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 {
2475 // Normalize the value returned by getBoolean in the following cases
2476 if (mismatched ||
2477 heap_base_oop == top() || // - heap_base_oop is NULL or
2478 (can_access_non_heap && alias_type->field() == NULL) // - heap_base_oop is potentially NULL
2479 // and the unsafe access is made to large offset
2480 // (i.e., larger than the maximum offset necessary for any
2481 // field access)
2482 ) {
2483 IdealKit ideal = IdealKit(this);
2484 #define __ ideal.
2485 IdealVariable normalized_result(ideal);
2486 __ declarations_done();
2487 __ set(normalized_result, p);
2488 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2489 __ set(normalized_result, ideal.ConI(1));
2490 ideal.end_if();
2491 final_sync(ideal);
2492 p = __ value(normalized_result);
2493 #undef __
2494 }
2495 }
2496 case T_CHAR:
2497 case T_BYTE:
2498 case T_SHORT:
2499 case T_INT:
2500 case T_LONG:
2501 case T_FLOAT:
2502 case T_DOUBLE:
2503 break;
2504 case T_OBJECT:
2505 if (need_read_barrier) {
2506 // We do not require a mem bar inside pre_barrier if need_mem_bar
2507 // is set: the barriers would be emitted by us.
2508 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2509 }
2510 break;
2511 case T_ADDRESS:
2512 // Cast to an int type.
2513 p = _gvn.transform(new CastP2XNode(NULL, p));
2514 p = ConvX2UL(p);
2515 break;
2516 default:
2517 fatal("unexpected type %d: %s", type, type2name(type));
2518 break;
2519 }
2520 }
2521 // The load node has the control of the preceding MemBarCPUOrder. All
2522 // following nodes will have the control of the MemBarCPUOrder inserted at
2523 // the end of this method. So, pushing the load onto the stack at a later
2524 // point is fine.
2525 set_result(p);
2526 } else {
2527 // place effect of store into memory
2528 switch (type) {
2529 case T_DOUBLE:
2530 val = dstore_rounding(val);
2531 break;
2532 case T_ADDRESS:
2533 // Repackage the long as a pointer.
2534 val = ConvL2X(val);
2535 val = _gvn.transform(new CastX2PNode(val));
2536 break;
2537 }
2538
2539 if (type == T_OBJECT) {
2540 store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2541 } else {
2542 store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2543 }
2544 }
2545
2546 switch(kind) {
2547 case Relaxed:
2548 case Opaque:
2549 case Release:
2550 break;
2551 case Acquire:
2552 case Volatile:
2553 if (!is_store) {
2554 insert_mem_bar(Op_MemBarAcquire);
2555 } else {
2556 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2557 insert_mem_bar(Op_MemBarVolatile);
2558 }
2559 }
2560 break;
2561 default:
2562 ShouldNotReachHere();
2563 }
2564
2565 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2566
2567 return true;
2568 }
2569
2570 //----------------------------inline_unsafe_load_store----------------------------
2571 // This method serves a couple of different customers (depending on LoadStoreKind):
2572 //
2573 // LS_cmp_swap:
2574 //
2575 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2576 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2577 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2578 //
2579 // LS_cmp_swap_weak:
2580 //
2581 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2582 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2583 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2584 //
2585 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2586 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2587 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2588 //
2589 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2590 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2591 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2592 //
2593 // LS_cmp_exchange:
2594 //
2595 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2596 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2597 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2598 //
2599 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2600 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2601 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2602 //
2603 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2604 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2605 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2606 //
2607 // LS_get_add:
2608 //
2609 // int getAndAddInt( Object o, long offset, int delta)
2610 // long getAndAddLong(Object o, long offset, long delta)
2611 //
2612 // LS_get_set:
2613 //
2614 // int getAndSet(Object o, long offset, int newValue)
2615 // long getAndSet(Object o, long offset, long newValue)
2616 // Object getAndSet(Object o, long offset, Object newValue)
2617 //
2618 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2619 // This basic scheme here is the same as inline_unsafe_access, but
2620 // differs in enough details that combining them would make the code
2621 // overly confusing. (This is a true fact! I originally combined
2622 // them, but even I was confused by it!) As much code/comments as
2623 // possible are retained from inline_unsafe_access though to make
2624 // the correspondences clearer. - dl
2625
2626 if (callee()->is_static()) return false; // caller must have the capability!
2627
2628 #ifndef PRODUCT
2629 BasicType rtype;
2630 {
2631 ResourceMark rm;
2632 // Check the signatures.
2633 ciSignature* sig = callee()->signature();
2634 rtype = sig->return_type()->basic_type();
2635 switch(kind) {
2636 case LS_get_add:
2637 case LS_get_set: {
2638 // Check the signatures.
2639 #ifdef ASSERT
2640 assert(rtype == type, "get and set must return the expected type");
2641 assert(sig->count() == 3, "get and set has 3 arguments");
2642 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2643 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2644 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2645 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2646 #endif // ASSERT
2647 break;
2648 }
2649 case LS_cmp_swap:
2650 case LS_cmp_swap_weak: {
2651 // Check the signatures.
2652 #ifdef ASSERT
2653 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2654 assert(sig->count() == 4, "CAS has 4 arguments");
2655 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2656 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2657 #endif // ASSERT
2658 break;
2659 }
2660 case LS_cmp_exchange: {
2661 // Check the signatures.
2662 #ifdef ASSERT
2663 assert(rtype == type, "CAS must return the expected type");
2664 assert(sig->count() == 4, "CAS has 4 arguments");
2665 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2666 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2667 #endif // ASSERT
2668 break;
2669 }
2670 default:
2671 ShouldNotReachHere();
2672 }
2673 }
2674 #endif //PRODUCT
2675
2676 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2677
2678 // Get arguments:
2679 Node* receiver = NULL;
2680 Node* base = NULL;
2681 Node* offset = NULL;
2682 Node* oldval = NULL;
2683 Node* newval = NULL;
2684 switch(kind) {
2685 case LS_cmp_swap:
2686 case LS_cmp_swap_weak:
2687 case LS_cmp_exchange: {
2688 const bool two_slot_type = type2size[type] == 2;
2689 receiver = argument(0); // type: oop
2690 base = argument(1); // type: oop
2691 offset = argument(2); // type: long
2692 oldval = argument(4); // type: oop, int, or long
2693 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2694 break;
2695 }
2696 case LS_get_add:
2697 case LS_get_set: {
2698 receiver = argument(0); // type: oop
2699 base = argument(1); // type: oop
2700 offset = argument(2); // type: long
2701 oldval = NULL;
2702 newval = argument(4); // type: oop, int, or long
2703 break;
2704 }
2705 default:
2706 ShouldNotReachHere();
2707 }
2708
2709 // Build field offset expression.
2710 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2711 // to be plain byte offsets, which are also the same as those accepted
2712 // by oopDesc::field_base.
2713 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2714 // 32-bit machines ignore the high half of long offsets
2715 offset = ConvL2X(offset);
2716 Node* adr = make_unsafe_address(base, offset);
2717 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2718
2719 Compile::AliasType* alias_type = C->alias_type(adr_type);
2720 BasicType bt = alias_type->basic_type();
2721 if (bt != T_ILLEGAL &&
2722 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2723 // Don't intrinsify mismatched object accesses.
2724 return false;
2725 }
2726
2727 // For CAS, unlike inline_unsafe_access, there seems no point in
2728 // trying to refine types. Just use the coarse types here.
2729 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2730 const Type *value_type = Type::get_const_basic_type(type);
2731
2732 switch (kind) {
2733 case LS_get_set:
2734 case LS_cmp_exchange: {
2735 if (type == T_OBJECT) {
2736 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2737 if (tjp != NULL) {
2738 value_type = tjp;
2739 }
2740 }
2741 break;
2742 }
2743 case LS_cmp_swap:
2744 case LS_cmp_swap_weak:
2745 case LS_get_add:
2746 break;
2747 default:
2748 ShouldNotReachHere();
2749 }
2750
2751 // Null check receiver.
2752 receiver = null_check(receiver);
2753 if (stopped()) {
2754 return true;
2755 }
2756
2757 int alias_idx = C->get_alias_index(adr_type);
2758
2759 // Memory-model-wise, a LoadStore acts like a little synchronized
2760 // block, so needs barriers on each side. These don't translate
2761 // into actual barriers on most machines, but we still need rest of
2762 // compiler to respect ordering.
2763
2764 switch (access_kind) {
2765 case Relaxed:
2766 case Acquire:
2767 break;
2768 case Release:
2769 insert_mem_bar(Op_MemBarRelease);
2770 break;
2771 case Volatile:
2772 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2773 insert_mem_bar(Op_MemBarVolatile);
2774 } else {
2775 insert_mem_bar(Op_MemBarRelease);
2776 }
2777 break;
2778 default:
2779 ShouldNotReachHere();
2780 }
2781 insert_mem_bar(Op_MemBarCPUOrder);
2782
2783 // Figure out the memory ordering.
2784 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2785
2786 // 4984716: MemBars must be inserted before this
2787 // memory node in order to avoid a false
2788 // dependency which will confuse the scheduler.
2789 Node *mem = memory(alias_idx);
2790
2791 // For now, we handle only those cases that actually exist: ints,
2792 // longs, and Object. Adding others should be straightforward.
2793 Node* load_store = NULL;
2794 switch(type) {
2795 case T_BYTE:
2796 switch(kind) {
2797 case LS_get_add:
2798 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2799 break;
2800 case LS_get_set:
2801 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2802 break;
2803 case LS_cmp_swap_weak:
2804 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2805 break;
2806 case LS_cmp_swap:
2807 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2808 break;
2809 case LS_cmp_exchange:
2810 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2811 break;
2812 default:
2813 ShouldNotReachHere();
2814 }
2815 break;
2816 case T_SHORT:
2817 switch(kind) {
2818 case LS_get_add:
2819 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2820 break;
2821 case LS_get_set:
2822 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2823 break;
2824 case LS_cmp_swap_weak:
2825 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2826 break;
2827 case LS_cmp_swap:
2828 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2829 break;
2830 case LS_cmp_exchange:
2831 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2832 break;
2833 default:
2834 ShouldNotReachHere();
2835 }
2836 break;
2837 case T_INT:
2838 switch(kind) {
2839 case LS_get_add:
2840 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2841 break;
2842 case LS_get_set:
2843 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2844 break;
2845 case LS_cmp_swap_weak:
2846 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2847 break;
2848 case LS_cmp_swap:
2849 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2850 break;
2851 case LS_cmp_exchange:
2852 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2853 break;
2854 default:
2855 ShouldNotReachHere();
2856 }
2857 break;
2858 case T_LONG:
2859 switch(kind) {
2860 case LS_get_add:
2861 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2862 break;
2863 case LS_get_set:
2864 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2865 break;
2866 case LS_cmp_swap_weak:
2867 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2868 break;
2869 case LS_cmp_swap:
2870 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2871 break;
2872 case LS_cmp_exchange:
2873 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2874 break;
2875 default:
2876 ShouldNotReachHere();
2877 }
2878 break;
2879 case T_OBJECT:
2880 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2881 // could be delayed during Parse (for example, in adjust_map_after_if()).
2882 // Execute transformation here to avoid barrier generation in such case.
2883 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2884 newval = _gvn.makecon(TypePtr::NULL_PTR);
2885
2886 // Reference stores need a store barrier.
2887 switch(kind) {
2888 case LS_get_set: {
2889 // If pre-barrier must execute before the oop store, old value will require do_load here.
2890 if (!can_move_pre_barrier()) {
2891 pre_barrier(true /* do_load*/,
2892 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2893 NULL /* pre_val*/,
2894 T_OBJECT);
2895 } // Else move pre_barrier to use load_store value, see below.
2896 break;
2897 }
2898 case LS_cmp_swap_weak:
2899 case LS_cmp_swap:
2900 case LS_cmp_exchange: {
2901 // Same as for newval above:
2902 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2903 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2904 }
2905 // The only known value which might get overwritten is oldval.
2906 pre_barrier(false /* do_load */,
2907 control(), NULL, NULL, max_juint, NULL, NULL,
2908 oldval /* pre_val */,
2909 T_OBJECT);
2910 break;
2911 }
2912 default:
2913 ShouldNotReachHere();
2914 }
2915
2916 #ifdef _LP64
2917 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2918 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2919
2920 switch(kind) {
2921 case LS_get_set:
2922 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2923 break;
2924 case LS_cmp_swap_weak: {
2925 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2926 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2927 break;
2928 }
2929 case LS_cmp_swap: {
2930 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2931 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2932 break;
2933 }
2934 case LS_cmp_exchange: {
2935 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2936 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2937 break;
2938 }
2939 default:
2940 ShouldNotReachHere();
2941 }
2942 } else
2943 #endif
2944 switch (kind) {
2945 case LS_get_set:
2946 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2947 break;
2948 case LS_cmp_swap_weak:
2949 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2950 break;
2951 case LS_cmp_swap:
2952 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2953 break;
2954 case LS_cmp_exchange:
2955 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2956 break;
2957 default:
2958 ShouldNotReachHere();
2959 }
2960
2961 // Emit the post barrier only when the actual store happened. This makes sense
2962 // to check only for LS_cmp_* that can fail to set the value.
2963 // LS_cmp_exchange does not produce any branches by default, so there is no
2964 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
2965 // CompareAndExchange and move branches here, it would make sense to conditionalize
2966 // post_barriers for LS_cmp_exchange as well.
2967 //
2968 // CAS success path is marked more likely since we anticipate this is a performance
2969 // critical path, while CAS failure path can use the penalty for going through unlikely
2970 // path as backoff. Which is still better than doing a store barrier there.
2971 switch (kind) {
2972 case LS_get_set:
2973 case LS_cmp_exchange: {
2974 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2975 break;
2976 }
2977 case LS_cmp_swap_weak:
2978 case LS_cmp_swap: {
2979 IdealKit ideal(this);
2980 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
2981 sync_kit(ideal);
2982 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2983 ideal.sync_kit(this);
2984 } ideal.end_if();
2985 final_sync(ideal);
2986 break;
2987 }
2988 default:
2989 ShouldNotReachHere();
2990 }
2991 break;
2992 default:
2993 fatal("unexpected type %d: %s", type, type2name(type));
2994 break;
2995 }
2996
2997 // SCMemProjNodes represent the memory state of a LoadStore. Their
2998 // main role is to prevent LoadStore nodes from being optimized away
2999 // when their results aren't used.
3000 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3001 set_memory(proj, alias_idx);
3002
3003 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3004 #ifdef _LP64
3005 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3006 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3007 }
3008 #endif
3009 if (can_move_pre_barrier()) {
3010 // Don't need to load pre_val. The old value is returned by load_store.
3011 // The pre_barrier can execute after the xchg as long as no safepoint
3012 // gets inserted between them.
3013 pre_barrier(false /* do_load */,
3014 control(), NULL, NULL, max_juint, NULL, NULL,
3015 load_store /* pre_val */,
3016 T_OBJECT);
3017 }
3018 }
3019
3020 // Add the trailing membar surrounding the access
3021 insert_mem_bar(Op_MemBarCPUOrder);
3022
3023 switch (access_kind) {
3024 case Relaxed:
3025 case Release:
3026 break; // do nothing
3027 case Acquire:
3028 case Volatile:
3029 insert_mem_bar(Op_MemBarAcquire);
3030 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3031 break;
3032 default:
3033 ShouldNotReachHere();
3034 }
3035
3036 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3037 set_result(load_store);
3038 return true;
3039 }
3040
3041 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3042 MemNode::MemOrd mo = MemNode::unset;
3043 switch(kind) {
3044 case Opaque:
3045 case Relaxed: mo = MemNode::unordered; break;
3046 case Acquire: mo = MemNode::acquire; break;
3047 case Release: mo = MemNode::release; break;
3048 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3049 default:
3050 ShouldNotReachHere();
3051 }
3052 guarantee(mo != MemNode::unset, "Should select memory ordering");
3053 return mo;
3054 }
3055
3056 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3057 MemNode::MemOrd mo = MemNode::unset;
3058 switch(kind) {
3059 case Opaque:
3060 case Relaxed: mo = MemNode::unordered; break;
3061 case Acquire: mo = MemNode::acquire; break;
3062 case Release: mo = MemNode::release; break;
3063 case Volatile: mo = MemNode::seqcst; break;
3064 default:
3065 ShouldNotReachHere();
3066 }
3067 guarantee(mo != MemNode::unset, "Should select memory ordering");
3068 return mo;
3069 }
3070
3071 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3072 // Regardless of form, don't allow previous ld/st to move down,
3073 // then issue acquire, release, or volatile mem_bar.
3074 insert_mem_bar(Op_MemBarCPUOrder);
3075 switch(id) {
3076 case vmIntrinsics::_loadFence:
3077 insert_mem_bar(Op_LoadFence);
3078 return true;
3079 case vmIntrinsics::_storeFence:
3080 insert_mem_bar(Op_StoreFence);
3081 return true;
3082 case vmIntrinsics::_fullFence:
3083 insert_mem_bar(Op_MemBarVolatile);
3084 return true;
3085 default:
3086 fatal_unexpected_iid(id);
3087 return false;
3088 }
3089 }
3090
3091 bool LibraryCallKit::inline_onspinwait() {
3092 insert_mem_bar(Op_OnSpinWait);
3093 return true;
3094 }
3095
3096 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3097 if (!kls->is_Con()) {
3098 return true;
3099 }
3100 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3101 if (klsptr == NULL) {
3102 return true;
3103 }
3104 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3105 // don't need a guard for a klass that is already initialized
3106 return !ik->is_initialized();
3107 }
3108
3109 //----------------------------inline_unsafe_allocate---------------------------
3110 // public native Object Unsafe.allocateInstance(Class<?> cls);
3111 bool LibraryCallKit::inline_unsafe_allocate() {
3112 if (callee()->is_static()) return false; // caller must have the capability!
3113
3114 null_check_receiver(); // null-check, then ignore
3115 Node* cls = null_check(argument(1));
3116 if (stopped()) return true;
3117
3118 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3119 kls = null_check(kls);
3120 if (stopped()) return true; // argument was like int.class
3121
3122 Node* test = NULL;
3123 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3124 // Note: The argument might still be an illegal value like
3125 // Serializable.class or Object[].class. The runtime will handle it.
3126 // But we must make an explicit check for initialization.
3127 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3128 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3129 // can generate code to load it as unsigned byte.
3130 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3131 Node* bits = intcon(InstanceKlass::fully_initialized);
3132 test = _gvn.transform(new SubINode(inst, bits));
3133 // The 'test' is non-zero if we need to take a slow path.
3134 }
3135
3136 Node* obj = new_instance(kls, test);
3137 set_result(obj);
3138 return true;
3139 }
3140
3141 //------------------------inline_native_time_funcs--------------
3142 // inline code for System.currentTimeMillis() and System.nanoTime()
3143 // these have the same type and signature
3144 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3145 const TypeFunc* tf = OptoRuntime::void_long_Type();
3146 const TypePtr* no_memory_effects = NULL;
3147 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3148 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3149 #ifdef ASSERT
3150 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3151 assert(value_top == top(), "second value must be top");
3152 #endif
3153 set_result(value);
3154 return true;
3155 }
3156
3157 //------------------------inline_native_currentThread------------------
3158 bool LibraryCallKit::inline_native_currentThread() {
3159 Node* junk = NULL;
3160 set_result(generate_current_thread(junk));
3161 return true;
3162 }
3163
3164 //------------------------inline_native_isInterrupted------------------
3165 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3166 bool LibraryCallKit::inline_native_isInterrupted() {
3167 // Add a fast path to t.isInterrupted(clear_int):
3168 // (t == Thread.current() &&
3169 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3170 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3171 // So, in the common case that the interrupt bit is false,
3172 // we avoid making a call into the VM. Even if the interrupt bit
3173 // is true, if the clear_int argument is false, we avoid the VM call.
3174 // However, if the receiver is not currentThread, we must call the VM,
3175 // because there must be some locking done around the operation.
3176
3177 // We only go to the fast case code if we pass two guards.
3178 // Paths which do not pass are accumulated in the slow_region.
3179
3180 enum {
3181 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3182 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3183 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3184 PATH_LIMIT
3185 };
3186
3187 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3188 // out of the function.
3189 insert_mem_bar(Op_MemBarCPUOrder);
3190
3191 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3192 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3193
3194 RegionNode* slow_region = new RegionNode(1);
3195 record_for_igvn(slow_region);
3196
3197 // (a) Receiving thread must be the current thread.
3198 Node* rec_thr = argument(0);
3199 Node* tls_ptr = NULL;
3200 Node* cur_thr = generate_current_thread(tls_ptr);
3201 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3202 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3203
3204 generate_slow_guard(bol_thr, slow_region);
3205
3206 // (b) Interrupt bit on TLS must be false.
3207 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3208 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3209 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3210
3211 // Set the control input on the field _interrupted read to prevent it floating up.
3212 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3213 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3214 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3215
3216 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3217
3218 // First fast path: if (!TLS._interrupted) return false;
3219 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3220 result_rgn->init_req(no_int_result_path, false_bit);
3221 result_val->init_req(no_int_result_path, intcon(0));
3222
3223 // drop through to next case
3224 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3225
3226 #ifndef _WINDOWS
3227 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3228 Node* clr_arg = argument(1);
3229 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3230 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3231 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3232
3233 // Second fast path: ... else if (!clear_int) return true;
3234 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3235 result_rgn->init_req(no_clear_result_path, false_arg);
3236 result_val->init_req(no_clear_result_path, intcon(1));
3237
3238 // drop through to next case
3239 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3240 #else
3241 // To return true on Windows you must read the _interrupted field
3242 // and check the event state i.e. take the slow path.
3243 #endif // _WINDOWS
3244
3245 // (d) Otherwise, go to the slow path.
3246 slow_region->add_req(control());
3247 set_control( _gvn.transform(slow_region));
3248
3249 if (stopped()) {
3250 // There is no slow path.
3251 result_rgn->init_req(slow_result_path, top());
3252 result_val->init_req(slow_result_path, top());
3253 } else {
3254 // non-virtual because it is a private non-static
3255 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3256
3257 Node* slow_val = set_results_for_java_call(slow_call);
3258 // this->control() comes from set_results_for_java_call
3259
3260 Node* fast_io = slow_call->in(TypeFunc::I_O);
3261 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3262
3263 // These two phis are pre-filled with copies of of the fast IO and Memory
3264 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3265 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3266
3267 result_rgn->init_req(slow_result_path, control());
3268 result_io ->init_req(slow_result_path, i_o());
3269 result_mem->init_req(slow_result_path, reset_memory());
3270 result_val->init_req(slow_result_path, slow_val);
3271
3272 set_all_memory(_gvn.transform(result_mem));
3273 set_i_o( _gvn.transform(result_io));
3274 }
3275
3276 C->set_has_split_ifs(true); // Has chance for split-if optimization
3277 set_result(result_rgn, result_val);
3278 return true;
3279 }
3280
3281 //---------------------------load_mirror_from_klass----------------------------
3282 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3283 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3284 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3285 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3286 }
3287
3288 //-----------------------load_klass_from_mirror_common-------------------------
3289 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3290 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3291 // and branch to the given path on the region.
3292 // If never_see_null, take an uncommon trap on null, so we can optimistically
3293 // compile for the non-null case.
3294 // If the region is NULL, force never_see_null = true.
3295 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3296 bool never_see_null,
3297 RegionNode* region,
3298 int null_path,
3299 int offset) {
3300 if (region == NULL) never_see_null = true;
3301 Node* p = basic_plus_adr(mirror, offset);
3302 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3303 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3304 Node* null_ctl = top();
3305 kls = null_check_oop(kls, &null_ctl, never_see_null);
3306 if (region != NULL) {
3307 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3308 region->init_req(null_path, null_ctl);
3309 } else {
3310 assert(null_ctl == top(), "no loose ends");
3311 }
3312 return kls;
3313 }
3314
3315 //--------------------(inline_native_Class_query helpers)---------------------
3316 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3317 // Fall through if (mods & mask) == bits, take the guard otherwise.
3318 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3319 // Branch around if the given klass has the given modifier bit set.
3320 // Like generate_guard, adds a new path onto the region.
3321 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3322 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3323 Node* mask = intcon(modifier_mask);
3324 Node* bits = intcon(modifier_bits);
3325 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3326 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3327 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3328 return generate_fair_guard(bol, region);
3329 }
3330 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3331 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3332 }
3333
3334 //-------------------------inline_native_Class_query-------------------
3335 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3336 const Type* return_type = TypeInt::BOOL;
3337 Node* prim_return_value = top(); // what happens if it's a primitive class?
3338 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3339 bool expect_prim = false; // most of these guys expect to work on refs
3340
3341 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3342
3343 Node* mirror = argument(0);
3344 Node* obj = top();
3345
3346 switch (id) {
3347 case vmIntrinsics::_isInstance:
3348 // nothing is an instance of a primitive type
3349 prim_return_value = intcon(0);
3350 obj = argument(1);
3351 break;
3352 case vmIntrinsics::_getModifiers:
3353 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3354 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3355 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3356 break;
3357 case vmIntrinsics::_isInterface:
3358 prim_return_value = intcon(0);
3359 break;
3360 case vmIntrinsics::_isArray:
3361 prim_return_value = intcon(0);
3362 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3363 break;
3364 case vmIntrinsics::_isPrimitive:
3365 prim_return_value = intcon(1);
3366 expect_prim = true; // obviously
3367 break;
3368 case vmIntrinsics::_getSuperclass:
3369 prim_return_value = null();
3370 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3371 break;
3372 case vmIntrinsics::_getClassAccessFlags:
3373 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3374 return_type = TypeInt::INT; // not bool! 6297094
3375 break;
3376 default:
3377 fatal_unexpected_iid(id);
3378 break;
3379 }
3380
3381 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3382 if (mirror_con == NULL) return false; // cannot happen?
3383
3384 #ifndef PRODUCT
3385 if (C->print_intrinsics() || C->print_inlining()) {
3386 ciType* k = mirror_con->java_mirror_type();
3387 if (k) {
3388 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3389 k->print_name();
3390 tty->cr();
3391 }
3392 }
3393 #endif
3394
3395 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3396 RegionNode* region = new RegionNode(PATH_LIMIT);
3397 record_for_igvn(region);
3398 PhiNode* phi = new PhiNode(region, return_type);
3399
3400 // The mirror will never be null of Reflection.getClassAccessFlags, however
3401 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3402 // if it is. See bug 4774291.
3403
3404 // For Reflection.getClassAccessFlags(), the null check occurs in
3405 // the wrong place; see inline_unsafe_access(), above, for a similar
3406 // situation.
3407 mirror = null_check(mirror);
3408 // If mirror or obj is dead, only null-path is taken.
3409 if (stopped()) return true;
3410
3411 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3412
3413 // Now load the mirror's klass metaobject, and null-check it.
3414 // Side-effects region with the control path if the klass is null.
3415 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3416 // If kls is null, we have a primitive mirror.
3417 phi->init_req(_prim_path, prim_return_value);
3418 if (stopped()) { set_result(region, phi); return true; }
3419 bool safe_for_replace = (region->in(_prim_path) == top());
3420
3421 Node* p; // handy temp
3422 Node* null_ctl;
3423
3424 // Now that we have the non-null klass, we can perform the real query.
3425 // For constant classes, the query will constant-fold in LoadNode::Value.
3426 Node* query_value = top();
3427 switch (id) {
3428 case vmIntrinsics::_isInstance:
3429 // nothing is an instance of a primitive type
3430 query_value = gen_instanceof(obj, kls, safe_for_replace);
3431 break;
3432
3433 case vmIntrinsics::_getModifiers:
3434 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3435 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3436 break;
3437
3438 case vmIntrinsics::_isInterface:
3439 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3440 if (generate_interface_guard(kls, region) != NULL)
3441 // A guard was added. If the guard is taken, it was an interface.
3442 phi->add_req(intcon(1));
3443 // If we fall through, it's a plain class.
3444 query_value = intcon(0);
3445 break;
3446
3447 case vmIntrinsics::_isArray:
3448 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3449 if (generate_array_guard(kls, region) != NULL)
3450 // A guard was added. If the guard is taken, it was an array.
3451 phi->add_req(intcon(1));
3452 // If we fall through, it's a plain class.
3453 query_value = intcon(0);
3454 break;
3455
3456 case vmIntrinsics::_isPrimitive:
3457 query_value = intcon(0); // "normal" path produces false
3458 break;
3459
3460 case vmIntrinsics::_getSuperclass:
3461 // The rules here are somewhat unfortunate, but we can still do better
3462 // with random logic than with a JNI call.
3463 // Interfaces store null or Object as _super, but must report null.
3464 // Arrays store an intermediate super as _super, but must report Object.
3465 // Other types can report the actual _super.
3466 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3467 if (generate_interface_guard(kls, region) != NULL)
3468 // A guard was added. If the guard is taken, it was an interface.
3469 phi->add_req(null());
3470 if (generate_array_guard(kls, region) != NULL)
3471 // A guard was added. If the guard is taken, it was an array.
3472 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3473 // If we fall through, it's a plain class. Get its _super.
3474 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3475 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3476 null_ctl = top();
3477 kls = null_check_oop(kls, &null_ctl);
3478 if (null_ctl != top()) {
3479 // If the guard is taken, Object.superClass is null (both klass and mirror).
3480 region->add_req(null_ctl);
3481 phi ->add_req(null());
3482 }
3483 if (!stopped()) {
3484 query_value = load_mirror_from_klass(kls);
3485 }
3486 break;
3487
3488 case vmIntrinsics::_getClassAccessFlags:
3489 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3490 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3491 break;
3492
3493 default:
3494 fatal_unexpected_iid(id);
3495 break;
3496 }
3497
3498 // Fall-through is the normal case of a query to a real class.
3499 phi->init_req(1, query_value);
3500 region->init_req(1, control());
3501
3502 C->set_has_split_ifs(true); // Has chance for split-if optimization
3503 set_result(region, phi);
3504 return true;
3505 }
3506
3507 //-------------------------inline_Class_cast-------------------
3508 bool LibraryCallKit::inline_Class_cast() {
3509 Node* mirror = argument(0); // Class
3510 Node* obj = argument(1);
3511 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3512 if (mirror_con == NULL) {
3513 return false; // dead path (mirror->is_top()).
3514 }
3515 if (obj == NULL || obj->is_top()) {
3516 return false; // dead path
3517 }
3518 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3519
3520 // First, see if Class.cast() can be folded statically.
3521 // java_mirror_type() returns non-null for compile-time Class constants.
3522 ciType* tm = mirror_con->java_mirror_type();
3523 if (tm != NULL && tm->is_klass() &&
3524 tp != NULL && tp->klass() != NULL) {
3525 if (!tp->klass()->is_loaded()) {
3526 // Don't use intrinsic when class is not loaded.
3527 return false;
3528 } else {
3529 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3530 if (static_res == Compile::SSC_always_true) {
3531 // isInstance() is true - fold the code.
3532 set_result(obj);
3533 return true;
3534 } else if (static_res == Compile::SSC_always_false) {
3535 // Don't use intrinsic, have to throw ClassCastException.
3536 // If the reference is null, the non-intrinsic bytecode will
3537 // be optimized appropriately.
3538 return false;
3539 }
3540 }
3541 }
3542
3543 // Bailout intrinsic and do normal inlining if exception path is frequent.
3544 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3545 return false;
3546 }
3547
3548 // Generate dynamic checks.
3549 // Class.cast() is java implementation of _checkcast bytecode.
3550 // Do checkcast (Parse::do_checkcast()) optimizations here.
3551
3552 mirror = null_check(mirror);
3553 // If mirror is dead, only null-path is taken.
3554 if (stopped()) {
3555 return true;
3556 }
3557
3558 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3559 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3560 RegionNode* region = new RegionNode(PATH_LIMIT);
3561 record_for_igvn(region);
3562
3563 // Now load the mirror's klass metaobject, and null-check it.
3564 // If kls is null, we have a primitive mirror and
3565 // nothing is an instance of a primitive type.
3566 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3567
3568 Node* res = top();
3569 if (!stopped()) {
3570 Node* bad_type_ctrl = top();
3571 // Do checkcast optimizations.
3572 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3573 region->init_req(_bad_type_path, bad_type_ctrl);
3574 }
3575 if (region->in(_prim_path) != top() ||
3576 region->in(_bad_type_path) != top()) {
3577 // Let Interpreter throw ClassCastException.
3578 PreserveJVMState pjvms(this);
3579 set_control(_gvn.transform(region));
3580 uncommon_trap(Deoptimization::Reason_intrinsic,
3581 Deoptimization::Action_maybe_recompile);
3582 }
3583 if (!stopped()) {
3584 set_result(res);
3585 }
3586 return true;
3587 }
3588
3589
3590 //--------------------------inline_native_subtype_check------------------------
3591 // This intrinsic takes the JNI calls out of the heart of
3592 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3593 bool LibraryCallKit::inline_native_subtype_check() {
3594 // Pull both arguments off the stack.
3595 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3596 args[0] = argument(0);
3597 args[1] = argument(1);
3598 Node* klasses[2]; // corresponding Klasses: superk, subk
3599 klasses[0] = klasses[1] = top();
3600
3601 enum {
3602 // A full decision tree on {superc is prim, subc is prim}:
3603 _prim_0_path = 1, // {P,N} => false
3604 // {P,P} & superc!=subc => false
3605 _prim_same_path, // {P,P} & superc==subc => true
3606 _prim_1_path, // {N,P} => false
3607 _ref_subtype_path, // {N,N} & subtype check wins => true
3608 _both_ref_path, // {N,N} & subtype check loses => false
3609 PATH_LIMIT
3610 };
3611
3612 RegionNode* region = new RegionNode(PATH_LIMIT);
3613 Node* phi = new PhiNode(region, TypeInt::BOOL);
3614 record_for_igvn(region);
3615
3616 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3617 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3618 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3619
3620 // First null-check both mirrors and load each mirror's klass metaobject.
3621 int which_arg;
3622 for (which_arg = 0; which_arg <= 1; which_arg++) {
3623 Node* arg = args[which_arg];
3624 arg = null_check(arg);
3625 if (stopped()) break;
3626 args[which_arg] = arg;
3627
3628 Node* p = basic_plus_adr(arg, class_klass_offset);
3629 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3630 klasses[which_arg] = _gvn.transform(kls);
3631 }
3632
3633 // Having loaded both klasses, test each for null.
3634 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3635 for (which_arg = 0; which_arg <= 1; which_arg++) {
3636 Node* kls = klasses[which_arg];
3637 Node* null_ctl = top();
3638 kls = null_check_oop(kls, &null_ctl, never_see_null);
3639 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3640 region->init_req(prim_path, null_ctl);
3641 if (stopped()) break;
3642 klasses[which_arg] = kls;
3643 }
3644
3645 if (!stopped()) {
3646 // now we have two reference types, in klasses[0..1]
3647 Node* subk = klasses[1]; // the argument to isAssignableFrom
3648 Node* superk = klasses[0]; // the receiver
3649 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3650 // now we have a successful reference subtype check
3651 region->set_req(_ref_subtype_path, control());
3652 }
3653
3654 // If both operands are primitive (both klasses null), then
3655 // we must return true when they are identical primitives.
3656 // It is convenient to test this after the first null klass check.
3657 set_control(region->in(_prim_0_path)); // go back to first null check
3658 if (!stopped()) {
3659 // Since superc is primitive, make a guard for the superc==subc case.
3660 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3661 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3662 generate_guard(bol_eq, region, PROB_FAIR);
3663 if (region->req() == PATH_LIMIT+1) {
3664 // A guard was added. If the added guard is taken, superc==subc.
3665 region->swap_edges(PATH_LIMIT, _prim_same_path);
3666 region->del_req(PATH_LIMIT);
3667 }
3668 region->set_req(_prim_0_path, control()); // Not equal after all.
3669 }
3670
3671 // these are the only paths that produce 'true':
3672 phi->set_req(_prim_same_path, intcon(1));
3673 phi->set_req(_ref_subtype_path, intcon(1));
3674
3675 // pull together the cases:
3676 assert(region->req() == PATH_LIMIT, "sane region");
3677 for (uint i = 1; i < region->req(); i++) {
3678 Node* ctl = region->in(i);
3679 if (ctl == NULL || ctl == top()) {
3680 region->set_req(i, top());
3681 phi ->set_req(i, top());
3682 } else if (phi->in(i) == NULL) {
3683 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3684 }
3685 }
3686
3687 set_control(_gvn.transform(region));
3688 set_result(_gvn.transform(phi));
3689 return true;
3690 }
3691
3692 //---------------------generate_array_guard_common------------------------
3693 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3694 bool obj_array, bool not_array) {
3695
3696 if (stopped()) {
3697 return NULL;
3698 }
3699
3700 // If obj_array/non_array==false/false:
3701 // Branch around if the given klass is in fact an array (either obj or prim).
3702 // If obj_array/non_array==false/true:
3703 // Branch around if the given klass is not an array klass of any kind.
3704 // If obj_array/non_array==true/true:
3705 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3706 // If obj_array/non_array==true/false:
3707 // Branch around if the kls is an oop array (Object[] or subtype)
3708 //
3709 // Like generate_guard, adds a new path onto the region.
3710 jint layout_con = 0;
3711 Node* layout_val = get_layout_helper(kls, layout_con);
3712 if (layout_val == NULL) {
3713 bool query = (obj_array
3714 ? Klass::layout_helper_is_objArray(layout_con)
3715 : Klass::layout_helper_is_array(layout_con));
3716 if (query == not_array) {
3717 return NULL; // never a branch
3718 } else { // always a branch
3719 Node* always_branch = control();
3720 if (region != NULL)
3721 region->add_req(always_branch);
3722 set_control(top());
3723 return always_branch;
3724 }
3725 }
3726 // Now test the correct condition.
3727 jint nval = (obj_array
3728 ? (jint)(Klass::_lh_array_tag_type_value
3729 << Klass::_lh_array_tag_shift)
3730 : Klass::_lh_neutral_value);
3731 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3732 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3733 // invert the test if we are looking for a non-array
3734 if (not_array) btest = BoolTest(btest).negate();
3735 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3736 return generate_fair_guard(bol, region);
3737 }
3738
3739
3740 //-----------------------inline_native_newArray--------------------------
3741 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3742 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3743 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3744 Node* mirror;
3745 Node* count_val;
3746 if (uninitialized) {
3747 mirror = argument(1);
3748 count_val = argument(2);
3749 } else {
3750 mirror = argument(0);
3751 count_val = argument(1);
3752 }
3753
3754 mirror = null_check(mirror);
3755 // If mirror or obj is dead, only null-path is taken.
3756 if (stopped()) return true;
3757
3758 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3759 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3760 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3761 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3762 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3763
3764 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3765 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3766 result_reg, _slow_path);
3767 Node* normal_ctl = control();
3768 Node* no_array_ctl = result_reg->in(_slow_path);
3769
3770 // Generate code for the slow case. We make a call to newArray().
3771 set_control(no_array_ctl);
3772 if (!stopped()) {
3773 // Either the input type is void.class, or else the
3774 // array klass has not yet been cached. Either the
3775 // ensuing call will throw an exception, or else it
3776 // will cache the array klass for next time.
3777 PreserveJVMState pjvms(this);
3778 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3779 Node* slow_result = set_results_for_java_call(slow_call);
3780 // this->control() comes from set_results_for_java_call
3781 result_reg->set_req(_slow_path, control());
3782 result_val->set_req(_slow_path, slow_result);
3783 result_io ->set_req(_slow_path, i_o());
3784 result_mem->set_req(_slow_path, reset_memory());
3785 }
3786
3787 set_control(normal_ctl);
3788 if (!stopped()) {
3789 // Normal case: The array type has been cached in the java.lang.Class.
3790 // The following call works fine even if the array type is polymorphic.
3791 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3792 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3793 result_reg->init_req(_normal_path, control());
3794 result_val->init_req(_normal_path, obj);
3795 result_io ->init_req(_normal_path, i_o());
3796 result_mem->init_req(_normal_path, reset_memory());
3797
3798 if (uninitialized) {
3799 // Mark the allocation so that zeroing is skipped
3800 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3801 alloc->maybe_set_complete(&_gvn);
3802 }
3803 }
3804
3805 // Return the combined state.
3806 set_i_o( _gvn.transform(result_io) );
3807 set_all_memory( _gvn.transform(result_mem));
3808
3809 C->set_has_split_ifs(true); // Has chance for split-if optimization
3810 set_result(result_reg, result_val);
3811 return true;
3812 }
3813
3814 //----------------------inline_native_getLength--------------------------
3815 // public static native int java.lang.reflect.Array.getLength(Object array);
3816 bool LibraryCallKit::inline_native_getLength() {
3817 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3818
3819 Node* array = null_check(argument(0));
3820 // If array is dead, only null-path is taken.
3821 if (stopped()) return true;
3822
3823 // Deoptimize if it is a non-array.
3824 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3825
3826 if (non_array != NULL) {
3827 PreserveJVMState pjvms(this);
3828 set_control(non_array);
3829 uncommon_trap(Deoptimization::Reason_intrinsic,
3830 Deoptimization::Action_maybe_recompile);
3831 }
3832
3833 // If control is dead, only non-array-path is taken.
3834 if (stopped()) return true;
3835
3836 // The works fine even if the array type is polymorphic.
3837 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3838 Node* result = load_array_length(array);
3839
3840 C->set_has_split_ifs(true); // Has chance for split-if optimization
3841 set_result(result);
3842 return true;
3843 }
3844
3845 //------------------------inline_array_copyOf----------------------------
3846 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3847 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3848 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3849 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3850
3851 // Get the arguments.
3852 Node* original = argument(0);
3853 Node* start = is_copyOfRange? argument(1): intcon(0);
3854 Node* end = is_copyOfRange? argument(2): argument(1);
3855 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3856
3857 Node* newcopy = NULL;
3858
3859 // Set the original stack and the reexecute bit for the interpreter to reexecute
3860 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3861 { PreserveReexecuteState preexecs(this);
3862 jvms()->set_should_reexecute(true);
3863
3864 array_type_mirror = null_check(array_type_mirror);
3865 original = null_check(original);
3866
3867 // Check if a null path was taken unconditionally.
3868 if (stopped()) return true;
3869
3870 Node* orig_length = load_array_length(original);
3871
3872 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3873 klass_node = null_check(klass_node);
3874
3875 RegionNode* bailout = new RegionNode(1);
3876 record_for_igvn(bailout);
3877
3878 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3879 // Bail out if that is so.
3880 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3881 if (not_objArray != NULL) {
3882 // Improve the klass node's type from the new optimistic assumption:
3883 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3884 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3885 Node* cast = new CastPPNode(klass_node, akls);
3886 cast->init_req(0, control());
3887 klass_node = _gvn.transform(cast);
3888 }
3889
3890 // Bail out if either start or end is negative.
3891 generate_negative_guard(start, bailout, &start);
3892 generate_negative_guard(end, bailout, &end);
3893
3894 Node* length = end;
3895 if (_gvn.type(start) != TypeInt::ZERO) {
3896 length = _gvn.transform(new SubINode(end, start));
3897 }
3898
3899 // Bail out if length is negative.
3900 // Without this the new_array would throw
3901 // NegativeArraySizeException but IllegalArgumentException is what
3902 // should be thrown
3903 generate_negative_guard(length, bailout, &length);
3904
3905 if (bailout->req() > 1) {
3906 PreserveJVMState pjvms(this);
3907 set_control(_gvn.transform(bailout));
3908 uncommon_trap(Deoptimization::Reason_intrinsic,
3909 Deoptimization::Action_maybe_recompile);
3910 }
3911
3912 if (!stopped()) {
3913 // How many elements will we copy from the original?
3914 // The answer is MinI(orig_length - start, length).
3915 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3916 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3917
3918 // Generate a direct call to the right arraycopy function(s).
3919 // We know the copy is disjoint but we might not know if the
3920 // oop stores need checking.
3921 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3922 // This will fail a store-check if x contains any non-nulls.
3923
3924 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3925 // loads/stores but it is legal only if we're sure the
3926 // Arrays.copyOf would succeed. So we need all input arguments
3927 // to the copyOf to be validated, including that the copy to the
3928 // new array won't trigger an ArrayStoreException. That subtype
3929 // check can be optimized if we know something on the type of
3930 // the input array from type speculation.
3931 if (_gvn.type(klass_node)->singleton()) {
3932 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3933 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3934
3935 int test = C->static_subtype_check(superk, subk);
3936 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3937 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3938 if (t_original->speculative_type() != NULL) {
3939 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3940 }
3941 }
3942 }
3943
3944 bool validated = false;
3945 // Reason_class_check rather than Reason_intrinsic because we
3946 // want to intrinsify even if this traps.
3947 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3948 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3949 klass_node);
3950
3951 if (not_subtype_ctrl != top()) {
3952 PreserveJVMState pjvms(this);
3953 set_control(not_subtype_ctrl);
3954 uncommon_trap(Deoptimization::Reason_class_check,
3955 Deoptimization::Action_make_not_entrant);
3956 assert(stopped(), "Should be stopped");
3957 }
3958 validated = true;
3959 }
3960
3961 if (!stopped()) {
3962 newcopy = new_array(klass_node, length, 0); // no arguments to push
3963
3964 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
3965 load_object_klass(original), klass_node);
3966 if (!is_copyOfRange) {
3967 ac->set_copyof(validated);
3968 } else {
3969 ac->set_copyofrange(validated);
3970 }
3971 Node* n = _gvn.transform(ac);
3972 if (n == ac) {
3973 ac->connect_outputs(this);
3974 } else {
3975 assert(validated, "shouldn't transform if all arguments not validated");
3976 set_all_memory(n);
3977 }
3978 }
3979 }
3980 } // original reexecute is set back here
3981
3982 C->set_has_split_ifs(true); // Has chance for split-if optimization
3983 if (!stopped()) {
3984 set_result(newcopy);
3985 }
3986 return true;
3987 }
3988
3989
3990 //----------------------generate_virtual_guard---------------------------
3991 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3992 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3993 RegionNode* slow_region) {
3994 ciMethod* method = callee();
3995 int vtable_index = method->vtable_index();
3996 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3997 "bad index %d", vtable_index);
3998 // Get the Method* out of the appropriate vtable entry.
3999 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4000 vtable_index*vtableEntry::size_in_bytes() +
4001 vtableEntry::method_offset_in_bytes();
4002 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4003 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4004
4005 // Compare the target method with the expected method (e.g., Object.hashCode).
4006 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4007
4008 Node* native_call = makecon(native_call_addr);
4009 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4010 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4011
4012 return generate_slow_guard(test_native, slow_region);
4013 }
4014
4015 //-----------------------generate_method_call----------------------------
4016 // Use generate_method_call to make a slow-call to the real
4017 // method if the fast path fails. An alternative would be to
4018 // use a stub like OptoRuntime::slow_arraycopy_Java.
4019 // This only works for expanding the current library call,
4020 // not another intrinsic. (E.g., don't use this for making an
4021 // arraycopy call inside of the copyOf intrinsic.)
4022 CallJavaNode*
4023 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4024 // When compiling the intrinsic method itself, do not use this technique.
4025 guarantee(callee() != C->method(), "cannot make slow-call to self");
4026
4027 ciMethod* method = callee();
4028 // ensure the JVMS we have will be correct for this call
4029 guarantee(method_id == method->intrinsic_id(), "must match");
4030
4031 const TypeFunc* tf = TypeFunc::make(method);
4032 CallJavaNode* slow_call;
4033 if (is_static) {
4034 assert(!is_virtual, "");
4035 slow_call = new CallStaticJavaNode(C, tf,
4036 SharedRuntime::get_resolve_static_call_stub(),
4037 method, bci());
4038 } else if (is_virtual) {
4039 null_check_receiver();
4040 int vtable_index = Method::invalid_vtable_index;
4041 if (UseInlineCaches) {
4042 // Suppress the vtable call
4043 } else {
4044 // hashCode and clone are not a miranda methods,
4045 // so the vtable index is fixed.
4046 // No need to use the linkResolver to get it.
4047 vtable_index = method->vtable_index();
4048 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4049 "bad index %d", vtable_index);
4050 }
4051 slow_call = new CallDynamicJavaNode(tf,
4052 SharedRuntime::get_resolve_virtual_call_stub(),
4053 method, vtable_index, bci());
4054 } else { // neither virtual nor static: opt_virtual
4055 null_check_receiver();
4056 slow_call = new CallStaticJavaNode(C, tf,
4057 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4058 method, bci());
4059 slow_call->set_optimized_virtual(true);
4060 }
4061 set_arguments_for_java_call(slow_call);
4062 set_edges_for_java_call(slow_call);
4063 return slow_call;
4064 }
4065
4066
4067 /**
4068 * Build special case code for calls to hashCode on an object. This call may
4069 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4070 * slightly different code.
4071 */
4072 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4073 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4074 assert(!(is_virtual && is_static), "either virtual, special, or static");
4075
4076 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4077
4078 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4079 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4080 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4081 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4082 Node* obj = NULL;
4083 if (!is_static) {
4084 // Check for hashing null object
4085 obj = null_check_receiver();
4086 if (stopped()) return true; // unconditionally null
4087 result_reg->init_req(_null_path, top());
4088 result_val->init_req(_null_path, top());
4089 } else {
4090 // Do a null check, and return zero if null.
4091 // System.identityHashCode(null) == 0
4092 obj = argument(0);
4093 Node* null_ctl = top();
4094 obj = null_check_oop(obj, &null_ctl);
4095 result_reg->init_req(_null_path, null_ctl);
4096 result_val->init_req(_null_path, _gvn.intcon(0));
4097 }
4098
4099 // Unconditionally null? Then return right away.
4100 if (stopped()) {
4101 set_control( result_reg->in(_null_path));
4102 if (!stopped())
4103 set_result(result_val->in(_null_path));
4104 return true;
4105 }
4106
4107 // We only go to the fast case code if we pass a number of guards. The
4108 // paths which do not pass are accumulated in the slow_region.
4109 RegionNode* slow_region = new RegionNode(1);
4110 record_for_igvn(slow_region);
4111
4112 // If this is a virtual call, we generate a funny guard. We pull out
4113 // the vtable entry corresponding to hashCode() from the target object.
4114 // If the target method which we are calling happens to be the native
4115 // Object hashCode() method, we pass the guard. We do not need this
4116 // guard for non-virtual calls -- the caller is known to be the native
4117 // Object hashCode().
4118 if (is_virtual) {
4119 // After null check, get the object's klass.
4120 Node* obj_klass = load_object_klass(obj);
4121 generate_virtual_guard(obj_klass, slow_region);
4122 }
4123
4124 // Get the header out of the object, use LoadMarkNode when available
4125 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4126 // The control of the load must be NULL. Otherwise, the load can move before
4127 // the null check after castPP removal.
4128 Node* no_ctrl = NULL;
4129 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4130
4131 // Test the header to see if it is unlocked.
4132 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4133 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4134 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4135 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4136 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4137
4138 generate_slow_guard(test_unlocked, slow_region);
4139
4140 // Get the hash value and check to see that it has been properly assigned.
4141 // We depend on hash_mask being at most 32 bits and avoid the use of
4142 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4143 // vm: see markOop.hpp.
4144 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4145 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4146 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4147 // This hack lets the hash bits live anywhere in the mark object now, as long
4148 // as the shift drops the relevant bits into the low 32 bits. Note that
4149 // Java spec says that HashCode is an int so there's no point in capturing
4150 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4151 hshifted_header = ConvX2I(hshifted_header);
4152 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4153
4154 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4155 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4156 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4157
4158 generate_slow_guard(test_assigned, slow_region);
4159
4160 Node* init_mem = reset_memory();
4161 // fill in the rest of the null path:
4162 result_io ->init_req(_null_path, i_o());
4163 result_mem->init_req(_null_path, init_mem);
4164
4165 result_val->init_req(_fast_path, hash_val);
4166 result_reg->init_req(_fast_path, control());
4167 result_io ->init_req(_fast_path, i_o());
4168 result_mem->init_req(_fast_path, init_mem);
4169
4170 // Generate code for the slow case. We make a call to hashCode().
4171 set_control(_gvn.transform(slow_region));
4172 if (!stopped()) {
4173 // No need for PreserveJVMState, because we're using up the present state.
4174 set_all_memory(init_mem);
4175 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4176 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4177 Node* slow_result = set_results_for_java_call(slow_call);
4178 // this->control() comes from set_results_for_java_call
4179 result_reg->init_req(_slow_path, control());
4180 result_val->init_req(_slow_path, slow_result);
4181 result_io ->set_req(_slow_path, i_o());
4182 result_mem ->set_req(_slow_path, reset_memory());
4183 }
4184
4185 // Return the combined state.
4186 set_i_o( _gvn.transform(result_io) );
4187 set_all_memory( _gvn.transform(result_mem));
4188
4189 set_result(result_reg, result_val);
4190 return true;
4191 }
4192
4193 //---------------------------inline_native_getClass----------------------------
4194 // public final native Class<?> java.lang.Object.getClass();
4195 //
4196 // Build special case code for calls to getClass on an object.
4197 bool LibraryCallKit::inline_native_getClass() {
4198 Node* obj = null_check_receiver();
4199 if (stopped()) return true;
4200 set_result(load_mirror_from_klass(load_object_klass(obj)));
4201 return true;
4202 }
4203
4204 //-----------------inline_native_Reflection_getCallerClass---------------------
4205 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4206 //
4207 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4208 //
4209 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4210 // in that it must skip particular security frames and checks for
4211 // caller sensitive methods.
4212 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4213 #ifndef PRODUCT
4214 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4215 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4216 }
4217 #endif
4218
4219 if (!jvms()->has_method()) {
4220 #ifndef PRODUCT
4221 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4222 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4223 }
4224 #endif
4225 return false;
4226 }
4227
4228 // Walk back up the JVM state to find the caller at the required
4229 // depth.
4230 JVMState* caller_jvms = jvms();
4231
4232 // Cf. JVM_GetCallerClass
4233 // NOTE: Start the loop at depth 1 because the current JVM state does
4234 // not include the Reflection.getCallerClass() frame.
4235 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4236 ciMethod* m = caller_jvms->method();
4237 switch (n) {
4238 case 0:
4239 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4240 break;
4241 case 1:
4242 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4243 if (!m->caller_sensitive()) {
4244 #ifndef PRODUCT
4245 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4246 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4247 }
4248 #endif
4249 return false; // bail-out; let JVM_GetCallerClass do the work
4250 }
4251 break;
4252 default:
4253 if (!m->is_ignored_by_security_stack_walk()) {
4254 // We have reached the desired frame; return the holder class.
4255 // Acquire method holder as java.lang.Class and push as constant.
4256 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4257 ciInstance* caller_mirror = caller_klass->java_mirror();
4258 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4259
4260 #ifndef PRODUCT
4261 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4262 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());
4263 tty->print_cr(" JVM state at this point:");
4264 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4265 ciMethod* m = jvms()->of_depth(i)->method();
4266 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4267 }
4268 }
4269 #endif
4270 return true;
4271 }
4272 break;
4273 }
4274 }
4275
4276 #ifndef PRODUCT
4277 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4278 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4279 tty->print_cr(" JVM state at this point:");
4280 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4281 ciMethod* m = jvms()->of_depth(i)->method();
4282 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4283 }
4284 }
4285 #endif
4286
4287 return false; // bail-out; let JVM_GetCallerClass do the work
4288 }
4289
4290 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4291 Node* arg = argument(0);
4292 Node* result = NULL;
4293
4294 switch (id) {
4295 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4296 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4297 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4298 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4299
4300 case vmIntrinsics::_doubleToLongBits: {
4301 // two paths (plus control) merge in a wood
4302 RegionNode *r = new RegionNode(3);
4303 Node *phi = new PhiNode(r, TypeLong::LONG);
4304
4305 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4306 // Build the boolean node
4307 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4308
4309 // Branch either way.
4310 // NaN case is less traveled, which makes all the difference.
4311 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4312 Node *opt_isnan = _gvn.transform(ifisnan);
4313 assert( opt_isnan->is_If(), "Expect an IfNode");
4314 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4315 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4316
4317 set_control(iftrue);
4318
4319 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4320 Node *slow_result = longcon(nan_bits); // return NaN
4321 phi->init_req(1, _gvn.transform( slow_result ));
4322 r->init_req(1, iftrue);
4323
4324 // Else fall through
4325 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4326 set_control(iffalse);
4327
4328 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4329 r->init_req(2, iffalse);
4330
4331 // Post merge
4332 set_control(_gvn.transform(r));
4333 record_for_igvn(r);
4334
4335 C->set_has_split_ifs(true); // Has chance for split-if optimization
4336 result = phi;
4337 assert(result->bottom_type()->isa_long(), "must be");
4338 break;
4339 }
4340
4341 case vmIntrinsics::_floatToIntBits: {
4342 // two paths (plus control) merge in a wood
4343 RegionNode *r = new RegionNode(3);
4344 Node *phi = new PhiNode(r, TypeInt::INT);
4345
4346 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4347 // Build the boolean node
4348 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4349
4350 // Branch either way.
4351 // NaN case is less traveled, which makes all the difference.
4352 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4353 Node *opt_isnan = _gvn.transform(ifisnan);
4354 assert( opt_isnan->is_If(), "Expect an IfNode");
4355 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4356 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4357
4358 set_control(iftrue);
4359
4360 static const jint nan_bits = 0x7fc00000;
4361 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4362 phi->init_req(1, _gvn.transform( slow_result ));
4363 r->init_req(1, iftrue);
4364
4365 // Else fall through
4366 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4367 set_control(iffalse);
4368
4369 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4370 r->init_req(2, iffalse);
4371
4372 // Post merge
4373 set_control(_gvn.transform(r));
4374 record_for_igvn(r);
4375
4376 C->set_has_split_ifs(true); // Has chance for split-if optimization
4377 result = phi;
4378 assert(result->bottom_type()->isa_int(), "must be");
4379 break;
4380 }
4381
4382 default:
4383 fatal_unexpected_iid(id);
4384 break;
4385 }
4386 set_result(_gvn.transform(result));
4387 return true;
4388 }
4389
4390 //----------------------inline_unsafe_copyMemory-------------------------
4391 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4392 bool LibraryCallKit::inline_unsafe_copyMemory() {
4393 if (callee()->is_static()) return false; // caller must have the capability!
4394 null_check_receiver(); // null-check receiver
4395 if (stopped()) return true;
4396
4397 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4398
4399 Node* src_ptr = argument(1); // type: oop
4400 Node* src_off = ConvL2X(argument(2)); // type: long
4401 Node* dst_ptr = argument(4); // type: oop
4402 Node* dst_off = ConvL2X(argument(5)); // type: long
4403 Node* size = ConvL2X(argument(7)); // type: long
4404
4405 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4406 "fieldOffset must be byte-scaled");
4407
4408 Node* src = make_unsafe_address(src_ptr, src_off);
4409 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4410
4411 // Conservatively insert a memory barrier on all memory slices.
4412 // Do not let writes of the copy source or destination float below the copy.
4413 insert_mem_bar(Op_MemBarCPUOrder);
4414
4415 // Call it. Note that the length argument is not scaled.
4416 make_runtime_call(RC_LEAF|RC_NO_FP,
4417 OptoRuntime::fast_arraycopy_Type(),
4418 StubRoutines::unsafe_arraycopy(),
4419 "unsafe_arraycopy",
4420 TypeRawPtr::BOTTOM,
4421 src, dst, size XTOP);
4422
4423 // Do not let reads of the copy destination float above the copy.
4424 insert_mem_bar(Op_MemBarCPUOrder);
4425
4426 return true;
4427 }
4428
4429 //------------------------clone_coping-----------------------------------
4430 // Helper function for inline_native_clone.
4431 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4432 assert(obj_size != NULL, "");
4433 Node* raw_obj = alloc_obj->in(1);
4434 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4435
4436 AllocateNode* alloc = NULL;
4437 if (ReduceBulkZeroing) {
4438 // We will be completely responsible for initializing this object -
4439 // mark Initialize node as complete.
4440 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4441 // The object was just allocated - there should be no any stores!
4442 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4443 // Mark as complete_with_arraycopy so that on AllocateNode
4444 // expansion, we know this AllocateNode is initialized by an array
4445 // copy and a StoreStore barrier exists after the array copy.
4446 alloc->initialization()->set_complete_with_arraycopy();
4447 }
4448
4449 // Copy the fastest available way.
4450 // TODO: generate fields copies for small objects instead.
4451 Node* src = obj;
4452 Node* dest = alloc_obj;
4453 Node* size = _gvn.transform(obj_size);
4454
4455 // Exclude the header but include array length to copy by 8 bytes words.
4456 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4457 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4458 instanceOopDesc::base_offset_in_bytes();
4459 // base_off:
4460 // 8 - 32-bit VM
4461 // 12 - 64-bit VM, compressed klass
4462 // 16 - 64-bit VM, normal klass
4463 if (base_off % BytesPerLong != 0) {
4464 assert(UseCompressedClassPointers, "");
4465 if (is_array) {
4466 // Exclude length to copy by 8 bytes words.
4467 base_off += sizeof(int);
4468 } else {
4469 // Include klass to copy by 8 bytes words.
4470 base_off = instanceOopDesc::klass_offset_in_bytes();
4471 }
4472 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4473 }
4474 src = basic_plus_adr(src, base_off);
4475 dest = basic_plus_adr(dest, base_off);
4476
4477 // Compute the length also, if needed:
4478 Node* countx = size;
4479 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4480 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4481
4482 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4483
4484 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4485 ac->set_clonebasic();
4486 Node* n = _gvn.transform(ac);
4487 if (n == ac) {
4488 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4489 } else {
4490 set_all_memory(n);
4491 }
4492
4493 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4494 if (card_mark) {
4495 assert(!is_array, "");
4496 // Put in store barrier for any and all oops we are sticking
4497 // into this object. (We could avoid this if we could prove
4498 // that the object type contains no oop fields at all.)
4499 Node* no_particular_value = NULL;
4500 Node* no_particular_field = NULL;
4501 int raw_adr_idx = Compile::AliasIdxRaw;
4502 post_barrier(control(),
4503 memory(raw_adr_type),
4504 alloc_obj,
4505 no_particular_field,
4506 raw_adr_idx,
4507 no_particular_value,
4508 T_OBJECT,
4509 false);
4510 }
4511
4512 // Do not let reads from the cloned object float above the arraycopy.
4513 if (alloc != NULL) {
4514 // Do not let stores that initialize this object be reordered with
4515 // a subsequent store that would make this object accessible by
4516 // other threads.
4517 // Record what AllocateNode this StoreStore protects so that
4518 // escape analysis can go from the MemBarStoreStoreNode to the
4519 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4520 // based on the escape status of the AllocateNode.
4521 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4522 } else {
4523 insert_mem_bar(Op_MemBarCPUOrder);
4524 }
4525 }
4526
4527 //------------------------inline_native_clone----------------------------
4528 // protected native Object java.lang.Object.clone();
4529 //
4530 // Here are the simple edge cases:
4531 // null receiver => normal trap
4532 // virtual and clone was overridden => slow path to out-of-line clone
4533 // not cloneable or finalizer => slow path to out-of-line Object.clone
4534 //
4535 // The general case has two steps, allocation and copying.
4536 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4537 //
4538 // Copying also has two cases, oop arrays and everything else.
4539 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4540 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4541 //
4542 // These steps fold up nicely if and when the cloned object's klass
4543 // can be sharply typed as an object array, a type array, or an instance.
4544 //
4545 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4546 PhiNode* result_val;
4547
4548 // Set the reexecute bit for the interpreter to reexecute
4549 // the bytecode that invokes Object.clone if deoptimization happens.
4550 { PreserveReexecuteState preexecs(this);
4551 jvms()->set_should_reexecute(true);
4552
4553 Node* obj = null_check_receiver();
4554 if (stopped()) return true;
4555
4556 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4557
4558 // If we are going to clone an instance, we need its exact type to
4559 // know the number and types of fields to convert the clone to
4560 // loads/stores. Maybe a speculative type can help us.
4561 if (!obj_type->klass_is_exact() &&
4562 obj_type->speculative_type() != NULL &&
4563 obj_type->speculative_type()->is_instance_klass()) {
4564 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4565 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4566 !spec_ik->has_injected_fields()) {
4567 ciKlass* k = obj_type->klass();
4568 if (!k->is_instance_klass() ||
4569 k->as_instance_klass()->is_interface() ||
4570 k->as_instance_klass()->has_subklass()) {
4571 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4572 }
4573 }
4574 }
4575
4576 Node* obj_klass = load_object_klass(obj);
4577 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4578 const TypeOopPtr* toop = ((tklass != NULL)
4579 ? tklass->as_instance_type()
4580 : TypeInstPtr::NOTNULL);
4581
4582 // Conservatively insert a memory barrier on all memory slices.
4583 // Do not let writes into the original float below the clone.
4584 insert_mem_bar(Op_MemBarCPUOrder);
4585
4586 // paths into result_reg:
4587 enum {
4588 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4589 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4590 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4591 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4592 PATH_LIMIT
4593 };
4594 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4595 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4596 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4597 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4598 record_for_igvn(result_reg);
4599
4600 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4601 int raw_adr_idx = Compile::AliasIdxRaw;
4602
4603 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4604 if (array_ctl != NULL) {
4605 // It's an array.
4606 PreserveJVMState pjvms(this);
4607 set_control(array_ctl);
4608 Node* obj_length = load_array_length(obj);
4609 Node* obj_size = NULL;
4610 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4611
4612 if (!use_ReduceInitialCardMarks()) {
4613 // If it is an oop array, it requires very special treatment,
4614 // because card marking is required on each card of the array.
4615 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4616 if (is_obja != NULL) {
4617 PreserveJVMState pjvms2(this);
4618 set_control(is_obja);
4619 // Generate a direct call to the right arraycopy function(s).
4620 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4621 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4622 ac->set_cloneoop();
4623 Node* n = _gvn.transform(ac);
4624 assert(n == ac, "cannot disappear");
4625 ac->connect_outputs(this);
4626
4627 result_reg->init_req(_objArray_path, control());
4628 result_val->init_req(_objArray_path, alloc_obj);
4629 result_i_o ->set_req(_objArray_path, i_o());
4630 result_mem ->set_req(_objArray_path, reset_memory());
4631 }
4632 }
4633 // Otherwise, there are no card marks to worry about.
4634 // (We can dispense with card marks if we know the allocation
4635 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4636 // causes the non-eden paths to take compensating steps to
4637 // simulate a fresh allocation, so that no further
4638 // card marks are required in compiled code to initialize
4639 // the object.)
4640
4641 if (!stopped()) {
4642 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4643
4644 // Present the results of the copy.
4645 result_reg->init_req(_array_path, control());
4646 result_val->init_req(_array_path, alloc_obj);
4647 result_i_o ->set_req(_array_path, i_o());
4648 result_mem ->set_req(_array_path, reset_memory());
4649 }
4650 }
4651
4652 // We only go to the instance fast case code if we pass a number of guards.
4653 // The paths which do not pass are accumulated in the slow_region.
4654 RegionNode* slow_region = new RegionNode(1);
4655 record_for_igvn(slow_region);
4656 if (!stopped()) {
4657 // It's an instance (we did array above). Make the slow-path tests.
4658 // If this is a virtual call, we generate a funny guard. We grab
4659 // the vtable entry corresponding to clone() from the target object.
4660 // If the target method which we are calling happens to be the
4661 // Object clone() method, we pass the guard. We do not need this
4662 // guard for non-virtual calls; the caller is known to be the native
4663 // Object clone().
4664 if (is_virtual) {
4665 generate_virtual_guard(obj_klass, slow_region);
4666 }
4667
4668 // The object must be easily cloneable and must not have a finalizer.
4669 // Both of these conditions may be checked in a single test.
4670 // We could optimize the test further, but we don't care.
4671 generate_access_flags_guard(obj_klass,
4672 // Test both conditions:
4673 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4674 // Must be cloneable but not finalizer:
4675 JVM_ACC_IS_CLONEABLE_FAST,
4676 slow_region);
4677 }
4678
4679 if (!stopped()) {
4680 // It's an instance, and it passed the slow-path tests.
4681 PreserveJVMState pjvms(this);
4682 Node* obj_size = NULL;
4683 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4684 // is reexecuted if deoptimization occurs and there could be problems when merging
4685 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4686 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4687
4688 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4689
4690 // Present the results of the slow call.
4691 result_reg->init_req(_instance_path, control());
4692 result_val->init_req(_instance_path, alloc_obj);
4693 result_i_o ->set_req(_instance_path, i_o());
4694 result_mem ->set_req(_instance_path, reset_memory());
4695 }
4696
4697 // Generate code for the slow case. We make a call to clone().
4698 set_control(_gvn.transform(slow_region));
4699 if (!stopped()) {
4700 PreserveJVMState pjvms(this);
4701 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4702 Node* slow_result = set_results_for_java_call(slow_call);
4703 // this->control() comes from set_results_for_java_call
4704 result_reg->init_req(_slow_path, control());
4705 result_val->init_req(_slow_path, slow_result);
4706 result_i_o ->set_req(_slow_path, i_o());
4707 result_mem ->set_req(_slow_path, reset_memory());
4708 }
4709
4710 // Return the combined state.
4711 set_control( _gvn.transform(result_reg));
4712 set_i_o( _gvn.transform(result_i_o));
4713 set_all_memory( _gvn.transform(result_mem));
4714 } // original reexecute is set back here
4715
4716 set_result(_gvn.transform(result_val));
4717 return true;
4718 }
4719
4720 // If we have a tighly coupled allocation, the arraycopy may take care
4721 // of the array initialization. If one of the guards we insert between
4722 // the allocation and the arraycopy causes a deoptimization, an
4723 // unitialized array will escape the compiled method. To prevent that
4724 // we set the JVM state for uncommon traps between the allocation and
4725 // the arraycopy to the state before the allocation so, in case of
4726 // deoptimization, we'll reexecute the allocation and the
4727 // initialization.
4728 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4729 if (alloc != NULL) {
4730 ciMethod* trap_method = alloc->jvms()->method();
4731 int trap_bci = alloc->jvms()->bci();
4732
4733 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4734 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4735 // Make sure there's no store between the allocation and the
4736 // arraycopy otherwise visible side effects could be rexecuted
4737 // in case of deoptimization and cause incorrect execution.
4738 bool no_interfering_store = true;
4739 Node* mem = alloc->in(TypeFunc::Memory);
4740 if (mem->is_MergeMem()) {
4741 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4742 Node* n = mms.memory();
4743 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4744 assert(n->is_Store(), "what else?");
4745 no_interfering_store = false;
4746 break;
4747 }
4748 }
4749 } else {
4750 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4751 Node* n = mms.memory();
4752 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4753 assert(n->is_Store(), "what else?");
4754 no_interfering_store = false;
4755 break;
4756 }
4757 }
4758 }
4759
4760 if (no_interfering_store) {
4761 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4762 uint size = alloc->req();
4763 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4764 old_jvms->set_map(sfpt);
4765 for (uint i = 0; i < size; i++) {
4766 sfpt->init_req(i, alloc->in(i));
4767 }
4768 // re-push array length for deoptimization
4769 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4770 old_jvms->set_sp(old_jvms->sp()+1);
4771 old_jvms->set_monoff(old_jvms->monoff()+1);
4772 old_jvms->set_scloff(old_jvms->scloff()+1);
4773 old_jvms->set_endoff(old_jvms->endoff()+1);
4774 old_jvms->set_should_reexecute(true);
4775
4776 sfpt->set_i_o(map()->i_o());
4777 sfpt->set_memory(map()->memory());
4778 sfpt->set_control(map()->control());
4779
4780 JVMState* saved_jvms = jvms();
4781 saved_reexecute_sp = _reexecute_sp;
4782
4783 set_jvms(sfpt->jvms());
4784 _reexecute_sp = jvms()->sp();
4785
4786 return saved_jvms;
4787 }
4788 }
4789 }
4790 return NULL;
4791 }
4792
4793 // In case of a deoptimization, we restart execution at the
4794 // allocation, allocating a new array. We would leave an uninitialized
4795 // array in the heap that GCs wouldn't expect. Move the allocation
4796 // after the traps so we don't allocate the array if we
4797 // deoptimize. This is possible because tightly_coupled_allocation()
4798 // guarantees there's no observer of the allocated array at this point
4799 // and the control flow is simple enough.
4800 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4801 if (saved_jvms != NULL && !stopped()) {
4802 assert(alloc != NULL, "only with a tightly coupled allocation");
4803 // restore JVM state to the state at the arraycopy
4804 saved_jvms->map()->set_control(map()->control());
4805 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4806 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4807 // If we've improved the types of some nodes (null check) while
4808 // emitting the guards, propagate them to the current state
4809 map()->replaced_nodes().apply(saved_jvms->map());
4810 set_jvms(saved_jvms);
4811 _reexecute_sp = saved_reexecute_sp;
4812
4813 // Remove the allocation from above the guards
4814 CallProjections callprojs;
4815 alloc->extract_projections(&callprojs, true);
4816 InitializeNode* init = alloc->initialization();
4817 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4818 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4819 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4820 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4821
4822 // move the allocation here (after the guards)
4823 _gvn.hash_delete(alloc);
4824 alloc->set_req(TypeFunc::Control, control());
4825 alloc->set_req(TypeFunc::I_O, i_o());
4826 Node *mem = reset_memory();
4827 set_all_memory(mem);
4828 alloc->set_req(TypeFunc::Memory, mem);
4829 set_control(init->proj_out(TypeFunc::Control));
4830 set_i_o(callprojs.fallthrough_ioproj);
4831
4832 // Update memory as done in GraphKit::set_output_for_allocation()
4833 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4834 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4835 if (ary_type->isa_aryptr() && length_type != NULL) {
4836 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4837 }
4838 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4839 int elemidx = C->get_alias_index(telemref);
4840 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4841 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4842
4843 Node* allocx = _gvn.transform(alloc);
4844 assert(allocx == alloc, "where has the allocation gone?");
4845 assert(dest->is_CheckCastPP(), "not an allocation result?");
4846
4847 _gvn.hash_delete(dest);
4848 dest->set_req(0, control());
4849 Node* destx = _gvn.transform(dest);
4850 assert(destx == dest, "where has the allocation result gone?");
4851 }
4852 }
4853
4854
4855 //------------------------------inline_arraycopy-----------------------
4856 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4857 // Object dest, int destPos,
4858 // int length);
4859 bool LibraryCallKit::inline_arraycopy() {
4860 // Get the arguments.
4861 Node* src = argument(0); // type: oop
4862 Node* src_offset = argument(1); // type: int
4863 Node* dest = argument(2); // type: oop
4864 Node* dest_offset = argument(3); // type: int
4865 Node* length = argument(4); // type: int
4866
4867
4868 // Check for allocation before we add nodes that would confuse
4869 // tightly_coupled_allocation()
4870 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4871
4872 int saved_reexecute_sp = -1;
4873 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4874 // See arraycopy_restore_alloc_state() comment
4875 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4876 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4877 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4878 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4879
4880 // The following tests must be performed
4881 // (1) src and dest are arrays.
4882 // (2) src and dest arrays must have elements of the same BasicType
4883 // (3) src and dest must not be null.
4884 // (4) src_offset must not be negative.
4885 // (5) dest_offset must not be negative.
4886 // (6) length must not be negative.
4887 // (7) src_offset + length must not exceed length of src.
4888 // (8) dest_offset + length must not exceed length of dest.
4889 // (9) each element of an oop array must be assignable
4890
4891 // (3) src and dest must not be null.
4892 // always do this here because we need the JVM state for uncommon traps
4893 Node* null_ctl = top();
4894 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4895 assert(null_ctl->is_top(), "no null control here");
4896 dest = null_check(dest, T_ARRAY);
4897
4898 if (!can_emit_guards) {
4899 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4900 // guards but the arraycopy node could still take advantage of a
4901 // tightly allocated allocation. tightly_coupled_allocation() is
4902 // called again to make sure it takes the null check above into
4903 // account: the null check is mandatory and if it caused an
4904 // uncommon trap to be emitted then the allocation can't be
4905 // considered tightly coupled in this context.
4906 alloc = tightly_coupled_allocation(dest, NULL);
4907 }
4908
4909 bool validated = false;
4910
4911 const Type* src_type = _gvn.type(src);
4912 const Type* dest_type = _gvn.type(dest);
4913 const TypeAryPtr* top_src = src_type->isa_aryptr();
4914 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4915
4916 // Do we have the type of src?
4917 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4918 // Do we have the type of dest?
4919 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4920 // Is the type for src from speculation?
4921 bool src_spec = false;
4922 // Is the type for dest from speculation?
4923 bool dest_spec = false;
4924
4925 if ((!has_src || !has_dest) && can_emit_guards) {
4926 // We don't have sufficient type information, let's see if
4927 // speculative types can help. We need to have types for both src
4928 // and dest so that it pays off.
4929
4930 // Do we already have or could we have type information for src
4931 bool could_have_src = has_src;
4932 // Do we already have or could we have type information for dest
4933 bool could_have_dest = has_dest;
4934
4935 ciKlass* src_k = NULL;
4936 if (!has_src) {
4937 src_k = src_type->speculative_type_not_null();
4938 if (src_k != NULL && src_k->is_array_klass()) {
4939 could_have_src = true;
4940 }
4941 }
4942
4943 ciKlass* dest_k = NULL;
4944 if (!has_dest) {
4945 dest_k = dest_type->speculative_type_not_null();
4946 if (dest_k != NULL && dest_k->is_array_klass()) {
4947 could_have_dest = true;
4948 }
4949 }
4950
4951 if (could_have_src && could_have_dest) {
4952 // This is going to pay off so emit the required guards
4953 if (!has_src) {
4954 src = maybe_cast_profiled_obj(src, src_k, true);
4955 src_type = _gvn.type(src);
4956 top_src = src_type->isa_aryptr();
4957 has_src = (top_src != NULL && top_src->klass() != NULL);
4958 src_spec = true;
4959 }
4960 if (!has_dest) {
4961 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4962 dest_type = _gvn.type(dest);
4963 top_dest = dest_type->isa_aryptr();
4964 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4965 dest_spec = true;
4966 }
4967 }
4968 }
4969
4970 if (has_src && has_dest && can_emit_guards) {
4971 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4972 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4973 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4974 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4975
4976 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4977 // If both arrays are object arrays then having the exact types
4978 // for both will remove the need for a subtype check at runtime
4979 // before the call and may make it possible to pick a faster copy
4980 // routine (without a subtype check on every element)
4981 // Do we have the exact type of src?
4982 bool could_have_src = src_spec;
4983 // Do we have the exact type of dest?
4984 bool could_have_dest = dest_spec;
4985 ciKlass* src_k = top_src->klass();
4986 ciKlass* dest_k = top_dest->klass();
4987 if (!src_spec) {
4988 src_k = src_type->speculative_type_not_null();
4989 if (src_k != NULL && src_k->is_array_klass()) {
4990 could_have_src = true;
4991 }
4992 }
4993 if (!dest_spec) {
4994 dest_k = dest_type->speculative_type_not_null();
4995 if (dest_k != NULL && dest_k->is_array_klass()) {
4996 could_have_dest = true;
4997 }
4998 }
4999 if (could_have_src && could_have_dest) {
5000 // If we can have both exact types, emit the missing guards
5001 if (could_have_src && !src_spec) {
5002 src = maybe_cast_profiled_obj(src, src_k, true);
5003 }
5004 if (could_have_dest && !dest_spec) {
5005 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5006 }
5007 }
5008 }
5009 }
5010
5011 ciMethod* trap_method = method();
5012 int trap_bci = bci();
5013 if (saved_jvms != NULL) {
5014 trap_method = alloc->jvms()->method();
5015 trap_bci = alloc->jvms()->bci();
5016 }
5017
5018 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5019 can_emit_guards &&
5020 !src->is_top() && !dest->is_top()) {
5021 // validate arguments: enables transformation the ArrayCopyNode
5022 validated = true;
5023
5024 RegionNode* slow_region = new RegionNode(1);
5025 record_for_igvn(slow_region);
5026
5027 // (1) src and dest are arrays.
5028 generate_non_array_guard(load_object_klass(src), slow_region);
5029 generate_non_array_guard(load_object_klass(dest), slow_region);
5030
5031 // (2) src and dest arrays must have elements of the same BasicType
5032 // done at macro expansion or at Ideal transformation time
5033
5034 // (4) src_offset must not be negative.
5035 generate_negative_guard(src_offset, slow_region);
5036
5037 // (5) dest_offset must not be negative.
5038 generate_negative_guard(dest_offset, slow_region);
5039
5040 // (7) src_offset + length must not exceed length of src.
5041 generate_limit_guard(src_offset, length,
5042 load_array_length(src),
5043 slow_region);
5044
5045 // (8) dest_offset + length must not exceed length of dest.
5046 generate_limit_guard(dest_offset, length,
5047 load_array_length(dest),
5048 slow_region);
5049
5050 // (9) each element of an oop array must be assignable
5051 Node* src_klass = load_object_klass(src);
5052 Node* dest_klass = load_object_klass(dest);
5053 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5054
5055 if (not_subtype_ctrl != top()) {
5056 PreserveJVMState pjvms(this);
5057 set_control(not_subtype_ctrl);
5058 uncommon_trap(Deoptimization::Reason_intrinsic,
5059 Deoptimization::Action_make_not_entrant);
5060 assert(stopped(), "Should be stopped");
5061 }
5062 {
5063 PreserveJVMState pjvms(this);
5064 set_control(_gvn.transform(slow_region));
5065 uncommon_trap(Deoptimization::Reason_intrinsic,
5066 Deoptimization::Action_make_not_entrant);
5067 assert(stopped(), "Should be stopped");
5068 }
5069 }
5070
5071 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5072
5073 if (stopped()) {
5074 return true;
5075 }
5076
5077 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5078 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5079 // so the compiler has a chance to eliminate them: during macro expansion,
5080 // we have to set their control (CastPP nodes are eliminated).
5081 load_object_klass(src), load_object_klass(dest),
5082 load_array_length(src), load_array_length(dest));
5083
5084 ac->set_arraycopy(validated);
5085
5086 Node* n = _gvn.transform(ac);
5087 if (n == ac) {
5088 ac->connect_outputs(this);
5089 } else {
5090 assert(validated, "shouldn't transform if all arguments not validated");
5091 set_all_memory(n);
5092 }
5093
5094 return true;
5095 }
5096
5097
5098 // Helper function which determines if an arraycopy immediately follows
5099 // an allocation, with no intervening tests or other escapes for the object.
5100 AllocateArrayNode*
5101 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5102 RegionNode* slow_region) {
5103 if (stopped()) return NULL; // no fast path
5104 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5105
5106 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5107 if (alloc == NULL) return NULL;
5108
5109 Node* rawmem = memory(Compile::AliasIdxRaw);
5110 // Is the allocation's memory state untouched?
5111 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5112 // Bail out if there have been raw-memory effects since the allocation.
5113 // (Example: There might have been a call or safepoint.)
5114 return NULL;
5115 }
5116 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5117 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5118 return NULL;
5119 }
5120
5121 // There must be no unexpected observers of this allocation.
5122 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5123 Node* obs = ptr->fast_out(i);
5124 if (obs != this->map()) {
5125 return NULL;
5126 }
5127 }
5128
5129 // This arraycopy must unconditionally follow the allocation of the ptr.
5130 Node* alloc_ctl = ptr->in(0);
5131 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5132
5133 Node* ctl = control();
5134 while (ctl != alloc_ctl) {
5135 // There may be guards which feed into the slow_region.
5136 // Any other control flow means that we might not get a chance
5137 // to finish initializing the allocated object.
5138 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5139 IfNode* iff = ctl->in(0)->as_If();
5140 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5141 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5142 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5143 ctl = iff->in(0); // This test feeds the known slow_region.
5144 continue;
5145 }
5146 // One more try: Various low-level checks bottom out in
5147 // uncommon traps. If the debug-info of the trap omits
5148 // any reference to the allocation, as we've already
5149 // observed, then there can be no objection to the trap.
5150 bool found_trap = false;
5151 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5152 Node* obs = not_ctl->fast_out(j);
5153 if (obs->in(0) == not_ctl && obs->is_Call() &&
5154 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5155 found_trap = true; break;
5156 }
5157 }
5158 if (found_trap) {
5159 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5160 continue;
5161 }
5162 }
5163 return NULL;
5164 }
5165
5166 // If we get this far, we have an allocation which immediately
5167 // precedes the arraycopy, and we can take over zeroing the new object.
5168 // The arraycopy will finish the initialization, and provide
5169 // a new control state to which we will anchor the destination pointer.
5170
5171 return alloc;
5172 }
5173
5174 //-------------inline_encodeISOArray-----------------------------------
5175 // encode char[] to byte[] in ISO_8859_1
5176 bool LibraryCallKit::inline_encodeISOArray() {
5177 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5178 // no receiver since it is static method
5179 Node *src = argument(0);
5180 Node *src_offset = argument(1);
5181 Node *dst = argument(2);
5182 Node *dst_offset = argument(3);
5183 Node *length = argument(4);
5184
5185 const Type* src_type = src->Value(&_gvn);
5186 const Type* dst_type = dst->Value(&_gvn);
5187 const TypeAryPtr* top_src = src_type->isa_aryptr();
5188 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5189 if (top_src == NULL || top_src->klass() == NULL ||
5190 top_dest == NULL || top_dest->klass() == NULL) {
5191 // failed array check
5192 return false;
5193 }
5194
5195 // Figure out the size and type of the elements we will be copying.
5196 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5197 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5198 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5199 return false;
5200 }
5201
5202 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5203 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5204 // 'src_start' points to src array + scaled offset
5205 // 'dst_start' points to dst array + scaled offset
5206
5207 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5208 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5209 enc = _gvn.transform(enc);
5210 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5211 set_memory(res_mem, mtype);
5212 set_result(enc);
5213 return true;
5214 }
5215
5216 //-------------inline_multiplyToLen-----------------------------------
5217 bool LibraryCallKit::inline_multiplyToLen() {
5218 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5219
5220 address stubAddr = StubRoutines::multiplyToLen();
5221 if (stubAddr == NULL) {
5222 return false; // Intrinsic's stub is not implemented on this platform
5223 }
5224 const char* stubName = "multiplyToLen";
5225
5226 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5227
5228 // no receiver because it is a static method
5229 Node* x = argument(0);
5230 Node* xlen = argument(1);
5231 Node* y = argument(2);
5232 Node* ylen = argument(3);
5233 Node* z = argument(4);
5234
5235 const Type* x_type = x->Value(&_gvn);
5236 const Type* y_type = y->Value(&_gvn);
5237 const TypeAryPtr* top_x = x_type->isa_aryptr();
5238 const TypeAryPtr* top_y = y_type->isa_aryptr();
5239 if (top_x == NULL || top_x->klass() == NULL ||
5240 top_y == NULL || top_y->klass() == NULL) {
5241 // failed array check
5242 return false;
5243 }
5244
5245 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5246 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5247 if (x_elem != T_INT || y_elem != T_INT) {
5248 return false;
5249 }
5250
5251 // Set the original stack and the reexecute bit for the interpreter to reexecute
5252 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5253 // on the return from z array allocation in runtime.
5254 { PreserveReexecuteState preexecs(this);
5255 jvms()->set_should_reexecute(true);
5256
5257 Node* x_start = array_element_address(x, intcon(0), x_elem);
5258 Node* y_start = array_element_address(y, intcon(0), y_elem);
5259 // 'x_start' points to x array + scaled xlen
5260 // 'y_start' points to y array + scaled ylen
5261
5262 // Allocate the result array
5263 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5264 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5265 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5266
5267 IdealKit ideal(this);
5268
5269 #define __ ideal.
5270 Node* one = __ ConI(1);
5271 Node* zero = __ ConI(0);
5272 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5273 __ set(need_alloc, zero);
5274 __ set(z_alloc, z);
5275 __ if_then(z, BoolTest::eq, null()); {
5276 __ increment (need_alloc, one);
5277 } __ else_(); {
5278 // Update graphKit memory and control from IdealKit.
5279 sync_kit(ideal);
5280 Node* zlen_arg = load_array_length(z);
5281 // Update IdealKit memory and control from graphKit.
5282 __ sync_kit(this);
5283 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5284 __ increment (need_alloc, one);
5285 } __ end_if();
5286 } __ end_if();
5287
5288 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5289 // Update graphKit memory and control from IdealKit.
5290 sync_kit(ideal);
5291 Node * narr = new_array(klass_node, zlen, 1);
5292 // Update IdealKit memory and control from graphKit.
5293 __ sync_kit(this);
5294 __ set(z_alloc, narr);
5295 } __ end_if();
5296
5297 sync_kit(ideal);
5298 z = __ value(z_alloc);
5299 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5300 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5301 // Final sync IdealKit and GraphKit.
5302 final_sync(ideal);
5303 #undef __
5304
5305 Node* z_start = array_element_address(z, intcon(0), T_INT);
5306
5307 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5308 OptoRuntime::multiplyToLen_Type(),
5309 stubAddr, stubName, TypePtr::BOTTOM,
5310 x_start, xlen, y_start, ylen, z_start, zlen);
5311 } // original reexecute is set back here
5312
5313 C->set_has_split_ifs(true); // Has chance for split-if optimization
5314 set_result(z);
5315 return true;
5316 }
5317
5318 //-------------inline_squareToLen------------------------------------
5319 bool LibraryCallKit::inline_squareToLen() {
5320 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5321
5322 address stubAddr = StubRoutines::squareToLen();
5323 if (stubAddr == NULL) {
5324 return false; // Intrinsic's stub is not implemented on this platform
5325 }
5326 const char* stubName = "squareToLen";
5327
5328 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5329
5330 Node* x = argument(0);
5331 Node* len = argument(1);
5332 Node* z = argument(2);
5333 Node* zlen = argument(3);
5334
5335 const Type* x_type = x->Value(&_gvn);
5336 const Type* z_type = z->Value(&_gvn);
5337 const TypeAryPtr* top_x = x_type->isa_aryptr();
5338 const TypeAryPtr* top_z = z_type->isa_aryptr();
5339 if (top_x == NULL || top_x->klass() == NULL ||
5340 top_z == NULL || top_z->klass() == NULL) {
5341 // failed array check
5342 return false;
5343 }
5344
5345 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5346 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5347 if (x_elem != T_INT || z_elem != T_INT) {
5348 return false;
5349 }
5350
5351
5352 Node* x_start = array_element_address(x, intcon(0), x_elem);
5353 Node* z_start = array_element_address(z, intcon(0), z_elem);
5354
5355 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5356 OptoRuntime::squareToLen_Type(),
5357 stubAddr, stubName, TypePtr::BOTTOM,
5358 x_start, len, z_start, zlen);
5359
5360 set_result(z);
5361 return true;
5362 }
5363
5364 //-------------inline_mulAdd------------------------------------------
5365 bool LibraryCallKit::inline_mulAdd() {
5366 assert(UseMulAddIntrinsic, "not implemented on this platform");
5367
5368 address stubAddr = StubRoutines::mulAdd();
5369 if (stubAddr == NULL) {
5370 return false; // Intrinsic's stub is not implemented on this platform
5371 }
5372 const char* stubName = "mulAdd";
5373
5374 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5375
5376 Node* out = argument(0);
5377 Node* in = argument(1);
5378 Node* offset = argument(2);
5379 Node* len = argument(3);
5380 Node* k = argument(4);
5381
5382 const Type* out_type = out->Value(&_gvn);
5383 const Type* in_type = in->Value(&_gvn);
5384 const TypeAryPtr* top_out = out_type->isa_aryptr();
5385 const TypeAryPtr* top_in = in_type->isa_aryptr();
5386 if (top_out == NULL || top_out->klass() == NULL ||
5387 top_in == NULL || top_in->klass() == NULL) {
5388 // failed array check
5389 return false;
5390 }
5391
5392 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5393 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5394 if (out_elem != T_INT || in_elem != T_INT) {
5395 return false;
5396 }
5397
5398 Node* outlen = load_array_length(out);
5399 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5400 Node* out_start = array_element_address(out, intcon(0), out_elem);
5401 Node* in_start = array_element_address(in, intcon(0), in_elem);
5402
5403 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5404 OptoRuntime::mulAdd_Type(),
5405 stubAddr, stubName, TypePtr::BOTTOM,
5406 out_start,in_start, new_offset, len, k);
5407 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5408 set_result(result);
5409 return true;
5410 }
5411
5412 //-------------inline_montgomeryMultiply-----------------------------------
5413 bool LibraryCallKit::inline_montgomeryMultiply() {
5414 address stubAddr = StubRoutines::montgomeryMultiply();
5415 if (stubAddr == NULL) {
5416 return false; // Intrinsic's stub is not implemented on this platform
5417 }
5418
5419 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5420 const char* stubName = "montgomery_square";
5421
5422 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5423
5424 Node* a = argument(0);
5425 Node* b = argument(1);
5426 Node* n = argument(2);
5427 Node* len = argument(3);
5428 Node* inv = argument(4);
5429 Node* m = argument(6);
5430
5431 const Type* a_type = a->Value(&_gvn);
5432 const TypeAryPtr* top_a = a_type->isa_aryptr();
5433 const Type* b_type = b->Value(&_gvn);
5434 const TypeAryPtr* top_b = b_type->isa_aryptr();
5435 const Type* n_type = a->Value(&_gvn);
5436 const TypeAryPtr* top_n = n_type->isa_aryptr();
5437 const Type* m_type = a->Value(&_gvn);
5438 const TypeAryPtr* top_m = m_type->isa_aryptr();
5439 if (top_a == NULL || top_a->klass() == NULL ||
5440 top_b == NULL || top_b->klass() == NULL ||
5441 top_n == NULL || top_n->klass() == NULL ||
5442 top_m == NULL || top_m->klass() == NULL) {
5443 // failed array check
5444 return false;
5445 }
5446
5447 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5448 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5449 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5450 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5451 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5452 return false;
5453 }
5454
5455 // Make the call
5456 {
5457 Node* a_start = array_element_address(a, intcon(0), a_elem);
5458 Node* b_start = array_element_address(b, intcon(0), b_elem);
5459 Node* n_start = array_element_address(n, intcon(0), n_elem);
5460 Node* m_start = array_element_address(m, intcon(0), m_elem);
5461
5462 Node* call = make_runtime_call(RC_LEAF,
5463 OptoRuntime::montgomeryMultiply_Type(),
5464 stubAddr, stubName, TypePtr::BOTTOM,
5465 a_start, b_start, n_start, len, inv, top(),
5466 m_start);
5467 set_result(m);
5468 }
5469
5470 return true;
5471 }
5472
5473 bool LibraryCallKit::inline_montgomerySquare() {
5474 address stubAddr = StubRoutines::montgomerySquare();
5475 if (stubAddr == NULL) {
5476 return false; // Intrinsic's stub is not implemented on this platform
5477 }
5478
5479 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5480 const char* stubName = "montgomery_square";
5481
5482 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5483
5484 Node* a = argument(0);
5485 Node* n = argument(1);
5486 Node* len = argument(2);
5487 Node* inv = argument(3);
5488 Node* m = argument(5);
5489
5490 const Type* a_type = a->Value(&_gvn);
5491 const TypeAryPtr* top_a = a_type->isa_aryptr();
5492 const Type* n_type = a->Value(&_gvn);
5493 const TypeAryPtr* top_n = n_type->isa_aryptr();
5494 const Type* m_type = a->Value(&_gvn);
5495 const TypeAryPtr* top_m = m_type->isa_aryptr();
5496 if (top_a == NULL || top_a->klass() == NULL ||
5497 top_n == NULL || top_n->klass() == NULL ||
5498 top_m == NULL || top_m->klass() == NULL) {
5499 // failed array check
5500 return false;
5501 }
5502
5503 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5504 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5505 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5506 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5507 return false;
5508 }
5509
5510 // Make the call
5511 {
5512 Node* a_start = array_element_address(a, intcon(0), a_elem);
5513 Node* n_start = array_element_address(n, intcon(0), n_elem);
5514 Node* m_start = array_element_address(m, intcon(0), m_elem);
5515
5516 Node* call = make_runtime_call(RC_LEAF,
5517 OptoRuntime::montgomerySquare_Type(),
5518 stubAddr, stubName, TypePtr::BOTTOM,
5519 a_start, n_start, len, inv, top(),
5520 m_start);
5521 set_result(m);
5522 }
5523
5524 return true;
5525 }
5526
5527 //-------------inline_vectorizedMismatch------------------------------
5528 bool LibraryCallKit::inline_vectorizedMismatch() {
5529 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5530
5531 address stubAddr = StubRoutines::vectorizedMismatch();
5532 if (stubAddr == NULL) {
5533 return false; // Intrinsic's stub is not implemented on this platform
5534 }
5535 const char* stubName = "vectorizedMismatch";
5536 int size_l = callee()->signature()->size();
5537 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5538
5539 Node* obja = argument(0);
5540 Node* aoffset = argument(1);
5541 Node* objb = argument(3);
5542 Node* boffset = argument(4);
5543 Node* length = argument(6);
5544 Node* scale = argument(7);
5545
5546 const Type* a_type = obja->Value(&_gvn);
5547 const Type* b_type = objb->Value(&_gvn);
5548 const TypeAryPtr* top_a = a_type->isa_aryptr();
5549 const TypeAryPtr* top_b = b_type->isa_aryptr();
5550 if (top_a == NULL || top_a->klass() == NULL ||
5551 top_b == NULL || top_b->klass() == NULL) {
5552 // failed array check
5553 return false;
5554 }
5555
5556 Node* call;
5557 jvms()->set_should_reexecute(true);
5558
5559 Node* obja_adr = make_unsafe_address(obja, aoffset);
5560 Node* objb_adr = make_unsafe_address(objb, boffset);
5561
5562 call = make_runtime_call(RC_LEAF,
5563 OptoRuntime::vectorizedMismatch_Type(),
5564 stubAddr, stubName, TypePtr::BOTTOM,
5565 obja_adr, objb_adr, length, scale);
5566
5567 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5568 set_result(result);
5569 return true;
5570 }
5571
5572 /**
5573 * Calculate CRC32 for byte.
5574 * int java.util.zip.CRC32.update(int crc, int b)
5575 */
5576 bool LibraryCallKit::inline_updateCRC32() {
5577 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5578 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5579 // no receiver since it is static method
5580 Node* crc = argument(0); // type: int
5581 Node* b = argument(1); // type: int
5582
5583 /*
5584 * int c = ~ crc;
5585 * b = timesXtoThe32[(b ^ c) & 0xFF];
5586 * b = b ^ (c >>> 8);
5587 * crc = ~b;
5588 */
5589
5590 Node* M1 = intcon(-1);
5591 crc = _gvn.transform(new XorINode(crc, M1));
5592 Node* result = _gvn.transform(new XorINode(crc, b));
5593 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5594
5595 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5596 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5597 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5598 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5599
5600 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5601 result = _gvn.transform(new XorINode(crc, result));
5602 result = _gvn.transform(new XorINode(result, M1));
5603 set_result(result);
5604 return true;
5605 }
5606
5607 /**
5608 * Calculate CRC32 for byte[] array.
5609 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5610 */
5611 bool LibraryCallKit::inline_updateBytesCRC32() {
5612 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5613 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5614 // no receiver since it is static method
5615 Node* crc = argument(0); // type: int
5616 Node* src = argument(1); // type: oop
5617 Node* offset = argument(2); // type: int
5618 Node* length = argument(3); // type: int
5619
5620 const Type* src_type = src->Value(&_gvn);
5621 const TypeAryPtr* top_src = src_type->isa_aryptr();
5622 if (top_src == NULL || top_src->klass() == NULL) {
5623 // failed array check
5624 return false;
5625 }
5626
5627 // Figure out the size and type of the elements we will be copying.
5628 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5629 if (src_elem != T_BYTE) {
5630 return false;
5631 }
5632
5633 // 'src_start' points to src array + scaled offset
5634 Node* src_start = array_element_address(src, offset, src_elem);
5635
5636 // We assume that range check is done by caller.
5637 // TODO: generate range check (offset+length < src.length) in debug VM.
5638
5639 // Call the stub.
5640 address stubAddr = StubRoutines::updateBytesCRC32();
5641 const char *stubName = "updateBytesCRC32";
5642
5643 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5644 stubAddr, stubName, TypePtr::BOTTOM,
5645 crc, src_start, length);
5646 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5647 set_result(result);
5648 return true;
5649 }
5650
5651 /**
5652 * Calculate CRC32 for ByteBuffer.
5653 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5654 */
5655 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5656 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5657 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5658 // no receiver since it is static method
5659 Node* crc = argument(0); // type: int
5660 Node* src = argument(1); // type: long
5661 Node* offset = argument(3); // type: int
5662 Node* length = argument(4); // type: int
5663
5664 src = ConvL2X(src); // adjust Java long to machine word
5665 Node* base = _gvn.transform(new CastX2PNode(src));
5666 offset = ConvI2X(offset);
5667
5668 // 'src_start' points to src array + scaled offset
5669 Node* src_start = basic_plus_adr(top(), base, offset);
5670
5671 // Call the stub.
5672 address stubAddr = StubRoutines::updateBytesCRC32();
5673 const char *stubName = "updateBytesCRC32";
5674
5675 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5676 stubAddr, stubName, TypePtr::BOTTOM,
5677 crc, src_start, length);
5678 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5679 set_result(result);
5680 return true;
5681 }
5682
5683 //------------------------------get_table_from_crc32c_class-----------------------
5684 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5685 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5686 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5687
5688 return table;
5689 }
5690
5691 //------------------------------inline_updateBytesCRC32C-----------------------
5692 //
5693 // Calculate CRC32C for byte[] array.
5694 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5695 //
5696 bool LibraryCallKit::inline_updateBytesCRC32C() {
5697 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5698 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5699 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5700 // no receiver since it is a static method
5701 Node* crc = argument(0); // type: int
5702 Node* src = argument(1); // type: oop
5703 Node* offset = argument(2); // type: int
5704 Node* end = argument(3); // type: int
5705
5706 Node* length = _gvn.transform(new SubINode(end, offset));
5707
5708 const Type* src_type = src->Value(&_gvn);
5709 const TypeAryPtr* top_src = src_type->isa_aryptr();
5710 if (top_src == NULL || top_src->klass() == NULL) {
5711 // failed array check
5712 return false;
5713 }
5714
5715 // Figure out the size and type of the elements we will be copying.
5716 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5717 if (src_elem != T_BYTE) {
5718 return false;
5719 }
5720
5721 // 'src_start' points to src array + scaled offset
5722 Node* src_start = array_element_address(src, offset, src_elem);
5723
5724 // static final int[] byteTable in class CRC32C
5725 Node* table = get_table_from_crc32c_class(callee()->holder());
5726 Node* table_start = array_element_address(table, intcon(0), T_INT);
5727
5728 // We assume that range check is done by caller.
5729 // TODO: generate range check (offset+length < src.length) in debug VM.
5730
5731 // Call the stub.
5732 address stubAddr = StubRoutines::updateBytesCRC32C();
5733 const char *stubName = "updateBytesCRC32C";
5734
5735 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5736 stubAddr, stubName, TypePtr::BOTTOM,
5737 crc, src_start, length, table_start);
5738 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5739 set_result(result);
5740 return true;
5741 }
5742
5743 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5744 //
5745 // Calculate CRC32C for DirectByteBuffer.
5746 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5747 //
5748 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5749 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5750 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5751 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5752 // no receiver since it is a static method
5753 Node* crc = argument(0); // type: int
5754 Node* src = argument(1); // type: long
5755 Node* offset = argument(3); // type: int
5756 Node* end = argument(4); // type: int
5757
5758 Node* length = _gvn.transform(new SubINode(end, offset));
5759
5760 src = ConvL2X(src); // adjust Java long to machine word
5761 Node* base = _gvn.transform(new CastX2PNode(src));
5762 offset = ConvI2X(offset);
5763
5764 // 'src_start' points to src array + scaled offset
5765 Node* src_start = basic_plus_adr(top(), base, offset);
5766
5767 // static final int[] byteTable in class CRC32C
5768 Node* table = get_table_from_crc32c_class(callee()->holder());
5769 Node* table_start = array_element_address(table, intcon(0), T_INT);
5770
5771 // Call the stub.
5772 address stubAddr = StubRoutines::updateBytesCRC32C();
5773 const char *stubName = "updateBytesCRC32C";
5774
5775 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5776 stubAddr, stubName, TypePtr::BOTTOM,
5777 crc, src_start, length, table_start);
5778 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5779 set_result(result);
5780 return true;
5781 }
5782
5783 //------------------------------inline_updateBytesAdler32----------------------
5784 //
5785 // Calculate Adler32 checksum for byte[] array.
5786 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5787 //
5788 bool LibraryCallKit::inline_updateBytesAdler32() {
5789 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5790 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5791 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5792 // no receiver since it is static method
5793 Node* crc = argument(0); // type: int
5794 Node* src = argument(1); // type: oop
5795 Node* offset = argument(2); // type: int
5796 Node* length = argument(3); // type: int
5797
5798 const Type* src_type = src->Value(&_gvn);
5799 const TypeAryPtr* top_src = src_type->isa_aryptr();
5800 if (top_src == NULL || top_src->klass() == NULL) {
5801 // failed array check
5802 return false;
5803 }
5804
5805 // Figure out the size and type of the elements we will be copying.
5806 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5807 if (src_elem != T_BYTE) {
5808 return false;
5809 }
5810
5811 // 'src_start' points to src array + scaled offset
5812 Node* src_start = array_element_address(src, offset, src_elem);
5813
5814 // We assume that range check is done by caller.
5815 // TODO: generate range check (offset+length < src.length) in debug VM.
5816
5817 // Call the stub.
5818 address stubAddr = StubRoutines::updateBytesAdler32();
5819 const char *stubName = "updateBytesAdler32";
5820
5821 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5822 stubAddr, stubName, TypePtr::BOTTOM,
5823 crc, src_start, length);
5824 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5825 set_result(result);
5826 return true;
5827 }
5828
5829 //------------------------------inline_updateByteBufferAdler32---------------
5830 //
5831 // Calculate Adler32 checksum for DirectByteBuffer.
5832 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5833 //
5834 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5835 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5836 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5837 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5838 // no receiver since it is static method
5839 Node* crc = argument(0); // type: int
5840 Node* src = argument(1); // type: long
5841 Node* offset = argument(3); // type: int
5842 Node* length = argument(4); // type: int
5843
5844 src = ConvL2X(src); // adjust Java long to machine word
5845 Node* base = _gvn.transform(new CastX2PNode(src));
5846 offset = ConvI2X(offset);
5847
5848 // 'src_start' points to src array + scaled offset
5849 Node* src_start = basic_plus_adr(top(), base, offset);
5850
5851 // Call the stub.
5852 address stubAddr = StubRoutines::updateBytesAdler32();
5853 const char *stubName = "updateBytesAdler32";
5854
5855 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5856 stubAddr, stubName, TypePtr::BOTTOM,
5857 crc, src_start, length);
5858
5859 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5860 set_result(result);
5861 return true;
5862 }
5863
5864 //----------------------------inline_reference_get----------------------------
5865 // public T java.lang.ref.Reference.get();
5866 bool LibraryCallKit::inline_reference_get() {
5867 const int referent_offset = java_lang_ref_Reference::referent_offset;
5868 guarantee(referent_offset > 0, "should have already been set");
5869
5870 // Get the argument:
5871 Node* reference_obj = null_check_receiver();
5872 if (stopped()) return true;
5873
5874 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5875
5876 ciInstanceKlass* klass = env()->Object_klass();
5877 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5878
5879 Node* no_ctrl = NULL;
5880 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5881
5882 // Use the pre-barrier to record the value in the referent field
5883 pre_barrier(false /* do_load */,
5884 control(),
5885 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5886 result /* pre_val */,
5887 T_OBJECT);
5888
5889 // Add memory barrier to prevent commoning reads from this field
5890 // across safepoint since GC can change its value.
5891 insert_mem_bar(Op_MemBarCPUOrder);
5892
5893 set_result(result);
5894 return true;
5895 }
5896
5897
5898 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5899 bool is_exact=true, bool is_static=false,
5900 ciInstanceKlass * fromKls=NULL) {
5901 if (fromKls == NULL) {
5902 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5903 assert(tinst != NULL, "obj is null");
5904 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5905 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5906 fromKls = tinst->klass()->as_instance_klass();
5907 } else {
5908 assert(is_static, "only for static field access");
5909 }
5910 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5911 ciSymbol::make(fieldTypeString),
5912 is_static);
5913
5914 assert (field != NULL, "undefined field");
5915 if (field == NULL) return (Node *) NULL;
5916
5917 if (is_static) {
5918 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5919 fromObj = makecon(tip);
5920 }
5921
5922 // Next code copied from Parse::do_get_xxx():
5923
5924 // Compute address and memory type.
5925 int offset = field->offset_in_bytes();
5926 bool is_vol = field->is_volatile();
5927 ciType* field_klass = field->type();
5928 assert(field_klass->is_loaded(), "should be loaded");
5929 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5930 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5931 BasicType bt = field->layout_type();
5932
5933 // Build the resultant type of the load
5934 const Type *type;
5935 if (bt == T_OBJECT) {
5936 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5937 } else {
5938 type = Type::get_const_basic_type(bt);
5939 }
5940
5941 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5942 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
5943 }
5944 // Build the load.
5945 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5946 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5947 // If reference is volatile, prevent following memory ops from
5948 // floating up past the volatile read. Also prevents commoning
5949 // another volatile read.
5950 if (is_vol) {
5951 // Memory barrier includes bogus read of value to force load BEFORE membar
5952 insert_mem_bar(Op_MemBarAcquire, loadedField);
5953 }
5954 return loadedField;
5955 }
5956
5957 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5958 bool is_exact = true, bool is_static = false,
5959 ciInstanceKlass * fromKls = NULL) {
5960 if (fromKls == NULL) {
5961 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5962 assert(tinst != NULL, "obj is null");
5963 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5964 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5965 fromKls = tinst->klass()->as_instance_klass();
5966 }
5967 else {
5968 assert(is_static, "only for static field access");
5969 }
5970 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5971 ciSymbol::make(fieldTypeString),
5972 is_static);
5973
5974 assert(field != NULL, "undefined field");
5975 assert(!field->is_volatile(), "not defined for volatile fields");
5976
5977 if (is_static) {
5978 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5979 fromObj = makecon(tip);
5980 }
5981
5982 // Next code copied from Parse::do_get_xxx():
5983
5984 // Compute address and memory type.
5985 int offset = field->offset_in_bytes();
5986 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5987
5988 return adr;
5989 }
5990
5991 //------------------------------inline_aescrypt_Block-----------------------
5992 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5993 address stubAddr = NULL;
5994 const char *stubName;
5995 assert(UseAES, "need AES instruction support");
5996
5997 switch(id) {
5998 case vmIntrinsics::_aescrypt_encryptBlock:
5999 stubAddr = StubRoutines::aescrypt_encryptBlock();
6000 stubName = "aescrypt_encryptBlock";
6001 break;
6002 case vmIntrinsics::_aescrypt_decryptBlock:
6003 stubAddr = StubRoutines::aescrypt_decryptBlock();
6004 stubName = "aescrypt_decryptBlock";
6005 break;
6006 }
6007 if (stubAddr == NULL) return false;
6008
6009 Node* aescrypt_object = argument(0);
6010 Node* src = argument(1);
6011 Node* src_offset = argument(2);
6012 Node* dest = argument(3);
6013 Node* dest_offset = argument(4);
6014
6015 // (1) src and dest are arrays.
6016 const Type* src_type = src->Value(&_gvn);
6017 const Type* dest_type = dest->Value(&_gvn);
6018 const TypeAryPtr* top_src = src_type->isa_aryptr();
6019 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6020 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6021
6022 // for the quick and dirty code we will skip all the checks.
6023 // we are just trying to get the call to be generated.
6024 Node* src_start = src;
6025 Node* dest_start = dest;
6026 if (src_offset != NULL || dest_offset != NULL) {
6027 assert(src_offset != NULL && dest_offset != NULL, "");
6028 src_start = array_element_address(src, src_offset, T_BYTE);
6029 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6030 }
6031
6032 // now need to get the start of its expanded key array
6033 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6034 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6035 if (k_start == NULL) return false;
6036
6037 if (Matcher::pass_original_key_for_aes()) {
6038 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6039 // compatibility issues between Java key expansion and SPARC crypto instructions
6040 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6041 if (original_k_start == NULL) return false;
6042
6043 // Call the stub.
6044 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6045 stubAddr, stubName, TypePtr::BOTTOM,
6046 src_start, dest_start, k_start, original_k_start);
6047 } else {
6048 // Call the stub.
6049 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6050 stubAddr, stubName, TypePtr::BOTTOM,
6051 src_start, dest_start, k_start);
6052 }
6053
6054 return true;
6055 }
6056
6057 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6058 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6059 address stubAddr = NULL;
6060 const char *stubName = NULL;
6061
6062 assert(UseAES, "need AES instruction support");
6063
6064 switch(id) {
6065 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6066 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6067 stubName = "cipherBlockChaining_encryptAESCrypt";
6068 break;
6069 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6070 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6071 stubName = "cipherBlockChaining_decryptAESCrypt";
6072 break;
6073 }
6074 if (stubAddr == NULL) return false;
6075
6076 Node* cipherBlockChaining_object = argument(0);
6077 Node* src = argument(1);
6078 Node* src_offset = argument(2);
6079 Node* len = argument(3);
6080 Node* dest = argument(4);
6081 Node* dest_offset = argument(5);
6082
6083 // (1) src and dest are arrays.
6084 const Type* src_type = src->Value(&_gvn);
6085 const Type* dest_type = dest->Value(&_gvn);
6086 const TypeAryPtr* top_src = src_type->isa_aryptr();
6087 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6088 assert (top_src != NULL && top_src->klass() != NULL
6089 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6090
6091 // checks are the responsibility of the caller
6092 Node* src_start = src;
6093 Node* dest_start = dest;
6094 if (src_offset != NULL || dest_offset != NULL) {
6095 assert(src_offset != NULL && dest_offset != NULL, "");
6096 src_start = array_element_address(src, src_offset, T_BYTE);
6097 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6098 }
6099
6100 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6101 // (because of the predicated logic executed earlier).
6102 // so we cast it here safely.
6103 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6104
6105 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6106 if (embeddedCipherObj == NULL) return false;
6107
6108 // cast it to what we know it will be at runtime
6109 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6110 assert(tinst != NULL, "CBC obj is null");
6111 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6112 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6113 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6114
6115 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6116 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6117 const TypeOopPtr* xtype = aklass->as_instance_type();
6118 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6119 aescrypt_object = _gvn.transform(aescrypt_object);
6120
6121 // we need to get the start of the aescrypt_object's expanded key array
6122 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6123 if (k_start == NULL) return false;
6124
6125 // similarly, get the start address of the r vector
6126 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6127 if (objRvec == NULL) return false;
6128 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6129
6130 Node* cbcCrypt;
6131 if (Matcher::pass_original_key_for_aes()) {
6132 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6133 // compatibility issues between Java key expansion and SPARC crypto instructions
6134 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6135 if (original_k_start == NULL) return false;
6136
6137 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6138 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6139 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6140 stubAddr, stubName, TypePtr::BOTTOM,
6141 src_start, dest_start, k_start, r_start, len, original_k_start);
6142 } else {
6143 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6144 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6145 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6146 stubAddr, stubName, TypePtr::BOTTOM,
6147 src_start, dest_start, k_start, r_start, len);
6148 }
6149
6150 // return cipher length (int)
6151 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6152 set_result(retvalue);
6153 return true;
6154 }
6155
6156 //------------------------------inline_counterMode_AESCrypt-----------------------
6157 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6158 assert(UseAES, "need AES instruction support");
6159 if (!UseAESCTRIntrinsics) return false;
6160
6161 address stubAddr = NULL;
6162 const char *stubName = NULL;
6163 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6164 stubAddr = StubRoutines::counterMode_AESCrypt();
6165 stubName = "counterMode_AESCrypt";
6166 }
6167 if (stubAddr == NULL) return false;
6168
6169 Node* counterMode_object = argument(0);
6170 Node* src = argument(1);
6171 Node* src_offset = argument(2);
6172 Node* len = argument(3);
6173 Node* dest = argument(4);
6174 Node* dest_offset = argument(5);
6175
6176 // (1) src and dest are arrays.
6177 const Type* src_type = src->Value(&_gvn);
6178 const Type* dest_type = dest->Value(&_gvn);
6179 const TypeAryPtr* top_src = src_type->isa_aryptr();
6180 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6181 assert(top_src != NULL && top_src->klass() != NULL &&
6182 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6183
6184 // checks are the responsibility of the caller
6185 Node* src_start = src;
6186 Node* dest_start = dest;
6187 if (src_offset != NULL || dest_offset != NULL) {
6188 assert(src_offset != NULL && dest_offset != NULL, "");
6189 src_start = array_element_address(src, src_offset, T_BYTE);
6190 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6191 }
6192
6193 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6194 // (because of the predicated logic executed earlier).
6195 // so we cast it here safely.
6196 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6197 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6198 if (embeddedCipherObj == NULL) return false;
6199 // cast it to what we know it will be at runtime
6200 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6201 assert(tinst != NULL, "CTR obj is null");
6202 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6203 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6204 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6205 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6206 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6207 const TypeOopPtr* xtype = aklass->as_instance_type();
6208 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6209 aescrypt_object = _gvn.transform(aescrypt_object);
6210 // we need to get the start of the aescrypt_object's expanded key array
6211 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6212 if (k_start == NULL) return false;
6213 // similarly, get the start address of the r vector
6214 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6215 if (obj_counter == NULL) return false;
6216 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6217
6218 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6219 if (saved_encCounter == NULL) return false;
6220 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6221 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6222
6223 Node* ctrCrypt;
6224 if (Matcher::pass_original_key_for_aes()) {
6225 // no SPARC version for AES/CTR intrinsics now.
6226 return false;
6227 }
6228 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6229 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6230 OptoRuntime::counterMode_aescrypt_Type(),
6231 stubAddr, stubName, TypePtr::BOTTOM,
6232 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6233
6234 // return cipher length (int)
6235 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6236 set_result(retvalue);
6237 return true;
6238 }
6239
6240 //------------------------------get_key_start_from_aescrypt_object-----------------------
6241 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6242 #ifdef PPC64
6243 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6244 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6245 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6246 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6247 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6248 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6249 if (objSessionK == NULL) {
6250 return (Node *) NULL;
6251 }
6252 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6253 #else
6254 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6255 #endif // PPC64
6256 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6257 if (objAESCryptKey == NULL) return (Node *) NULL;
6258
6259 // now have the array, need to get the start address of the K array
6260 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6261 return k_start;
6262 }
6263
6264 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6265 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6266 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6267 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6268 if (objAESCryptKey == NULL) return (Node *) NULL;
6269
6270 // now have the array, need to get the start address of the lastKey array
6271 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6272 return original_k_start;
6273 }
6274
6275 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6276 // Return node representing slow path of predicate check.
6277 // the pseudo code we want to emulate with this predicate is:
6278 // for encryption:
6279 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6280 // for decryption:
6281 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6282 // note cipher==plain is more conservative than the original java code but that's OK
6283 //
6284 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6285 // The receiver was checked for NULL already.
6286 Node* objCBC = argument(0);
6287
6288 // Load embeddedCipher field of CipherBlockChaining object.
6289 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6290
6291 // get AESCrypt klass for instanceOf check
6292 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6293 // will have same classloader as CipherBlockChaining object
6294 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6295 assert(tinst != NULL, "CBCobj is null");
6296 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6297
6298 // we want to do an instanceof comparison against the AESCrypt class
6299 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6300 if (!klass_AESCrypt->is_loaded()) {
6301 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6302 Node* ctrl = control();
6303 set_control(top()); // no regular fast path
6304 return ctrl;
6305 }
6306 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6307
6308 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6309 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6310 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6311
6312 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6313
6314 // for encryption, we are done
6315 if (!decrypting)
6316 return instof_false; // even if it is NULL
6317
6318 // for decryption, we need to add a further check to avoid
6319 // taking the intrinsic path when cipher and plain are the same
6320 // see the original java code for why.
6321 RegionNode* region = new RegionNode(3);
6322 region->init_req(1, instof_false);
6323 Node* src = argument(1);
6324 Node* dest = argument(4);
6325 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6326 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6327 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6328 region->init_req(2, src_dest_conjoint);
6329
6330 record_for_igvn(region);
6331 return _gvn.transform(region);
6332 }
6333
6334 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6335 // Return node representing slow path of predicate check.
6336 // the pseudo code we want to emulate with this predicate is:
6337 // for encryption:
6338 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6339 // for decryption:
6340 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6341 // note cipher==plain is more conservative than the original java code but that's OK
6342 //
6343
6344 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6345 // The receiver was checked for NULL already.
6346 Node* objCTR = argument(0);
6347
6348 // Load embeddedCipher field of CipherBlockChaining object.
6349 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6350
6351 // get AESCrypt klass for instanceOf check
6352 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6353 // will have same classloader as CipherBlockChaining object
6354 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6355 assert(tinst != NULL, "CTRobj is null");
6356 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6357
6358 // we want to do an instanceof comparison against the AESCrypt class
6359 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6360 if (!klass_AESCrypt->is_loaded()) {
6361 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6362 Node* ctrl = control();
6363 set_control(top()); // no regular fast path
6364 return ctrl;
6365 }
6366
6367 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6368 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6369 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6370 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6371 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6372
6373 return instof_false; // even if it is NULL
6374 }
6375
6376 //------------------------------inline_ghash_processBlocks
6377 bool LibraryCallKit::inline_ghash_processBlocks() {
6378 address stubAddr;
6379 const char *stubName;
6380 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6381
6382 stubAddr = StubRoutines::ghash_processBlocks();
6383 stubName = "ghash_processBlocks";
6384
6385 Node* data = argument(0);
6386 Node* offset = argument(1);
6387 Node* len = argument(2);
6388 Node* state = argument(3);
6389 Node* subkeyH = argument(4);
6390
6391 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6392 assert(state_start, "state is NULL");
6393 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6394 assert(subkeyH_start, "subkeyH is NULL");
6395 Node* data_start = array_element_address(data, offset, T_BYTE);
6396 assert(data_start, "data is NULL");
6397
6398 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6399 OptoRuntime::ghash_processBlocks_Type(),
6400 stubAddr, stubName, TypePtr::BOTTOM,
6401 state_start, subkeyH_start, data_start, len);
6402 return true;
6403 }
6404
6405 //------------------------------inline_sha_implCompress-----------------------
6406 //
6407 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6408 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6409 //
6410 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6411 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6412 //
6413 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6414 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6415 //
6416 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6417 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6418
6419 Node* sha_obj = argument(0);
6420 Node* src = argument(1); // type oop
6421 Node* ofs = argument(2); // type int
6422
6423 const Type* src_type = src->Value(&_gvn);
6424 const TypeAryPtr* top_src = src_type->isa_aryptr();
6425 if (top_src == NULL || top_src->klass() == NULL) {
6426 // failed array check
6427 return false;
6428 }
6429 // Figure out the size and type of the elements we will be copying.
6430 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6431 if (src_elem != T_BYTE) {
6432 return false;
6433 }
6434 // 'src_start' points to src array + offset
6435 Node* src_start = array_element_address(src, ofs, src_elem);
6436 Node* state = NULL;
6437 address stubAddr;
6438 const char *stubName;
6439
6440 switch(id) {
6441 case vmIntrinsics::_sha_implCompress:
6442 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6443 state = get_state_from_sha_object(sha_obj);
6444 stubAddr = StubRoutines::sha1_implCompress();
6445 stubName = "sha1_implCompress";
6446 break;
6447 case vmIntrinsics::_sha2_implCompress:
6448 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6449 state = get_state_from_sha_object(sha_obj);
6450 stubAddr = StubRoutines::sha256_implCompress();
6451 stubName = "sha256_implCompress";
6452 break;
6453 case vmIntrinsics::_sha5_implCompress:
6454 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6455 state = get_state_from_sha5_object(sha_obj);
6456 stubAddr = StubRoutines::sha512_implCompress();
6457 stubName = "sha512_implCompress";
6458 break;
6459 default:
6460 fatal_unexpected_iid(id);
6461 return false;
6462 }
6463 if (state == NULL) return false;
6464
6465 // Call the stub.
6466 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6467 stubAddr, stubName, TypePtr::BOTTOM,
6468 src_start, state);
6469
6470 return true;
6471 }
6472
6473 //------------------------------inline_digestBase_implCompressMB-----------------------
6474 //
6475 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6476 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6477 //
6478 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6479 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6480 "need SHA1/SHA256/SHA512 instruction support");
6481 assert((uint)predicate < 3, "sanity");
6482 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6483
6484 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6485 Node* src = argument(1); // byte[] array
6486 Node* ofs = argument(2); // type int
6487 Node* limit = argument(3); // type int
6488
6489 const Type* src_type = src->Value(&_gvn);
6490 const TypeAryPtr* top_src = src_type->isa_aryptr();
6491 if (top_src == NULL || top_src->klass() == NULL) {
6492 // failed array check
6493 return false;
6494 }
6495 // Figure out the size and type of the elements we will be copying.
6496 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6497 if (src_elem != T_BYTE) {
6498 return false;
6499 }
6500 // 'src_start' points to src array + offset
6501 Node* src_start = array_element_address(src, ofs, src_elem);
6502
6503 const char* klass_SHA_name = NULL;
6504 const char* stub_name = NULL;
6505 address stub_addr = NULL;
6506 bool long_state = false;
6507
6508 switch (predicate) {
6509 case 0:
6510 if (UseSHA1Intrinsics) {
6511 klass_SHA_name = "sun/security/provider/SHA";
6512 stub_name = "sha1_implCompressMB";
6513 stub_addr = StubRoutines::sha1_implCompressMB();
6514 }
6515 break;
6516 case 1:
6517 if (UseSHA256Intrinsics) {
6518 klass_SHA_name = "sun/security/provider/SHA2";
6519 stub_name = "sha256_implCompressMB";
6520 stub_addr = StubRoutines::sha256_implCompressMB();
6521 }
6522 break;
6523 case 2:
6524 if (UseSHA512Intrinsics) {
6525 klass_SHA_name = "sun/security/provider/SHA5";
6526 stub_name = "sha512_implCompressMB";
6527 stub_addr = StubRoutines::sha512_implCompressMB();
6528 long_state = true;
6529 }
6530 break;
6531 default:
6532 fatal("unknown SHA intrinsic predicate: %d", predicate);
6533 }
6534 if (klass_SHA_name != NULL) {
6535 // get DigestBase klass to lookup for SHA klass
6536 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6537 assert(tinst != NULL, "digestBase_obj is not instance???");
6538 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6539
6540 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6541 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6542 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6543 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6544 }
6545 return false;
6546 }
6547 //------------------------------inline_sha_implCompressMB-----------------------
6548 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6549 bool long_state, address stubAddr, const char *stubName,
6550 Node* src_start, Node* ofs, Node* limit) {
6551 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6552 const TypeOopPtr* xtype = aklass->as_instance_type();
6553 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6554 sha_obj = _gvn.transform(sha_obj);
6555
6556 Node* state;
6557 if (long_state) {
6558 state = get_state_from_sha5_object(sha_obj);
6559 } else {
6560 state = get_state_from_sha_object(sha_obj);
6561 }
6562 if (state == NULL) return false;
6563
6564 // Call the stub.
6565 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6566 OptoRuntime::digestBase_implCompressMB_Type(),
6567 stubAddr, stubName, TypePtr::BOTTOM,
6568 src_start, state, ofs, limit);
6569 // return ofs (int)
6570 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6571 set_result(result);
6572
6573 return true;
6574 }
6575
6576 //------------------------------get_state_from_sha_object-----------------------
6577 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6578 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6579 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6580 if (sha_state == NULL) return (Node *) NULL;
6581
6582 // now have the array, need to get the start address of the state array
6583 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6584 return state;
6585 }
6586
6587 //------------------------------get_state_from_sha5_object-----------------------
6588 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6589 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6590 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6591 if (sha_state == NULL) return (Node *) NULL;
6592
6593 // now have the array, need to get the start address of the state array
6594 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6595 return state;
6596 }
6597
6598 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6599 // Return node representing slow path of predicate check.
6600 // the pseudo code we want to emulate with this predicate is:
6601 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6602 //
6603 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6604 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6605 "need SHA1/SHA256/SHA512 instruction support");
6606 assert((uint)predicate < 3, "sanity");
6607
6608 // The receiver was checked for NULL already.
6609 Node* digestBaseObj = argument(0);
6610
6611 // get DigestBase klass for instanceOf check
6612 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6613 assert(tinst != NULL, "digestBaseObj is null");
6614 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6615
6616 const char* klass_SHA_name = NULL;
6617 switch (predicate) {
6618 case 0:
6619 if (UseSHA1Intrinsics) {
6620 // we want to do an instanceof comparison against the SHA class
6621 klass_SHA_name = "sun/security/provider/SHA";
6622 }
6623 break;
6624 case 1:
6625 if (UseSHA256Intrinsics) {
6626 // we want to do an instanceof comparison against the SHA2 class
6627 klass_SHA_name = "sun/security/provider/SHA2";
6628 }
6629 break;
6630 case 2:
6631 if (UseSHA512Intrinsics) {
6632 // we want to do an instanceof comparison against the SHA5 class
6633 klass_SHA_name = "sun/security/provider/SHA5";
6634 }
6635 break;
6636 default:
6637 fatal("unknown SHA intrinsic predicate: %d", predicate);
6638 }
6639
6640 ciKlass* klass_SHA = NULL;
6641 if (klass_SHA_name != NULL) {
6642 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6643 }
6644 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6645 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6646 Node* ctrl = control();
6647 set_control(top()); // no intrinsic path
6648 return ctrl;
6649 }
6650 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6651
6652 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6653 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6654 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6655 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6656
6657 return instof_false; // even if it is NULL
6658 }
6659
6660 bool LibraryCallKit::inline_profileBoolean() {
6661 Node* counts = argument(1);
6662 const TypeAryPtr* ary = NULL;
6663 ciArray* aobj = NULL;
6664 if (counts->is_Con()
6665 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6666 && (aobj = ary->const_oop()->as_array()) != NULL
6667 && (aobj->length() == 2)) {
6668 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6669 jint false_cnt = aobj->element_value(0).as_int();
6670 jint true_cnt = aobj->element_value(1).as_int();
6671
6672 if (C->log() != NULL) {
6673 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6674 false_cnt, true_cnt);
6675 }
6676
6677 if (false_cnt + true_cnt == 0) {
6678 // According to profile, never executed.
6679 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6680 Deoptimization::Action_reinterpret);
6681 return true;
6682 }
6683
6684 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6685 // is a number of each value occurrences.
6686 Node* result = argument(0);
6687 if (false_cnt == 0 || true_cnt == 0) {
6688 // According to profile, one value has been never seen.
6689 int expected_val = (false_cnt == 0) ? 1 : 0;
6690
6691 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6692 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6693
6694 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6695 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6696 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6697
6698 { // Slow path: uncommon trap for never seen value and then reexecute
6699 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6700 // the value has been seen at least once.
6701 PreserveJVMState pjvms(this);
6702 PreserveReexecuteState preexecs(this);
6703 jvms()->set_should_reexecute(true);
6704
6705 set_control(slow_path);
6706 set_i_o(i_o());
6707
6708 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6709 Deoptimization::Action_reinterpret);
6710 }
6711 // The guard for never seen value enables sharpening of the result and
6712 // returning a constant. It allows to eliminate branches on the same value
6713 // later on.
6714 set_control(fast_path);
6715 result = intcon(expected_val);
6716 }
6717 // Stop profiling.
6718 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6719 // By replacing method body with profile data (represented as ProfileBooleanNode
6720 // on IR level) we effectively disable profiling.
6721 // It enables full speed execution once optimized code is generated.
6722 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6723 C->record_for_igvn(profile);
6724 set_result(profile);
6725 return true;
6726 } else {
6727 // Continue profiling.
6728 // Profile data isn't available at the moment. So, execute method's bytecode version.
6729 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6730 // is compiled and counters aren't available since corresponding MethodHandle
6731 // isn't a compile-time constant.
6732 return false;
6733 }
6734 }
6735
6736 bool LibraryCallKit::inline_isCompileConstant() {
6737 Node* n = argument(0);
6738 set_result(n->is_Con() ? intcon(1) : intcon(0));
6739 return true;
6740 }