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