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, bool is_native_ptr = false);
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_trig(vmIntrinsics::ID id);
226 bool inline_math(vmIntrinsics::ID id);
227 template <typename OverflowOp>
228 bool inline_math_overflow(Node* arg1, Node* arg2);
229 void inline_math_mathExact(Node* math, Node* test);
230 bool inline_math_addExactI(bool is_increment);
231 bool inline_math_addExactL(bool is_increment);
232 bool inline_math_multiplyExactI();
233 bool inline_math_multiplyExactL();
234 bool inline_math_negateExactI();
235 bool inline_math_negateExactL();
236 bool inline_math_subtractExactI(bool is_decrement);
237 bool inline_math_subtractExactL(bool is_decrement);
238 bool inline_min_max(vmIntrinsics::ID id);
239 bool inline_notify(vmIntrinsics::ID id);
240 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
241 // This returns Type::AnyPtr, RawPtr, or OopPtr.
242 int classify_unsafe_addr(Node* &base, Node* &offset);
243 Node* make_unsafe_address(Node* base, Node* offset);
244 // Helper for inline_unsafe_access.
245 // Generates the guards that check whether the result of
246 // Unsafe.getObject should be recorded in an SATB log buffer.
247 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
248
249 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
250 bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
251 static bool klass_needs_init_guard(Node* kls);
252 bool inline_unsafe_allocate();
253 bool inline_unsafe_newArray(bool uninitialized);
254 bool inline_unsafe_copyMemory();
255 bool inline_native_currentThread();
256
257 bool inline_native_time_funcs(address method, const char* funcName);
258 bool inline_native_isInterrupted();
259 bool inline_native_Class_query(vmIntrinsics::ID id);
260 bool inline_native_subtype_check();
261 bool inline_native_getLength();
262 bool inline_array_copyOf(bool is_copyOfRange);
263 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
264 bool inline_objects_checkIndex();
265 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
266 bool inline_native_clone(bool is_virtual);
267 bool inline_native_Reflection_getCallerClass();
268 // Helper function for inlining native object hash method
269 bool inline_native_hashcode(bool is_virtual, bool is_static);
270 bool inline_native_getClass();
271
272 // Helper functions for inlining arraycopy
273 bool inline_arraycopy();
274 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
275 RegionNode* slow_region);
276 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
277 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
278
279 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
280 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
281 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
282 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
283 bool inline_unsafe_fence(vmIntrinsics::ID id);
284 bool inline_onspinwait();
285 bool inline_fp_conversions(vmIntrinsics::ID id);
286 bool inline_number_methods(vmIntrinsics::ID id);
287 bool inline_reference_get();
288 bool inline_Class_cast();
289 bool inline_aescrypt_Block(vmIntrinsics::ID id);
290 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
291 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
292 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
293 Node* inline_counterMode_AESCrypt_predicate();
294 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
295 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
296 bool inline_ghash_processBlocks();
297 bool inline_sha_implCompress(vmIntrinsics::ID id);
298 bool inline_digestBase_implCompressMB(int predicate);
299 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
300 bool long_state, address stubAddr, const char *stubName,
301 Node* src_start, Node* ofs, Node* limit);
302 Node* get_state_from_sha_object(Node *sha_object);
303 Node* get_state_from_sha5_object(Node *sha_object);
304 Node* inline_digestBase_implCompressMB_predicate(int predicate);
305 bool inline_encodeISOArray();
306 bool inline_updateCRC32();
307 bool inline_updateBytesCRC32();
308 bool inline_updateByteBufferCRC32();
309 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
310 bool inline_updateBytesCRC32C();
311 bool inline_updateDirectByteBufferCRC32C();
312 bool inline_updateBytesAdler32();
313 bool inline_updateByteBufferAdler32();
314 bool inline_multiplyToLen();
315 bool inline_hasNegatives();
316 bool inline_squareToLen();
317 bool inline_mulAdd();
318 bool inline_montgomeryMultiply();
319 bool inline_montgomerySquare();
320 bool inline_vectorizedMismatch();
321
322 bool inline_profileBoolean();
323 bool inline_isCompileConstant();
324 };
325
326 //---------------------------make_vm_intrinsic----------------------------
327 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
328 vmIntrinsics::ID id = m->intrinsic_id();
329 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
330
331 if (!m->is_loaded()) {
332 // Do not attempt to inline unloaded methods.
333 return NULL;
334 }
335
336 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
337 bool is_available = false;
338
339 {
340 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
341 // the compiler must transition to '_thread_in_vm' state because both
342 // methods access VM-internal data.
343 VM_ENTRY_MARK;
344 methodHandle mh(THREAD, m->get_Method());
345 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
346 !C->directive()->is_intrinsic_disabled(mh) &&
347 !vmIntrinsics::is_disabled_by_flags(mh);
348
349 }
350
351 if (is_available) {
352 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
353 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
354 return new LibraryIntrinsic(m, is_virtual,
355 vmIntrinsics::predicates_needed(id),
356 vmIntrinsics::does_virtual_dispatch(id),
357 (vmIntrinsics::ID) id);
358 } else {
359 return NULL;
360 }
361 }
362
363 //----------------------register_library_intrinsics-----------------------
364 // Initialize this file's data structures, for each Compile instance.
365 void Compile::register_library_intrinsics() {
366 // Nothing to do here.
367 }
368
369 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
370 LibraryCallKit kit(jvms, this);
371 Compile* C = kit.C;
372 int nodes = C->unique();
373 #ifndef PRODUCT
374 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
375 char buf[1000];
376 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
377 tty->print_cr("Intrinsic %s", str);
378 }
379 #endif
380 ciMethod* callee = kit.callee();
381 const int bci = kit.bci();
382
383 // Try to inline the intrinsic.
384 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
385 kit.try_to_inline(_last_predicate)) {
386 if (C->print_intrinsics() || C->print_inlining()) {
387 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
388 }
389 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
390 if (C->log()) {
391 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
392 vmIntrinsics::name_at(intrinsic_id()),
393 (is_virtual() ? " virtual='1'" : ""),
394 C->unique() - nodes);
395 }
396 // Push the result from the inlined method onto the stack.
397 kit.push_result();
398 C->print_inlining_update(this);
399 return kit.transfer_exceptions_into_jvms();
400 }
401
402 // The intrinsic bailed out
403 if (C->print_intrinsics() || C->print_inlining()) {
404 if (jvms->has_method()) {
405 // Not a root compile.
406 const char* msg;
407 if (callee->intrinsic_candidate()) {
408 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
409 } else {
410 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
411 : "failed to inline (intrinsic), method not annotated";
412 }
413 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
414 } else {
415 // Root compile
416 tty->print("Did not generate intrinsic %s%s at bci:%d in",
417 vmIntrinsics::name_at(intrinsic_id()),
418 (is_virtual() ? " (virtual)" : ""), bci);
419 }
420 }
421 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
422 C->print_inlining_update(this);
423 return NULL;
424 }
425
426 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
427 LibraryCallKit kit(jvms, this);
428 Compile* C = kit.C;
429 int nodes = C->unique();
430 _last_predicate = predicate;
431 #ifndef PRODUCT
432 assert(is_predicated() && predicate < predicates_count(), "sanity");
433 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
434 char buf[1000];
435 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
436 tty->print_cr("Predicate for intrinsic %s", str);
437 }
438 #endif
439 ciMethod* callee = kit.callee();
440 const int bci = kit.bci();
441
442 Node* slow_ctl = kit.try_to_predicate(predicate);
443 if (!kit.failing()) {
444 if (C->print_intrinsics() || C->print_inlining()) {
445 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
446 }
447 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
448 if (C->log()) {
449 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
450 vmIntrinsics::name_at(intrinsic_id()),
451 (is_virtual() ? " virtual='1'" : ""),
452 C->unique() - nodes);
453 }
454 return slow_ctl; // Could be NULL if the check folds.
455 }
456
457 // The intrinsic bailed out
458 if (C->print_intrinsics() || C->print_inlining()) {
459 if (jvms->has_method()) {
460 // Not a root compile.
461 const char* msg = "failed to generate predicate for intrinsic";
462 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
463 } else {
464 // Root compile
465 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
466 vmIntrinsics::name_at(intrinsic_id()),
467 (is_virtual() ? " (virtual)" : ""), bci);
468 }
469 }
470 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
471 return NULL;
472 }
473
474 bool LibraryCallKit::try_to_inline(int predicate) {
475 // Handle symbolic names for otherwise undistinguished boolean switches:
476 const bool is_store = true;
477 const bool is_compress = true;
478 const bool is_native_ptr = true;
479 const bool is_static = true;
480 const bool is_volatile = true;
481
482 if (!jvms()->has_method()) {
483 // Root JVMState has a null method.
484 assert(map()->memory()->Opcode() == Op_Parm, "");
485 // Insert the memory aliasing node
486 set_all_memory(reset_memory());
487 }
488 assert(merged_memory(), "");
489
490
491 switch (intrinsic_id()) {
492 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
493 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
494 case vmIntrinsics::_getClass: return inline_native_getClass();
495
496 case vmIntrinsics::_dsin:
497 case vmIntrinsics::_dcos:
498 case vmIntrinsics::_dtan:
499 case vmIntrinsics::_dabs:
500 case vmIntrinsics::_datan2:
501 case vmIntrinsics::_dsqrt:
502 case vmIntrinsics::_dexp:
503 case vmIntrinsics::_dlog:
504 case vmIntrinsics::_dlog10:
505 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
506
507 case vmIntrinsics::_min:
508 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
509
510 case vmIntrinsics::_notify:
511 case vmIntrinsics::_notifyAll:
512 if (InlineNotify) {
513 return inline_notify(intrinsic_id());
514 }
515 return false;
516
517 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
518 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
519 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
520 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
521 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
522 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
523 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
524 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
525 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
526 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
527 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
528 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
529
530 case vmIntrinsics::_arraycopy: return inline_arraycopy();
531
532 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
533 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
534 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
535 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
536
537 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
538 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
539 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
540 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
541 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
542 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
543 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
544
545 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
546 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
547
548 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
549 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
550 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
551 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
552
553 case vmIntrinsics::_compressStringC:
554 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
555 case vmIntrinsics::_inflateStringC:
556 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
557
558 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Relaxed, false);
559 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Relaxed, false);
560 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Relaxed, false);
561 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, false);
562 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, false);
563 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, false);
564 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, false);
565 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
566 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
567
568 case vmIntrinsics::_putObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Relaxed, false);
569 case vmIntrinsics::_putBoolean: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Relaxed, false);
570 case vmIntrinsics::_putByte: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Relaxed, false);
571 case vmIntrinsics::_putShort: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, false);
572 case vmIntrinsics::_putChar: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, false);
573 case vmIntrinsics::_putInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, false);
574 case vmIntrinsics::_putLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, false);
575 case vmIntrinsics::_putFloat: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Relaxed, false);
576 case vmIntrinsics::_putDouble: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
577
578 case vmIntrinsics::_getByte_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE, Relaxed, false);
579 case vmIntrinsics::_getShort_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT, Relaxed, false);
580 case vmIntrinsics::_getChar_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR, Relaxed, false);
581 case vmIntrinsics::_getInt_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_INT, Relaxed, false);
582 case vmIntrinsics::_getLong_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_LONG, Relaxed, false);
583 case vmIntrinsics::_getFloat_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
584 case vmIntrinsics::_getDouble_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
585 case vmIntrinsics::_getAddress_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, Relaxed, false);
586
587 case vmIntrinsics::_putByte_raw: return inline_unsafe_access( is_native_ptr, is_store, T_BYTE, Relaxed, false);
588 case vmIntrinsics::_putShort_raw: return inline_unsafe_access( is_native_ptr, is_store, T_SHORT, Relaxed, false);
589 case vmIntrinsics::_putChar_raw: return inline_unsafe_access( is_native_ptr, is_store, T_CHAR, Relaxed, false);
590 case vmIntrinsics::_putInt_raw: return inline_unsafe_access( is_native_ptr, is_store, T_INT, Relaxed, false);
591 case vmIntrinsics::_putLong_raw: return inline_unsafe_access( is_native_ptr, is_store, T_LONG, Relaxed, false);
592 case vmIntrinsics::_putFloat_raw: return inline_unsafe_access( is_native_ptr, is_store, T_FLOAT, Relaxed, false);
593 case vmIntrinsics::_putDouble_raw: return inline_unsafe_access( is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
594 case vmIntrinsics::_putAddress_raw: return inline_unsafe_access( is_native_ptr, is_store, T_ADDRESS, Relaxed, false);
595
596 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Volatile, false);
597 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Volatile, false);
598 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Volatile, false);
599 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Volatile, false);
600 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Volatile, false);
601 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Volatile, false);
602 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Volatile, false);
603 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Volatile, false);
604 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Volatile, false);
605
606 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Volatile, false);
607 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Volatile, false);
608 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Volatile, false);
609 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Volatile, false);
610 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Volatile, false);
611 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Volatile, false);
612 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Volatile, false);
613 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Volatile, false);
614 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Volatile, false);
615
616 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, true);
617 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, true);
618 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, true);
619 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, true);
620
621 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, true);
622 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, true);
623 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, true);
624 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, true);
625
626 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Acquire, false);
627 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Acquire, false);
628 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Acquire, false);
629 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Acquire, false);
630 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Acquire, false);
631 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Acquire, false);
632 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Acquire, false);
633 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Acquire, false);
634 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Acquire, false);
635
636 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
637 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Release, false);
638 case vmIntrinsics::_putByteRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Release, false);
639 case vmIntrinsics::_putShortRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Release, false);
640 case vmIntrinsics::_putCharRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Release, false);
641 case vmIntrinsics::_putIntRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
642 case vmIntrinsics::_putLongRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
643 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Release, false);
644 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Release, false);
645
646 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Opaque, false);
647 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Opaque, false);
648 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Opaque, false);
649 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Opaque, false);
650 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Opaque, false);
651 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Opaque, false);
652 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Opaque, false);
653 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Opaque, false);
654 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Opaque, false);
655
656 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Opaque, false);
657 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Opaque, false);
658 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Opaque, false);
659 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Opaque, false);
660 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Opaque, false);
661 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Opaque, false);
662 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Opaque, false);
663 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Opaque, false);
664 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Opaque, false);
665
666 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
667 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
668 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
669
670 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
671 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
672 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
673 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
674 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
675 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
676 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
677 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
678 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
679
680 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
681 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
682 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
683 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
684 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
685 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
686 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
687 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
688 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
689
690 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
691 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
692 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
693 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
694 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
695
696 case vmIntrinsics::_loadFence:
697 case vmIntrinsics::_storeFence:
698 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
699
700 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
701
702 case vmIntrinsics::_currentThread: return inline_native_currentThread();
703 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
704
705 #ifdef TRACE_HAVE_INTRINSICS
706 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
707 #endif
708 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
709 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
710 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
711 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
712 case vmIntrinsics::_getLength: return inline_native_getLength();
713 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
714 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
715 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
716 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
717 case vmIntrinsics::_Objects_checkIndex: return inline_objects_checkIndex();
718 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
719
720 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
721 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
722
723 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
724
725 case vmIntrinsics::_isInstance:
726 case vmIntrinsics::_getModifiers:
727 case vmIntrinsics::_isInterface:
728 case vmIntrinsics::_isArray:
729 case vmIntrinsics::_isPrimitive:
730 case vmIntrinsics::_getSuperclass:
731 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
732
733 case vmIntrinsics::_floatToRawIntBits:
734 case vmIntrinsics::_floatToIntBits:
735 case vmIntrinsics::_intBitsToFloat:
736 case vmIntrinsics::_doubleToRawLongBits:
737 case vmIntrinsics::_doubleToLongBits:
738 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
739
740 case vmIntrinsics::_numberOfLeadingZeros_i:
741 case vmIntrinsics::_numberOfLeadingZeros_l:
742 case vmIntrinsics::_numberOfTrailingZeros_i:
743 case vmIntrinsics::_numberOfTrailingZeros_l:
744 case vmIntrinsics::_bitCount_i:
745 case vmIntrinsics::_bitCount_l:
746 case vmIntrinsics::_reverseBytes_i:
747 case vmIntrinsics::_reverseBytes_l:
748 case vmIntrinsics::_reverseBytes_s:
749 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
750
751 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
752
753 case vmIntrinsics::_Reference_get: return inline_reference_get();
754
755 case vmIntrinsics::_Class_cast: return inline_Class_cast();
756
757 case vmIntrinsics::_aescrypt_encryptBlock:
758 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
759
760 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
761 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
762 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
763
764 case vmIntrinsics::_counterMode_AESCrypt:
765 return inline_counterMode_AESCrypt(intrinsic_id());
766
767 case vmIntrinsics::_sha_implCompress:
768 case vmIntrinsics::_sha2_implCompress:
769 case vmIntrinsics::_sha5_implCompress:
770 return inline_sha_implCompress(intrinsic_id());
771
772 case vmIntrinsics::_digestBase_implCompressMB:
773 return inline_digestBase_implCompressMB(predicate);
774
775 case vmIntrinsics::_multiplyToLen:
776 return inline_multiplyToLen();
777
778 case vmIntrinsics::_squareToLen:
779 return inline_squareToLen();
780
781 case vmIntrinsics::_mulAdd:
782 return inline_mulAdd();
783
784 case vmIntrinsics::_montgomeryMultiply:
785 return inline_montgomeryMultiply();
786 case vmIntrinsics::_montgomerySquare:
787 return inline_montgomerySquare();
788
789 case vmIntrinsics::_vectorizedMismatch:
790 return inline_vectorizedMismatch();
791
792 case vmIntrinsics::_ghash_processBlocks:
793 return inline_ghash_processBlocks();
794
795 case vmIntrinsics::_encodeISOArray:
796 case vmIntrinsics::_encodeByteISOArray:
797 return inline_encodeISOArray();
798
799 case vmIntrinsics::_updateCRC32:
800 return inline_updateCRC32();
801 case vmIntrinsics::_updateBytesCRC32:
802 return inline_updateBytesCRC32();
803 case vmIntrinsics::_updateByteBufferCRC32:
804 return inline_updateByteBufferCRC32();
805
806 case vmIntrinsics::_updateBytesCRC32C:
807 return inline_updateBytesCRC32C();
808 case vmIntrinsics::_updateDirectByteBufferCRC32C:
809 return inline_updateDirectByteBufferCRC32C();
810
811 case vmIntrinsics::_updateBytesAdler32:
812 return inline_updateBytesAdler32();
813 case vmIntrinsics::_updateByteBufferAdler32:
814 return inline_updateByteBufferAdler32();
815
816 case vmIntrinsics::_profileBoolean:
817 return inline_profileBoolean();
818 case vmIntrinsics::_isCompileConstant:
819 return inline_isCompileConstant();
820
821 case vmIntrinsics::_hasNegatives:
822 return inline_hasNegatives();
823
824 default:
825 // If you get here, it may be that someone has added a new intrinsic
826 // to the list in vmSymbols.hpp without implementing it here.
827 #ifndef PRODUCT
828 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
829 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
830 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
831 }
832 #endif
833 return false;
834 }
835 }
836
837 Node* LibraryCallKit::try_to_predicate(int predicate) {
838 if (!jvms()->has_method()) {
839 // Root JVMState has a null method.
840 assert(map()->memory()->Opcode() == Op_Parm, "");
841 // Insert the memory aliasing node
842 set_all_memory(reset_memory());
843 }
844 assert(merged_memory(), "");
845
846 switch (intrinsic_id()) {
847 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
848 return inline_cipherBlockChaining_AESCrypt_predicate(false);
849 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
850 return inline_cipherBlockChaining_AESCrypt_predicate(true);
851 case vmIntrinsics::_counterMode_AESCrypt:
852 return inline_counterMode_AESCrypt_predicate();
853 case vmIntrinsics::_digestBase_implCompressMB:
854 return inline_digestBase_implCompressMB_predicate(predicate);
855
856 default:
857 // If you get here, it may be that someone has added a new intrinsic
858 // to the list in vmSymbols.hpp without implementing it here.
859 #ifndef PRODUCT
860 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
861 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
862 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
863 }
864 #endif
865 Node* slow_ctl = control();
866 set_control(top()); // No fast path instrinsic
867 return slow_ctl;
868 }
869 }
870
871 //------------------------------set_result-------------------------------
872 // Helper function for finishing intrinsics.
873 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
874 record_for_igvn(region);
875 set_control(_gvn.transform(region));
876 set_result( _gvn.transform(value));
877 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
878 }
879
880 //------------------------------generate_guard---------------------------
881 // Helper function for generating guarded fast-slow graph structures.
882 // The given 'test', if true, guards a slow path. If the test fails
883 // then a fast path can be taken. (We generally hope it fails.)
884 // In all cases, GraphKit::control() is updated to the fast path.
885 // The returned value represents the control for the slow path.
886 // The return value is never 'top'; it is either a valid control
887 // or NULL if it is obvious that the slow path can never be taken.
888 // Also, if region and the slow control are not NULL, the slow edge
889 // is appended to the region.
890 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
891 if (stopped()) {
892 // Already short circuited.
893 return NULL;
894 }
895
896 // Build an if node and its projections.
897 // If test is true we take the slow path, which we assume is uncommon.
898 if (_gvn.type(test) == TypeInt::ZERO) {
899 // The slow branch is never taken. No need to build this guard.
900 return NULL;
901 }
902
903 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
904
905 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
906 if (if_slow == top()) {
907 // The slow branch is never taken. No need to build this guard.
908 return NULL;
909 }
910
911 if (region != NULL)
912 region->add_req(if_slow);
913
914 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
915 set_control(if_fast);
916
917 return if_slow;
918 }
919
920 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
921 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
922 }
923 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
924 return generate_guard(test, region, PROB_FAIR);
925 }
926
927 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
928 Node* *pos_index) {
929 if (stopped())
930 return NULL; // already stopped
931 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
932 return NULL; // index is already adequately typed
933 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
934 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
935 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
936 if (is_neg != NULL && pos_index != NULL) {
937 // Emulate effect of Parse::adjust_map_after_if.
938 Node* ccast = new CastIINode(index, TypeInt::POS);
939 ccast->set_req(0, control());
940 (*pos_index) = _gvn.transform(ccast);
941 }
942 return is_neg;
943 }
944
945 // Make sure that 'position' is a valid limit index, in [0..length].
946 // There are two equivalent plans for checking this:
947 // A. (offset + copyLength) unsigned<= arrayLength
948 // B. offset <= (arrayLength - copyLength)
949 // We require that all of the values above, except for the sum and
950 // difference, are already known to be non-negative.
951 // Plan A is robust in the face of overflow, if offset and copyLength
952 // are both hugely positive.
953 //
954 // Plan B is less direct and intuitive, but it does not overflow at
955 // all, since the difference of two non-negatives is always
956 // representable. Whenever Java methods must perform the equivalent
957 // check they generally use Plan B instead of Plan A.
958 // For the moment we use Plan A.
959 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
960 Node* subseq_length,
961 Node* array_length,
962 RegionNode* region) {
963 if (stopped())
964 return NULL; // already stopped
965 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
966 if (zero_offset && subseq_length->eqv_uncast(array_length))
967 return NULL; // common case of whole-array copy
968 Node* last = subseq_length;
969 if (!zero_offset) // last += offset
970 last = _gvn.transform(new AddINode(last, offset));
971 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
972 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
973 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
974 return is_over;
975 }
976
977 // Emit range checks for the given String.value byte array
978 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
979 if (stopped()) {
980 return; // already stopped
981 }
982 RegionNode* bailout = new RegionNode(1);
983 record_for_igvn(bailout);
984 if (char_count) {
985 // Convert char count to byte count
986 count = _gvn.transform(new LShiftINode(count, intcon(1)));
987 }
988
989 // Offset and count must not be negative
990 generate_negative_guard(offset, bailout);
991 generate_negative_guard(count, bailout);
992 // Offset + count must not exceed length of array
993 generate_limit_guard(offset, count, load_array_length(array), bailout);
994
995 if (bailout->req() > 1) {
996 PreserveJVMState pjvms(this);
997 set_control(_gvn.transform(bailout));
998 uncommon_trap(Deoptimization::Reason_intrinsic,
999 Deoptimization::Action_maybe_recompile);
1000 }
1001 }
1002
1003 //--------------------------generate_current_thread--------------------
1004 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1005 ciKlass* thread_klass = env()->Thread_klass();
1006 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1007 Node* thread = _gvn.transform(new ThreadLocalNode());
1008 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1009 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1010 tls_output = thread;
1011 return threadObj;
1012 }
1013
1014
1015 //------------------------------make_string_method_node------------------------
1016 // Helper method for String intrinsic functions. This version is called with
1017 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1018 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1019 // containing the lengths of str1 and str2.
1020 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1021 Node* result = NULL;
1022 switch (opcode) {
1023 case Op_StrIndexOf:
1024 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1025 str1_start, cnt1, str2_start, cnt2, ae);
1026 break;
1027 case Op_StrComp:
1028 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1029 str1_start, cnt1, str2_start, cnt2, ae);
1030 break;
1031 case Op_StrEquals:
1032 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1033 // Use the constant length if there is one because optimized match rule may exist.
1034 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1035 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1036 break;
1037 default:
1038 ShouldNotReachHere();
1039 return NULL;
1040 }
1041
1042 // All these intrinsics have checks.
1043 C->set_has_split_ifs(true); // Has chance for split-if optimization
1044
1045 return _gvn.transform(result);
1046 }
1047
1048 //------------------------------inline_string_compareTo------------------------
1049 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1050 Node* arg1 = argument(0);
1051 Node* arg2 = argument(1);
1052
1053 // Get start addr and length of first argument
1054 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1055 Node* arg1_cnt = load_array_length(arg1);
1056
1057 // Get start addr and length of second argument
1058 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1059 Node* arg2_cnt = load_array_length(arg2);
1060
1061 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1062 set_result(result);
1063 return true;
1064 }
1065
1066 //------------------------------inline_string_equals------------------------
1067 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1068 Node* arg1 = argument(0);
1069 Node* arg2 = argument(1);
1070
1071 // paths (plus control) merge
1072 RegionNode* region = new RegionNode(3);
1073 Node* phi = new PhiNode(region, TypeInt::BOOL);
1074
1075 if (!stopped()) {
1076 // Get start addr and length of first argument
1077 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1078 Node* arg1_cnt = load_array_length(arg1);
1079
1080 // Get start addr and length of second argument
1081 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1082 Node* arg2_cnt = load_array_length(arg2);
1083
1084 // Check for arg1_cnt != arg2_cnt
1085 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1086 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1087 Node* if_ne = generate_slow_guard(bol, NULL);
1088 if (if_ne != NULL) {
1089 phi->init_req(2, intcon(0));
1090 region->init_req(2, if_ne);
1091 }
1092
1093 // Check for count == 0 is done by assembler code for StrEquals.
1094
1095 if (!stopped()) {
1096 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1097 phi->init_req(1, equals);
1098 region->init_req(1, control());
1099 }
1100 }
1101
1102 // post merge
1103 set_control(_gvn.transform(region));
1104 record_for_igvn(region);
1105
1106 set_result(_gvn.transform(phi));
1107 return true;
1108 }
1109
1110 //------------------------------inline_array_equals----------------------------
1111 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1112 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1113 Node* arg1 = argument(0);
1114 Node* arg2 = argument(1);
1115
1116 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1117 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1118 return true;
1119 }
1120
1121 //------------------------------inline_hasNegatives------------------------------
1122 bool LibraryCallKit::inline_hasNegatives() {
1123 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1124 return false;
1125 }
1126
1127 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1128 // no receiver since it is static method
1129 Node* ba = argument(0);
1130 Node* offset = argument(1);
1131 Node* len = argument(2);
1132
1133 // Range checks
1134 generate_string_range_check(ba, offset, len, false);
1135 if (stopped()) {
1136 return true;
1137 }
1138 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1139 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1140 set_result(_gvn.transform(result));
1141 return true;
1142 }
1143
1144 bool LibraryCallKit::inline_objects_checkIndex() {
1145 Node* index = argument(0);
1146 Node* length = argument(1);
1147 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1148 return false;
1149 }
1150
1151 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1152 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1153
1154 {
1155 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1156 uncommon_trap(Deoptimization::Reason_intrinsic,
1157 Deoptimization::Action_make_not_entrant);
1158 }
1159
1160 if (stopped()) {
1161 return false;
1162 }
1163
1164 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1165 BoolTest::mask btest = BoolTest::lt;
1166 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1167 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1168 _gvn.set_type(rc, rc->Value(&_gvn));
1169 if (!rc_bool->is_Con()) {
1170 record_for_igvn(rc);
1171 }
1172 set_control(_gvn.transform(new IfTrueNode(rc)));
1173 {
1174 PreserveJVMState pjvms(this);
1175 set_control(_gvn.transform(new IfFalseNode(rc)));
1176 uncommon_trap(Deoptimization::Reason_range_check,
1177 Deoptimization::Action_make_not_entrant);
1178 }
1179
1180 if (stopped()) {
1181 return false;
1182 }
1183
1184 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1185 result->set_req(0, control());
1186 result = _gvn.transform(result);
1187 set_result(result);
1188 replace_in_map(index, result);
1189 return true;
1190 }
1191
1192 //------------------------------inline_string_indexOf------------------------
1193 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1194 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1195 return false;
1196 }
1197 Node* src = argument(0);
1198 Node* tgt = argument(1);
1199
1200 // Make the merge point
1201 RegionNode* result_rgn = new RegionNode(4);
1202 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1203
1204 // Get start addr and length of source string
1205 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1206 Node* src_count = load_array_length(src);
1207
1208 // Get start addr and length of substring
1209 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1210 Node* tgt_count = load_array_length(tgt);
1211
1212 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1213 // Divide src size by 2 if String is UTF16 encoded
1214 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1215 }
1216 if (ae == StrIntrinsicNode::UU) {
1217 // Divide substring size by 2 if String is UTF16 encoded
1218 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1219 }
1220
1221 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1222 if (result != NULL) {
1223 result_phi->init_req(3, result);
1224 result_rgn->init_req(3, control());
1225 }
1226 set_control(_gvn.transform(result_rgn));
1227 record_for_igvn(result_rgn);
1228 set_result(_gvn.transform(result_phi));
1229
1230 return true;
1231 }
1232
1233 //-----------------------------inline_string_indexOf-----------------------
1234 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1235 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1236 return false;
1237 }
1238 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1239 return false;
1240 }
1241 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1242 Node* src = argument(0); // byte[]
1243 Node* src_count = argument(1); // char count
1244 Node* tgt = argument(2); // byte[]
1245 Node* tgt_count = argument(3); // char count
1246 Node* from_index = argument(4); // char index
1247
1248 // Multiply byte array index by 2 if String is UTF16 encoded
1249 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1250 src_count = _gvn.transform(new SubINode(src_count, from_index));
1251 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1252 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1253
1254 // Range checks
1255 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1256 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1257 if (stopped()) {
1258 return true;
1259 }
1260
1261 RegionNode* region = new RegionNode(5);
1262 Node* phi = new PhiNode(region, TypeInt::INT);
1263
1264 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1265 if (result != NULL) {
1266 // The result is index relative to from_index if substring was found, -1 otherwise.
1267 // Generate code which will fold into cmove.
1268 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1269 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1270
1271 Node* if_lt = generate_slow_guard(bol, NULL);
1272 if (if_lt != NULL) {
1273 // result == -1
1274 phi->init_req(3, result);
1275 region->init_req(3, if_lt);
1276 }
1277 if (!stopped()) {
1278 result = _gvn.transform(new AddINode(result, from_index));
1279 phi->init_req(4, result);
1280 region->init_req(4, control());
1281 }
1282 }
1283
1284 set_control(_gvn.transform(region));
1285 record_for_igvn(region);
1286 set_result(_gvn.transform(phi));
1287
1288 return true;
1289 }
1290
1291 // Create StrIndexOfNode with fast path checks
1292 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1293 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1294 // Check for substr count > string count
1295 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1296 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1297 Node* if_gt = generate_slow_guard(bol, NULL);
1298 if (if_gt != NULL) {
1299 phi->init_req(1, intcon(-1));
1300 region->init_req(1, if_gt);
1301 }
1302 if (!stopped()) {
1303 // Check for substr count == 0
1304 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1305 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1306 Node* if_zero = generate_slow_guard(bol, NULL);
1307 if (if_zero != NULL) {
1308 phi->init_req(2, intcon(0));
1309 region->init_req(2, if_zero);
1310 }
1311 }
1312 if (!stopped()) {
1313 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1314 }
1315 return NULL;
1316 }
1317
1318 //-----------------------------inline_string_indexOfChar-----------------------
1319 bool LibraryCallKit::inline_string_indexOfChar() {
1320 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1321 return false;
1322 }
1323 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1324 return false;
1325 }
1326 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1327 Node* src = argument(0); // byte[]
1328 Node* tgt = argument(1); // tgt is int ch
1329 Node* from_index = argument(2);
1330 Node* max = argument(3);
1331
1332 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1333 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1334 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1335
1336 // Range checks
1337 generate_string_range_check(src, src_offset, src_count, true);
1338 if (stopped()) {
1339 return true;
1340 }
1341
1342 RegionNode* region = new RegionNode(3);
1343 Node* phi = new PhiNode(region, TypeInt::INT);
1344
1345 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1346 C->set_has_split_ifs(true); // Has chance for split-if optimization
1347 _gvn.transform(result);
1348
1349 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1350 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1351
1352 Node* if_lt = generate_slow_guard(bol, NULL);
1353 if (if_lt != NULL) {
1354 // result == -1
1355 phi->init_req(2, result);
1356 region->init_req(2, if_lt);
1357 }
1358 if (!stopped()) {
1359 result = _gvn.transform(new AddINode(result, from_index));
1360 phi->init_req(1, result);
1361 region->init_req(1, control());
1362 }
1363 set_control(_gvn.transform(region));
1364 record_for_igvn(region);
1365 set_result(_gvn.transform(phi));
1366
1367 return true;
1368 }
1369 //---------------------------inline_string_copy---------------------
1370 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1371 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1372 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1373 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1374 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1375 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1376 bool LibraryCallKit::inline_string_copy(bool compress) {
1377 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1378 return false;
1379 }
1380 int nargs = 5; // 2 oops, 3 ints
1381 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1382
1383 Node* src = argument(0);
1384 Node* src_offset = argument(1);
1385 Node* dst = argument(2);
1386 Node* dst_offset = argument(3);
1387 Node* length = argument(4);
1388
1389 // Check for allocation before we add nodes that would confuse
1390 // tightly_coupled_allocation()
1391 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1392
1393 // Figure out the size and type of the elements we will be copying.
1394 const Type* src_type = src->Value(&_gvn);
1395 const Type* dst_type = dst->Value(&_gvn);
1396 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1397 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1398 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1399 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1400 "Unsupported array types for inline_string_copy");
1401
1402 // Range checks
1403 generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1404 generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1405 if (stopped()) {
1406 return true;
1407 }
1408
1409 // Convert char[] offsets to byte[] offsets
1410 if (compress && src_elem == T_BYTE) {
1411 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1412 } else if (!compress && dst_elem == T_BYTE) {
1413 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1414 }
1415
1416 Node* src_start = array_element_address(src, src_offset, src_elem);
1417 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1418 // 'src_start' points to src array + scaled offset
1419 // 'dst_start' points to dst array + scaled offset
1420 Node* count = NULL;
1421 if (compress) {
1422 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1423 } else {
1424 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1425 }
1426
1427 if (alloc != NULL) {
1428 if (alloc->maybe_set_complete(&_gvn)) {
1429 // "You break it, you buy it."
1430 InitializeNode* init = alloc->initialization();
1431 assert(init->is_complete(), "we just did this");
1432 init->set_complete_with_arraycopy();
1433 assert(dst->is_CheckCastPP(), "sanity");
1434 assert(dst->in(0)->in(0) == init, "dest pinned");
1435 }
1436 // Do not let stores that initialize this object be reordered with
1437 // a subsequent store that would make this object accessible by
1438 // other threads.
1439 // Record what AllocateNode this StoreStore protects so that
1440 // escape analysis can go from the MemBarStoreStoreNode to the
1441 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1442 // based on the escape status of the AllocateNode.
1443 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1444 }
1445 if (compress) {
1446 set_result(_gvn.transform(count));
1447 }
1448 return true;
1449 }
1450
1451 #ifdef _LP64
1452 #define XTOP ,top() /*additional argument*/
1453 #else //_LP64
1454 #define XTOP /*no additional argument*/
1455 #endif //_LP64
1456
1457 //------------------------inline_string_toBytesU--------------------------
1458 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1459 bool LibraryCallKit::inline_string_toBytesU() {
1460 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1461 return false;
1462 }
1463 // Get the arguments.
1464 Node* value = argument(0);
1465 Node* offset = argument(1);
1466 Node* length = argument(2);
1467
1468 Node* newcopy = NULL;
1469
1470 // Set the original stack and the reexecute bit for the interpreter to reexecute
1471 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1472 { PreserveReexecuteState preexecs(this);
1473 jvms()->set_should_reexecute(true);
1474
1475 // Check if a null path was taken unconditionally.
1476 value = null_check(value);
1477
1478 RegionNode* bailout = new RegionNode(1);
1479 record_for_igvn(bailout);
1480
1481 // Range checks
1482 generate_negative_guard(offset, bailout);
1483 generate_negative_guard(length, bailout);
1484 generate_limit_guard(offset, length, load_array_length(value), bailout);
1485 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1486 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1487
1488 if (bailout->req() > 1) {
1489 PreserveJVMState pjvms(this);
1490 set_control(_gvn.transform(bailout));
1491 uncommon_trap(Deoptimization::Reason_intrinsic,
1492 Deoptimization::Action_maybe_recompile);
1493 }
1494 if (stopped()) {
1495 return true;
1496 }
1497
1498 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1499 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1500 newcopy = new_array(klass_node, size, 0); // no arguments to push
1501 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1502
1503 // Calculate starting addresses.
1504 Node* src_start = array_element_address(value, offset, T_CHAR);
1505 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1506
1507 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1508 const TypeInt* toffset = gvn().type(offset)->is_int();
1509 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1510
1511 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1512 const char* copyfunc_name = "arraycopy";
1513 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1514 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1515 OptoRuntime::fast_arraycopy_Type(),
1516 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1517 src_start, dst_start, ConvI2X(length) XTOP);
1518 // Do not let reads from the cloned object float above the arraycopy.
1519 if (alloc != NULL) {
1520 if (alloc->maybe_set_complete(&_gvn)) {
1521 // "You break it, you buy it."
1522 InitializeNode* init = alloc->initialization();
1523 assert(init->is_complete(), "we just did this");
1524 init->set_complete_with_arraycopy();
1525 assert(newcopy->is_CheckCastPP(), "sanity");
1526 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1527 }
1528 // Do not let stores that initialize this object be reordered with
1529 // a subsequent store that would make this object accessible by
1530 // other threads.
1531 // Record what AllocateNode this StoreStore protects so that
1532 // escape analysis can go from the MemBarStoreStoreNode to the
1533 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1534 // based on the escape status of the AllocateNode.
1535 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1536 } else {
1537 insert_mem_bar(Op_MemBarCPUOrder);
1538 }
1539 } // original reexecute is set back here
1540
1541 C->set_has_split_ifs(true); // Has chance for split-if optimization
1542 if (!stopped()) {
1543 set_result(newcopy);
1544 }
1545 return true;
1546 }
1547
1548 //------------------------inline_string_getCharsU--------------------------
1549 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1550 bool LibraryCallKit::inline_string_getCharsU() {
1551 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1552 return false;
1553 }
1554
1555 // Get the arguments.
1556 Node* src = argument(0);
1557 Node* src_begin = argument(1);
1558 Node* src_end = argument(2); // exclusive offset (i < src_end)
1559 Node* dst = argument(3);
1560 Node* dst_begin = argument(4);
1561
1562 // Check for allocation before we add nodes that would confuse
1563 // tightly_coupled_allocation()
1564 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1565
1566 // Check if a null path was taken unconditionally.
1567 src = null_check(src);
1568 dst = null_check(dst);
1569 if (stopped()) {
1570 return true;
1571 }
1572
1573 // Get length and convert char[] offset to byte[] offset
1574 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1575 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1576
1577 // Range checks
1578 generate_string_range_check(src, src_begin, length, true);
1579 generate_string_range_check(dst, dst_begin, length, false);
1580 if (stopped()) {
1581 return true;
1582 }
1583
1584 if (!stopped()) {
1585 // Calculate starting addresses.
1586 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1587 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1588
1589 // Check if array addresses are aligned to HeapWordSize
1590 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1591 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1592 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1593 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1594
1595 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1596 const char* copyfunc_name = "arraycopy";
1597 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1598 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1599 OptoRuntime::fast_arraycopy_Type(),
1600 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1601 src_start, dst_start, ConvI2X(length) XTOP);
1602 // Do not let reads from the cloned object float above the arraycopy.
1603 if (alloc != NULL) {
1604 if (alloc->maybe_set_complete(&_gvn)) {
1605 // "You break it, you buy it."
1606 InitializeNode* init = alloc->initialization();
1607 assert(init->is_complete(), "we just did this");
1608 init->set_complete_with_arraycopy();
1609 assert(dst->is_CheckCastPP(), "sanity");
1610 assert(dst->in(0)->in(0) == init, "dest pinned");
1611 }
1612 // Do not let stores that initialize this object be reordered with
1613 // a subsequent store that would make this object accessible by
1614 // other threads.
1615 // Record what AllocateNode this StoreStore protects so that
1616 // escape analysis can go from the MemBarStoreStoreNode to the
1617 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1618 // based on the escape status of the AllocateNode.
1619 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1620 } else {
1621 insert_mem_bar(Op_MemBarCPUOrder);
1622 }
1623 }
1624
1625 C->set_has_split_ifs(true); // Has chance for split-if optimization
1626 return true;
1627 }
1628
1629 //----------------------inline_string_char_access----------------------------
1630 // Store/Load char to/from byte[] array.
1631 // static void StringUTF16.putChar(byte[] val, int index, int c)
1632 // static char StringUTF16.getChar(byte[] val, int index)
1633 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1634 Node* value = argument(0);
1635 Node* index = argument(1);
1636 Node* ch = is_store ? argument(2) : NULL;
1637
1638 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1639 // correctly requires matched array shapes.
1640 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1641 "sanity: byte[] and char[] bases agree");
1642 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1643 "sanity: byte[] and char[] scales agree");
1644
1645 // Bail when getChar over constants is requested: constant folding would
1646 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1647 // Java method would constant fold nicely instead.
1648 if (!is_store && value->is_Con() && index->is_Con()) {
1649 return false;
1650 }
1651
1652 Node* adr = array_element_address(value, index, T_CHAR);
1653 if (is_store) {
1654 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1655 false, false, true /* mismatched */);
1656 } else {
1657 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1658 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1659 set_result(ch);
1660 }
1661 return true;
1662 }
1663
1664 //--------------------------round_double_node--------------------------------
1665 // Round a double node if necessary.
1666 Node* LibraryCallKit::round_double_node(Node* n) {
1667 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1668 n = _gvn.transform(new RoundDoubleNode(0, n));
1669 return n;
1670 }
1671
1672 //------------------------------inline_math-----------------------------------
1673 // public static double Math.abs(double)
1674 // public static double Math.sqrt(double)
1675 // public static double Math.log(double)
1676 // public static double Math.log10(double)
1677 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1678 Node* arg = round_double_node(argument(0));
1679 Node* n = NULL;
1680 switch (id) {
1681 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1682 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1683 default: fatal_unexpected_iid(id); break;
1684 }
1685 set_result(_gvn.transform(n));
1686 return true;
1687 }
1688
1689 //------------------------------inline_trig----------------------------------
1690 // Inline sin/cos/tan instructions, if possible. If rounding is required, do
1691 // argument reduction which will turn into a fast/slow diamond.
1692 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1693 Node* arg = round_double_node(argument(0));
1694 Node* n = NULL;
1695
1696 n = _gvn.transform(n);
1697
1698 // Rounding required? Check for argument reduction!
1699 if (Matcher::strict_fp_requires_explicit_rounding) {
1700 static const double pi_4 = 0.7853981633974483;
1701 static const double neg_pi_4 = -0.7853981633974483;
1702 // pi/2 in 80-bit extended precision
1703 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1704 // -pi/2 in 80-bit extended precision
1705 // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00};
1706 // Cutoff value for using this argument reduction technique
1707 //static const double pi_2_minus_epsilon = 1.564660403643354;
1708 //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1709
1710 // Pseudocode for sin:
1711 // if (x <= Math.PI / 4.0) {
1712 // if (x >= -Math.PI / 4.0) return fsin(x);
1713 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1714 // } else {
1715 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0);
1716 // }
1717 // return StrictMath.sin(x);
1718
1719 // Pseudocode for cos:
1720 // if (x <= Math.PI / 4.0) {
1721 // if (x >= -Math.PI / 4.0) return fcos(x);
1722 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0);
1723 // } else {
1724 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1725 // }
1726 // return StrictMath.cos(x);
1727
1728 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1729 // requires a special machine instruction to load it. Instead we'll try
1730 // the 'easy' case. If we really need the extra range +/- PI/2 we'll
1731 // probably do the math inside the SIN encoding.
1732
1733 // Make the merge point
1734 RegionNode* r = new RegionNode(3);
1735 Node* phi = new PhiNode(r, Type::DOUBLE);
1736
1737 // Flatten arg so we need only 1 test
1738 Node *abs = _gvn.transform(new AbsDNode(arg));
1739 // Node for PI/4 constant
1740 Node *pi4 = makecon(TypeD::make(pi_4));
1741 // Check PI/4 : abs(arg)
1742 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1743 // Check: If PI/4 < abs(arg) then go slow
1744 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1745 // Branch either way
1746 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1747 set_control(opt_iff(r,iff));
1748
1749 // Set fast path result
1750 phi->init_req(2, n);
1751
1752 // Slow path - non-blocking leaf call
1753 Node* call = NULL;
1754 switch (id) {
1755 case vmIntrinsics::_dtan:
1756 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1757 CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1758 "Tan", NULL, arg, top());
1759 break;
1760 }
1761 assert(control()->in(0) == call, "");
1762 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1763 r->init_req(1, control());
1764 phi->init_req(1, slow_result);
1765
1766 // Post-merge
1767 set_control(_gvn.transform(r));
1768 record_for_igvn(r);
1769 n = _gvn.transform(phi);
1770
1771 C->set_has_split_ifs(true); // Has chance for split-if optimization
1772 }
1773 set_result(n);
1774 return true;
1775 }
1776
1777 //------------------------------runtime_math-----------------------------
1778 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1779 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1780 "must be (DD)D or (D)D type");
1781
1782 // Inputs
1783 Node* a = round_double_node(argument(0));
1784 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1785
1786 const TypePtr* no_memory_effects = NULL;
1787 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1788 no_memory_effects,
1789 a, top(), b, b ? top() : NULL);
1790 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1791 #ifdef ASSERT
1792 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1793 assert(value_top == top(), "second value must be top");
1794 #endif
1795
1796 set_result(value);
1797 return true;
1798 }
1799
1800 //------------------------------inline_math_native-----------------------------
1801 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1802 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1803 switch (id) {
1804 // These intrinsics are not properly supported on all hardware
1805 case vmIntrinsics::_dsin:
1806 return StubRoutines::dsin() != NULL ?
1807 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1808 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1809 case vmIntrinsics::_dcos:
1810 return StubRoutines::dcos() != NULL ?
1811 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1812 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1813 case vmIntrinsics::_dtan:
1814 return StubRoutines::dtan() != NULL ?
1815 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1816 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1817 case vmIntrinsics::_dlog:
1818 return StubRoutines::dlog() != NULL ?
1819 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1820 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1821 case vmIntrinsics::_dlog10:
1822 return StubRoutines::dlog10() != NULL ?
1823 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1824 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1825
1826 // These intrinsics are supported on all hardware
1827 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1828 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1829
1830 case vmIntrinsics::_dexp:
1831 return StubRoutines::dexp() != NULL ?
1832 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1833 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1834 case vmIntrinsics::_dpow:
1835 return StubRoutines::dpow() != NULL ?
1836 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1837 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1838 #undef FN_PTR
1839
1840 // These intrinsics are not yet correctly implemented
1841 case vmIntrinsics::_datan2:
1842 return false;
1843
1844 default:
1845 fatal_unexpected_iid(id);
1846 return false;
1847 }
1848 }
1849
1850 static bool is_simple_name(Node* n) {
1851 return (n->req() == 1 // constant
1852 || (n->is_Type() && n->as_Type()->type()->singleton())
1853 || n->is_Proj() // parameter or return value
1854 || n->is_Phi() // local of some sort
1855 );
1856 }
1857
1858 //----------------------------inline_notify-----------------------------------*
1859 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1860 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1861 address func;
1862 if (id == vmIntrinsics::_notify) {
1863 func = OptoRuntime::monitor_notify_Java();
1864 } else {
1865 func = OptoRuntime::monitor_notifyAll_Java();
1866 }
1867 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1868 make_slow_call_ex(call, env()->Throwable_klass(), false);
1869 return true;
1870 }
1871
1872
1873 //----------------------------inline_min_max-----------------------------------
1874 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1875 set_result(generate_min_max(id, argument(0), argument(1)));
1876 return true;
1877 }
1878
1879 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1880 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1881 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1882 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1883 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1884
1885 {
1886 PreserveJVMState pjvms(this);
1887 PreserveReexecuteState preexecs(this);
1888 jvms()->set_should_reexecute(true);
1889
1890 set_control(slow_path);
1891 set_i_o(i_o());
1892
1893 uncommon_trap(Deoptimization::Reason_intrinsic,
1894 Deoptimization::Action_none);
1895 }
1896
1897 set_control(fast_path);
1898 set_result(math);
1899 }
1900
1901 template <typename OverflowOp>
1902 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1903 typedef typename OverflowOp::MathOp MathOp;
1904
1905 MathOp* mathOp = new MathOp(arg1, arg2);
1906 Node* operation = _gvn.transform( mathOp );
1907 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1908 inline_math_mathExact(operation, ofcheck);
1909 return true;
1910 }
1911
1912 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1913 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1914 }
1915
1916 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1917 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1918 }
1919
1920 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1921 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1922 }
1923
1924 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1925 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1926 }
1927
1928 bool LibraryCallKit::inline_math_negateExactI() {
1929 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1930 }
1931
1932 bool LibraryCallKit::inline_math_negateExactL() {
1933 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1934 }
1935
1936 bool LibraryCallKit::inline_math_multiplyExactI() {
1937 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1938 }
1939
1940 bool LibraryCallKit::inline_math_multiplyExactL() {
1941 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1942 }
1943
1944 Node*
1945 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1946 // These are the candidate return value:
1947 Node* xvalue = x0;
1948 Node* yvalue = y0;
1949
1950 if (xvalue == yvalue) {
1951 return xvalue;
1952 }
1953
1954 bool want_max = (id == vmIntrinsics::_max);
1955
1956 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1957 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1958 if (txvalue == NULL || tyvalue == NULL) return top();
1959 // This is not really necessary, but it is consistent with a
1960 // hypothetical MaxINode::Value method:
1961 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1962
1963 // %%% This folding logic should (ideally) be in a different place.
1964 // Some should be inside IfNode, and there to be a more reliable
1965 // transformation of ?: style patterns into cmoves. We also want
1966 // more powerful optimizations around cmove and min/max.
1967
1968 // Try to find a dominating comparison of these guys.
1969 // It can simplify the index computation for Arrays.copyOf
1970 // and similar uses of System.arraycopy.
1971 // First, compute the normalized version of CmpI(x, y).
1972 int cmp_op = Op_CmpI;
1973 Node* xkey = xvalue;
1974 Node* ykey = yvalue;
1975 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1976 if (ideal_cmpxy->is_Cmp()) {
1977 // E.g., if we have CmpI(length - offset, count),
1978 // it might idealize to CmpI(length, count + offset)
1979 cmp_op = ideal_cmpxy->Opcode();
1980 xkey = ideal_cmpxy->in(1);
1981 ykey = ideal_cmpxy->in(2);
1982 }
1983
1984 // Start by locating any relevant comparisons.
1985 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1986 Node* cmpxy = NULL;
1987 Node* cmpyx = NULL;
1988 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1989 Node* cmp = start_from->fast_out(k);
1990 if (cmp->outcnt() > 0 && // must have prior uses
1991 cmp->in(0) == NULL && // must be context-independent
1992 cmp->Opcode() == cmp_op) { // right kind of compare
1993 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1994 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1995 }
1996 }
1997
1998 const int NCMPS = 2;
1999 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2000 int cmpn;
2001 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2002 if (cmps[cmpn] != NULL) break; // find a result
2003 }
2004 if (cmpn < NCMPS) {
2005 // Look for a dominating test that tells us the min and max.
2006 int depth = 0; // Limit search depth for speed
2007 Node* dom = control();
2008 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2009 if (++depth >= 100) break;
2010 Node* ifproj = dom;
2011 if (!ifproj->is_Proj()) continue;
2012 Node* iff = ifproj->in(0);
2013 if (!iff->is_If()) continue;
2014 Node* bol = iff->in(1);
2015 if (!bol->is_Bool()) continue;
2016 Node* cmp = bol->in(1);
2017 if (cmp == NULL) continue;
2018 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2019 if (cmps[cmpn] == cmp) break;
2020 if (cmpn == NCMPS) continue;
2021 BoolTest::mask btest = bol->as_Bool()->_test._test;
2022 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2023 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2024 // At this point, we know that 'x btest y' is true.
2025 switch (btest) {
2026 case BoolTest::eq:
2027 // They are proven equal, so we can collapse the min/max.
2028 // Either value is the answer. Choose the simpler.
2029 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2030 return yvalue;
2031 return xvalue;
2032 case BoolTest::lt: // x < y
2033 case BoolTest::le: // x <= y
2034 return (want_max ? yvalue : xvalue);
2035 case BoolTest::gt: // x > y
2036 case BoolTest::ge: // x >= y
2037 return (want_max ? xvalue : yvalue);
2038 }
2039 }
2040 }
2041
2042 // We failed to find a dominating test.
2043 // Let's pick a test that might GVN with prior tests.
2044 Node* best_bol = NULL;
2045 BoolTest::mask best_btest = BoolTest::illegal;
2046 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2047 Node* cmp = cmps[cmpn];
2048 if (cmp == NULL) continue;
2049 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2050 Node* bol = cmp->fast_out(j);
2051 if (!bol->is_Bool()) continue;
2052 BoolTest::mask btest = bol->as_Bool()->_test._test;
2053 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2054 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2055 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2056 best_bol = bol->as_Bool();
2057 best_btest = btest;
2058 }
2059 }
2060 }
2061
2062 Node* answer_if_true = NULL;
2063 Node* answer_if_false = NULL;
2064 switch (best_btest) {
2065 default:
2066 if (cmpxy == NULL)
2067 cmpxy = ideal_cmpxy;
2068 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2069 // and fall through:
2070 case BoolTest::lt: // x < y
2071 case BoolTest::le: // x <= y
2072 answer_if_true = (want_max ? yvalue : xvalue);
2073 answer_if_false = (want_max ? xvalue : yvalue);
2074 break;
2075 case BoolTest::gt: // x > y
2076 case BoolTest::ge: // x >= y
2077 answer_if_true = (want_max ? xvalue : yvalue);
2078 answer_if_false = (want_max ? yvalue : xvalue);
2079 break;
2080 }
2081
2082 jint hi, lo;
2083 if (want_max) {
2084 // We can sharpen the minimum.
2085 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2086 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2087 } else {
2088 // We can sharpen the maximum.
2089 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2090 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2091 }
2092
2093 // Use a flow-free graph structure, to avoid creating excess control edges
2094 // which could hinder other optimizations.
2095 // Since Math.min/max is often used with arraycopy, we want
2096 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2097 Node* cmov = CMoveNode::make(NULL, best_bol,
2098 answer_if_false, answer_if_true,
2099 TypeInt::make(lo, hi, widen));
2100
2101 return _gvn.transform(cmov);
2102
2103 /*
2104 // This is not as desirable as it may seem, since Min and Max
2105 // nodes do not have a full set of optimizations.
2106 // And they would interfere, anyway, with 'if' optimizations
2107 // and with CMoveI canonical forms.
2108 switch (id) {
2109 case vmIntrinsics::_min:
2110 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2111 case vmIntrinsics::_max:
2112 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2113 default:
2114 ShouldNotReachHere();
2115 }
2116 */
2117 }
2118
2119 inline int
2120 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2121 const TypePtr* base_type = TypePtr::NULL_PTR;
2122 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2123 if (base_type == NULL) {
2124 // Unknown type.
2125 return Type::AnyPtr;
2126 } else if (base_type == TypePtr::NULL_PTR) {
2127 // Since this is a NULL+long form, we have to switch to a rawptr.
2128 base = _gvn.transform(new CastX2PNode(offset));
2129 offset = MakeConX(0);
2130 return Type::RawPtr;
2131 } else if (base_type->base() == Type::RawPtr) {
2132 return Type::RawPtr;
2133 } else if (base_type->isa_oopptr()) {
2134 // Base is never null => always a heap address.
2135 if (base_type->ptr() == TypePtr::NotNull) {
2136 return Type::OopPtr;
2137 }
2138 // Offset is small => always a heap address.
2139 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2140 if (offset_type != NULL &&
2141 base_type->offset() == 0 && // (should always be?)
2142 offset_type->_lo >= 0 &&
2143 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2144 return Type::OopPtr;
2145 }
2146 // Otherwise, it might either be oop+off or NULL+addr.
2147 return Type::AnyPtr;
2148 } else {
2149 // No information:
2150 return Type::AnyPtr;
2151 }
2152 }
2153
2154 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2155 int kind = classify_unsafe_addr(base, offset);
2156 if (kind == Type::RawPtr) {
2157 return basic_plus_adr(top(), base, offset);
2158 } else {
2159 return basic_plus_adr(base, offset);
2160 }
2161 }
2162
2163 //--------------------------inline_number_methods-----------------------------
2164 // inline int Integer.numberOfLeadingZeros(int)
2165 // inline int Long.numberOfLeadingZeros(long)
2166 //
2167 // inline int Integer.numberOfTrailingZeros(int)
2168 // inline int Long.numberOfTrailingZeros(long)
2169 //
2170 // inline int Integer.bitCount(int)
2171 // inline int Long.bitCount(long)
2172 //
2173 // inline char Character.reverseBytes(char)
2174 // inline short Short.reverseBytes(short)
2175 // inline int Integer.reverseBytes(int)
2176 // inline long Long.reverseBytes(long)
2177 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2178 Node* arg = argument(0);
2179 Node* n = NULL;
2180 switch (id) {
2181 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2182 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2183 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2184 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2185 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2186 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2187 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2188 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2189 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2190 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2191 default: fatal_unexpected_iid(id); break;
2192 }
2193 set_result(_gvn.transform(n));
2194 return true;
2195 }
2196
2197 //----------------------------inline_unsafe_access----------------------------
2198
2199 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2200
2201 // Helper that guards and inserts a pre-barrier.
2202 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2203 Node* pre_val, bool need_mem_bar) {
2204 // We could be accessing the referent field of a reference object. If so, when G1
2205 // is enabled, we need to log the value in the referent field in an SATB buffer.
2206 // This routine performs some compile time filters and generates suitable
2207 // runtime filters that guard the pre-barrier code.
2208 // Also add memory barrier for non volatile load from the referent field
2209 // to prevent commoning of loads across safepoint.
2210 if (!UseG1GC && !need_mem_bar)
2211 return;
2212
2213 // Some compile time checks.
2214
2215 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2216 const TypeX* otype = offset->find_intptr_t_type();
2217 if (otype != NULL && otype->is_con() &&
2218 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2219 // Constant offset but not the reference_offset so just return
2220 return;
2221 }
2222
2223 // We only need to generate the runtime guards for instances.
2224 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2225 if (btype != NULL) {
2226 if (btype->isa_aryptr()) {
2227 // Array type so nothing to do
2228 return;
2229 }
2230
2231 const TypeInstPtr* itype = btype->isa_instptr();
2232 if (itype != NULL) {
2233 // Can the klass of base_oop be statically determined to be
2234 // _not_ a sub-class of Reference and _not_ Object?
2235 ciKlass* klass = itype->klass();
2236 if ( klass->is_loaded() &&
2237 !klass->is_subtype_of(env()->Reference_klass()) &&
2238 !env()->Object_klass()->is_subtype_of(klass)) {
2239 return;
2240 }
2241 }
2242 }
2243
2244 // The compile time filters did not reject base_oop/offset so
2245 // we need to generate the following runtime filters
2246 //
2247 // if (offset == java_lang_ref_Reference::_reference_offset) {
2248 // if (instance_of(base, java.lang.ref.Reference)) {
2249 // pre_barrier(_, pre_val, ...);
2250 // }
2251 // }
2252
2253 float likely = PROB_LIKELY( 0.999);
2254 float unlikely = PROB_UNLIKELY(0.999);
2255
2256 IdealKit ideal(this);
2257 #define __ ideal.
2258
2259 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2260
2261 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2262 // Update graphKit memory and control from IdealKit.
2263 sync_kit(ideal);
2264
2265 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2266 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2267
2268 // Update IdealKit memory and control from graphKit.
2269 __ sync_kit(this);
2270
2271 Node* one = __ ConI(1);
2272 // is_instof == 0 if base_oop == NULL
2273 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2274
2275 // Update graphKit from IdeakKit.
2276 sync_kit(ideal);
2277
2278 // Use the pre-barrier to record the value in the referent field
2279 pre_barrier(false /* do_load */,
2280 __ ctrl(),
2281 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2282 pre_val /* pre_val */,
2283 T_OBJECT);
2284 if (need_mem_bar) {
2285 // Add memory barrier to prevent commoning reads from this field
2286 // across safepoint since GC can change its value.
2287 insert_mem_bar(Op_MemBarCPUOrder);
2288 }
2289 // Update IdealKit from graphKit.
2290 __ sync_kit(this);
2291
2292 } __ end_if(); // _ref_type != ref_none
2293 } __ end_if(); // offset == referent_offset
2294
2295 // Final sync IdealKit and GraphKit.
2296 final_sync(ideal);
2297 #undef __
2298 }
2299
2300
2301 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2302 // Attempt to infer a sharper value type from the offset and base type.
2303 ciKlass* sharpened_klass = NULL;
2304
2305 // See if it is an instance field, with an object type.
2306 if (alias_type->field() != NULL) {
2307 assert(!is_native_ptr, "native pointer op cannot use a java address");
2308 if (alias_type->field()->type()->is_klass()) {
2309 sharpened_klass = alias_type->field()->type()->as_klass();
2310 }
2311 }
2312
2313 // See if it is a narrow oop array.
2314 if (adr_type->isa_aryptr()) {
2315 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2316 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2317 if (elem_type != NULL) {
2318 sharpened_klass = elem_type->klass();
2319 }
2320 }
2321 }
2322
2323 // The sharpened class might be unloaded if there is no class loader
2324 // contraint in place.
2325 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2326 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2327
2328 #ifndef PRODUCT
2329 if (C->print_intrinsics() || C->print_inlining()) {
2330 tty->print(" from base type: "); adr_type->dump();
2331 tty->print(" sharpened value: "); tjp->dump();
2332 }
2333 #endif
2334 // Sharpen the value type.
2335 return tjp;
2336 }
2337 return NULL;
2338 }
2339
2340 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2341 if (callee()->is_static()) return false; // caller must have the capability!
2342 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2343 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2344
2345 #ifndef PRODUCT
2346 {
2347 ResourceMark rm;
2348 // Check the signatures.
2349 ciSignature* sig = callee()->signature();
2350 #ifdef ASSERT
2351 if (!is_store) {
2352 // Object getObject(Object base, int/long offset), etc.
2353 BasicType rtype = sig->return_type()->basic_type();
2354 if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2355 rtype = T_ADDRESS; // it is really a C void*
2356 assert(rtype == type, "getter must return the expected value");
2357 if (!is_native_ptr) {
2358 assert(sig->count() == 2, "oop getter has 2 arguments");
2359 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2360 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2361 } else {
2362 assert(sig->count() == 1, "native getter has 1 argument");
2363 assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2364 }
2365 } else {
2366 // void putObject(Object base, int/long offset, Object x), etc.
2367 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2368 if (!is_native_ptr) {
2369 assert(sig->count() == 3, "oop putter has 3 arguments");
2370 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2371 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2372 } else {
2373 assert(sig->count() == 2, "native putter has 2 arguments");
2374 assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2375 }
2376 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2377 if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2378 vtype = T_ADDRESS; // it is really a C void*
2379 assert(vtype == type, "putter must accept the expected value");
2380 }
2381 #endif // ASSERT
2382 }
2383 #endif //PRODUCT
2384
2385 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2386
2387 Node* receiver = argument(0); // type: oop
2388
2389 // Build address expression.
2390 Node* adr;
2391 Node* heap_base_oop = top();
2392 Node* offset = top();
2393 Node* val;
2394
2395 if (!is_native_ptr) {
2396 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2397 Node* base = argument(1); // type: oop
2398 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2399 offset = argument(2); // type: long
2400 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2401 // to be plain byte offsets, which are also the same as those accepted
2402 // by oopDesc::field_base.
2403 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2404 "fieldOffset must be byte-scaled");
2405 // 32-bit machines ignore the high half!
2406 offset = ConvL2X(offset);
2407 adr = make_unsafe_address(base, offset);
2408 heap_base_oop = base;
2409 val = is_store ? argument(4) : NULL;
2410 } else {
2411 Node* ptr = argument(1); // type: long
2412 ptr = ConvL2X(ptr); // adjust Java long to machine word
2413 adr = make_unsafe_address(NULL, ptr);
2414 val = is_store ? argument(3) : NULL;
2415 }
2416
2417 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2418
2419 // First guess at the value type.
2420 const Type *value_type = Type::get_const_basic_type(type);
2421
2422 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM,
2423 // there was not enough information to nail it down.
2424 Compile::AliasType* alias_type = C->alias_type(adr_type);
2425 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2426
2427 // We will need memory barriers unless we can determine a unique
2428 // alias category for this reference. (Note: If for some reason
2429 // the barriers get omitted and the unsafe reference begins to "pollute"
2430 // the alias analysis of the rest of the graph, either Compile::can_alias
2431 // or Compile::must_alias will throw a diagnostic assert.)
2432 bool need_mem_bar;
2433 switch (kind) {
2434 case Relaxed:
2435 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2436 break;
2437 case Opaque:
2438 // Opaque uses CPUOrder membars for protection against code movement.
2439 case Acquire:
2440 case Release:
2441 case Volatile:
2442 need_mem_bar = true;
2443 break;
2444 default:
2445 ShouldNotReachHere();
2446 }
2447
2448 // Some accesses require access atomicity for all types, notably longs and doubles.
2449 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2450 bool requires_atomic_access = false;
2451 switch (kind) {
2452 case Relaxed:
2453 case Opaque:
2454 requires_atomic_access = AlwaysAtomicAccesses;
2455 break;
2456 case Acquire:
2457 case Release:
2458 case Volatile:
2459 requires_atomic_access = true;
2460 break;
2461 default:
2462 ShouldNotReachHere();
2463 }
2464
2465 // Figure out the memory ordering.
2466 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2467 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2468
2469 // If we are reading the value of the referent field of a Reference
2470 // object (either by using Unsafe directly or through reflection)
2471 // then, if G1 is enabled, we need to record the referent in an
2472 // SATB log buffer using the pre-barrier mechanism.
2473 // Also we need to add memory barrier to prevent commoning reads
2474 // from this field across safepoint since GC can change its value.
2475 bool need_read_barrier = !is_native_ptr && !is_store &&
2476 offset != top() && heap_base_oop != top();
2477
2478 if (!is_store && type == T_OBJECT) {
2479 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2480 if (tjp != NULL) {
2481 value_type = tjp;
2482 }
2483 }
2484
2485 receiver = null_check(receiver);
2486 if (stopped()) {
2487 return true;
2488 }
2489 // Heap pointers get a null-check from the interpreter,
2490 // as a courtesy. However, this is not guaranteed by Unsafe,
2491 // and it is not possible to fully distinguish unintended nulls
2492 // from intended ones in this API.
2493
2494 // We need to emit leading and trailing CPU membars (see below) in
2495 // addition to memory membars for special access modes. This is a little
2496 // too strong, but avoids the need to insert per-alias-type
2497 // volatile membars (for stores; compare Parse::do_put_xxx), which
2498 // we cannot do effectively here because we probably only have a
2499 // rough approximation of type.
2500
2501 switch(kind) {
2502 case Relaxed:
2503 case Opaque:
2504 case Acquire:
2505 break;
2506 case Release:
2507 case Volatile:
2508 if (is_store) {
2509 insert_mem_bar(Op_MemBarRelease);
2510 } else {
2511 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2512 insert_mem_bar(Op_MemBarVolatile);
2513 }
2514 }
2515 break;
2516 default:
2517 ShouldNotReachHere();
2518 }
2519
2520 // Memory barrier to prevent normal and 'unsafe' accesses from
2521 // bypassing each other. Happens after null checks, so the
2522 // exception paths do not take memory state from the memory barrier,
2523 // so there's no problems making a strong assert about mixing users
2524 // of safe & unsafe memory.
2525 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2526
2527 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2528 alias_type->field() != NULL || alias_type->element() != NULL, "field, array element or unknown");
2529 bool mismatched = false;
2530 if (alias_type->element() != NULL || alias_type->field() != NULL) {
2531 BasicType bt;
2532 if (alias_type->element() != NULL) {
2533 // Use address type to get the element type. Alias type doesn't provide
2534 // enough information (e.g., doesn't differentiate between byte[] and boolean[]).
2535 const Type* element = adr_type->is_aryptr()->elem();
2536 bt = element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type();
2537 } else {
2538 bt = alias_type->field()->layout_type();
2539 }
2540 if (bt == T_ARRAY) {
2541 // accessing an array field with getObject is not a mismatch
2542 bt = T_OBJECT;
2543 }
2544 if (bt != type) {
2545 mismatched = true;
2546 }
2547 }
2548 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2549
2550 if (!is_store) {
2551 Node* p = NULL;
2552 // Try to constant fold a load from a constant field
2553 ciField* field = alias_type->field();
2554 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2555 // final or stable field
2556 p = make_constant_from_field(field, heap_base_oop);
2557 }
2558 if (p == NULL) {
2559 // To be valid, unsafe loads may depend on other conditions than
2560 // the one that guards them: pin the Load node
2561 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2562 // load value
2563 switch (type) {
2564 case T_BOOLEAN:
2565 case T_CHAR:
2566 case T_BYTE:
2567 case T_SHORT:
2568 case T_INT:
2569 case T_LONG:
2570 case T_FLOAT:
2571 case T_DOUBLE:
2572 break;
2573 case T_OBJECT:
2574 if (need_read_barrier) {
2575 // We do not require a mem bar inside pre_barrier if need_mem_bar
2576 // is set: the barriers would be emitted by us.
2577 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2578 }
2579 break;
2580 case T_ADDRESS:
2581 // Cast to an int type.
2582 p = _gvn.transform(new CastP2XNode(NULL, p));
2583 p = ConvX2UL(p);
2584 break;
2585 default:
2586 fatal("unexpected type %d: %s", type, type2name(type));
2587 break;
2588 }
2589 }
2590 // The load node has the control of the preceding MemBarCPUOrder. All
2591 // following nodes will have the control of the MemBarCPUOrder inserted at
2592 // the end of this method. So, pushing the load onto the stack at a later
2593 // point is fine.
2594 set_result(p);
2595 } else {
2596 // place effect of store into memory
2597 switch (type) {
2598 case T_DOUBLE:
2599 val = dstore_rounding(val);
2600 break;
2601 case T_ADDRESS:
2602 // Repackage the long as a pointer.
2603 val = ConvL2X(val);
2604 val = _gvn.transform(new CastX2PNode(val));
2605 break;
2606 }
2607
2608 if (type != T_OBJECT) {
2609 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2610 } else {
2611 // Possibly an oop being stored to Java heap or native memory
2612 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2613 // oop to Java heap.
2614 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2615 } else {
2616 // We can't tell at compile time if we are storing in the Java heap or outside
2617 // of it. So we need to emit code to conditionally do the proper type of
2618 // store.
2619
2620 IdealKit ideal(this);
2621 #define __ ideal.
2622 // QQQ who knows what probability is here??
2623 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2624 // Sync IdealKit and graphKit.
2625 sync_kit(ideal);
2626 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2627 // Update IdealKit memory.
2628 __ sync_kit(this);
2629 } __ else_(); {
2630 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2631 } __ end_if();
2632 // Final sync IdealKit and GraphKit.
2633 final_sync(ideal);
2634 #undef __
2635 }
2636 }
2637 }
2638
2639 switch(kind) {
2640 case Relaxed:
2641 case Opaque:
2642 case Release:
2643 break;
2644 case Acquire:
2645 case Volatile:
2646 if (!is_store) {
2647 insert_mem_bar(Op_MemBarAcquire);
2648 } else {
2649 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2650 insert_mem_bar(Op_MemBarVolatile);
2651 }
2652 }
2653 break;
2654 default:
2655 ShouldNotReachHere();
2656 }
2657
2658 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2659
2660 return true;
2661 }
2662
2663 //----------------------------inline_unsafe_load_store----------------------------
2664 // This method serves a couple of different customers (depending on LoadStoreKind):
2665 //
2666 // LS_cmp_swap:
2667 //
2668 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2669 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2670 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2671 //
2672 // LS_cmp_swap_weak:
2673 //
2674 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2675 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2676 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2677 //
2678 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2679 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2680 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2681 //
2682 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2683 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2684 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2685 //
2686 // LS_cmp_exchange:
2687 //
2688 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2689 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2690 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2691 //
2692 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2693 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2694 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2695 //
2696 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2697 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2698 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2699 //
2700 // LS_get_add:
2701 //
2702 // int getAndAddInt( Object o, long offset, int delta)
2703 // long getAndAddLong(Object o, long offset, long delta)
2704 //
2705 // LS_get_set:
2706 //
2707 // int getAndSet(Object o, long offset, int newValue)
2708 // long getAndSet(Object o, long offset, long newValue)
2709 // Object getAndSet(Object o, long offset, Object newValue)
2710 //
2711 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2712 // This basic scheme here is the same as inline_unsafe_access, but
2713 // differs in enough details that combining them would make the code
2714 // overly confusing. (This is a true fact! I originally combined
2715 // them, but even I was confused by it!) As much code/comments as
2716 // possible are retained from inline_unsafe_access though to make
2717 // the correspondences clearer. - dl
2718
2719 if (callee()->is_static()) return false; // caller must have the capability!
2720
2721 #ifndef PRODUCT
2722 BasicType rtype;
2723 {
2724 ResourceMark rm;
2725 // Check the signatures.
2726 ciSignature* sig = callee()->signature();
2727 rtype = sig->return_type()->basic_type();
2728 switch(kind) {
2729 case LS_get_add:
2730 case LS_get_set: {
2731 // Check the signatures.
2732 #ifdef ASSERT
2733 assert(rtype == type, "get and set must return the expected type");
2734 assert(sig->count() == 3, "get and set has 3 arguments");
2735 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2736 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2737 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2738 #endif // ASSERT
2739 break;
2740 }
2741 case LS_cmp_swap:
2742 case LS_cmp_swap_weak: {
2743 // Check the signatures.
2744 #ifdef ASSERT
2745 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2746 assert(sig->count() == 4, "CAS has 4 arguments");
2747 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2748 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2749 #endif // ASSERT
2750 break;
2751 }
2752 case LS_cmp_exchange: {
2753 // Check the signatures.
2754 #ifdef ASSERT
2755 assert(rtype == type, "CAS must return the expected type");
2756 assert(sig->count() == 4, "CAS has 4 arguments");
2757 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2758 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2759 #endif // ASSERT
2760 break;
2761 }
2762 default:
2763 ShouldNotReachHere();
2764 }
2765 }
2766 #endif //PRODUCT
2767
2768 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2769
2770 // Get arguments:
2771 Node* receiver = NULL;
2772 Node* base = NULL;
2773 Node* offset = NULL;
2774 Node* oldval = NULL;
2775 Node* newval = NULL;
2776 switch(kind) {
2777 case LS_cmp_swap:
2778 case LS_cmp_swap_weak:
2779 case LS_cmp_exchange: {
2780 const bool two_slot_type = type2size[type] == 2;
2781 receiver = argument(0); // type: oop
2782 base = argument(1); // type: oop
2783 offset = argument(2); // type: long
2784 oldval = argument(4); // type: oop, int, or long
2785 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2786 break;
2787 }
2788 case LS_get_add:
2789 case LS_get_set: {
2790 receiver = argument(0); // type: oop
2791 base = argument(1); // type: oop
2792 offset = argument(2); // type: long
2793 oldval = NULL;
2794 newval = argument(4); // type: oop, int, or long
2795 break;
2796 }
2797 default:
2798 ShouldNotReachHere();
2799 }
2800
2801 // Null check receiver.
2802 receiver = null_check(receiver);
2803 if (stopped()) {
2804 return true;
2805 }
2806
2807 // Build field offset expression.
2808 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2809 // to be plain byte offsets, which are also the same as those accepted
2810 // by oopDesc::field_base.
2811 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2812 // 32-bit machines ignore the high half of long offsets
2813 offset = ConvL2X(offset);
2814 Node* adr = make_unsafe_address(base, offset);
2815 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2816
2817 // For CAS, unlike inline_unsafe_access, there seems no point in
2818 // trying to refine types. Just use the coarse types here.
2819 const Type *value_type = Type::get_const_basic_type(type);
2820 Compile::AliasType* alias_type = C->alias_type(adr_type);
2821 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2822
2823 switch (kind) {
2824 case LS_get_set:
2825 case LS_cmp_exchange: {
2826 if (type == T_OBJECT) {
2827 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2828 if (tjp != NULL) {
2829 value_type = tjp;
2830 }
2831 }
2832 break;
2833 }
2834 case LS_cmp_swap:
2835 case LS_cmp_swap_weak:
2836 case LS_get_add:
2837 break;
2838 default:
2839 ShouldNotReachHere();
2840 }
2841
2842 int alias_idx = C->get_alias_index(adr_type);
2843
2844 // Memory-model-wise, a LoadStore acts like a little synchronized
2845 // block, so needs barriers on each side. These don't translate
2846 // into actual barriers on most machines, but we still need rest of
2847 // compiler to respect ordering.
2848
2849 switch (access_kind) {
2850 case Relaxed:
2851 case Acquire:
2852 break;
2853 case Release:
2854 case Volatile:
2855 insert_mem_bar(Op_MemBarRelease);
2856 break;
2857 default:
2858 ShouldNotReachHere();
2859 }
2860 insert_mem_bar(Op_MemBarCPUOrder);
2861
2862 // Figure out the memory ordering.
2863 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2864
2865 // 4984716: MemBars must be inserted before this
2866 // memory node in order to avoid a false
2867 // dependency which will confuse the scheduler.
2868 Node *mem = memory(alias_idx);
2869
2870 // For now, we handle only those cases that actually exist: ints,
2871 // longs, and Object. Adding others should be straightforward.
2872 Node* load_store = NULL;
2873 switch(type) {
2874 case T_INT:
2875 switch(kind) {
2876 case LS_get_add:
2877 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2878 break;
2879 case LS_get_set:
2880 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2881 break;
2882 case LS_cmp_swap_weak:
2883 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2884 break;
2885 case LS_cmp_swap:
2886 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2887 break;
2888 case LS_cmp_exchange:
2889 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2890 break;
2891 default:
2892 ShouldNotReachHere();
2893 }
2894 break;
2895 case T_LONG:
2896 switch(kind) {
2897 case LS_get_add:
2898 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2899 break;
2900 case LS_get_set:
2901 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2902 break;
2903 case LS_cmp_swap_weak:
2904 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2905 break;
2906 case LS_cmp_swap:
2907 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2908 break;
2909 case LS_cmp_exchange:
2910 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2911 break;
2912 default:
2913 ShouldNotReachHere();
2914 }
2915 break;
2916 case T_OBJECT:
2917 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2918 // could be delayed during Parse (for example, in adjust_map_after_if()).
2919 // Execute transformation here to avoid barrier generation in such case.
2920 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2921 newval = _gvn.makecon(TypePtr::NULL_PTR);
2922
2923 // Reference stores need a store barrier.
2924 switch(kind) {
2925 case LS_get_set: {
2926 // If pre-barrier must execute before the oop store, old value will require do_load here.
2927 if (!can_move_pre_barrier()) {
2928 pre_barrier(true /* do_load*/,
2929 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2930 NULL /* pre_val*/,
2931 T_OBJECT);
2932 } // Else move pre_barrier to use load_store value, see below.
2933 break;
2934 }
2935 case LS_cmp_swap_weak:
2936 case LS_cmp_swap:
2937 case LS_cmp_exchange: {
2938 // Same as for newval above:
2939 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2940 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2941 }
2942 // The only known value which might get overwritten is oldval.
2943 pre_barrier(false /* do_load */,
2944 control(), NULL, NULL, max_juint, NULL, NULL,
2945 oldval /* pre_val */,
2946 T_OBJECT);
2947 break;
2948 }
2949 default:
2950 ShouldNotReachHere();
2951 }
2952
2953 #ifdef _LP64
2954 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2955 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2956
2957 switch(kind) {
2958 case LS_get_set:
2959 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2960 break;
2961 case LS_cmp_swap_weak: {
2962 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2963 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2964 break;
2965 }
2966 case LS_cmp_swap: {
2967 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2968 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2969 break;
2970 }
2971 case LS_cmp_exchange: {
2972 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2973 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2974 break;
2975 }
2976 default:
2977 ShouldNotReachHere();
2978 }
2979 } else
2980 #endif
2981 switch (kind) {
2982 case LS_get_set:
2983 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2984 break;
2985 case LS_cmp_swap_weak:
2986 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2987 break;
2988 case LS_cmp_swap:
2989 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2990 break;
2991 case LS_cmp_exchange:
2992 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2993 break;
2994 default:
2995 ShouldNotReachHere();
2996 }
2997
2998 // Emit the post barrier only when the actual store happened. This makes sense
2999 // to check only for LS_cmp_* that can fail to set the value.
3000 // LS_cmp_exchange does not produce any branches by default, so there is no
3001 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3002 // CompareAndExchange and move branches here, it would make sense to conditionalize
3003 // post_barriers for LS_cmp_exchange as well.
3004 //
3005 // CAS success path is marked more likely since we anticipate this is a performance
3006 // critical path, while CAS failure path can use the penalty for going through unlikely
3007 // path as backoff. Which is still better than doing a store barrier there.
3008 switch (kind) {
3009 case LS_get_set:
3010 case LS_cmp_exchange: {
3011 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3012 break;
3013 }
3014 case LS_cmp_swap_weak:
3015 case LS_cmp_swap: {
3016 IdealKit ideal(this);
3017 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3018 sync_kit(ideal);
3019 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3020 ideal.sync_kit(this);
3021 } ideal.end_if();
3022 final_sync(ideal);
3023 break;
3024 }
3025 default:
3026 ShouldNotReachHere();
3027 }
3028 break;
3029 default:
3030 fatal("unexpected type %d: %s", type, type2name(type));
3031 break;
3032 }
3033
3034 // SCMemProjNodes represent the memory state of a LoadStore. Their
3035 // main role is to prevent LoadStore nodes from being optimized away
3036 // when their results aren't used.
3037 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3038 set_memory(proj, alias_idx);
3039
3040 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3041 #ifdef _LP64
3042 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3043 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3044 }
3045 #endif
3046 if (can_move_pre_barrier()) {
3047 // Don't need to load pre_val. The old value is returned by load_store.
3048 // The pre_barrier can execute after the xchg as long as no safepoint
3049 // gets inserted between them.
3050 pre_barrier(false /* do_load */,
3051 control(), NULL, NULL, max_juint, NULL, NULL,
3052 load_store /* pre_val */,
3053 T_OBJECT);
3054 }
3055 }
3056
3057 // Add the trailing membar surrounding the access
3058 insert_mem_bar(Op_MemBarCPUOrder);
3059
3060 switch (access_kind) {
3061 case Relaxed:
3062 case Release:
3063 break; // do nothing
3064 case Acquire:
3065 case Volatile:
3066 insert_mem_bar(Op_MemBarAcquire);
3067 break;
3068 default:
3069 ShouldNotReachHere();
3070 }
3071
3072 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3073 set_result(load_store);
3074 return true;
3075 }
3076
3077 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3078 MemNode::MemOrd mo = MemNode::unset;
3079 switch(kind) {
3080 case Opaque:
3081 case Relaxed: mo = MemNode::unordered; break;
3082 case Acquire: mo = MemNode::acquire; break;
3083 case Release: mo = MemNode::release; break;
3084 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3085 default:
3086 ShouldNotReachHere();
3087 }
3088 guarantee(mo != MemNode::unset, "Should select memory ordering");
3089 return mo;
3090 }
3091
3092 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3093 MemNode::MemOrd mo = MemNode::unset;
3094 switch(kind) {
3095 case Opaque:
3096 case Relaxed: mo = MemNode::unordered; break;
3097 case Acquire: mo = MemNode::acquire; break;
3098 case Release: mo = MemNode::release; break;
3099 case Volatile: mo = MemNode::seqcst; break;
3100 default:
3101 ShouldNotReachHere();
3102 }
3103 guarantee(mo != MemNode::unset, "Should select memory ordering");
3104 return mo;
3105 }
3106
3107 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3108 // Regardless of form, don't allow previous ld/st to move down,
3109 // then issue acquire, release, or volatile mem_bar.
3110 insert_mem_bar(Op_MemBarCPUOrder);
3111 switch(id) {
3112 case vmIntrinsics::_loadFence:
3113 insert_mem_bar(Op_LoadFence);
3114 return true;
3115 case vmIntrinsics::_storeFence:
3116 insert_mem_bar(Op_StoreFence);
3117 return true;
3118 case vmIntrinsics::_fullFence:
3119 insert_mem_bar(Op_MemBarVolatile);
3120 return true;
3121 default:
3122 fatal_unexpected_iid(id);
3123 return false;
3124 }
3125 }
3126
3127 bool LibraryCallKit::inline_onspinwait() {
3128 insert_mem_bar(Op_OnSpinWait);
3129 return true;
3130 }
3131
3132 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3133 if (!kls->is_Con()) {
3134 return true;
3135 }
3136 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3137 if (klsptr == NULL) {
3138 return true;
3139 }
3140 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3141 // don't need a guard for a klass that is already initialized
3142 return !ik->is_initialized();
3143 }
3144
3145 //----------------------------inline_unsafe_allocate---------------------------
3146 // public native Object Unsafe.allocateInstance(Class<?> cls);
3147 bool LibraryCallKit::inline_unsafe_allocate() {
3148 if (callee()->is_static()) return false; // caller must have the capability!
3149
3150 null_check_receiver(); // null-check, then ignore
3151 Node* cls = null_check(argument(1));
3152 if (stopped()) return true;
3153
3154 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3155 kls = null_check(kls);
3156 if (stopped()) return true; // argument was like int.class
3157
3158 Node* test = NULL;
3159 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3160 // Note: The argument might still be an illegal value like
3161 // Serializable.class or Object[].class. The runtime will handle it.
3162 // But we must make an explicit check for initialization.
3163 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3164 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3165 // can generate code to load it as unsigned byte.
3166 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3167 Node* bits = intcon(InstanceKlass::fully_initialized);
3168 test = _gvn.transform(new SubINode(inst, bits));
3169 // The 'test' is non-zero if we need to take a slow path.
3170 }
3171
3172 Node* obj = new_instance(kls, test);
3173 set_result(obj);
3174 return true;
3175 }
3176
3177 //------------------------inline_native_time_funcs--------------
3178 // inline code for System.currentTimeMillis() and System.nanoTime()
3179 // these have the same type and signature
3180 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3181 const TypeFunc* tf = OptoRuntime::void_long_Type();
3182 const TypePtr* no_memory_effects = NULL;
3183 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3184 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3185 #ifdef ASSERT
3186 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3187 assert(value_top == top(), "second value must be top");
3188 #endif
3189 set_result(value);
3190 return true;
3191 }
3192
3193 //------------------------inline_native_currentThread------------------
3194 bool LibraryCallKit::inline_native_currentThread() {
3195 Node* junk = NULL;
3196 set_result(generate_current_thread(junk));
3197 return true;
3198 }
3199
3200 //------------------------inline_native_isInterrupted------------------
3201 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3202 bool LibraryCallKit::inline_native_isInterrupted() {
3203 // Add a fast path to t.isInterrupted(clear_int):
3204 // (t == Thread.current() &&
3205 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3206 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3207 // So, in the common case that the interrupt bit is false,
3208 // we avoid making a call into the VM. Even if the interrupt bit
3209 // is true, if the clear_int argument is false, we avoid the VM call.
3210 // However, if the receiver is not currentThread, we must call the VM,
3211 // because there must be some locking done around the operation.
3212
3213 // We only go to the fast case code if we pass two guards.
3214 // Paths which do not pass are accumulated in the slow_region.
3215
3216 enum {
3217 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3218 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3219 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3220 PATH_LIMIT
3221 };
3222
3223 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3224 // out of the function.
3225 insert_mem_bar(Op_MemBarCPUOrder);
3226
3227 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3228 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3229
3230 RegionNode* slow_region = new RegionNode(1);
3231 record_for_igvn(slow_region);
3232
3233 // (a) Receiving thread must be the current thread.
3234 Node* rec_thr = argument(0);
3235 Node* tls_ptr = NULL;
3236 Node* cur_thr = generate_current_thread(tls_ptr);
3237 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3238 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3239
3240 generate_slow_guard(bol_thr, slow_region);
3241
3242 // (b) Interrupt bit on TLS must be false.
3243 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3244 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3245 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3246
3247 // Set the control input on the field _interrupted read to prevent it floating up.
3248 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3249 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3250 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3251
3252 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3253
3254 // First fast path: if (!TLS._interrupted) return false;
3255 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3256 result_rgn->init_req(no_int_result_path, false_bit);
3257 result_val->init_req(no_int_result_path, intcon(0));
3258
3259 // drop through to next case
3260 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3261
3262 #ifndef TARGET_OS_FAMILY_windows
3263 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3264 Node* clr_arg = argument(1);
3265 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3266 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3267 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3268
3269 // Second fast path: ... else if (!clear_int) return true;
3270 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3271 result_rgn->init_req(no_clear_result_path, false_arg);
3272 result_val->init_req(no_clear_result_path, intcon(1));
3273
3274 // drop through to next case
3275 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3276 #else
3277 // To return true on Windows you must read the _interrupted field
3278 // and check the event state i.e. take the slow path.
3279 #endif // TARGET_OS_FAMILY_windows
3280
3281 // (d) Otherwise, go to the slow path.
3282 slow_region->add_req(control());
3283 set_control( _gvn.transform(slow_region));
3284
3285 if (stopped()) {
3286 // There is no slow path.
3287 result_rgn->init_req(slow_result_path, top());
3288 result_val->init_req(slow_result_path, top());
3289 } else {
3290 // non-virtual because it is a private non-static
3291 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3292
3293 Node* slow_val = set_results_for_java_call(slow_call);
3294 // this->control() comes from set_results_for_java_call
3295
3296 Node* fast_io = slow_call->in(TypeFunc::I_O);
3297 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3298
3299 // These two phis are pre-filled with copies of of the fast IO and Memory
3300 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3301 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3302
3303 result_rgn->init_req(slow_result_path, control());
3304 result_io ->init_req(slow_result_path, i_o());
3305 result_mem->init_req(slow_result_path, reset_memory());
3306 result_val->init_req(slow_result_path, slow_val);
3307
3308 set_all_memory(_gvn.transform(result_mem));
3309 set_i_o( _gvn.transform(result_io));
3310 }
3311
3312 C->set_has_split_ifs(true); // Has chance for split-if optimization
3313 set_result(result_rgn, result_val);
3314 return true;
3315 }
3316
3317 //---------------------------load_mirror_from_klass----------------------------
3318 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3319 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3320 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3321 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3322 }
3323
3324 //-----------------------load_klass_from_mirror_common-------------------------
3325 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3326 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3327 // and branch to the given path on the region.
3328 // If never_see_null, take an uncommon trap on null, so we can optimistically
3329 // compile for the non-null case.
3330 // If the region is NULL, force never_see_null = true.
3331 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3332 bool never_see_null,
3333 RegionNode* region,
3334 int null_path,
3335 int offset) {
3336 if (region == NULL) never_see_null = true;
3337 Node* p = basic_plus_adr(mirror, offset);
3338 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3339 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3340 Node* null_ctl = top();
3341 kls = null_check_oop(kls, &null_ctl, never_see_null);
3342 if (region != NULL) {
3343 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3344 region->init_req(null_path, null_ctl);
3345 } else {
3346 assert(null_ctl == top(), "no loose ends");
3347 }
3348 return kls;
3349 }
3350
3351 //--------------------(inline_native_Class_query helpers)---------------------
3352 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3353 // Fall through if (mods & mask) == bits, take the guard otherwise.
3354 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3355 // Branch around if the given klass has the given modifier bit set.
3356 // Like generate_guard, adds a new path onto the region.
3357 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3358 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3359 Node* mask = intcon(modifier_mask);
3360 Node* bits = intcon(modifier_bits);
3361 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3362 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3363 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3364 return generate_fair_guard(bol, region);
3365 }
3366 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3367 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3368 }
3369
3370 //-------------------------inline_native_Class_query-------------------
3371 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3372 const Type* return_type = TypeInt::BOOL;
3373 Node* prim_return_value = top(); // what happens if it's a primitive class?
3374 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3375 bool expect_prim = false; // most of these guys expect to work on refs
3376
3377 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3378
3379 Node* mirror = argument(0);
3380 Node* obj = top();
3381
3382 switch (id) {
3383 case vmIntrinsics::_isInstance:
3384 // nothing is an instance of a primitive type
3385 prim_return_value = intcon(0);
3386 obj = argument(1);
3387 break;
3388 case vmIntrinsics::_getModifiers:
3389 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3390 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3391 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3392 break;
3393 case vmIntrinsics::_isInterface:
3394 prim_return_value = intcon(0);
3395 break;
3396 case vmIntrinsics::_isArray:
3397 prim_return_value = intcon(0);
3398 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3399 break;
3400 case vmIntrinsics::_isPrimitive:
3401 prim_return_value = intcon(1);
3402 expect_prim = true; // obviously
3403 break;
3404 case vmIntrinsics::_getSuperclass:
3405 prim_return_value = null();
3406 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3407 break;
3408 case vmIntrinsics::_getClassAccessFlags:
3409 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3410 return_type = TypeInt::INT; // not bool! 6297094
3411 break;
3412 default:
3413 fatal_unexpected_iid(id);
3414 break;
3415 }
3416
3417 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3418 if (mirror_con == NULL) return false; // cannot happen?
3419
3420 #ifndef PRODUCT
3421 if (C->print_intrinsics() || C->print_inlining()) {
3422 ciType* k = mirror_con->java_mirror_type();
3423 if (k) {
3424 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3425 k->print_name();
3426 tty->cr();
3427 }
3428 }
3429 #endif
3430
3431 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3432 RegionNode* region = new RegionNode(PATH_LIMIT);
3433 record_for_igvn(region);
3434 PhiNode* phi = new PhiNode(region, return_type);
3435
3436 // The mirror will never be null of Reflection.getClassAccessFlags, however
3437 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3438 // if it is. See bug 4774291.
3439
3440 // For Reflection.getClassAccessFlags(), the null check occurs in
3441 // the wrong place; see inline_unsafe_access(), above, for a similar
3442 // situation.
3443 mirror = null_check(mirror);
3444 // If mirror or obj is dead, only null-path is taken.
3445 if (stopped()) return true;
3446
3447 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3448
3449 // Now load the mirror's klass metaobject, and null-check it.
3450 // Side-effects region with the control path if the klass is null.
3451 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3452 // If kls is null, we have a primitive mirror.
3453 phi->init_req(_prim_path, prim_return_value);
3454 if (stopped()) { set_result(region, phi); return true; }
3455 bool safe_for_replace = (region->in(_prim_path) == top());
3456
3457 Node* p; // handy temp
3458 Node* null_ctl;
3459
3460 // Now that we have the non-null klass, we can perform the real query.
3461 // For constant classes, the query will constant-fold in LoadNode::Value.
3462 Node* query_value = top();
3463 switch (id) {
3464 case vmIntrinsics::_isInstance:
3465 // nothing is an instance of a primitive type
3466 query_value = gen_instanceof(obj, kls, safe_for_replace);
3467 break;
3468
3469 case vmIntrinsics::_getModifiers:
3470 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3471 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3472 break;
3473
3474 case vmIntrinsics::_isInterface:
3475 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3476 if (generate_interface_guard(kls, region) != NULL)
3477 // A guard was added. If the guard is taken, it was an interface.
3478 phi->add_req(intcon(1));
3479 // If we fall through, it's a plain class.
3480 query_value = intcon(0);
3481 break;
3482
3483 case vmIntrinsics::_isArray:
3484 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3485 if (generate_array_guard(kls, region) != NULL)
3486 // A guard was added. If the guard is taken, it was an array.
3487 phi->add_req(intcon(1));
3488 // If we fall through, it's a plain class.
3489 query_value = intcon(0);
3490 break;
3491
3492 case vmIntrinsics::_isPrimitive:
3493 query_value = intcon(0); // "normal" path produces false
3494 break;
3495
3496 case vmIntrinsics::_getSuperclass:
3497 // The rules here are somewhat unfortunate, but we can still do better
3498 // with random logic than with a JNI call.
3499 // Interfaces store null or Object as _super, but must report null.
3500 // Arrays store an intermediate super as _super, but must report Object.
3501 // Other types can report the actual _super.
3502 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3503 if (generate_interface_guard(kls, region) != NULL)
3504 // A guard was added. If the guard is taken, it was an interface.
3505 phi->add_req(null());
3506 if (generate_array_guard(kls, region) != NULL)
3507 // A guard was added. If the guard is taken, it was an array.
3508 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3509 // If we fall through, it's a plain class. Get its _super.
3510 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3511 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3512 null_ctl = top();
3513 kls = null_check_oop(kls, &null_ctl);
3514 if (null_ctl != top()) {
3515 // If the guard is taken, Object.superClass is null (both klass and mirror).
3516 region->add_req(null_ctl);
3517 phi ->add_req(null());
3518 }
3519 if (!stopped()) {
3520 query_value = load_mirror_from_klass(kls);
3521 }
3522 break;
3523
3524 case vmIntrinsics::_getClassAccessFlags:
3525 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3526 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3527 break;
3528
3529 default:
3530 fatal_unexpected_iid(id);
3531 break;
3532 }
3533
3534 // Fall-through is the normal case of a query to a real class.
3535 phi->init_req(1, query_value);
3536 region->init_req(1, control());
3537
3538 C->set_has_split_ifs(true); // Has chance for split-if optimization
3539 set_result(region, phi);
3540 return true;
3541 }
3542
3543 //-------------------------inline_Class_cast-------------------
3544 bool LibraryCallKit::inline_Class_cast() {
3545 Node* mirror = argument(0); // Class
3546 Node* obj = argument(1);
3547 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3548 if (mirror_con == NULL) {
3549 return false; // dead path (mirror->is_top()).
3550 }
3551 if (obj == NULL || obj->is_top()) {
3552 return false; // dead path
3553 }
3554 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3555
3556 // First, see if Class.cast() can be folded statically.
3557 // java_mirror_type() returns non-null for compile-time Class constants.
3558 ciType* tm = mirror_con->java_mirror_type();
3559 if (tm != NULL && tm->is_klass() &&
3560 tp != NULL && tp->klass() != NULL) {
3561 if (!tp->klass()->is_loaded()) {
3562 // Don't use intrinsic when class is not loaded.
3563 return false;
3564 } else {
3565 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3566 if (static_res == Compile::SSC_always_true) {
3567 // isInstance() is true - fold the code.
3568 set_result(obj);
3569 return true;
3570 } else if (static_res == Compile::SSC_always_false) {
3571 // Don't use intrinsic, have to throw ClassCastException.
3572 // If the reference is null, the non-intrinsic bytecode will
3573 // be optimized appropriately.
3574 return false;
3575 }
3576 }
3577 }
3578
3579 // Bailout intrinsic and do normal inlining if exception path is frequent.
3580 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3581 return false;
3582 }
3583
3584 // Generate dynamic checks.
3585 // Class.cast() is java implementation of _checkcast bytecode.
3586 // Do checkcast (Parse::do_checkcast()) optimizations here.
3587
3588 mirror = null_check(mirror);
3589 // If mirror is dead, only null-path is taken.
3590 if (stopped()) {
3591 return true;
3592 }
3593
3594 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3595 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3596 RegionNode* region = new RegionNode(PATH_LIMIT);
3597 record_for_igvn(region);
3598
3599 // Now load the mirror's klass metaobject, and null-check it.
3600 // If kls is null, we have a primitive mirror and
3601 // nothing is an instance of a primitive type.
3602 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3603
3604 Node* res = top();
3605 if (!stopped()) {
3606 Node* bad_type_ctrl = top();
3607 // Do checkcast optimizations.
3608 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3609 region->init_req(_bad_type_path, bad_type_ctrl);
3610 }
3611 if (region->in(_prim_path) != top() ||
3612 region->in(_bad_type_path) != top()) {
3613 // Let Interpreter throw ClassCastException.
3614 PreserveJVMState pjvms(this);
3615 set_control(_gvn.transform(region));
3616 uncommon_trap(Deoptimization::Reason_intrinsic,
3617 Deoptimization::Action_maybe_recompile);
3618 }
3619 if (!stopped()) {
3620 set_result(res);
3621 }
3622 return true;
3623 }
3624
3625
3626 //--------------------------inline_native_subtype_check------------------------
3627 // This intrinsic takes the JNI calls out of the heart of
3628 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3629 bool LibraryCallKit::inline_native_subtype_check() {
3630 // Pull both arguments off the stack.
3631 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3632 args[0] = argument(0);
3633 args[1] = argument(1);
3634 Node* klasses[2]; // corresponding Klasses: superk, subk
3635 klasses[0] = klasses[1] = top();
3636
3637 enum {
3638 // A full decision tree on {superc is prim, subc is prim}:
3639 _prim_0_path = 1, // {P,N} => false
3640 // {P,P} & superc!=subc => false
3641 _prim_same_path, // {P,P} & superc==subc => true
3642 _prim_1_path, // {N,P} => false
3643 _ref_subtype_path, // {N,N} & subtype check wins => true
3644 _both_ref_path, // {N,N} & subtype check loses => false
3645 PATH_LIMIT
3646 };
3647
3648 RegionNode* region = new RegionNode(PATH_LIMIT);
3649 Node* phi = new PhiNode(region, TypeInt::BOOL);
3650 record_for_igvn(region);
3651
3652 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3653 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3654 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3655
3656 // First null-check both mirrors and load each mirror's klass metaobject.
3657 int which_arg;
3658 for (which_arg = 0; which_arg <= 1; which_arg++) {
3659 Node* arg = args[which_arg];
3660 arg = null_check(arg);
3661 if (stopped()) break;
3662 args[which_arg] = arg;
3663
3664 Node* p = basic_plus_adr(arg, class_klass_offset);
3665 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3666 klasses[which_arg] = _gvn.transform(kls);
3667 }
3668
3669 // Having loaded both klasses, test each for null.
3670 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3671 for (which_arg = 0; which_arg <= 1; which_arg++) {
3672 Node* kls = klasses[which_arg];
3673 Node* null_ctl = top();
3674 kls = null_check_oop(kls, &null_ctl, never_see_null);
3675 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3676 region->init_req(prim_path, null_ctl);
3677 if (stopped()) break;
3678 klasses[which_arg] = kls;
3679 }
3680
3681 if (!stopped()) {
3682 // now we have two reference types, in klasses[0..1]
3683 Node* subk = klasses[1]; // the argument to isAssignableFrom
3684 Node* superk = klasses[0]; // the receiver
3685 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3686 // now we have a successful reference subtype check
3687 region->set_req(_ref_subtype_path, control());
3688 }
3689
3690 // If both operands are primitive (both klasses null), then
3691 // we must return true when they are identical primitives.
3692 // It is convenient to test this after the first null klass check.
3693 set_control(region->in(_prim_0_path)); // go back to first null check
3694 if (!stopped()) {
3695 // Since superc is primitive, make a guard for the superc==subc case.
3696 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3697 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3698 generate_guard(bol_eq, region, PROB_FAIR);
3699 if (region->req() == PATH_LIMIT+1) {
3700 // A guard was added. If the added guard is taken, superc==subc.
3701 region->swap_edges(PATH_LIMIT, _prim_same_path);
3702 region->del_req(PATH_LIMIT);
3703 }
3704 region->set_req(_prim_0_path, control()); // Not equal after all.
3705 }
3706
3707 // these are the only paths that produce 'true':
3708 phi->set_req(_prim_same_path, intcon(1));
3709 phi->set_req(_ref_subtype_path, intcon(1));
3710
3711 // pull together the cases:
3712 assert(region->req() == PATH_LIMIT, "sane region");
3713 for (uint i = 1; i < region->req(); i++) {
3714 Node* ctl = region->in(i);
3715 if (ctl == NULL || ctl == top()) {
3716 region->set_req(i, top());
3717 phi ->set_req(i, top());
3718 } else if (phi->in(i) == NULL) {
3719 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3720 }
3721 }
3722
3723 set_control(_gvn.transform(region));
3724 set_result(_gvn.transform(phi));
3725 return true;
3726 }
3727
3728 //---------------------generate_array_guard_common------------------------
3729 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3730 bool obj_array, bool not_array) {
3731
3732 if (stopped()) {
3733 return NULL;
3734 }
3735
3736 // If obj_array/non_array==false/false:
3737 // Branch around if the given klass is in fact an array (either obj or prim).
3738 // If obj_array/non_array==false/true:
3739 // Branch around if the given klass is not an array klass of any kind.
3740 // If obj_array/non_array==true/true:
3741 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3742 // If obj_array/non_array==true/false:
3743 // Branch around if the kls is an oop array (Object[] or subtype)
3744 //
3745 // Like generate_guard, adds a new path onto the region.
3746 jint layout_con = 0;
3747 Node* layout_val = get_layout_helper(kls, layout_con);
3748 if (layout_val == NULL) {
3749 bool query = (obj_array
3750 ? Klass::layout_helper_is_objArray(layout_con)
3751 : Klass::layout_helper_is_array(layout_con));
3752 if (query == not_array) {
3753 return NULL; // never a branch
3754 } else { // always a branch
3755 Node* always_branch = control();
3756 if (region != NULL)
3757 region->add_req(always_branch);
3758 set_control(top());
3759 return always_branch;
3760 }
3761 }
3762 // Now test the correct condition.
3763 jint nval = (obj_array
3764 ? ((jint)Klass::_lh_array_tag_type_value
3765 << Klass::_lh_array_tag_shift)
3766 : Klass::_lh_neutral_value);
3767 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3768 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3769 // invert the test if we are looking for a non-array
3770 if (not_array) btest = BoolTest(btest).negate();
3771 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3772 return generate_fair_guard(bol, region);
3773 }
3774
3775
3776 //-----------------------inline_native_newArray--------------------------
3777 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3778 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3779 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3780 Node* mirror;
3781 Node* count_val;
3782 if (uninitialized) {
3783 mirror = argument(1);
3784 count_val = argument(2);
3785 } else {
3786 mirror = argument(0);
3787 count_val = argument(1);
3788 }
3789
3790 mirror = null_check(mirror);
3791 // If mirror or obj is dead, only null-path is taken.
3792 if (stopped()) return true;
3793
3794 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3795 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3796 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3797 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3798 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3799
3800 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3801 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3802 result_reg, _slow_path);
3803 Node* normal_ctl = control();
3804 Node* no_array_ctl = result_reg->in(_slow_path);
3805
3806 // Generate code for the slow case. We make a call to newArray().
3807 set_control(no_array_ctl);
3808 if (!stopped()) {
3809 // Either the input type is void.class, or else the
3810 // array klass has not yet been cached. Either the
3811 // ensuing call will throw an exception, or else it
3812 // will cache the array klass for next time.
3813 PreserveJVMState pjvms(this);
3814 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3815 Node* slow_result = set_results_for_java_call(slow_call);
3816 // this->control() comes from set_results_for_java_call
3817 result_reg->set_req(_slow_path, control());
3818 result_val->set_req(_slow_path, slow_result);
3819 result_io ->set_req(_slow_path, i_o());
3820 result_mem->set_req(_slow_path, reset_memory());
3821 }
3822
3823 set_control(normal_ctl);
3824 if (!stopped()) {
3825 // Normal case: The array type has been cached in the java.lang.Class.
3826 // The following call works fine even if the array type is polymorphic.
3827 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3828 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3829 result_reg->init_req(_normal_path, control());
3830 result_val->init_req(_normal_path, obj);
3831 result_io ->init_req(_normal_path, i_o());
3832 result_mem->init_req(_normal_path, reset_memory());
3833
3834 if (uninitialized) {
3835 // Mark the allocation so that zeroing is skipped
3836 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3837 alloc->maybe_set_complete(&_gvn);
3838 }
3839 }
3840
3841 // Return the combined state.
3842 set_i_o( _gvn.transform(result_io) );
3843 set_all_memory( _gvn.transform(result_mem));
3844
3845 C->set_has_split_ifs(true); // Has chance for split-if optimization
3846 set_result(result_reg, result_val);
3847 return true;
3848 }
3849
3850 //----------------------inline_native_getLength--------------------------
3851 // public static native int java.lang.reflect.Array.getLength(Object array);
3852 bool LibraryCallKit::inline_native_getLength() {
3853 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3854
3855 Node* array = null_check(argument(0));
3856 // If array is dead, only null-path is taken.
3857 if (stopped()) return true;
3858
3859 // Deoptimize if it is a non-array.
3860 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3861
3862 if (non_array != NULL) {
3863 PreserveJVMState pjvms(this);
3864 set_control(non_array);
3865 uncommon_trap(Deoptimization::Reason_intrinsic,
3866 Deoptimization::Action_maybe_recompile);
3867 }
3868
3869 // If control is dead, only non-array-path is taken.
3870 if (stopped()) return true;
3871
3872 // The works fine even if the array type is polymorphic.
3873 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3874 Node* result = load_array_length(array);
3875
3876 C->set_has_split_ifs(true); // Has chance for split-if optimization
3877 set_result(result);
3878 return true;
3879 }
3880
3881 //------------------------inline_array_copyOf----------------------------
3882 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3883 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3884 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3885 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3886
3887 // Get the arguments.
3888 Node* original = argument(0);
3889 Node* start = is_copyOfRange? argument(1): intcon(0);
3890 Node* end = is_copyOfRange? argument(2): argument(1);
3891 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3892
3893 Node* newcopy = NULL;
3894
3895 // Set the original stack and the reexecute bit for the interpreter to reexecute
3896 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3897 { PreserveReexecuteState preexecs(this);
3898 jvms()->set_should_reexecute(true);
3899
3900 array_type_mirror = null_check(array_type_mirror);
3901 original = null_check(original);
3902
3903 // Check if a null path was taken unconditionally.
3904 if (stopped()) return true;
3905
3906 Node* orig_length = load_array_length(original);
3907
3908 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3909 klass_node = null_check(klass_node);
3910
3911 RegionNode* bailout = new RegionNode(1);
3912 record_for_igvn(bailout);
3913
3914 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3915 // Bail out if that is so.
3916 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3917 if (not_objArray != NULL) {
3918 // Improve the klass node's type from the new optimistic assumption:
3919 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3920 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3921 Node* cast = new CastPPNode(klass_node, akls);
3922 cast->init_req(0, control());
3923 klass_node = _gvn.transform(cast);
3924 }
3925
3926 // Bail out if either start or end is negative.
3927 generate_negative_guard(start, bailout, &start);
3928 generate_negative_guard(end, bailout, &end);
3929
3930 Node* length = end;
3931 if (_gvn.type(start) != TypeInt::ZERO) {
3932 length = _gvn.transform(new SubINode(end, start));
3933 }
3934
3935 // Bail out if length is negative.
3936 // Without this the new_array would throw
3937 // NegativeArraySizeException but IllegalArgumentException is what
3938 // should be thrown
3939 generate_negative_guard(length, bailout, &length);
3940
3941 if (bailout->req() > 1) {
3942 PreserveJVMState pjvms(this);
3943 set_control(_gvn.transform(bailout));
3944 uncommon_trap(Deoptimization::Reason_intrinsic,
3945 Deoptimization::Action_maybe_recompile);
3946 }
3947
3948 if (!stopped()) {
3949 // How many elements will we copy from the original?
3950 // The answer is MinI(orig_length - start, length).
3951 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3952 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3953
3954 // Generate a direct call to the right arraycopy function(s).
3955 // We know the copy is disjoint but we might not know if the
3956 // oop stores need checking.
3957 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3958 // This will fail a store-check if x contains any non-nulls.
3959
3960 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3961 // loads/stores but it is legal only if we're sure the
3962 // Arrays.copyOf would succeed. So we need all input arguments
3963 // to the copyOf to be validated, including that the copy to the
3964 // new array won't trigger an ArrayStoreException. That subtype
3965 // check can be optimized if we know something on the type of
3966 // the input array from type speculation.
3967 if (_gvn.type(klass_node)->singleton()) {
3968 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3969 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3970
3971 int test = C->static_subtype_check(superk, subk);
3972 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3973 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3974 if (t_original->speculative_type() != NULL) {
3975 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3976 }
3977 }
3978 }
3979
3980 bool validated = false;
3981 // Reason_class_check rather than Reason_intrinsic because we
3982 // want to intrinsify even if this traps.
3983 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3984 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3985 klass_node);
3986
3987 if (not_subtype_ctrl != top()) {
3988 PreserveJVMState pjvms(this);
3989 set_control(not_subtype_ctrl);
3990 uncommon_trap(Deoptimization::Reason_class_check,
3991 Deoptimization::Action_make_not_entrant);
3992 assert(stopped(), "Should be stopped");
3993 }
3994 validated = true;
3995 }
3996
3997 if (!stopped()) {
3998 newcopy = new_array(klass_node, length, 0); // no arguments to push
3999
4000 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
4001 load_object_klass(original), klass_node);
4002 if (!is_copyOfRange) {
4003 ac->set_copyof(validated);
4004 } else {
4005 ac->set_copyofrange(validated);
4006 }
4007 Node* n = _gvn.transform(ac);
4008 if (n == ac) {
4009 ac->connect_outputs(this);
4010 } else {
4011 assert(validated, "shouldn't transform if all arguments not validated");
4012 set_all_memory(n);
4013 }
4014 }
4015 }
4016 } // original reexecute is set back here
4017
4018 C->set_has_split_ifs(true); // Has chance for split-if optimization
4019 if (!stopped()) {
4020 set_result(newcopy);
4021 }
4022 return true;
4023 }
4024
4025
4026 //----------------------generate_virtual_guard---------------------------
4027 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4028 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4029 RegionNode* slow_region) {
4030 ciMethod* method = callee();
4031 int vtable_index = method->vtable_index();
4032 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4033 "bad index %d", vtable_index);
4034 // Get the Method* out of the appropriate vtable entry.
4035 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4036 vtable_index*vtableEntry::size_in_bytes() +
4037 vtableEntry::method_offset_in_bytes();
4038 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4039 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4040
4041 // Compare the target method with the expected method (e.g., Object.hashCode).
4042 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4043
4044 Node* native_call = makecon(native_call_addr);
4045 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4046 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4047
4048 return generate_slow_guard(test_native, slow_region);
4049 }
4050
4051 //-----------------------generate_method_call----------------------------
4052 // Use generate_method_call to make a slow-call to the real
4053 // method if the fast path fails. An alternative would be to
4054 // use a stub like OptoRuntime::slow_arraycopy_Java.
4055 // This only works for expanding the current library call,
4056 // not another intrinsic. (E.g., don't use this for making an
4057 // arraycopy call inside of the copyOf intrinsic.)
4058 CallJavaNode*
4059 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4060 // When compiling the intrinsic method itself, do not use this technique.
4061 guarantee(callee() != C->method(), "cannot make slow-call to self");
4062
4063 ciMethod* method = callee();
4064 // ensure the JVMS we have will be correct for this call
4065 guarantee(method_id == method->intrinsic_id(), "must match");
4066
4067 const TypeFunc* tf = TypeFunc::make(method);
4068 CallJavaNode* slow_call;
4069 if (is_static) {
4070 assert(!is_virtual, "");
4071 slow_call = new CallStaticJavaNode(C, tf,
4072 SharedRuntime::get_resolve_static_call_stub(),
4073 method, bci());
4074 } else if (is_virtual) {
4075 null_check_receiver();
4076 int vtable_index = Method::invalid_vtable_index;
4077 if (UseInlineCaches) {
4078 // Suppress the vtable call
4079 } else {
4080 // hashCode and clone are not a miranda methods,
4081 // so the vtable index is fixed.
4082 // No need to use the linkResolver to get it.
4083 vtable_index = method->vtable_index();
4084 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4085 "bad index %d", vtable_index);
4086 }
4087 slow_call = new CallDynamicJavaNode(tf,
4088 SharedRuntime::get_resolve_virtual_call_stub(),
4089 method, vtable_index, bci());
4090 } else { // neither virtual nor static: opt_virtual
4091 null_check_receiver();
4092 slow_call = new CallStaticJavaNode(C, tf,
4093 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4094 method, bci());
4095 slow_call->set_optimized_virtual(true);
4096 }
4097 set_arguments_for_java_call(slow_call);
4098 set_edges_for_java_call(slow_call);
4099 return slow_call;
4100 }
4101
4102
4103 /**
4104 * Build special case code for calls to hashCode on an object. This call may
4105 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4106 * slightly different code.
4107 */
4108 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4109 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4110 assert(!(is_virtual && is_static), "either virtual, special, or static");
4111
4112 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4113
4114 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4115 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4116 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4117 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4118 Node* obj = NULL;
4119 if (!is_static) {
4120 // Check for hashing null object
4121 obj = null_check_receiver();
4122 if (stopped()) return true; // unconditionally null
4123 result_reg->init_req(_null_path, top());
4124 result_val->init_req(_null_path, top());
4125 } else {
4126 // Do a null check, and return zero if null.
4127 // System.identityHashCode(null) == 0
4128 obj = argument(0);
4129 Node* null_ctl = top();
4130 obj = null_check_oop(obj, &null_ctl);
4131 result_reg->init_req(_null_path, null_ctl);
4132 result_val->init_req(_null_path, _gvn.intcon(0));
4133 }
4134
4135 // Unconditionally null? Then return right away.
4136 if (stopped()) {
4137 set_control( result_reg->in(_null_path));
4138 if (!stopped())
4139 set_result(result_val->in(_null_path));
4140 return true;
4141 }
4142
4143 // We only go to the fast case code if we pass a number of guards. The
4144 // paths which do not pass are accumulated in the slow_region.
4145 RegionNode* slow_region = new RegionNode(1);
4146 record_for_igvn(slow_region);
4147
4148 // If this is a virtual call, we generate a funny guard. We pull out
4149 // the vtable entry corresponding to hashCode() from the target object.
4150 // If the target method which we are calling happens to be the native
4151 // Object hashCode() method, we pass the guard. We do not need this
4152 // guard for non-virtual calls -- the caller is known to be the native
4153 // Object hashCode().
4154 if (is_virtual) {
4155 // After null check, get the object's klass.
4156 Node* obj_klass = load_object_klass(obj);
4157 generate_virtual_guard(obj_klass, slow_region);
4158 }
4159
4160 // Get the header out of the object, use LoadMarkNode when available
4161 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4162 // The control of the load must be NULL. Otherwise, the load can move before
4163 // the null check after castPP removal.
4164 Node* no_ctrl = NULL;
4165 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4166
4167 // Test the header to see if it is unlocked.
4168 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4169 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4170 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4171 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4172 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4173
4174 generate_slow_guard(test_unlocked, slow_region);
4175
4176 // Get the hash value and check to see that it has been properly assigned.
4177 // We depend on hash_mask being at most 32 bits and avoid the use of
4178 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4179 // vm: see markOop.hpp.
4180 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4181 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4182 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4183 // This hack lets the hash bits live anywhere in the mark object now, as long
4184 // as the shift drops the relevant bits into the low 32 bits. Note that
4185 // Java spec says that HashCode is an int so there's no point in capturing
4186 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4187 hshifted_header = ConvX2I(hshifted_header);
4188 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4189
4190 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4191 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4192 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4193
4194 generate_slow_guard(test_assigned, slow_region);
4195
4196 Node* init_mem = reset_memory();
4197 // fill in the rest of the null path:
4198 result_io ->init_req(_null_path, i_o());
4199 result_mem->init_req(_null_path, init_mem);
4200
4201 result_val->init_req(_fast_path, hash_val);
4202 result_reg->init_req(_fast_path, control());
4203 result_io ->init_req(_fast_path, i_o());
4204 result_mem->init_req(_fast_path, init_mem);
4205
4206 // Generate code for the slow case. We make a call to hashCode().
4207 set_control(_gvn.transform(slow_region));
4208 if (!stopped()) {
4209 // No need for PreserveJVMState, because we're using up the present state.
4210 set_all_memory(init_mem);
4211 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4212 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4213 Node* slow_result = set_results_for_java_call(slow_call);
4214 // this->control() comes from set_results_for_java_call
4215 result_reg->init_req(_slow_path, control());
4216 result_val->init_req(_slow_path, slow_result);
4217 result_io ->set_req(_slow_path, i_o());
4218 result_mem ->set_req(_slow_path, reset_memory());
4219 }
4220
4221 // Return the combined state.
4222 set_i_o( _gvn.transform(result_io) );
4223 set_all_memory( _gvn.transform(result_mem));
4224
4225 set_result(result_reg, result_val);
4226 return true;
4227 }
4228
4229 //---------------------------inline_native_getClass----------------------------
4230 // public final native Class<?> java.lang.Object.getClass();
4231 //
4232 // Build special case code for calls to getClass on an object.
4233 bool LibraryCallKit::inline_native_getClass() {
4234 Node* obj = null_check_receiver();
4235 if (stopped()) return true;
4236 set_result(load_mirror_from_klass(load_object_klass(obj)));
4237 return true;
4238 }
4239
4240 //-----------------inline_native_Reflection_getCallerClass---------------------
4241 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4242 //
4243 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4244 //
4245 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4246 // in that it must skip particular security frames and checks for
4247 // caller sensitive methods.
4248 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4249 #ifndef PRODUCT
4250 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4251 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4252 }
4253 #endif
4254
4255 if (!jvms()->has_method()) {
4256 #ifndef PRODUCT
4257 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4258 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4259 }
4260 #endif
4261 return false;
4262 }
4263
4264 // Walk back up the JVM state to find the caller at the required
4265 // depth.
4266 JVMState* caller_jvms = jvms();
4267
4268 // Cf. JVM_GetCallerClass
4269 // NOTE: Start the loop at depth 1 because the current JVM state does
4270 // not include the Reflection.getCallerClass() frame.
4271 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4272 ciMethod* m = caller_jvms->method();
4273 switch (n) {
4274 case 0:
4275 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4276 break;
4277 case 1:
4278 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4279 if (!m->caller_sensitive()) {
4280 #ifndef PRODUCT
4281 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4282 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4283 }
4284 #endif
4285 return false; // bail-out; let JVM_GetCallerClass do the work
4286 }
4287 break;
4288 default:
4289 if (!m->is_ignored_by_security_stack_walk()) {
4290 // We have reached the desired frame; return the holder class.
4291 // Acquire method holder as java.lang.Class and push as constant.
4292 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4293 ciInstance* caller_mirror = caller_klass->java_mirror();
4294 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4295
4296 #ifndef PRODUCT
4297 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4298 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());
4299 tty->print_cr(" JVM state at this point:");
4300 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4301 ciMethod* m = jvms()->of_depth(i)->method();
4302 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4303 }
4304 }
4305 #endif
4306 return true;
4307 }
4308 break;
4309 }
4310 }
4311
4312 #ifndef PRODUCT
4313 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4314 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4315 tty->print_cr(" JVM state at this point:");
4316 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4317 ciMethod* m = jvms()->of_depth(i)->method();
4318 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4319 }
4320 }
4321 #endif
4322
4323 return false; // bail-out; let JVM_GetCallerClass do the work
4324 }
4325
4326 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4327 Node* arg = argument(0);
4328 Node* result = NULL;
4329
4330 switch (id) {
4331 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4332 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4333 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4334 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4335
4336 case vmIntrinsics::_doubleToLongBits: {
4337 // two paths (plus control) merge in a wood
4338 RegionNode *r = new RegionNode(3);
4339 Node *phi = new PhiNode(r, TypeLong::LONG);
4340
4341 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4342 // Build the boolean node
4343 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4344
4345 // Branch either way.
4346 // NaN case is less traveled, which makes all the difference.
4347 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4348 Node *opt_isnan = _gvn.transform(ifisnan);
4349 assert( opt_isnan->is_If(), "Expect an IfNode");
4350 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4351 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4352
4353 set_control(iftrue);
4354
4355 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4356 Node *slow_result = longcon(nan_bits); // return NaN
4357 phi->init_req(1, _gvn.transform( slow_result ));
4358 r->init_req(1, iftrue);
4359
4360 // Else fall through
4361 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4362 set_control(iffalse);
4363
4364 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4365 r->init_req(2, iffalse);
4366
4367 // Post merge
4368 set_control(_gvn.transform(r));
4369 record_for_igvn(r);
4370
4371 C->set_has_split_ifs(true); // Has chance for split-if optimization
4372 result = phi;
4373 assert(result->bottom_type()->isa_long(), "must be");
4374 break;
4375 }
4376
4377 case vmIntrinsics::_floatToIntBits: {
4378 // two paths (plus control) merge in a wood
4379 RegionNode *r = new RegionNode(3);
4380 Node *phi = new PhiNode(r, TypeInt::INT);
4381
4382 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4383 // Build the boolean node
4384 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4385
4386 // Branch either way.
4387 // NaN case is less traveled, which makes all the difference.
4388 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4389 Node *opt_isnan = _gvn.transform(ifisnan);
4390 assert( opt_isnan->is_If(), "Expect an IfNode");
4391 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4392 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4393
4394 set_control(iftrue);
4395
4396 static const jint nan_bits = 0x7fc00000;
4397 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4398 phi->init_req(1, _gvn.transform( slow_result ));
4399 r->init_req(1, iftrue);
4400
4401 // Else fall through
4402 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4403 set_control(iffalse);
4404
4405 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4406 r->init_req(2, iffalse);
4407
4408 // Post merge
4409 set_control(_gvn.transform(r));
4410 record_for_igvn(r);
4411
4412 C->set_has_split_ifs(true); // Has chance for split-if optimization
4413 result = phi;
4414 assert(result->bottom_type()->isa_int(), "must be");
4415 break;
4416 }
4417
4418 default:
4419 fatal_unexpected_iid(id);
4420 break;
4421 }
4422 set_result(_gvn.transform(result));
4423 return true;
4424 }
4425
4426 //----------------------inline_unsafe_copyMemory-------------------------
4427 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4428 bool LibraryCallKit::inline_unsafe_copyMemory() {
4429 if (callee()->is_static()) return false; // caller must have the capability!
4430 null_check_receiver(); // null-check receiver
4431 if (stopped()) return true;
4432
4433 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4434
4435 Node* src_ptr = argument(1); // type: oop
4436 Node* src_off = ConvL2X(argument(2)); // type: long
4437 Node* dst_ptr = argument(4); // type: oop
4438 Node* dst_off = ConvL2X(argument(5)); // type: long
4439 Node* size = ConvL2X(argument(7)); // type: long
4440
4441 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4442 "fieldOffset must be byte-scaled");
4443
4444 Node* src = make_unsafe_address(src_ptr, src_off);
4445 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4446
4447 // Conservatively insert a memory barrier on all memory slices.
4448 // Do not let writes of the copy source or destination float below the copy.
4449 insert_mem_bar(Op_MemBarCPUOrder);
4450
4451 // Call it. Note that the length argument is not scaled.
4452 make_runtime_call(RC_LEAF|RC_NO_FP,
4453 OptoRuntime::fast_arraycopy_Type(),
4454 StubRoutines::unsafe_arraycopy(),
4455 "unsafe_arraycopy",
4456 TypeRawPtr::BOTTOM,
4457 src, dst, size XTOP);
4458
4459 // Do not let reads of the copy destination float above the copy.
4460 insert_mem_bar(Op_MemBarCPUOrder);
4461
4462 return true;
4463 }
4464
4465 //------------------------clone_coping-----------------------------------
4466 // Helper function for inline_native_clone.
4467 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4468 assert(obj_size != NULL, "");
4469 Node* raw_obj = alloc_obj->in(1);
4470 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4471
4472 AllocateNode* alloc = NULL;
4473 if (ReduceBulkZeroing) {
4474 // We will be completely responsible for initializing this object -
4475 // mark Initialize node as complete.
4476 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4477 // The object was just allocated - there should be no any stores!
4478 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4479 // Mark as complete_with_arraycopy so that on AllocateNode
4480 // expansion, we know this AllocateNode is initialized by an array
4481 // copy and a StoreStore barrier exists after the array copy.
4482 alloc->initialization()->set_complete_with_arraycopy();
4483 }
4484
4485 // Copy the fastest available way.
4486 // TODO: generate fields copies for small objects instead.
4487 Node* src = obj;
4488 Node* dest = alloc_obj;
4489 Node* size = _gvn.transform(obj_size);
4490
4491 // Exclude the header but include array length to copy by 8 bytes words.
4492 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4493 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4494 instanceOopDesc::base_offset_in_bytes();
4495 // base_off:
4496 // 8 - 32-bit VM
4497 // 12 - 64-bit VM, compressed klass
4498 // 16 - 64-bit VM, normal klass
4499 if (base_off % BytesPerLong != 0) {
4500 assert(UseCompressedClassPointers, "");
4501 if (is_array) {
4502 // Exclude length to copy by 8 bytes words.
4503 base_off += sizeof(int);
4504 } else {
4505 // Include klass to copy by 8 bytes words.
4506 base_off = instanceOopDesc::klass_offset_in_bytes();
4507 }
4508 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4509 }
4510 src = basic_plus_adr(src, base_off);
4511 dest = basic_plus_adr(dest, base_off);
4512
4513 // Compute the length also, if needed:
4514 Node* countx = size;
4515 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4516 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4517
4518 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4519
4520 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4521 ac->set_clonebasic();
4522 Node* n = _gvn.transform(ac);
4523 if (n == ac) {
4524 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4525 } else {
4526 set_all_memory(n);
4527 }
4528
4529 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4530 if (card_mark) {
4531 assert(!is_array, "");
4532 // Put in store barrier for any and all oops we are sticking
4533 // into this object. (We could avoid this if we could prove
4534 // that the object type contains no oop fields at all.)
4535 Node* no_particular_value = NULL;
4536 Node* no_particular_field = NULL;
4537 int raw_adr_idx = Compile::AliasIdxRaw;
4538 post_barrier(control(),
4539 memory(raw_adr_type),
4540 alloc_obj,
4541 no_particular_field,
4542 raw_adr_idx,
4543 no_particular_value,
4544 T_OBJECT,
4545 false);
4546 }
4547
4548 // Do not let reads from the cloned object float above the arraycopy.
4549 if (alloc != NULL) {
4550 // Do not let stores that initialize this object be reordered with
4551 // a subsequent store that would make this object accessible by
4552 // other threads.
4553 // Record what AllocateNode this StoreStore protects so that
4554 // escape analysis can go from the MemBarStoreStoreNode to the
4555 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4556 // based on the escape status of the AllocateNode.
4557 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4558 } else {
4559 insert_mem_bar(Op_MemBarCPUOrder);
4560 }
4561 }
4562
4563 //------------------------inline_native_clone----------------------------
4564 // protected native Object java.lang.Object.clone();
4565 //
4566 // Here are the simple edge cases:
4567 // null receiver => normal trap
4568 // virtual and clone was overridden => slow path to out-of-line clone
4569 // not cloneable or finalizer => slow path to out-of-line Object.clone
4570 //
4571 // The general case has two steps, allocation and copying.
4572 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4573 //
4574 // Copying also has two cases, oop arrays and everything else.
4575 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4576 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4577 //
4578 // These steps fold up nicely if and when the cloned object's klass
4579 // can be sharply typed as an object array, a type array, or an instance.
4580 //
4581 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4582 PhiNode* result_val;
4583
4584 // Set the reexecute bit for the interpreter to reexecute
4585 // the bytecode that invokes Object.clone if deoptimization happens.
4586 { PreserveReexecuteState preexecs(this);
4587 jvms()->set_should_reexecute(true);
4588
4589 Node* obj = null_check_receiver();
4590 if (stopped()) return true;
4591
4592 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4593
4594 // If we are going to clone an instance, we need its exact type to
4595 // know the number and types of fields to convert the clone to
4596 // loads/stores. Maybe a speculative type can help us.
4597 if (!obj_type->klass_is_exact() &&
4598 obj_type->speculative_type() != NULL &&
4599 obj_type->speculative_type()->is_instance_klass()) {
4600 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4601 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4602 !spec_ik->has_injected_fields()) {
4603 ciKlass* k = obj_type->klass();
4604 if (!k->is_instance_klass() ||
4605 k->as_instance_klass()->is_interface() ||
4606 k->as_instance_klass()->has_subklass()) {
4607 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4608 }
4609 }
4610 }
4611
4612 Node* obj_klass = load_object_klass(obj);
4613 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4614 const TypeOopPtr* toop = ((tklass != NULL)
4615 ? tklass->as_instance_type()
4616 : TypeInstPtr::NOTNULL);
4617
4618 // Conservatively insert a memory barrier on all memory slices.
4619 // Do not let writes into the original float below the clone.
4620 insert_mem_bar(Op_MemBarCPUOrder);
4621
4622 // paths into result_reg:
4623 enum {
4624 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4625 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4626 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4627 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4628 PATH_LIMIT
4629 };
4630 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4631 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4632 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4633 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4634 record_for_igvn(result_reg);
4635
4636 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4637 int raw_adr_idx = Compile::AliasIdxRaw;
4638
4639 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4640 if (array_ctl != NULL) {
4641 // It's an array.
4642 PreserveJVMState pjvms(this);
4643 set_control(array_ctl);
4644 Node* obj_length = load_array_length(obj);
4645 Node* obj_size = NULL;
4646 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4647
4648 if (!use_ReduceInitialCardMarks()) {
4649 // If it is an oop array, it requires very special treatment,
4650 // because card marking is required on each card of the array.
4651 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4652 if (is_obja != NULL) {
4653 PreserveJVMState pjvms2(this);
4654 set_control(is_obja);
4655 // Generate a direct call to the right arraycopy function(s).
4656 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4657 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4658 ac->set_cloneoop();
4659 Node* n = _gvn.transform(ac);
4660 assert(n == ac, "cannot disappear");
4661 ac->connect_outputs(this);
4662
4663 result_reg->init_req(_objArray_path, control());
4664 result_val->init_req(_objArray_path, alloc_obj);
4665 result_i_o ->set_req(_objArray_path, i_o());
4666 result_mem ->set_req(_objArray_path, reset_memory());
4667 }
4668 }
4669 // Otherwise, there are no card marks to worry about.
4670 // (We can dispense with card marks if we know the allocation
4671 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4672 // causes the non-eden paths to take compensating steps to
4673 // simulate a fresh allocation, so that no further
4674 // card marks are required in compiled code to initialize
4675 // the object.)
4676
4677 if (!stopped()) {
4678 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4679
4680 // Present the results of the copy.
4681 result_reg->init_req(_array_path, control());
4682 result_val->init_req(_array_path, alloc_obj);
4683 result_i_o ->set_req(_array_path, i_o());
4684 result_mem ->set_req(_array_path, reset_memory());
4685 }
4686 }
4687
4688 // We only go to the instance fast case code if we pass a number of guards.
4689 // The paths which do not pass are accumulated in the slow_region.
4690 RegionNode* slow_region = new RegionNode(1);
4691 record_for_igvn(slow_region);
4692 if (!stopped()) {
4693 // It's an instance (we did array above). Make the slow-path tests.
4694 // If this is a virtual call, we generate a funny guard. We grab
4695 // the vtable entry corresponding to clone() from the target object.
4696 // If the target method which we are calling happens to be the
4697 // Object clone() method, we pass the guard. We do not need this
4698 // guard for non-virtual calls; the caller is known to be the native
4699 // Object clone().
4700 if (is_virtual) {
4701 generate_virtual_guard(obj_klass, slow_region);
4702 }
4703
4704 // The object must be easily cloneable and must not have a finalizer.
4705 // Both of these conditions may be checked in a single test.
4706 // We could optimize the test further, but we don't care.
4707 generate_access_flags_guard(obj_klass,
4708 // Test both conditions:
4709 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4710 // Must be cloneable but not finalizer:
4711 JVM_ACC_IS_CLONEABLE_FAST,
4712 slow_region);
4713 }
4714
4715 if (!stopped()) {
4716 // It's an instance, and it passed the slow-path tests.
4717 PreserveJVMState pjvms(this);
4718 Node* obj_size = NULL;
4719 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4720 // is reexecuted if deoptimization occurs and there could be problems when merging
4721 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4722 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4723
4724 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4725
4726 // Present the results of the slow call.
4727 result_reg->init_req(_instance_path, control());
4728 result_val->init_req(_instance_path, alloc_obj);
4729 result_i_o ->set_req(_instance_path, i_o());
4730 result_mem ->set_req(_instance_path, reset_memory());
4731 }
4732
4733 // Generate code for the slow case. We make a call to clone().
4734 set_control(_gvn.transform(slow_region));
4735 if (!stopped()) {
4736 PreserveJVMState pjvms(this);
4737 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4738 Node* slow_result = set_results_for_java_call(slow_call);
4739 // this->control() comes from set_results_for_java_call
4740 result_reg->init_req(_slow_path, control());
4741 result_val->init_req(_slow_path, slow_result);
4742 result_i_o ->set_req(_slow_path, i_o());
4743 result_mem ->set_req(_slow_path, reset_memory());
4744 }
4745
4746 // Return the combined state.
4747 set_control( _gvn.transform(result_reg));
4748 set_i_o( _gvn.transform(result_i_o));
4749 set_all_memory( _gvn.transform(result_mem));
4750 } // original reexecute is set back here
4751
4752 set_result(_gvn.transform(result_val));
4753 return true;
4754 }
4755
4756 // If we have a tighly coupled allocation, the arraycopy may take care
4757 // of the array initialization. If one of the guards we insert between
4758 // the allocation and the arraycopy causes a deoptimization, an
4759 // unitialized array will escape the compiled method. To prevent that
4760 // we set the JVM state for uncommon traps between the allocation and
4761 // the arraycopy to the state before the allocation so, in case of
4762 // deoptimization, we'll reexecute the allocation and the
4763 // initialization.
4764 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4765 if (alloc != NULL) {
4766 ciMethod* trap_method = alloc->jvms()->method();
4767 int trap_bci = alloc->jvms()->bci();
4768
4769 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4770 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4771 // Make sure there's no store between the allocation and the
4772 // arraycopy otherwise visible side effects could be rexecuted
4773 // in case of deoptimization and cause incorrect execution.
4774 bool no_interfering_store = true;
4775 Node* mem = alloc->in(TypeFunc::Memory);
4776 if (mem->is_MergeMem()) {
4777 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4778 Node* n = mms.memory();
4779 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4780 assert(n->is_Store(), "what else?");
4781 no_interfering_store = false;
4782 break;
4783 }
4784 }
4785 } else {
4786 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4787 Node* n = mms.memory();
4788 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4789 assert(n->is_Store(), "what else?");
4790 no_interfering_store = false;
4791 break;
4792 }
4793 }
4794 }
4795
4796 if (no_interfering_store) {
4797 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4798 uint size = alloc->req();
4799 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4800 old_jvms->set_map(sfpt);
4801 for (uint i = 0; i < size; i++) {
4802 sfpt->init_req(i, alloc->in(i));
4803 }
4804 // re-push array length for deoptimization
4805 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4806 old_jvms->set_sp(old_jvms->sp()+1);
4807 old_jvms->set_monoff(old_jvms->monoff()+1);
4808 old_jvms->set_scloff(old_jvms->scloff()+1);
4809 old_jvms->set_endoff(old_jvms->endoff()+1);
4810 old_jvms->set_should_reexecute(true);
4811
4812 sfpt->set_i_o(map()->i_o());
4813 sfpt->set_memory(map()->memory());
4814 sfpt->set_control(map()->control());
4815
4816 JVMState* saved_jvms = jvms();
4817 saved_reexecute_sp = _reexecute_sp;
4818
4819 set_jvms(sfpt->jvms());
4820 _reexecute_sp = jvms()->sp();
4821
4822 return saved_jvms;
4823 }
4824 }
4825 }
4826 return NULL;
4827 }
4828
4829 // In case of a deoptimization, we restart execution at the
4830 // allocation, allocating a new array. We would leave an uninitialized
4831 // array in the heap that GCs wouldn't expect. Move the allocation
4832 // after the traps so we don't allocate the array if we
4833 // deoptimize. This is possible because tightly_coupled_allocation()
4834 // guarantees there's no observer of the allocated array at this point
4835 // and the control flow is simple enough.
4836 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4837 if (saved_jvms != NULL && !stopped()) {
4838 assert(alloc != NULL, "only with a tightly coupled allocation");
4839 // restore JVM state to the state at the arraycopy
4840 saved_jvms->map()->set_control(map()->control());
4841 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4842 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4843 // If we've improved the types of some nodes (null check) while
4844 // emitting the guards, propagate them to the current state
4845 map()->replaced_nodes().apply(saved_jvms->map());
4846 set_jvms(saved_jvms);
4847 _reexecute_sp = saved_reexecute_sp;
4848
4849 // Remove the allocation from above the guards
4850 CallProjections callprojs;
4851 alloc->extract_projections(&callprojs, true);
4852 InitializeNode* init = alloc->initialization();
4853 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4854 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4855 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4856 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4857
4858 // move the allocation here (after the guards)
4859 _gvn.hash_delete(alloc);
4860 alloc->set_req(TypeFunc::Control, control());
4861 alloc->set_req(TypeFunc::I_O, i_o());
4862 Node *mem = reset_memory();
4863 set_all_memory(mem);
4864 alloc->set_req(TypeFunc::Memory, mem);
4865 set_control(init->proj_out(TypeFunc::Control));
4866 set_i_o(callprojs.fallthrough_ioproj);
4867
4868 // Update memory as done in GraphKit::set_output_for_allocation()
4869 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4870 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4871 if (ary_type->isa_aryptr() && length_type != NULL) {
4872 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4873 }
4874 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4875 int elemidx = C->get_alias_index(telemref);
4876 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4877 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4878
4879 Node* allocx = _gvn.transform(alloc);
4880 assert(allocx == alloc, "where has the allocation gone?");
4881 assert(dest->is_CheckCastPP(), "not an allocation result?");
4882
4883 _gvn.hash_delete(dest);
4884 dest->set_req(0, control());
4885 Node* destx = _gvn.transform(dest);
4886 assert(destx == dest, "where has the allocation result gone?");
4887 }
4888 }
4889
4890
4891 //------------------------------inline_arraycopy-----------------------
4892 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4893 // Object dest, int destPos,
4894 // int length);
4895 bool LibraryCallKit::inline_arraycopy() {
4896 // Get the arguments.
4897 Node* src = argument(0); // type: oop
4898 Node* src_offset = argument(1); // type: int
4899 Node* dest = argument(2); // type: oop
4900 Node* dest_offset = argument(3); // type: int
4901 Node* length = argument(4); // type: int
4902
4903
4904 // Check for allocation before we add nodes that would confuse
4905 // tightly_coupled_allocation()
4906 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4907
4908 int saved_reexecute_sp = -1;
4909 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4910 // See arraycopy_restore_alloc_state() comment
4911 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4912 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4913 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4914 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4915
4916 // The following tests must be performed
4917 // (1) src and dest are arrays.
4918 // (2) src and dest arrays must have elements of the same BasicType
4919 // (3) src and dest must not be null.
4920 // (4) src_offset must not be negative.
4921 // (5) dest_offset must not be negative.
4922 // (6) length must not be negative.
4923 // (7) src_offset + length must not exceed length of src.
4924 // (8) dest_offset + length must not exceed length of dest.
4925 // (9) each element of an oop array must be assignable
4926
4927 // (3) src and dest must not be null.
4928 // always do this here because we need the JVM state for uncommon traps
4929 Node* null_ctl = top();
4930 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4931 assert(null_ctl->is_top(), "no null control here");
4932 dest = null_check(dest, T_ARRAY);
4933
4934 if (!can_emit_guards) {
4935 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4936 // guards but the arraycopy node could still take advantage of a
4937 // tightly allocated allocation. tightly_coupled_allocation() is
4938 // called again to make sure it takes the null check above into
4939 // account: the null check is mandatory and if it caused an
4940 // uncommon trap to be emitted then the allocation can't be
4941 // considered tightly coupled in this context.
4942 alloc = tightly_coupled_allocation(dest, NULL);
4943 }
4944
4945 bool validated = false;
4946
4947 const Type* src_type = _gvn.type(src);
4948 const Type* dest_type = _gvn.type(dest);
4949 const TypeAryPtr* top_src = src_type->isa_aryptr();
4950 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4951
4952 // Do we have the type of src?
4953 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4954 // Do we have the type of dest?
4955 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4956 // Is the type for src from speculation?
4957 bool src_spec = false;
4958 // Is the type for dest from speculation?
4959 bool dest_spec = false;
4960
4961 if ((!has_src || !has_dest) && can_emit_guards) {
4962 // We don't have sufficient type information, let's see if
4963 // speculative types can help. We need to have types for both src
4964 // and dest so that it pays off.
4965
4966 // Do we already have or could we have type information for src
4967 bool could_have_src = has_src;
4968 // Do we already have or could we have type information for dest
4969 bool could_have_dest = has_dest;
4970
4971 ciKlass* src_k = NULL;
4972 if (!has_src) {
4973 src_k = src_type->speculative_type_not_null();
4974 if (src_k != NULL && src_k->is_array_klass()) {
4975 could_have_src = true;
4976 }
4977 }
4978
4979 ciKlass* dest_k = NULL;
4980 if (!has_dest) {
4981 dest_k = dest_type->speculative_type_not_null();
4982 if (dest_k != NULL && dest_k->is_array_klass()) {
4983 could_have_dest = true;
4984 }
4985 }
4986
4987 if (could_have_src && could_have_dest) {
4988 // This is going to pay off so emit the required guards
4989 if (!has_src) {
4990 src = maybe_cast_profiled_obj(src, src_k, true);
4991 src_type = _gvn.type(src);
4992 top_src = src_type->isa_aryptr();
4993 has_src = (top_src != NULL && top_src->klass() != NULL);
4994 src_spec = true;
4995 }
4996 if (!has_dest) {
4997 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4998 dest_type = _gvn.type(dest);
4999 top_dest = dest_type->isa_aryptr();
5000 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5001 dest_spec = true;
5002 }
5003 }
5004 }
5005
5006 if (has_src && has_dest && can_emit_guards) {
5007 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5008 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5009 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5010 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5011
5012 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5013 // If both arrays are object arrays then having the exact types
5014 // for both will remove the need for a subtype check at runtime
5015 // before the call and may make it possible to pick a faster copy
5016 // routine (without a subtype check on every element)
5017 // Do we have the exact type of src?
5018 bool could_have_src = src_spec;
5019 // Do we have the exact type of dest?
5020 bool could_have_dest = dest_spec;
5021 ciKlass* src_k = top_src->klass();
5022 ciKlass* dest_k = top_dest->klass();
5023 if (!src_spec) {
5024 src_k = src_type->speculative_type_not_null();
5025 if (src_k != NULL && src_k->is_array_klass()) {
5026 could_have_src = true;
5027 }
5028 }
5029 if (!dest_spec) {
5030 dest_k = dest_type->speculative_type_not_null();
5031 if (dest_k != NULL && dest_k->is_array_klass()) {
5032 could_have_dest = true;
5033 }
5034 }
5035 if (could_have_src && could_have_dest) {
5036 // If we can have both exact types, emit the missing guards
5037 if (could_have_src && !src_spec) {
5038 src = maybe_cast_profiled_obj(src, src_k, true);
5039 }
5040 if (could_have_dest && !dest_spec) {
5041 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5042 }
5043 }
5044 }
5045 }
5046
5047 ciMethod* trap_method = method();
5048 int trap_bci = bci();
5049 if (saved_jvms != NULL) {
5050 trap_method = alloc->jvms()->method();
5051 trap_bci = alloc->jvms()->bci();
5052 }
5053
5054 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5055 can_emit_guards &&
5056 !src->is_top() && !dest->is_top()) {
5057 // validate arguments: enables transformation the ArrayCopyNode
5058 validated = true;
5059
5060 RegionNode* slow_region = new RegionNode(1);
5061 record_for_igvn(slow_region);
5062
5063 // (1) src and dest are arrays.
5064 generate_non_array_guard(load_object_klass(src), slow_region);
5065 generate_non_array_guard(load_object_klass(dest), slow_region);
5066
5067 // (2) src and dest arrays must have elements of the same BasicType
5068 // done at macro expansion or at Ideal transformation time
5069
5070 // (4) src_offset must not be negative.
5071 generate_negative_guard(src_offset, slow_region);
5072
5073 // (5) dest_offset must not be negative.
5074 generate_negative_guard(dest_offset, slow_region);
5075
5076 // (7) src_offset + length must not exceed length of src.
5077 generate_limit_guard(src_offset, length,
5078 load_array_length(src),
5079 slow_region);
5080
5081 // (8) dest_offset + length must not exceed length of dest.
5082 generate_limit_guard(dest_offset, length,
5083 load_array_length(dest),
5084 slow_region);
5085
5086 // (9) each element of an oop array must be assignable
5087 Node* src_klass = load_object_klass(src);
5088 Node* dest_klass = load_object_klass(dest);
5089 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5090
5091 if (not_subtype_ctrl != top()) {
5092 PreserveJVMState pjvms(this);
5093 set_control(not_subtype_ctrl);
5094 uncommon_trap(Deoptimization::Reason_intrinsic,
5095 Deoptimization::Action_make_not_entrant);
5096 assert(stopped(), "Should be stopped");
5097 }
5098 {
5099 PreserveJVMState pjvms(this);
5100 set_control(_gvn.transform(slow_region));
5101 uncommon_trap(Deoptimization::Reason_intrinsic,
5102 Deoptimization::Action_make_not_entrant);
5103 assert(stopped(), "Should be stopped");
5104 }
5105 }
5106
5107 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5108
5109 if (stopped()) {
5110 return true;
5111 }
5112
5113 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5114 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5115 // so the compiler has a chance to eliminate them: during macro expansion,
5116 // we have to set their control (CastPP nodes are eliminated).
5117 load_object_klass(src), load_object_klass(dest),
5118 load_array_length(src), load_array_length(dest));
5119
5120 ac->set_arraycopy(validated);
5121
5122 Node* n = _gvn.transform(ac);
5123 if (n == ac) {
5124 ac->connect_outputs(this);
5125 } else {
5126 assert(validated, "shouldn't transform if all arguments not validated");
5127 set_all_memory(n);
5128 }
5129
5130 return true;
5131 }
5132
5133
5134 // Helper function which determines if an arraycopy immediately follows
5135 // an allocation, with no intervening tests or other escapes for the object.
5136 AllocateArrayNode*
5137 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5138 RegionNode* slow_region) {
5139 if (stopped()) return NULL; // no fast path
5140 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5141
5142 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5143 if (alloc == NULL) return NULL;
5144
5145 Node* rawmem = memory(Compile::AliasIdxRaw);
5146 // Is the allocation's memory state untouched?
5147 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5148 // Bail out if there have been raw-memory effects since the allocation.
5149 // (Example: There might have been a call or safepoint.)
5150 return NULL;
5151 }
5152 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5153 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5154 return NULL;
5155 }
5156
5157 // There must be no unexpected observers of this allocation.
5158 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5159 Node* obs = ptr->fast_out(i);
5160 if (obs != this->map()) {
5161 return NULL;
5162 }
5163 }
5164
5165 // This arraycopy must unconditionally follow the allocation of the ptr.
5166 Node* alloc_ctl = ptr->in(0);
5167 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5168
5169 Node* ctl = control();
5170 while (ctl != alloc_ctl) {
5171 // There may be guards which feed into the slow_region.
5172 // Any other control flow means that we might not get a chance
5173 // to finish initializing the allocated object.
5174 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5175 IfNode* iff = ctl->in(0)->as_If();
5176 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5177 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5178 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5179 ctl = iff->in(0); // This test feeds the known slow_region.
5180 continue;
5181 }
5182 // One more try: Various low-level checks bottom out in
5183 // uncommon traps. If the debug-info of the trap omits
5184 // any reference to the allocation, as we've already
5185 // observed, then there can be no objection to the trap.
5186 bool found_trap = false;
5187 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5188 Node* obs = not_ctl->fast_out(j);
5189 if (obs->in(0) == not_ctl && obs->is_Call() &&
5190 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5191 found_trap = true; break;
5192 }
5193 }
5194 if (found_trap) {
5195 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5196 continue;
5197 }
5198 }
5199 return NULL;
5200 }
5201
5202 // If we get this far, we have an allocation which immediately
5203 // precedes the arraycopy, and we can take over zeroing the new object.
5204 // The arraycopy will finish the initialization, and provide
5205 // a new control state to which we will anchor the destination pointer.
5206
5207 return alloc;
5208 }
5209
5210 //-------------inline_encodeISOArray-----------------------------------
5211 // encode char[] to byte[] in ISO_8859_1
5212 bool LibraryCallKit::inline_encodeISOArray() {
5213 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5214 // no receiver since it is static method
5215 Node *src = argument(0);
5216 Node *src_offset = argument(1);
5217 Node *dst = argument(2);
5218 Node *dst_offset = argument(3);
5219 Node *length = argument(4);
5220
5221 const Type* src_type = src->Value(&_gvn);
5222 const Type* dst_type = dst->Value(&_gvn);
5223 const TypeAryPtr* top_src = src_type->isa_aryptr();
5224 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5225 if (top_src == NULL || top_src->klass() == NULL ||
5226 top_dest == NULL || top_dest->klass() == NULL) {
5227 // failed array check
5228 return false;
5229 }
5230
5231 // Figure out the size and type of the elements we will be copying.
5232 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5233 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5234 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5235 return false;
5236 }
5237
5238 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5239 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5240 // 'src_start' points to src array + scaled offset
5241 // 'dst_start' points to dst array + scaled offset
5242
5243 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5244 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5245 enc = _gvn.transform(enc);
5246 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5247 set_memory(res_mem, mtype);
5248 set_result(enc);
5249 return true;
5250 }
5251
5252 //-------------inline_multiplyToLen-----------------------------------
5253 bool LibraryCallKit::inline_multiplyToLen() {
5254 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5255
5256 address stubAddr = StubRoutines::multiplyToLen();
5257 if (stubAddr == NULL) {
5258 return false; // Intrinsic's stub is not implemented on this platform
5259 }
5260 const char* stubName = "multiplyToLen";
5261
5262 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5263
5264 // no receiver because it is a static method
5265 Node* x = argument(0);
5266 Node* xlen = argument(1);
5267 Node* y = argument(2);
5268 Node* ylen = argument(3);
5269 Node* z = argument(4);
5270
5271 const Type* x_type = x->Value(&_gvn);
5272 const Type* y_type = y->Value(&_gvn);
5273 const TypeAryPtr* top_x = x_type->isa_aryptr();
5274 const TypeAryPtr* top_y = y_type->isa_aryptr();
5275 if (top_x == NULL || top_x->klass() == NULL ||
5276 top_y == NULL || top_y->klass() == NULL) {
5277 // failed array check
5278 return false;
5279 }
5280
5281 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5282 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5283 if (x_elem != T_INT || y_elem != T_INT) {
5284 return false;
5285 }
5286
5287 // Set the original stack and the reexecute bit for the interpreter to reexecute
5288 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5289 // on the return from z array allocation in runtime.
5290 { PreserveReexecuteState preexecs(this);
5291 jvms()->set_should_reexecute(true);
5292
5293 Node* x_start = array_element_address(x, intcon(0), x_elem);
5294 Node* y_start = array_element_address(y, intcon(0), y_elem);
5295 // 'x_start' points to x array + scaled xlen
5296 // 'y_start' points to y array + scaled ylen
5297
5298 // Allocate the result array
5299 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5300 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5301 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5302
5303 IdealKit ideal(this);
5304
5305 #define __ ideal.
5306 Node* one = __ ConI(1);
5307 Node* zero = __ ConI(0);
5308 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5309 __ set(need_alloc, zero);
5310 __ set(z_alloc, z);
5311 __ if_then(z, BoolTest::eq, null()); {
5312 __ increment (need_alloc, one);
5313 } __ else_(); {
5314 // Update graphKit memory and control from IdealKit.
5315 sync_kit(ideal);
5316 Node* zlen_arg = load_array_length(z);
5317 // Update IdealKit memory and control from graphKit.
5318 __ sync_kit(this);
5319 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5320 __ increment (need_alloc, one);
5321 } __ end_if();
5322 } __ end_if();
5323
5324 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5325 // Update graphKit memory and control from IdealKit.
5326 sync_kit(ideal);
5327 Node * narr = new_array(klass_node, zlen, 1);
5328 // Update IdealKit memory and control from graphKit.
5329 __ sync_kit(this);
5330 __ set(z_alloc, narr);
5331 } __ end_if();
5332
5333 sync_kit(ideal);
5334 z = __ value(z_alloc);
5335 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5336 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5337 // Final sync IdealKit and GraphKit.
5338 final_sync(ideal);
5339 #undef __
5340
5341 Node* z_start = array_element_address(z, intcon(0), T_INT);
5342
5343 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5344 OptoRuntime::multiplyToLen_Type(),
5345 stubAddr, stubName, TypePtr::BOTTOM,
5346 x_start, xlen, y_start, ylen, z_start, zlen);
5347 } // original reexecute is set back here
5348
5349 C->set_has_split_ifs(true); // Has chance for split-if optimization
5350 set_result(z);
5351 return true;
5352 }
5353
5354 //-------------inline_squareToLen------------------------------------
5355 bool LibraryCallKit::inline_squareToLen() {
5356 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5357
5358 address stubAddr = StubRoutines::squareToLen();
5359 if (stubAddr == NULL) {
5360 return false; // Intrinsic's stub is not implemented on this platform
5361 }
5362 const char* stubName = "squareToLen";
5363
5364 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5365
5366 Node* x = argument(0);
5367 Node* len = argument(1);
5368 Node* z = argument(2);
5369 Node* zlen = argument(3);
5370
5371 const Type* x_type = x->Value(&_gvn);
5372 const Type* z_type = z->Value(&_gvn);
5373 const TypeAryPtr* top_x = x_type->isa_aryptr();
5374 const TypeAryPtr* top_z = z_type->isa_aryptr();
5375 if (top_x == NULL || top_x->klass() == NULL ||
5376 top_z == NULL || top_z->klass() == NULL) {
5377 // failed array check
5378 return false;
5379 }
5380
5381 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5382 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5383 if (x_elem != T_INT || z_elem != T_INT) {
5384 return false;
5385 }
5386
5387
5388 Node* x_start = array_element_address(x, intcon(0), x_elem);
5389 Node* z_start = array_element_address(z, intcon(0), z_elem);
5390
5391 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5392 OptoRuntime::squareToLen_Type(),
5393 stubAddr, stubName, TypePtr::BOTTOM,
5394 x_start, len, z_start, zlen);
5395
5396 set_result(z);
5397 return true;
5398 }
5399
5400 //-------------inline_mulAdd------------------------------------------
5401 bool LibraryCallKit::inline_mulAdd() {
5402 assert(UseMulAddIntrinsic, "not implemented on this platform");
5403
5404 address stubAddr = StubRoutines::mulAdd();
5405 if (stubAddr == NULL) {
5406 return false; // Intrinsic's stub is not implemented on this platform
5407 }
5408 const char* stubName = "mulAdd";
5409
5410 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5411
5412 Node* out = argument(0);
5413 Node* in = argument(1);
5414 Node* offset = argument(2);
5415 Node* len = argument(3);
5416 Node* k = argument(4);
5417
5418 const Type* out_type = out->Value(&_gvn);
5419 const Type* in_type = in->Value(&_gvn);
5420 const TypeAryPtr* top_out = out_type->isa_aryptr();
5421 const TypeAryPtr* top_in = in_type->isa_aryptr();
5422 if (top_out == NULL || top_out->klass() == NULL ||
5423 top_in == NULL || top_in->klass() == NULL) {
5424 // failed array check
5425 return false;
5426 }
5427
5428 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5429 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5430 if (out_elem != T_INT || in_elem != T_INT) {
5431 return false;
5432 }
5433
5434 Node* outlen = load_array_length(out);
5435 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5436 Node* out_start = array_element_address(out, intcon(0), out_elem);
5437 Node* in_start = array_element_address(in, intcon(0), in_elem);
5438
5439 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5440 OptoRuntime::mulAdd_Type(),
5441 stubAddr, stubName, TypePtr::BOTTOM,
5442 out_start,in_start, new_offset, len, k);
5443 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5444 set_result(result);
5445 return true;
5446 }
5447
5448 //-------------inline_montgomeryMultiply-----------------------------------
5449 bool LibraryCallKit::inline_montgomeryMultiply() {
5450 address stubAddr = StubRoutines::montgomeryMultiply();
5451 if (stubAddr == NULL) {
5452 return false; // Intrinsic's stub is not implemented on this platform
5453 }
5454
5455 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5456 const char* stubName = "montgomery_square";
5457
5458 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5459
5460 Node* a = argument(0);
5461 Node* b = argument(1);
5462 Node* n = argument(2);
5463 Node* len = argument(3);
5464 Node* inv = argument(4);
5465 Node* m = argument(6);
5466
5467 const Type* a_type = a->Value(&_gvn);
5468 const TypeAryPtr* top_a = a_type->isa_aryptr();
5469 const Type* b_type = b->Value(&_gvn);
5470 const TypeAryPtr* top_b = b_type->isa_aryptr();
5471 const Type* n_type = a->Value(&_gvn);
5472 const TypeAryPtr* top_n = n_type->isa_aryptr();
5473 const Type* m_type = a->Value(&_gvn);
5474 const TypeAryPtr* top_m = m_type->isa_aryptr();
5475 if (top_a == NULL || top_a->klass() == NULL ||
5476 top_b == NULL || top_b->klass() == NULL ||
5477 top_n == NULL || top_n->klass() == NULL ||
5478 top_m == NULL || top_m->klass() == NULL) {
5479 // failed array check
5480 return false;
5481 }
5482
5483 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5484 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5485 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5486 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5487 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5488 return false;
5489 }
5490
5491 // Make the call
5492 {
5493 Node* a_start = array_element_address(a, intcon(0), a_elem);
5494 Node* b_start = array_element_address(b, intcon(0), b_elem);
5495 Node* n_start = array_element_address(n, intcon(0), n_elem);
5496 Node* m_start = array_element_address(m, intcon(0), m_elem);
5497
5498 Node* call = make_runtime_call(RC_LEAF,
5499 OptoRuntime::montgomeryMultiply_Type(),
5500 stubAddr, stubName, TypePtr::BOTTOM,
5501 a_start, b_start, n_start, len, inv, top(),
5502 m_start);
5503 set_result(m);
5504 }
5505
5506 return true;
5507 }
5508
5509 bool LibraryCallKit::inline_montgomerySquare() {
5510 address stubAddr = StubRoutines::montgomerySquare();
5511 if (stubAddr == NULL) {
5512 return false; // Intrinsic's stub is not implemented on this platform
5513 }
5514
5515 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5516 const char* stubName = "montgomery_square";
5517
5518 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5519
5520 Node* a = argument(0);
5521 Node* n = argument(1);
5522 Node* len = argument(2);
5523 Node* inv = argument(3);
5524 Node* m = argument(5);
5525
5526 const Type* a_type = a->Value(&_gvn);
5527 const TypeAryPtr* top_a = a_type->isa_aryptr();
5528 const Type* n_type = a->Value(&_gvn);
5529 const TypeAryPtr* top_n = n_type->isa_aryptr();
5530 const Type* m_type = a->Value(&_gvn);
5531 const TypeAryPtr* top_m = m_type->isa_aryptr();
5532 if (top_a == NULL || top_a->klass() == NULL ||
5533 top_n == NULL || top_n->klass() == NULL ||
5534 top_m == NULL || top_m->klass() == NULL) {
5535 // failed array check
5536 return false;
5537 }
5538
5539 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5540 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5541 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5542 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5543 return false;
5544 }
5545
5546 // Make the call
5547 {
5548 Node* a_start = array_element_address(a, intcon(0), a_elem);
5549 Node* n_start = array_element_address(n, intcon(0), n_elem);
5550 Node* m_start = array_element_address(m, intcon(0), m_elem);
5551
5552 Node* call = make_runtime_call(RC_LEAF,
5553 OptoRuntime::montgomerySquare_Type(),
5554 stubAddr, stubName, TypePtr::BOTTOM,
5555 a_start, n_start, len, inv, top(),
5556 m_start);
5557 set_result(m);
5558 }
5559
5560 return true;
5561 }
5562
5563 //-------------inline_vectorizedMismatch------------------------------
5564 bool LibraryCallKit::inline_vectorizedMismatch() {
5565 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5566
5567 address stubAddr = StubRoutines::vectorizedMismatch();
5568 if (stubAddr == NULL) {
5569 return false; // Intrinsic's stub is not implemented on this platform
5570 }
5571 const char* stubName = "vectorizedMismatch";
5572 int size_l = callee()->signature()->size();
5573 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5574
5575 Node* obja = argument(0);
5576 Node* aoffset = argument(1);
5577 Node* objb = argument(3);
5578 Node* boffset = argument(4);
5579 Node* length = argument(6);
5580 Node* scale = argument(7);
5581
5582 const Type* a_type = obja->Value(&_gvn);
5583 const Type* b_type = objb->Value(&_gvn);
5584 const TypeAryPtr* top_a = a_type->isa_aryptr();
5585 const TypeAryPtr* top_b = b_type->isa_aryptr();
5586 if (top_a == NULL || top_a->klass() == NULL ||
5587 top_b == NULL || top_b->klass() == NULL) {
5588 // failed array check
5589 return false;
5590 }
5591
5592 Node* call;
5593 jvms()->set_should_reexecute(true);
5594
5595 Node* obja_adr = make_unsafe_address(obja, aoffset);
5596 Node* objb_adr = make_unsafe_address(objb, boffset);
5597
5598 call = make_runtime_call(RC_LEAF,
5599 OptoRuntime::vectorizedMismatch_Type(),
5600 stubAddr, stubName, TypePtr::BOTTOM,
5601 obja_adr, objb_adr, length, scale);
5602
5603 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5604 set_result(result);
5605 return true;
5606 }
5607
5608 /**
5609 * Calculate CRC32 for byte.
5610 * int java.util.zip.CRC32.update(int crc, int b)
5611 */
5612 bool LibraryCallKit::inline_updateCRC32() {
5613 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5614 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5615 // no receiver since it is static method
5616 Node* crc = argument(0); // type: int
5617 Node* b = argument(1); // type: int
5618
5619 /*
5620 * int c = ~ crc;
5621 * b = timesXtoThe32[(b ^ c) & 0xFF];
5622 * b = b ^ (c >>> 8);
5623 * crc = ~b;
5624 */
5625
5626 Node* M1 = intcon(-1);
5627 crc = _gvn.transform(new XorINode(crc, M1));
5628 Node* result = _gvn.transform(new XorINode(crc, b));
5629 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5630
5631 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5632 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5633 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5634 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5635
5636 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5637 result = _gvn.transform(new XorINode(crc, result));
5638 result = _gvn.transform(new XorINode(result, M1));
5639 set_result(result);
5640 return true;
5641 }
5642
5643 /**
5644 * Calculate CRC32 for byte[] array.
5645 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5646 */
5647 bool LibraryCallKit::inline_updateBytesCRC32() {
5648 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5649 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5650 // no receiver since it is static method
5651 Node* crc = argument(0); // type: int
5652 Node* src = argument(1); // type: oop
5653 Node* offset = argument(2); // type: int
5654 Node* length = argument(3); // type: int
5655
5656 const Type* src_type = src->Value(&_gvn);
5657 const TypeAryPtr* top_src = src_type->isa_aryptr();
5658 if (top_src == NULL || top_src->klass() == NULL) {
5659 // failed array check
5660 return false;
5661 }
5662
5663 // Figure out the size and type of the elements we will be copying.
5664 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5665 if (src_elem != T_BYTE) {
5666 return false;
5667 }
5668
5669 // 'src_start' points to src array + scaled offset
5670 Node* src_start = array_element_address(src, offset, src_elem);
5671
5672 // We assume that range check is done by caller.
5673 // TODO: generate range check (offset+length < src.length) in debug VM.
5674
5675 // Call the stub.
5676 address stubAddr = StubRoutines::updateBytesCRC32();
5677 const char *stubName = "updateBytesCRC32";
5678
5679 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5680 stubAddr, stubName, TypePtr::BOTTOM,
5681 crc, src_start, length);
5682 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5683 set_result(result);
5684 return true;
5685 }
5686
5687 /**
5688 * Calculate CRC32 for ByteBuffer.
5689 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5690 */
5691 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5692 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5693 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5694 // no receiver since it is static method
5695 Node* crc = argument(0); // type: int
5696 Node* src = argument(1); // type: long
5697 Node* offset = argument(3); // type: int
5698 Node* length = argument(4); // type: int
5699
5700 src = ConvL2X(src); // adjust Java long to machine word
5701 Node* base = _gvn.transform(new CastX2PNode(src));
5702 offset = ConvI2X(offset);
5703
5704 // 'src_start' points to src array + scaled offset
5705 Node* src_start = basic_plus_adr(top(), base, offset);
5706
5707 // Call the stub.
5708 address stubAddr = StubRoutines::updateBytesCRC32();
5709 const char *stubName = "updateBytesCRC32";
5710
5711 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5712 stubAddr, stubName, TypePtr::BOTTOM,
5713 crc, src_start, length);
5714 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5715 set_result(result);
5716 return true;
5717 }
5718
5719 //------------------------------get_table_from_crc32c_class-----------------------
5720 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5721 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5722 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5723
5724 return table;
5725 }
5726
5727 //------------------------------inline_updateBytesCRC32C-----------------------
5728 //
5729 // Calculate CRC32C for byte[] array.
5730 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5731 //
5732 bool LibraryCallKit::inline_updateBytesCRC32C() {
5733 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5734 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5735 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5736 // no receiver since it is a static method
5737 Node* crc = argument(0); // type: int
5738 Node* src = argument(1); // type: oop
5739 Node* offset = argument(2); // type: int
5740 Node* end = argument(3); // type: int
5741
5742 Node* length = _gvn.transform(new SubINode(end, offset));
5743
5744 const Type* src_type = src->Value(&_gvn);
5745 const TypeAryPtr* top_src = src_type->isa_aryptr();
5746 if (top_src == NULL || top_src->klass() == NULL) {
5747 // failed array check
5748 return false;
5749 }
5750
5751 // Figure out the size and type of the elements we will be copying.
5752 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5753 if (src_elem != T_BYTE) {
5754 return false;
5755 }
5756
5757 // 'src_start' points to src array + scaled offset
5758 Node* src_start = array_element_address(src, offset, src_elem);
5759
5760 // static final int[] byteTable in class CRC32C
5761 Node* table = get_table_from_crc32c_class(callee()->holder());
5762 Node* table_start = array_element_address(table, intcon(0), T_INT);
5763
5764 // We assume that range check is done by caller.
5765 // TODO: generate range check (offset+length < src.length) in debug VM.
5766
5767 // Call the stub.
5768 address stubAddr = StubRoutines::updateBytesCRC32C();
5769 const char *stubName = "updateBytesCRC32C";
5770
5771 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5772 stubAddr, stubName, TypePtr::BOTTOM,
5773 crc, src_start, length, table_start);
5774 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5775 set_result(result);
5776 return true;
5777 }
5778
5779 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5780 //
5781 // Calculate CRC32C for DirectByteBuffer.
5782 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5783 //
5784 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5785 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5786 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5787 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5788 // no receiver since it is a static method
5789 Node* crc = argument(0); // type: int
5790 Node* src = argument(1); // type: long
5791 Node* offset = argument(3); // type: int
5792 Node* end = argument(4); // type: int
5793
5794 Node* length = _gvn.transform(new SubINode(end, offset));
5795
5796 src = ConvL2X(src); // adjust Java long to machine word
5797 Node* base = _gvn.transform(new CastX2PNode(src));
5798 offset = ConvI2X(offset);
5799
5800 // 'src_start' points to src array + scaled offset
5801 Node* src_start = basic_plus_adr(top(), base, offset);
5802
5803 // static final int[] byteTable in class CRC32C
5804 Node* table = get_table_from_crc32c_class(callee()->holder());
5805 Node* table_start = array_element_address(table, intcon(0), T_INT);
5806
5807 // Call the stub.
5808 address stubAddr = StubRoutines::updateBytesCRC32C();
5809 const char *stubName = "updateBytesCRC32C";
5810
5811 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5812 stubAddr, stubName, TypePtr::BOTTOM,
5813 crc, src_start, length, table_start);
5814 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5815 set_result(result);
5816 return true;
5817 }
5818
5819 //------------------------------inline_updateBytesAdler32----------------------
5820 //
5821 // Calculate Adler32 checksum for byte[] array.
5822 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5823 //
5824 bool LibraryCallKit::inline_updateBytesAdler32() {
5825 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5826 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5827 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5828 // no receiver since it is static method
5829 Node* crc = argument(0); // type: int
5830 Node* src = argument(1); // type: oop
5831 Node* offset = argument(2); // type: int
5832 Node* length = argument(3); // type: int
5833
5834 const Type* src_type = src->Value(&_gvn);
5835 const TypeAryPtr* top_src = src_type->isa_aryptr();
5836 if (top_src == NULL || top_src->klass() == NULL) {
5837 // failed array check
5838 return false;
5839 }
5840
5841 // Figure out the size and type of the elements we will be copying.
5842 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5843 if (src_elem != T_BYTE) {
5844 return false;
5845 }
5846
5847 // 'src_start' points to src array + scaled offset
5848 Node* src_start = array_element_address(src, offset, src_elem);
5849
5850 // We assume that range check is done by caller.
5851 // TODO: generate range check (offset+length < src.length) in debug VM.
5852
5853 // Call the stub.
5854 address stubAddr = StubRoutines::updateBytesAdler32();
5855 const char *stubName = "updateBytesAdler32";
5856
5857 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5858 stubAddr, stubName, TypePtr::BOTTOM,
5859 crc, src_start, length);
5860 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5861 set_result(result);
5862 return true;
5863 }
5864
5865 //------------------------------inline_updateByteBufferAdler32---------------
5866 //
5867 // Calculate Adler32 checksum for DirectByteBuffer.
5868 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5869 //
5870 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5871 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5872 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5873 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5874 // no receiver since it is static method
5875 Node* crc = argument(0); // type: int
5876 Node* src = argument(1); // type: long
5877 Node* offset = argument(3); // type: int
5878 Node* length = argument(4); // type: int
5879
5880 src = ConvL2X(src); // adjust Java long to machine word
5881 Node* base = _gvn.transform(new CastX2PNode(src));
5882 offset = ConvI2X(offset);
5883
5884 // 'src_start' points to src array + scaled offset
5885 Node* src_start = basic_plus_adr(top(), base, offset);
5886
5887 // Call the stub.
5888 address stubAddr = StubRoutines::updateBytesAdler32();
5889 const char *stubName = "updateBytesAdler32";
5890
5891 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5892 stubAddr, stubName, TypePtr::BOTTOM,
5893 crc, src_start, length);
5894
5895 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5896 set_result(result);
5897 return true;
5898 }
5899
5900 //----------------------------inline_reference_get----------------------------
5901 // public T java.lang.ref.Reference.get();
5902 bool LibraryCallKit::inline_reference_get() {
5903 const int referent_offset = java_lang_ref_Reference::referent_offset;
5904 guarantee(referent_offset > 0, "should have already been set");
5905
5906 // Get the argument:
5907 Node* reference_obj = null_check_receiver();
5908 if (stopped()) return true;
5909
5910 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5911
5912 ciInstanceKlass* klass = env()->Object_klass();
5913 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5914
5915 Node* no_ctrl = NULL;
5916 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5917
5918 // Use the pre-barrier to record the value in the referent field
5919 pre_barrier(false /* do_load */,
5920 control(),
5921 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5922 result /* pre_val */,
5923 T_OBJECT);
5924
5925 // Add memory barrier to prevent commoning reads from this field
5926 // across safepoint since GC can change its value.
5927 insert_mem_bar(Op_MemBarCPUOrder);
5928
5929 set_result(result);
5930 return true;
5931 }
5932
5933
5934 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5935 bool is_exact=true, bool is_static=false,
5936 ciInstanceKlass * fromKls=NULL) {
5937 if (fromKls == NULL) {
5938 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5939 assert(tinst != NULL, "obj is null");
5940 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5941 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5942 fromKls = tinst->klass()->as_instance_klass();
5943 } else {
5944 assert(is_static, "only for static field access");
5945 }
5946 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5947 ciSymbol::make(fieldTypeString),
5948 is_static);
5949
5950 assert (field != NULL, "undefined field");
5951 if (field == NULL) return (Node *) NULL;
5952
5953 if (is_static) {
5954 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5955 fromObj = makecon(tip);
5956 }
5957
5958 // Next code copied from Parse::do_get_xxx():
5959
5960 // Compute address and memory type.
5961 int offset = field->offset_in_bytes();
5962 bool is_vol = field->is_volatile();
5963 ciType* field_klass = field->type();
5964 assert(field_klass->is_loaded(), "should be loaded");
5965 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5966 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5967 BasicType bt = field->layout_type();
5968
5969 // Build the resultant type of the load
5970 const Type *type;
5971 if (bt == T_OBJECT) {
5972 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5973 } else {
5974 type = Type::get_const_basic_type(bt);
5975 }
5976
5977 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5978 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
5979 }
5980 // Build the load.
5981 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5982 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5983 // If reference is volatile, prevent following memory ops from
5984 // floating up past the volatile read. Also prevents commoning
5985 // another volatile read.
5986 if (is_vol) {
5987 // Memory barrier includes bogus read of value to force load BEFORE membar
5988 insert_mem_bar(Op_MemBarAcquire, loadedField);
5989 }
5990 return loadedField;
5991 }
5992
5993 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5994 bool is_exact = true, bool is_static = false,
5995 ciInstanceKlass * fromKls = NULL) {
5996 if (fromKls == NULL) {
5997 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5998 assert(tinst != NULL, "obj is null");
5999 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6000 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6001 fromKls = tinst->klass()->as_instance_klass();
6002 }
6003 else {
6004 assert(is_static, "only for static field access");
6005 }
6006 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6007 ciSymbol::make(fieldTypeString),
6008 is_static);
6009
6010 assert(field != NULL, "undefined field");
6011 assert(!field->is_volatile(), "not defined for volatile fields");
6012
6013 if (is_static) {
6014 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6015 fromObj = makecon(tip);
6016 }
6017
6018 // Next code copied from Parse::do_get_xxx():
6019
6020 // Compute address and memory type.
6021 int offset = field->offset_in_bytes();
6022 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6023
6024 return adr;
6025 }
6026
6027 //------------------------------inline_aescrypt_Block-----------------------
6028 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6029 address stubAddr = NULL;
6030 const char *stubName;
6031 assert(UseAES, "need AES instruction support");
6032
6033 switch(id) {
6034 case vmIntrinsics::_aescrypt_encryptBlock:
6035 stubAddr = StubRoutines::aescrypt_encryptBlock();
6036 stubName = "aescrypt_encryptBlock";
6037 break;
6038 case vmIntrinsics::_aescrypt_decryptBlock:
6039 stubAddr = StubRoutines::aescrypt_decryptBlock();
6040 stubName = "aescrypt_decryptBlock";
6041 break;
6042 }
6043 if (stubAddr == NULL) return false;
6044
6045 Node* aescrypt_object = argument(0);
6046 Node* src = argument(1);
6047 Node* src_offset = argument(2);
6048 Node* dest = argument(3);
6049 Node* dest_offset = argument(4);
6050
6051 // (1) src and dest are arrays.
6052 const Type* src_type = src->Value(&_gvn);
6053 const Type* dest_type = dest->Value(&_gvn);
6054 const TypeAryPtr* top_src = src_type->isa_aryptr();
6055 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6056 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6057
6058 // for the quick and dirty code we will skip all the checks.
6059 // we are just trying to get the call to be generated.
6060 Node* src_start = src;
6061 Node* dest_start = dest;
6062 if (src_offset != NULL || dest_offset != NULL) {
6063 assert(src_offset != NULL && dest_offset != NULL, "");
6064 src_start = array_element_address(src, src_offset, T_BYTE);
6065 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6066 }
6067
6068 // now need to get the start of its expanded key array
6069 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6070 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6071 if (k_start == NULL) return false;
6072
6073 if (Matcher::pass_original_key_for_aes()) {
6074 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6075 // compatibility issues between Java key expansion and SPARC crypto instructions
6076 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6077 if (original_k_start == NULL) return false;
6078
6079 // Call the stub.
6080 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6081 stubAddr, stubName, TypePtr::BOTTOM,
6082 src_start, dest_start, k_start, original_k_start);
6083 } else {
6084 // Call the stub.
6085 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6086 stubAddr, stubName, TypePtr::BOTTOM,
6087 src_start, dest_start, k_start);
6088 }
6089
6090 return true;
6091 }
6092
6093 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6094 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6095 address stubAddr = NULL;
6096 const char *stubName = NULL;
6097
6098 assert(UseAES, "need AES instruction support");
6099
6100 switch(id) {
6101 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6102 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6103 stubName = "cipherBlockChaining_encryptAESCrypt";
6104 break;
6105 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6106 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6107 stubName = "cipherBlockChaining_decryptAESCrypt";
6108 break;
6109 }
6110 if (stubAddr == NULL) return false;
6111
6112 Node* cipherBlockChaining_object = argument(0);
6113 Node* src = argument(1);
6114 Node* src_offset = argument(2);
6115 Node* len = argument(3);
6116 Node* dest = argument(4);
6117 Node* dest_offset = argument(5);
6118
6119 // (1) src and dest are arrays.
6120 const Type* src_type = src->Value(&_gvn);
6121 const Type* dest_type = dest->Value(&_gvn);
6122 const TypeAryPtr* top_src = src_type->isa_aryptr();
6123 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6124 assert (top_src != NULL && top_src->klass() != NULL
6125 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6126
6127 // checks are the responsibility of the caller
6128 Node* src_start = src;
6129 Node* dest_start = dest;
6130 if (src_offset != NULL || dest_offset != NULL) {
6131 assert(src_offset != NULL && dest_offset != NULL, "");
6132 src_start = array_element_address(src, src_offset, T_BYTE);
6133 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6134 }
6135
6136 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6137 // (because of the predicated logic executed earlier).
6138 // so we cast it here safely.
6139 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6140
6141 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6142 if (embeddedCipherObj == NULL) return false;
6143
6144 // cast it to what we know it will be at runtime
6145 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6146 assert(tinst != NULL, "CBC obj is null");
6147 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6148 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6149 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6150
6151 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6152 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6153 const TypeOopPtr* xtype = aklass->as_instance_type();
6154 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6155 aescrypt_object = _gvn.transform(aescrypt_object);
6156
6157 // we need to get the start of the aescrypt_object's expanded key array
6158 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6159 if (k_start == NULL) return false;
6160
6161 // similarly, get the start address of the r vector
6162 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6163 if (objRvec == NULL) return false;
6164 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6165
6166 Node* cbcCrypt;
6167 if (Matcher::pass_original_key_for_aes()) {
6168 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6169 // compatibility issues between Java key expansion and SPARC crypto instructions
6170 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6171 if (original_k_start == NULL) return false;
6172
6173 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6174 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6175 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6176 stubAddr, stubName, TypePtr::BOTTOM,
6177 src_start, dest_start, k_start, r_start, len, original_k_start);
6178 } else {
6179 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6180 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6181 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6182 stubAddr, stubName, TypePtr::BOTTOM,
6183 src_start, dest_start, k_start, r_start, len);
6184 }
6185
6186 // return cipher length (int)
6187 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6188 set_result(retvalue);
6189 return true;
6190 }
6191
6192 //------------------------------inline_counterMode_AESCrypt-----------------------
6193 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6194 assert(UseAES, "need AES instruction support");
6195 if (!UseAESCTRIntrinsics) return false;
6196
6197 address stubAddr = NULL;
6198 const char *stubName = NULL;
6199 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6200 stubAddr = StubRoutines::counterMode_AESCrypt();
6201 stubName = "counterMode_AESCrypt";
6202 }
6203 if (stubAddr == NULL) return false;
6204
6205 Node* counterMode_object = argument(0);
6206 Node* src = argument(1);
6207 Node* src_offset = argument(2);
6208 Node* len = argument(3);
6209 Node* dest = argument(4);
6210 Node* dest_offset = argument(5);
6211
6212 // (1) src and dest are arrays.
6213 const Type* src_type = src->Value(&_gvn);
6214 const Type* dest_type = dest->Value(&_gvn);
6215 const TypeAryPtr* top_src = src_type->isa_aryptr();
6216 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6217 assert(top_src != NULL && top_src->klass() != NULL &&
6218 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6219
6220 // checks are the responsibility of the caller
6221 Node* src_start = src;
6222 Node* dest_start = dest;
6223 if (src_offset != NULL || dest_offset != NULL) {
6224 assert(src_offset != NULL && dest_offset != NULL, "");
6225 src_start = array_element_address(src, src_offset, T_BYTE);
6226 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6227 }
6228
6229 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6230 // (because of the predicated logic executed earlier).
6231 // so we cast it here safely.
6232 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6233 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6234 if (embeddedCipherObj == NULL) return false;
6235 // cast it to what we know it will be at runtime
6236 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6237 assert(tinst != NULL, "CTR obj is null");
6238 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6239 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6240 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6241 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6242 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6243 const TypeOopPtr* xtype = aklass->as_instance_type();
6244 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6245 aescrypt_object = _gvn.transform(aescrypt_object);
6246 // we need to get the start of the aescrypt_object's expanded key array
6247 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6248 if (k_start == NULL) return false;
6249 // similarly, get the start address of the r vector
6250 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6251 if (obj_counter == NULL) return false;
6252 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6253
6254 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6255 if (saved_encCounter == NULL) return false;
6256 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6257 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6258
6259 Node* ctrCrypt;
6260 if (Matcher::pass_original_key_for_aes()) {
6261 // no SPARC version for AES/CTR intrinsics now.
6262 return false;
6263 }
6264 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6265 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6266 OptoRuntime::counterMode_aescrypt_Type(),
6267 stubAddr, stubName, TypePtr::BOTTOM,
6268 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6269
6270 // return cipher length (int)
6271 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6272 set_result(retvalue);
6273 return true;
6274 }
6275
6276 //------------------------------get_key_start_from_aescrypt_object-----------------------
6277 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6278 #ifdef PPC64
6279 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6280 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6281 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6282 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6283 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6284 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6285 if (objSessionK == NULL) {
6286 return (Node *) NULL;
6287 }
6288 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6289 #else
6290 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6291 #endif // PPC64
6292 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6293 if (objAESCryptKey == NULL) return (Node *) NULL;
6294
6295 // now have the array, need to get the start address of the K array
6296 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6297 return k_start;
6298 }
6299
6300 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6301 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6302 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6303 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6304 if (objAESCryptKey == NULL) return (Node *) NULL;
6305
6306 // now have the array, need to get the start address of the lastKey array
6307 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6308 return original_k_start;
6309 }
6310
6311 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6312 // Return node representing slow path of predicate check.
6313 // the pseudo code we want to emulate with this predicate is:
6314 // for encryption:
6315 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6316 // for decryption:
6317 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6318 // note cipher==plain is more conservative than the original java code but that's OK
6319 //
6320 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6321 // The receiver was checked for NULL already.
6322 Node* objCBC = argument(0);
6323
6324 // Load embeddedCipher field of CipherBlockChaining object.
6325 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6326
6327 // get AESCrypt klass for instanceOf check
6328 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6329 // will have same classloader as CipherBlockChaining object
6330 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6331 assert(tinst != NULL, "CBCobj is null");
6332 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6333
6334 // we want to do an instanceof comparison against the AESCrypt class
6335 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6336 if (!klass_AESCrypt->is_loaded()) {
6337 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6338 Node* ctrl = control();
6339 set_control(top()); // no regular fast path
6340 return ctrl;
6341 }
6342 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6343
6344 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6345 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6346 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6347
6348 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6349
6350 // for encryption, we are done
6351 if (!decrypting)
6352 return instof_false; // even if it is NULL
6353
6354 // for decryption, we need to add a further check to avoid
6355 // taking the intrinsic path when cipher and plain are the same
6356 // see the original java code for why.
6357 RegionNode* region = new RegionNode(3);
6358 region->init_req(1, instof_false);
6359 Node* src = argument(1);
6360 Node* dest = argument(4);
6361 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6362 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6363 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6364 region->init_req(2, src_dest_conjoint);
6365
6366 record_for_igvn(region);
6367 return _gvn.transform(region);
6368 }
6369
6370 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6371 // Return node representing slow path of predicate check.
6372 // the pseudo code we want to emulate with this predicate is:
6373 // for encryption:
6374 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6375 // for decryption:
6376 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6377 // note cipher==plain is more conservative than the original java code but that's OK
6378 //
6379
6380 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6381 // The receiver was checked for NULL already.
6382 Node* objCTR = argument(0);
6383
6384 // Load embeddedCipher field of CipherBlockChaining object.
6385 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6386
6387 // get AESCrypt klass for instanceOf check
6388 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6389 // will have same classloader as CipherBlockChaining object
6390 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6391 assert(tinst != NULL, "CTRobj is null");
6392 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6393
6394 // we want to do an instanceof comparison against the AESCrypt class
6395 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6396 if (!klass_AESCrypt->is_loaded()) {
6397 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6398 Node* ctrl = control();
6399 set_control(top()); // no regular fast path
6400 return ctrl;
6401 }
6402
6403 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6404 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6405 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6406 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6407 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6408
6409 return instof_false; // even if it is NULL
6410 }
6411
6412 //------------------------------inline_ghash_processBlocks
6413 bool LibraryCallKit::inline_ghash_processBlocks() {
6414 address stubAddr;
6415 const char *stubName;
6416 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6417
6418 stubAddr = StubRoutines::ghash_processBlocks();
6419 stubName = "ghash_processBlocks";
6420
6421 Node* data = argument(0);
6422 Node* offset = argument(1);
6423 Node* len = argument(2);
6424 Node* state = argument(3);
6425 Node* subkeyH = argument(4);
6426
6427 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6428 assert(state_start, "state is NULL");
6429 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6430 assert(subkeyH_start, "subkeyH is NULL");
6431 Node* data_start = array_element_address(data, offset, T_BYTE);
6432 assert(data_start, "data is NULL");
6433
6434 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6435 OptoRuntime::ghash_processBlocks_Type(),
6436 stubAddr, stubName, TypePtr::BOTTOM,
6437 state_start, subkeyH_start, data_start, len);
6438 return true;
6439 }
6440
6441 //------------------------------inline_sha_implCompress-----------------------
6442 //
6443 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6444 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6445 //
6446 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6447 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6448 //
6449 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6450 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6451 //
6452 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6453 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6454
6455 Node* sha_obj = argument(0);
6456 Node* src = argument(1); // type oop
6457 Node* ofs = argument(2); // type int
6458
6459 const Type* src_type = src->Value(&_gvn);
6460 const TypeAryPtr* top_src = src_type->isa_aryptr();
6461 if (top_src == NULL || top_src->klass() == NULL) {
6462 // failed array check
6463 return false;
6464 }
6465 // Figure out the size and type of the elements we will be copying.
6466 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6467 if (src_elem != T_BYTE) {
6468 return false;
6469 }
6470 // 'src_start' points to src array + offset
6471 Node* src_start = array_element_address(src, ofs, src_elem);
6472 Node* state = NULL;
6473 address stubAddr;
6474 const char *stubName;
6475
6476 switch(id) {
6477 case vmIntrinsics::_sha_implCompress:
6478 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6479 state = get_state_from_sha_object(sha_obj);
6480 stubAddr = StubRoutines::sha1_implCompress();
6481 stubName = "sha1_implCompress";
6482 break;
6483 case vmIntrinsics::_sha2_implCompress:
6484 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6485 state = get_state_from_sha_object(sha_obj);
6486 stubAddr = StubRoutines::sha256_implCompress();
6487 stubName = "sha256_implCompress";
6488 break;
6489 case vmIntrinsics::_sha5_implCompress:
6490 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6491 state = get_state_from_sha5_object(sha_obj);
6492 stubAddr = StubRoutines::sha512_implCompress();
6493 stubName = "sha512_implCompress";
6494 break;
6495 default:
6496 fatal_unexpected_iid(id);
6497 return false;
6498 }
6499 if (state == NULL) return false;
6500
6501 // Call the stub.
6502 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6503 stubAddr, stubName, TypePtr::BOTTOM,
6504 src_start, state);
6505
6506 return true;
6507 }
6508
6509 //------------------------------inline_digestBase_implCompressMB-----------------------
6510 //
6511 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6512 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6513 //
6514 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6515 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6516 "need SHA1/SHA256/SHA512 instruction support");
6517 assert((uint)predicate < 3, "sanity");
6518 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6519
6520 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6521 Node* src = argument(1); // byte[] array
6522 Node* ofs = argument(2); // type int
6523 Node* limit = argument(3); // type int
6524
6525 const Type* src_type = src->Value(&_gvn);
6526 const TypeAryPtr* top_src = src_type->isa_aryptr();
6527 if (top_src == NULL || top_src->klass() == NULL) {
6528 // failed array check
6529 return false;
6530 }
6531 // Figure out the size and type of the elements we will be copying.
6532 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6533 if (src_elem != T_BYTE) {
6534 return false;
6535 }
6536 // 'src_start' points to src array + offset
6537 Node* src_start = array_element_address(src, ofs, src_elem);
6538
6539 const char* klass_SHA_name = NULL;
6540 const char* stub_name = NULL;
6541 address stub_addr = NULL;
6542 bool long_state = false;
6543
6544 switch (predicate) {
6545 case 0:
6546 if (UseSHA1Intrinsics) {
6547 klass_SHA_name = "sun/security/provider/SHA";
6548 stub_name = "sha1_implCompressMB";
6549 stub_addr = StubRoutines::sha1_implCompressMB();
6550 }
6551 break;
6552 case 1:
6553 if (UseSHA256Intrinsics) {
6554 klass_SHA_name = "sun/security/provider/SHA2";
6555 stub_name = "sha256_implCompressMB";
6556 stub_addr = StubRoutines::sha256_implCompressMB();
6557 }
6558 break;
6559 case 2:
6560 if (UseSHA512Intrinsics) {
6561 klass_SHA_name = "sun/security/provider/SHA5";
6562 stub_name = "sha512_implCompressMB";
6563 stub_addr = StubRoutines::sha512_implCompressMB();
6564 long_state = true;
6565 }
6566 break;
6567 default:
6568 fatal("unknown SHA intrinsic predicate: %d", predicate);
6569 }
6570 if (klass_SHA_name != NULL) {
6571 // get DigestBase klass to lookup for SHA klass
6572 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6573 assert(tinst != NULL, "digestBase_obj is not instance???");
6574 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6575
6576 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6577 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6578 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6579 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6580 }
6581 return false;
6582 }
6583 //------------------------------inline_sha_implCompressMB-----------------------
6584 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6585 bool long_state, address stubAddr, const char *stubName,
6586 Node* src_start, Node* ofs, Node* limit) {
6587 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6588 const TypeOopPtr* xtype = aklass->as_instance_type();
6589 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6590 sha_obj = _gvn.transform(sha_obj);
6591
6592 Node* state;
6593 if (long_state) {
6594 state = get_state_from_sha5_object(sha_obj);
6595 } else {
6596 state = get_state_from_sha_object(sha_obj);
6597 }
6598 if (state == NULL) return false;
6599
6600 // Call the stub.
6601 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6602 OptoRuntime::digestBase_implCompressMB_Type(),
6603 stubAddr, stubName, TypePtr::BOTTOM,
6604 src_start, state, ofs, limit);
6605 // return ofs (int)
6606 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6607 set_result(result);
6608
6609 return true;
6610 }
6611
6612 //------------------------------get_state_from_sha_object-----------------------
6613 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6614 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6615 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6616 if (sha_state == NULL) return (Node *) NULL;
6617
6618 // now have the array, need to get the start address of the state array
6619 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6620 return state;
6621 }
6622
6623 //------------------------------get_state_from_sha5_object-----------------------
6624 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6625 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6626 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6627 if (sha_state == NULL) return (Node *) NULL;
6628
6629 // now have the array, need to get the start address of the state array
6630 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6631 return state;
6632 }
6633
6634 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6635 // Return node representing slow path of predicate check.
6636 // the pseudo code we want to emulate with this predicate is:
6637 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6638 //
6639 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6640 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6641 "need SHA1/SHA256/SHA512 instruction support");
6642 assert((uint)predicate < 3, "sanity");
6643
6644 // The receiver was checked for NULL already.
6645 Node* digestBaseObj = argument(0);
6646
6647 // get DigestBase klass for instanceOf check
6648 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6649 assert(tinst != NULL, "digestBaseObj is null");
6650 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6651
6652 const char* klass_SHA_name = NULL;
6653 switch (predicate) {
6654 case 0:
6655 if (UseSHA1Intrinsics) {
6656 // we want to do an instanceof comparison against the SHA class
6657 klass_SHA_name = "sun/security/provider/SHA";
6658 }
6659 break;
6660 case 1:
6661 if (UseSHA256Intrinsics) {
6662 // we want to do an instanceof comparison against the SHA2 class
6663 klass_SHA_name = "sun/security/provider/SHA2";
6664 }
6665 break;
6666 case 2:
6667 if (UseSHA512Intrinsics) {
6668 // we want to do an instanceof comparison against the SHA5 class
6669 klass_SHA_name = "sun/security/provider/SHA5";
6670 }
6671 break;
6672 default:
6673 fatal("unknown SHA intrinsic predicate: %d", predicate);
6674 }
6675
6676 ciKlass* klass_SHA = NULL;
6677 if (klass_SHA_name != NULL) {
6678 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6679 }
6680 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6681 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6682 Node* ctrl = control();
6683 set_control(top()); // no intrinsic path
6684 return ctrl;
6685 }
6686 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6687
6688 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6689 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6690 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6691 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6692
6693 return instof_false; // even if it is NULL
6694 }
6695
6696 bool LibraryCallKit::inline_profileBoolean() {
6697 Node* counts = argument(1);
6698 const TypeAryPtr* ary = NULL;
6699 ciArray* aobj = NULL;
6700 if (counts->is_Con()
6701 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6702 && (aobj = ary->const_oop()->as_array()) != NULL
6703 && (aobj->length() == 2)) {
6704 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6705 jint false_cnt = aobj->element_value(0).as_int();
6706 jint true_cnt = aobj->element_value(1).as_int();
6707
6708 if (C->log() != NULL) {
6709 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6710 false_cnt, true_cnt);
6711 }
6712
6713 if (false_cnt + true_cnt == 0) {
6714 // According to profile, never executed.
6715 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6716 Deoptimization::Action_reinterpret);
6717 return true;
6718 }
6719
6720 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6721 // is a number of each value occurrences.
6722 Node* result = argument(0);
6723 if (false_cnt == 0 || true_cnt == 0) {
6724 // According to profile, one value has been never seen.
6725 int expected_val = (false_cnt == 0) ? 1 : 0;
6726
6727 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6728 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6729
6730 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6731 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6732 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6733
6734 { // Slow path: uncommon trap for never seen value and then reexecute
6735 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6736 // the value has been seen at least once.
6737 PreserveJVMState pjvms(this);
6738 PreserveReexecuteState preexecs(this);
6739 jvms()->set_should_reexecute(true);
6740
6741 set_control(slow_path);
6742 set_i_o(i_o());
6743
6744 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6745 Deoptimization::Action_reinterpret);
6746 }
6747 // The guard for never seen value enables sharpening of the result and
6748 // returning a constant. It allows to eliminate branches on the same value
6749 // later on.
6750 set_control(fast_path);
6751 result = intcon(expected_val);
6752 }
6753 // Stop profiling.
6754 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6755 // By replacing method body with profile data (represented as ProfileBooleanNode
6756 // on IR level) we effectively disable profiling.
6757 // It enables full speed execution once optimized code is generated.
6758 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6759 C->record_for_igvn(profile);
6760 set_result(profile);
6761 return true;
6762 } else {
6763 // Continue profiling.
6764 // Profile data isn't available at the moment. So, execute method's bytecode version.
6765 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6766 // is compiled and counters aren't available since corresponding MethodHandle
6767 // isn't a compile-time constant.
6768 return false;
6769 }
6770 }
6771
6772 bool LibraryCallKit::inline_isCompileConstant() {
6773 Node* n = argument(0);
6774 set_result(n->is_Con() ? intcon(1) : intcon(0));
6775 return true;
6776 }