate)";
463 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
464 if (C->print_intrinsics() || C->print_inlining()) {
465 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
466 }
467 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
468 if (C->log()) {
469 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
470 vmIntrinsics::name_at(intrinsic_id()),
471 (is_virtual() ? " virtual='1'" : ""),
472 C->unique() - nodes);
473 }
474 return slow_ctl; // Could be NULL if the check folds.
475 }
476
477 // The intrinsic bailed out
478 if (jvms->has_method()) {
479 // Not a root compile.
480 const char* msg = "failed to generate predicate for intrinsic";
481 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
482 if (C->print_intrinsics() || C->print_inlining()) {
483 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
484 }
485 } else {
486 // Root compile
487 ResourceMark rm;
488 stringStream msg_stream;
489 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
490 vmIntrinsics::name_at(intrinsic_id()),
491 is_virtual() ? " (virtual)" : "", bci);
492 const char *msg = msg_stream.as_string();
493 log_debug(jit, inlining)("%s", msg);
494 if (C->print_intrinsics() || C->print_inlining()) {
495 C->print_inlining_stream()->print("%s", msg);
496 }
497 }
498 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
499 return NULL;
500 }
501
502 bool LibraryCallKit::try_to_inline(int predicate) {
503 // Handle symbolic names for otherwise undistinguished boolean switches:
504 const bool is_store = true;
505 const bool is_compress = true;
506 const bool is_static = true;
507 const bool is_volatile = true;
508
509 if (!jvms()->has_method()) {
510 // Root JVMState has a null method.
511 assert(map()->memory()->Opcode() == Op_Parm, "");
512 // Insert the memory aliasing node
513 set_all_memory(reset_memory());
514 }
515 assert(merged_memory(), "");
516
517
518 switch (intrinsic_id()) {
519 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
520 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
521 case vmIntrinsics::_getClass: return inline_native_getClass();
522
523 case vmIntrinsics::_dsin:
524 case vmIntrinsics::_dcos:
525 case vmIntrinsics::_dtan:
526 case vmIntrinsics::_dabs:
527 case vmIntrinsics::_datan2:
528 case vmIntrinsics::_dsqrt:
529 case vmIntrinsics::_dexp:
530 case vmIntrinsics::_dlog:
531 case vmIntrinsics::_dlog10:
532 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
533
534 case vmIntrinsics::_min:
535 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
536
537 case vmIntrinsics::_notify:
538 case vmIntrinsics::_notifyAll:
539 if (InlineNotify) {
540 return inline_notify(intrinsic_id());
541 }
542 return false;
543
544 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
545 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
546 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
547 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
548 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
549 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
550 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
551 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
552 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
553 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
554 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
555 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
556
557 case vmIntrinsics::_arraycopy: return inline_arraycopy();
558
559 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
560 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
561 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
562 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
563
564 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
565 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
566 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
567 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
568 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
569 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
570 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
571
572 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
573 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
574
575 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
576 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
577 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
578 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
579
580 case vmIntrinsics::_compressStringC:
581 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
582 case vmIntrinsics::_inflateStringC:
583 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
584
585 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
586 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
587 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
588 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
589 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
590 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
591 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
592 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
593 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
594
595 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
596 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
597 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
598 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
599 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
600 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
601 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
602 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
603 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
604
605 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
606 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
607 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
608 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
609 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
610 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
611 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
612 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
613 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
614
615 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
616 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
617 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
618 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
619 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
620 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
621 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
622 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
623 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
624
625 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
626 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
627 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
628 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
629
630 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
631 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
632 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
633 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
634
635 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
636 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
637 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
638 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
639 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
640 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
641 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
642 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
643 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
644
645 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
646 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
647 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
648 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
649 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
650 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
651 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
652 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
653 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
654
655 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
656 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
657 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
658 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
659 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
660 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
661 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
662 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
663 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
664
665 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
666 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
667 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
668 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
669 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
670 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
671 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
672 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
673 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
674
675 case vmIntrinsics::_compareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
676 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
677 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
678 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
679 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
680
681 case vmIntrinsics::_weakCompareAndSetObjectPlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
682 case vmIntrinsics::_weakCompareAndSetObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
683 case vmIntrinsics::_weakCompareAndSetObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
684 case vmIntrinsics::_weakCompareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
685 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
686 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
687 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
688 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
689 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
690 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
691 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
692 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
693 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
694 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
695 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
696 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
697 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
698 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
699 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
700 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
701
702 case vmIntrinsics::_compareAndExchangeObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
703 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
704 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
705 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
706 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
707 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
708 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
709 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
710 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
711 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
712 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
713 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
714 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
715 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
716 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
717
718 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
719 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
720 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
721 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
722
723 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
724 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
725 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
726 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
727 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
728
729 case vmIntrinsics::_loadFence:
730 case vmIntrinsics::_storeFence:
731 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
732
733 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
734
735 case vmIntrinsics::_currentThread: return inline_native_currentThread();
736 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
737
738 #ifdef TRACE_HAVE_INTRINSICS
739 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
740 case vmIntrinsics::_getClassId: return inline_native_classID();
741 case vmIntrinsics::_getBufferWriter: return inline_native_getBufferWriter();
742 #endif
743 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
744 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
745 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
746 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
747 case vmIntrinsics::_getLength: return inline_native_getLength();
748 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
749 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
750 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
751 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
752 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
753 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
754
755 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
756 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
757
758 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
759
760 case vmIntrinsics::_isInstance:
761 case vmIntrinsics::_getModifiers:
762 case vmIntrinsics::_isInterface:
763 case vmIntrinsics::_isArray:
764 case vmIntrinsics::_isPrimitive:
765 case vmIntrinsics::_getSuperclass:
766 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
767
768 case vmIntrinsics::_floatToRawIntBits:
769 case vmIntrinsics::_floatToIntBits:
770 case vmIntrinsics::_intBitsToFloat:
771 case vmIntrinsics::_doubleToRawLongBits:
772 case vmIntrinsics::_doubleToLongBits:
773 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
774
775 case vmIntrinsics::_numberOfLeadingZeros_i:
776 case vmIntrinsics::_numberOfLeadingZeros_l:
777 case vmIntrinsics::_numberOfTrailingZeros_i:
778 case vmIntrinsics::_numberOfTrailingZeros_l:
779 case vmIntrinsics::_bitCount_i:
780 case vmIntrinsics::_bitCount_l:
781 case vmIntrinsics::_reverseBytes_i:
782 case vmIntrinsics::_reverseBytes_l:
783 case vmIntrinsics::_reverseBytes_s:
784 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
785
786 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
787
788 case vmIntrinsics::_Reference_get: return inline_reference_get();
789
790 case vmIntrinsics::_Class_cast: return inline_Class_cast();
791
792 case vmIntrinsics::_aescrypt_encryptBlock:
793 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
794
795 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
796 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
797 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
798
799 case vmIntrinsics::_counterMode_AESCrypt:
800 return inline_counterMode_AESCrypt(intrinsic_id());
801
802 case vmIntrinsics::_sha_implCompress:
803 case vmIntrinsics::_sha2_implCompress:
804 case vmIntrinsics::_sha5_implCompress:
805 return inline_sha_implCompress(intrinsic_id());
806
807 case vmIntrinsics::_digestBase_implCompressMB:
808 return inline_digestBase_implCompressMB(predicate);
809
810 case vmIntrinsics::_multiplyToLen:
811 return inline_multiplyToLen();
812
813 case vmIntrinsics::_squareToLen:
814 return inline_squareToLen();
815
816 case vmIntrinsics::_mulAdd:
817 return inline_mulAdd();
818
819 case vmIntrinsics::_montgomeryMultiply:
820 return inline_montgomeryMultiply();
821 case vmIntrinsics::_montgomerySquare:
822 return inline_montgomerySquare();
823
824 case vmIntrinsics::_vectorizedMismatch:
825 return inline_vectorizedMismatch();
826
827 case vmIntrinsics::_ghash_processBlocks:
828 return inline_ghash_processBlocks();
829
830 case vmIntrinsics::_encodeISOArray:
831 case vmIntrinsics::_encodeByteISOArray:
832 return inline_encodeISOArray();
833
834 case vmIntrinsics::_updateCRC32:
835 return inline_updateCRC32();
836 case vmIntrinsics::_updateBytesCRC32:
837 return inline_updateBytesCRC32();
838 case vmIntrinsics::_updateByteBufferCRC32:
839 return inline_updateByteBufferCRC32();
840
841 case vmIntrinsics::_updateBytesCRC32C:
842 return inline_updateBytesCRC32C();
843 case vmIntrinsics::_updateDirectByteBufferCRC32C:
844 return inline_updateDirectByteBufferCRC32C();
845
846 case vmIntrinsics::_updateBytesAdler32:
847 return inline_updateBytesAdler32();
848 case vmIntrinsics::_updateByteBufferAdler32:
849 return inline_updateByteBufferAdler32();
850
851 case vmIntrinsics::_profileBoolean:
852 return inline_profileBoolean();
853 case vmIntrinsics::_isCompileConstant:
854 return inline_isCompileConstant();
855
856 case vmIntrinsics::_hasNegatives:
857 return inline_hasNegatives();
858
859 case vmIntrinsics::_fmaD:
860 case vmIntrinsics::_fmaF:
861 return inline_fma(intrinsic_id());
862
863 default:
864 // If you get here, it may be that someone has added a new intrinsic
865 // to the list in vmSymbols.hpp without implementing it here.
866 #ifndef PRODUCT
867 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
868 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
869 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
870 }
871 #endif
872 return false;
873 }
874 }
875
876 Node* LibraryCallKit::try_to_predicate(int predicate) {
877 if (!jvms()->has_method()) {
878 // Root JVMState has a null method.
879 assert(map()->memory()->Opcode() == Op_Parm, "");
880 // Insert the memory aliasing node
881 set_all_memory(reset_memory());
882 }
883 assert(merged_memory(), "");
884
885 switch (intrinsic_id()) {
886 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
887 return inline_cipherBlockChaining_AESCrypt_predicate(false);
888 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
889 return inline_cipherBlockChaining_AESCrypt_predicate(true);
890 case vmIntrinsics::_counterMode_AESCrypt:
891 return inline_counterMode_AESCrypt_predicate();
892 case vmIntrinsics::_digestBase_implCompressMB:
893 return inline_digestBase_implCompressMB_predicate(predicate);
894
895 default:
896 // If you get here, it may be that someone has added a new intrinsic
897 // to the list in vmSymbols.hpp without implementing it here.
898 #ifndef PRODUCT
899 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
900 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
901 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
902 }
903 #endif
904 Node* slow_ctl = control();
905 set_control(top()); // No fast path instrinsic
906 return slow_ctl;
907 }
908 }
909
910 //------------------------------set_result-------------------------------
911 // Helper function for finishing intrinsics.
912 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
913 record_for_igvn(region);
914 set_control(_gvn.transform(region));
915 set_result( _gvn.transform(value));
916 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
917 }
918
919 //------------------------------generate_guard---------------------------
920 // Helper function for generating guarded fast-slow graph structures.
921 // The given 'test', if true, guards a slow path. If the test fails
922 // then a fast path can be taken. (We generally hope it fails.)
923 // In all cases, GraphKit::control() is updated to the fast path.
924 // The returned value represents the control for the slow path.
925 // The return value is never 'top'; it is either a valid control
926 // or NULL if it is obvious that the slow path can never be taken.
927 // Also, if region and the slow control are not NULL, the slow edge
928 // is appended to the region.
929 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
930 if (stopped()) {
931 // Already short circuited.
932 return NULL;
933 }
934
935 // Build an if node and its projections.
936 // If test is true we take the slow path, which we assume is uncommon.
937 if (_gvn.type(test) == TypeInt::ZERO) {
938 // The slow branch is never taken. No need to build this guard.
939 return NULL;
940 }
941
942 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
943
944 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
945 if (if_slow == top()) {
946 // The slow branch is never taken. No need to build this guard.
947 return NULL;
948 }
949
950 if (region != NULL)
951 region->add_req(if_slow);
952
953 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
954 set_control(if_fast);
955
956 return if_slow;
957 }
958
959 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
960 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
961 }
962 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
963 return generate_guard(test, region, PROB_FAIR);
964 }
965
966 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
967 Node* *pos_index) {
968 if (stopped())
969 return NULL; // already stopped
970 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
971 return NULL; // index is already adequately typed
972 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
973 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
974 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
975 if (is_neg != NULL && pos_index != NULL) {
976 // Emulate effect of Parse::adjust_map_after_if.
977 Node* ccast = new CastIINode(index, TypeInt::POS);
978 ccast->set_req(0, control());
979 (*pos_index) = _gvn.transform(ccast);
980 }
981 return is_neg;
982 }
983
984 // Make sure that 'position' is a valid limit index, in [0..length].
985 // There are two equivalent plans for checking this:
986 // A. (offset + copyLength) unsigned<= arrayLength
987 // B. offset <= (arrayLength - copyLength)
988 // We require that all of the values above, except for the sum and
989 // difference, are already known to be non-negative.
990 // Plan A is robust in the face of overflow, if offset and copyLength
991 // are both hugely positive.
992 //
993 // Plan B is less direct and intuitive, but it does not overflow at
994 // all, since the difference of two non-negatives is always
995 // representable. Whenever Java methods must perform the equivalent
996 // check they generally use Plan B instead of Plan A.
997 // For the moment we use Plan A.
998 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
999 Node* subseq_length,
1000 Node* array_length,
1001 RegionNode* region) {
1002 if (stopped())
1003 return NULL; // already stopped
1004 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1005 if (zero_offset && subseq_length->eqv_uncast(array_length))
1006 return NULL; // common case of whole-array copy
1007 Node* last = subseq_length;
1008 if (!zero_offset) // last += offset
1009 last = _gvn.transform(new AddINode(last, offset));
1010 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1011 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1012 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1013 return is_over;
1014 }
1015
1016 // Emit range checks for the given String.value byte array
1017 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1018 if (stopped()) {
1019 return; // already stopped
1020 }
1021 RegionNode* bailout = new RegionNode(1);
1022 record_for_igvn(bailout);
1023 if (char_count) {
1024 // Convert char count to byte count
1025 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1026 }
1027
1028 // Offset and count must not be negative
1029 generate_negative_guard(offset, bailout);
1030 generate_negative_guard(count, bailout);
1031 // Offset + count must not exceed length of array
1032 generate_limit_guard(offset, count, load_array_length(array), bailout);
1033
1034 if (bailout->req() > 1) {
1035 PreserveJVMState pjvms(this);
1036 set_control(_gvn.transform(bailout));
1037 uncommon_trap(Deoptimization::Reason_intrinsic,
1038 Deoptimization::Action_maybe_recompile);
1039 }
1040 }
1041
1042 //--------------------------generate_current_thread--------------------
1043 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1044 ciKlass* thread_klass = env()->Thread_klass();
1045 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1046 Node* thread = _gvn.transform(new ThreadLocalNode());
1047 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1048 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1049 tls_output = thread;
1050 return threadObj;
1051 }
1052
1053
1054 //------------------------------make_string_method_node------------------------
1055 // Helper method for String intrinsic functions. This version is called with
1056 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1057 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1058 // containing the lengths of str1 and str2.
1059 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1060 Node* result = NULL;
1061 switch (opcode) {
1062 case Op_StrIndexOf:
1063 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1064 str1_start, cnt1, str2_start, cnt2, ae);
1065 break;
1066 case Op_StrComp:
1067 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1068 str1_start, cnt1, str2_start, cnt2, ae);
1069 break;
1070 case Op_StrEquals:
1071 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1072 // Use the constant length if there is one because optimized match rule may exist.
1073 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1074 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1075 break;
1076 default:
1077 ShouldNotReachHere();
1078 return NULL;
1079 }
1080
1081 // All these intrinsics have checks.
1082 C->set_has_split_ifs(true); // Has chance for split-if optimization
1083
1084 return _gvn.transform(result);
1085 }
1086
1087 //------------------------------inline_string_compareTo------------------------
1088 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1089 Node* arg1 = argument(0);
1090 Node* arg2 = argument(1);
1091
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1101 set_result(result);
1102 return true;
1103 }
1104
1105 //------------------------------inline_string_equals------------------------
1106 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1107 Node* arg1 = argument(0);
1108 Node* arg2 = argument(1);
1109
1110 // paths (plus control) merge
1111 RegionNode* region = new RegionNode(3);
1112 Node* phi = new PhiNode(region, TypeInt::BOOL);
1113
1114 if (!stopped()) {
1115 // Get start addr and length of first argument
1116 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1117 Node* arg1_cnt = load_array_length(arg1);
1118
1119 // Get start addr and length of second argument
1120 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1121 Node* arg2_cnt = load_array_length(arg2);
1122
1123 // Check for arg1_cnt != arg2_cnt
1124 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1125 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1126 Node* if_ne = generate_slow_guard(bol, NULL);
1127 if (if_ne != NULL) {
1128 phi->init_req(2, intcon(0));
1129 region->init_req(2, if_ne);
1130 }
1131
1132 // Check for count == 0 is done by assembler code for StrEquals.
1133
1134 if (!stopped()) {
1135 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1136 phi->init_req(1, equals);
1137 region->init_req(1, control());
1138 }
1139 }
1140
1141 // post merge
1142 set_control(_gvn.transform(region));
1143 record_for_igvn(region);
1144
1145 set_result(_gvn.transform(phi));
1146 return true;
1147 }
1148
1149 //------------------------------inline_array_equals----------------------------
1150 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1151 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1152 Node* arg1 = argument(0);
1153 Node* arg2 = argument(1);
1154
1155 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1156 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1157 return true;
1158 }
1159
1160 //------------------------------inline_hasNegatives------------------------------
1161 bool LibraryCallKit::inline_hasNegatives() {
1162 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1163 return false;
1164 }
1165
1166 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1167 // no receiver since it is static method
1168 Node* ba = argument(0);
1169 Node* offset = argument(1);
1170 Node* len = argument(2);
1171
1172 // Range checks
1173 generate_string_range_check(ba, offset, len, false);
1174 if (stopped()) {
1175 return true;
1176 }
1177 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1178 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1179 set_result(_gvn.transform(result));
1180 return true;
1181 }
1182
1183 bool LibraryCallKit::inline_preconditions_checkIndex() {
1184 Node* index = argument(0);
1185 Node* length = argument(1);
1186 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1187 return false;
1188 }
1189
1190 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1191 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1192
1193 {
1194 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1195 uncommon_trap(Deoptimization::Reason_intrinsic,
1196 Deoptimization::Action_make_not_entrant);
1197 }
1198
1199 if (stopped()) {
1200 return false;
1201 }
1202
1203 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1204 BoolTest::mask btest = BoolTest::lt;
1205 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1206 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1207 _gvn.set_type(rc, rc->Value(&_gvn));
1208 if (!rc_bool->is_Con()) {
1209 record_for_igvn(rc);
1210 }
1211 set_control(_gvn.transform(new IfTrueNode(rc)));
1212 {
1213 PreserveJVMState pjvms(this);
1214 set_control(_gvn.transform(new IfFalseNode(rc)));
1215 uncommon_trap(Deoptimization::Reason_range_check,
1216 Deoptimization::Action_make_not_entrant);
1217 }
1218
1219 if (stopped()) {
1220 return false;
1221 }
1222
1223 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1224 result->set_req(0, control());
1225 result = _gvn.transform(result);
1226 set_result(result);
1227 replace_in_map(index, result);
1228 return true;
1229 }
1230
1231 //------------------------------inline_string_indexOf------------------------
1232 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1233 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1234 return false;
1235 }
1236 Node* src = argument(0);
1237 Node* tgt = argument(1);
1238
1239 // Make the merge point
1240 RegionNode* result_rgn = new RegionNode(4);
1241 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1242
1243 // Get start addr and length of source string
1244 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1245 Node* src_count = load_array_length(src);
1246
1247 // Get start addr and length of substring
1248 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1249 Node* tgt_count = load_array_length(tgt);
1250
1251 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1252 // Divide src size by 2 if String is UTF16 encoded
1253 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1254 }
1255 if (ae == StrIntrinsicNode::UU) {
1256 // Divide substring size by 2 if String is UTF16 encoded
1257 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1258 }
1259
1260 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1261 if (result != NULL) {
1262 result_phi->init_req(3, result);
1263 result_rgn->init_req(3, control());
1264 }
1265 set_control(_gvn.transform(result_rgn));
1266 record_for_igvn(result_rgn);
1267 set_result(_gvn.transform(result_phi));
1268
1269 return true;
1270 }
1271
1272 //-----------------------------inline_string_indexOf-----------------------
1273 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1274 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1275 return false;
1276 }
1277 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1278 return false;
1279 }
1280 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1281 Node* src = argument(0); // byte[]
1282 Node* src_count = argument(1); // char count
1283 Node* tgt = argument(2); // byte[]
1284 Node* tgt_count = argument(3); // char count
1285 Node* from_index = argument(4); // char index
1286
1287 // Multiply byte array index by 2 if String is UTF16 encoded
1288 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1289 src_count = _gvn.transform(new SubINode(src_count, from_index));
1290 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1291 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1292
1293 // Range checks
1294 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1295 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1296 if (stopped()) {
1297 return true;
1298 }
1299
1300 RegionNode* region = new RegionNode(5);
1301 Node* phi = new PhiNode(region, TypeInt::INT);
1302
1303 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1304 if (result != NULL) {
1305 // The result is index relative to from_index if substring was found, -1 otherwise.
1306 // Generate code which will fold into cmove.
1307 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1308 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1309
1310 Node* if_lt = generate_slow_guard(bol, NULL);
1311 if (if_lt != NULL) {
1312 // result == -1
1313 phi->init_req(3, result);
1314 region->init_req(3, if_lt);
1315 }
1316 if (!stopped()) {
1317 result = _gvn.transform(new AddINode(result, from_index));
1318 phi->init_req(4, result);
1319 region->init_req(4, control());
1320 }
1321 }
1322
1323 set_control(_gvn.transform(region));
1324 record_for_igvn(region);
1325 set_result(_gvn.transform(phi));
1326
1327 return true;
1328 }
1329
1330 // Create StrIndexOfNode with fast path checks
1331 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1332 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1333 // Check for substr count > string count
1334 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1335 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1336 Node* if_gt = generate_slow_guard(bol, NULL);
1337 if (if_gt != NULL) {
1338 phi->init_req(1, intcon(-1));
1339 region->init_req(1, if_gt);
1340 }
1341 if (!stopped()) {
1342 // Check for substr count == 0
1343 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1344 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1345 Node* if_zero = generate_slow_guard(bol, NULL);
1346 if (if_zero != NULL) {
1347 phi->init_req(2, intcon(0));
1348 region->init_req(2, if_zero);
1349 }
1350 }
1351 if (!stopped()) {
1352 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1353 }
1354 return NULL;
1355 }
1356
1357 //-----------------------------inline_string_indexOfChar-----------------------
1358 bool LibraryCallKit::inline_string_indexOfChar() {
1359 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1360 return false;
1361 }
1362 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1363 return false;
1364 }
1365 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1366 Node* src = argument(0); // byte[]
1367 Node* tgt = argument(1); // tgt is int ch
1368 Node* from_index = argument(2);
1369 Node* max = argument(3);
1370
1371 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1372 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1373 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1374
1375 // Range checks
1376 generate_string_range_check(src, src_offset, src_count, true);
1377 if (stopped()) {
1378 return true;
1379 }
1380
1381 RegionNode* region = new RegionNode(3);
1382 Node* phi = new PhiNode(region, TypeInt::INT);
1383
1384 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1385 C->set_has_split_ifs(true); // Has chance for split-if optimization
1386 _gvn.transform(result);
1387
1388 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1389 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1390
1391 Node* if_lt = generate_slow_guard(bol, NULL);
1392 if (if_lt != NULL) {
1393 // result == -1
1394 phi->init_req(2, result);
1395 region->init_req(2, if_lt);
1396 }
1397 if (!stopped()) {
1398 result = _gvn.transform(new AddINode(result, from_index));
1399 phi->init_req(1, result);
1400 region->init_req(1, control());
1401 }
1402 set_control(_gvn.transform(region));
1403 record_for_igvn(region);
1404 set_result(_gvn.transform(phi));
1405
1406 return true;
1407 }
1408 //---------------------------inline_string_copy---------------------
1409 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1410 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1411 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1412 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1413 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1414 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1415 bool LibraryCallKit::inline_string_copy(bool compress) {
1416 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1417 return false;
1418 }
1419 int nargs = 5; // 2 oops, 3 ints
1420 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1421
1422 Node* src = argument(0);
1423 Node* src_offset = argument(1);
1424 Node* dst = argument(2);
1425 Node* dst_offset = argument(3);
1426 Node* length = argument(4);
1427
1428 // Check for allocation before we add nodes that would confuse
1429 // tightly_coupled_allocation()
1430 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1431
1432 // Figure out the size and type of the elements we will be copying.
1433 const Type* src_type = src->Value(&_gvn);
1434 const Type* dst_type = dst->Value(&_gvn);
1435 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1436 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1437 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1438 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1439 "Unsupported array types for inline_string_copy");
1440
1441 // Convert char[] offsets to byte[] offsets
1442 bool convert_src = (compress && src_elem == T_BYTE);
1443 bool convert_dst = (!compress && dst_elem == T_BYTE);
1444 if (convert_src) {
1445 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1446 } else if (convert_dst) {
1447 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1448 }
1449
1450 // Range checks
1451 generate_string_range_check(src, src_offset, length, convert_src);
1452 generate_string_range_check(dst, dst_offset, length, convert_dst);
1453 if (stopped()) {
1454 return true;
1455 }
1456
1457 Node* src_start = array_element_address(src, src_offset, src_elem);
1458 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1459 // 'src_start' points to src array + scaled offset
1460 // 'dst_start' points to dst array + scaled offset
1461 Node* count = NULL;
1462 if (compress) {
1463 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1464 } else {
1465 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1466 }
1467
1468 if (alloc != NULL) {
1469 if (alloc->maybe_set_complete(&_gvn)) {
1470 // "You break it, you buy it."
1471 InitializeNode* init = alloc->initialization();
1472 assert(init->is_complete(), "we just did this");
1473 init->set_complete_with_arraycopy();
1474 assert(dst->is_CheckCastPP(), "sanity");
1475 assert(dst->in(0)->in(0) == init, "dest pinned");
1476 }
1477 // Do not let stores that initialize this object be reordered with
1478 // a subsequent store that would make this object accessible by
1479 // other threads.
1480 // Record what AllocateNode this StoreStore protects so that
1481 // escape analysis can go from the MemBarStoreStoreNode to the
1482 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1483 // based on the escape status of the AllocateNode.
1484 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1485 }
1486 if (compress) {
1487 set_result(_gvn.transform(count));
1488 }
1489 return true;
1490 }
1491
1492 #ifdef _LP64
1493 #define XTOP ,top() /*additional argument*/
1494 #else //_LP64
1495 #define XTOP /*no additional argument*/
1496 #endif //_LP64
1497
1498 //------------------------inline_string_toBytesU--------------------------
1499 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1500 bool LibraryCallKit::inline_string_toBytesU() {
1501 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1502 return false;
1503 }
1504 // Get the arguments.
1505 Node* value = argument(0);
1506 Node* offset = argument(1);
1507 Node* length = argument(2);
1508
1509 Node* newcopy = NULL;
1510
1511 // Set the original stack and the reexecute bit for the interpreter to reexecute
1512 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1513 { PreserveReexecuteState preexecs(this);
1514 jvms()->set_should_reexecute(true);
1515
1516 // Check if a null path was taken unconditionally.
1517 value = null_check(value);
1518
1519 RegionNode* bailout = new RegionNode(1);
1520 record_for_igvn(bailout);
1521
1522 // Range checks
1523 generate_negative_guard(offset, bailout);
1524 generate_negative_guard(length, bailout);
1525 generate_limit_guard(offset, length, load_array_length(value), bailout);
1526 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1527 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1528
1529 if (bailout->req() > 1) {
1530 PreserveJVMState pjvms(this);
1531 set_control(_gvn.transform(bailout));
1532 uncommon_trap(Deoptimization::Reason_intrinsic,
1533 Deoptimization::Action_maybe_recompile);
1534 }
1535 if (stopped()) {
1536 return true;
1537 }
1538
1539 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1540 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1541 newcopy = new_array(klass_node, size, 0); // no arguments to push
1542 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1543
1544 // Calculate starting addresses.
1545 Node* src_start = array_element_address(value, offset, T_CHAR);
1546 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1547
1548 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1549 const TypeInt* toffset = gvn().type(offset)->is_int();
1550 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1551
1552 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1553 const char* copyfunc_name = "arraycopy";
1554 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1555 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1556 OptoRuntime::fast_arraycopy_Type(),
1557 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1558 src_start, dst_start, ConvI2X(length) XTOP);
1559 // Do not let reads from the cloned object float above the arraycopy.
1560 if (alloc != NULL) {
1561 if (alloc->maybe_set_complete(&_gvn)) {
1562 // "You break it, you buy it."
1563 InitializeNode* init = alloc->initialization();
1564 assert(init->is_complete(), "we just did this");
1565 init->set_complete_with_arraycopy();
1566 assert(newcopy->is_CheckCastPP(), "sanity");
1567 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1568 }
1569 // Do not let stores that initialize this object be reordered with
1570 // a subsequent store that would make this object accessible by
1571 // other threads.
1572 // Record what AllocateNode this StoreStore protects so that
1573 // escape analysis can go from the MemBarStoreStoreNode to the
1574 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1575 // based on the escape status of the AllocateNode.
1576 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1577 } else {
1578 insert_mem_bar(Op_MemBarCPUOrder);
1579 }
1580 } // original reexecute is set back here
1581
1582 C->set_has_split_ifs(true); // Has chance for split-if optimization
1583 if (!stopped()) {
1584 set_result(newcopy);
1585 }
1586 return true;
1587 }
1588
1589 //------------------------inline_string_getCharsU--------------------------
1590 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1591 bool LibraryCallKit::inline_string_getCharsU() {
1592 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1593 return false;
1594 }
1595
1596 // Get the arguments.
1597 Node* src = argument(0);
1598 Node* src_begin = argument(1);
1599 Node* src_end = argument(2); // exclusive offset (i < src_end)
1600 Node* dst = argument(3);
1601 Node* dst_begin = argument(4);
1602
1603 // Check for allocation before we add nodes that would confuse
1604 // tightly_coupled_allocation()
1605 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1606
1607 // Check if a null path was taken unconditionally.
1608 src = null_check(src);
1609 dst = null_check(dst);
1610 if (stopped()) {
1611 return true;
1612 }
1613
1614 // Get length and convert char[] offset to byte[] offset
1615 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1616 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1617
1618 // Range checks
1619 generate_string_range_check(src, src_begin, length, true);
1620 generate_string_range_check(dst, dst_begin, length, false);
1621 if (stopped()) {
1622 return true;
1623 }
1624
1625 if (!stopped()) {
1626 // Calculate starting addresses.
1627 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1628 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1629
1630 // Check if array addresses are aligned to HeapWordSize
1631 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1632 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1633 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1634 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1635
1636 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1637 const char* copyfunc_name = "arraycopy";
1638 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1639 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1640 OptoRuntime::fast_arraycopy_Type(),
1641 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1642 src_start, dst_start, ConvI2X(length) XTOP);
1643 // Do not let reads from the cloned object float above the arraycopy.
1644 if (alloc != NULL) {
1645 if (alloc->maybe_set_complete(&_gvn)) {
1646 // "You break it, you buy it."
1647 InitializeNode* init = alloc->initialization();
1648 assert(init->is_complete(), "we just did this");
1649 init->set_complete_with_arraycopy();
1650 assert(dst->is_CheckCastPP(), "sanity");
1651 assert(dst->in(0)->in(0) == init, "dest pinned");
1652 }
1653 // Do not let stores that initialize this object be reordered with
1654 // a subsequent store that would make this object accessible by
1655 // other threads.
1656 // Record what AllocateNode this StoreStore protects so that
1657 // escape analysis can go from the MemBarStoreStoreNode to the
1658 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1659 // based on the escape status of the AllocateNode.
1660 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1661 } else {
1662 insert_mem_bar(Op_MemBarCPUOrder);
1663 }
1664 }
1665
1666 C->set_has_split_ifs(true); // Has chance for split-if optimization
1667 return true;
1668 }
1669
1670 //----------------------inline_string_char_access----------------------------
1671 // Store/Load char to/from byte[] array.
1672 // static void StringUTF16.putChar(byte[] val, int index, int c)
1673 // static char StringUTF16.getChar(byte[] val, int index)
1674 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1675 Node* value = argument(0);
1676 Node* index = argument(1);
1677 Node* ch = is_store ? argument(2) : NULL;
1678
1679 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1680 // correctly requires matched array shapes.
1681 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1682 "sanity: byte[] and char[] bases agree");
1683 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1684 "sanity: byte[] and char[] scales agree");
1685
1686 // Bail when getChar over constants is requested: constant folding would
1687 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1688 // Java method would constant fold nicely instead.
1689 if (!is_store && value->is_Con() && index->is_Con()) {
1690 return false;
1691 }
1692
1693 Node* adr = array_element_address(value, index, T_CHAR);
1694 if (adr->is_top()) {
1695 return false;
1696 }
1697 if (is_store) {
1698 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1699 false, false, true /* mismatched */);
1700 } else {
1701 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1702 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1703 set_result(ch);
1704 }
1705 return true;
1706 }
1707
1708 //--------------------------round_double_node--------------------------------
1709 // Round a double node if necessary.
1710 Node* LibraryCallKit::round_double_node(Node* n) {
1711 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1712 n = _gvn.transform(new RoundDoubleNode(0, n));
1713 return n;
1714 }
1715
1716 //------------------------------inline_math-----------------------------------
1717 // public static double Math.abs(double)
1718 // public static double Math.sqrt(double)
1719 // public static double Math.log(double)
1720 // public static double Math.log10(double)
1721 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1722 Node* arg = round_double_node(argument(0));
1723 Node* n = NULL;
1724 switch (id) {
1725 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1726 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1727 default: fatal_unexpected_iid(id); break;
1728 }
1729 set_result(_gvn.transform(n));
1730 return true;
1731 }
1732
1733 //------------------------------runtime_math-----------------------------
1734 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1735 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1736 "must be (DD)D or (D)D type");
1737
1738 // Inputs
1739 Node* a = round_double_node(argument(0));
1740 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1741
1742 const TypePtr* no_memory_effects = NULL;
1743 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1744 no_memory_effects,
1745 a, top(), b, b ? top() : NULL);
1746 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1747 #ifdef ASSERT
1748 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1749 assert(value_top == top(), "second value must be top");
1750 #endif
1751
1752 set_result(value);
1753 return true;
1754 }
1755
1756 //------------------------------inline_math_native-----------------------------
1757 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1758 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1759 switch (id) {
1760 // These intrinsics are not properly supported on all hardware
1761 case vmIntrinsics::_dsin:
1762 return StubRoutines::dsin() != NULL ?
1763 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1764 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1765 case vmIntrinsics::_dcos:
1766 return StubRoutines::dcos() != NULL ?
1767 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1768 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1769 case vmIntrinsics::_dtan:
1770 return StubRoutines::dtan() != NULL ?
1771 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1772 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1773 case vmIntrinsics::_dlog:
1774 return StubRoutines::dlog() != NULL ?
1775 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1776 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1777 case vmIntrinsics::_dlog10:
1778 return StubRoutines::dlog10() != NULL ?
1779 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1780 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1781
1782 // These intrinsics are supported on all hardware
1783 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1784 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1785
1786 case vmIntrinsics::_dexp:
1787 return StubRoutines::dexp() != NULL ?
1788 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1789 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1790 case vmIntrinsics::_dpow:
1791 return StubRoutines::dpow() != NULL ?
1792 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1793 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1794 #undef FN_PTR
1795
1796 // These intrinsics are not yet correctly implemented
1797 case vmIntrinsics::_datan2:
1798 return false;
1799
1800 default:
1801 fatal_unexpected_iid(id);
1802 return false;
1803 }
1804 }
1805
1806 static bool is_simple_name(Node* n) {
1807 return (n->req() == 1 // constant
1808 || (n->is_Type() && n->as_Type()->type()->singleton())
1809 || n->is_Proj() // parameter or return value
1810 || n->is_Phi() // local of some sort
1811 );
1812 }
1813
1814 //----------------------------inline_notify-----------------------------------*
1815 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1816 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1817 address func;
1818 if (id == vmIntrinsics::_notify) {
1819 func = OptoRuntime::monitor_notify_Java();
1820 } else {
1821 func = OptoRuntime::monitor_notifyAll_Java();
1822 }
1823 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1824 make_slow_call_ex(call, env()->Throwable_klass(), false);
1825 return true;
1826 }
1827
1828
1829 //----------------------------inline_min_max-----------------------------------
1830 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1831 set_result(generate_min_max(id, argument(0), argument(1)));
1832 return true;
1833 }
1834
1835 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1836 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1837 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1838 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1839 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1840
1841 {
1842 PreserveJVMState pjvms(this);
1843 PreserveReexecuteState preexecs(this);
1844 jvms()->set_should_reexecute(true);
1845
1846 set_control(slow_path);
1847 set_i_o(i_o());
1848
1849 uncommon_trap(Deoptimization::Reason_intrinsic,
1850 Deoptimization::Action_none);
1851 }
1852
1853 set_control(fast_path);
1854 set_result(math);
1855 }
1856
1857 template <typename OverflowOp>
1858 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1859 typedef typename OverflowOp::MathOp MathOp;
1860
1861 MathOp* mathOp = new MathOp(arg1, arg2);
1862 Node* operation = _gvn.transform( mathOp );
1863 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1864 inline_math_mathExact(operation, ofcheck);
1865 return true;
1866 }
1867
1868 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1869 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1870 }
1871
1872 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1873 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1874 }
1875
1876 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1877 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1878 }
1879
1880 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1881 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1882 }
1883
1884 bool LibraryCallKit::inline_math_negateExactI() {
1885 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1886 }
1887
1888 bool LibraryCallKit::inline_math_negateExactL() {
1889 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1890 }
1891
1892 bool LibraryCallKit::inline_math_multiplyExactI() {
1893 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1894 }
1895
1896 bool LibraryCallKit::inline_math_multiplyExactL() {
1897 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1898 }
1899
1900 Node*
1901 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1902 // These are the candidate return value:
1903 Node* xvalue = x0;
1904 Node* yvalue = y0;
1905
1906 if (xvalue == yvalue) {
1907 return xvalue;
1908 }
1909
1910 bool want_max = (id == vmIntrinsics::_max);
1911
1912 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1913 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1914 if (txvalue == NULL || tyvalue == NULL) return top();
1915 // This is not really necessary, but it is consistent with a
1916 // hypothetical MaxINode::Value method:
1917 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1918
1919 // %%% This folding logic should (ideally) be in a different place.
1920 // Some should be inside IfNode, and there to be a more reliable
1921 // transformation of ?: style patterns into cmoves. We also want
1922 // more powerful optimizations around cmove and min/max.
1923
1924 // Try to find a dominating comparison of these guys.
1925 // It can simplify the index computation for Arrays.copyOf
1926 // and similar uses of System.arraycopy.
1927 // First, compute the normalized version of CmpI(x, y).
1928 int cmp_op = Op_CmpI;
1929 Node* xkey = xvalue;
1930 Node* ykey = yvalue;
1931 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1932 if (ideal_cmpxy->is_Cmp()) {
1933 // E.g., if we have CmpI(length - offset, count),
1934 // it might idealize to CmpI(length, count + offset)
1935 cmp_op = ideal_cmpxy->Opcode();
1936 xkey = ideal_cmpxy->in(1);
1937 ykey = ideal_cmpxy->in(2);
1938 }
1939
1940 // Start by locating any relevant comparisons.
1941 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1942 Node* cmpxy = NULL;
1943 Node* cmpyx = NULL;
1944 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1945 Node* cmp = start_from->fast_out(k);
1946 if (cmp->outcnt() > 0 && // must have prior uses
1947 cmp->in(0) == NULL && // must be context-independent
1948 cmp->Opcode() == cmp_op) { // right kind of compare
1949 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1950 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1951 }
1952 }
1953
1954 const int NCMPS = 2;
1955 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1956 int cmpn;
1957 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1958 if (cmps[cmpn] != NULL) break; // find a result
1959 }
1960 if (cmpn < NCMPS) {
1961 // Look for a dominating test that tells us the min and max.
1962 int depth = 0; // Limit search depth for speed
1963 Node* dom = control();
1964 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1965 if (++depth >= 100) break;
1966 Node* ifproj = dom;
1967 if (!ifproj->is_Proj()) continue;
1968 Node* iff = ifproj->in(0);
1969 if (!iff->is_If()) continue;
1970 Node* bol = iff->in(1);
1971 if (!bol->is_Bool()) continue;
1972 Node* cmp = bol->in(1);
1973 if (cmp == NULL) continue;
1974 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1975 if (cmps[cmpn] == cmp) break;
1976 if (cmpn == NCMPS) continue;
1977 BoolTest::mask btest = bol->as_Bool()->_test._test;
1978 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1979 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1980 // At this point, we know that 'x btest y' is true.
1981 switch (btest) {
1982 case BoolTest::eq:
1983 // They are proven equal, so we can collapse the min/max.
1984 // Either value is the answer. Choose the simpler.
1985 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1986 return yvalue;
1987 return xvalue;
1988 case BoolTest::lt: // x < y
1989 case BoolTest::le: // x <= y
1990 return (want_max ? yvalue : xvalue);
1991 case BoolTest::gt: // x > y
1992 case BoolTest::ge: // x >= y
1993 return (want_max ? xvalue : yvalue);
1994 }
1995 }
1996 }
1997
1998 // We failed to find a dominating test.
1999 // Let's pick a test that might GVN with prior tests.
2000 Node* best_bol = NULL;
2001 BoolTest::mask best_btest = BoolTest::illegal;
2002 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2003 Node* cmp = cmps[cmpn];
2004 if (cmp == NULL) continue;
2005 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2006 Node* bol = cmp->fast_out(j);
2007 if (!bol->is_Bool()) continue;
2008 BoolTest::mask btest = bol->as_Bool()->_test._test;
2009 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2010 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2011 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2012 best_bol = bol->as_Bool();
2013 best_btest = btest;
2014 }
2015 }
2016 }
2017
2018 Node* answer_if_true = NULL;
2019 Node* answer_if_false = NULL;
2020 switch (best_btest) {
2021 default:
2022 if (cmpxy == NULL)
2023 cmpxy = ideal_cmpxy;
2024 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2025 // and fall through:
2026 case BoolTest::lt: // x < y
2027 case BoolTest::le: // x <= y
2028 answer_if_true = (want_max ? yvalue : xvalue);
2029 answer_if_false = (want_max ? xvalue : yvalue);
2030 break;
2031 case BoolTest::gt: // x > y
2032 case BoolTest::ge: // x >= y
2033 answer_if_true = (want_max ? xvalue : yvalue);
2034 answer_if_false = (want_max ? yvalue : xvalue);
2035 break;
2036 }
2037
2038 jint hi, lo;
2039 if (want_max) {
2040 // We can sharpen the minimum.
2041 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2042 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2043 } else {
2044 // We can sharpen the maximum.
2045 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2046 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2047 }
2048
2049 // Use a flow-free graph structure, to avoid creating excess control edges
2050 // which could hinder other optimizations.
2051 // Since Math.min/max is often used with arraycopy, we want
2052 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2053 Node* cmov = CMoveNode::make(NULL, best_bol,
2054 answer_if_false, answer_if_true,
2055 TypeInt::make(lo, hi, widen));
2056
2057 return _gvn.transform(cmov);
2058
2059 /*
2060 // This is not as desirable as it may seem, since Min and Max
2061 // nodes do not have a full set of optimizations.
2062 // And they would interfere, anyway, with 'if' optimizations
2063 // and with CMoveI canonical forms.
2064 switch (id) {
2065 case vmIntrinsics::_min:
2066 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2067 case vmIntrinsics::_max:
2068 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2069 default:
2070 ShouldNotReachHere();
2071 }
2072 */
2073 }
2074
2075 inline int
2076 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2077 const TypePtr* base_type = TypePtr::NULL_PTR;
2078 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2079 if (base_type == NULL) {
2080 // Unknown type.
2081 return Type::AnyPtr;
2082 } else if (base_type == TypePtr::NULL_PTR) {
2083 // Since this is a NULL+long form, we have to switch to a rawptr.
2084 base = _gvn.transform(new CastX2PNode(offset));
2085 offset = MakeConX(0);
2086 return Type::RawPtr;
2087 } else if (base_type->base() == Type::RawPtr) {
2088 return Type::RawPtr;
2089 } else if (base_type->isa_oopptr()) {
2090 // Base is never null => always a heap address.
2091 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2092 return Type::OopPtr;
2093 }
2094 // Offset is small => always a heap address.
2095 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2096 if (offset_type != NULL &&
2097 base_type->offset() == 0 && // (should always be?)
2098 offset_type->_lo >= 0 &&
2099 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2100 return Type::OopPtr;
2101 } else if (type == T_OBJECT) {
2102 // off heap access to an oop doesn't make any sense. Has to be on
2103 // heap.
2104 return Type::OopPtr;
2105 }
2106 // Otherwise, it might either be oop+off or NULL+addr.
2107 return Type::AnyPtr;
2108 } else {
2109 // No information:
2110 return Type::AnyPtr;
2111 }
2112 }
2113
2114 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type) {
2115 Node* uncasted_base = base;
2116 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2117 if (kind == Type::RawPtr) {
2118 return basic_plus_adr(top(), uncasted_base, offset);
2119 } else if (kind == Type::AnyPtr) {
2120 assert(base == uncasted_base, "unexpected base change");
2121 // We don't know if it's an on heap or off heap access. Fall back
2122 // to raw memory access.
2123 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2124 return basic_plus_adr(top(), raw, offset);
2125 } else {
2126 assert(base == uncasted_base, "unexpected base change");
2127 // We know it's an on heap access so base can't be null
2128 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2129 base = must_be_not_null(base, true);
2130 }
2131 return basic_plus_adr(base, offset);
2132 }
2133 }
2134
2135 //--------------------------inline_number_methods-----------------------------
2136 // inline int Integer.numberOfLeadingZeros(int)
2137 // inline int Long.numberOfLeadingZeros(long)
2138 //
2139 // inline int Integer.numberOfTrailingZeros(int)
2140 // inline int Long.numberOfTrailingZeros(long)
2141 //
2142 // inline int Integer.bitCount(int)
2143 // inline int Long.bitCount(long)
2144 //
2145 // inline char Character.reverseBytes(char)
2146 // inline short Short.reverseBytes(short)
2147 // inline int Integer.reverseBytes(int)
2148 // inline long Long.reverseBytes(long)
2149 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2150 Node* arg = argument(0);
2151 Node* n = NULL;
2152 switch (id) {
2153 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2154 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2155 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2156 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2157 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2158 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2159 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2160 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2161 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2162 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2163 default: fatal_unexpected_iid(id); break;
2164 }
2165 set_result(_gvn.transform(n));
2166 return true;
2167 }
2168
2169 //----------------------------inline_unsafe_access----------------------------
2170
2171 // Helper that guards and inserts a pre-barrier.
2172 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2173 Node* pre_val, bool need_mem_bar) {
2174 // We could be accessing the referent field of a reference object. If so, when G1
2175 // is enabled, we need to log the value in the referent field in an SATB buffer.
2176 // This routine performs some compile time filters and generates suitable
2177 // runtime filters that guard the pre-barrier code.
2178 // Also add memory barrier for non volatile load from the referent field
2179 // to prevent commoning of loads across safepoint.
2180 if (!UseG1GC && !need_mem_bar)
2181 return;
2182
2183 // Some compile time checks.
2184
2185 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2186 const TypeX* otype = offset->find_intptr_t_type();
2187 if (otype != NULL && otype->is_con() &&
2188 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2189 // Constant offset but not the reference_offset so just return
2190 return;
2191 }
2192
2193 // We only need to generate the runtime guards for instances.
2194 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2195 if (btype != NULL) {
2196 if (btype->isa_aryptr()) {
2197 // Array type so nothing to do
2198 return;
2199 }
2200
2201 const TypeInstPtr* itype = btype->isa_instptr();
2202 if (itype != NULL) {
2203 // Can the klass of base_oop be statically determined to be
2204 // _not_ a sub-class of Reference and _not_ Object?
2205 ciKlass* klass = itype->klass();
2206 if ( klass->is_loaded() &&
2207 !klass->is_subtype_of(env()->Reference_klass()) &&
2208 !env()->Object_klass()->is_subtype_of(klass)) {
2209 return;
2210 }
2211 }
2212 }
2213
2214 // The compile time filters did not reject base_oop/offset so
2215 // we need to generate the following runtime filters
2216 //
2217 // if (offset == java_lang_ref_Reference::_reference_offset) {
2218 // if (instance_of(base, java.lang.ref.Reference)) {
2219 // pre_barrier(_, pre_val, ...);
2220 // }
2221 // }
2222
2223 float likely = PROB_LIKELY( 0.999);
2224 float unlikely = PROB_UNLIKELY(0.999);
2225
2226 IdealKit ideal(this);
2227 #define __ ideal.
2228
2229 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2230
2231 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2232 // Update graphKit memory and control from IdealKit.
2233 sync_kit(ideal);
2234
2235 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2236 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2237
2238 // Update IdealKit memory and control from graphKit.
2239 __ sync_kit(this);
2240
2241 Node* one = __ ConI(1);
2242 // is_instof == 0 if base_oop == NULL
2243 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2244
2245 // Update graphKit from IdeakKit.
2246 sync_kit(ideal);
2247
2248 // Use the pre-barrier to record the value in the referent field
2249 pre_barrier(false /* do_load */,
2250 __ ctrl(),
2251 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2252 pre_val /* pre_val */,
2253 T_OBJECT);
2254 if (need_mem_bar) {
2255 // Add memory barrier to prevent commoning reads from this field
2256 // across safepoint since GC can change its value.
2257 insert_mem_bar(Op_MemBarCPUOrder);
2258 }
2259 // Update IdealKit from graphKit.
2260 __ sync_kit(this);
2261
2262 } __ end_if(); // _ref_type != ref_none
2263 } __ end_if(); // offset == referent_offset
2264
2265 // Final sync IdealKit and GraphKit.
2266 final_sync(ideal);
2267 #undef __
2268 }
2269
2270
2271 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2272 // Attempt to infer a sharper value type from the offset and base type.
2273 ciKlass* sharpened_klass = NULL;
2274
2275 // See if it is an instance field, with an object type.
2276 if (alias_type->field() != NULL) {
2277 if (alias_type->field()->type()->is_klass()) {
2278 sharpened_klass = alias_type->field()->type()->as_klass();
2279 }
2280 }
2281
2282 // See if it is a narrow oop array.
2283 if (adr_type->isa_aryptr()) {
2284 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2285 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2286 if (elem_type != NULL) {
2287 sharpened_klass = elem_type->klass();
2288 }
2289 }
2290 }
2291
2292 // The sharpened class might be unloaded if there is no class loader
2293 // contraint in place.
2294 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2295 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2296
2297 #ifndef PRODUCT
2298 if (C->print_intrinsics() || C->print_inlining()) {
2299 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2300 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2301 }
2302 #endif
2303 // Sharpen the value type.
2304 return tjp;
2305 }
2306 return NULL;
2307 }
2308
2309 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2310 if (callee()->is_static()) return false; // caller must have the capability!
2311 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2312 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2313 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2314
2315 #ifndef PRODUCT
2316 {
2317 ResourceMark rm;
2318 // Check the signatures.
2319 ciSignature* sig = callee()->signature();
2320 #ifdef ASSERT
2321 if (!is_store) {
2322 // Object getObject(Object base, int/long offset), etc.
2323 BasicType rtype = sig->return_type()->basic_type();
2324 assert(rtype == type, "getter must return the expected value");
2325 assert(sig->count() == 2, "oop getter has 2 arguments");
2326 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2327 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2328 } else {
2329 // void putObject(Object base, int/long offset, Object x), etc.
2330 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2331 assert(sig->count() == 3, "oop putter has 3 arguments");
2332 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2333 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2334 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2335 assert(vtype == type, "putter must accept the expected value");
2336 }
2337 #endif // ASSERT
2338 }
2339 #endif //PRODUCT
2340
2341 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2342
2343 Node* receiver = argument(0); // type: oop
2344
2345 // Build address expression.
2346 Node* adr;
2347 Node* heap_base_oop = top();
2348 Node* offset = top();
2349 Node* val;
2350
2351 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2352 Node* base = argument(1); // type: oop
2353 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2354 offset = argument(2); // type: long
2355 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2356 // to be plain byte offsets, which are also the same as those accepted
2357 // by oopDesc::field_base.
2358 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2359 "fieldOffset must be byte-scaled");
2360 // 32-bit machines ignore the high half!
2361 offset = ConvL2X(offset);
2362 adr = make_unsafe_address(base, offset, type);
2363 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2364 heap_base_oop = base;
2365 } else if (type == T_OBJECT) {
2366 return false; // off-heap oop accesses are not supported
2367 }
2368
2369 // Can base be NULL? Otherwise, always on-heap access.
2370 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2371
2372 val = is_store ? argument(4) : NULL;
2373
2374 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2375
2376 // Try to categorize the address.
2377 Compile::AliasType* alias_type = C->alias_type(adr_type);
2378 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2379
2380 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2381 alias_type->adr_type() == TypeAryPtr::RANGE) {
2382 return false; // not supported
2383 }
2384
2385 bool mismatched = false;
2386 BasicType bt = alias_type->basic_type();
2387 if (bt != T_ILLEGAL) {
2388 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2389 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2390 // Alias type doesn't differentiate between byte[] and boolean[]).
2391 // Use address type to get the element type.
2392 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2393 }
2394 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2395 // accessing an array field with getObject is not a mismatch
2396 bt = T_OBJECT;
2397 }
2398 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2399 // Don't intrinsify mismatched object accesses
2400 return false;
2401 }
2402 mismatched = (bt != type);
2403 } else if (alias_type->adr_type()->isa_oopptr()) {
2404 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2405 }
2406
2407 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2408
2409 // First guess at the value type.
2410 const Type *value_type = Type::get_const_basic_type(type);
2411
2412 // We will need memory barriers unless we can determine a unique
2413 // alias category for this reference. (Note: If for some reason
2414 // the barriers get omitted and the unsafe reference begins to "pollute"
2415 // the alias analysis of the rest of the graph, either Compile::can_alias
2416 // or Compile::must_alias will throw a diagnostic assert.)
2417 bool need_mem_bar = false;
2418 switch (kind) {
2419 case Relaxed:
2420 need_mem_bar = mismatched && !adr_type->isa_aryptr();
2421 break;
2422 case Opaque:
2423 // Opaque uses CPUOrder membars for protection against code movement.
2424 case Acquire:
2425 case Release:
2426 case Volatile:
2427 need_mem_bar = true;
2428 break;
2429 default:
2430 ShouldNotReachHere();
2431 }
2432
2433 // Some accesses require access atomicity for all types, notably longs and doubles.
2434 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2435 bool requires_atomic_access = false;
2436 switch (kind) {
2437 case Relaxed:
2438 requires_atomic_access = AlwaysAtomicAccesses;
2439 break;
2440 case Opaque:
2441 // Opaque accesses are atomic.
2442 case Acquire:
2443 case Release:
2444 case Volatile:
2445 requires_atomic_access = true;
2446 break;
2447 default:
2448 ShouldNotReachHere();
2449 }
2450
2451 // Figure out the memory ordering.
2452 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2453 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2454
2455 // If we are reading the value of the referent field of a Reference
2456 // object (either by using Unsafe directly or through reflection)
2457 // then, if G1 is enabled, we need to record the referent in an
2458 // SATB log buffer using the pre-barrier mechanism.
2459 // Also we need to add memory barrier to prevent commoning reads
2460 // from this field across safepoint since GC can change its value.
2461 bool need_read_barrier = !is_store &&
2462 offset != top() && heap_base_oop != top();
2463
2464 if (!is_store && type == T_OBJECT) {
2465 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2466 if (tjp != NULL) {
2467 value_type = tjp;
2468 }
2469 }
2470
2471 receiver = null_check(receiver);
2472 if (stopped()) {
2473 return true;
2474 }
2475 // Heap pointers get a null-check from the interpreter,
2476 // as a courtesy. However, this is not guaranteed by Unsafe,
2477 // and it is not possible to fully distinguish unintended nulls
2478 // from intended ones in this API.
2479
2480 // We need to emit leading and trailing CPU membars (see below) in
2481 // addition to memory membars for special access modes. This is a little
2482 // too strong, but avoids the need to insert per-alias-type
2483 // volatile membars (for stores; compare Parse::do_put_xxx), which
2484 // we cannot do effectively here because we probably only have a
2485 // rough approximation of type.
2486
2487 switch(kind) {
2488 case Relaxed:
2489 case Opaque:
2490 case Acquire:
2491 break;
2492 case Release:
2493 case Volatile:
2494 if (is_store) {
2495 insert_mem_bar(Op_MemBarRelease);
2496 } else {
2497 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2498 insert_mem_bar(Op_MemBarVolatile);
2499 }
2500 }
2501 break;
2502 default:
2503 ShouldNotReachHere();
2504 }
2505
2506 // Memory barrier to prevent normal and 'unsafe' accesses from
2507 // bypassing each other. Happens after null checks, so the
2508 // exception paths do not take memory state from the memory barrier,
2509 // so there's no problems making a strong assert about mixing users
2510 // of safe & unsafe memory.
2511 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2512
2513 if (!is_store) {
2514 Node* p = NULL;
2515 // Try to constant fold a load from a constant field
2516 ciField* field = alias_type->field();
2517 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2518 // final or stable field
2519 p = make_constant_from_field(field, heap_base_oop);
2520 }
2521 if (p == NULL) {
2522 // To be valid, unsafe loads may depend on other conditions than
2523 // the one that guards them: pin the Load node
2524 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2525 // load value
2526 switch (type) {
2527 case T_BOOLEAN:
2528 {
2529 // Normalize the value returned by getBoolean in the following cases
2530 if (mismatched ||
2531 heap_base_oop == top() || // - heap_base_oop is NULL or
2532 (can_access_non_heap && alias_type->field() == NULL) // - heap_base_oop is potentially NULL
2533 // and the unsafe access is made to large offset
2534 // (i.e., larger than the maximum offset necessary for any
2535 // field access)
2536 ) {
2537 IdealKit ideal = IdealKit(this);
2538 #define __ ideal.
2539 IdealVariable normalized_result(ideal);
2540 __ declarations_done();
2541 __ set(normalized_result, p);
2542 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2543 __ set(normalized_result, ideal.ConI(1));
2544 ideal.end_if();
2545 final_sync(ideal);
2546 p = __ value(normalized_result);
2547 #undef __
2548 }
2549 }
2550 case T_CHAR:
2551 case T_BYTE:
2552 case T_SHORT:
2553 case T_INT:
2554 case T_LONG:
2555 case T_FLOAT:
2556 case T_DOUBLE:
2557 break;
2558 case T_OBJECT:
2559 if (need_read_barrier) {
2560 // We do not require a mem bar inside pre_barrier if need_mem_bar
2561 // is set: the barriers would be emitted by us.
2562 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2563 }
2564 break;
2565 case T_ADDRESS:
2566 // Cast to an int type.
2567 p = _gvn.transform(new CastP2XNode(NULL, p));
2568 p = ConvX2UL(p);
2569 break;
2570 default:
2571 fatal("unexpected type %d: %s", type, type2name(type));
2572 break;
2573 }
2574 }
2575 // The load node has the control of the preceding MemBarCPUOrder. All
2576 // following nodes will have the control of the MemBarCPUOrder inserted at
2577 // the end of this method. So, pushing the load onto the stack at a later
2578 // point is fine.
2579 set_result(p);
2580 } else {
2581 // place effect of store into memory
2582 switch (type) {
2583 case T_DOUBLE:
2584 val = dstore_rounding(val);
2585 break;
2586 case T_ADDRESS:
2587 // Repackage the long as a pointer.
2588 val = ConvL2X(val);
2589 val = _gvn.transform(new CastX2PNode(val));
2590 break;
2591 }
2592
2593 if (type == T_OBJECT) {
2594 store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2595 } else {
2596 store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2597 }
2598 }
2599
2600 switch(kind) {
2601 case Relaxed:
2602 case Opaque:
2603 case Release:
2604 break;
2605 case Acquire:
2606 case Volatile:
2607 if (!is_store) {
2608 insert_mem_bar(Op_MemBarAcquire);
2609 } else {
2610 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2611 insert_mem_bar(Op_MemBarVolatile);
2612 }
2613 }
2614 break;
2615 default:
2616 ShouldNotReachHere();
2617 }
2618
2619 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2620
2621 return true;
2622 }
2623
2624 //----------------------------inline_unsafe_load_store----------------------------
2625 // This method serves a couple of different customers (depending on LoadStoreKind):
2626 //
2627 // LS_cmp_swap:
2628 //
2629 // boolean compareAndSetObject(Object o, long offset, Object expected, Object x);
2630 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2631 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2632 //
2633 // LS_cmp_swap_weak:
2634 //
2635 // boolean weakCompareAndSetObject( Object o, long offset, Object expected, Object x);
2636 // boolean weakCompareAndSetObjectPlain( Object o, long offset, Object expected, Object x);
2637 // boolean weakCompareAndSetObjectAcquire(Object o, long offset, Object expected, Object x);
2638 // boolean weakCompareAndSetObjectRelease(Object o, long offset, Object expected, Object x);
2639 //
2640 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2641 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2642 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2643 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2644 //
2645 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2646 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2647 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2648 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2649 //
2650 // LS_cmp_exchange:
2651 //
2652 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2653 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2654 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2655 //
2656 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2657 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2658 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2659 //
2660 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2661 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2662 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2663 //
2664 // LS_get_add:
2665 //
2666 // int getAndAddInt( Object o, long offset, int delta)
2667 // long getAndAddLong(Object o, long offset, long delta)
2668 //
2669 // LS_get_set:
2670 //
2671 // int getAndSet(Object o, long offset, int newValue)
2672 // long getAndSet(Object o, long offset, long newValue)
2673 // Object getAndSet(Object o, long offset, Object newValue)
2674 //
2675 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2676 // This basic scheme here is the same as inline_unsafe_access, but
2677 // differs in enough details that combining them would make the code
2678 // overly confusing. (This is a true fact! I originally combined
2679 // them, but even I was confused by it!) As much code/comments as
2680 // possible are retained from inline_unsafe_access though to make
2681 // the correspondences clearer. - dl
2682
2683 if (callee()->is_static()) return false; // caller must have the capability!
2684
2685 #ifndef PRODUCT
2686 BasicType rtype;
2687 {
2688 ResourceMark rm;
2689 // Check the signatures.
2690 ciSignature* sig = callee()->signature();
2691 rtype = sig->return_type()->basic_type();
2692 switch(kind) {
2693 case LS_get_add:
2694 case LS_get_set: {
2695 // Check the signatures.
2696 #ifdef ASSERT
2697 assert(rtype == type, "get and set must return the expected type");
2698 assert(sig->count() == 3, "get and set has 3 arguments");
2699 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2700 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2701 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2702 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2703 #endif // ASSERT
2704 break;
2705 }
2706 case LS_cmp_swap:
2707 case LS_cmp_swap_weak: {
2708 // Check the signatures.
2709 #ifdef ASSERT
2710 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2711 assert(sig->count() == 4, "CAS has 4 arguments");
2712 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2713 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2714 #endif // ASSERT
2715 break;
2716 }
2717 case LS_cmp_exchange: {
2718 // Check the signatures.
2719 #ifdef ASSERT
2720 assert(rtype == type, "CAS must return the expected type");
2721 assert(sig->count() == 4, "CAS has 4 arguments");
2722 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2723 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2724 #endif // ASSERT
2725 break;
2726 }
2727 default:
2728 ShouldNotReachHere();
2729 }
2730 }
2731 #endif //PRODUCT
2732
2733 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2734
2735 // Get arguments:
2736 Node* receiver = NULL;
2737 Node* base = NULL;
2738 Node* offset = NULL;
2739 Node* oldval = NULL;
2740 Node* newval = NULL;
2741 switch(kind) {
2742 case LS_cmp_swap:
2743 case LS_cmp_swap_weak:
2744 case LS_cmp_exchange: {
2745 const bool two_slot_type = type2size[type] == 2;
2746 receiver = argument(0); // type: oop
2747 base = argument(1); // type: oop
2748 offset = argument(2); // type: long
2749 oldval = argument(4); // type: oop, int, or long
2750 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2751 break;
2752 }
2753 case LS_get_add:
2754 case LS_get_set: {
2755 receiver = argument(0); // type: oop
2756 base = argument(1); // type: oop
2757 offset = argument(2); // type: long
2758 oldval = NULL;
2759 newval = argument(4); // type: oop, int, or long
2760 break;
2761 }
2762 default:
2763 ShouldNotReachHere();
2764 }
2765
2766 // Build field offset expression.
2767 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2768 // to be plain byte offsets, which are also the same as those accepted
2769 // by oopDesc::field_base.
2770 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2771 // 32-bit machines ignore the high half of long offsets
2772 offset = ConvL2X(offset);
2773 Node* adr = make_unsafe_address(base, offset, type);
2774 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2775
2776 Compile::AliasType* alias_type = C->alias_type(adr_type);
2777 BasicType bt = alias_type->basic_type();
2778 if (bt != T_ILLEGAL &&
2779 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2780 // Don't intrinsify mismatched object accesses.
2781 return false;
2782 }
2783
2784 // For CAS, unlike inline_unsafe_access, there seems no point in
2785 // trying to refine types. Just use the coarse types here.
2786 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2787 const Type *value_type = Type::get_const_basic_type(type);
2788
2789 switch (kind) {
2790 case LS_get_set:
2791 case LS_cmp_exchange: {
2792 if (type == T_OBJECT) {
2793 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2794 if (tjp != NULL) {
2795 value_type = tjp;
2796 }
2797 }
2798 break;
2799 }
2800 case LS_cmp_swap:
2801 case LS_cmp_swap_weak:
2802 case LS_get_add:
2803 break;
2804 default:
2805 ShouldNotReachHere();
2806 }
2807
2808 // Null check receiver.
2809 receiver = null_check(receiver);
2810 if (stopped()) {
2811 return true;
2812 }
2813
2814 int alias_idx = C->get_alias_index(adr_type);
2815
2816 // Memory-model-wise, a LoadStore acts like a little synchronized
2817 // block, so needs barriers on each side. These don't translate
2818 // into actual barriers on most machines, but we still need rest of
2819 // compiler to respect ordering.
2820
2821 switch (access_kind) {
2822 case Relaxed:
2823 case Acquire:
2824 break;
2825 case Release:
2826 insert_mem_bar(Op_MemBarRelease);
2827 break;
2828 case Volatile:
2829 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2830 insert_mem_bar(Op_MemBarVolatile);
2831 } else {
2832 insert_mem_bar(Op_MemBarRelease);
2833 }
2834 break;
2835 default:
2836 ShouldNotReachHere();
2837 }
2838 insert_mem_bar(Op_MemBarCPUOrder);
2839
2840 // Figure out the memory ordering.
2841 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2842
2843 // 4984716: MemBars must be inserted before this
2844 // memory node in order to avoid a false
2845 // dependency which will confuse the scheduler.
2846 Node *mem = memory(alias_idx);
2847
2848 // For now, we handle only those cases that actually exist: ints,
2849 // longs, and Object. Adding others should be straightforward.
2850 Node* load_store = NULL;
2851 switch(type) {
2852 case T_BYTE:
2853 switch(kind) {
2854 case LS_get_add:
2855 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2856 break;
2857 case LS_get_set:
2858 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2859 break;
2860 case LS_cmp_swap_weak:
2861 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2862 break;
2863 case LS_cmp_swap:
2864 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2865 break;
2866 case LS_cmp_exchange:
2867 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2868 break;
2869 default:
2870 ShouldNotReachHere();
2871 }
2872 break;
2873 case T_SHORT:
2874 switch(kind) {
2875 case LS_get_add:
2876 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2877 break;
2878 case LS_get_set:
2879 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2880 break;
2881 case LS_cmp_swap_weak:
2882 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2883 break;
2884 case LS_cmp_swap:
2885 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2886 break;
2887 case LS_cmp_exchange:
2888 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2889 break;
2890 default:
2891 ShouldNotReachHere();
2892 }
2893 break;
2894 case T_INT:
2895 switch(kind) {
2896 case LS_get_add:
2897 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2898 break;
2899 case LS_get_set:
2900 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2901 break;
2902 case LS_cmp_swap_weak:
2903 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2904 break;
2905 case LS_cmp_swap:
2906 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2907 break;
2908 case LS_cmp_exchange:
2909 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2910 break;
2911 default:
2912 ShouldNotReachHere();
2913 }
2914 break;
2915 case T_LONG:
2916 switch(kind) {
2917 case LS_get_add:
2918 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2919 break;
2920 case LS_get_set:
2921 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2922 break;
2923 case LS_cmp_swap_weak:
2924 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2925 break;
2926 case LS_cmp_swap:
2927 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2928 break;
2929 case LS_cmp_exchange:
2930 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2931 break;
2932 default:
2933 ShouldNotReachHere();
2934 }
2935 break;
2936 case T_OBJECT:
2937 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2938 // could be delayed during Parse (for example, in adjust_map_after_if()).
2939 // Execute transformation here to avoid barrier generation in such case.
2940 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2941 newval = _gvn.makecon(TypePtr::NULL_PTR);
2942
2943 // Reference stores need a store barrier.
2944 switch(kind) {
2945 case LS_get_set: {
2946 // If pre-barrier must execute before the oop store, old value will require do_load here.
2947 if (!can_move_pre_barrier()) {
2948 pre_barrier(true /* do_load*/,
2949 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2950 NULL /* pre_val*/,
2951 T_OBJECT);
2952 } // Else move pre_barrier to use load_store value, see below.
2953 break;
2954 }
2955 case LS_cmp_swap_weak:
2956 case LS_cmp_swap:
2957 case LS_cmp_exchange: {
2958 // Same as for newval above:
2959 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2960 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2961 }
2962 // The only known value which might get overwritten is oldval.
2963 pre_barrier(false /* do_load */,
2964 control(), NULL, NULL, max_juint, NULL, NULL,
2965 oldval /* pre_val */,
2966 T_OBJECT);
2967 break;
2968 }
2969 default:
2970 ShouldNotReachHere();
2971 }
2972
2973 #ifdef _LP64
2974 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2975 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2976
2977 switch(kind) {
2978 case LS_get_set:
2979 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2980 break;
2981 case LS_cmp_swap_weak: {
2982 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2983 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2984 break;
2985 }
2986 case LS_cmp_swap: {
2987 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2988 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2989 break;
2990 }
2991 case LS_cmp_exchange: {
2992 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2993 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2994 break;
2995 }
2996 default:
2997 ShouldNotReachHere();
2998 }
2999 } else
3000 #endif
3001 switch (kind) {
3002 case LS_get_set:
3003 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
3004 break;
3005 case LS_cmp_swap_weak:
3006 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3007 break;
3008 case LS_cmp_swap:
3009 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3010 break;
3011 case LS_cmp_exchange:
3012 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
3013 break;
3014 default:
3015 ShouldNotReachHere();
3016 }
3017
3018 // Emit the post barrier only when the actual store happened. This makes sense
3019 // to check only for LS_cmp_* that can fail to set the value.
3020 // LS_cmp_exchange does not produce any branches by default, so there is no
3021 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3022 // CompareAndExchange and move branches here, it would make sense to conditionalize
3023 // post_barriers for LS_cmp_exchange as well.
3024 //
3025 // CAS success path is marked more likely since we anticipate this is a performance
3026 // critical path, while CAS failure path can use the penalty for going through unlikely
3027 // path as backoff. Which is still better than doing a store barrier there.
3028 switch (kind) {
3029 case LS_get_set:
3030 case LS_cmp_exchange: {
3031 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3032 break;
3033 }
3034 case LS_cmp_swap_weak:
3035 case LS_cmp_swap: {
3036 IdealKit ideal(this);
3037 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3038 sync_kit(ideal);
3039 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3040 ideal.sync_kit(this);
3041 } ideal.end_if();
3042 final_sync(ideal);
3043 break;
3044 }
3045 default:
3046 ShouldNotReachHere();
3047 }
3048 break;
3049 default:
3050 fatal("unexpected type %d: %s", type, type2name(type));
3051 break;
3052 }
3053
3054 // SCMemProjNodes represent the memory state of a LoadStore. Their
3055 // main role is to prevent LoadStore nodes from being optimized away
3056 // when their results aren't used.
3057 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3058 set_memory(proj, alias_idx);
3059
3060 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3061 #ifdef _LP64
3062 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3063 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3064 }
3065 #endif
3066 if (can_move_pre_barrier() && kind == LS_get_set) {
3067 // Don't need to load pre_val. The old value is returned by load_store.
3068 // The pre_barrier can execute after the xchg as long as no safepoint
3069 // gets inserted between them.
3070 pre_barrier(false /* do_load */,
3071 control(), NULL, NULL, max_juint, NULL, NULL,
3072 load_store /* pre_val */,
3073 T_OBJECT);
3074 }
3075 }
3076
3077 // Add the trailing membar surrounding the access
3078 insert_mem_bar(Op_MemBarCPUOrder);
3079
3080 switch (access_kind) {
3081 case Relaxed:
3082 case Release:
3083 break; // do nothing
3084 case Acquire:
3085 case Volatile:
3086 insert_mem_bar(Op_MemBarAcquire);
3087 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3088 break;
3089 default:
3090 ShouldNotReachHere();
3091 }
3092
3093 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3094 set_result(load_store);
3095 return true;
3096 }
3097
3098 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3099 MemNode::MemOrd mo = MemNode::unset;
3100 switch(kind) {
3101 case Opaque:
3102 case Relaxed: mo = MemNode::unordered; break;
3103 case Acquire: mo = MemNode::acquire; break;
3104 case Release: mo = MemNode::release; break;
3105 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3106 default:
3107 ShouldNotReachHere();
3108 }
3109 guarantee(mo != MemNode::unset, "Should select memory ordering");
3110 return mo;
3111 }
3112
3113 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3114 MemNode::MemOrd mo = MemNode::unset;
3115 switch(kind) {
3116 case Opaque:
3117 case Relaxed: mo = MemNode::unordered; break;
3118 case Acquire: mo = MemNode::acquire; break;
3119 case Release: mo = MemNode::release; break;
3120 case Volatile: mo = MemNode::seqcst; break;
3121 default:
3122 ShouldNotReachHere();
3123 }
3124 guarantee(mo != MemNode::unset, "Should select memory ordering");
3125 return mo;
3126 }
3127
3128 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3129 // Regardless of form, don't allow previous ld/st to move down,
3130 // then issue acquire, release, or volatile mem_bar.
3131 insert_mem_bar(Op_MemBarCPUOrder);
3132 switch(id) {
3133 case vmIntrinsics::_loadFence:
3134 insert_mem_bar(Op_LoadFence);
3135 return true;
3136 case vmIntrinsics::_storeFence:
3137 insert_mem_bar(Op_StoreFence);
3138 return true;
3139 case vmIntrinsics::_fullFence:
3140 insert_mem_bar(Op_MemBarVolatile);
3141 return true;
3142 default:
3143 fatal_unexpected_iid(id);
3144 return false;
3145 }
3146 }
3147
3148 bool LibraryCallKit::inline_onspinwait() {
3149 insert_mem_bar(Op_OnSpinWait);
3150 return true;
3151 }
3152
3153 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3154 if (!kls->is_Con()) {
3155 return true;
3156 }
3157 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3158 if (klsptr == NULL) {
3159 return true;
3160 }
3161 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3162 // don't need a guard for a klass that is already initialized
3163 return !ik->is_initialized();
3164 }
3165
3166 //----------------------------inline_unsafe_allocate---------------------------
3167 // public native Object Unsafe.allocateInstance(Class<?> cls);
3168 bool LibraryCallKit::inline_unsafe_allocate() {
3169 if (callee()->is_static()) return false; // caller must have the capability!
3170
3171 null_check_receiver(); // null-check, then ignore
3172 Node* cls = null_check(argument(1));
3173 if (stopped()) return true;
3174
3175 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3176 kls = null_check(kls);
3177 if (stopped()) return true; // argument was like int.class
3178
3179 Node* test = NULL;
3180 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3181 // Note: The argument might still be an illegal value like
3182 // Serializable.class or Object[].class. The runtime will handle it.
3183 // But we must make an explicit check for initialization.
3184 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3185 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3186 // can generate code to load it as unsigned byte.
3187 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3188 Node* bits = intcon(InstanceKlass::fully_initialized);
3189 test = _gvn.transform(new SubINode(inst, bits));
3190 // The 'test' is non-zero if we need to take a slow path.
3191 }
3192
3193 Node* obj = new_instance(kls, test);
3194 set_result(obj);
3195 return true;
3196 }
3197
3198 //------------------------inline_native_time_funcs--------------
3199 // inline code for System.currentTimeMillis() and System.nanoTime()
3200 // these have the same type and signature
3201 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3202 const TypeFunc* tf = OptoRuntime::void_long_Type();
3203 const TypePtr* no_memory_effects = NULL;
3204 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3205 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3206 #ifdef ASSERT
3207 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3208 assert(value_top == top(), "second value must be top");
3209 #endif
3210 set_result(value);
3211 return true;
3212 }
3213
3214 #ifdef TRACE_HAVE_INTRINSICS
3215
3216 /*
3217 * oop -> myklass
3218 * myklass->trace_id |= USED
3219 * return myklass->trace_id & ~0x3
3220 */
3221 bool LibraryCallKit::inline_native_classID() {
3222 Node* cls = null_check(argument(0), T_OBJECT);
3223 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3224 kls = null_check(kls, T_OBJECT);
3225
3226 ByteSize offset = TRACE_KLASS_TRACE_ID_OFFSET;
3227 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3228 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3229
3230 Node* clsused = longcon(0x01l); // set the class bit
3231 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3232 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3233 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3234
3235 #ifdef TRACE_ID_META_BITS
3236 Node* mbits = longcon(~TRACE_ID_META_BITS);
3237 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3238 #endif
3239 #ifdef TRACE_ID_CLASS_SHIFT
3240 Node* cbits = intcon(TRACE_ID_CLASS_SHIFT);
3241 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3242 #endif
3243
3244 set_result(tvalue);
3245 return true;
3246
3247 }
3248
3249 bool LibraryCallKit::inline_native_getBufferWriter() {
3250 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3251
3252 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3253 in_bytes(TRACE_THREAD_DATA_WRITER_OFFSET)
3254 );
3255
3256 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3257
3258 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3259 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3260
3261 IfNode* iff_jobj_null =
3262 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3263
3264 enum { _normal_path = 1,
3265 _null_path = 2,
3266 PATH_LIMIT };
3267
3268 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3269 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
3270
3271 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3272 result_rgn->init_req(_null_path, jobj_is_null);
3273 result_val->init_req(_null_path, null());
3274
3275 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3276 result_rgn->init_req(_normal_path, jobj_is_not_null);
3277
3278 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
3279 result_val->init_req(_normal_path, res);
3280
3281 set_result(result_rgn, result_val);
3282
3283 return true;
3284 }
3285
3286 #endif
3287
3288 //------------------------inline_native_currentThread------------------
3289 bool LibraryCallKit::inline_native_currentThread() {
3290 Node* junk = NULL;
3291 set_result(generate_current_thread(junk));
3292 return true;
3293 }
3294
3295 //------------------------inline_native_isInterrupted------------------
3296 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3297 bool LibraryCallKit::inline_native_isInterrupted() {
3298 // Add a fast path to t.isInterrupted(clear_int):
3299 // (t == Thread.current() &&
3300 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3301 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3302 // So, in the common case that the interrupt bit is false,
3303 // we avoid making a call into the VM. Even if the interrupt bit
3304 // is true, if the clear_int argument is false, we avoid the VM call.
3305 // However, if the receiver is not currentThread, we must call the VM,
3306 // because there must be some locking done around the operation.
3307
3308 // We only go to the fast case code if we pass two guards.
3309 // Paths which do not pass are accumulated in the slow_region.
3310
3311 enum {
3312 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3313 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3314 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3315 PATH_LIMIT
3316 };
3317
3318 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3319 // out of the function.
3320 insert_mem_bar(Op_MemBarCPUOrder);
3321
3322 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3323 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3324
3325 RegionNode* slow_region = new RegionNode(1);
3326 record_for_igvn(slow_region);
3327
3328 // (a) Receiving thread must be the current thread.
3329 Node* rec_thr = argument(0);
3330 Node* tls_ptr = NULL;
3331 Node* cur_thr = generate_current_thread(tls_ptr);
3332 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3333 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3334
3335 generate_slow_guard(bol_thr, slow_region);
3336
3337 // (b) Interrupt bit on TLS must be false.
3338 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3339 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3340 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3341
3342 // Set the control input on the field _interrupted read to prevent it floating up.
3343 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3344 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3345 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3346
3347 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3348
3349 // First fast path: if (!TLS._interrupted) return false;
3350 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3351 result_rgn->init_req(no_int_result_path, false_bit);
3352 result_val->init_req(no_int_result_path, intcon(0));
3353
3354 // drop through to next case
3355 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3356
3357 #ifndef _WINDOWS
3358 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3359 Node* clr_arg = argument(1);
3360 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3361 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3362 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3363
3364 // Second fast path: ... else if (!clear_int) return true;
3365 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3366 result_rgn->init_req(no_clear_result_path, false_arg);
3367 result_val->init_req(no_clear_result_path, intcon(1));
3368
3369 // drop through to next case
3370 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3371 #else
3372 // To return true on Windows you must read the _interrupted field
3373 // and check the event state i.e. take the slow path.
3374 #endif // _WINDOWS
3375
3376 // (d) Otherwise, go to the slow path.
3377 slow_region->add_req(control());
3378 set_control( _gvn.transform(slow_region));
3379
3380 if (stopped()) {
3381 // There is no slow path.
3382 result_rgn->init_req(slow_result_path, top());
3383 result_val->init_req(slow_result_path, top());
3384 } else {
3385 // non-virtual because it is a private non-static
3386 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3387
3388 Node* slow_val = set_results_for_java_call(slow_call);
3389 // this->control() comes from set_results_for_java_call
3390
3391 Node* fast_io = slow_call->in(TypeFunc::I_O);
3392 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3393
3394 // These two phis are pre-filled with copies of of the fast IO and Memory
3395 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3396 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3397
3398 result_rgn->init_req(slow_result_path, control());
3399 result_io ->init_req(slow_result_path, i_o());
3400 result_mem->init_req(slow_result_path, reset_memory());
3401 result_val->init_req(slow_result_path, slow_val);
3402
3403 set_all_memory(_gvn.transform(result_mem));
3404 set_i_o( _gvn.transform(result_io));
3405 }
3406
3407 C->set_has_split_ifs(true); // Has chance for split-if optimization
3408 set_result(result_rgn, result_val);
3409 return true;
3410 }
3411
3412 //---------------------------load_mirror_from_klass----------------------------
3413 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3414 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3415 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3416 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3417 }
3418
3419 //-----------------------load_klass_from_mirror_common-------------------------
3420 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3421 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3422 // and branch to the given path on the region.
3423 // If never_see_null, take an uncommon trap on null, so we can optimistically
3424 // compile for the non-null case.
3425 // If the region is NULL, force never_see_null = true.
3426 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3427 bool never_see_null,
3428 RegionNode* region,
3429 int null_path,
3430 int offset) {
3431 if (region == NULL) never_see_null = true;
3432 Node* p = basic_plus_adr(mirror, offset);
3433 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3434 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3435 Node* null_ctl = top();
3436 kls = null_check_oop(kls, &null_ctl, never_see_null);
3437 if (region != NULL) {
3438 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3439 region->init_req(null_path, null_ctl);
3440 } else {
3441 assert(null_ctl == top(), "no loose ends");
3442 }
3443 return kls;
3444 }
3445
3446 //--------------------(inline_native_Class_query helpers)---------------------
3447 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3448 // Fall through if (mods & mask) == bits, take the guard otherwise.
3449 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3450 // Branch around if the given klass has the given modifier bit set.
3451 // Like generate_guard, adds a new path onto the region.
3452 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3453 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3454 Node* mask = intcon(modifier_mask);
3455 Node* bits = intcon(modifier_bits);
3456 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3457 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3458 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3459 return generate_fair_guard(bol, region);
3460 }
3461 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3462 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3463 }
3464
3465 //-------------------------inline_native_Class_query-------------------
3466 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3467 const Type* return_type = TypeInt::BOOL;
3468 Node* prim_return_value = top(); // what happens if it's a primitive class?
3469 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3470 bool expect_prim = false; // most of these guys expect to work on refs
3471
3472 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3473
3474 Node* mirror = argument(0);
3475 Node* obj = top();
3476
3477 switch (id) {
3478 case vmIntrinsics::_isInstance:
3479 // nothing is an instance of a primitive type
3480 prim_return_value = intcon(0);
3481 obj = argument(1);
3482 break;
3483 case vmIntrinsics::_getModifiers:
3484 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3485 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3486 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3487 break;
3488 case vmIntrinsics::_isInterface:
3489 prim_return_value = intcon(0);
3490 break;
3491 case vmIntrinsics::_isArray:
3492 prim_return_value = intcon(0);
3493 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3494 break;
3495 case vmIntrinsics::_isPrimitive:
3496 prim_return_value = intcon(1);
3497 expect_prim = true; // obviously
3498 break;
3499 case vmIntrinsics::_getSuperclass:
3500 prim_return_value = null();
3501 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3502 break;
3503 case vmIntrinsics::_getClassAccessFlags:
3504 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3505 return_type = TypeInt::INT; // not bool! 6297094
3506 break;
3507 default:
3508 fatal_unexpected_iid(id);
3509 break;
3510 }
3511
3512 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3513 if (mirror_con == NULL) return false; // cannot happen?
3514
3515 #ifndef PRODUCT
3516 if (C->print_intrinsics() || C->print_inlining()) {
3517 ciType* k = mirror_con->java_mirror_type();
3518 if (k) {
3519 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3520 k->print_name();
3521 tty->cr();
3522 }
3523 }
3524 #endif
3525
3526 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3527 RegionNode* region = new RegionNode(PATH_LIMIT);
3528 record_for_igvn(region);
3529 PhiNode* phi = new PhiNode(region, return_type);
3530
3531 // The mirror will never be null of Reflection.getClassAccessFlags, however
3532 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3533 // if it is. See bug 4774291.
3534
3535 // For Reflection.getClassAccessFlags(), the null check occurs in
3536 // the wrong place; see inline_unsafe_access(), above, for a similar
3537 // situation.
3538 mirror = null_check(mirror);
3539 // If mirror or obj is dead, only null-path is taken.
3540 if (stopped()) return true;
3541
3542 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3543
3544 // Now load the mirror's klass metaobject, and null-check it.
3545 // Side-effects region with the control path if the klass is null.
3546 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3547 // If kls is null, we have a primitive mirror.
3548 phi->init_req(_prim_path, prim_return_value);
3549 if (stopped()) { set_result(region, phi); return true; }
3550 bool safe_for_replace = (region->in(_prim_path) == top());
3551
3552 Node* p; // handy temp
3553 Node* null_ctl;
3554
3555 // Now that we have the non-null klass, we can perform the real query.
3556 // For constant classes, the query will constant-fold in LoadNode::Value.
3557 Node* query_value = top();
3558 switch (id) {
3559 case vmIntrinsics::_isInstance:
3560 // nothing is an instance of a primitive type
3561 query_value = gen_instanceof(obj, kls, safe_for_replace);
3562 break;
3563
3564 case vmIntrinsics::_getModifiers:
3565 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3566 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3567 break;
3568
3569 case vmIntrinsics::_isInterface:
3570 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3571 if (generate_interface_guard(kls, region) != NULL)
3572 // A guard was added. If the guard is taken, it was an interface.
3573 phi->add_req(intcon(1));
3574 // If we fall through, it's a plain class.
3575 query_value = intcon(0);
3576 break;
3577
3578 case vmIntrinsics::_isArray:
3579 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3580 if (generate_array_guard(kls, region) != NULL)
3581 // A guard was added. If the guard is taken, it was an array.
3582 phi->add_req(intcon(1));
3583 // If we fall through, it's a plain class.
3584 query_value = intcon(0);
3585 break;
3586
3587 case vmIntrinsics::_isPrimitive:
3588 query_value = intcon(0); // "normal" path produces false
3589 break;
3590
3591 case vmIntrinsics::_getSuperclass:
3592 // The rules here are somewhat unfortunate, but we can still do better
3593 // with random logic than with a JNI call.
3594 // Interfaces store null or Object as _super, but must report null.
3595 // Arrays store an intermediate super as _super, but must report Object.
3596 // Other types can report the actual _super.
3597 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3598 if (generate_interface_guard(kls, region) != NULL)
3599 // A guard was added. If the guard is taken, it was an interface.
3600 phi->add_req(null());
3601 if (generate_array_guard(kls, region) != NULL)
3602 // A guard was added. If the guard is taken, it was an array.
3603 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3604 // If we fall through, it's a plain class. Get its _super.
3605 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3606 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3607 null_ctl = top();
3608 kls = null_check_oop(kls, &null_ctl);
3609 if (null_ctl != top()) {
3610 // If the guard is taken, Object.superClass is null (both klass and mirror).
3611 region->add_req(null_ctl);
3612 phi ->add_req(null());
3613 }
3614 if (!stopped()) {
3615 query_value = load_mirror_from_klass(kls);
3616 }
3617 break;
3618
3619 case vmIntrinsics::_getClassAccessFlags:
3620 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3621 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3622 break;
3623
3624 default:
3625 fatal_unexpected_iid(id);
3626 break;
3627 }
3628
3629 // Fall-through is the normal case of a query to a real class.
3630 phi->init_req(1, query_value);
3631 region->init_req(1, control());
3632
3633 C->set_has_split_ifs(true); // Has chance for split-if optimization
3634 set_result(region, phi);
3635 return true;
3636 }
3637
3638 //-------------------------inline_Class_cast-------------------
3639 bool LibraryCallKit::inline_Class_cast() {
3640 Node* mirror = argument(0); // Class
3641 Node* obj = argument(1);
3642 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3643 if (mirror_con == NULL) {
3644 return false; // dead path (mirror->is_top()).
3645 }
3646 if (obj == NULL || obj->is_top()) {
3647 return false; // dead path
3648 }
3649 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3650
3651 // First, see if Class.cast() can be folded statically.
3652 // java_mirror_type() returns non-null for compile-time Class constants.
3653 ciType* tm = mirror_con->java_mirror_type();
3654 if (tm != NULL && tm->is_klass() &&
3655 tp != NULL && tp->klass() != NULL) {
3656 if (!tp->klass()->is_loaded()) {
3657 // Don't use intrinsic when class is not loaded.
3658 return false;
3659 } else {
3660 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3661 if (static_res == Compile::SSC_always_true) {
3662 // isInstance() is true - fold the code.
3663 set_result(obj);
3664 return true;
3665 } else if (static_res == Compile::SSC_always_false) {
3666 // Don't use intrinsic, have to throw ClassCastException.
3667 // If the reference is null, the non-intrinsic bytecode will
3668 // be optimized appropriately.
3669 return false;
3670 }
3671 }
3672 }
3673
3674 // Bailout intrinsic and do normal inlining if exception path is frequent.
3675 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3676 return false;
3677 }
3678
3679 // Generate dynamic checks.
3680 // Class.cast() is java implementation of _checkcast bytecode.
3681 // Do checkcast (Parse::do_checkcast()) optimizations here.
3682
3683 mirror = null_check(mirror);
3684 // If mirror is dead, only null-path is taken.
3685 if (stopped()) {
3686 return true;
3687 }
3688
3689 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3690 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3691 RegionNode* region = new RegionNode(PATH_LIMIT);
3692 record_for_igvn(region);
3693
3694 // Now load the mirror's klass metaobject, and null-check it.
3695 // If kls is null, we have a primitive mirror and
3696 // nothing is an instance of a primitive type.
3697 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3698
3699 Node* res = top();
3700 if (!stopped()) {
3701 Node* bad_type_ctrl = top();
3702 // Do checkcast optimizations.
3703 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3704 region->init_req(_bad_type_path, bad_type_ctrl);
3705 }
3706 if (region->in(_prim_path) != top() ||
3707 region->in(_bad_type_path) != top()) {
3708 // Let Interpreter throw ClassCastException.
3709 PreserveJVMState pjvms(this);
3710 set_control(_gvn.transform(region));
3711 uncommon_trap(Deoptimization::Reason_intrinsic,
3712 Deoptimization::Action_maybe_recompile);
3713 }
3714 if (!stopped()) {
3715 set_result(res);
3716 }
3717 return true;
3718 }
3719
3720
3721 //--------------------------inline_native_subtype_check------------------------
3722 // This intrinsic takes the JNI calls out of the heart of
3723 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3724 bool LibraryCallKit::inline_native_subtype_check() {
3725 // Pull both arguments off the stack.
3726 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3727 args[0] = argument(0);
3728 args[1] = argument(1);
3729 Node* klasses[2]; // corresponding Klasses: superk, subk
3730 klasses[0] = klasses[1] = top();
3731
3732 enum {
3733 // A full decision tree on {superc is prim, subc is prim}:
3734 _prim_0_path = 1, // {P,N} => false
3735 // {P,P} & superc!=subc => false
3736 _prim_same_path, // {P,P} & superc==subc => true
3737 _prim_1_path, // {N,P} => false
3738 _ref_subtype_path, // {N,N} & subtype check wins => true
3739 _both_ref_path, // {N,N} & subtype check loses => false
3740 PATH_LIMIT
3741 };
3742
3743 RegionNode* region = new RegionNode(PATH_LIMIT);
3744 Node* phi = new PhiNode(region, TypeInt::BOOL);
3745 record_for_igvn(region);
3746
3747 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3748 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3749 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3750
3751 // First null-check both mirrors and load each mirror's klass metaobject.
3752 int which_arg;
3753 for (which_arg = 0; which_arg <= 1; which_arg++) {
3754 Node* arg = args[which_arg];
3755 arg = null_check(arg);
3756 if (stopped()) break;
3757 args[which_arg] = arg;
3758
3759 Node* p = basic_plus_adr(arg, class_klass_offset);
3760 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3761 klasses[which_arg] = _gvn.transform(kls);
3762 }
3763
3764 // Having loaded both klasses, test each for null.
3765 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3766 for (which_arg = 0; which_arg <= 1; which_arg++) {
3767 Node* kls = klasses[which_arg];
3768 Node* null_ctl = top();
3769 kls = null_check_oop(kls, &null_ctl, never_see_null);
3770 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3771 region->init_req(prim_path, null_ctl);
3772 if (stopped()) break;
3773 klasses[which_arg] = kls;
3774 }
3775
3776 if (!stopped()) {
3777 // now we have two reference types, in klasses[0..1]
3778 Node* subk = klasses[1]; // the argument to isAssignableFrom
3779 Node* superk = klasses[0]; // the receiver
3780 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3781 // now we have a successful reference subtype check
3782 region->set_req(_ref_subtype_path, control());
3783 }
3784
3785 // If both operands are primitive (both klasses null), then
3786 // we must return true when they are identical primitives.
3787 // It is convenient to test this after the first null klass check.
3788 set_control(region->in(_prim_0_path)); // go back to first null check
3789 if (!stopped()) {
3790 // Since superc is primitive, make a guard for the superc==subc case.
3791 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3792 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3793 generate_guard(bol_eq, region, PROB_FAIR);
3794 if (region->req() == PATH_LIMIT+1) {
3795 // A guard was added. If the added guard is taken, superc==subc.
3796 region->swap_edges(PATH_LIMIT, _prim_same_path);
3797 region->del_req(PATH_LIMIT);
3798 }
3799 region->set_req(_prim_0_path, control()); // Not equal after all.
3800 }
3801
3802 // these are the only paths that produce 'true':
3803 phi->set_req(_prim_same_path, intcon(1));
3804 phi->set_req(_ref_subtype_path, intcon(1));
3805
3806 // pull together the cases:
3807 assert(region->req() == PATH_LIMIT, "sane region");
3808 for (uint i = 1; i < region->req(); i++) {
3809 Node* ctl = region->in(i);
3810 if (ctl == NULL || ctl == top()) {
3811 region->set_req(i, top());
3812 phi ->set_req(i, top());
3813 } else if (phi->in(i) == NULL) {
3814 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3815 }
3816 }
3817
3818 set_control(_gvn.transform(region));
3819 set_result(_gvn.transform(phi));
3820 return true;
3821 }
3822
3823 //---------------------generate_array_guard_common------------------------
3824 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3825 bool obj_array, bool not_array) {
3826
3827 if (stopped()) {
3828 return NULL;
3829 }
3830
3831 // If obj_array/non_array==false/false:
3832 // Branch around if the given klass is in fact an array (either obj or prim).
3833 // If obj_array/non_array==false/true:
3834 // Branch around if the given klass is not an array klass of any kind.
3835 // If obj_array/non_array==true/true:
3836 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3837 // If obj_array/non_array==true/false:
3838 // Branch around if the kls is an oop array (Object[] or subtype)
3839 //
3840 // Like generate_guard, adds a new path onto the region.
3841 jint layout_con = 0;
3842 Node* layout_val = get_layout_helper(kls, layout_con);
3843 if (layout_val == NULL) {
3844 bool query = (obj_array
3845 ? Klass::layout_helper_is_objArray(layout_con)
3846 : Klass::layout_helper_is_array(layout_con));
3847 if (query == not_array) {
3848 return NULL; // never a branch
3849 } else { // always a branch
3850 Node* always_branch = control();
3851 if (region != NULL)
3852 region->add_req(always_branch);
3853 set_control(top());
3854 return always_branch;
3855 }
3856 }
3857 // Now test the correct condition.
3858 jint nval = (obj_array
3859 ? (jint)(Klass::_lh_array_tag_type_value
3860 << Klass::_lh_array_tag_shift)
3861 : Klass::_lh_neutral_value);
3862 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3863 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3864 // invert the test if we are looking for a non-array
3865 if (not_array) btest = BoolTest(btest).negate();
3866 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3867 return generate_fair_guard(bol, region);
3868 }
3869
3870
3871 //-----------------------inline_native_newArray--------------------------
3872 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3873 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3874 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3875 Node* mirror;
3876 Node* count_val;
3877 if (uninitialized) {
3878 mirror = argument(1);
3879 count_val = argument(2);
3880 } else {
3881 mirror = argument(0);
3882 count_val = argument(1);
3883 }
3884
3885 mirror = null_check(mirror);
3886 // If mirror or obj is dead, only null-path is taken.
3887 if (stopped()) return true;
3888
3889 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3890 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3891 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3892 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3893 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3894
3895 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3896 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3897 result_reg, _slow_path);
3898 Node* normal_ctl = control();
3899 Node* no_array_ctl = result_reg->in(_slow_path);
3900
3901 // Generate code for the slow case. We make a call to newArray().
3902 set_control(no_array_ctl);
3903 if (!stopped()) {
3904 // Either the input type is void.class, or else the
3905 // array klass has not yet been cached. Either the
3906 // ensuing call will throw an exception, or else it
3907 // will cache the array klass for next time.
3908 PreserveJVMState pjvms(this);
3909 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3910 Node* slow_result = set_results_for_java_call(slow_call);
3911 // this->control() comes from set_results_for_java_call
3912 result_reg->set_req(_slow_path, control());
3913 result_val->set_req(_slow_path, slow_result);
3914 result_io ->set_req(_slow_path, i_o());
3915 result_mem->set_req(_slow_path, reset_memory());
3916 }
3917
3918 set_control(normal_ctl);
3919 if (!stopped()) {
3920 // Normal case: The array type has been cached in the java.lang.Class.
3921 // The following call works fine even if the array type is polymorphic.
3922 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3923 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3924 result_reg->init_req(_normal_path, control());
3925 result_val->init_req(_normal_path, obj);
3926 result_io ->init_req(_normal_path, i_o());
3927 result_mem->init_req(_normal_path, reset_memory());
3928
3929 if (uninitialized) {
3930 // Mark the allocation so that zeroing is skipped
3931 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3932 alloc->maybe_set_complete(&_gvn);
3933 }
3934 }
3935
3936 // Return the combined state.
3937 set_i_o( _gvn.transform(result_io) );
3938 set_all_memory( _gvn.transform(result_mem));
3939
3940 C->set_has_split_ifs(true); // Has chance for split-if optimization
3941 set_result(result_reg, result_val);
3942 return true;
3943 }
3944
3945 //----------------------inline_native_getLength--------------------------
3946 // public static native int java.lang.reflect.Array.getLength(Object array);
3947 bool LibraryCallKit::inline_native_getLength() {
3948 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3949
3950 Node* array = null_check(argument(0));
3951 // If array is dead, only null-path is taken.
3952 if (stopped()) return true;
3953
3954 // Deoptimize if it is a non-array.
3955 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3956
3957 if (non_array != NULL) {
3958 PreserveJVMState pjvms(this);
3959 set_control(non_array);
3960 uncommon_trap(Deoptimization::Reason_intrinsic,
3961 Deoptimization::Action_maybe_recompile);
3962 }
3963
3964 // If control is dead, only non-array-path is taken.
3965 if (stopped()) return true;
3966
3967 // The works fine even if the array type is polymorphic.
3968 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3969 Node* result = load_array_length(array);
3970
3971 C->set_has_split_ifs(true); // Has chance for split-if optimization
3972 set_result(result);
3973 return true;
3974 }
3975
3976 //------------------------inline_array_copyOf----------------------------
3977 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3978 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3979 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3980 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3981
3982 // Get the arguments.
3983 Node* original = argument(0);
3984 Node* start = is_copyOfRange? argument(1): intcon(0);
3985 Node* end = is_copyOfRange? argument(2): argument(1);
3986 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3987
3988 Node* newcopy = NULL;
3989
3990 // Set the original stack and the reexecute bit for the interpreter to reexecute
3991 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3992 { PreserveReexecuteState preexecs(this);
3993 jvms()->set_should_reexecute(true);
3994
3995 array_type_mirror = null_check(array_type_mirror);
3996 original = null_check(original);
3997
3998 // Check if a null path was taken unconditionally.
3999 if (stopped()) return true;
4000
4001 Node* orig_length = load_array_length(original);
4002
4003 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4004 klass_node = null_check(klass_node);
4005
4006 RegionNode* bailout = new RegionNode(1);
4007 record_for_igvn(bailout);
4008
4009 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4010 // Bail out if that is so.
4011 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4012 if (not_objArray != NULL) {
4013 // Improve the klass node's type from the new optimistic assumption:
4014 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4015 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4016 Node* cast = new CastPPNode(klass_node, akls);
4017 cast->init_req(0, control());
4018 klass_node = _gvn.transform(cast);
4019 }
4020
4021 // Bail out if either start or end is negative.
4022 generate_negative_guard(start, bailout, &start);
4023 generate_negative_guard(end, bailout, &end);
4024
4025 Node* length = end;
4026 if (_gvn.type(start) != TypeInt::ZERO) {
4027 length = _gvn.transform(new SubINode(end, start));
4028 }
4029
4030 // Bail out if length is negative.
4031 // Without this the new_array would throw
4032 // NegativeArraySizeException but IllegalArgumentException is what
4033 // should be thrown
4034 generate_negative_guard(length, bailout, &length);
4035
4036 if (bailout->req() > 1) {
4037 PreserveJVMState pjvms(this);
4038 set_control(_gvn.transform(bailout));
4039 uncommon_trap(Deoptimization::Reason_intrinsic,
4040 Deoptimization::Action_maybe_recompile);
4041 }
4042
4043 if (!stopped()) {
4044 // How many elements will we copy from the original?
4045 // The answer is MinI(orig_length - start, length).
4046 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4047 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4048
4049 // Generate a direct call to the right arraycopy function(s).
4050 // We know the copy is disjoint but we might not know if the
4051 // oop stores need checking.
4052 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4053 // This will fail a store-check if x contains any non-nulls.
4054
4055 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4056 // loads/stores but it is legal only if we're sure the
4057 // Arrays.copyOf would succeed. So we need all input arguments
4058 // to the copyOf to be validated, including that the copy to the
4059 // new array won't trigger an ArrayStoreException. That subtype
4060 // check can be optimized if we know something on the type of
4061 // the input array from type speculation.
4062 if (_gvn.type(klass_node)->singleton()) {
4063 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4064 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4065
4066 int test = C->static_subtype_check(superk, subk);
4067 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4068 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4069 if (t_original->speculative_type() != NULL) {
4070 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4071 }
4072 }
4073 }
4074
4075 bool validated = false;
4076 // Reason_class_check rather than Reason_intrinsic because we
4077 // want to intrinsify even if this traps.
4078 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4079 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4080 klass_node);
4081
4082 if (not_subtype_ctrl != top()) {
4083 PreserveJVMState pjvms(this);
4084 set_control(not_subtype_ctrl);
4085 uncommon_trap(Deoptimization::Reason_class_check,
4086 Deoptimization::Action_make_not_entrant);
4087 assert(stopped(), "Should be stopped");
4088 }
4089 validated = true;
4090 }
4091
4092 if (!stopped()) {
4093 newcopy = new_array(klass_node, length, 0); // no arguments to push
4094
4095 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4096 load_object_klass(original), klass_node);
4097 if (!is_copyOfRange) {
4098 ac->set_copyof(validated);
4099 } else {
4100 ac->set_copyofrange(validated);
4101 }
4102 Node* n = _gvn.transform(ac);
4103 if (n == ac) {
4104 ac->connect_outputs(this);
4105 } else {
4106 assert(validated, "shouldn't transform if all arguments not validated");
4107 set_all_memory(n);
4108 }
4109 }
4110 }
4111 } // original reexecute is set back here
4112
4113 C->set_has_split_ifs(true); // Has chance for split-if optimization
4114 if (!stopped()) {
4115 set_result(newcopy);
4116 }
4117 return true;
4118 }
4119
4120
4121 //----------------------generate_virtual_guard---------------------------
4122 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4123 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4124 RegionNode* slow_region) {
4125 ciMethod* method = callee();
4126 int vtable_index = method->vtable_index();
4127 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4128 "bad index %d", vtable_index);
4129 // Get the Method* out of the appropriate vtable entry.
4130 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4131 vtable_index*vtableEntry::size_in_bytes() +
4132 vtableEntry::method_offset_in_bytes();
4133 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4134 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4135
4136 // Compare the target method with the expected method (e.g., Object.hashCode).
4137 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4138
4139 Node* native_call = makecon(native_call_addr);
4140 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4141 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4142
4143 return generate_slow_guard(test_native, slow_region);
4144 }
4145
4146 //-----------------------generate_method_call----------------------------
4147 // Use generate_method_call to make a slow-call to the real
4148 // method if the fast path fails. An alternative would be to
4149 // use a stub like OptoRuntime::slow_arraycopy_Java.
4150 // This only works for expanding the current library call,
4151 // not another intrinsic. (E.g., don't use this for making an
4152 // arraycopy call inside of the copyOf intrinsic.)
4153 CallJavaNode*
4154 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4155 // When compiling the intrinsic method itself, do not use this technique.
4156 guarantee(callee() != C->method(), "cannot make slow-call to self");
4157
4158 ciMethod* method = callee();
4159 // ensure the JVMS we have will be correct for this call
4160 guarantee(method_id == method->intrinsic_id(), "must match");
4161
4162 const TypeFunc* tf = TypeFunc::make(method);
4163 CallJavaNode* slow_call;
4164 if (is_static) {
4165 assert(!is_virtual, "");
4166 slow_call = new CallStaticJavaNode(C, tf,
4167 SharedRuntime::get_resolve_static_call_stub(),
4168 method, bci());
4169 } else if (is_virtual) {
4170 null_check_receiver();
4171 int vtable_index = Method::invalid_vtable_index;
4172 if (UseInlineCaches) {
4173 // Suppress the vtable call
4174 } else {
4175 // hashCode and clone are not a miranda methods,
4176 // so the vtable index is fixed.
4177 // No need to use the linkResolver to get it.
4178 vtable_index = method->vtable_index();
4179 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4180 "bad index %d", vtable_index);
4181 }
4182 slow_call = new CallDynamicJavaNode(tf,
4183 SharedRuntime::get_resolve_virtual_call_stub(),
4184 method, vtable_index, bci());
4185 } else { // neither virtual nor static: opt_virtual
4186 null_check_receiver();
4187 slow_call = new CallStaticJavaNode(C, tf,
4188 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4189 method, bci());
4190 slow_call->set_optimized_virtual(true);
4191 }
4192 set_arguments_for_java_call(slow_call);
4193 set_edges_for_java_call(slow_call);
4194 return slow_call;
4195 }
4196
4197
4198 /**
4199 * Build special case code for calls to hashCode on an object. This call may
4200 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4201 * slightly different code.
4202 */
4203 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4204 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4205 assert(!(is_virtual && is_static), "either virtual, special, or static");
4206
4207 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4208
4209 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4210 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4211 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4212 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4213 Node* obj = NULL;
4214 if (!is_static) {
4215 // Check for hashing null object
4216 obj = null_check_receiver();
4217 if (stopped()) return true; // unconditionally null
4218 result_reg->init_req(_null_path, top());
4219 result_val->init_req(_null_path, top());
4220 } else {
4221 // Do a null check, and return zero if null.
4222 // System.identityHashCode(null) == 0
4223 obj = argument(0);
4224 Node* null_ctl = top();
4225 obj = null_check_oop(obj, &null_ctl);
4226 result_reg->init_req(_null_path, null_ctl);
4227 result_val->init_req(_null_path, _gvn.intcon(0));
4228 }
4229
4230 // Unconditionally null? Then return right away.
4231 if (stopped()) {
4232 set_control( result_reg->in(_null_path));
4233 if (!stopped())
4234 set_result(result_val->in(_null_path));
4235 return true;
4236 }
4237
4238 // We only go to the fast case code if we pass a number of guards. The
4239 // paths which do not pass are accumulated in the slow_region.
4240 RegionNode* slow_region = new RegionNode(1);
4241 record_for_igvn(slow_region);
4242
4243 // If this is a virtual call, we generate a funny guard. We pull out
4244 // the vtable entry corresponding to hashCode() from the target object.
4245 // If the target method which we are calling happens to be the native
4246 // Object hashCode() method, we pass the guard. We do not need this
4247 // guard for non-virtual calls -- the caller is known to be the native
4248 // Object hashCode().
4249 if (is_virtual) {
4250 // After null check, get the object's klass.
4251 Node* obj_klass = load_object_klass(obj);
4252 generate_virtual_guard(obj_klass, slow_region);
4253 }
4254
4255 // Get the header out of the object, use LoadMarkNode when available
4256 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4257 // The control of the load must be NULL. Otherwise, the load can move before
4258 // the null check after castPP removal.
4259 Node* no_ctrl = NULL;
4260 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4261
4262 // Test the header to see if it is unlocked.
4263 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4264 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4265 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4266 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4267 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4268
4269 generate_slow_guard(test_unlocked, slow_region);
4270
4271 // Get the hash value and check to see that it has been properly assigned.
4272 // We depend on hash_mask being at most 32 bits and avoid the use of
4273 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4274 // vm: see markOop.hpp.
4275 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4276 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4277 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4278 // This hack lets the hash bits live anywhere in the mark object now, as long
4279 // as the shift drops the relevant bits into the low 32 bits. Note that
4280 // Java spec says that HashCode is an int so there's no point in capturing
4281 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4282 hshifted_header = ConvX2I(hshifted_header);
4283 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4284
4285 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4286 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4287 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4288
4289 generate_slow_guard(test_assigned, slow_region);
4290
4291 Node* init_mem = reset_memory();
4292 // fill in the rest of the null path:
4293 result_io ->init_req(_null_path, i_o());
4294 result_mem->init_req(_null_path, init_mem);
4295
4296 result_val->init_req(_fast_path, hash_val);
4297 result_reg->init_req(_fast_path, control());
4298 result_io ->init_req(_fast_path, i_o());
4299 result_mem->init_req(_fast_path, init_mem);
4300
4301 // Generate code for the slow case. We make a call to hashCode().
4302 set_control(_gvn.transform(slow_region));
4303 if (!stopped()) {
4304 // No need for PreserveJVMState, because we're using up the present state.
4305 set_all_memory(init_mem);
4306 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4307 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4308 Node* slow_result = set_results_for_java_call(slow_call);
4309 // this->control() comes from set_results_for_java_call
4310 result_reg->init_req(_slow_path, control());
4311 result_val->init_req(_slow_path, slow_result);
4312 result_io ->set_req(_slow_path, i_o());
4313 result_mem ->set_req(_slow_path, reset_memory());
4314 }
4315
4316 // Return the combined state.
4317 set_i_o( _gvn.transform(result_io) );
4318 set_all_memory( _gvn.transform(result_mem));
4319
4320 set_result(result_reg, result_val);
4321 return true;
4322 }
4323
4324 //---------------------------inline_native_getClass----------------------------
4325 // public final native Class<?> java.lang.Object.getClass();
4326 //
4327 // Build special case code for calls to getClass on an object.
4328 bool LibraryCallKit::inline_native_getClass() {
4329 Node* obj = null_check_receiver();
4330 if (stopped()) return true;
4331 set_result(load_mirror_from_klass(load_object_klass(obj)));
4332 return true;
4333 }
4334
4335 //-----------------inline_native_Reflection_getCallerClass---------------------
4336 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4337 //
4338 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4339 //
4340 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4341 // in that it must skip particular security frames and checks for
4342 // caller sensitive methods.
4343 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4344 #ifndef PRODUCT
4345 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4346 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4347 }
4348 #endif
4349
4350 if (!jvms()->has_method()) {
4351 #ifndef PRODUCT
4352 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4353 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4354 }
4355 #endif
4356 return false;
4357 }
4358
4359 // Walk back up the JVM state to find the caller at the required
4360 // depth.
4361 JVMState* caller_jvms = jvms();
4362
4363 // Cf. JVM_GetCallerClass
4364 // NOTE: Start the loop at depth 1 because the current JVM state does
4365 // not include the Reflection.getCallerClass() frame.
4366 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4367 ciMethod* m = caller_jvms->method();
4368 switch (n) {
4369 case 0:
4370 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4371 break;
4372 case 1:
4373 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4374 if (!m->caller_sensitive()) {
4375 #ifndef PRODUCT
4376 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4377 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4378 }
4379 #endif
4380 return false; // bail-out; let JVM_GetCallerClass do the work
4381 }
4382 break;
4383 default:
4384 if (!m->is_ignored_by_security_stack_walk()) {
4385 // We have reached the desired frame; return the holder class.
4386 // Acquire method holder as java.lang.Class and push as constant.
4387 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4388 ciInstance* caller_mirror = caller_klass->java_mirror();
4389 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4390
4391 #ifndef PRODUCT
4392 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4393 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
4394 tty->print_cr(" JVM state at this point:");
4395 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4396 ciMethod* m = jvms()->of_depth(i)->method();
4397 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4398 }
4399 }
4400 #endif
4401 return true;
4402 }
4403 break;
4404 }
4405 }
4406
4407 #ifndef PRODUCT
4408 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4409 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4410 tty->print_cr(" JVM state at this point:");
4411 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4412 ciMethod* m = jvms()->of_depth(i)->method();
4413 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4414 }
4415 }
4416 #endif
4417
4418 return false; // bail-out; let JVM_GetCallerClass do the work
4419 }
4420
4421 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4422 Node* arg = argument(0);
4423 Node* result = NULL;
4424
4425 switch (id) {
4426 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4427 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4428 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4429 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4430
4431 case vmIntrinsics::_doubleToLongBits: {
4432 // two paths (plus control) merge in a wood
4433 RegionNode *r = new RegionNode(3);
4434 Node *phi = new PhiNode(r, TypeLong::LONG);
4435
4436 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4437 // Build the boolean node
4438 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4439
4440 // Branch either way.
4441 // NaN case is less traveled, which makes all the difference.
4442 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4443 Node *opt_isnan = _gvn.transform(ifisnan);
4444 assert( opt_isnan->is_If(), "Expect an IfNode");
4445 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4446 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4447
4448 set_control(iftrue);
4449
4450 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4451 Node *slow_result = longcon(nan_bits); // return NaN
4452 phi->init_req(1, _gvn.transform( slow_result ));
4453 r->init_req(1, iftrue);
4454
4455 // Else fall through
4456 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4457 set_control(iffalse);
4458
4459 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4460 r->init_req(2, iffalse);
4461
4462 // Post merge
4463 set_control(_gvn.transform(r));
4464 record_for_igvn(r);
4465
4466 C->set_has_split_ifs(true); // Has chance for split-if optimization
4467 result = phi;
4468 assert(result->bottom_type()->isa_long(), "must be");
4469 break;
4470 }
4471
4472 case vmIntrinsics::_floatToIntBits: {
4473 // two paths (plus control) merge in a wood
4474 RegionNode *r = new RegionNode(3);
4475 Node *phi = new PhiNode(r, TypeInt::INT);
4476
4477 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4478 // Build the boolean node
4479 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4480
4481 // Branch either way.
4482 // NaN case is less traveled, which makes all the difference.
4483 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4484 Node *opt_isnan = _gvn.transform(ifisnan);
4485 assert( opt_isnan->is_If(), "Expect an IfNode");
4486 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4487 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4488
4489 set_control(iftrue);
4490
4491 static const jint nan_bits = 0x7fc00000;
4492 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4493 phi->init_req(1, _gvn.transform( slow_result ));
4494 r->init_req(1, iftrue);
4495
4496 // Else fall through
4497 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4498 set_control(iffalse);
4499
4500 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4501 r->init_req(2, iffalse);
4502
4503 // Post merge
4504 set_control(_gvn.transform(r));
4505 record_for_igvn(r);
4506
4507 C->set_has_split_ifs(true); // Has chance for split-if optimization
4508 result = phi;
4509 assert(result->bottom_type()->isa_int(), "must be");
4510 break;
4511 }
4512
4513 default:
4514 fatal_unexpected_iid(id);
4515 break;
4516 }
4517 set_result(_gvn.transform(result));
4518 return true;
4519 }
4520
4521 //----------------------inline_unsafe_copyMemory-------------------------
4522 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4523 bool LibraryCallKit::inline_unsafe_copyMemory() {
4524 if (callee()->is_static()) return false; // caller must have the capability!
4525 null_check_receiver(); // null-check receiver
4526 if (stopped()) return true;
4527
4528 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4529
4530 Node* src_ptr = argument(1); // type: oop
4531 Node* src_off = ConvL2X(argument(2)); // type: long
4532 Node* dst_ptr = argument(4); // type: oop
4533 Node* dst_off = ConvL2X(argument(5)); // type: long
4534 Node* size = ConvL2X(argument(7)); // type: long
4535
4536 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4537 "fieldOffset must be byte-scaled");
4538
4539 Node* src = make_unsafe_address(src_ptr, src_off);
4540 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4541
4542 // Conservatively insert a memory barrier on all memory slices.
4543 // Do not let writes of the copy source or destination float below the copy.
4544 insert_mem_bar(Op_MemBarCPUOrder);
4545
4546 // Call it. Note that the length argument is not scaled.
4547 make_runtime_call(RC_LEAF|RC_NO_FP,
4548 OptoRuntime::fast_arraycopy_Type(),
4549 StubRoutines::unsafe_arraycopy(),
4550 "unsafe_arraycopy",
4551 TypeRawPtr::BOTTOM,
4552 src, dst, size XTOP);
4553
4554 // Do not let reads of the copy destination float above the copy.
4555 insert_mem_bar(Op_MemBarCPUOrder);
4556
4557 return true;
4558 }
4559
4560 //------------------------clone_coping-----------------------------------
4561 // Helper function for inline_native_clone.
4562 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4563 assert(obj_size != NULL, "");
4564 Node* raw_obj = alloc_obj->in(1);
4565 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4566
4567 AllocateNode* alloc = NULL;
4568 if (ReduceBulkZeroing) {
4569 // We will be completely responsible for initializing this object -
4570 // mark Initialize node as complete.
4571 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4572 // The object was just allocated - there should be no any stores!
4573 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4574 // Mark as complete_with_arraycopy so that on AllocateNode
4575 // expansion, we know this AllocateNode is initialized by an array
4576 // copy and a StoreStore barrier exists after the array copy.
4577 alloc->initialization()->set_complete_with_arraycopy();
4578 }
4579
4580 // Copy the fastest available way.
4581 // TODO: generate fields copies for small objects instead.
4582 Node* src = obj;
4583 Node* dest = alloc_obj;
4584 Node* size = _gvn.transform(obj_size);
4585
4586 // Exclude the header but include array length to copy by 8 bytes words.
4587 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4588 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4589 instanceOopDesc::base_offset_in_bytes();
4590 // base_off:
4591 // 8 - 32-bit VM
4592 // 12 - 64-bit VM, compressed klass
4593 // 16 - 64-bit VM, normal klass
4594 if (base_off % BytesPerLong != 0) {
4595 assert(UseCompressedClassPointers, "");
4596 if (is_array) {
4597 // Exclude length to copy by 8 bytes words.
4598 base_off += sizeof(int);
4599 } else {
4600 // Include klass to copy by 8 bytes words.
4601 base_off = instanceOopDesc::klass_offset_in_bytes();
4602 }
4603 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4604 }
4605 src = basic_plus_adr(src, base_off);
4606 dest = basic_plus_adr(dest, base_off);
4607
4608 // Compute the length also, if needed:
4609 Node* countx = size;
4610 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4611 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4612
4613 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4614
4615 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false, false);
4616 ac->set_clonebasic();
4617 Node* n = _gvn.transform(ac);
4618 if (n == ac) {
4619 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4620 } else {
4621 set_all_memory(n);
4622 }
4623
4624 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4625 if (card_mark) {
4626 assert(!is_array, "");
4627 // Put in store barrier for any and all oops we are sticking
4628 // into this object. (We could avoid this if we could prove
4629 // that the object type contains no oop fields at all.)
4630 Node* no_particular_value = NULL;
4631 Node* no_particular_field = NULL;
4632 int raw_adr_idx = Compile::AliasIdxRaw;
4633 post_barrier(control(),
4634 memory(raw_adr_type),
4635 alloc_obj,
4636 no_particular_field,
4637 raw_adr_idx,
4638 no_particular_value,
4639 T_OBJECT,
4640 false);
4641 }
4642
4643 // Do not let reads from the cloned object float above the arraycopy.
4644 if (alloc != NULL) {
4645 // Do not let stores that initialize this object be reordered with
4646 // a subsequent store that would make this object accessible by
4647 // other threads.
4648 // Record what AllocateNode this StoreStore protects so that
4649 // escape analysis can go from the MemBarStoreStoreNode to the
4650 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4651 // based on the escape status of the AllocateNode.
4652 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4653 } else {
4654 insert_mem_bar(Op_MemBarCPUOrder);
4655 }
4656 }
4657
4658 //------------------------inline_native_clone----------------------------
4659 // protected native Object java.lang.Object.clone();
4660 //
4661 // Here are the simple edge cases:
4662 // null receiver => normal trap
4663 // virtual and clone was overridden => slow path to out-of-line clone
4664 // not cloneable or finalizer => slow path to out-of-line Object.clone
4665 //
4666 // The general case has two steps, allocation and copying.
4667 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4668 //
4669 // Copying also has two cases, oop arrays and everything else.
4670 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4671 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4672 //
4673 // These steps fold up nicely if and when the cloned object's klass
4674 // can be sharply typed as an object array, a type array, or an instance.
4675 //
4676 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4677 PhiNode* result_val;
4678
4679 // Set the reexecute bit for the interpreter to reexecute
4680 // the bytecode that invokes Object.clone if deoptimization happens.
4681 { PreserveReexecuteState preexecs(this);
4682 jvms()->set_should_reexecute(true);
4683
4684 Node* obj = null_check_receiver();
4685 if (stopped()) return true;
4686
4687 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4688
4689 // If we are going to clone an instance, we need its exact type to
4690 // know the number and types of fields to convert the clone to
4691 // loads/stores. Maybe a speculative type can help us.
4692 if (!obj_type->klass_is_exact() &&
4693 obj_type->speculative_type() != NULL &&
4694 obj_type->speculative_type()->is_instance_klass()) {
4695 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4696 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4697 !spec_ik->has_injected_fields()) {
4698 ciKlass* k = obj_type->klass();
4699 if (!k->is_instance_klass() ||
4700 k->as_instance_klass()->is_interface() ||
4701 k->as_instance_klass()->has_subklass()) {
4702 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4703 }
4704 }
4705 }
4706
4707 Node* obj_klass = load_object_klass(obj);
4708 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4709 const TypeOopPtr* toop = ((tklass != NULL)
4710 ? tklass->as_instance_type()
4711 : TypeInstPtr::NOTNULL);
4712
4713 // Conservatively insert a memory barrier on all memory slices.
4714 // Do not let writes into the original float below the clone.
4715 insert_mem_bar(Op_MemBarCPUOrder);
4716
4717 // paths into result_reg:
4718 enum {
4719 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4720 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4721 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4722 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4723 PATH_LIMIT
4724 };
4725 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4726 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4727 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4728 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4729 record_for_igvn(result_reg);
4730
4731 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4732 int raw_adr_idx = Compile::AliasIdxRaw;
4733
4734 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4735 if (array_ctl != NULL) {
4736 // It's an array.
4737 PreserveJVMState pjvms(this);
4738 set_control(array_ctl);
4739 Node* obj_length = load_array_length(obj);
4740 Node* obj_size = NULL;
4741 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4742
4743 if (!use_ReduceInitialCardMarks()) {
4744 // If it is an oop array, it requires very special treatment,
4745 // because card marking is required on each card of the array.
4746 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4747 if (is_obja != NULL) {
4748 PreserveJVMState pjvms2(this);
4749 set_control(is_obja);
4750 // Generate a direct call to the right arraycopy function(s).
4751 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4752 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4753 ac->set_cloneoop();
4754 Node* n = _gvn.transform(ac);
4755 assert(n == ac, "cannot disappear");
4756 ac->connect_outputs(this);
4757
4758 result_reg->init_req(_objArray_path, control());
4759 result_val->init_req(_objArray_path, alloc_obj);
4760 result_i_o ->set_req(_objArray_path, i_o());
4761 result_mem ->set_req(_objArray_path, reset_memory());
4762 }
4763 }
4764 // Otherwise, there are no card marks to worry about.
4765 // (We can dispense with card marks if we know the allocation
4766 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4767 // causes the non-eden paths to take compensating steps to
4768 // simulate a fresh allocation, so that no further
4769 // card marks are required in compiled code to initialize
4770 // the object.)
4771
4772 if (!stopped()) {
4773 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4774
4775 // Present the results of the copy.
4776 result_reg->init_req(_array_path, control());
4777 result_val->init_req(_array_path, alloc_obj);
4778 result_i_o ->set_req(_array_path, i_o());
4779 result_mem ->set_req(_array_path, reset_memory());
4780 }
4781 }
4782
4783 // We only go to the instance fast case code if we pass a number of guards.
4784 // The paths which do not pass are accumulated in the slow_region.
4785 RegionNode* slow_region = new RegionNode(1);
4786 record_for_igvn(slow_region);
4787 if (!stopped()) {
4788 // It's an instance (we did array above). Make the slow-path tests.
4789 // If this is a virtual call, we generate a funny guard. We grab
4790 // the vtable entry corresponding to clone() from the target object.
4791 // If the target method which we are calling happens to be the
4792 // Object clone() method, we pass the guard. We do not need this
4793 // guard for non-virtual calls; the caller is known to be the native
4794 // Object clone().
4795 if (is_virtual) {
4796 generate_virtual_guard(obj_klass, slow_region);
4797 }
4798
4799 // The object must be easily cloneable and must not have a finalizer.
4800 // Both of these conditions may be checked in a single test.
4801 // We could optimize the test further, but we don't care.
4802 generate_access_flags_guard(obj_klass,
4803 // Test both conditions:
4804 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4805 // Must be cloneable but not finalizer:
4806 JVM_ACC_IS_CLONEABLE_FAST,
4807 slow_region);
4808 }
4809
4810 if (!stopped()) {
4811 // It's an instance, and it passed the slow-path tests.
4812 PreserveJVMState pjvms(this);
4813 Node* obj_size = NULL;
4814 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4815 // is reexecuted if deoptimization occurs and there could be problems when merging
4816 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4817 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4818
4819 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4820
4821 // Present the results of the slow call.
4822 result_reg->init_req(_instance_path, control());
4823 result_val->init_req(_instance_path, alloc_obj);
4824 result_i_o ->set_req(_instance_path, i_o());
4825 result_mem ->set_req(_instance_path, reset_memory());
4826 }
4827
4828 // Generate code for the slow case. We make a call to clone().
4829 set_control(_gvn.transform(slow_region));
4830 if (!stopped()) {
4831 PreserveJVMState pjvms(this);
4832 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4833 Node* slow_result = set_results_for_java_call(slow_call);
4834 // this->control() comes from set_results_for_java_call
4835 result_reg->init_req(_slow_path, control());
4836 result_val->init_req(_slow_path, slow_result);
4837 result_i_o ->set_req(_slow_path, i_o());
4838 result_mem ->set_req(_slow_path, reset_memory());
4839 }
4840
4841 // Return the combined state.
4842 set_control( _gvn.transform(result_reg));
4843 set_i_o( _gvn.transform(result_i_o));
4844 set_all_memory( _gvn.transform(result_mem));
4845 } // original reexecute is set back here
4846
4847 set_result(_gvn.transform(result_val));
4848 return true;
4849 }
4850
4851 // If we have a tighly coupled allocation, the arraycopy may take care
4852 // of the array initialization. If one of the guards we insert between
4853 // the allocation and the arraycopy causes a deoptimization, an
4854 // unitialized array will escape the compiled method. To prevent that
4855 // we set the JVM state for uncommon traps between the allocation and
4856 // the arraycopy to the state before the allocation so, in case of
4857 // deoptimization, we'll reexecute the allocation and the
4858 // initialization.
4859 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4860 if (alloc != NULL) {
4861 ciMethod* trap_method = alloc->jvms()->method();
4862 int trap_bci = alloc->jvms()->bci();
4863
4864 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4865 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4866 // Make sure there's no store between the allocation and the
4867 // arraycopy otherwise visible side effects could be rexecuted
4868 // in case of deoptimization and cause incorrect execution.
4869 bool no_interfering_store = true;
4870 Node* mem = alloc->in(TypeFunc::Memory);
4871 if (mem->is_MergeMem()) {
4872 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4873 Node* n = mms.memory();
4874 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4875 assert(n->is_Store(), "what else?");
4876 no_interfering_store = false;
4877 break;
4878 }
4879 }
4880 } else {
4881 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4882 Node* n = mms.memory();
4883 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4884 assert(n->is_Store(), "what else?");
4885 no_interfering_store = false;
4886 break;
4887 }
4888 }
4889 }
4890
4891 if (no_interfering_store) {
4892 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4893 uint size = alloc->req();
4894 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4895 old_jvms->set_map(sfpt);
4896 for (uint i = 0; i < size; i++) {
4897 sfpt->init_req(i, alloc->in(i));
4898 }
4899 // re-push array length for deoptimization
4900 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4901 old_jvms->set_sp(old_jvms->sp()+1);
4902 old_jvms->set_monoff(old_jvms->monoff()+1);
4903 old_jvms->set_scloff(old_jvms->scloff()+1);
4904 old_jvms->set_endoff(old_jvms->endoff()+1);
4905 old_jvms->set_should_reexecute(true);
4906
4907 sfpt->set_i_o(map()->i_o());
4908 sfpt->set_memory(map()->memory());
4909 sfpt->set_control(map()->control());
4910
4911 JVMState* saved_jvms = jvms();
4912 saved_reexecute_sp = _reexecute_sp;
4913
4914 set_jvms(sfpt->jvms());
4915 _reexecute_sp = jvms()->sp();
4916
4917 return saved_jvms;
4918 }
4919 }
4920 }
4921 return NULL;
4922 }
4923
4924 // In case of a deoptimization, we restart execution at the
4925 // allocation, allocating a new array. We would leave an uninitialized
4926 // array in the heap that GCs wouldn't expect. Move the allocation
4927 // after the traps so we don't allocate the array if we
4928 // deoptimize. This is possible because tightly_coupled_allocation()
4929 // guarantees there's no observer of the allocated array at this point
4930 // and the control flow is simple enough.
4931 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4932 int saved_reexecute_sp, uint new_idx) {
4933 if (saved_jvms != NULL && !stopped()) {
4934 assert(alloc != NULL, "only with a tightly coupled allocation");
4935 // restore JVM state to the state at the arraycopy
4936 saved_jvms->map()->set_control(map()->control());
4937 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4938 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4939 // If we've improved the types of some nodes (null check) while
4940 // emitting the guards, propagate them to the current state
4941 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4942 set_jvms(saved_jvms);
4943 _reexecute_sp = saved_reexecute_sp;
4944
4945 // Remove the allocation from above the guards
4946 CallProjections callprojs;
4947 alloc->extract_projections(&callprojs, true);
4948 InitializeNode* init = alloc->initialization();
4949 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4950 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4951 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4952 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4953
4954 // move the allocation here (after the guards)
4955 _gvn.hash_delete(alloc);
4956 alloc->set_req(TypeFunc::Control, control());
4957 alloc->set_req(TypeFunc::I_O, i_o());
4958 Node *mem = reset_memory();
4959 set_all_memory(mem);
4960 alloc->set_req(TypeFunc::Memory, mem);
4961 set_control(init->proj_out(TypeFunc::Control));
4962 set_i_o(callprojs.fallthrough_ioproj);
4963
4964 // Update memory as done in GraphKit::set_output_for_allocation()
4965 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4966 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4967 if (ary_type->isa_aryptr() && length_type != NULL) {
4968 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4969 }
4970 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4971 int elemidx = C->get_alias_index(telemref);
4972 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4973 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4974
4975 Node* allocx = _gvn.transform(alloc);
4976 assert(allocx == alloc, "where has the allocation gone?");
4977 assert(dest->is_CheckCastPP(), "not an allocation result?");
4978
4979 _gvn.hash_delete(dest);
4980 dest->set_req(0, control());
4981 Node* destx = _gvn.transform(dest);
4982 assert(destx == dest, "where has the allocation result gone?");
4983 }
4984 }
4985
4986
4987 //------------------------------inline_arraycopy-----------------------
4988 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4989 // Object dest, int destPos,
4990 // int length);
4991 bool LibraryCallKit::inline_arraycopy() {
4992 // Get the arguments.
4993 Node* src = argument(0); // type: oop
4994 Node* src_offset = argument(1); // type: int
4995 Node* dest = argument(2); // type: oop
4996 Node* dest_offset = argument(3); // type: int
4997 Node* length = argument(4); // type: int
4998
4999 uint new_idx = C->unique();
5000
5001 // Check for allocation before we add nodes that would confuse
5002 // tightly_coupled_allocation()
5003 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
5004
5005 int saved_reexecute_sp = -1;
5006 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5007 // See arraycopy_restore_alloc_state() comment
5008 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5009 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5010 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5011 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5012
5013 // The following tests must be performed
5014 // (1) src and dest are arrays.
5015 // (2) src and dest arrays must have elements of the same BasicType
5016 // (3) src and dest must not be null.
5017 // (4) src_offset must not be negative.
5018 // (5) dest_offset must not be negative.
5019 // (6) length must not be negative.
5020 // (7) src_offset + length must not exceed length of src.
5021 // (8) dest_offset + length must not exceed length of dest.
5022 // (9) each element of an oop array must be assignable
5023
5024 // (3) src and dest must not be null.
5025 // always do this here because we need the JVM state for uncommon traps
5026 Node* null_ctl = top();
5027 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5028 assert(null_ctl->is_top(), "no null control here");
5029 dest = null_check(dest, T_ARRAY);
5030
5031 if (!can_emit_guards) {
5032 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5033 // guards but the arraycopy node could still take advantage of a
5034 // tightly allocated allocation. tightly_coupled_allocation() is
5035 // called again to make sure it takes the null check above into
5036 // account: the null check is mandatory and if it caused an
5037 // uncommon trap to be emitted then the allocation can't be
5038 // considered tightly coupled in this context.
5039 alloc = tightly_coupled_allocation(dest, NULL);
5040 }
5041
5042 bool validated = false;
5043
5044 const Type* src_type = _gvn.type(src);
5045 const Type* dest_type = _gvn.type(dest);
5046 const TypeAryPtr* top_src = src_type->isa_aryptr();
5047 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5048
5049 // Do we have the type of src?
5050 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5051 // Do we have the type of dest?
5052 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5053 // Is the type for src from speculation?
5054 bool src_spec = false;
5055 // Is the type for dest from speculation?
5056 bool dest_spec = false;
5057
5058 if ((!has_src || !has_dest) && can_emit_guards) {
5059 // We don't have sufficient type information, let's see if
5060 // speculative types can help. We need to have types for both src
5061 // and dest so that it pays off.
5062
5063 // Do we already have or could we have type information for src
5064 bool could_have_src = has_src;
5065 // Do we already have or could we have type information for dest
5066 bool could_have_dest = has_dest;
5067
5068 ciKlass* src_k = NULL;
5069 if (!has_src) {
5070 src_k = src_type->speculative_type_not_null();
5071 if (src_k != NULL && src_k->is_array_klass()) {
5072 could_have_src = true;
5073 }
5074 }
5075
5076 ciKlass* dest_k = NULL;
5077 if (!has_dest) {
5078 dest_k = dest_type->speculative_type_not_null();
5079 if (dest_k != NULL && dest_k->is_array_klass()) {
5080 could_have_dest = true;
5081 }
5082 }
5083
5084 if (could_have_src && could_have_dest) {
5085 // This is going to pay off so emit the required guards
5086 if (!has_src) {
5087 src = maybe_cast_profiled_obj(src, src_k, true);
5088 src_type = _gvn.type(src);
5089 top_src = src_type->isa_aryptr();
5090 has_src = (top_src != NULL && top_src->klass() != NULL);
5091 src_spec = true;
5092 }
5093 if (!has_dest) {
5094 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5095 dest_type = _gvn.type(dest);
5096 top_dest = dest_type->isa_aryptr();
5097 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5098 dest_spec = true;
5099 }
5100 }
5101 }
5102
5103 if (has_src && has_dest && can_emit_guards) {
5104 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5105 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5106 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5107 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5108
5109 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5110 // If both arrays are object arrays then having the exact types
5111 // for both will remove the need for a subtype check at runtime
5112 // before the call and may make it possible to pick a faster copy
5113 // routine (without a subtype check on every element)
5114 // Do we have the exact type of src?
5115 bool could_have_src = src_spec;
5116 // Do we have the exact type of dest?
5117 bool could_have_dest = dest_spec;
5118 ciKlass* src_k = top_src->klass();
5119 ciKlass* dest_k = top_dest->klass();
5120 if (!src_spec) {
5121 src_k = src_type->speculative_type_not_null();
5122 if (src_k != NULL && src_k->is_array_klass()) {
5123 could_have_src = true;
5124 }
5125 }
5126 if (!dest_spec) {
5127 dest_k = dest_type->speculative_type_not_null();
5128 if (dest_k != NULL && dest_k->is_array_klass()) {
5129 could_have_dest = true;
5130 }
5131 }
5132 if (could_have_src && could_have_dest) {
5133 // If we can have both exact types, emit the missing guards
5134 if (could_have_src && !src_spec) {
5135 src = maybe_cast_profiled_obj(src, src_k, true);
5136 }
5137 if (could_have_dest && !dest_spec) {
5138 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5139 }
5140 }
5141 }
5142 }
5143
5144 ciMethod* trap_method = method();
5145 int trap_bci = bci();
5146 if (saved_jvms != NULL) {
5147 trap_method = alloc->jvms()->method();
5148 trap_bci = alloc->jvms()->bci();
5149 }
5150
5151 bool negative_length_guard_generated = false;
5152
5153 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5154 can_emit_guards &&
5155 !src->is_top() && !dest->is_top()) {
5156 // validate arguments: enables transformation the ArrayCopyNode
5157 validated = true;
5158
5159 RegionNode* slow_region = new RegionNode(1);
5160 record_for_igvn(slow_region);
5161
5162 // (1) src and dest are arrays.
5163 generate_non_array_guard(load_object_klass(src), slow_region);
5164 generate_non_array_guard(load_object_klass(dest), slow_region);
5165
5166 // (2) src and dest arrays must have elements of the same BasicType
5167 // done at macro expansion or at Ideal transformation time
5168
5169 // (4) src_offset must not be negative.
5170 generate_negative_guard(src_offset, slow_region);
5171
5172 // (5) dest_offset must not be negative.
5173 generate_negative_guard(dest_offset, slow_region);
5174
5175 // (7) src_offset + length must not exceed length of src.
5176 generate_limit_guard(src_offset, length,
5177 load_array_length(src),
5178 slow_region);
5179
5180 // (8) dest_offset + length must not exceed length of dest.
5181 generate_limit_guard(dest_offset, length,
5182 load_array_length(dest),
5183 slow_region);
5184
5185 // (6) length must not be negative.
5186 // This is also checked in generate_arraycopy() during macro expansion, but
5187 // we also have to check it here for the case where the ArrayCopyNode will
5188 // be eliminated by Escape Analysis.
5189 if (EliminateAllocations) {
5190 generate_negative_guard(length, slow_region);
5191 negative_length_guard_generated = true;
5192 }
5193
5194 // (9) each element of an oop array must be assignable
5195 Node* src_klass = load_object_klass(src);
5196 Node* dest_klass = load_object_klass(dest);
5197 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5198
5199 if (not_subtype_ctrl != top()) {
5200 PreserveJVMState pjvms(this);
5201 set_control(not_subtype_ctrl);
5202 uncommon_trap(Deoptimization::Reason_intrinsic,
5203 Deoptimization::Action_make_not_entrant);
5204 assert(stopped(), "Should be stopped");
5205 }
5206 {
5207 PreserveJVMState pjvms(this);
5208 set_control(_gvn.transform(slow_region));
5209 uncommon_trap(Deoptimization::Reason_intrinsic,
5210 Deoptimization::Action_make_not_entrant);
5211 assert(stopped(), "Should be stopped");
5212 }
5213 }
5214
5215 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5216
5217 if (stopped()) {
5218 return true;
5219 }
5220
5221 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5222 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5223 // so the compiler has a chance to eliminate them: during macro expansion,
5224 // we have to set their control (CastPP nodes are eliminated).
5225 load_object_klass(src), load_object_klass(dest),
5226 load_array_length(src), load_array_length(dest));
5227
5228 ac->set_arraycopy(validated);
5229
5230 Node* n = _gvn.transform(ac);
5231 if (n == ac) {
5232 ac->connect_outputs(this);
5233 } else {
5234 assert(validated, "shouldn't transform if all arguments not validated");
5235 set_all_memory(n);
5236 }
5237
5238 return true;
5239 }
5240
5241
5242 // Helper function which determines if an arraycopy immediately follows
5243 // an allocation, with no intervening tests or other escapes for the object.
5244 AllocateArrayNode*
5245 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5246 RegionNode* slow_region) {
5247 if (stopped()) return NULL; // no fast path
5248 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5249
5250 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5251 if (alloc == NULL) return NULL;
5252
5253 Node* rawmem = memory(Compile::AliasIdxRaw);
5254 // Is the allocation's memory state untouched?
5255 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5256 // Bail out if there have been raw-memory effects since the allocation.
5257 // (Example: There might have been a call or safepoint.)
5258 return NULL;
5259 }
5260 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5261 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5262 return NULL;
5263 }
5264
5265 // There must be no unexpected observers of this allocation.
5266 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5267 Node* obs = ptr->fast_out(i);
5268 if (obs != this->map()) {
5269 return NULL;
5270 }
5271 }
5272
5273 // This arraycopy must unconditionally follow the allocation of the ptr.
5274 Node* alloc_ctl = ptr->in(0);
5275 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5276
5277 Node* ctl = control();
5278 while (ctl != alloc_ctl) {
5279 // There may be guards which feed into the slow_region.
5280 // Any other control flow means that we might not get a chance
5281 // to finish initializing the allocated object.
5282 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5283 IfNode* iff = ctl->in(0)->as_If();
5284 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5285 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5286 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5287 ctl = iff->in(0); // This test feeds the known slow_region.
5288 continue;
5289 }
5290 // One more try: Various low-level checks bottom out in
5291 // uncommon traps. If the debug-info of the trap omits
5292 // any reference to the allocation, as we've already
5293 // observed, then there can be no objection to the trap.
5294 bool found_trap = false;
5295 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5296 Node* obs = not_ctl->fast_out(j);
5297 if (obs->in(0) == not_ctl && obs->is_Call() &&
5298 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5299 found_trap = true; break;
5300 }
5301 }
5302 if (found_trap) {
5303 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5304 continue;
5305 }
5306 }
5307 return NULL;
5308 }
5309
5310 // If we get this far, we have an allocation which immediately
5311 // precedes the arraycopy, and we can take over zeroing the new object.
5312 // The arraycopy will finish the initialization, and provide
5313 // a new control state to which we will anchor the destination pointer.
5314
5315 return alloc;
5316 }
5317
5318 //-------------inline_encodeISOArray-----------------------------------
5319 // encode char[] to byte[] in ISO_8859_1
5320 bool LibraryCallKit::inline_encodeISOArray() {
5321 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5322 // no receiver since it is static method
5323 Node *src = argument(0);
5324 Node *src_offset = argument(1);
5325 Node *dst = argument(2);
5326 Node *dst_offset = argument(3);
5327 Node *length = argument(4);
5328
5329 const Type* src_type = src->Value(&_gvn);
5330 const Type* dst_type = dst->Value(&_gvn);
5331 const TypeAryPtr* top_src = src_type->isa_aryptr();
5332 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5333 if (top_src == NULL || top_src->klass() == NULL ||
5334 top_dest == NULL || top_dest->klass() == NULL) {
5335 // failed array check
5336 return false;
5337 }
5338
5339 // Figure out the size and type of the elements we will be copying.
5340 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5341 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5342 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5343 return false;
5344 }
5345
5346 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5347 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5348 // 'src_start' points to src array + scaled offset
5349 // 'dst_start' points to dst array + scaled offset
5350
5351 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5352 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5353 enc = _gvn.transform(enc);
5354 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5355 set_memory(res_mem, mtype);
5356 set_result(enc);
5357 return true;
5358 }
5359
5360 //-------------inline_multiplyToLen-----------------------------------
5361 bool LibraryCallKit::inline_multiplyToLen() {
5362 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5363
5364 address stubAddr = StubRoutines::multiplyToLen();
5365 if (stubAddr == NULL) {
5366 return false; // Intrinsic's stub is not implemented on this platform
5367 }
5368 const char* stubName = "multiplyToLen";
5369
5370 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5371
5372 // no receiver because it is a static method
5373 Node* x = argument(0);
5374 Node* xlen = argument(1);
5375 Node* y = argument(2);
5376 Node* ylen = argument(3);
5377 Node* z = argument(4);
5378
5379 const Type* x_type = x->Value(&_gvn);
5380 const Type* y_type = y->Value(&_gvn);
5381 const TypeAryPtr* top_x = x_type->isa_aryptr();
5382 const TypeAryPtr* top_y = y_type->isa_aryptr();
5383 if (top_x == NULL || top_x->klass() == NULL ||
5384 top_y == NULL || top_y->klass() == NULL) {
5385 // failed array check
5386 return false;
5387 }
5388
5389 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5390 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5391 if (x_elem != T_INT || y_elem != T_INT) {
5392 return false;
5393 }
5394
5395 // Set the original stack and the reexecute bit for the interpreter to reexecute
5396 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5397 // on the return from z array allocation in runtime.
5398 { PreserveReexecuteState preexecs(this);
5399 jvms()->set_should_reexecute(true);
5400
5401 Node* x_start = array_element_address(x, intcon(0), x_elem);
5402 Node* y_start = array_element_address(y, intcon(0), y_elem);
5403 // 'x_start' points to x array + scaled xlen
5404 // 'y_start' points to y array + scaled ylen
5405
5406 // Allocate the result array
5407 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5408 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5409 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5410
5411 IdealKit ideal(this);
5412
5413 #define __ ideal.
5414 Node* one = __ ConI(1);
5415 Node* zero = __ ConI(0);
5416 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5417 __ set(need_alloc, zero);
5418 __ set(z_alloc, z);
5419 __ if_then(z, BoolTest::eq, null()); {
5420 __ increment (need_alloc, one);
5421 } __ else_(); {
5422 // Update graphKit memory and control from IdealKit.
5423 sync_kit(ideal);
5424 Node* zlen_arg = load_array_length(z);
5425 // Update IdealKit memory and control from graphKit.
5426 __ sync_kit(this);
5427 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5428 __ increment (need_alloc, one);
5429 } __ end_if();
5430 } __ end_if();
5431
5432 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5433 // Update graphKit memory and control from IdealKit.
5434 sync_kit(ideal);
5435 Node * narr = new_array(klass_node, zlen, 1);
5436 // Update IdealKit memory and control from graphKit.
5437 __ sync_kit(this);
5438 __ set(z_alloc, narr);
5439 } __ end_if();
5440
5441 sync_kit(ideal);
5442 z = __ value(z_alloc);
5443 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5444 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5445 // Final sync IdealKit and GraphKit.
5446 final_sync(ideal);
5447 #undef __
5448
5449 Node* z_start = array_element_address(z, intcon(0), T_INT);
5450
5451 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5452 OptoRuntime::multiplyToLen_Type(),
5453 stubAddr, stubName, TypePtr::BOTTOM,
5454 x_start, xlen, y_start, ylen, z_start, zlen);
5455 } // original reexecute is set back here
5456
5457 C->set_has_split_ifs(true); // Has chance for split-if optimization
5458 set_result(z);
5459 return true;
5460 }
5461
5462 //-------------inline_squareToLen------------------------------------
5463 bool LibraryCallKit::inline_squareToLen() {
5464 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5465
5466 address stubAddr = StubRoutines::squareToLen();
5467 if (stubAddr == NULL) {
5468 return false; // Intrinsic's stub is not implemented on this platform
5469 }
5470 const char* stubName = "squareToLen";
5471
5472 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5473
5474 Node* x = argument(0);
5475 Node* len = argument(1);
5476 Node* z = argument(2);
5477 Node* zlen = argument(3);
5478
5479 const Type* x_type = x->Value(&_gvn);
5480 const Type* z_type = z->Value(&_gvn);
5481 const TypeAryPtr* top_x = x_type->isa_aryptr();
5482 const TypeAryPtr* top_z = z_type->isa_aryptr();
5483 if (top_x == NULL || top_x->klass() == NULL ||
5484 top_z == NULL || top_z->klass() == NULL) {
5485 // failed array check
5486 return false;
5487 }
5488
5489 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5490 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5491 if (x_elem != T_INT || z_elem != T_INT) {
5492 return false;
5493 }
5494
5495
5496 Node* x_start = array_element_address(x, intcon(0), x_elem);
5497 Node* z_start = array_element_address(z, intcon(0), z_elem);
5498
5499 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5500 OptoRuntime::squareToLen_Type(),
5501 stubAddr, stubName, TypePtr::BOTTOM,
5502 x_start, len, z_start, zlen);
5503
5504 set_result(z);
5505 return true;
5506 }
5507
5508 //-------------inline_mulAdd------------------------------------------
5509 bool LibraryCallKit::inline_mulAdd() {
5510 assert(UseMulAddIntrinsic, "not implemented on this platform");
5511
5512 address stubAddr = StubRoutines::mulAdd();
5513 if (stubAddr == NULL) {
5514 return false; // Intrinsic's stub is not implemented on this platform
5515 }
5516 const char* stubName = "mulAdd";
5517
5518 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5519
5520 Node* out = argument(0);
5521 Node* in = argument(1);
5522 Node* offset = argument(2);
5523 Node* len = argument(3);
5524 Node* k = argument(4);
5525
5526 const Type* out_type = out->Value(&_gvn);
5527 const Type* in_type = in->Value(&_gvn);
5528 const TypeAryPtr* top_out = out_type->isa_aryptr();
5529 const TypeAryPtr* top_in = in_type->isa_aryptr();
5530 if (top_out == NULL || top_out->klass() == NULL ||
5531 top_in == NULL || top_in->klass() == NULL) {
5532 // failed array check
5533 return false;
5534 }
5535
5536 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5537 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5538 if (out_elem != T_INT || in_elem != T_INT) {
5539 return false;
5540 }
5541
5542 Node* outlen = load_array_length(out);
5543 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5544 Node* out_start = array_element_address(out, intcon(0), out_elem);
5545 Node* in_start = array_element_address(in, intcon(0), in_elem);
5546
5547 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5548 OptoRuntime::mulAdd_Type(),
5549 stubAddr, stubName, TypePtr::BOTTOM,
5550 out_start,in_start, new_offset, len, k);
5551 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5552 set_result(result);
5553 return true;
5554 }
5555
5556 //-------------inline_montgomeryMultiply-----------------------------------
5557 bool LibraryCallKit::inline_montgomeryMultiply() {
5558 address stubAddr = StubRoutines::montgomeryMultiply();
5559 if (stubAddr == NULL) {
5560 return false; // Intrinsic's stub is not implemented on this platform
5561 }
5562
5563 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5564 const char* stubName = "montgomery_multiply";
5565
5566 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5567
5568 Node* a = argument(0);
5569 Node* b = argument(1);
5570 Node* n = argument(2);
5571 Node* len = argument(3);
5572 Node* inv = argument(4);
5573 Node* m = argument(6);
5574
5575 const Type* a_type = a->Value(&_gvn);
5576 const TypeAryPtr* top_a = a_type->isa_aryptr();
5577 const Type* b_type = b->Value(&_gvn);
5578 const TypeAryPtr* top_b = b_type->isa_aryptr();
5579 const Type* n_type = a->Value(&_gvn);
5580 const TypeAryPtr* top_n = n_type->isa_aryptr();
5581 const Type* m_type = a->Value(&_gvn);
5582 const TypeAryPtr* top_m = m_type->isa_aryptr();
5583 if (top_a == NULL || top_a->klass() == NULL ||
5584 top_b == NULL || top_b->klass() == NULL ||
5585 top_n == NULL || top_n->klass() == NULL ||
5586 top_m == NULL || top_m->klass() == NULL) {
5587 // failed array check
5588 return false;
5589 }
5590
5591 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5592 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5593 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5594 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5595 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5596 return false;
5597 }
5598
5599 // Make the call
5600 {
5601 Node* a_start = array_element_address(a, intcon(0), a_elem);
5602 Node* b_start = array_element_address(b, intcon(0), b_elem);
5603 Node* n_start = array_element_address(n, intcon(0), n_elem);
5604 Node* m_start = array_element_address(m, intcon(0), m_elem);
5605
5606 Node* call = make_runtime_call(RC_LEAF,
5607 OptoRuntime::montgomeryMultiply_Type(),
5608 stubAddr, stubName, TypePtr::BOTTOM,
5609 a_start, b_start, n_start, len, inv, top(),
5610 m_start);
5611 set_result(m);
5612 }
5613
5614 return true;
5615 }
5616
5617 bool LibraryCallKit::inline_montgomerySquare() {
5618 address stubAddr = StubRoutines::montgomerySquare();
5619 if (stubAddr == NULL) {
5620 return false; // Intrinsic's stub is not implemented on this platform
5621 }
5622
5623 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5624 const char* stubName = "montgomery_square";
5625
5626 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5627
5628 Node* a = argument(0);
5629 Node* n = argument(1);
5630 Node* len = argument(2);
5631 Node* inv = argument(3);
5632 Node* m = argument(5);
5633
5634 const Type* a_type = a->Value(&_gvn);
5635 const TypeAryPtr* top_a = a_type->isa_aryptr();
5636 const Type* n_type = a->Value(&_gvn);
5637 const TypeAryPtr* top_n = n_type->isa_aryptr();
5638 const Type* m_type = a->Value(&_gvn);
5639 const TypeAryPtr* top_m = m_type->isa_aryptr();
5640 if (top_a == NULL || top_a->klass() == NULL ||
5641 top_n == NULL || top_n->klass() == NULL ||
5642 top_m == NULL || top_m->klass() == NULL) {
5643 // failed array check
5644 return false;
5645 }
5646
5647 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5648 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5649 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5650 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5651 return false;
5652 }
5653
5654 // Make the call
5655 {
5656 Node* a_start = array_element_address(a, intcon(0), a_elem);
5657 Node* n_start = array_element_address(n, intcon(0), n_elem);
5658 Node* m_start = array_element_address(m, intcon(0), m_elem);
5659
5660 Node* call = make_runtime_call(RC_LEAF,
5661 OptoRuntime::montgomerySquare_Type(),
5662 stubAddr, stubName, TypePtr::BOTTOM,
5663 a_start, n_start, len, inv, top(),
5664 m_start);
5665 set_result(m);
5666 }
5667
5668 return true;
5669 }
5670
5671 //-------------inline_vectorizedMismatch------------------------------
5672 bool LibraryCallKit::inline_vectorizedMismatch() {
5673 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5674
5675 address stubAddr = StubRoutines::vectorizedMismatch();
5676 if (stubAddr == NULL) {
5677 return false; // Intrinsic's stub is not implemented on this platform
5678 }
5679 const char* stubName = "vectorizedMismatch";
5680 int size_l = callee()->signature()->size();
5681 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5682
5683 Node* obja = argument(0);
5684 Node* aoffset = argument(1);
5685 Node* objb = argument(3);
5686 Node* boffset = argument(4);
5687 Node* length = argument(6);
5688 Node* scale = argument(7);
5689
5690 const Type* a_type = obja->Value(&_gvn);
5691 const Type* b_type = objb->Value(&_gvn);
5692 const TypeAryPtr* top_a = a_type->isa_aryptr();
5693 const TypeAryPtr* top_b = b_type->isa_aryptr();
5694 if (top_a == NULL || top_a->klass() == NULL ||
5695 top_b == NULL || top_b->klass() == NULL) {
5696 // failed array check
5697 return false;
5698 }
5699
5700 Node* call;
5701 jvms()->set_should_reexecute(true);
5702
5703 Node* obja_adr = make_unsafe_address(obja, aoffset);
5704 Node* objb_adr = make_unsafe_address(objb, boffset);
5705
5706 call = make_runtime_call(RC_LEAF,
5707 OptoRuntime::vectorizedMismatch_Type(),
5708 stubAddr, stubName, TypePtr::BOTTOM,
5709 obja_adr, objb_adr, length, scale);
5710
5711 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5712 set_result(result);
5713 return true;
5714 }
5715
5716 /**
5717 * Calculate CRC32 for byte.
5718 * int java.util.zip.CRC32.update(int crc, int b)
5719 */
5720 bool LibraryCallKit::inline_updateCRC32() {
5721 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5722 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5723 // no receiver since it is static method
5724 Node* crc = argument(0); // type: int
5725 Node* b = argument(1); // type: int
5726
5727 /*
5728 * int c = ~ crc;
5729 * b = timesXtoThe32[(b ^ c) & 0xFF];
5730 * b = b ^ (c >>> 8);
5731 * crc = ~b;
5732 */
5733
5734 Node* M1 = intcon(-1);
5735 crc = _gvn.transform(new XorINode(crc, M1));
5736 Node* result = _gvn.transform(new XorINode(crc, b));
5737 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5738
5739 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5740 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5741 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5742 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5743
5744 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5745 result = _gvn.transform(new XorINode(crc, result));
5746 result = _gvn.transform(new XorINode(result, M1));
5747 set_result(result);
5748 return true;
5749 }
5750
5751 /**
5752 * Calculate CRC32 for byte[] array.
5753 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5754 */
5755 bool LibraryCallKit::inline_updateBytesCRC32() {
5756 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5757 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5758 // no receiver since it is static method
5759 Node* crc = argument(0); // type: int
5760 Node* src = argument(1); // type: oop
5761 Node* offset = argument(2); // type: int
5762 Node* length = argument(3); // type: int
5763
5764 const Type* src_type = src->Value(&_gvn);
5765 const TypeAryPtr* top_src = src_type->isa_aryptr();
5766 if (top_src == NULL || top_src->klass() == NULL) {
5767 // failed array check
5768 return false;
5769 }
5770
5771 // Figure out the size and type of the elements we will be copying.
5772 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5773 if (src_elem != T_BYTE) {
5774 return false;
5775 }
5776
5777 // 'src_start' points to src array + scaled offset
5778 Node* src_start = array_element_address(src, offset, src_elem);
5779
5780 // We assume that range check is done by caller.
5781 // TODO: generate range check (offset+length < src.length) in debug VM.
5782
5783 // Call the stub.
5784 address stubAddr = StubRoutines::updateBytesCRC32();
5785 const char *stubName = "updateBytesCRC32";
5786
5787 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5788 stubAddr, stubName, TypePtr::BOTTOM,
5789 crc, src_start, length);
5790 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5791 set_result(result);
5792 return true;
5793 }
5794
5795 /**
5796 * Calculate CRC32 for ByteBuffer.
5797 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5798 */
5799 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5800 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5801 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5802 // no receiver since it is static method
5803 Node* crc = argument(0); // type: int
5804 Node* src = argument(1); // type: long
5805 Node* offset = argument(3); // type: int
5806 Node* length = argument(4); // type: int
5807
5808 src = ConvL2X(src); // adjust Java long to machine word
5809 Node* base = _gvn.transform(new CastX2PNode(src));
5810 offset = ConvI2X(offset);
5811
5812 // 'src_start' points to src array + scaled offset
5813 Node* src_start = basic_plus_adr(top(), base, offset);
5814
5815 // Call the stub.
5816 address stubAddr = StubRoutines::updateBytesCRC32();
5817 const char *stubName = "updateBytesCRC32";
5818
5819 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5820 stubAddr, stubName, TypePtr::BOTTOM,
5821 crc, src_start, length);
5822 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5823 set_result(result);
5824 return true;
5825 }
5826
5827 //------------------------------get_table_from_crc32c_class-----------------------
5828 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5829 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5830 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5831
5832 return table;
5833 }
5834
5835 //------------------------------inline_updateBytesCRC32C-----------------------
5836 //
5837 // Calculate CRC32C for byte[] array.
5838 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5839 //
5840 bool LibraryCallKit::inline_updateBytesCRC32C() {
5841 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5842 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5843 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5844 // no receiver since it is a static method
5845 Node* crc = argument(0); // type: int
5846 Node* src = argument(1); // type: oop
5847 Node* offset = argument(2); // type: int
5848 Node* end = argument(3); // type: int
5849
5850 Node* length = _gvn.transform(new SubINode(end, offset));
5851
5852 const Type* src_type = src->Value(&_gvn);
5853 const TypeAryPtr* top_src = src_type->isa_aryptr();
5854 if (top_src == NULL || top_src->klass() == NULL) {
5855 // failed array check
5856 return false;
5857 }
5858
5859 // Figure out the size and type of the elements we will be copying.
5860 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5861 if (src_elem != T_BYTE) {
5862 return false;
5863 }
5864
5865 // 'src_start' points to src array + scaled offset
5866 Node* src_start = array_element_address(src, offset, src_elem);
5867
5868 // static final int[] byteTable in class CRC32C
5869 Node* table = get_table_from_crc32c_class(callee()->holder());
5870 Node* table_start = array_element_address(table, intcon(0), T_INT);
5871
5872 // We assume that range check is done by caller.
5873 // TODO: generate range check (offset+length < src.length) in debug VM.
5874
5875 // Call the stub.
5876 address stubAddr = StubRoutines::updateBytesCRC32C();
5877 const char *stubName = "updateBytesCRC32C";
5878
5879 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5880 stubAddr, stubName, TypePtr::BOTTOM,
5881 crc, src_start, length, table_start);
5882 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5883 set_result(result);
5884 return true;
5885 }
5886
5887 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5888 //
5889 // Calculate CRC32C for DirectByteBuffer.
5890 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5891 //
5892 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5893 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5894 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5895 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5896 // no receiver since it is a static method
5897 Node* crc = argument(0); // type: int
5898 Node* src = argument(1); // type: long
5899 Node* offset = argument(3); // type: int
5900 Node* end = argument(4); // type: int
5901
5902 Node* length = _gvn.transform(new SubINode(end, offset));
5903
5904 src = ConvL2X(src); // adjust Java long to machine word
5905 Node* base = _gvn.transform(new CastX2PNode(src));
5906 offset = ConvI2X(offset);
5907
5908 // 'src_start' points to src array + scaled offset
5909 Node* src_start = basic_plus_adr(top(), base, offset);
5910
5911 // static final int[] byteTable in class CRC32C
5912 Node* table = get_table_from_crc32c_class(callee()->holder());
5913 Node* table_start = array_element_address(table, intcon(0), T_INT);
5914
5915 // Call the stub.
5916 address stubAddr = StubRoutines::updateBytesCRC32C();
5917 const char *stubName = "updateBytesCRC32C";
5918
5919 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5920 stubAddr, stubName, TypePtr::BOTTOM,
5921 crc, src_start, length, table_start);
5922 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5923 set_result(result);
5924 return true;
5925 }
5926
5927 //------------------------------inline_updateBytesAdler32----------------------
5928 //
5929 // Calculate Adler32 checksum for byte[] array.
5930 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5931 //
5932 bool LibraryCallKit::inline_updateBytesAdler32() {
5933 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5934 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5935 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5936 // no receiver since it is static method
5937 Node* crc = argument(0); // type: int
5938 Node* src = argument(1); // type: oop
5939 Node* offset = argument(2); // type: int
5940 Node* length = argument(3); // type: int
5941
5942 const Type* src_type = src->Value(&_gvn);
5943 const TypeAryPtr* top_src = src_type->isa_aryptr();
5944 if (top_src == NULL || top_src->klass() == NULL) {
5945 // failed array check
5946 return false;
5947 }
5948
5949 // Figure out the size and type of the elements we will be copying.
5950 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5951 if (src_elem != T_BYTE) {
5952 return false;
5953 }
5954
5955 // 'src_start' points to src array + scaled offset
5956 Node* src_start = array_element_address(src, offset, src_elem);
5957
5958 // We assume that range check is done by caller.
5959 // TODO: generate range check (offset+length < src.length) in debug VM.
5960
5961 // Call the stub.
5962 address stubAddr = StubRoutines::updateBytesAdler32();
5963 const char *stubName = "updateBytesAdler32";
5964
5965 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5966 stubAddr, stubName, TypePtr::BOTTOM,
5967 crc, src_start, length);
5968 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5969 set_result(result);
5970 return true;
5971 }
5972
5973 //------------------------------inline_updateByteBufferAdler32---------------
5974 //
5975 // Calculate Adler32 checksum for DirectByteBuffer.
5976 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5977 //
5978 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5979 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5980 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5981 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5982 // no receiver since it is static method
5983 Node* crc = argument(0); // type: int
5984 Node* src = argument(1); // type: long
5985 Node* offset = argument(3); // type: int
5986 Node* length = argument(4); // type: int
5987
5988 src = ConvL2X(src); // adjust Java long to machine word
5989 Node* base = _gvn.transform(new CastX2PNode(src));
5990 offset = ConvI2X(offset);
5991
5992 // 'src_start' points to src array + scaled offset
5993 Node* src_start = basic_plus_adr(top(), base, offset);
5994
5995 // Call the stub.
5996 address stubAddr = StubRoutines::updateBytesAdler32();
5997 const char *stubName = "updateBytesAdler32";
5998
5999 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6000 stubAddr, stubName, TypePtr::BOTTOM,
6001 crc, src_start, length);
6002
6003 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6004 set_result(result);
6005 return true;
6006 }
6007
6008 //----------------------------inline_reference_get----------------------------
6009 // public T java.lang.ref.Reference.get();
6010 bool LibraryCallKit::inline_reference_get() {
6011 const int referent_offset = java_lang_ref_Reference::referent_offset;
6012 guarantee(referent_offset > 0, "should have already been set");
6013
6014 // Get the argument:
6015 Node* reference_obj = null_check_receiver();
6016 if (stopped()) return true;
6017
6018 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
6019
6020 ciInstanceKlass* klass = env()->Object_klass();
6021 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
6022
6023 Node* no_ctrl = NULL;
6024 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
6025
6026 // Use the pre-barrier to record the value in the referent field
6027 pre_barrier(false /* do_load */,
6028 control(),
6029 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
6030 result /* pre_val */,
6031 T_OBJECT);
6032
6033 // Add memory barrier to prevent commoning reads from this field
6034 // across safepoint since GC can change its value.
6035 insert_mem_bar(Op_MemBarCPUOrder);
6036
6037 set_result(result);
6038 return true;
6039 }
6040
6041
6042 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6043 bool is_exact=true, bool is_static=false,
6044 ciInstanceKlass * fromKls=NULL) {
6045 if (fromKls == NULL) {
6046 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6047 assert(tinst != NULL, "obj is null");
6048 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6049 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6050 fromKls = tinst->klass()->as_instance_klass();
6051 } else {
6052 assert(is_static, "only for static field access");
6053 }
6054 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6055 ciSymbol::make(fieldTypeString),
6056 is_static);
6057
6058 assert (field != NULL, "undefined field");
6059 if (field == NULL) return (Node *) NULL;
6060
6061 if (is_static) {
6062 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6063 fromObj = makecon(tip);
6064 }
6065
6066 // Next code copied from Parse::do_get_xxx():
6067
6068 // Compute address and memory type.
6069 int offset = field->offset_in_bytes();
6070 bool is_vol = field->is_volatile();
6071 ciType* field_klass = field->type();
6072 assert(field_klass->is_loaded(), "should be loaded");
6073 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6074 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6075 BasicType bt = field->layout_type();
6076
6077 // Build the resultant type of the load
6078 const Type *type;
6079 if (bt == T_OBJECT) {
6080 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6081 } else {
6082 type = Type::get_const_basic_type(bt);
6083 }
6084
6085 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6086 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6087 }
6088 // Build the load.
6089 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6090 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6091 // If reference is volatile, prevent following memory ops from
6092 // floating up past the volatile read. Also prevents commoning
6093 // another volatile read.
6094 if (is_vol) {
6095 // Memory barrier includes bogus read of value to force load BEFORE membar
6096 insert_mem_bar(Op_MemBarAcquire, loadedField);
6097 }
6098 return loadedField;
6099 }
6100
6101 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6102 bool is_exact = true, bool is_static = false,
6103 ciInstanceKlass * fromKls = NULL) {
6104 if (fromKls == NULL) {
6105 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6106 assert(tinst != NULL, "obj is null");
6107 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6108 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6109 fromKls = tinst->klass()->as_instance_klass();
6110 }
6111 else {
6112 assert(is_static, "only for static field access");
6113 }
6114 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6115 ciSymbol::make(fieldTypeString),
6116 is_static);
6117
6118 assert(field != NULL, "undefined field");
6119 assert(!field->is_volatile(), "not defined for volatile fields");
6120
6121 if (is_static) {
6122 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6123 fromObj = makecon(tip);
6124 }
6125
6126 // Next code copied from Parse::do_get_xxx():
6127
6128 // Compute address and memory type.
6129 int offset = field->offset_in_bytes();
6130 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6131
6132 return adr;
6133 }
6134
6135 //------------------------------inline_aescrypt_Block-----------------------
6136 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6137 address stubAddr = NULL;
6138 const char *stubName;
6139 assert(UseAES, "need AES instruction support");
6140
6141 switch(id) {
6142 case vmIntrinsics::_aescrypt_encryptBlock:
6143 stubAddr = StubRoutines::aescrypt_encryptBlock();
6144 stubName = "aescrypt_encryptBlock";
6145 break;
6146 case vmIntrinsics::_aescrypt_decryptBlock:
6147 stubAddr = StubRoutines::aescrypt_decryptBlock();
6148 stubName = "aescrypt_decryptBlock";
6149 break;
6150 }
6151 if (stubAddr == NULL) return false;
6152
6153 Node* aescrypt_object = argument(0);
6154 Node* src = argument(1);
6155 Node* src_offset = argument(2);
6156 Node* dest = argument(3);
6157 Node* dest_offset = argument(4);
6158
6159 // (1) src and dest are arrays.
6160 const Type* src_type = src->Value(&_gvn);
6161 const Type* dest_type = dest->Value(&_gvn);
6162 const TypeAryPtr* top_src = src_type->isa_aryptr();
6163 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6164 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6165
6166 // for the quick and dirty code we will skip all the checks.
6167 // we are just trying to get the call to be generated.
6168 Node* src_start = src;
6169 Node* dest_start = dest;
6170 if (src_offset != NULL || dest_offset != NULL) {
6171 assert(src_offset != NULL && dest_offset != NULL, "");
6172 src_start = array_element_address(src, src_offset, T_BYTE);
6173 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6174 }
6175
6176 // now need to get the start of its expanded key array
6177 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6178 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6179 if (k_start == NULL) return false;
6180
6181 if (Matcher::pass_original_key_for_aes()) {
6182 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6183 // compatibility issues between Java key expansion and SPARC crypto instructions
6184 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6185 if (original_k_start == NULL) return false;
6186
6187 // Call the stub.
6188 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6189 stubAddr, stubName, TypePtr::BOTTOM,
6190 src_start, dest_start, k_start, original_k_start);
6191 } else {
6192 // Call the stub.
6193 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6194 stubAddr, stubName, TypePtr::BOTTOM,
6195 src_start, dest_start, k_start);
6196 }
6197
6198 return true;
6199 }
6200
6201 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6202 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6203 address stubAddr = NULL;
6204 const char *stubName = NULL;
6205
6206 assert(UseAES, "need AES instruction support");
6207
6208 switch(id) {
6209 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6210 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6211 stubName = "cipherBlockChaining_encryptAESCrypt";
6212 break;
6213 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6214 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6215 stubName = "cipherBlockChaining_decryptAESCrypt";
6216 break;
6217 }
6218 if (stubAddr == NULL) return false;
6219
6220 Node* cipherBlockChaining_object = argument(0);
6221 Node* src = argument(1);
6222 Node* src_offset = argument(2);
6223 Node* len = argument(3);
6224 Node* dest = argument(4);
6225 Node* dest_offset = argument(5);
6226
6227 // (1) src and dest are arrays.
6228 const Type* src_type = src->Value(&_gvn);
6229 const Type* dest_type = dest->Value(&_gvn);
6230 const TypeAryPtr* top_src = src_type->isa_aryptr();
6231 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6232 assert (top_src != NULL && top_src->klass() != NULL
6233 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6234
6235 // checks are the responsibility of the caller
6236 Node* src_start = src;
6237 Node* dest_start = dest;
6238 if (src_offset != NULL || dest_offset != NULL) {
6239 assert(src_offset != NULL && dest_offset != NULL, "");
6240 src_start = array_element_address(src, src_offset, T_BYTE);
6241 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6242 }
6243
6244 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6245 // (because of the predicated logic executed earlier).
6246 // so we cast it here safely.
6247 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6248
6249 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6250 if (embeddedCipherObj == NULL) return false;
6251
6252 // cast it to what we know it will be at runtime
6253 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6254 assert(tinst != NULL, "CBC obj is null");
6255 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6256 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6257 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6258
6259 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6260 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6261 const TypeOopPtr* xtype = aklass->as_instance_type();
6262 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6263 aescrypt_object = _gvn.transform(aescrypt_object);
6264
6265 // we need to get the start of the aescrypt_object's expanded key array
6266 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6267 if (k_start == NULL) return false;
6268
6269 // similarly, get the start address of the r vector
6270 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6271 if (objRvec == NULL) return false;
6272 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6273
6274 Node* cbcCrypt;
6275 if (Matcher::pass_original_key_for_aes()) {
6276 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6277 // compatibility issues between Java key expansion and SPARC crypto instructions
6278 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6279 if (original_k_start == NULL) return false;
6280
6281 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6282 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6283 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6284 stubAddr, stubName, TypePtr::BOTTOM,
6285 src_start, dest_start, k_start, r_start, len, original_k_start);
6286 } else {
6287 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6288 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6289 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6290 stubAddr, stubName, TypePtr::BOTTOM,
6291 src_start, dest_start, k_start, r_start, len);
6292 }
6293
6294 // return cipher length (int)
6295 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6296 set_result(retvalue);
6297 return true;
6298 }
6299
6300 //------------------------------inline_counterMode_AESCrypt-----------------------
6301 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6302 assert(UseAES, "need AES instruction support");
6303 if (!UseAESCTRIntrinsics) return false;
6304
6305 address stubAddr = NULL;
6306 const char *stubName = NULL;
6307 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6308 stubAddr = StubRoutines::counterMode_AESCrypt();
6309 stubName = "counterMode_AESCrypt";
6310 }
6311 if (stubAddr == NULL) return false;
6312
6313 Node* counterMode_object = argument(0);
6314 Node* src = argument(1);
6315 Node* src_offset = argument(2);
6316 Node* len = argument(3);
6317 Node* dest = argument(4);
6318 Node* dest_offset = argument(5);
6319
6320 // (1) src and dest are arrays.
6321 const Type* src_type = src->Value(&_gvn);
6322 const Type* dest_type = dest->Value(&_gvn);
6323 const TypeAryPtr* top_src = src_type->isa_aryptr();
6324 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6325 assert(top_src != NULL && top_src->klass() != NULL &&
6326 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6327
6328 // checks are the responsibility of the caller
6329 Node* src_start = src;
6330 Node* dest_start = dest;
6331 if (src_offset != NULL || dest_offset != NULL) {
6332 assert(src_offset != NULL && dest_offset != NULL, "");
6333 src_start = array_element_address(src, src_offset, T_BYTE);
6334 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6335 }
6336
6337 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6338 // (because of the predicated logic executed earlier).
6339 // so we cast it here safely.
6340 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6341 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6342 if (embeddedCipherObj == NULL) return false;
6343 // cast it to what we know it will be at runtime
6344 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6345 assert(tinst != NULL, "CTR obj is null");
6346 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6347 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6348 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6349 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6350 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6351 const TypeOopPtr* xtype = aklass->as_instance_type();
6352 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6353 aescrypt_object = _gvn.transform(aescrypt_object);
6354 // we need to get the start of the aescrypt_object's expanded key array
6355 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6356 if (k_start == NULL) return false;
6357 // similarly, get the start address of the r vector
6358 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6359 if (obj_counter == NULL) return false;
6360 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6361
6362 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6363 if (saved_encCounter == NULL) return false;
6364 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6365 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6366
6367 Node* ctrCrypt;
6368 if (Matcher::pass_original_key_for_aes()) {
6369 // no SPARC version for AES/CTR intrinsics now.
6370 return false;
6371 }
6372 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6373 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6374 OptoRuntime::counterMode_aescrypt_Type(),
6375 stubAddr, stubName, TypePtr::BOTTOM,
6376 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6377
6378 // return cipher length (int)
6379 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6380 set_result(retvalue);
6381 return true;
6382 }
6383
6384 //------------------------------get_key_start_from_aescrypt_object-----------------------
6385 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6386 #if defined(PPC64) || defined(S390)
6387 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6388 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6389 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6390 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6391 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6392 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6393 if (objSessionK == NULL) {
6394 return (Node *) NULL;
6395 }
6396 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6397 #else
6398 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6399 #endif // PPC64
6400 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6401 if (objAESCryptKey == NULL) return (Node *) NULL;
6402
6403 // now have the array, need to get the start address of the K array
6404 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6405 return k_start;
6406 }
6407
6408 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6409 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6410 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6411 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6412 if (objAESCryptKey == NULL) return (Node *) NULL;
6413
6414 // now have the array, need to get the start address of the lastKey array
6415 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6416 return original_k_start;
6417 }
6418
6419 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6420 // Return node representing slow path of predicate check.
6421 // the pseudo code we want to emulate with this predicate is:
6422 // for encryption:
6423 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6424 // for decryption:
6425 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6426 // note cipher==plain is more conservative than the original java code but that's OK
6427 //
6428 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6429 // The receiver was checked for NULL already.
6430 Node* objCBC = argument(0);
6431
6432 // Load embeddedCipher field of CipherBlockChaining object.
6433 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6434
6435 // get AESCrypt klass for instanceOf check
6436 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6437 // will have same classloader as CipherBlockChaining object
6438 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6439 assert(tinst != NULL, "CBCobj is null");
6440 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6441
6442 // we want to do an instanceof comparison against the AESCrypt class
6443 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6444 if (!klass_AESCrypt->is_loaded()) {
6445 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6446 Node* ctrl = control();
6447 set_control(top()); // no regular fast path
6448 return ctrl;
6449 }
6450 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6451
6452 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6453 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6454 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6455
6456 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6457
6458 // for encryption, we are done
6459 if (!decrypting)
6460 return instof_false; // even if it is NULL
6461
6462 // for decryption, we need to add a further check to avoid
6463 // taking the intrinsic path when cipher and plain are the same
6464 // see the original java code for why.
6465 RegionNode* region = new RegionNode(3);
6466 region->init_req(1, instof_false);
6467 Node* src = argument(1);
6468 Node* dest = argument(4);
6469 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6470 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6471 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6472 region->init_req(2, src_dest_conjoint);
6473
6474 record_for_igvn(region);
6475 return _gvn.transform(region);
6476 }
6477
6478 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6479 // Return node representing slow path of predicate check.
6480 // the pseudo code we want to emulate with this predicate is:
6481 // for encryption:
6482 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6483 // for decryption:
6484 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6485 // note cipher==plain is more conservative than the original java code but that's OK
6486 //
6487
6488 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6489 // The receiver was checked for NULL already.
6490 Node* objCTR = argument(0);
6491
6492 // Load embeddedCipher field of CipherBlockChaining object.
6493 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6494
6495 // get AESCrypt klass for instanceOf check
6496 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6497 // will have same classloader as CipherBlockChaining object
6498 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6499 assert(tinst != NULL, "CTRobj is null");
6500 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6501
6502 // we want to do an instanceof comparison against the AESCrypt class
6503 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6504 if (!klass_AESCrypt->is_loaded()) {
6505 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6506 Node* ctrl = control();
6507 set_control(top()); // no regular fast path
6508 return ctrl;
6509 }
6510
6511 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6512 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6513 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6514 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6515 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6516
6517 return instof_false; // even if it is NULL
6518 }
6519
6520 //------------------------------inline_ghash_processBlocks
6521 bool LibraryCallKit::inline_ghash_processBlocks() {
6522 address stubAddr;
6523 const char *stubName;
6524 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6525
6526 stubAddr = StubRoutines::ghash_processBlocks();
6527 stubName = "ghash_processBlocks";
6528
6529 Node* data = argument(0);
6530 Node* offset = argument(1);
6531 Node* len = argument(2);
6532 Node* state = argument(3);
6533 Node* subkeyH = argument(4);
6534
6535 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6536 assert(state_start, "state is NULL");
6537 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6538 assert(subkeyH_start, "subkeyH is NULL");
6539 Node* data_start = array_element_address(data, offset, T_BYTE);
6540 assert(data_start, "data is NULL");
6541
6542 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6543 OptoRuntime::ghash_processBlocks_Type(),
6544 stubAddr, stubName, TypePtr::BOTTOM,
6545 state_start, subkeyH_start, data_start, len);
6546 return true;
6547 }
6548
6549 //------------------------------inline_sha_implCompress-----------------------
6550 //
6551 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6552 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6553 //
6554 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6555 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6556 //
6557 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6558 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6559 //
6560 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6561 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6562
6563 Node* sha_obj = argument(0);
6564 Node* src = argument(1); // type oop
6565 Node* ofs = argument(2); // type int
6566
6567 const Type* src_type = src->Value(&_gvn);
6568 const TypeAryPtr* top_src = src_type->isa_aryptr();
6569 if (top_src == NULL || top_src->klass() == NULL) {
6570 // failed array check
6571 return false;
6572 }
6573 // Figure out the size and type of the elements we will be copying.
6574 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6575 if (src_elem != T_BYTE) {
6576 return false;
6577 }
6578 // 'src_start' points to src array + offset
6579 Node* src_start = array_element_address(src, ofs, src_elem);
6580 Node* state = NULL;
6581 address stubAddr;
6582 const char *stubName;
6583
6584 switch(id) {
6585 case vmIntrinsics::_sha_implCompress:
6586 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6587 state = get_state_from_sha_object(sha_obj);
6588 stubAddr = StubRoutines::sha1_implCompress();
6589 stubName = "sha1_implCompress";
6590 break;
6591 case vmIntrinsics::_sha2_implCompress:
6592 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6593 state = get_state_from_sha_object(sha_obj);
6594 stubAddr = StubRoutines::sha256_implCompress();
6595 stubName = "sha256_implCompress";
6596 break;
6597 case vmIntrinsics::_sha5_implCompress:
6598 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6599 state = get_state_from_sha5_object(sha_obj);
6600 stubAddr = StubRoutines::sha512_implCompress();
6601 stubName = "sha512_implCompress";
6602 break;
6603 default:
6604 fatal_unexpected_iid(id);
6605 return false;
6606 }
6607 if (state == NULL) return false;
6608
6609 // Call the stub.
6610 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6611 stubAddr, stubName, TypePtr::BOTTOM,
6612 src_start, state);
6613
6614 return true;
6615 }
6616
6617 //------------------------------inline_digestBase_implCompressMB-----------------------
6618 //
6619 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6620 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6621 //
6622 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6623 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6624 "need SHA1/SHA256/SHA512 instruction support");
6625 assert((uint)predicate < 3, "sanity");
6626 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6627
6628 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6629 Node* src = argument(1); // byte[] array
6630 Node* ofs = argument(2); // type int
6631 Node* limit = argument(3); // type int
6632
6633 const Type* src_type = src->Value(&_gvn);
6634 const TypeAryPtr* top_src = src_type->isa_aryptr();
6635 if (top_src == NULL || top_src->klass() == NULL) {
6636 // failed array check
6637 return false;
6638 }
6639 // Figure out the size and type of the elements we will be copying.
6640 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6641 if (src_elem != T_BYTE) {
6642 return false;
6643 }
6644 // 'src_start' points to src array + offset
6645 Node* src_start = array_element_address(src, ofs, src_elem);
6646
6647 const char* klass_SHA_name = NULL;
6648 const char* stub_name = NULL;
6649 address stub_addr = NULL;
6650 bool long_state = false;
6651
6652 switch (predicate) {
6653 case 0:
6654 if (UseSHA1Intrinsics) {
6655 klass_SHA_name = "sun/security/provider/SHA";
6656 stub_name = "sha1_implCompressMB";
6657 stub_addr = StubRoutines::sha1_implCompressMB();
6658 }
6659 break;
6660 case 1:
6661 if (UseSHA256Intrinsics) {
6662 klass_SHA_name = "sun/security/provider/SHA2";
6663 stub_name = "sha256_implCompressMB";
6664 stub_addr = StubRoutines::sha256_implCompressMB();
6665 }
6666 break;
6667 case 2:
6668 if (UseSHA512Intrinsics) {
6669 klass_SHA_name = "sun/security/provider/SHA5";
6670 stub_name = "sha512_implCompressMB";
6671 stub_addr = StubRoutines::sha512_implCompressMB();
6672 long_state = true;
6673 }
6674 break;
6675 default:
6676 fatal("unknown SHA intrinsic predicate: %d", predicate);
6677 }
6678 if (klass_SHA_name != NULL) {
6679 // get DigestBase klass to lookup for SHA klass
6680 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6681 assert(tinst != NULL, "digestBase_obj is not instance???");
6682 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6683
6684 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6685 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6686 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6687 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6688 }
6689 return false;
6690 }
6691 //------------------------------inline_sha_implCompressMB-----------------------
6692 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6693 bool long_state, address stubAddr, const char *stubName,
6694 Node* src_start, Node* ofs, Node* limit) {
6695 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6696 const TypeOopPtr* xtype = aklass->as_instance_type();
6697 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6698 sha_obj = _gvn.transform(sha_obj);
6699
6700 Node* state;
6701 if (long_state) {
6702 state = get_state_from_sha5_object(sha_obj);
6703 } else {
6704 state = get_state_from_sha_object(sha_obj);
6705 }
6706 if (state == NULL) return false;
6707
6708 // Call the stub.
6709 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6710 OptoRuntime::digestBase_implCompressMB_Type(),
6711 stubAddr, stubName, TypePtr::BOTTOM,
6712 src_start, state, ofs, limit);
6713 // return ofs (int)
6714 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6715 set_result(result);
6716
6717 return true;
6718 }
6719
6720 //------------------------------get_state_from_sha_object-----------------------
6721 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6722 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6723 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6724 if (sha_state == NULL) return (Node *) NULL;
6725
6726 // now have the array, need to get the start address of the state array
6727 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6728 return state;
6729 }
6730
6731 //------------------------------get_state_from_sha5_object-----------------------
6732 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6733 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6734 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6735 if (sha_state == NULL) return (Node *) NULL;
6736
6737 // now have the array, need to get the start address of the state array
6738 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6739 return state;
6740 }
6741
6742 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6743 // Return node representing slow path of predicate check.
6744 // the pseudo code we want to emulate with this predicate is:
6745 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6746 //
6747 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6748 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6749 "need SHA1/SHA256/SHA512 instruction support");
6750 assert((uint)predicate < 3, "sanity");
6751
6752 // The receiver was checked for NULL already.
6753 Node* digestBaseObj = argument(0);
6754
6755 // get DigestBase klass for instanceOf check
6756 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6757 assert(tinst != NULL, "digestBaseObj is null");
6758 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6759
6760 const char* klass_SHA_name = NULL;
6761 switch (predicate) {
6762 case 0:
6763 if (UseSHA1Intrinsics) {
6764 // we want to do an instanceof comparison against the SHA class
6765 klass_SHA_name = "sun/security/provider/SHA";
6766 }
6767 break;
6768 case 1:
6769 if (UseSHA256Intrinsics) {
6770 // we want to do an instanceof comparison against the SHA2 class
6771 klass_SHA_name = "sun/security/provider/SHA2";
6772 }
6773 break;
6774 case 2:
6775 if (UseSHA512Intrinsics) {
6776 // we want to do an instanceof comparison against the SHA5 class
6777 klass_SHA_name = "sun/security/provider/SHA5";
6778 }
6779 break;
6780 default:
6781 fatal("unknown SHA intrinsic predicate: %d", predicate);
6782 }
6783
6784 ciKlass* klass_SHA = NULL;
6785 if (klass_SHA_name != NULL) {
6786 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6787 }
6788 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6789 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6790 Node* ctrl = control();
6791 set_control(top()); // no intrinsic path
6792 return ctrl;
6793 }
6794 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6795
6796 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6797 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6798 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6799 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6800
6801 return instof_false; // even if it is NULL
6802 }
6803
6804 //-------------inline_fma-----------------------------------
6805 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6806 Node *a = NULL;
6807 Node *b = NULL;
6808 Node *c = NULL;
6809 Node* result = NULL;
6810 switch (id) {
6811 case vmIntrinsics::_fmaD:
6812 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6813 // no receiver since it is static method
6814 a = round_double_node(argument(0));
6815 b = round_double_node(argument(2));
6816 c = round_double_node(argument(4));
6817 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6818 break;
6819 case vmIntrinsics::_fmaF:
6820 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6821 a = argument(0);
6822 b = argument(1);
6823 c = argument(2);
6824 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6825 break;
6826 default:
6827 fatal_unexpected_iid(id); break;
6828 }
6829 set_result(result);
6830 return true;
6831 }
6832
6833 bool LibraryCallKit::inline_profileBoolean() {
6834 Node* counts = argument(1);
6835 const TypeAryPtr* ary = NULL;
6836 ciArray* aobj = NULL;
6837 if (counts->is_Con()
6838 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6839 && (aobj = ary->const_oop()->as_array()) != NULL
6840 && (aobj->length() == 2)) {
6841 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6842 jint false_cnt = aobj->element_value(0).as_int();
6843 jint true_cnt = aobj->element_value(1).as_int();
6844
6845 if (C->log() != NULL) {
6846 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6847 false_cnt, true_cnt);
6848 }
6849
6850 if (false_cnt + true_cnt == 0) {
6851 // According to profile, never executed.
6852 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6853 Deoptimization::Action_reinterpret);
6854 return true;
6855 }
6856
6857 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6858 // is a number of each value occurrences.
6859 Node* result = argument(0);
6860 if (false_cnt == 0 || true_cnt == 0) {
6861 // According to profile, one value has been never seen.
6862 int expected_val = (false_cnt == 0) ? 1 : 0;
6863
6864 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6865 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6866
6867 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6868 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6869 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6870
6871 { // Slow path: uncommon trap for never seen value and then reexecute
6872 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6873 // the value has been seen at least once.
6874 PreserveJVMState pjvms(this);
6875 PreserveReexecuteState preexecs(this);
6876 jvms()->set_should_reexecute(true);
6877
6878 set_control(slow_path);
6879 set_i_o(i_o());
6880
6881 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6882 Deoptimization::Action_reinterpret);
6883 }
6884 // The guard for never seen value enables sharpening of the result and
6885 // returning a constant. It allows to eliminate branches on the same value
6886 // later on.
6887 set_control(fast_path);
6888 result = intcon(expected_val);
6889 }
6890 // Stop profiling.
6891 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6892 // By replacing method body with profile data (represented as ProfileBooleanNode
6893 // on IR level) we effectively disable profiling.
6894 // It enables full speed execution once optimized code is generated.
6895 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6896 C->record_for_igvn(profile);
6897 set_result(profile);
6898 return true;
6899 } else {
6900 // Continue profiling.
6901 // Profile data isn't available at the moment. So, execute method's bytecode version.
6902 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6903 // is compiled and counters aren't available since corresponding MethodHandle
6904 // isn't a compile-time constant.
6905 return false;
6906 }
6907 }
6908
6909 bool LibraryCallKit::inline_isCompileConstant() {
6910 Node* n = argument(0);
6911 set_result(n->is_Con() ? intcon(1) : intcon(0));
6912 return true;
6913 }