1 /*
   2  * Copyright (c) 1999, 2015, 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 "oops/objArrayKlass.hpp"
  32 #include "opto/addnode.hpp"
  33 #include "opto/arraycopynode.hpp"
  34 #include "opto/callGenerator.hpp"
  35 #include "opto/castnode.hpp"
  36 #include "opto/cfgnode.hpp"
  37 #include "opto/convertnode.hpp"
  38 #include "opto/countbitsnode.hpp"
  39 #include "opto/intrinsicnode.hpp"
  40 #include "opto/idealKit.hpp"
  41 #include "opto/mathexactnode.hpp"
  42 #include "opto/movenode.hpp"
  43 #include "opto/mulnode.hpp"
  44 #include "opto/narrowptrnode.hpp"
  45 #include "opto/opaquenode.hpp"
  46 #include "opto/parse.hpp"
  47 #include "opto/runtime.hpp"
  48 #include "opto/subnode.hpp"
  49 #include "prims/nativeLookup.hpp"
  50 #include "runtime/sharedRuntime.hpp"
  51 #include "trace/traceMacros.hpp"
  52 
  53 class LibraryIntrinsic : public InlineCallGenerator {
  54   // Extend the set of intrinsics known to the runtime:
  55  public:
  56  private:
  57   bool             _is_virtual;
  58   bool             _does_virtual_dispatch;
  59   int8_t           _predicates_count;  // Intrinsic is predicated by several conditions
  60   int8_t           _last_predicate; // Last generated predicate
  61   vmIntrinsics::ID _intrinsic_id;
  62 
  63  public:
  64   LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
  65     : InlineCallGenerator(m),
  66       _is_virtual(is_virtual),
  67       _does_virtual_dispatch(does_virtual_dispatch),
  68       _predicates_count((int8_t)predicates_count),
  69       _last_predicate((int8_t)-1),
  70       _intrinsic_id(id)
  71   {
  72   }
  73   virtual bool is_intrinsic() const { return true; }
  74   virtual bool is_virtual()   const { return _is_virtual; }
  75   virtual bool is_predicated() const { return _predicates_count > 0; }
  76   virtual int  predicates_count() const { return _predicates_count; }
  77   virtual bool does_virtual_dispatch()   const { return _does_virtual_dispatch; }
  78   virtual JVMState* generate(JVMState* jvms);
  79   virtual Node* generate_predicate(JVMState* jvms, int predicate);
  80   vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
  81 };
  82 
  83 
  84 // Local helper class for LibraryIntrinsic:
  85 class LibraryCallKit : public GraphKit {
  86  private:
  87   LibraryIntrinsic* _intrinsic;     // the library intrinsic being called
  88   Node*             _result;        // the result node, if any
  89   int               _reexecute_sp;  // the stack pointer when bytecode needs to be reexecuted
  90 
  91   const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);
  92 
  93  public:
  94   LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
  95     : GraphKit(jvms),
  96       _intrinsic(intrinsic),
  97       _result(NULL)
  98   {
  99     // Check if this is a root compile.  In that case we don't have a caller.
 100     if (!jvms->has_method()) {
 101       _reexecute_sp = sp();
 102     } else {
 103       // Find out how many arguments the interpreter needs when deoptimizing
 104       // and save the stack pointer value so it can used by uncommon_trap.
 105       // We find the argument count by looking at the declared signature.
 106       bool ignored_will_link;
 107       ciSignature* declared_signature = NULL;
 108       ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
 109       const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
 110       _reexecute_sp = sp() + nargs;  // "push" arguments back on stack
 111     }
 112   }
 113 
 114   virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
 115 
 116   ciMethod*         caller()    const    { return jvms()->method(); }
 117   int               bci()       const    { return jvms()->bci(); }
 118   LibraryIntrinsic* intrinsic() const    { return _intrinsic; }
 119   vmIntrinsics::ID  intrinsic_id() const { return _intrinsic->intrinsic_id(); }
 120   ciMethod*         callee()    const    { return _intrinsic->method(); }
 121 
 122   bool  try_to_inline(int predicate);
 123   Node* try_to_predicate(int predicate);
 124 
 125   void push_result() {
 126     // Push the result onto the stack.
 127     if (!stopped() && result() != NULL) {
 128       BasicType bt = result()->bottom_type()->basic_type();
 129       push_node(bt, result());
 130     }
 131   }
 132 
 133  private:
 134   void fatal_unexpected_iid(vmIntrinsics::ID iid) {
 135     fatal(err_msg_res("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)));
 136   }
 137 
 138   void  set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
 139   void  set_result(RegionNode* region, PhiNode* value);
 140   Node*     result() { return _result; }
 141 
 142   virtual int reexecute_sp() { return _reexecute_sp; }
 143 
 144   // Helper functions to inline natives
 145   Node* generate_guard(Node* test, RegionNode* region, float true_prob);
 146   Node* generate_slow_guard(Node* test, RegionNode* region);
 147   Node* generate_fair_guard(Node* test, RegionNode* region);
 148   Node* generate_negative_guard(Node* index, RegionNode* region,
 149                                 // resulting CastII of index:
 150                                 Node* *pos_index = NULL);
 151   Node* generate_limit_guard(Node* offset, Node* subseq_length,
 152                              Node* array_length,
 153                              RegionNode* region);
 154   Node* generate_current_thread(Node* &tls_output);
 155   Node* load_mirror_from_klass(Node* klass);
 156   Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
 157                                       RegionNode* region, int null_path,
 158                                       int offset);
 159   Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
 160                                RegionNode* region, int null_path) {
 161     int offset = java_lang_Class::klass_offset_in_bytes();
 162     return load_klass_from_mirror_common(mirror, never_see_null,
 163                                          region, null_path,
 164                                          offset);
 165   }
 166   Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
 167                                      RegionNode* region, int null_path) {
 168     int offset = java_lang_Class::array_klass_offset_in_bytes();
 169     return load_klass_from_mirror_common(mirror, never_see_null,
 170                                          region, null_path,
 171                                          offset);
 172   }
 173   Node* generate_access_flags_guard(Node* kls,
 174                                     int modifier_mask, int modifier_bits,
 175                                     RegionNode* region);
 176   Node* generate_interface_guard(Node* kls, RegionNode* region);
 177   Node* generate_array_guard(Node* kls, RegionNode* region) {
 178     return generate_array_guard_common(kls, region, false, false);
 179   }
 180   Node* generate_non_array_guard(Node* kls, RegionNode* region) {
 181     return generate_array_guard_common(kls, region, false, true);
 182   }
 183   Node* generate_objArray_guard(Node* kls, RegionNode* region) {
 184     return generate_array_guard_common(kls, region, true, false);
 185   }
 186   Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
 187     return generate_array_guard_common(kls, region, true, true);
 188   }
 189   Node* generate_array_guard_common(Node* kls, RegionNode* region,
 190                                     bool obj_array, bool not_array);
 191   Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
 192   CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
 193                                      bool is_virtual = false, bool is_static = false);
 194   CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
 195     return generate_method_call(method_id, false, true);
 196   }
 197   CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
 198     return generate_method_call(method_id, true, false);
 199   }
 200   Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static);
 201 
 202   Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2);
 203   Node* make_string_method_node(int opcode, Node* str1, Node* str2);
 204   bool inline_string_compareTo();
 205   bool inline_string_indexOf();
 206   Node* string_indexOf(Node* string_object, ciTypeArray* target_array, jint offset, jint cache_i, jint md2_i);
 207   bool inline_string_equals();
 208   Node* round_double_node(Node* n);
 209   bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
 210   bool inline_math_native(vmIntrinsics::ID id);
 211   bool inline_trig(vmIntrinsics::ID id);
 212   bool inline_math(vmIntrinsics::ID id);
 213   template <typename OverflowOp>
 214   bool inline_math_overflow(Node* arg1, Node* arg2);
 215   void inline_math_mathExact(Node* math, Node* test);
 216   bool inline_math_addExactI(bool is_increment);
 217   bool inline_math_addExactL(bool is_increment);
 218   bool inline_math_multiplyExactI();
 219   bool inline_math_multiplyExactL();
 220   bool inline_math_negateExactI();
 221   bool inline_math_negateExactL();
 222   bool inline_math_subtractExactI(bool is_decrement);
 223   bool inline_math_subtractExactL(bool is_decrement);
 224   bool inline_exp();
 225   bool inline_pow();
 226   Node* finish_pow_exp(Node* result, Node* x, Node* y, const TypeFunc* call_type, address funcAddr, const char* funcName);
 227   bool inline_min_max(vmIntrinsics::ID id);
 228   Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
 229   // This returns Type::AnyPtr, RawPtr, or OopPtr.
 230   int classify_unsafe_addr(Node* &base, Node* &offset);
 231   Node* make_unsafe_address(Node* base, Node* offset);
 232   // Helper for inline_unsafe_access.
 233   // Generates the guards that check whether the result of
 234   // Unsafe.getObject should be recorded in an SATB log buffer.
 235   void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
 236   bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile);
 237   static bool klass_needs_init_guard(Node* kls);
 238   bool inline_unsafe_allocate();
 239   bool inline_unsafe_copyMemory();
 240   bool inline_native_currentThread();
 241 #ifdef TRACE_HAVE_INTRINSICS
 242   bool inline_native_classID();
 243   bool inline_native_threadID();
 244 #endif
 245   bool inline_native_time_funcs(address method, const char* funcName);
 246   bool inline_native_isInterrupted();
 247   bool inline_native_Class_query(vmIntrinsics::ID id);
 248   bool inline_native_subtype_check();
 249 
 250   bool inline_native_newArray();
 251   bool inline_native_getLength();
 252   bool inline_array_copyOf(bool is_copyOfRange);
 253   bool inline_array_equals();
 254   void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
 255   bool inline_native_clone(bool is_virtual);
 256   bool inline_native_Reflection_getCallerClass();
 257   // Helper function for inlining native object hash method
 258   bool inline_native_hashcode(bool is_virtual, bool is_static);
 259   bool inline_native_getClass();
 260 
 261   // Helper functions for inlining arraycopy
 262   bool inline_arraycopy();
 263   AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
 264                                                 RegionNode* slow_region);
 265   JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
 266   void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
 267 
 268   typedef enum { LS_xadd, LS_xchg, LS_cmpxchg } LoadStoreKind;
 269   bool inline_unsafe_load_store(BasicType type,  LoadStoreKind kind);
 270   bool inline_unsafe_ordered_store(BasicType type);
 271   bool inline_unsafe_fence(vmIntrinsics::ID id);
 272   bool inline_fp_conversions(vmIntrinsics::ID id);
 273   bool inline_number_methods(vmIntrinsics::ID id);
 274   bool inline_reference_get();
 275   bool inline_Class_cast();
 276   bool inline_aescrypt_Block(vmIntrinsics::ID id);
 277   bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
 278   Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
 279   Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
 280   Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
 281   bool inline_sha_implCompress(vmIntrinsics::ID id);
 282   bool inline_digestBase_implCompressMB(int predicate);
 283   bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
 284                                  bool long_state, address stubAddr, const char *stubName,
 285                                  Node* src_start, Node* ofs, Node* limit);
 286   Node* get_state_from_sha_object(Node *sha_object);
 287   Node* get_state_from_sha5_object(Node *sha_object);
 288   Node* inline_digestBase_implCompressMB_predicate(int predicate);
 289   bool inline_encodeISOArray();
 290   bool inline_updateCRC32();
 291   bool inline_updateBytesCRC32();
 292   bool inline_updateByteBufferCRC32();
 293   bool inline_multiplyToLen();
 294   bool inline_squareToLen();
 295   bool inline_mulAdd();
 296 
 297   bool inline_profileBoolean();
 298   bool inline_isCompileConstant();
 299 };
 300 
 301 
 302 //---------------------------make_vm_intrinsic----------------------------
 303 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
 304   vmIntrinsics::ID id = m->intrinsic_id();
 305   assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
 306 
 307   ccstr disable_intr = NULL;
 308 
 309   if ((DisableIntrinsic[0] != '\0'
 310        && strstr(DisableIntrinsic, vmIntrinsics::name_at(id)) != NULL) ||
 311       (method_has_option_value("DisableIntrinsic", disable_intr)
 312        && strstr(disable_intr, vmIntrinsics::name_at(id)) != NULL)) {
 313     // disabled by a user request on the command line:
 314     // example: -XX:DisableIntrinsic=_hashCode,_getClass
 315     return NULL;
 316   }
 317 
 318   if (!m->is_loaded()) {
 319     // do not attempt to inline unloaded methods
 320     return NULL;
 321   }
 322 
 323   // Only a few intrinsics implement a virtual dispatch.
 324   // They are expensive calls which are also frequently overridden.
 325   if (is_virtual) {
 326     switch (id) {
 327     case vmIntrinsics::_hashCode:
 328     case vmIntrinsics::_clone:
 329       // OK, Object.hashCode and Object.clone intrinsics come in both flavors
 330       break;
 331     default:
 332       return NULL;
 333     }
 334   }
 335 
 336   // -XX:-InlineNatives disables nearly all intrinsics:
 337   if (!InlineNatives) {
 338     switch (id) {
 339     case vmIntrinsics::_indexOf:
 340     case vmIntrinsics::_compareTo:
 341     case vmIntrinsics::_equals:
 342     case vmIntrinsics::_equalsC:
 343     case vmIntrinsics::_getAndAddInt:
 344     case vmIntrinsics::_getAndAddLong:
 345     case vmIntrinsics::_getAndSetInt:
 346     case vmIntrinsics::_getAndSetLong:
 347     case vmIntrinsics::_getAndSetObject:
 348     case vmIntrinsics::_loadFence:
 349     case vmIntrinsics::_storeFence:
 350     case vmIntrinsics::_fullFence:
 351       break;  // InlineNatives does not control String.compareTo
 352     case vmIntrinsics::_Reference_get:
 353       break;  // InlineNatives does not control Reference.get
 354     default:
 355       return NULL;
 356     }
 357   }
 358 
 359   int predicates = 0;
 360   bool does_virtual_dispatch = false;
 361 
 362   switch (id) {
 363   case vmIntrinsics::_compareTo:
 364     if (!SpecialStringCompareTo)  return NULL;
 365     if (!Matcher::match_rule_supported(Op_StrComp))  return NULL;
 366     break;
 367   case vmIntrinsics::_indexOf:
 368     if (!SpecialStringIndexOf)  return NULL;
 369     break;
 370   case vmIntrinsics::_equals:
 371     if (!SpecialStringEquals)  return NULL;
 372     if (!Matcher::match_rule_supported(Op_StrEquals))  return NULL;
 373     break;
 374   case vmIntrinsics::_equalsC:
 375     if (!SpecialArraysEquals)  return NULL;
 376     if (!Matcher::match_rule_supported(Op_AryEq))  return NULL;
 377     break;
 378   case vmIntrinsics::_arraycopy:
 379     if (!InlineArrayCopy)  return NULL;
 380     break;
 381   case vmIntrinsics::_copyMemory:
 382     if (StubRoutines::unsafe_arraycopy() == NULL)  return NULL;
 383     if (!InlineArrayCopy)  return NULL;
 384     break;
 385   case vmIntrinsics::_hashCode:
 386     if (!InlineObjectHash)  return NULL;
 387     does_virtual_dispatch = true;
 388     break;
 389   case vmIntrinsics::_clone:
 390     does_virtual_dispatch = true;
 391   case vmIntrinsics::_copyOf:
 392   case vmIntrinsics::_copyOfRange:
 393     if (!InlineObjectCopy)  return NULL;
 394     // These also use the arraycopy intrinsic mechanism:
 395     if (!InlineArrayCopy)  return NULL;
 396     break;
 397   case vmIntrinsics::_encodeISOArray:
 398     if (!SpecialEncodeISOArray)  return NULL;
 399     if (!Matcher::match_rule_supported(Op_EncodeISOArray))  return NULL;
 400     break;
 401   case vmIntrinsics::_checkIndex:
 402     // We do not intrinsify this.  The optimizer does fine with it.
 403     return NULL;
 404 
 405   case vmIntrinsics::_getCallerClass:
 406     if (!InlineReflectionGetCallerClass)  return NULL;
 407     if (SystemDictionary::reflect_CallerSensitive_klass() == NULL)  return NULL;
 408     break;
 409 
 410   case vmIntrinsics::_bitCount_i:
 411     if (!Matcher::match_rule_supported(Op_PopCountI)) return NULL;
 412     break;
 413 
 414   case vmIntrinsics::_bitCount_l:
 415     if (!Matcher::match_rule_supported(Op_PopCountL)) return NULL;
 416     break;
 417 
 418   case vmIntrinsics::_numberOfLeadingZeros_i:
 419     if (!Matcher::match_rule_supported(Op_CountLeadingZerosI)) return NULL;
 420     break;
 421 
 422   case vmIntrinsics::_numberOfLeadingZeros_l:
 423     if (!Matcher::match_rule_supported(Op_CountLeadingZerosL)) return NULL;
 424     break;
 425 
 426   case vmIntrinsics::_numberOfTrailingZeros_i:
 427     if (!Matcher::match_rule_supported(Op_CountTrailingZerosI)) return NULL;
 428     break;
 429 
 430   case vmIntrinsics::_numberOfTrailingZeros_l:
 431     if (!Matcher::match_rule_supported(Op_CountTrailingZerosL)) return NULL;
 432     break;
 433 
 434   case vmIntrinsics::_reverseBytes_c:
 435     if (!Matcher::match_rule_supported(Op_ReverseBytesUS)) return NULL;
 436     break;
 437   case vmIntrinsics::_reverseBytes_s:
 438     if (!Matcher::match_rule_supported(Op_ReverseBytesS))  return NULL;
 439     break;
 440   case vmIntrinsics::_reverseBytes_i:
 441     if (!Matcher::match_rule_supported(Op_ReverseBytesI))  return NULL;
 442     break;
 443   case vmIntrinsics::_reverseBytes_l:
 444     if (!Matcher::match_rule_supported(Op_ReverseBytesL))  return NULL;
 445     break;
 446 
 447   case vmIntrinsics::_Reference_get:
 448     // Use the intrinsic version of Reference.get() so that the value in
 449     // the referent field can be registered by the G1 pre-barrier code.
 450     // Also add memory barrier to prevent commoning reads from this field
 451     // across safepoint since GC can change it value.
 452     break;
 453 
 454   case vmIntrinsics::_compareAndSwapObject:
 455 #ifdef _LP64
 456     if (!UseCompressedOops && !Matcher::match_rule_supported(Op_CompareAndSwapP)) return NULL;
 457 #endif
 458     break;
 459 
 460   case vmIntrinsics::_compareAndSwapLong:
 461     if (!Matcher::match_rule_supported(Op_CompareAndSwapL)) return NULL;
 462     break;
 463 
 464   case vmIntrinsics::_getAndAddInt:
 465     if (!Matcher::match_rule_supported(Op_GetAndAddI)) return NULL;
 466     break;
 467 
 468   case vmIntrinsics::_getAndAddLong:
 469     if (!Matcher::match_rule_supported(Op_GetAndAddL)) return NULL;
 470     break;
 471 
 472   case vmIntrinsics::_getAndSetInt:
 473     if (!Matcher::match_rule_supported(Op_GetAndSetI)) return NULL;
 474     break;
 475 
 476   case vmIntrinsics::_getAndSetLong:
 477     if (!Matcher::match_rule_supported(Op_GetAndSetL)) return NULL;
 478     break;
 479 
 480   case vmIntrinsics::_getAndSetObject:
 481 #ifdef _LP64
 482     if (!UseCompressedOops && !Matcher::match_rule_supported(Op_GetAndSetP)) return NULL;
 483     if (UseCompressedOops && !Matcher::match_rule_supported(Op_GetAndSetN)) return NULL;
 484     break;
 485 #else
 486     if (!Matcher::match_rule_supported(Op_GetAndSetP)) return NULL;
 487     break;
 488 #endif
 489 
 490   case vmIntrinsics::_aescrypt_encryptBlock:
 491   case vmIntrinsics::_aescrypt_decryptBlock:
 492     if (!UseAESIntrinsics) return NULL;
 493     break;
 494 
 495   case vmIntrinsics::_multiplyToLen:
 496     if (!UseMultiplyToLenIntrinsic) return NULL;
 497     break;
 498 
 499   case vmIntrinsics::_squareToLen:
 500     if (!UseSquareToLenIntrinsic) return NULL;
 501     break;
 502 
 503   case vmIntrinsics::_mulAdd:
 504     if (!UseMulAddIntrinsic) return NULL;
 505     break;
 506 
 507   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 508   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 509     if (!UseAESIntrinsics) return NULL;
 510     // these two require the predicated logic
 511     predicates = 1;
 512     break;
 513 
 514   case vmIntrinsics::_sha_implCompress:
 515     if (!UseSHA1Intrinsics) return NULL;
 516     break;
 517 
 518   case vmIntrinsics::_sha2_implCompress:
 519     if (!UseSHA256Intrinsics) return NULL;
 520     break;
 521 
 522   case vmIntrinsics::_sha5_implCompress:
 523     if (!UseSHA512Intrinsics) return NULL;
 524     break;
 525 
 526   case vmIntrinsics::_digestBase_implCompressMB:
 527     if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) return NULL;
 528     predicates = 3;
 529     break;
 530 
 531   case vmIntrinsics::_updateCRC32:
 532   case vmIntrinsics::_updateBytesCRC32:
 533   case vmIntrinsics::_updateByteBufferCRC32:
 534     if (!UseCRC32Intrinsics) return NULL;
 535     break;
 536 
 537   case vmIntrinsics::_incrementExactI:
 538   case vmIntrinsics::_addExactI:
 539     if (!Matcher::match_rule_supported(Op_OverflowAddI) || !UseMathExactIntrinsics) return NULL;
 540     break;
 541   case vmIntrinsics::_incrementExactL:
 542   case vmIntrinsics::_addExactL:
 543     if (!Matcher::match_rule_supported(Op_OverflowAddL) || !UseMathExactIntrinsics) return NULL;
 544     break;
 545   case vmIntrinsics::_decrementExactI:
 546   case vmIntrinsics::_subtractExactI:
 547     if (!Matcher::match_rule_supported(Op_OverflowSubI) || !UseMathExactIntrinsics) return NULL;
 548     break;
 549   case vmIntrinsics::_decrementExactL:
 550   case vmIntrinsics::_subtractExactL:
 551     if (!Matcher::match_rule_supported(Op_OverflowSubL) || !UseMathExactIntrinsics) return NULL;
 552     break;
 553   case vmIntrinsics::_negateExactI:
 554     if (!Matcher::match_rule_supported(Op_OverflowSubI) || !UseMathExactIntrinsics) return NULL;
 555     break;
 556   case vmIntrinsics::_negateExactL:
 557     if (!Matcher::match_rule_supported(Op_OverflowSubL) || !UseMathExactIntrinsics) return NULL;
 558     break;
 559   case vmIntrinsics::_multiplyExactI:
 560     if (!Matcher::match_rule_supported(Op_OverflowMulI) || !UseMathExactIntrinsics) return NULL;
 561     break;
 562   case vmIntrinsics::_multiplyExactL:
 563     if (!Matcher::match_rule_supported(Op_OverflowMulL) || !UseMathExactIntrinsics) return NULL;
 564     break;
 565 
 566   case vmIntrinsics::_getShortUnaligned:
 567   case vmIntrinsics::_getCharUnaligned:
 568   case vmIntrinsics::_getIntUnaligned:
 569   case vmIntrinsics::_getLongUnaligned:
 570   case vmIntrinsics::_putShortUnaligned:
 571   case vmIntrinsics::_putCharUnaligned:
 572   case vmIntrinsics::_putIntUnaligned:
 573   case vmIntrinsics::_putLongUnaligned:
 574     if (!UseUnalignedAccesses) return NULL;
 575     break;
 576 
 577  default:
 578     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
 579     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
 580     break;
 581   }
 582 
 583   // -XX:-InlineClassNatives disables natives from the Class class.
 584   // The flag applies to all reflective calls, notably Array.newArray
 585   // (visible to Java programmers as Array.newInstance).
 586   if (m->holder()->name() == ciSymbol::java_lang_Class() ||
 587       m->holder()->name() == ciSymbol::java_lang_reflect_Array()) {
 588     if (!InlineClassNatives)  return NULL;
 589   }
 590 
 591   // -XX:-InlineThreadNatives disables natives from the Thread class.
 592   if (m->holder()->name() == ciSymbol::java_lang_Thread()) {
 593     if (!InlineThreadNatives)  return NULL;
 594   }
 595 
 596   // -XX:-InlineMathNatives disables natives from the Math,Float and Double classes.
 597   if (m->holder()->name() == ciSymbol::java_lang_Math() ||
 598       m->holder()->name() == ciSymbol::java_lang_Float() ||
 599       m->holder()->name() == ciSymbol::java_lang_Double()) {
 600     if (!InlineMathNatives)  return NULL;
 601   }
 602 
 603   // -XX:-InlineUnsafeOps disables natives from the Unsafe class.
 604   if (m->holder()->name() == ciSymbol::sun_misc_Unsafe()) {
 605     if (!InlineUnsafeOps)  return NULL;
 606   }
 607 
 608   return new LibraryIntrinsic(m, is_virtual, predicates, does_virtual_dispatch, (vmIntrinsics::ID) id);
 609 }
 610 
 611 //----------------------register_library_intrinsics-----------------------
 612 // Initialize this file's data structures, for each Compile instance.
 613 void Compile::register_library_intrinsics() {
 614   // Nothing to do here.
 615 }
 616 
 617 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
 618   LibraryCallKit kit(jvms, this);
 619   Compile* C = kit.C;
 620   int nodes = C->unique();
 621 #ifndef PRODUCT
 622   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 623     char buf[1000];
 624     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 625     tty->print_cr("Intrinsic %s", str);
 626   }
 627 #endif
 628   ciMethod* callee = kit.callee();
 629   const int bci    = kit.bci();
 630 
 631   // Try to inline the intrinsic.
 632   if (kit.try_to_inline(_last_predicate)) {
 633     if (C->print_intrinsics() || C->print_inlining()) {
 634       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
 635     }
 636     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 637     if (C->log()) {
 638       C->log()->elem("intrinsic id='%s'%s nodes='%d'",
 639                      vmIntrinsics::name_at(intrinsic_id()),
 640                      (is_virtual() ? " virtual='1'" : ""),
 641                      C->unique() - nodes);
 642     }
 643     // Push the result from the inlined method onto the stack.
 644     kit.push_result();
 645     C->print_inlining_update(this);
 646     return kit.transfer_exceptions_into_jvms();
 647   }
 648 
 649   // The intrinsic bailed out
 650   if (C->print_intrinsics() || C->print_inlining()) {
 651     if (jvms->has_method()) {
 652       // Not a root compile.
 653       const char* msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
 654       C->print_inlining(callee, jvms->depth() - 1, bci, msg);
 655     } else {
 656       // Root compile
 657       tty->print("Did not generate intrinsic %s%s at bci:%d in",
 658                vmIntrinsics::name_at(intrinsic_id()),
 659                (is_virtual() ? " (virtual)" : ""), bci);
 660     }
 661   }
 662   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 663   C->print_inlining_update(this);
 664   return NULL;
 665 }
 666 
 667 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
 668   LibraryCallKit kit(jvms, this);
 669   Compile* C = kit.C;
 670   int nodes = C->unique();
 671   _last_predicate = predicate;
 672 #ifndef PRODUCT
 673   assert(is_predicated() && predicate < predicates_count(), "sanity");
 674   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 675     char buf[1000];
 676     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 677     tty->print_cr("Predicate for intrinsic %s", str);
 678   }
 679 #endif
 680   ciMethod* callee = kit.callee();
 681   const int bci    = kit.bci();
 682 
 683   Node* slow_ctl = kit.try_to_predicate(predicate);
 684   if (!kit.failing()) {
 685     if (C->print_intrinsics() || C->print_inlining()) {
 686       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
 687     }
 688     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 689     if (C->log()) {
 690       C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
 691                      vmIntrinsics::name_at(intrinsic_id()),
 692                      (is_virtual() ? " virtual='1'" : ""),
 693                      C->unique() - nodes);
 694     }
 695     return slow_ctl; // Could be NULL if the check folds.
 696   }
 697 
 698   // The intrinsic bailed out
 699   if (C->print_intrinsics() || C->print_inlining()) {
 700     if (jvms->has_method()) {
 701       // Not a root compile.
 702       const char* msg = "failed to generate predicate for intrinsic";
 703       C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
 704     } else {
 705       // Root compile
 706       C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
 707                                         vmIntrinsics::name_at(intrinsic_id()),
 708                                         (is_virtual() ? " (virtual)" : ""), bci);
 709     }
 710   }
 711   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 712   return NULL;
 713 }
 714 
 715 bool LibraryCallKit::try_to_inline(int predicate) {
 716   // Handle symbolic names for otherwise undistinguished boolean switches:
 717   const bool is_store       = true;
 718   const bool is_native_ptr  = true;
 719   const bool is_static      = true;
 720   const bool is_volatile    = true;
 721 
 722   if (!jvms()->has_method()) {
 723     // Root JVMState has a null method.
 724     assert(map()->memory()->Opcode() == Op_Parm, "");
 725     // Insert the memory aliasing node
 726     set_all_memory(reset_memory());
 727   }
 728   assert(merged_memory(), "");
 729 
 730 
 731   switch (intrinsic_id()) {
 732   case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
 733   case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
 734   case vmIntrinsics::_getClass:                 return inline_native_getClass();
 735 
 736   case vmIntrinsics::_dsin:
 737   case vmIntrinsics::_dcos:
 738   case vmIntrinsics::_dtan:
 739   case vmIntrinsics::_dabs:
 740   case vmIntrinsics::_datan2:
 741   case vmIntrinsics::_dsqrt:
 742   case vmIntrinsics::_dexp:
 743   case vmIntrinsics::_dlog:
 744   case vmIntrinsics::_dlog10:
 745   case vmIntrinsics::_dpow:                     return inline_math_native(intrinsic_id());
 746 
 747   case vmIntrinsics::_min:
 748   case vmIntrinsics::_max:                      return inline_min_max(intrinsic_id());
 749 
 750   case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
 751   case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
 752   case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
 753   case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
 754   case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
 755   case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
 756   case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
 757   case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
 758   case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
 759   case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
 760   case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
 761   case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
 762 
 763   case vmIntrinsics::_arraycopy:                return inline_arraycopy();
 764 
 765   case vmIntrinsics::_compareTo:                return inline_string_compareTo();
 766   case vmIntrinsics::_indexOf:                  return inline_string_indexOf();
 767   case vmIntrinsics::_equals:                   return inline_string_equals();
 768 
 769   case vmIntrinsics::_getObject:                return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,  !is_volatile);
 770   case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, !is_volatile);
 771   case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,    !is_volatile);
 772   case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,   !is_volatile);
 773   case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,    !is_volatile);
 774   case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,     !is_volatile);
 775   case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,    !is_volatile);
 776   case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,   !is_volatile);
 777   case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,  !is_volatile);
 778 
 779   case vmIntrinsics::_putObject:                return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,  !is_volatile);
 780   case vmIntrinsics::_putBoolean:               return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN, !is_volatile);
 781   case vmIntrinsics::_putByte:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,    !is_volatile);
 782   case vmIntrinsics::_putShort:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,   !is_volatile);
 783   case vmIntrinsics::_putChar:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,    !is_volatile);
 784   case vmIntrinsics::_putInt:                   return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,     !is_volatile);
 785   case vmIntrinsics::_putLong:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,    !is_volatile);
 786   case vmIntrinsics::_putFloat:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,   !is_volatile);
 787   case vmIntrinsics::_putDouble:                return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,  !is_volatile);
 788 
 789   case vmIntrinsics::_getByte_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE,    !is_volatile);
 790   case vmIntrinsics::_getShort_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT,   !is_volatile);
 791   case vmIntrinsics::_getChar_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR,    !is_volatile);
 792   case vmIntrinsics::_getInt_raw:               return inline_unsafe_access( is_native_ptr, !is_store, T_INT,     !is_volatile);
 793   case vmIntrinsics::_getLong_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_LONG,    !is_volatile);
 794   case vmIntrinsics::_getFloat_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT,   !is_volatile);
 795   case vmIntrinsics::_getDouble_raw:            return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE,  !is_volatile);
 796   case vmIntrinsics::_getAddress_raw:           return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, !is_volatile);
 797 
 798   case vmIntrinsics::_putByte_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_BYTE,    !is_volatile);
 799   case vmIntrinsics::_putShort_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_SHORT,   !is_volatile);
 800   case vmIntrinsics::_putChar_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_CHAR,    !is_volatile);
 801   case vmIntrinsics::_putInt_raw:               return inline_unsafe_access( is_native_ptr,  is_store, T_INT,     !is_volatile);
 802   case vmIntrinsics::_putLong_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_LONG,    !is_volatile);
 803   case vmIntrinsics::_putFloat_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_FLOAT,   !is_volatile);
 804   case vmIntrinsics::_putDouble_raw:            return inline_unsafe_access( is_native_ptr,  is_store, T_DOUBLE,  !is_volatile);
 805   case vmIntrinsics::_putAddress_raw:           return inline_unsafe_access( is_native_ptr,  is_store, T_ADDRESS, !is_volatile);
 806 
 807   case vmIntrinsics::_getObjectVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   is_volatile);
 808   case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  is_volatile);
 809   case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     is_volatile);
 810   case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    is_volatile);
 811   case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     is_volatile);
 812   case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      is_volatile);
 813   case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     is_volatile);
 814   case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    is_volatile);
 815   case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   is_volatile);
 816 
 817   case vmIntrinsics::_putObjectVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   is_volatile);
 818   case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  is_volatile);
 819   case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     is_volatile);
 820   case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    is_volatile);
 821   case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     is_volatile);
 822   case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      is_volatile);
 823   case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     is_volatile);
 824   case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    is_volatile);
 825   case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   is_volatile);
 826 
 827   case vmIntrinsics::_getShortUnaligned:        return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,   !is_volatile);
 828   case vmIntrinsics::_getCharUnaligned:         return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,    !is_volatile);
 829   case vmIntrinsics::_getIntUnaligned:          return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,     !is_volatile);
 830   case vmIntrinsics::_getLongUnaligned:         return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,    !is_volatile);
 831 
 832   case vmIntrinsics::_putShortUnaligned:        return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,   !is_volatile);
 833   case vmIntrinsics::_putCharUnaligned:         return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,    !is_volatile);
 834   case vmIntrinsics::_putIntUnaligned:          return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,     !is_volatile);
 835   case vmIntrinsics::_putLongUnaligned:         return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,    !is_volatile);
 836 
 837   case vmIntrinsics::_compareAndSwapObject:     return inline_unsafe_load_store(T_OBJECT, LS_cmpxchg);
 838   case vmIntrinsics::_compareAndSwapInt:        return inline_unsafe_load_store(T_INT,    LS_cmpxchg);
 839   case vmIntrinsics::_compareAndSwapLong:       return inline_unsafe_load_store(T_LONG,   LS_cmpxchg);
 840 
 841   case vmIntrinsics::_putOrderedObject:         return inline_unsafe_ordered_store(T_OBJECT);
 842   case vmIntrinsics::_putOrderedInt:            return inline_unsafe_ordered_store(T_INT);
 843   case vmIntrinsics::_putOrderedLong:           return inline_unsafe_ordered_store(T_LONG);
 844 
 845   case vmIntrinsics::_getAndAddInt:             return inline_unsafe_load_store(T_INT,    LS_xadd);
 846   case vmIntrinsics::_getAndAddLong:            return inline_unsafe_load_store(T_LONG,   LS_xadd);
 847   case vmIntrinsics::_getAndSetInt:             return inline_unsafe_load_store(T_INT,    LS_xchg);
 848   case vmIntrinsics::_getAndSetLong:            return inline_unsafe_load_store(T_LONG,   LS_xchg);
 849   case vmIntrinsics::_getAndSetObject:          return inline_unsafe_load_store(T_OBJECT, LS_xchg);
 850 
 851   case vmIntrinsics::_loadFence:
 852   case vmIntrinsics::_storeFence:
 853   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 854 
 855   case vmIntrinsics::_currentThread:            return inline_native_currentThread();
 856   case vmIntrinsics::_isInterrupted:            return inline_native_isInterrupted();
 857 
 858 #ifdef TRACE_HAVE_INTRINSICS
 859   case vmIntrinsics::_classID:                  return inline_native_classID();
 860   case vmIntrinsics::_threadID:                 return inline_native_threadID();
 861   case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
 862 #endif
 863   case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
 864   case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
 865   case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
 866   case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
 867   case vmIntrinsics::_newArray:                 return inline_native_newArray();
 868   case vmIntrinsics::_getLength:                return inline_native_getLength();
 869   case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
 870   case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
 871   case vmIntrinsics::_equalsC:                  return inline_array_equals();
 872   case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());
 873 
 874   case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
 875 
 876   case vmIntrinsics::_isInstance:
 877   case vmIntrinsics::_getModifiers:
 878   case vmIntrinsics::_isInterface:
 879   case vmIntrinsics::_isArray:
 880   case vmIntrinsics::_isPrimitive:
 881   case vmIntrinsics::_getSuperclass:
 882   case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
 883 
 884   case vmIntrinsics::_floatToRawIntBits:
 885   case vmIntrinsics::_floatToIntBits:
 886   case vmIntrinsics::_intBitsToFloat:
 887   case vmIntrinsics::_doubleToRawLongBits:
 888   case vmIntrinsics::_doubleToLongBits:
 889   case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
 890 
 891   case vmIntrinsics::_numberOfLeadingZeros_i:
 892   case vmIntrinsics::_numberOfLeadingZeros_l:
 893   case vmIntrinsics::_numberOfTrailingZeros_i:
 894   case vmIntrinsics::_numberOfTrailingZeros_l:
 895   case vmIntrinsics::_bitCount_i:
 896   case vmIntrinsics::_bitCount_l:
 897   case vmIntrinsics::_reverseBytes_i:
 898   case vmIntrinsics::_reverseBytes_l:
 899   case vmIntrinsics::_reverseBytes_s:
 900   case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
 901 
 902   case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
 903 
 904   case vmIntrinsics::_Reference_get:            return inline_reference_get();
 905 
 906   case vmIntrinsics::_Class_cast:               return inline_Class_cast();
 907 
 908   case vmIntrinsics::_aescrypt_encryptBlock:
 909   case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
 910 
 911   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 912   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 913     return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
 914 
 915   case vmIntrinsics::_sha_implCompress:
 916   case vmIntrinsics::_sha2_implCompress:
 917   case vmIntrinsics::_sha5_implCompress:
 918     return inline_sha_implCompress(intrinsic_id());
 919 
 920   case vmIntrinsics::_digestBase_implCompressMB:
 921     return inline_digestBase_implCompressMB(predicate);
 922 
 923   case vmIntrinsics::_multiplyToLen:
 924     return inline_multiplyToLen();
 925 
 926   case vmIntrinsics::_squareToLen:
 927     return inline_squareToLen();
 928 
 929   case vmIntrinsics::_mulAdd:
 930     return inline_mulAdd();
 931 
 932   case vmIntrinsics::_encodeISOArray:
 933     return inline_encodeISOArray();
 934 
 935   case vmIntrinsics::_updateCRC32:
 936     return inline_updateCRC32();
 937   case vmIntrinsics::_updateBytesCRC32:
 938     return inline_updateBytesCRC32();
 939   case vmIntrinsics::_updateByteBufferCRC32:
 940     return inline_updateByteBufferCRC32();
 941 
 942   case vmIntrinsics::_profileBoolean:
 943     return inline_profileBoolean();
 944   case vmIntrinsics::_isCompileConstant:
 945     return inline_isCompileConstant();
 946 
 947   default:
 948     // If you get here, it may be that someone has added a new intrinsic
 949     // to the list in vmSymbols.hpp without implementing it here.
 950 #ifndef PRODUCT
 951     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 952       tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
 953                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 954     }
 955 #endif
 956     return false;
 957   }
 958 }
 959 
 960 Node* LibraryCallKit::try_to_predicate(int predicate) {
 961   if (!jvms()->has_method()) {
 962     // Root JVMState has a null method.
 963     assert(map()->memory()->Opcode() == Op_Parm, "");
 964     // Insert the memory aliasing node
 965     set_all_memory(reset_memory());
 966   }
 967   assert(merged_memory(), "");
 968 
 969   switch (intrinsic_id()) {
 970   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 971     return inline_cipherBlockChaining_AESCrypt_predicate(false);
 972   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 973     return inline_cipherBlockChaining_AESCrypt_predicate(true);
 974   case vmIntrinsics::_digestBase_implCompressMB:
 975     return inline_digestBase_implCompressMB_predicate(predicate);
 976 
 977   default:
 978     // If you get here, it may be that someone has added a new intrinsic
 979     // to the list in vmSymbols.hpp without implementing it here.
 980 #ifndef PRODUCT
 981     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 982       tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
 983                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 984     }
 985 #endif
 986     Node* slow_ctl = control();
 987     set_control(top()); // No fast path instrinsic
 988     return slow_ctl;
 989   }
 990 }
 991 
 992 //------------------------------set_result-------------------------------
 993 // Helper function for finishing intrinsics.
 994 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
 995   record_for_igvn(region);
 996   set_control(_gvn.transform(region));
 997   set_result( _gvn.transform(value));
 998   assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
 999 }
1000 
1001 //------------------------------generate_guard---------------------------
1002 // Helper function for generating guarded fast-slow graph structures.
1003 // The given 'test', if true, guards a slow path.  If the test fails
1004 // then a fast path can be taken.  (We generally hope it fails.)
1005 // In all cases, GraphKit::control() is updated to the fast path.
1006 // The returned value represents the control for the slow path.
1007 // The return value is never 'top'; it is either a valid control
1008 // or NULL if it is obvious that the slow path can never be taken.
1009 // Also, if region and the slow control are not NULL, the slow edge
1010 // is appended to the region.
1011 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
1012   if (stopped()) {
1013     // Already short circuited.
1014     return NULL;
1015   }
1016 
1017   // Build an if node and its projections.
1018   // If test is true we take the slow path, which we assume is uncommon.
1019   if (_gvn.type(test) == TypeInt::ZERO) {
1020     // The slow branch is never taken.  No need to build this guard.
1021     return NULL;
1022   }
1023 
1024   IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
1025 
1026   Node* if_slow = _gvn.transform(new IfTrueNode(iff));
1027   if (if_slow == top()) {
1028     // The slow branch is never taken.  No need to build this guard.
1029     return NULL;
1030   }
1031 
1032   if (region != NULL)
1033     region->add_req(if_slow);
1034 
1035   Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1036   set_control(if_fast);
1037 
1038   return if_slow;
1039 }
1040 
1041 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1042   return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1043 }
1044 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1045   return generate_guard(test, region, PROB_FAIR);
1046 }
1047 
1048 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1049                                                      Node* *pos_index) {
1050   if (stopped())
1051     return NULL;                // already stopped
1052   if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1053     return NULL;                // index is already adequately typed
1054   Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1055   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1056   Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1057   if (is_neg != NULL && pos_index != NULL) {
1058     // Emulate effect of Parse::adjust_map_after_if.
1059     Node* ccast = new CastIINode(index, TypeInt::POS);
1060     ccast->set_req(0, control());
1061     (*pos_index) = _gvn.transform(ccast);
1062   }
1063   return is_neg;
1064 }
1065 
1066 // Make sure that 'position' is a valid limit index, in [0..length].
1067 // There are two equivalent plans for checking this:
1068 //   A. (offset + copyLength)  unsigned<=  arrayLength
1069 //   B. offset  <=  (arrayLength - copyLength)
1070 // We require that all of the values above, except for the sum and
1071 // difference, are already known to be non-negative.
1072 // Plan A is robust in the face of overflow, if offset and copyLength
1073 // are both hugely positive.
1074 //
1075 // Plan B is less direct and intuitive, but it does not overflow at
1076 // all, since the difference of two non-negatives is always
1077 // representable.  Whenever Java methods must perform the equivalent
1078 // check they generally use Plan B instead of Plan A.
1079 // For the moment we use Plan A.
1080 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1081                                                   Node* subseq_length,
1082                                                   Node* array_length,
1083                                                   RegionNode* region) {
1084   if (stopped())
1085     return NULL;                // already stopped
1086   bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1087   if (zero_offset && subseq_length->eqv_uncast(array_length))
1088     return NULL;                // common case of whole-array copy
1089   Node* last = subseq_length;
1090   if (!zero_offset)             // last += offset
1091     last = _gvn.transform(new AddINode(last, offset));
1092   Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1093   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1094   Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1095   return is_over;
1096 }
1097 
1098 
1099 //--------------------------generate_current_thread--------------------
1100 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1101   ciKlass*    thread_klass = env()->Thread_klass();
1102   const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1103   Node* thread = _gvn.transform(new ThreadLocalNode());
1104   Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1105   Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1106   tls_output = thread;
1107   return threadObj;
1108 }
1109 
1110 
1111 //------------------------------make_string_method_node------------------------
1112 // Helper method for String intrinsic functions. This version is called
1113 // with str1 and str2 pointing to String object nodes.
1114 //
1115 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1, Node* str2) {
1116   Node* no_ctrl = NULL;
1117 
1118   // Get start addr of string
1119   Node* str1_value   = load_String_value(no_ctrl, str1);
1120   Node* str1_offset  = load_String_offset(no_ctrl, str1);
1121   Node* str1_start   = array_element_address(str1_value, str1_offset, T_CHAR);
1122 
1123   // Get length of string 1
1124   Node* str1_len  = load_String_length(no_ctrl, str1);
1125 
1126   Node* str2_value   = load_String_value(no_ctrl, str2);
1127   Node* str2_offset  = load_String_offset(no_ctrl, str2);
1128   Node* str2_start   = array_element_address(str2_value, str2_offset, T_CHAR);
1129 
1130   Node* str2_len = NULL;
1131   Node* result = NULL;
1132 
1133   switch (opcode) {
1134   case Op_StrIndexOf:
1135     // Get length of string 2
1136     str2_len = load_String_length(no_ctrl, str2);
1137 
1138     result = new StrIndexOfNode(control(), memory(TypeAryPtr::CHARS),
1139                                 str1_start, str1_len, str2_start, str2_len);
1140     break;
1141   case Op_StrComp:
1142     // Get length of string 2
1143     str2_len = load_String_length(no_ctrl, str2);
1144 
1145     result = new StrCompNode(control(), memory(TypeAryPtr::CHARS),
1146                              str1_start, str1_len, str2_start, str2_len);
1147     break;
1148   case Op_StrEquals:
1149     result = new StrEqualsNode(control(), memory(TypeAryPtr::CHARS),
1150                                str1_start, str2_start, str1_len);
1151     break;
1152   default:
1153     ShouldNotReachHere();
1154     return NULL;
1155   }
1156 
1157   // All these intrinsics have checks.
1158   C->set_has_split_ifs(true); // Has chance for split-if optimization
1159 
1160   return _gvn.transform(result);
1161 }
1162 
1163 // Helper method for String intrinsic functions. This version is called
1164 // with str1 and str2 pointing to char[] nodes, with cnt1 and cnt2 pointing
1165 // to Int nodes containing the lenghts of str1 and str2.
1166 //
1167 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2) {
1168   Node* result = NULL;
1169   switch (opcode) {
1170   case Op_StrIndexOf:
1171     result = new StrIndexOfNode(control(), memory(TypeAryPtr::CHARS),
1172                                 str1_start, cnt1, str2_start, cnt2);
1173     break;
1174   case Op_StrComp:
1175     result = new StrCompNode(control(), memory(TypeAryPtr::CHARS),
1176                              str1_start, cnt1, str2_start, cnt2);
1177     break;
1178   case Op_StrEquals:
1179     result = new StrEqualsNode(control(), memory(TypeAryPtr::CHARS),
1180                                str1_start, str2_start, cnt1);
1181     break;
1182   default:
1183     ShouldNotReachHere();
1184     return NULL;
1185   }
1186 
1187   // All these intrinsics have checks.
1188   C->set_has_split_ifs(true); // Has chance for split-if optimization
1189 
1190   return _gvn.transform(result);
1191 }
1192 
1193 //------------------------------inline_string_compareTo------------------------
1194 // public int java.lang.String.compareTo(String anotherString);
1195 bool LibraryCallKit::inline_string_compareTo() {
1196   Node* receiver = null_check(argument(0));
1197   Node* arg      = null_check(argument(1));
1198   if (stopped()) {
1199     return true;
1200   }
1201   set_result(make_string_method_node(Op_StrComp, receiver, arg));
1202   return true;
1203 }
1204 
1205 //------------------------------inline_string_equals------------------------
1206 bool LibraryCallKit::inline_string_equals() {
1207   Node* receiver = null_check_receiver();
1208   // NOTE: Do not null check argument for String.equals() because spec
1209   // allows to specify NULL as argument.
1210   Node* argument = this->argument(1);
1211   if (stopped()) {
1212     return true;
1213   }
1214 
1215   // paths (plus control) merge
1216   RegionNode* region = new RegionNode(5);
1217   Node* phi = new PhiNode(region, TypeInt::BOOL);
1218 
1219   // does source == target string?
1220   Node* cmp = _gvn.transform(new CmpPNode(receiver, argument));
1221   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1222 
1223   Node* if_eq = generate_slow_guard(bol, NULL);
1224   if (if_eq != NULL) {
1225     // receiver == argument
1226     phi->init_req(2, intcon(1));
1227     region->init_req(2, if_eq);
1228   }
1229 
1230   // get String klass for instanceOf
1231   ciInstanceKlass* klass = env()->String_klass();
1232 
1233   if (!stopped()) {
1234     Node* inst = gen_instanceof(argument, makecon(TypeKlassPtr::make(klass)));
1235     Node* cmp  = _gvn.transform(new CmpINode(inst, intcon(1)));
1236     Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1237 
1238     Node* inst_false = generate_guard(bol, NULL, PROB_MIN);
1239     //instanceOf == true, fallthrough
1240 
1241     if (inst_false != NULL) {
1242       phi->init_req(3, intcon(0));
1243       region->init_req(3, inst_false);
1244     }
1245   }
1246 
1247   if (!stopped()) {
1248     const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);
1249 
1250     // Properly cast the argument to String
1251     argument = _gvn.transform(new CheckCastPPNode(control(), argument, string_type));
1252     // This path is taken only when argument's type is String:NotNull.
1253     argument = cast_not_null(argument, false);
1254 
1255     Node* no_ctrl = NULL;
1256 
1257     // Get start addr of receiver
1258     Node* receiver_val    = load_String_value(no_ctrl, receiver);
1259     Node* receiver_offset = load_String_offset(no_ctrl, receiver);
1260     Node* receiver_start = array_element_address(receiver_val, receiver_offset, T_CHAR);
1261 
1262     // Get length of receiver
1263     Node* receiver_cnt  = load_String_length(no_ctrl, receiver);
1264 
1265     // Get start addr of argument
1266     Node* argument_val    = load_String_value(no_ctrl, argument);
1267     Node* argument_offset = load_String_offset(no_ctrl, argument);
1268     Node* argument_start = array_element_address(argument_val, argument_offset, T_CHAR);
1269 
1270     // Get length of argument
1271     Node* argument_cnt  = load_String_length(no_ctrl, argument);
1272 
1273     // Check for receiver count != argument count
1274     Node* cmp = _gvn.transform(new CmpINode(receiver_cnt, argument_cnt));
1275     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1276     Node* if_ne = generate_slow_guard(bol, NULL);
1277     if (if_ne != NULL) {
1278       phi->init_req(4, intcon(0));
1279       region->init_req(4, if_ne);
1280     }
1281 
1282     // Check for count == 0 is done by assembler code for StrEquals.
1283 
1284     if (!stopped()) {
1285       Node* equals = make_string_method_node(Op_StrEquals, receiver_start, receiver_cnt, argument_start, argument_cnt);
1286       phi->init_req(1, equals);
1287       region->init_req(1, control());
1288     }
1289   }
1290 
1291   // post merge
1292   set_control(_gvn.transform(region));
1293   record_for_igvn(region);
1294 
1295   set_result(_gvn.transform(phi));
1296   return true;
1297 }
1298 
1299 //------------------------------inline_array_equals----------------------------
1300 bool LibraryCallKit::inline_array_equals() {
1301   Node* arg1 = argument(0);
1302   Node* arg2 = argument(1);
1303   set_result(_gvn.transform(new AryEqNode(control(), memory(TypeAryPtr::CHARS), arg1, arg2)));
1304   return true;
1305 }
1306 
1307 // Java version of String.indexOf(constant string)
1308 // class StringDecl {
1309 //   StringDecl(char[] ca) {
1310 //     offset = 0;
1311 //     count = ca.length;
1312 //     value = ca;
1313 //   }
1314 //   int offset;
1315 //   int count;
1316 //   char[] value;
1317 // }
1318 //
1319 // static int string_indexOf_J(StringDecl string_object, char[] target_object,
1320 //                             int targetOffset, int cache_i, int md2) {
1321 //   int cache = cache_i;
1322 //   int sourceOffset = string_object.offset;
1323 //   int sourceCount = string_object.count;
1324 //   int targetCount = target_object.length;
1325 //
1326 //   int targetCountLess1 = targetCount - 1;
1327 //   int sourceEnd = sourceOffset + sourceCount - targetCountLess1;
1328 //
1329 //   char[] source = string_object.value;
1330 //   char[] target = target_object;
1331 //   int lastChar = target[targetCountLess1];
1332 //
1333 //  outer_loop:
1334 //   for (int i = sourceOffset; i < sourceEnd; ) {
1335 //     int src = source[i + targetCountLess1];
1336 //     if (src == lastChar) {
1337 //       // With random strings and a 4-character alphabet,
1338 //       // reverse matching at this point sets up 0.8% fewer
1339 //       // frames, but (paradoxically) makes 0.3% more probes.
1340 //       // Since those probes are nearer the lastChar probe,
1341 //       // there is may be a net D$ win with reverse matching.
1342 //       // But, reversing loop inhibits unroll of inner loop
1343 //       // for unknown reason.  So, does running outer loop from
1344 //       // (sourceOffset - targetCountLess1) to (sourceOffset + sourceCount)
1345 //       for (int j = 0; j < targetCountLess1; j++) {
1346 //         if (target[targetOffset + j] != source[i+j]) {
1347 //           if ((cache & (1 << source[i+j])) == 0) {
1348 //             if (md2 < j+1) {
1349 //               i += j+1;
1350 //               continue outer_loop;
1351 //             }
1352 //           }
1353 //           i += md2;
1354 //           continue outer_loop;
1355 //         }
1356 //       }
1357 //       return i - sourceOffset;
1358 //     }
1359 //     if ((cache & (1 << src)) == 0) {
1360 //       i += targetCountLess1;
1361 //     } // using "i += targetCount;" and an "else i++;" causes a jump to jump.
1362 //     i++;
1363 //   }
1364 //   return -1;
1365 // }
1366 
1367 //------------------------------string_indexOf------------------------
1368 Node* LibraryCallKit::string_indexOf(Node* string_object, ciTypeArray* target_array, jint targetOffset_i,
1369                                      jint cache_i, jint md2_i) {
1370 
1371   Node* no_ctrl  = NULL;
1372   float likely   = PROB_LIKELY(0.9);
1373   float unlikely = PROB_UNLIKELY(0.9);
1374 
1375   const int nargs = 0; // no arguments to push back for uncommon trap in predicate
1376 
1377   Node* source        = load_String_value(no_ctrl, string_object);
1378   Node* sourceOffset  = load_String_offset(no_ctrl, string_object);
1379   Node* sourceCount   = load_String_length(no_ctrl, string_object);
1380 
1381   Node* target = _gvn.transform( makecon(TypeOopPtr::make_from_constant(target_array, true)));
1382   jint target_length = target_array->length();
1383   const TypeAry* target_array_type = TypeAry::make(TypeInt::CHAR, TypeInt::make(0, target_length, Type::WidenMin));
1384   const TypeAryPtr* target_type = TypeAryPtr::make(TypePtr::BotPTR, target_array_type, target_array->klass(), true, Type::OffsetBot);
1385 
1386   // String.value field is known to be @Stable.
1387   if (UseImplicitStableValues) {
1388     target = cast_array_to_stable(target, target_type);
1389   }
1390 
1391   IdealKit kit(this, false, true);
1392 #define __ kit.
1393   Node* zero             = __ ConI(0);
1394   Node* one              = __ ConI(1);
1395   Node* cache            = __ ConI(cache_i);
1396   Node* md2              = __ ConI(md2_i);
1397   Node* lastChar         = __ ConI(target_array->char_at(target_length - 1));
1398   Node* targetCountLess1 = __ ConI(target_length - 1);
1399   Node* targetOffset     = __ ConI(targetOffset_i);
1400   Node* sourceEnd        = __ SubI(__ AddI(sourceOffset, sourceCount), targetCountLess1);
1401 
1402   IdealVariable rtn(kit), i(kit), j(kit); __ declarations_done();
1403   Node* outer_loop = __ make_label(2 /* goto */);
1404   Node* return_    = __ make_label(1);
1405 
1406   __ set(rtn,__ ConI(-1));
1407   __ loop(this, nargs, i, sourceOffset, BoolTest::lt, sourceEnd); {
1408        Node* i2  = __ AddI(__ value(i), targetCountLess1);
1409        // pin to prohibit loading of "next iteration" value which may SEGV (rare)
1410        Node* src = load_array_element(__ ctrl(), source, i2, TypeAryPtr::CHARS);
1411        __ if_then(src, BoolTest::eq, lastChar, unlikely); {
1412          __ loop(this, nargs, j, zero, BoolTest::lt, targetCountLess1); {
1413               Node* tpj = __ AddI(targetOffset, __ value(j));
1414               Node* targ = load_array_element(no_ctrl, target, tpj, target_type);
1415               Node* ipj  = __ AddI(__ value(i), __ value(j));
1416               Node* src2 = load_array_element(no_ctrl, source, ipj, TypeAryPtr::CHARS);
1417               __ if_then(targ, BoolTest::ne, src2); {
1418                 __ if_then(__ AndI(cache, __ LShiftI(one, src2)), BoolTest::eq, zero); {
1419                   __ if_then(md2, BoolTest::lt, __ AddI(__ value(j), one)); {
1420                     __ increment(i, __ AddI(__ value(j), one));
1421                     __ goto_(outer_loop);
1422                   } __ end_if(); __ dead(j);
1423                 }__ end_if(); __ dead(j);
1424                 __ increment(i, md2);
1425                 __ goto_(outer_loop);
1426               }__ end_if();
1427               __ increment(j, one);
1428          }__ end_loop(); __ dead(j);
1429          __ set(rtn, __ SubI(__ value(i), sourceOffset)); __ dead(i);
1430          __ goto_(return_);
1431        }__ end_if();
1432        __ if_then(__ AndI(cache, __ LShiftI(one, src)), BoolTest::eq, zero, likely); {
1433          __ increment(i, targetCountLess1);
1434        }__ end_if();
1435        __ increment(i, one);
1436        __ bind(outer_loop);
1437   }__ end_loop(); __ dead(i);
1438   __ bind(return_);
1439 
1440   // Final sync IdealKit and GraphKit.
1441   final_sync(kit);
1442   Node* result = __ value(rtn);
1443 #undef __
1444   C->set_has_loops(true);
1445   return result;
1446 }
1447 
1448 //------------------------------inline_string_indexOf------------------------
1449 bool LibraryCallKit::inline_string_indexOf() {
1450   Node* receiver = argument(0);
1451   Node* arg      = argument(1);
1452 
1453   Node* result;
1454   if (Matcher::has_match_rule(Op_StrIndexOf) &&
1455       UseSSE42Intrinsics) {
1456     // Generate SSE4.2 version of indexOf
1457     // We currently only have match rules that use SSE4.2
1458 
1459     receiver = null_check(receiver);
1460     arg      = null_check(arg);
1461     if (stopped()) {
1462       return true;
1463     }
1464 
1465     // Make the merge point
1466     RegionNode* result_rgn = new RegionNode(4);
1467     Node*       result_phi = new PhiNode(result_rgn, TypeInt::INT);
1468     Node* no_ctrl  = NULL;
1469 
1470     // Get start addr of source string
1471     Node* source = load_String_value(no_ctrl, receiver);
1472     Node* source_offset = load_String_offset(no_ctrl, receiver);
1473     Node* source_start = array_element_address(source, source_offset, T_CHAR);
1474 
1475     // Get length of source string
1476     Node* source_cnt  = load_String_length(no_ctrl, receiver);
1477 
1478     // Get start addr of substring
1479     Node* substr = load_String_value(no_ctrl, arg);
1480     Node* substr_offset = load_String_offset(no_ctrl, arg);
1481     Node* substr_start = array_element_address(substr, substr_offset, T_CHAR);
1482 
1483     // Get length of source string
1484     Node* substr_cnt  = load_String_length(no_ctrl, arg);
1485 
1486     // Check for substr count > string count
1487     Node* cmp = _gvn.transform(new CmpINode(substr_cnt, source_cnt));
1488     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1489     Node* if_gt = generate_slow_guard(bol, NULL);
1490     if (if_gt != NULL) {
1491       result_phi->init_req(2, intcon(-1));
1492       result_rgn->init_req(2, if_gt);
1493     }
1494 
1495     if (!stopped()) {
1496       // Check for substr count == 0
1497       cmp = _gvn.transform(new CmpINode(substr_cnt, intcon(0)));
1498       bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1499       Node* if_zero = generate_slow_guard(bol, NULL);
1500       if (if_zero != NULL) {
1501         result_phi->init_req(3, intcon(0));
1502         result_rgn->init_req(3, if_zero);
1503       }
1504     }
1505 
1506     if (!stopped()) {
1507       result = make_string_method_node(Op_StrIndexOf, source_start, source_cnt, substr_start, substr_cnt);
1508       result_phi->init_req(1, result);
1509       result_rgn->init_req(1, control());
1510     }
1511     set_control(_gvn.transform(result_rgn));
1512     record_for_igvn(result_rgn);
1513     result = _gvn.transform(result_phi);
1514 
1515   } else { // Use LibraryCallKit::string_indexOf
1516     // don't intrinsify if argument isn't a constant string.
1517     if (!arg->is_Con()) {
1518      return false;
1519     }
1520     const TypeOopPtr* str_type = _gvn.type(arg)->isa_oopptr();
1521     if (str_type == NULL) {
1522       return false;
1523     }
1524     ciInstanceKlass* klass = env()->String_klass();
1525     ciObject* str_const = str_type->const_oop();
1526     if (str_const == NULL || str_const->klass() != klass) {
1527       return false;
1528     }
1529     ciInstance* str = str_const->as_instance();
1530     assert(str != NULL, "must be instance");
1531 
1532     ciObject* v = str->field_value_by_offset(java_lang_String::value_offset_in_bytes()).as_object();
1533     ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array
1534 
1535     int o;
1536     int c;
1537     if (java_lang_String::has_offset_field()) {
1538       o = str->field_value_by_offset(java_lang_String::offset_offset_in_bytes()).as_int();
1539       c = str->field_value_by_offset(java_lang_String::count_offset_in_bytes()).as_int();
1540     } else {
1541       o = 0;
1542       c = pat->length();
1543     }
1544 
1545     // constant strings have no offset and count == length which
1546     // simplifies the resulting code somewhat so lets optimize for that.
1547     if (o != 0 || c != pat->length()) {
1548      return false;
1549     }
1550 
1551     receiver = null_check(receiver, T_OBJECT);
1552     // NOTE: No null check on the argument is needed since it's a constant String oop.
1553     if (stopped()) {
1554       return true;
1555     }
1556 
1557     // The null string as a pattern always returns 0 (match at beginning of string)
1558     if (c == 0) {
1559       set_result(intcon(0));
1560       return true;
1561     }
1562 
1563     // Generate default indexOf
1564     jchar lastChar = pat->char_at(o + (c - 1));
1565     int cache = 0;
1566     int i;
1567     for (i = 0; i < c - 1; i++) {
1568       assert(i < pat->length(), "out of range");
1569       cache |= (1 << (pat->char_at(o + i) & (sizeof(cache) * BitsPerByte - 1)));
1570     }
1571 
1572     int md2 = c;
1573     for (i = 0; i < c - 1; i++) {
1574       assert(i < pat->length(), "out of range");
1575       if (pat->char_at(o + i) == lastChar) {
1576         md2 = (c - 1) - i;
1577       }
1578     }
1579 
1580     result = string_indexOf(receiver, pat, o, cache, md2);
1581   }
1582   set_result(result);
1583   return true;
1584 }
1585 
1586 //--------------------------round_double_node--------------------------------
1587 // Round a double node if necessary.
1588 Node* LibraryCallKit::round_double_node(Node* n) {
1589   if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1590     n = _gvn.transform(new RoundDoubleNode(0, n));
1591   return n;
1592 }
1593 
1594 //------------------------------inline_math-----------------------------------
1595 // public static double Math.abs(double)
1596 // public static double Math.sqrt(double)
1597 // public static double Math.log(double)
1598 // public static double Math.log10(double)
1599 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1600   Node* arg = round_double_node(argument(0));
1601   Node* n;
1602   switch (id) {
1603   case vmIntrinsics::_dabs:   n = new AbsDNode(                arg);  break;
1604   case vmIntrinsics::_dsqrt:  n = new SqrtDNode(C, control(),  arg);  break;
1605   case vmIntrinsics::_dlog:   n = new LogDNode(C, control(),   arg);  break;
1606   case vmIntrinsics::_dlog10: n = new Log10DNode(C, control(), arg);  break;
1607   default:  fatal_unexpected_iid(id);  break;
1608   }
1609   set_result(_gvn.transform(n));
1610   return true;
1611 }
1612 
1613 //------------------------------inline_trig----------------------------------
1614 // Inline sin/cos/tan instructions, if possible.  If rounding is required, do
1615 // argument reduction which will turn into a fast/slow diamond.
1616 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1617   Node* arg = round_double_node(argument(0));
1618   Node* n = NULL;
1619 
1620   switch (id) {
1621   case vmIntrinsics::_dsin:  n = new SinDNode(C, control(), arg);  break;
1622   case vmIntrinsics::_dcos:  n = new CosDNode(C, control(), arg);  break;
1623   case vmIntrinsics::_dtan:  n = new TanDNode(C, control(), arg);  break;
1624   default:  fatal_unexpected_iid(id);  break;
1625   }
1626   n = _gvn.transform(n);
1627 
1628   // Rounding required?  Check for argument reduction!
1629   if (Matcher::strict_fp_requires_explicit_rounding) {
1630     static const double     pi_4 =  0.7853981633974483;
1631     static const double neg_pi_4 = -0.7853981633974483;
1632     // pi/2 in 80-bit extended precision
1633     // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1634     // -pi/2 in 80-bit extended precision
1635     // 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};
1636     // Cutoff value for using this argument reduction technique
1637     //static const double    pi_2_minus_epsilon =  1.564660403643354;
1638     //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1639 
1640     // Pseudocode for sin:
1641     // if (x <= Math.PI / 4.0) {
1642     //   if (x >= -Math.PI / 4.0) return  fsin(x);
1643     //   if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1644     // } else {
1645     //   if (x <=  Math.PI / 2.0) return  fcos(x - Math.PI / 2.0);
1646     // }
1647     // return StrictMath.sin(x);
1648 
1649     // Pseudocode for cos:
1650     // if (x <= Math.PI / 4.0) {
1651     //   if (x >= -Math.PI / 4.0) return  fcos(x);
1652     //   if (x >= -Math.PI / 2.0) return  fsin(x + Math.PI / 2.0);
1653     // } else {
1654     //   if (x <=  Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1655     // }
1656     // return StrictMath.cos(x);
1657 
1658     // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1659     // requires a special machine instruction to load it.  Instead we'll try
1660     // the 'easy' case.  If we really need the extra range +/- PI/2 we'll
1661     // probably do the math inside the SIN encoding.
1662 
1663     // Make the merge point
1664     RegionNode* r = new RegionNode(3);
1665     Node* phi = new PhiNode(r, Type::DOUBLE);
1666 
1667     // Flatten arg so we need only 1 test
1668     Node *abs = _gvn.transform(new AbsDNode(arg));
1669     // Node for PI/4 constant
1670     Node *pi4 = makecon(TypeD::make(pi_4));
1671     // Check PI/4 : abs(arg)
1672     Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1673     // Check: If PI/4 < abs(arg) then go slow
1674     Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1675     // Branch either way
1676     IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1677     set_control(opt_iff(r,iff));
1678 
1679     // Set fast path result
1680     phi->init_req(2, n);
1681 
1682     // Slow path - non-blocking leaf call
1683     Node* call = NULL;
1684     switch (id) {
1685     case vmIntrinsics::_dsin:
1686       call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1687                                CAST_FROM_FN_PTR(address, SharedRuntime::dsin),
1688                                "Sin", NULL, arg, top());
1689       break;
1690     case vmIntrinsics::_dcos:
1691       call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1692                                CAST_FROM_FN_PTR(address, SharedRuntime::dcos),
1693                                "Cos", NULL, arg, top());
1694       break;
1695     case vmIntrinsics::_dtan:
1696       call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1697                                CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1698                                "Tan", NULL, arg, top());
1699       break;
1700     }
1701     assert(control()->in(0) == call, "");
1702     Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1703     r->init_req(1, control());
1704     phi->init_req(1, slow_result);
1705 
1706     // Post-merge
1707     set_control(_gvn.transform(r));
1708     record_for_igvn(r);
1709     n = _gvn.transform(phi);
1710 
1711     C->set_has_split_ifs(true); // Has chance for split-if optimization
1712   }
1713   set_result(n);
1714   return true;
1715 }
1716 
1717 Node* LibraryCallKit::finish_pow_exp(Node* result, Node* x, Node* y, const TypeFunc* call_type, address funcAddr, const char* funcName) {
1718   //-------------------
1719   //result=(result.isNaN())? funcAddr():result;
1720   // Check: If isNaN() by checking result!=result? then either trap
1721   // or go to runtime
1722   Node* cmpisnan = _gvn.transform(new CmpDNode(result, result));
1723   // Build the boolean node
1724   Node* bolisnum = _gvn.transform(new BoolNode(cmpisnan, BoolTest::eq));
1725 
1726   if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
1727     { BuildCutout unless(this, bolisnum, PROB_STATIC_FREQUENT);
1728       // The pow or exp intrinsic returned a NaN, which requires a call
1729       // to the runtime.  Recompile with the runtime call.
1730       uncommon_trap(Deoptimization::Reason_intrinsic,
1731                     Deoptimization::Action_make_not_entrant);
1732     }
1733     return result;
1734   } else {
1735     // If this inlining ever returned NaN in the past, we compile a call
1736     // to the runtime to properly handle corner cases
1737 
1738     IfNode* iff = create_and_xform_if(control(), bolisnum, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1739     Node* if_slow = _gvn.transform(new IfFalseNode(iff));
1740     Node* if_fast = _gvn.transform(new IfTrueNode(iff));
1741 
1742     if (!if_slow->is_top()) {
1743       RegionNode* result_region = new RegionNode(3);
1744       PhiNode*    result_val = new PhiNode(result_region, Type::DOUBLE);
1745 
1746       result_region->init_req(1, if_fast);
1747       result_val->init_req(1, result);
1748 
1749       set_control(if_slow);
1750 
1751       const TypePtr* no_memory_effects = NULL;
1752       Node* rt = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1753                                    no_memory_effects,
1754                                    x, top(), y, y ? top() : NULL);
1755       Node* value = _gvn.transform(new ProjNode(rt, TypeFunc::Parms+0));
1756 #ifdef ASSERT
1757       Node* value_top = _gvn.transform(new ProjNode(rt, TypeFunc::Parms+1));
1758       assert(value_top == top(), "second value must be top");
1759 #endif
1760 
1761       result_region->init_req(2, control());
1762       result_val->init_req(2, value);
1763       set_control(_gvn.transform(result_region));
1764       return _gvn.transform(result_val);
1765     } else {
1766       return result;
1767     }
1768   }
1769 }
1770 
1771 //------------------------------inline_exp-------------------------------------
1772 // Inline exp instructions, if possible.  The Intel hardware only misses
1773 // really odd corner cases (+/- Infinity).  Just uncommon-trap them.
1774 bool LibraryCallKit::inline_exp() {
1775   Node* arg = round_double_node(argument(0));
1776   Node* n   = _gvn.transform(new ExpDNode(C, control(), arg));
1777 
1778   n = finish_pow_exp(n, arg, NULL, OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1779   set_result(n);
1780 
1781   C->set_has_split_ifs(true); // Has chance for split-if optimization
1782   return true;
1783 }
1784 
1785 //------------------------------inline_pow-------------------------------------
1786 // Inline power instructions, if possible.
1787 bool LibraryCallKit::inline_pow() {
1788   // Pseudocode for pow
1789   // if (y == 2) {
1790   //   return x * x;
1791   // } else {
1792   //   if (x <= 0.0) {
1793   //     long longy = (long)y;
1794   //     if ((double)longy == y) { // if y is long
1795   //       if (y + 1 == y) longy = 0; // huge number: even
1796   //       result = ((1&longy) == 0)?-DPow(abs(x), y):DPow(abs(x), y);
1797   //     } else {
1798   //       result = NaN;
1799   //     }
1800   //   } else {
1801   //     result = DPow(x,y);
1802   //   }
1803   //   if (result != result)?  {
1804   //     result = uncommon_trap() or runtime_call();
1805   //   }
1806   //   return result;
1807   // }
1808 
1809   Node* x = round_double_node(argument(0));
1810   Node* y = round_double_node(argument(2));
1811 
1812   Node* result = NULL;
1813 
1814   Node*   const_two_node = makecon(TypeD::make(2.0));
1815   Node*   cmp_node       = _gvn.transform(new CmpDNode(y, const_two_node));
1816   Node*   bool_node      = _gvn.transform(new BoolNode(cmp_node, BoolTest::eq));
1817   IfNode* if_node        = create_and_xform_if(control(), bool_node, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
1818   Node*   if_true        = _gvn.transform(new IfTrueNode(if_node));
1819   Node*   if_false       = _gvn.transform(new IfFalseNode(if_node));
1820 
1821   RegionNode* region_node = new RegionNode(3);
1822   region_node->init_req(1, if_true);
1823 
1824   Node* phi_node = new PhiNode(region_node, Type::DOUBLE);
1825   // special case for x^y where y == 2, we can convert it to x * x
1826   phi_node->init_req(1, _gvn.transform(new MulDNode(x, x)));
1827 
1828   // set control to if_false since we will now process the false branch
1829   set_control(if_false);
1830 
1831   if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
1832     // Short form: skip the fancy tests and just check for NaN result.
1833     result = _gvn.transform(new PowDNode(C, control(), x, y));
1834   } else {
1835     // If this inlining ever returned NaN in the past, include all
1836     // checks + call to the runtime.
1837 
1838     // Set the merge point for If node with condition of (x <= 0.0)
1839     // There are four possible paths to region node and phi node
1840     RegionNode *r = new RegionNode(4);
1841     Node *phi = new PhiNode(r, Type::DOUBLE);
1842 
1843     // Build the first if node: if (x <= 0.0)
1844     // Node for 0 constant
1845     Node *zeronode = makecon(TypeD::ZERO);
1846     // Check x:0
1847     Node *cmp = _gvn.transform(new CmpDNode(x, zeronode));
1848     // Check: If (x<=0) then go complex path
1849     Node *bol1 = _gvn.transform(new BoolNode( cmp, BoolTest::le ));
1850     // Branch either way
1851     IfNode *if1 = create_and_xform_if(control(),bol1, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
1852     // Fast path taken; set region slot 3
1853     Node *fast_taken = _gvn.transform(new IfFalseNode(if1));
1854     r->init_req(3,fast_taken); // Capture fast-control
1855 
1856     // Fast path not-taken, i.e. slow path
1857     Node *complex_path = _gvn.transform(new IfTrueNode(if1));
1858 
1859     // Set fast path result
1860     Node *fast_result = _gvn.transform(new PowDNode(C, control(), x, y));
1861     phi->init_req(3, fast_result);
1862 
1863     // Complex path
1864     // Build the second if node (if y is long)
1865     // Node for (long)y
1866     Node *longy = _gvn.transform(new ConvD2LNode(y));
1867     // Node for (double)((long) y)
1868     Node *doublelongy= _gvn.transform(new ConvL2DNode(longy));
1869     // Check (double)((long) y) : y
1870     Node *cmplongy= _gvn.transform(new CmpDNode(doublelongy, y));
1871     // Check if (y isn't long) then go to slow path
1872 
1873     Node *bol2 = _gvn.transform(new BoolNode( cmplongy, BoolTest::ne ));
1874     // Branch either way
1875     IfNode *if2 = create_and_xform_if(complex_path,bol2, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
1876     Node* ylong_path = _gvn.transform(new IfFalseNode(if2));
1877 
1878     Node *slow_path = _gvn.transform(new IfTrueNode(if2));
1879 
1880     // Calculate DPow(abs(x), y)*(1 & (long)y)
1881     // Node for constant 1
1882     Node *conone = longcon(1);
1883     // 1& (long)y
1884     Node *signnode= _gvn.transform(new AndLNode(conone, longy));
1885 
1886     // A huge number is always even. Detect a huge number by checking
1887     // if y + 1 == y and set integer to be tested for parity to 0.
1888     // Required for corner case:
1889     // (long)9.223372036854776E18 = max_jlong
1890     // (double)(long)9.223372036854776E18 = 9.223372036854776E18
1891     // max_jlong is odd but 9.223372036854776E18 is even
1892     Node* yplus1 = _gvn.transform(new AddDNode(y, makecon(TypeD::make(1))));
1893     Node *cmpyplus1= _gvn.transform(new CmpDNode(yplus1, y));
1894     Node *bolyplus1 = _gvn.transform(new BoolNode( cmpyplus1, BoolTest::eq ));
1895     Node* correctedsign = NULL;
1896     if (ConditionalMoveLimit != 0) {
1897       correctedsign = _gvn.transform(CMoveNode::make(NULL, bolyplus1, signnode, longcon(0), TypeLong::LONG));
1898     } else {
1899       IfNode *ifyplus1 = create_and_xform_if(ylong_path,bolyplus1, PROB_FAIR, COUNT_UNKNOWN);
1900       RegionNode *r = new RegionNode(3);
1901       Node *phi = new PhiNode(r, TypeLong::LONG);
1902       r->init_req(1, _gvn.transform(new IfFalseNode(ifyplus1)));
1903       r->init_req(2, _gvn.transform(new IfTrueNode(ifyplus1)));
1904       phi->init_req(1, signnode);
1905       phi->init_req(2, longcon(0));
1906       correctedsign = _gvn.transform(phi);
1907       ylong_path = _gvn.transform(r);
1908       record_for_igvn(r);
1909     }
1910 
1911     // zero node
1912     Node *conzero = longcon(0);
1913     // Check (1&(long)y)==0?
1914     Node *cmpeq1 = _gvn.transform(new CmpLNode(correctedsign, conzero));
1915     // Check if (1&(long)y)!=0?, if so the result is negative
1916     Node *bol3 = _gvn.transform(new BoolNode( cmpeq1, BoolTest::ne ));
1917     // abs(x)
1918     Node *absx=_gvn.transform(new AbsDNode(x));
1919     // abs(x)^y
1920     Node *absxpowy = _gvn.transform(new PowDNode(C, control(), absx, y));
1921     // -abs(x)^y
1922     Node *negabsxpowy = _gvn.transform(new NegDNode (absxpowy));
1923     // (1&(long)y)==1?-DPow(abs(x), y):DPow(abs(x), y)
1924     Node *signresult = NULL;
1925     if (ConditionalMoveLimit != 0) {
1926       signresult = _gvn.transform(CMoveNode::make(NULL, bol3, absxpowy, negabsxpowy, Type::DOUBLE));
1927     } else {
1928       IfNode *ifyeven = create_and_xform_if(ylong_path,bol3, PROB_FAIR, COUNT_UNKNOWN);
1929       RegionNode *r = new RegionNode(3);
1930       Node *phi = new PhiNode(r, Type::DOUBLE);
1931       r->init_req(1, _gvn.transform(new IfFalseNode(ifyeven)));
1932       r->init_req(2, _gvn.transform(new IfTrueNode(ifyeven)));
1933       phi->init_req(1, absxpowy);
1934       phi->init_req(2, negabsxpowy);
1935       signresult = _gvn.transform(phi);
1936       ylong_path = _gvn.transform(r);
1937       record_for_igvn(r);
1938     }
1939     // Set complex path fast result
1940     r->init_req(2, ylong_path);
1941     phi->init_req(2, signresult);
1942 
1943     static const jlong nan_bits = CONST64(0x7ff8000000000000);
1944     Node *slow_result = makecon(TypeD::make(*(double*)&nan_bits)); // return NaN
1945     r->init_req(1,slow_path);
1946     phi->init_req(1,slow_result);
1947 
1948     // Post merge
1949     set_control(_gvn.transform(r));
1950     record_for_igvn(r);
1951     result = _gvn.transform(phi);
1952   }
1953 
1954   result = finish_pow_exp(result, x, y, OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1955 
1956   // control from finish_pow_exp is now input to the region node
1957   region_node->set_req(2, control());
1958   // the result from finish_pow_exp is now input to the phi node
1959   phi_node->init_req(2, result);
1960   set_control(_gvn.transform(region_node));
1961   record_for_igvn(region_node);
1962   set_result(_gvn.transform(phi_node));
1963 
1964   C->set_has_split_ifs(true); // Has chance for split-if optimization
1965   return true;
1966 }
1967 
1968 //------------------------------runtime_math-----------------------------
1969 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1970   assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1971          "must be (DD)D or (D)D type");
1972 
1973   // Inputs
1974   Node* a = round_double_node(argument(0));
1975   Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1976 
1977   const TypePtr* no_memory_effects = NULL;
1978   Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1979                                  no_memory_effects,
1980                                  a, top(), b, b ? top() : NULL);
1981   Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1982 #ifdef ASSERT
1983   Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1984   assert(value_top == top(), "second value must be top");
1985 #endif
1986 
1987   set_result(value);
1988   return true;
1989 }
1990 
1991 //------------------------------inline_math_native-----------------------------
1992 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1993 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1994   switch (id) {
1995     // These intrinsics are not properly supported on all hardware
1996   case vmIntrinsics::_dcos:   return Matcher::has_match_rule(Op_CosD)   ? inline_trig(id) :
1997     runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
1998   case vmIntrinsics::_dsin:   return Matcher::has_match_rule(Op_SinD)   ? inline_trig(id) :
1999     runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
2000   case vmIntrinsics::_dtan:   return Matcher::has_match_rule(Op_TanD)   ? inline_trig(id) :
2001     runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan),   "TAN");
2002 
2003   case vmIntrinsics::_dlog:   return Matcher::has_match_rule(Op_LogD)   ? inline_math(id) :
2004     runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
2005   case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
2006     runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
2007 
2008     // These intrinsics are supported on all hardware
2009   case vmIntrinsics::_dsqrt:  return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
2010   case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_math(id) : false;
2011 
2012   case vmIntrinsics::_dexp:   return Matcher::has_match_rule(Op_ExpD)   ? inline_exp()    :
2013     runtime_math(OptoRuntime::Math_D_D_Type(),  FN_PTR(SharedRuntime::dexp),  "EXP");
2014   case vmIntrinsics::_dpow:   return Matcher::has_match_rule(Op_PowD)   ? inline_pow()    :
2015     runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow),  "POW");
2016 #undef FN_PTR
2017 
2018    // These intrinsics are not yet correctly implemented
2019   case vmIntrinsics::_datan2:
2020     return false;
2021 
2022   default:
2023     fatal_unexpected_iid(id);
2024     return false;
2025   }
2026 }
2027 
2028 static bool is_simple_name(Node* n) {
2029   return (n->req() == 1         // constant
2030           || (n->is_Type() && n->as_Type()->type()->singleton())
2031           || n->is_Proj()       // parameter or return value
2032           || n->is_Phi()        // local of some sort
2033           );
2034 }
2035 
2036 //----------------------------inline_min_max-----------------------------------
2037 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
2038   set_result(generate_min_max(id, argument(0), argument(1)));
2039   return true;
2040 }
2041 
2042 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
2043   Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2044   IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2045   Node* fast_path = _gvn.transform( new IfFalseNode(check));
2046   Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2047 
2048   {
2049     PreserveJVMState pjvms(this);
2050     PreserveReexecuteState preexecs(this);
2051     jvms()->set_should_reexecute(true);
2052 
2053     set_control(slow_path);
2054     set_i_o(i_o());
2055 
2056     uncommon_trap(Deoptimization::Reason_intrinsic,
2057                   Deoptimization::Action_none);
2058   }
2059 
2060   set_control(fast_path);
2061   set_result(math);
2062 }
2063 
2064 template <typename OverflowOp>
2065 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2066   typedef typename OverflowOp::MathOp MathOp;
2067 
2068   MathOp* mathOp = new MathOp(arg1, arg2);
2069   Node* operation = _gvn.transform( mathOp );
2070   Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2071   inline_math_mathExact(operation, ofcheck);
2072   return true;
2073 }
2074 
2075 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2076   return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2077 }
2078 
2079 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2080   return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2081 }
2082 
2083 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2084   return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2085 }
2086 
2087 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2088   return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2089 }
2090 
2091 bool LibraryCallKit::inline_math_negateExactI() {
2092   return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2093 }
2094 
2095 bool LibraryCallKit::inline_math_negateExactL() {
2096   return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2097 }
2098 
2099 bool LibraryCallKit::inline_math_multiplyExactI() {
2100   return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2101 }
2102 
2103 bool LibraryCallKit::inline_math_multiplyExactL() {
2104   return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2105 }
2106 
2107 Node*
2108 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2109   // These are the candidate return value:
2110   Node* xvalue = x0;
2111   Node* yvalue = y0;
2112 
2113   if (xvalue == yvalue) {
2114     return xvalue;
2115   }
2116 
2117   bool want_max = (id == vmIntrinsics::_max);
2118 
2119   const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2120   const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2121   if (txvalue == NULL || tyvalue == NULL)  return top();
2122   // This is not really necessary, but it is consistent with a
2123   // hypothetical MaxINode::Value method:
2124   int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2125 
2126   // %%% This folding logic should (ideally) be in a different place.
2127   // Some should be inside IfNode, and there to be a more reliable
2128   // transformation of ?: style patterns into cmoves.  We also want
2129   // more powerful optimizations around cmove and min/max.
2130 
2131   // Try to find a dominating comparison of these guys.
2132   // It can simplify the index computation for Arrays.copyOf
2133   // and similar uses of System.arraycopy.
2134   // First, compute the normalized version of CmpI(x, y).
2135   int   cmp_op = Op_CmpI;
2136   Node* xkey = xvalue;
2137   Node* ykey = yvalue;
2138   Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2139   if (ideal_cmpxy->is_Cmp()) {
2140     // E.g., if we have CmpI(length - offset, count),
2141     // it might idealize to CmpI(length, count + offset)
2142     cmp_op = ideal_cmpxy->Opcode();
2143     xkey = ideal_cmpxy->in(1);
2144     ykey = ideal_cmpxy->in(2);
2145   }
2146 
2147   // Start by locating any relevant comparisons.
2148   Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2149   Node* cmpxy = NULL;
2150   Node* cmpyx = NULL;
2151   for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2152     Node* cmp = start_from->fast_out(k);
2153     if (cmp->outcnt() > 0 &&            // must have prior uses
2154         cmp->in(0) == NULL &&           // must be context-independent
2155         cmp->Opcode() == cmp_op) {      // right kind of compare
2156       if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
2157       if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
2158     }
2159   }
2160 
2161   const int NCMPS = 2;
2162   Node* cmps[NCMPS] = { cmpxy, cmpyx };
2163   int cmpn;
2164   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2165     if (cmps[cmpn] != NULL)  break;     // find a result
2166   }
2167   if (cmpn < NCMPS) {
2168     // Look for a dominating test that tells us the min and max.
2169     int depth = 0;                // Limit search depth for speed
2170     Node* dom = control();
2171     for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2172       if (++depth >= 100)  break;
2173       Node* ifproj = dom;
2174       if (!ifproj->is_Proj())  continue;
2175       Node* iff = ifproj->in(0);
2176       if (!iff->is_If())  continue;
2177       Node* bol = iff->in(1);
2178       if (!bol->is_Bool())  continue;
2179       Node* cmp = bol->in(1);
2180       if (cmp == NULL)  continue;
2181       for (cmpn = 0; cmpn < NCMPS; cmpn++)
2182         if (cmps[cmpn] == cmp)  break;
2183       if (cmpn == NCMPS)  continue;
2184       BoolTest::mask btest = bol->as_Bool()->_test._test;
2185       if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
2186       if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
2187       // At this point, we know that 'x btest y' is true.
2188       switch (btest) {
2189       case BoolTest::eq:
2190         // They are proven equal, so we can collapse the min/max.
2191         // Either value is the answer.  Choose the simpler.
2192         if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2193           return yvalue;
2194         return xvalue;
2195       case BoolTest::lt:          // x < y
2196       case BoolTest::le:          // x <= y
2197         return (want_max ? yvalue : xvalue);
2198       case BoolTest::gt:          // x > y
2199       case BoolTest::ge:          // x >= y
2200         return (want_max ? xvalue : yvalue);
2201       }
2202     }
2203   }
2204 
2205   // We failed to find a dominating test.
2206   // Let's pick a test that might GVN with prior tests.
2207   Node*          best_bol   = NULL;
2208   BoolTest::mask best_btest = BoolTest::illegal;
2209   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2210     Node* cmp = cmps[cmpn];
2211     if (cmp == NULL)  continue;
2212     for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2213       Node* bol = cmp->fast_out(j);
2214       if (!bol->is_Bool())  continue;
2215       BoolTest::mask btest = bol->as_Bool()->_test._test;
2216       if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
2217       if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
2218       if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2219         best_bol   = bol->as_Bool();
2220         best_btest = btest;
2221       }
2222     }
2223   }
2224 
2225   Node* answer_if_true  = NULL;
2226   Node* answer_if_false = NULL;
2227   switch (best_btest) {
2228   default:
2229     if (cmpxy == NULL)
2230       cmpxy = ideal_cmpxy;
2231     best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2232     // and fall through:
2233   case BoolTest::lt:          // x < y
2234   case BoolTest::le:          // x <= y
2235     answer_if_true  = (want_max ? yvalue : xvalue);
2236     answer_if_false = (want_max ? xvalue : yvalue);
2237     break;
2238   case BoolTest::gt:          // x > y
2239   case BoolTest::ge:          // x >= y
2240     answer_if_true  = (want_max ? xvalue : yvalue);
2241     answer_if_false = (want_max ? yvalue : xvalue);
2242     break;
2243   }
2244 
2245   jint hi, lo;
2246   if (want_max) {
2247     // We can sharpen the minimum.
2248     hi = MAX2(txvalue->_hi, tyvalue->_hi);
2249     lo = MAX2(txvalue->_lo, tyvalue->_lo);
2250   } else {
2251     // We can sharpen the maximum.
2252     hi = MIN2(txvalue->_hi, tyvalue->_hi);
2253     lo = MIN2(txvalue->_lo, tyvalue->_lo);
2254   }
2255 
2256   // Use a flow-free graph structure, to avoid creating excess control edges
2257   // which could hinder other optimizations.
2258   // Since Math.min/max is often used with arraycopy, we want
2259   // tightly_coupled_allocation to be able to see beyond min/max expressions.
2260   Node* cmov = CMoveNode::make(NULL, best_bol,
2261                                answer_if_false, answer_if_true,
2262                                TypeInt::make(lo, hi, widen));
2263 
2264   return _gvn.transform(cmov);
2265 
2266   /*
2267   // This is not as desirable as it may seem, since Min and Max
2268   // nodes do not have a full set of optimizations.
2269   // And they would interfere, anyway, with 'if' optimizations
2270   // and with CMoveI canonical forms.
2271   switch (id) {
2272   case vmIntrinsics::_min:
2273     result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2274   case vmIntrinsics::_max:
2275     result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2276   default:
2277     ShouldNotReachHere();
2278   }
2279   */
2280 }
2281 
2282 inline int
2283 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2284   const TypePtr* base_type = TypePtr::NULL_PTR;
2285   if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
2286   if (base_type == NULL) {
2287     // Unknown type.
2288     return Type::AnyPtr;
2289   } else if (base_type == TypePtr::NULL_PTR) {
2290     // Since this is a NULL+long form, we have to switch to a rawptr.
2291     base   = _gvn.transform(new CastX2PNode(offset));
2292     offset = MakeConX(0);
2293     return Type::RawPtr;
2294   } else if (base_type->base() == Type::RawPtr) {
2295     return Type::RawPtr;
2296   } else if (base_type->isa_oopptr()) {
2297     // Base is never null => always a heap address.
2298     if (base_type->ptr() == TypePtr::NotNull) {
2299       return Type::OopPtr;
2300     }
2301     // Offset is small => always a heap address.
2302     const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2303     if (offset_type != NULL &&
2304         base_type->offset() == 0 &&     // (should always be?)
2305         offset_type->_lo >= 0 &&
2306         !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2307       return Type::OopPtr;
2308     }
2309     // Otherwise, it might either be oop+off or NULL+addr.
2310     return Type::AnyPtr;
2311   } else {
2312     // No information:
2313     return Type::AnyPtr;
2314   }
2315 }
2316 
2317 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2318   int kind = classify_unsafe_addr(base, offset);
2319   if (kind == Type::RawPtr) {
2320     return basic_plus_adr(top(), base, offset);
2321   } else {
2322     return basic_plus_adr(base, offset);
2323   }
2324 }
2325 
2326 //--------------------------inline_number_methods-----------------------------
2327 // inline int     Integer.numberOfLeadingZeros(int)
2328 // inline int        Long.numberOfLeadingZeros(long)
2329 //
2330 // inline int     Integer.numberOfTrailingZeros(int)
2331 // inline int        Long.numberOfTrailingZeros(long)
2332 //
2333 // inline int     Integer.bitCount(int)
2334 // inline int        Long.bitCount(long)
2335 //
2336 // inline char  Character.reverseBytes(char)
2337 // inline short     Short.reverseBytes(short)
2338 // inline int     Integer.reverseBytes(int)
2339 // inline long       Long.reverseBytes(long)
2340 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2341   Node* arg = argument(0);
2342   Node* n;
2343   switch (id) {
2344   case vmIntrinsics::_numberOfLeadingZeros_i:   n = new CountLeadingZerosINode( arg);  break;
2345   case vmIntrinsics::_numberOfLeadingZeros_l:   n = new CountLeadingZerosLNode( arg);  break;
2346   case vmIntrinsics::_numberOfTrailingZeros_i:  n = new CountTrailingZerosINode(arg);  break;
2347   case vmIntrinsics::_numberOfTrailingZeros_l:  n = new CountTrailingZerosLNode(arg);  break;
2348   case vmIntrinsics::_bitCount_i:               n = new PopCountINode(          arg);  break;
2349   case vmIntrinsics::_bitCount_l:               n = new PopCountLNode(          arg);  break;
2350   case vmIntrinsics::_reverseBytes_c:           n = new ReverseBytesUSNode(0,   arg);  break;
2351   case vmIntrinsics::_reverseBytes_s:           n = new ReverseBytesSNode( 0,   arg);  break;
2352   case vmIntrinsics::_reverseBytes_i:           n = new ReverseBytesINode( 0,   arg);  break;
2353   case vmIntrinsics::_reverseBytes_l:           n = new ReverseBytesLNode( 0,   arg);  break;
2354   default:  fatal_unexpected_iid(id);  break;
2355   }
2356   set_result(_gvn.transform(n));
2357   return true;
2358 }
2359 
2360 //----------------------------inline_unsafe_access----------------------------
2361 
2362 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2363 
2364 // Helper that guards and inserts a pre-barrier.
2365 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2366                                         Node* pre_val, bool need_mem_bar) {
2367   // We could be accessing the referent field of a reference object. If so, when G1
2368   // is enabled, we need to log the value in the referent field in an SATB buffer.
2369   // This routine performs some compile time filters and generates suitable
2370   // runtime filters that guard the pre-barrier code.
2371   // Also add memory barrier for non volatile load from the referent field
2372   // to prevent commoning of loads across safepoint.
2373   if (!UseG1GC && !need_mem_bar)
2374     return;
2375 
2376   // Some compile time checks.
2377 
2378   // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2379   const TypeX* otype = offset->find_intptr_t_type();
2380   if (otype != NULL && otype->is_con() &&
2381       otype->get_con() != java_lang_ref_Reference::referent_offset) {
2382     // Constant offset but not the reference_offset so just return
2383     return;
2384   }
2385 
2386   // We only need to generate the runtime guards for instances.
2387   const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2388   if (btype != NULL) {
2389     if (btype->isa_aryptr()) {
2390       // Array type so nothing to do
2391       return;
2392     }
2393 
2394     const TypeInstPtr* itype = btype->isa_instptr();
2395     if (itype != NULL) {
2396       // Can the klass of base_oop be statically determined to be
2397       // _not_ a sub-class of Reference and _not_ Object?
2398       ciKlass* klass = itype->klass();
2399       if ( klass->is_loaded() &&
2400           !klass->is_subtype_of(env()->Reference_klass()) &&
2401           !env()->Object_klass()->is_subtype_of(klass)) {
2402         return;
2403       }
2404     }
2405   }
2406 
2407   // The compile time filters did not reject base_oop/offset so
2408   // we need to generate the following runtime filters
2409   //
2410   // if (offset == java_lang_ref_Reference::_reference_offset) {
2411   //   if (instance_of(base, java.lang.ref.Reference)) {
2412   //     pre_barrier(_, pre_val, ...);
2413   //   }
2414   // }
2415 
2416   float likely   = PROB_LIKELY(  0.999);
2417   float unlikely = PROB_UNLIKELY(0.999);
2418 
2419   IdealKit ideal(this);
2420 #define __ ideal.
2421 
2422   Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2423 
2424   __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2425       // Update graphKit memory and control from IdealKit.
2426       sync_kit(ideal);
2427 
2428       Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2429       Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2430 
2431       // Update IdealKit memory and control from graphKit.
2432       __ sync_kit(this);
2433 
2434       Node* one = __ ConI(1);
2435       // is_instof == 0 if base_oop == NULL
2436       __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2437 
2438         // Update graphKit from IdeakKit.
2439         sync_kit(ideal);
2440 
2441         // Use the pre-barrier to record the value in the referent field
2442         pre_barrier(false /* do_load */,
2443                     __ ctrl(),
2444                     NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2445                     pre_val /* pre_val */,
2446                     T_OBJECT);
2447         if (need_mem_bar) {
2448           // Add memory barrier to prevent commoning reads from this field
2449           // across safepoint since GC can change its value.
2450           insert_mem_bar(Op_MemBarCPUOrder);
2451         }
2452         // Update IdealKit from graphKit.
2453         __ sync_kit(this);
2454 
2455       } __ end_if(); // _ref_type != ref_none
2456   } __ end_if(); // offset == referent_offset
2457 
2458   // Final sync IdealKit and GraphKit.
2459   final_sync(ideal);
2460 #undef __
2461 }
2462 
2463 
2464 // Interpret Unsafe.fieldOffset cookies correctly:
2465 extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);
2466 
2467 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2468   // Attempt to infer a sharper value type from the offset and base type.
2469   ciKlass* sharpened_klass = NULL;
2470 
2471   // See if it is an instance field, with an object type.
2472   if (alias_type->field() != NULL) {
2473     assert(!is_native_ptr, "native pointer op cannot use a java address");
2474     if (alias_type->field()->type()->is_klass()) {
2475       sharpened_klass = alias_type->field()->type()->as_klass();
2476     }
2477   }
2478 
2479   // See if it is a narrow oop array.
2480   if (adr_type->isa_aryptr()) {
2481     if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2482       const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2483       if (elem_type != NULL) {
2484         sharpened_klass = elem_type->klass();
2485       }
2486     }
2487   }
2488 
2489   // The sharpened class might be unloaded if there is no class loader
2490   // contraint in place.
2491   if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2492     const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2493 
2494 #ifndef PRODUCT
2495     if (C->print_intrinsics() || C->print_inlining()) {
2496       tty->print("  from base type: ");  adr_type->dump();
2497       tty->print("  sharpened value: ");  tjp->dump();
2498     }
2499 #endif
2500     // Sharpen the value type.
2501     return tjp;
2502   }
2503   return NULL;
2504 }
2505 
2506 bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile) {
2507   if (callee()->is_static())  return false;  // caller must have the capability!
2508 
2509 #ifndef PRODUCT
2510   {
2511     ResourceMark rm;
2512     // Check the signatures.
2513     ciSignature* sig = callee()->signature();
2514 #ifdef ASSERT
2515     if (!is_store) {
2516       // Object getObject(Object base, int/long offset), etc.
2517       BasicType rtype = sig->return_type()->basic_type();
2518       if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2519           rtype = T_ADDRESS;  // it is really a C void*
2520       assert(rtype == type, "getter must return the expected value");
2521       if (!is_native_ptr) {
2522         assert(sig->count() == 2, "oop getter has 2 arguments");
2523         assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2524         assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2525       } else {
2526         assert(sig->count() == 1, "native getter has 1 argument");
2527         assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2528       }
2529     } else {
2530       // void putObject(Object base, int/long offset, Object x), etc.
2531       assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2532       if (!is_native_ptr) {
2533         assert(sig->count() == 3, "oop putter has 3 arguments");
2534         assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2535         assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2536       } else {
2537         assert(sig->count() == 2, "native putter has 2 arguments");
2538         assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2539       }
2540       BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2541       if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2542         vtype = T_ADDRESS;  // it is really a C void*
2543       assert(vtype == type, "putter must accept the expected value");
2544     }
2545 #endif // ASSERT
2546  }
2547 #endif //PRODUCT
2548 
2549   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2550 
2551   Node* receiver = argument(0);  // type: oop
2552 
2553   // Build address expression.
2554   Node* adr;
2555   Node* heap_base_oop = top();
2556   Node* offset = top();
2557   Node* val;
2558 
2559   if (!is_native_ptr) {
2560     // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2561     Node* base = argument(1);  // type: oop
2562     // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2563     offset = argument(2);  // type: long
2564     // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2565     // to be plain byte offsets, which are also the same as those accepted
2566     // by oopDesc::field_base.
2567     assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2568            "fieldOffset must be byte-scaled");
2569     // 32-bit machines ignore the high half!
2570     offset = ConvL2X(offset);
2571     adr = make_unsafe_address(base, offset);
2572     heap_base_oop = base;
2573     val = is_store ? argument(4) : NULL;
2574   } else {
2575     Node* ptr = argument(1);  // type: long
2576     ptr = ConvL2X(ptr);  // adjust Java long to machine word
2577     adr = make_unsafe_address(NULL, ptr);
2578     val = is_store ? argument(3) : NULL;
2579   }
2580 
2581   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2582 
2583   // First guess at the value type.
2584   const Type *value_type = Type::get_const_basic_type(type);
2585 
2586   // Try to categorize the address.  If it comes up as TypeJavaPtr::BOTTOM,
2587   // there was not enough information to nail it down.
2588   Compile::AliasType* alias_type = C->alias_type(adr_type);
2589   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2590 
2591   // We will need memory barriers unless we can determine a unique
2592   // alias category for this reference.  (Note:  If for some reason
2593   // the barriers get omitted and the unsafe reference begins to "pollute"
2594   // the alias analysis of the rest of the graph, either Compile::can_alias
2595   // or Compile::must_alias will throw a diagnostic assert.)
2596   bool need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2597 
2598   // If we are reading the value of the referent field of a Reference
2599   // object (either by using Unsafe directly or through reflection)
2600   // then, if G1 is enabled, we need to record the referent in an
2601   // SATB log buffer using the pre-barrier mechanism.
2602   // Also we need to add memory barrier to prevent commoning reads
2603   // from this field across safepoint since GC can change its value.
2604   bool need_read_barrier = !is_native_ptr && !is_store &&
2605                            offset != top() && heap_base_oop != top();
2606 
2607   if (!is_store && type == T_OBJECT) {
2608     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2609     if (tjp != NULL) {
2610       value_type = tjp;
2611     }
2612   }
2613 
2614   receiver = null_check(receiver);
2615   if (stopped()) {
2616     return true;
2617   }
2618   // Heap pointers get a null-check from the interpreter,
2619   // as a courtesy.  However, this is not guaranteed by Unsafe,
2620   // and it is not possible to fully distinguish unintended nulls
2621   // from intended ones in this API.
2622 
2623   if (is_volatile) {
2624     // We need to emit leading and trailing CPU membars (see below) in
2625     // addition to memory membars when is_volatile. This is a little
2626     // too strong, but avoids the need to insert per-alias-type
2627     // volatile membars (for stores; compare Parse::do_put_xxx), which
2628     // we cannot do effectively here because we probably only have a
2629     // rough approximation of type.
2630     need_mem_bar = true;
2631     // For Stores, place a memory ordering barrier now.
2632     if (is_store) {
2633       insert_mem_bar(Op_MemBarRelease);
2634     } else {
2635       if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2636         insert_mem_bar(Op_MemBarVolatile);
2637       }
2638     }
2639   }
2640 
2641   // Memory barrier to prevent normal and 'unsafe' accesses from
2642   // bypassing each other.  Happens after null checks, so the
2643   // exception paths do not take memory state from the memory barrier,
2644   // so there's no problems making a strong assert about mixing users
2645   // of safe & unsafe memory.
2646   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2647 
2648   if (!is_store) {
2649     MemNode::MemOrd mo = is_volatile ? MemNode::acquire : MemNode::unordered;
2650     // To be valid, unsafe loads may depend on other conditions than
2651     // the one that guards them: pin the Load node
2652     Node* p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, is_volatile);
2653     // load value
2654     switch (type) {
2655     case T_BOOLEAN:
2656     case T_CHAR:
2657     case T_BYTE:
2658     case T_SHORT:
2659     case T_INT:
2660     case T_LONG:
2661     case T_FLOAT:
2662     case T_DOUBLE:
2663       break;
2664     case T_OBJECT:
2665       if (need_read_barrier) {
2666         insert_pre_barrier(heap_base_oop, offset, p, !(is_volatile || need_mem_bar));
2667       }
2668       break;
2669     case T_ADDRESS:
2670       // Cast to an int type.
2671       p = _gvn.transform(new CastP2XNode(NULL, p));
2672       p = ConvX2UL(p);
2673       break;
2674     default:
2675       fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
2676       break;
2677     }
2678     // The load node has the control of the preceding MemBarCPUOrder.  All
2679     // following nodes will have the control of the MemBarCPUOrder inserted at
2680     // the end of this method.  So, pushing the load onto the stack at a later
2681     // point is fine.
2682     set_result(p);
2683   } else {
2684     // place effect of store into memory
2685     switch (type) {
2686     case T_DOUBLE:
2687       val = dstore_rounding(val);
2688       break;
2689     case T_ADDRESS:
2690       // Repackage the long as a pointer.
2691       val = ConvL2X(val);
2692       val = _gvn.transform(new CastX2PNode(val));
2693       break;
2694     }
2695 
2696     MemNode::MemOrd mo = is_volatile ? MemNode::release : MemNode::unordered;
2697     if (type != T_OBJECT ) {
2698       (void) store_to_memory(control(), adr, val, type, adr_type, mo, is_volatile);
2699     } else {
2700       // Possibly an oop being stored to Java heap or native memory
2701       if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2702         // oop to Java heap.
2703         (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo);
2704       } else {
2705         // We can't tell at compile time if we are storing in the Java heap or outside
2706         // of it. So we need to emit code to conditionally do the proper type of
2707         // store.
2708 
2709         IdealKit ideal(this);
2710 #define __ ideal.
2711         // QQQ who knows what probability is here??
2712         __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2713           // Sync IdealKit and graphKit.
2714           sync_kit(ideal);
2715           Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo);
2716           // Update IdealKit memory.
2717           __ sync_kit(this);
2718         } __ else_(); {
2719           __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, is_volatile);
2720         } __ end_if();
2721         // Final sync IdealKit and GraphKit.
2722         final_sync(ideal);
2723 #undef __
2724       }
2725     }
2726   }
2727 
2728   if (is_volatile) {
2729     if (!is_store) {
2730       insert_mem_bar(Op_MemBarAcquire);
2731     } else {
2732       if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2733         insert_mem_bar(Op_MemBarVolatile);
2734       }
2735     }
2736   }
2737 
2738   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2739 
2740   return true;
2741 }
2742 
2743 //----------------------------inline_unsafe_load_store----------------------------
2744 // This method serves a couple of different customers (depending on LoadStoreKind):
2745 //
2746 // LS_cmpxchg:
2747 //   public final native boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2748 //   public final native boolean compareAndSwapInt(   Object o, long offset, int    expected, int    x);
2749 //   public final native boolean compareAndSwapLong(  Object o, long offset, long   expected, long   x);
2750 //
2751 // LS_xadd:
2752 //   public int  getAndAddInt( Object o, long offset, int  delta)
2753 //   public long getAndAddLong(Object o, long offset, long delta)
2754 //
2755 // LS_xchg:
2756 //   int    getAndSet(Object o, long offset, int    newValue)
2757 //   long   getAndSet(Object o, long offset, long   newValue)
2758 //   Object getAndSet(Object o, long offset, Object newValue)
2759 //
2760 bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind) {
2761   // This basic scheme here is the same as inline_unsafe_access, but
2762   // differs in enough details that combining them would make the code
2763   // overly confusing.  (This is a true fact! I originally combined
2764   // them, but even I was confused by it!) As much code/comments as
2765   // possible are retained from inline_unsafe_access though to make
2766   // the correspondences clearer. - dl
2767 
2768   if (callee()->is_static())  return false;  // caller must have the capability!
2769 
2770 #ifndef PRODUCT
2771   BasicType rtype;
2772   {
2773     ResourceMark rm;
2774     // Check the signatures.
2775     ciSignature* sig = callee()->signature();
2776     rtype = sig->return_type()->basic_type();
2777     if (kind == LS_xadd || kind == LS_xchg) {
2778       // Check the signatures.
2779 #ifdef ASSERT
2780       assert(rtype == type, "get and set must return the expected type");
2781       assert(sig->count() == 3, "get and set has 3 arguments");
2782       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2783       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2784       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2785 #endif // ASSERT
2786     } else if (kind == LS_cmpxchg) {
2787       // Check the signatures.
2788 #ifdef ASSERT
2789       assert(rtype == T_BOOLEAN, "CAS must return boolean");
2790       assert(sig->count() == 4, "CAS has 4 arguments");
2791       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2792       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2793 #endif // ASSERT
2794     } else {
2795       ShouldNotReachHere();
2796     }
2797   }
2798 #endif //PRODUCT
2799 
2800   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2801 
2802   // Get arguments:
2803   Node* receiver = NULL;
2804   Node* base     = NULL;
2805   Node* offset   = NULL;
2806   Node* oldval   = NULL;
2807   Node* newval   = NULL;
2808   if (kind == LS_cmpxchg) {
2809     const bool two_slot_type = type2size[type] == 2;
2810     receiver = argument(0);  // type: oop
2811     base     = argument(1);  // type: oop
2812     offset   = argument(2);  // type: long
2813     oldval   = argument(4);  // type: oop, int, or long
2814     newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2815   } else if (kind == LS_xadd || kind == LS_xchg){
2816     receiver = argument(0);  // type: oop
2817     base     = argument(1);  // type: oop
2818     offset   = argument(2);  // type: long
2819     oldval   = NULL;
2820     newval   = argument(4);  // type: oop, int, or long
2821   }
2822 
2823   // Null check receiver.
2824   receiver = null_check(receiver);
2825   if (stopped()) {
2826     return true;
2827   }
2828 
2829   // Build field offset expression.
2830   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2831   // to be plain byte offsets, which are also the same as those accepted
2832   // by oopDesc::field_base.
2833   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2834   // 32-bit machines ignore the high half of long offsets
2835   offset = ConvL2X(offset);
2836   Node* adr = make_unsafe_address(base, offset);
2837   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2838 
2839   // For CAS, unlike inline_unsafe_access, there seems no point in
2840   // trying to refine types. Just use the coarse types here.
2841   const Type *value_type = Type::get_const_basic_type(type);
2842   Compile::AliasType* alias_type = C->alias_type(adr_type);
2843   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2844 
2845   if (kind == LS_xchg && type == T_OBJECT) {
2846     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2847     if (tjp != NULL) {
2848       value_type = tjp;
2849     }
2850   }
2851 
2852   int alias_idx = C->get_alias_index(adr_type);
2853 
2854   // Memory-model-wise, a LoadStore acts like a little synchronized
2855   // block, so needs barriers on each side.  These don't translate
2856   // into actual barriers on most machines, but we still need rest of
2857   // compiler to respect ordering.
2858 
2859   insert_mem_bar(Op_MemBarRelease);
2860   insert_mem_bar(Op_MemBarCPUOrder);
2861 
2862   // 4984716: MemBars must be inserted before this
2863   //          memory node in order to avoid a false
2864   //          dependency which will confuse the scheduler.
2865   Node *mem = memory(alias_idx);
2866 
2867   // For now, we handle only those cases that actually exist: ints,
2868   // longs, and Object. Adding others should be straightforward.
2869   Node* load_store;
2870   switch(type) {
2871   case T_INT:
2872     if (kind == LS_xadd) {
2873       load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2874     } else if (kind == LS_xchg) {
2875       load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2876     } else if (kind == LS_cmpxchg) {
2877       load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval));
2878     } else {
2879       ShouldNotReachHere();
2880     }
2881     break;
2882   case T_LONG:
2883     if (kind == LS_xadd) {
2884       load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2885     } else if (kind == LS_xchg) {
2886       load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2887     } else if (kind == LS_cmpxchg) {
2888       load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval));
2889     } else {
2890       ShouldNotReachHere();
2891     }
2892     break;
2893   case T_OBJECT:
2894     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2895     // could be delayed during Parse (for example, in adjust_map_after_if()).
2896     // Execute transformation here to avoid barrier generation in such case.
2897     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2898       newval = _gvn.makecon(TypePtr::NULL_PTR);
2899 
2900     // Reference stores need a store barrier.
2901     if (kind == LS_xchg) {
2902       // If pre-barrier must execute before the oop store, old value will require do_load here.
2903       if (!can_move_pre_barrier()) {
2904         pre_barrier(true /* do_load*/,
2905                     control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2906                     NULL /* pre_val*/,
2907                     T_OBJECT);
2908       } // Else move pre_barrier to use load_store value, see below.
2909     } else if (kind == LS_cmpxchg) {
2910       // Same as for newval above:
2911       if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2912         oldval = _gvn.makecon(TypePtr::NULL_PTR);
2913       }
2914       // The only known value which might get overwritten is oldval.
2915       pre_barrier(false /* do_load */,
2916                   control(), NULL, NULL, max_juint, NULL, NULL,
2917                   oldval /* pre_val */,
2918                   T_OBJECT);
2919     } else {
2920       ShouldNotReachHere();
2921     }
2922 
2923 #ifdef _LP64
2924     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2925       Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2926       if (kind == LS_xchg) {
2927         load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr,
2928                                                        newval_enc, adr_type, value_type->make_narrowoop()));
2929       } else {
2930         assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2931         Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2932         load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr,
2933                                                                 newval_enc, oldval_enc));
2934       }
2935     } else
2936 #endif
2937     {
2938       if (kind == LS_xchg) {
2939         load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2940       } else {
2941         assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2942         load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval));
2943       }
2944     }
2945     post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2946     break;
2947   default:
2948     fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
2949     break;
2950   }
2951 
2952   // SCMemProjNodes represent the memory state of a LoadStore. Their
2953   // main role is to prevent LoadStore nodes from being optimized away
2954   // when their results aren't used.
2955   Node* proj = _gvn.transform(new SCMemProjNode(load_store));
2956   set_memory(proj, alias_idx);
2957 
2958   if (type == T_OBJECT && kind == LS_xchg) {
2959 #ifdef _LP64
2960     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2961       load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
2962     }
2963 #endif
2964     if (can_move_pre_barrier()) {
2965       // Don't need to load pre_val. The old value is returned by load_store.
2966       // The pre_barrier can execute after the xchg as long as no safepoint
2967       // gets inserted between them.
2968       pre_barrier(false /* do_load */,
2969                   control(), NULL, NULL, max_juint, NULL, NULL,
2970                   load_store /* pre_val */,
2971                   T_OBJECT);
2972     }
2973   }
2974 
2975   // Add the trailing membar surrounding the access
2976   insert_mem_bar(Op_MemBarCPUOrder);
2977   insert_mem_bar(Op_MemBarAcquire);
2978 
2979   assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2980   set_result(load_store);
2981   return true;
2982 }
2983 
2984 //----------------------------inline_unsafe_ordered_store----------------------
2985 // public native void sun.misc.Unsafe.putOrderedObject(Object o, long offset, Object x);
2986 // public native void sun.misc.Unsafe.putOrderedInt(Object o, long offset, int x);
2987 // public native void sun.misc.Unsafe.putOrderedLong(Object o, long offset, long x);
2988 bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
2989   // This is another variant of inline_unsafe_access, differing in
2990   // that it always issues store-store ("release") barrier and ensures
2991   // store-atomicity (which only matters for "long").
2992 
2993   if (callee()->is_static())  return false;  // caller must have the capability!
2994 
2995 #ifndef PRODUCT
2996   {
2997     ResourceMark rm;
2998     // Check the signatures.
2999     ciSignature* sig = callee()->signature();
3000 #ifdef ASSERT
3001     BasicType rtype = sig->return_type()->basic_type();
3002     assert(rtype == T_VOID, "must return void");
3003     assert(sig->count() == 3, "has 3 arguments");
3004     assert(sig->type_at(0)->basic_type() == T_OBJECT, "base is object");
3005     assert(sig->type_at(1)->basic_type() == T_LONG, "offset is long");
3006 #endif // ASSERT
3007   }
3008 #endif //PRODUCT
3009 
3010   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
3011 
3012   // Get arguments:
3013   Node* receiver = argument(0);  // type: oop
3014   Node* base     = argument(1);  // type: oop
3015   Node* offset   = argument(2);  // type: long
3016   Node* val      = argument(4);  // type: oop, int, or long
3017 
3018   // Null check receiver.
3019   receiver = null_check(receiver);
3020   if (stopped()) {
3021     return true;
3022   }
3023 
3024   // Build field offset expression.
3025   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3026   // 32-bit machines ignore the high half of long offsets
3027   offset = ConvL2X(offset);
3028   Node* adr = make_unsafe_address(base, offset);
3029   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3030   const Type *value_type = Type::get_const_basic_type(type);
3031   Compile::AliasType* alias_type = C->alias_type(adr_type);
3032 
3033   insert_mem_bar(Op_MemBarRelease);
3034   insert_mem_bar(Op_MemBarCPUOrder);
3035   // Ensure that the store is atomic for longs:
3036   const bool require_atomic_access = true;
3037   Node* store;
3038   if (type == T_OBJECT) // reference stores need a store barrier.
3039     store = store_oop_to_unknown(control(), base, adr, adr_type, val, type, MemNode::release);
3040   else {
3041     store = store_to_memory(control(), adr, val, type, adr_type, MemNode::release, require_atomic_access);
3042   }
3043   insert_mem_bar(Op_MemBarCPUOrder);
3044   return true;
3045 }
3046 
3047 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3048   // Regardless of form, don't allow previous ld/st to move down,
3049   // then issue acquire, release, or volatile mem_bar.
3050   insert_mem_bar(Op_MemBarCPUOrder);
3051   switch(id) {
3052     case vmIntrinsics::_loadFence:
3053       insert_mem_bar(Op_LoadFence);
3054       return true;
3055     case vmIntrinsics::_storeFence:
3056       insert_mem_bar(Op_StoreFence);
3057       return true;
3058     case vmIntrinsics::_fullFence:
3059       insert_mem_bar(Op_MemBarVolatile);
3060       return true;
3061     default:
3062       fatal_unexpected_iid(id);
3063       return false;
3064   }
3065 }
3066 
3067 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3068   if (!kls->is_Con()) {
3069     return true;
3070   }
3071   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3072   if (klsptr == NULL) {
3073     return true;
3074   }
3075   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3076   // don't need a guard for a klass that is already initialized
3077   return !ik->is_initialized();
3078 }
3079 
3080 //----------------------------inline_unsafe_allocate---------------------------
3081 // public native Object sun.misc.Unsafe.allocateInstance(Class<?> cls);
3082 bool LibraryCallKit::inline_unsafe_allocate() {
3083   if (callee()->is_static())  return false;  // caller must have the capability!
3084 
3085   null_check_receiver();  // null-check, then ignore
3086   Node* cls = null_check(argument(1));
3087   if (stopped())  return true;
3088 
3089   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3090   kls = null_check(kls);
3091   if (stopped())  return true;  // argument was like int.class
3092 
3093   Node* test = NULL;
3094   if (LibraryCallKit::klass_needs_init_guard(kls)) {
3095     // Note:  The argument might still be an illegal value like
3096     // Serializable.class or Object[].class.   The runtime will handle it.
3097     // But we must make an explicit check for initialization.
3098     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3099     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3100     // can generate code to load it as unsigned byte.
3101     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3102     Node* bits = intcon(InstanceKlass::fully_initialized);
3103     test = _gvn.transform(new SubINode(inst, bits));
3104     // The 'test' is non-zero if we need to take a slow path.
3105   }
3106 
3107   Node* obj = new_instance(kls, test);
3108   set_result(obj);
3109   return true;
3110 }
3111 
3112 #ifdef TRACE_HAVE_INTRINSICS
3113 /*
3114  * oop -> myklass
3115  * myklass->trace_id |= USED
3116  * return myklass->trace_id & ~0x3
3117  */
3118 bool LibraryCallKit::inline_native_classID() {
3119   null_check_receiver();  // null-check, then ignore
3120   Node* cls = null_check(argument(1), T_OBJECT);
3121   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3122   kls = null_check(kls, T_OBJECT);
3123   ByteSize offset = TRACE_ID_OFFSET;
3124   Node* insp = basic_plus_adr(kls, in_bytes(offset));
3125   Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3126   Node* bits = longcon(~0x03l); // ignore bit 0 & 1
3127   Node* andl = _gvn.transform(new AndLNode(tvalue, bits));
3128   Node* clsused = longcon(0x01l); // set the class bit
3129   Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3130 
3131   const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3132   store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3133   set_result(andl);
3134   return true;
3135 }
3136 
3137 bool LibraryCallKit::inline_native_threadID() {
3138   Node* tls_ptr = NULL;
3139   Node* cur_thr = generate_current_thread(tls_ptr);
3140   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3141   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3142   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::thread_id_offset()));
3143 
3144   Node* threadid = NULL;
3145   size_t thread_id_size = OSThread::thread_id_size();
3146   if (thread_id_size == (size_t) BytesPerLong) {
3147     threadid = ConvL2I(make_load(control(), p, TypeLong::LONG, T_LONG, MemNode::unordered));
3148   } else if (thread_id_size == (size_t) BytesPerInt) {
3149     threadid = make_load(control(), p, TypeInt::INT, T_INT, MemNode::unordered);
3150   } else {
3151     ShouldNotReachHere();
3152   }
3153   set_result(threadid);
3154   return true;
3155 }
3156 #endif
3157 
3158 //------------------------inline_native_time_funcs--------------
3159 // inline code for System.currentTimeMillis() and System.nanoTime()
3160 // these have the same type and signature
3161 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3162   const TypeFunc* tf = OptoRuntime::void_long_Type();
3163   const TypePtr* no_memory_effects = NULL;
3164   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3165   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3166 #ifdef ASSERT
3167   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3168   assert(value_top == top(), "second value must be top");
3169 #endif
3170   set_result(value);
3171   return true;
3172 }
3173 
3174 //------------------------inline_native_currentThread------------------
3175 bool LibraryCallKit::inline_native_currentThread() {
3176   Node* junk = NULL;
3177   set_result(generate_current_thread(junk));
3178   return true;
3179 }
3180 
3181 //------------------------inline_native_isInterrupted------------------
3182 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3183 bool LibraryCallKit::inline_native_isInterrupted() {
3184   // Add a fast path to t.isInterrupted(clear_int):
3185   //   (t == Thread.current() &&
3186   //    (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3187   //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3188   // So, in the common case that the interrupt bit is false,
3189   // we avoid making a call into the VM.  Even if the interrupt bit
3190   // is true, if the clear_int argument is false, we avoid the VM call.
3191   // However, if the receiver is not currentThread, we must call the VM,
3192   // because there must be some locking done around the operation.
3193 
3194   // We only go to the fast case code if we pass two guards.
3195   // Paths which do not pass are accumulated in the slow_region.
3196 
3197   enum {
3198     no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
3199     no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
3200     slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
3201     PATH_LIMIT
3202   };
3203 
3204   // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3205   // out of the function.
3206   insert_mem_bar(Op_MemBarCPUOrder);
3207 
3208   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3209   PhiNode*    result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3210 
3211   RegionNode* slow_region = new RegionNode(1);
3212   record_for_igvn(slow_region);
3213 
3214   // (a) Receiving thread must be the current thread.
3215   Node* rec_thr = argument(0);
3216   Node* tls_ptr = NULL;
3217   Node* cur_thr = generate_current_thread(tls_ptr);
3218   Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3219   Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3220 
3221   generate_slow_guard(bol_thr, slow_region);
3222 
3223   // (b) Interrupt bit on TLS must be false.
3224   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3225   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3226   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3227 
3228   // Set the control input on the field _interrupted read to prevent it floating up.
3229   Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3230   Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3231   Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3232 
3233   IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3234 
3235   // First fast path:  if (!TLS._interrupted) return false;
3236   Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3237   result_rgn->init_req(no_int_result_path, false_bit);
3238   result_val->init_req(no_int_result_path, intcon(0));
3239 
3240   // drop through to next case
3241   set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3242 
3243 #ifndef TARGET_OS_FAMILY_windows
3244   // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3245   Node* clr_arg = argument(1);
3246   Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3247   Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3248   IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3249 
3250   // Second fast path:  ... else if (!clear_int) return true;
3251   Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3252   result_rgn->init_req(no_clear_result_path, false_arg);
3253   result_val->init_req(no_clear_result_path, intcon(1));
3254 
3255   // drop through to next case
3256   set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3257 #else
3258   // To return true on Windows you must read the _interrupted field
3259   // and check the the event state i.e. take the slow path.
3260 #endif // TARGET_OS_FAMILY_windows
3261 
3262   // (d) Otherwise, go to the slow path.
3263   slow_region->add_req(control());
3264   set_control( _gvn.transform(slow_region));
3265 
3266   if (stopped()) {
3267     // There is no slow path.
3268     result_rgn->init_req(slow_result_path, top());
3269     result_val->init_req(slow_result_path, top());
3270   } else {
3271     // non-virtual because it is a private non-static
3272     CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3273 
3274     Node* slow_val = set_results_for_java_call(slow_call);
3275     // this->control() comes from set_results_for_java_call
3276 
3277     Node* fast_io  = slow_call->in(TypeFunc::I_O);
3278     Node* fast_mem = slow_call->in(TypeFunc::Memory);
3279 
3280     // These two phis are pre-filled with copies of of the fast IO and Memory
3281     PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3282     PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
3283 
3284     result_rgn->init_req(slow_result_path, control());
3285     result_io ->init_req(slow_result_path, i_o());
3286     result_mem->init_req(slow_result_path, reset_memory());
3287     result_val->init_req(slow_result_path, slow_val);
3288 
3289     set_all_memory(_gvn.transform(result_mem));
3290     set_i_o(       _gvn.transform(result_io));
3291   }
3292 
3293   C->set_has_split_ifs(true); // Has chance for split-if optimization
3294   set_result(result_rgn, result_val);
3295   return true;
3296 }
3297 
3298 //---------------------------load_mirror_from_klass----------------------------
3299 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3300 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3301   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3302   return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3303 }
3304 
3305 //-----------------------load_klass_from_mirror_common-------------------------
3306 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3307 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3308 // and branch to the given path on the region.
3309 // If never_see_null, take an uncommon trap on null, so we can optimistically
3310 // compile for the non-null case.
3311 // If the region is NULL, force never_see_null = true.
3312 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3313                                                     bool never_see_null,
3314                                                     RegionNode* region,
3315                                                     int null_path,
3316                                                     int offset) {
3317   if (region == NULL)  never_see_null = true;
3318   Node* p = basic_plus_adr(mirror, offset);
3319   const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3320   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3321   Node* null_ctl = top();
3322   kls = null_check_oop(kls, &null_ctl, never_see_null);
3323   if (region != NULL) {
3324     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3325     region->init_req(null_path, null_ctl);
3326   } else {
3327     assert(null_ctl == top(), "no loose ends");
3328   }
3329   return kls;
3330 }
3331 
3332 //--------------------(inline_native_Class_query helpers)---------------------
3333 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE, JVM_ACC_HAS_FINALIZER.
3334 // Fall through if (mods & mask) == bits, take the guard otherwise.
3335 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3336   // Branch around if the given klass has the given modifier bit set.
3337   // Like generate_guard, adds a new path onto the region.
3338   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3339   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3340   Node* mask = intcon(modifier_mask);
3341   Node* bits = intcon(modifier_bits);
3342   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3343   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3344   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3345   return generate_fair_guard(bol, region);
3346 }
3347 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3348   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3349 }
3350 
3351 //-------------------------inline_native_Class_query-------------------
3352 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3353   const Type* return_type = TypeInt::BOOL;
3354   Node* prim_return_value = top();  // what happens if it's a primitive class?
3355   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3356   bool expect_prim = false;     // most of these guys expect to work on refs
3357 
3358   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3359 
3360   Node* mirror = argument(0);
3361   Node* obj    = top();
3362 
3363   switch (id) {
3364   case vmIntrinsics::_isInstance:
3365     // nothing is an instance of a primitive type
3366     prim_return_value = intcon(0);
3367     obj = argument(1);
3368     break;
3369   case vmIntrinsics::_getModifiers:
3370     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3371     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3372     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3373     break;
3374   case vmIntrinsics::_isInterface:
3375     prim_return_value = intcon(0);
3376     break;
3377   case vmIntrinsics::_isArray:
3378     prim_return_value = intcon(0);
3379     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3380     break;
3381   case vmIntrinsics::_isPrimitive:
3382     prim_return_value = intcon(1);
3383     expect_prim = true;  // obviously
3384     break;
3385   case vmIntrinsics::_getSuperclass:
3386     prim_return_value = null();
3387     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3388     break;
3389   case vmIntrinsics::_getClassAccessFlags:
3390     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3391     return_type = TypeInt::INT;  // not bool!  6297094
3392     break;
3393   default:
3394     fatal_unexpected_iid(id);
3395     break;
3396   }
3397 
3398   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3399   if (mirror_con == NULL)  return false;  // cannot happen?
3400 
3401 #ifndef PRODUCT
3402   if (C->print_intrinsics() || C->print_inlining()) {
3403     ciType* k = mirror_con->java_mirror_type();
3404     if (k) {
3405       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3406       k->print_name();
3407       tty->cr();
3408     }
3409   }
3410 #endif
3411 
3412   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3413   RegionNode* region = new RegionNode(PATH_LIMIT);
3414   record_for_igvn(region);
3415   PhiNode* phi = new PhiNode(region, return_type);
3416 
3417   // The mirror will never be null of Reflection.getClassAccessFlags, however
3418   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3419   // if it is. See bug 4774291.
3420 
3421   // For Reflection.getClassAccessFlags(), the null check occurs in
3422   // the wrong place; see inline_unsafe_access(), above, for a similar
3423   // situation.
3424   mirror = null_check(mirror);
3425   // If mirror or obj is dead, only null-path is taken.
3426   if (stopped())  return true;
3427 
3428   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3429 
3430   // Now load the mirror's klass metaobject, and null-check it.
3431   // Side-effects region with the control path if the klass is null.
3432   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3433   // If kls is null, we have a primitive mirror.
3434   phi->init_req(_prim_path, prim_return_value);
3435   if (stopped()) { set_result(region, phi); return true; }
3436   bool safe_for_replace = (region->in(_prim_path) == top());
3437 
3438   Node* p;  // handy temp
3439   Node* null_ctl;
3440 
3441   // Now that we have the non-null klass, we can perform the real query.
3442   // For constant classes, the query will constant-fold in LoadNode::Value.
3443   Node* query_value = top();
3444   switch (id) {
3445   case vmIntrinsics::_isInstance:
3446     // nothing is an instance of a primitive type
3447     query_value = gen_instanceof(obj, kls, safe_for_replace);
3448     break;
3449 
3450   case vmIntrinsics::_getModifiers:
3451     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3452     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3453     break;
3454 
3455   case vmIntrinsics::_isInterface:
3456     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3457     if (generate_interface_guard(kls, region) != NULL)
3458       // A guard was added.  If the guard is taken, it was an interface.
3459       phi->add_req(intcon(1));
3460     // If we fall through, it's a plain class.
3461     query_value = intcon(0);
3462     break;
3463 
3464   case vmIntrinsics::_isArray:
3465     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3466     if (generate_array_guard(kls, region) != NULL)
3467       // A guard was added.  If the guard is taken, it was an array.
3468       phi->add_req(intcon(1));
3469     // If we fall through, it's a plain class.
3470     query_value = intcon(0);
3471     break;
3472 
3473   case vmIntrinsics::_isPrimitive:
3474     query_value = intcon(0); // "normal" path produces false
3475     break;
3476 
3477   case vmIntrinsics::_getSuperclass:
3478     // The rules here are somewhat unfortunate, but we can still do better
3479     // with random logic than with a JNI call.
3480     // Interfaces store null or Object as _super, but must report null.
3481     // Arrays store an intermediate super as _super, but must report Object.
3482     // Other types can report the actual _super.
3483     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3484     if (generate_interface_guard(kls, region) != NULL)
3485       // A guard was added.  If the guard is taken, it was an interface.
3486       phi->add_req(null());
3487     if (generate_array_guard(kls, region) != NULL)
3488       // A guard was added.  If the guard is taken, it was an array.
3489       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3490     // If we fall through, it's a plain class.  Get its _super.
3491     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3492     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3493     null_ctl = top();
3494     kls = null_check_oop(kls, &null_ctl);
3495     if (null_ctl != top()) {
3496       // If the guard is taken, Object.superClass is null (both klass and mirror).
3497       region->add_req(null_ctl);
3498       phi   ->add_req(null());
3499     }
3500     if (!stopped()) {
3501       query_value = load_mirror_from_klass(kls);
3502     }
3503     break;
3504 
3505   case vmIntrinsics::_getClassAccessFlags:
3506     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3507     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3508     break;
3509 
3510   default:
3511     fatal_unexpected_iid(id);
3512     break;
3513   }
3514 
3515   // Fall-through is the normal case of a query to a real class.
3516   phi->init_req(1, query_value);
3517   region->init_req(1, control());
3518 
3519   C->set_has_split_ifs(true); // Has chance for split-if optimization
3520   set_result(region, phi);
3521   return true;
3522 }
3523 
3524 //-------------------------inline_Class_cast-------------------
3525 bool LibraryCallKit::inline_Class_cast() {
3526   Node* mirror = argument(0); // Class
3527   Node* obj    = argument(1);
3528   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3529   if (mirror_con == NULL) {
3530     return false;  // dead path (mirror->is_top()).
3531   }
3532   if (obj == NULL || obj->is_top()) {
3533     return false;  // dead path
3534   }
3535   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3536 
3537   // First, see if Class.cast() can be folded statically.
3538   // java_mirror_type() returns non-null for compile-time Class constants.
3539   ciType* tm = mirror_con->java_mirror_type();
3540   if (tm != NULL && tm->is_klass() &&
3541       tp != NULL && tp->klass() != NULL) {
3542     if (!tp->klass()->is_loaded()) {
3543       // Don't use intrinsic when class is not loaded.
3544       return false;
3545     } else {
3546       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3547       if (static_res == Compile::SSC_always_true) {
3548         // isInstance() is true - fold the code.
3549         set_result(obj);
3550         return true;
3551       } else if (static_res == Compile::SSC_always_false) {
3552         // Don't use intrinsic, have to throw ClassCastException.
3553         // If the reference is null, the non-intrinsic bytecode will
3554         // be optimized appropriately.
3555         return false;
3556       }
3557     }
3558   }
3559 
3560   // Bailout intrinsic and do normal inlining if exception path is frequent.
3561   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3562     return false;
3563   }
3564 
3565   // Generate dynamic checks.
3566   // Class.cast() is java implementation of _checkcast bytecode.
3567   // Do checkcast (Parse::do_checkcast()) optimizations here.
3568 
3569   mirror = null_check(mirror);
3570   // If mirror is dead, only null-path is taken.
3571   if (stopped()) {
3572     return true;
3573   }
3574 
3575   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3576   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3577   RegionNode* region = new RegionNode(PATH_LIMIT);
3578   record_for_igvn(region);
3579 
3580   // Now load the mirror's klass metaobject, and null-check it.
3581   // If kls is null, we have a primitive mirror and
3582   // nothing is an instance of a primitive type.
3583   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3584 
3585   Node* res = top();
3586   if (!stopped()) {
3587     Node* bad_type_ctrl = top();
3588     // Do checkcast optimizations.
3589     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3590     region->init_req(_bad_type_path, bad_type_ctrl);
3591   }
3592   if (region->in(_prim_path) != top() ||
3593       region->in(_bad_type_path) != top()) {
3594     // Let Interpreter throw ClassCastException.
3595     PreserveJVMState pjvms(this);
3596     set_control(_gvn.transform(region));
3597     uncommon_trap(Deoptimization::Reason_intrinsic,
3598                   Deoptimization::Action_maybe_recompile);
3599   }
3600   if (!stopped()) {
3601     set_result(res);
3602   }
3603   return true;
3604 }
3605 
3606 
3607 //--------------------------inline_native_subtype_check------------------------
3608 // This intrinsic takes the JNI calls out of the heart of
3609 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3610 bool LibraryCallKit::inline_native_subtype_check() {
3611   // Pull both arguments off the stack.
3612   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3613   args[0] = argument(0);
3614   args[1] = argument(1);
3615   Node* klasses[2];             // corresponding Klasses: superk, subk
3616   klasses[0] = klasses[1] = top();
3617 
3618   enum {
3619     // A full decision tree on {superc is prim, subc is prim}:
3620     _prim_0_path = 1,           // {P,N} => false
3621                                 // {P,P} & superc!=subc => false
3622     _prim_same_path,            // {P,P} & superc==subc => true
3623     _prim_1_path,               // {N,P} => false
3624     _ref_subtype_path,          // {N,N} & subtype check wins => true
3625     _both_ref_path,             // {N,N} & subtype check loses => false
3626     PATH_LIMIT
3627   };
3628 
3629   RegionNode* region = new RegionNode(PATH_LIMIT);
3630   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3631   record_for_igvn(region);
3632 
3633   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3634   const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3635   int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3636 
3637   // First null-check both mirrors and load each mirror's klass metaobject.
3638   int which_arg;
3639   for (which_arg = 0; which_arg <= 1; which_arg++) {
3640     Node* arg = args[which_arg];
3641     arg = null_check(arg);
3642     if (stopped())  break;
3643     args[which_arg] = arg;
3644 
3645     Node* p = basic_plus_adr(arg, class_klass_offset);
3646     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3647     klasses[which_arg] = _gvn.transform(kls);
3648   }
3649 
3650   // Having loaded both klasses, test each for null.
3651   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3652   for (which_arg = 0; which_arg <= 1; which_arg++) {
3653     Node* kls = klasses[which_arg];
3654     Node* null_ctl = top();
3655     kls = null_check_oop(kls, &null_ctl, never_see_null);
3656     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3657     region->init_req(prim_path, null_ctl);
3658     if (stopped())  break;
3659     klasses[which_arg] = kls;
3660   }
3661 
3662   if (!stopped()) {
3663     // now we have two reference types, in klasses[0..1]
3664     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3665     Node* superk = klasses[0];  // the receiver
3666     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3667     // now we have a successful reference subtype check
3668     region->set_req(_ref_subtype_path, control());
3669   }
3670 
3671   // If both operands are primitive (both klasses null), then
3672   // we must return true when they are identical primitives.
3673   // It is convenient to test this after the first null klass check.
3674   set_control(region->in(_prim_0_path)); // go back to first null check
3675   if (!stopped()) {
3676     // Since superc is primitive, make a guard for the superc==subc case.
3677     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3678     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3679     generate_guard(bol_eq, region, PROB_FAIR);
3680     if (region->req() == PATH_LIMIT+1) {
3681       // A guard was added.  If the added guard is taken, superc==subc.
3682       region->swap_edges(PATH_LIMIT, _prim_same_path);
3683       region->del_req(PATH_LIMIT);
3684     }
3685     region->set_req(_prim_0_path, control()); // Not equal after all.
3686   }
3687 
3688   // these are the only paths that produce 'true':
3689   phi->set_req(_prim_same_path,   intcon(1));
3690   phi->set_req(_ref_subtype_path, intcon(1));
3691 
3692   // pull together the cases:
3693   assert(region->req() == PATH_LIMIT, "sane region");
3694   for (uint i = 1; i < region->req(); i++) {
3695     Node* ctl = region->in(i);
3696     if (ctl == NULL || ctl == top()) {
3697       region->set_req(i, top());
3698       phi   ->set_req(i, top());
3699     } else if (phi->in(i) == NULL) {
3700       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3701     }
3702   }
3703 
3704   set_control(_gvn.transform(region));
3705   set_result(_gvn.transform(phi));
3706   return true;
3707 }
3708 
3709 //---------------------generate_array_guard_common------------------------
3710 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3711                                                   bool obj_array, bool not_array) {
3712 
3713   if (stopped()) {
3714     return NULL;
3715   }
3716 
3717   // If obj_array/non_array==false/false:
3718   // Branch around if the given klass is in fact an array (either obj or prim).
3719   // If obj_array/non_array==false/true:
3720   // Branch around if the given klass is not an array klass of any kind.
3721   // If obj_array/non_array==true/true:
3722   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3723   // If obj_array/non_array==true/false:
3724   // Branch around if the kls is an oop array (Object[] or subtype)
3725   //
3726   // Like generate_guard, adds a new path onto the region.
3727   jint  layout_con = 0;
3728   Node* layout_val = get_layout_helper(kls, layout_con);
3729   if (layout_val == NULL) {
3730     bool query = (obj_array
3731                   ? Klass::layout_helper_is_objArray(layout_con)
3732                   : Klass::layout_helper_is_array(layout_con));
3733     if (query == not_array) {
3734       return NULL;                       // never a branch
3735     } else {                             // always a branch
3736       Node* always_branch = control();
3737       if (region != NULL)
3738         region->add_req(always_branch);
3739       set_control(top());
3740       return always_branch;
3741     }
3742   }
3743   // Now test the correct condition.
3744   jint  nval = (obj_array
3745                 ? ((jint)Klass::_lh_array_tag_type_value
3746                    <<    Klass::_lh_array_tag_shift)
3747                 : Klass::_lh_neutral_value);
3748   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3749   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3750   // invert the test if we are looking for a non-array
3751   if (not_array)  btest = BoolTest(btest).negate();
3752   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3753   return generate_fair_guard(bol, region);
3754 }
3755 
3756 
3757 //-----------------------inline_native_newArray--------------------------
3758 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3759 bool LibraryCallKit::inline_native_newArray() {
3760   Node* mirror    = argument(0);
3761   Node* count_val = argument(1);
3762 
3763   mirror = null_check(mirror);
3764   // If mirror or obj is dead, only null-path is taken.
3765   if (stopped())  return true;
3766 
3767   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3768   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3769   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3770   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3771   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3772 
3773   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3774   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3775                                                   result_reg, _slow_path);
3776   Node* normal_ctl   = control();
3777   Node* no_array_ctl = result_reg->in(_slow_path);
3778 
3779   // Generate code for the slow case.  We make a call to newArray().
3780   set_control(no_array_ctl);
3781   if (!stopped()) {
3782     // Either the input type is void.class, or else the
3783     // array klass has not yet been cached.  Either the
3784     // ensuing call will throw an exception, or else it
3785     // will cache the array klass for next time.
3786     PreserveJVMState pjvms(this);
3787     CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3788     Node* slow_result = set_results_for_java_call(slow_call);
3789     // this->control() comes from set_results_for_java_call
3790     result_reg->set_req(_slow_path, control());
3791     result_val->set_req(_slow_path, slow_result);
3792     result_io ->set_req(_slow_path, i_o());
3793     result_mem->set_req(_slow_path, reset_memory());
3794   }
3795 
3796   set_control(normal_ctl);
3797   if (!stopped()) {
3798     // Normal case:  The array type has been cached in the java.lang.Class.
3799     // The following call works fine even if the array type is polymorphic.
3800     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3801     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3802     result_reg->init_req(_normal_path, control());
3803     result_val->init_req(_normal_path, obj);
3804     result_io ->init_req(_normal_path, i_o());
3805     result_mem->init_req(_normal_path, reset_memory());
3806   }
3807 
3808   // Return the combined state.
3809   set_i_o(        _gvn.transform(result_io)  );
3810   set_all_memory( _gvn.transform(result_mem));
3811 
3812   C->set_has_split_ifs(true); // Has chance for split-if optimization
3813   set_result(result_reg, result_val);
3814   return true;
3815 }
3816 
3817 //----------------------inline_native_getLength--------------------------
3818 // public static native int java.lang.reflect.Array.getLength(Object array);
3819 bool LibraryCallKit::inline_native_getLength() {
3820   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3821 
3822   Node* array = null_check(argument(0));
3823   // If array is dead, only null-path is taken.
3824   if (stopped())  return true;
3825 
3826   // Deoptimize if it is a non-array.
3827   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3828 
3829   if (non_array != NULL) {
3830     PreserveJVMState pjvms(this);
3831     set_control(non_array);
3832     uncommon_trap(Deoptimization::Reason_intrinsic,
3833                   Deoptimization::Action_maybe_recompile);
3834   }
3835 
3836   // If control is dead, only non-array-path is taken.
3837   if (stopped())  return true;
3838 
3839   // The works fine even if the array type is polymorphic.
3840   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3841   Node* result = load_array_length(array);
3842 
3843   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3844   set_result(result);
3845   return true;
3846 }
3847 
3848 //------------------------inline_array_copyOf----------------------------
3849 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3850 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3851 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3852   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3853 
3854   // Get the arguments.
3855   Node* original          = argument(0);
3856   Node* start             = is_copyOfRange? argument(1): intcon(0);
3857   Node* end               = is_copyOfRange? argument(2): argument(1);
3858   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3859 
3860   Node* newcopy;
3861 
3862   // Set the original stack and the reexecute bit for the interpreter to reexecute
3863   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3864   { PreserveReexecuteState preexecs(this);
3865     jvms()->set_should_reexecute(true);
3866 
3867     array_type_mirror = null_check(array_type_mirror);
3868     original          = null_check(original);
3869 
3870     // Check if a null path was taken unconditionally.
3871     if (stopped())  return true;
3872 
3873     Node* orig_length = load_array_length(original);
3874 
3875     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3876     klass_node = null_check(klass_node);
3877 
3878     RegionNode* bailout = new RegionNode(1);
3879     record_for_igvn(bailout);
3880 
3881     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3882     // Bail out if that is so.
3883     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3884     if (not_objArray != NULL) {
3885       // Improve the klass node's type from the new optimistic assumption:
3886       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3887       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3888       Node* cast = new CastPPNode(klass_node, akls);
3889       cast->init_req(0, control());
3890       klass_node = _gvn.transform(cast);
3891     }
3892 
3893     // Bail out if either start or end is negative.
3894     generate_negative_guard(start, bailout, &start);
3895     generate_negative_guard(end,   bailout, &end);
3896 
3897     Node* length = end;
3898     if (_gvn.type(start) != TypeInt::ZERO) {
3899       length = _gvn.transform(new SubINode(end, start));
3900     }
3901 
3902     // Bail out if length is negative.
3903     // Without this the new_array would throw
3904     // NegativeArraySizeException but IllegalArgumentException is what
3905     // should be thrown
3906     generate_negative_guard(length, bailout, &length);
3907 
3908     if (bailout->req() > 1) {
3909       PreserveJVMState pjvms(this);
3910       set_control(_gvn.transform(bailout));
3911       uncommon_trap(Deoptimization::Reason_intrinsic,
3912                     Deoptimization::Action_maybe_recompile);
3913     }
3914 
3915     if (!stopped()) {
3916       // How many elements will we copy from the original?
3917       // The answer is MinI(orig_length - start, length).
3918       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3919       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3920 
3921       // Generate a direct call to the right arraycopy function(s).
3922       // We know the copy is disjoint but we might not know if the
3923       // oop stores need checking.
3924       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3925       // This will fail a store-check if x contains any non-nulls.
3926 
3927       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3928       // loads/stores but it is legal only if we're sure the
3929       // Arrays.copyOf would succeed. So we need all input arguments
3930       // to the copyOf to be validated, including that the copy to the
3931       // new array won't trigger an ArrayStoreException. That subtype
3932       // check can be optimized if we know something on the type of
3933       // the input array from type speculation.
3934       if (_gvn.type(klass_node)->singleton()) {
3935         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3936         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3937 
3938         int test = C->static_subtype_check(superk, subk);
3939         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3940           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3941           if (t_original->speculative_type() != NULL) {
3942             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3943           }
3944         }
3945       }
3946 
3947       bool validated = false;
3948       // Reason_class_check rather than Reason_intrinsic because we
3949       // want to intrinsify even if this traps.
3950       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3951         Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3952                                                    klass_node);
3953 
3954         if (not_subtype_ctrl != top()) {
3955           PreserveJVMState pjvms(this);
3956           set_control(not_subtype_ctrl);
3957           uncommon_trap(Deoptimization::Reason_class_check,
3958                         Deoptimization::Action_make_not_entrant);
3959           assert(stopped(), "Should be stopped");
3960         }
3961         validated = true;
3962       }
3963 
3964       if (!stopped()) {
3965         newcopy = new_array(klass_node, length, 0);  // no arguments to push
3966 
3967         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
3968                                                 load_object_klass(original), klass_node);
3969         if (!is_copyOfRange) {
3970           ac->set_copyof(validated);
3971         } else {
3972           ac->set_copyofrange(validated);
3973         }
3974         Node* n = _gvn.transform(ac);
3975         if (n == ac) {
3976           ac->connect_outputs(this);
3977         } else {
3978           assert(validated, "shouldn't transform if all arguments not validated");
3979           set_all_memory(n);
3980         }
3981       }
3982     }
3983   } // original reexecute is set back here
3984 
3985   C->set_has_split_ifs(true); // Has chance for split-if optimization
3986   if (!stopped()) {
3987     set_result(newcopy);
3988   }
3989   return true;
3990 }
3991 
3992 
3993 //----------------------generate_virtual_guard---------------------------
3994 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
3995 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3996                                              RegionNode* slow_region) {
3997   ciMethod* method = callee();
3998   int vtable_index = method->vtable_index();
3999   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4000          err_msg_res("bad index %d", vtable_index));
4001   // Get the Method* out of the appropriate vtable entry.
4002   int entry_offset  = (InstanceKlass::vtable_start_offset() +
4003                      vtable_index*vtableEntry::size()) * wordSize +
4004                      vtableEntry::method_offset_in_bytes();
4005   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
4006   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4007 
4008   // Compare the target method with the expected method (e.g., Object.hashCode).
4009   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4010 
4011   Node* native_call = makecon(native_call_addr);
4012   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
4013   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4014 
4015   return generate_slow_guard(test_native, slow_region);
4016 }
4017 
4018 //-----------------------generate_method_call----------------------------
4019 // Use generate_method_call to make a slow-call to the real
4020 // method if the fast path fails.  An alternative would be to
4021 // use a stub like OptoRuntime::slow_arraycopy_Java.
4022 // This only works for expanding the current library call,
4023 // not another intrinsic.  (E.g., don't use this for making an
4024 // arraycopy call inside of the copyOf intrinsic.)
4025 CallJavaNode*
4026 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4027   // When compiling the intrinsic method itself, do not use this technique.
4028   guarantee(callee() != C->method(), "cannot make slow-call to self");
4029 
4030   ciMethod* method = callee();
4031   // ensure the JVMS we have will be correct for this call
4032   guarantee(method_id == method->intrinsic_id(), "must match");
4033 
4034   const TypeFunc* tf = TypeFunc::make(method);
4035   CallJavaNode* slow_call;
4036   if (is_static) {
4037     assert(!is_virtual, "");
4038     slow_call = new CallStaticJavaNode(C, tf,
4039                            SharedRuntime::get_resolve_static_call_stub(),
4040                            method, bci());
4041   } else if (is_virtual) {
4042     null_check_receiver();
4043     int vtable_index = Method::invalid_vtable_index;
4044     if (UseInlineCaches) {
4045       // Suppress the vtable call
4046     } else {
4047       // hashCode and clone are not a miranda methods,
4048       // so the vtable index is fixed.
4049       // No need to use the linkResolver to get it.
4050        vtable_index = method->vtable_index();
4051        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4052               err_msg_res("bad index %d", vtable_index));
4053     }
4054     slow_call = new CallDynamicJavaNode(tf,
4055                           SharedRuntime::get_resolve_virtual_call_stub(),
4056                           method, vtable_index, bci());
4057   } else {  // neither virtual nor static:  opt_virtual
4058     null_check_receiver();
4059     slow_call = new CallStaticJavaNode(C, tf,
4060                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4061                                 method, bci());
4062     slow_call->set_optimized_virtual(true);
4063   }
4064   set_arguments_for_java_call(slow_call);
4065   set_edges_for_java_call(slow_call);
4066   return slow_call;
4067 }
4068 
4069 
4070 /**
4071  * Build special case code for calls to hashCode on an object. This call may
4072  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4073  * slightly different code.
4074  */
4075 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4076   assert(is_static == callee()->is_static(), "correct intrinsic selection");
4077   assert(!(is_virtual && is_static), "either virtual, special, or static");
4078 
4079   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4080 
4081   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4082   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
4083   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
4084   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4085   Node* obj = NULL;
4086   if (!is_static) {
4087     // Check for hashing null object
4088     obj = null_check_receiver();
4089     if (stopped())  return true;        // unconditionally null
4090     result_reg->init_req(_null_path, top());
4091     result_val->init_req(_null_path, top());
4092   } else {
4093     // Do a null check, and return zero if null.
4094     // System.identityHashCode(null) == 0
4095     obj = argument(0);
4096     Node* null_ctl = top();
4097     obj = null_check_oop(obj, &null_ctl);
4098     result_reg->init_req(_null_path, null_ctl);
4099     result_val->init_req(_null_path, _gvn.intcon(0));
4100   }
4101 
4102   // Unconditionally null?  Then return right away.
4103   if (stopped()) {
4104     set_control( result_reg->in(_null_path));
4105     if (!stopped())
4106       set_result(result_val->in(_null_path));
4107     return true;
4108   }
4109 
4110   // We only go to the fast case code if we pass a number of guards.  The
4111   // paths which do not pass are accumulated in the slow_region.
4112   RegionNode* slow_region = new RegionNode(1);
4113   record_for_igvn(slow_region);
4114 
4115   // If this is a virtual call, we generate a funny guard.  We pull out
4116   // the vtable entry corresponding to hashCode() from the target object.
4117   // If the target method which we are calling happens to be the native
4118   // Object hashCode() method, we pass the guard.  We do not need this
4119   // guard for non-virtual calls -- the caller is known to be the native
4120   // Object hashCode().
4121   if (is_virtual) {
4122     // After null check, get the object's klass.
4123     Node* obj_klass = load_object_klass(obj);
4124     generate_virtual_guard(obj_klass, slow_region);
4125   }
4126 
4127   // Get the header out of the object, use LoadMarkNode when available
4128   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4129   // The control of the load must be NULL. Otherwise, the load can move before
4130   // the null check after castPP removal.
4131   Node* no_ctrl = NULL;
4132   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4133 
4134   // Test the header to see if it is unlocked.
4135   Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4136   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4137   Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
4138   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4139   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4140 
4141   generate_slow_guard(test_unlocked, slow_region);
4142 
4143   // Get the hash value and check to see that it has been properly assigned.
4144   // We depend on hash_mask being at most 32 bits and avoid the use of
4145   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4146   // vm: see markOop.hpp.
4147   Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
4148   Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
4149   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4150   // This hack lets the hash bits live anywhere in the mark object now, as long
4151   // as the shift drops the relevant bits into the low 32 bits.  Note that
4152   // Java spec says that HashCode is an int so there's no point in capturing
4153   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4154   hshifted_header      = ConvX2I(hshifted_header);
4155   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4156 
4157   Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
4158   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4159   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4160 
4161   generate_slow_guard(test_assigned, slow_region);
4162 
4163   Node* init_mem = reset_memory();
4164   // fill in the rest of the null path:
4165   result_io ->init_req(_null_path, i_o());
4166   result_mem->init_req(_null_path, init_mem);
4167 
4168   result_val->init_req(_fast_path, hash_val);
4169   result_reg->init_req(_fast_path, control());
4170   result_io ->init_req(_fast_path, i_o());
4171   result_mem->init_req(_fast_path, init_mem);
4172 
4173   // Generate code for the slow case.  We make a call to hashCode().
4174   set_control(_gvn.transform(slow_region));
4175   if (!stopped()) {
4176     // No need for PreserveJVMState, because we're using up the present state.
4177     set_all_memory(init_mem);
4178     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4179     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4180     Node* slow_result = set_results_for_java_call(slow_call);
4181     // this->control() comes from set_results_for_java_call
4182     result_reg->init_req(_slow_path, control());
4183     result_val->init_req(_slow_path, slow_result);
4184     result_io  ->set_req(_slow_path, i_o());
4185     result_mem ->set_req(_slow_path, reset_memory());
4186   }
4187 
4188   // Return the combined state.
4189   set_i_o(        _gvn.transform(result_io)  );
4190   set_all_memory( _gvn.transform(result_mem));
4191 
4192   set_result(result_reg, result_val);
4193   return true;
4194 }
4195 
4196 //---------------------------inline_native_getClass----------------------------
4197 // public final native Class<?> java.lang.Object.getClass();
4198 //
4199 // Build special case code for calls to getClass on an object.
4200 bool LibraryCallKit::inline_native_getClass() {
4201   Node* obj = null_check_receiver();
4202   if (stopped())  return true;
4203   set_result(load_mirror_from_klass(load_object_klass(obj)));
4204   return true;
4205 }
4206 
4207 //-----------------inline_native_Reflection_getCallerClass---------------------
4208 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4209 //
4210 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4211 //
4212 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4213 // in that it must skip particular security frames and checks for
4214 // caller sensitive methods.
4215 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4216 #ifndef PRODUCT
4217   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4218     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4219   }
4220 #endif
4221 
4222   if (!jvms()->has_method()) {
4223 #ifndef PRODUCT
4224     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4225       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
4226     }
4227 #endif
4228     return false;
4229   }
4230 
4231   // Walk back up the JVM state to find the caller at the required
4232   // depth.
4233   JVMState* caller_jvms = jvms();
4234 
4235   // Cf. JVM_GetCallerClass
4236   // NOTE: Start the loop at depth 1 because the current JVM state does
4237   // not include the Reflection.getCallerClass() frame.
4238   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4239     ciMethod* m = caller_jvms->method();
4240     switch (n) {
4241     case 0:
4242       fatal("current JVM state does not include the Reflection.getCallerClass frame");
4243       break;
4244     case 1:
4245       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4246       if (!m->caller_sensitive()) {
4247 #ifndef PRODUCT
4248         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4249           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
4250         }
4251 #endif
4252         return false;  // bail-out; let JVM_GetCallerClass do the work
4253       }
4254       break;
4255     default:
4256       if (!m->is_ignored_by_security_stack_walk()) {
4257         // We have reached the desired frame; return the holder class.
4258         // Acquire method holder as java.lang.Class and push as constant.
4259         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4260         ciInstance* caller_mirror = caller_klass->java_mirror();
4261         set_result(makecon(TypeInstPtr::make(caller_mirror)));
4262 
4263 #ifndef PRODUCT
4264         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4265           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());
4266           tty->print_cr("  JVM state at this point:");
4267           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4268             ciMethod* m = jvms()->of_depth(i)->method();
4269             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4270           }
4271         }
4272 #endif
4273         return true;
4274       }
4275       break;
4276     }
4277   }
4278 
4279 #ifndef PRODUCT
4280   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4281     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4282     tty->print_cr("  JVM state at this point:");
4283     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4284       ciMethod* m = jvms()->of_depth(i)->method();
4285       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4286     }
4287   }
4288 #endif
4289 
4290   return false;  // bail-out; let JVM_GetCallerClass do the work
4291 }
4292 
4293 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4294   Node* arg = argument(0);
4295   Node* result;
4296 
4297   switch (id) {
4298   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4299   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4300   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4301   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4302 
4303   case vmIntrinsics::_doubleToLongBits: {
4304     // two paths (plus control) merge in a wood
4305     RegionNode *r = new RegionNode(3);
4306     Node *phi = new PhiNode(r, TypeLong::LONG);
4307 
4308     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4309     // Build the boolean node
4310     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4311 
4312     // Branch either way.
4313     // NaN case is less traveled, which makes all the difference.
4314     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4315     Node *opt_isnan = _gvn.transform(ifisnan);
4316     assert( opt_isnan->is_If(), "Expect an IfNode");
4317     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4318     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4319 
4320     set_control(iftrue);
4321 
4322     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4323     Node *slow_result = longcon(nan_bits); // return NaN
4324     phi->init_req(1, _gvn.transform( slow_result ));
4325     r->init_req(1, iftrue);
4326 
4327     // Else fall through
4328     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4329     set_control(iffalse);
4330 
4331     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4332     r->init_req(2, iffalse);
4333 
4334     // Post merge
4335     set_control(_gvn.transform(r));
4336     record_for_igvn(r);
4337 
4338     C->set_has_split_ifs(true); // Has chance for split-if optimization
4339     result = phi;
4340     assert(result->bottom_type()->isa_long(), "must be");
4341     break;
4342   }
4343 
4344   case vmIntrinsics::_floatToIntBits: {
4345     // two paths (plus control) merge in a wood
4346     RegionNode *r = new RegionNode(3);
4347     Node *phi = new PhiNode(r, TypeInt::INT);
4348 
4349     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4350     // Build the boolean node
4351     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4352 
4353     // Branch either way.
4354     // NaN case is less traveled, which makes all the difference.
4355     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4356     Node *opt_isnan = _gvn.transform(ifisnan);
4357     assert( opt_isnan->is_If(), "Expect an IfNode");
4358     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4359     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4360 
4361     set_control(iftrue);
4362 
4363     static const jint nan_bits = 0x7fc00000;
4364     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4365     phi->init_req(1, _gvn.transform( slow_result ));
4366     r->init_req(1, iftrue);
4367 
4368     // Else fall through
4369     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4370     set_control(iffalse);
4371 
4372     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4373     r->init_req(2, iffalse);
4374 
4375     // Post merge
4376     set_control(_gvn.transform(r));
4377     record_for_igvn(r);
4378 
4379     C->set_has_split_ifs(true); // Has chance for split-if optimization
4380     result = phi;
4381     assert(result->bottom_type()->isa_int(), "must be");
4382     break;
4383   }
4384 
4385   default:
4386     fatal_unexpected_iid(id);
4387     break;
4388   }
4389   set_result(_gvn.transform(result));
4390   return true;
4391 }
4392 
4393 #ifdef _LP64
4394 #define XTOP ,top() /*additional argument*/
4395 #else  //_LP64
4396 #define XTOP        /*no additional argument*/
4397 #endif //_LP64
4398 
4399 //----------------------inline_unsafe_copyMemory-------------------------
4400 // public native void sun.misc.Unsafe.copyMemory(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4401 bool LibraryCallKit::inline_unsafe_copyMemory() {
4402   if (callee()->is_static())  return false;  // caller must have the capability!
4403   null_check_receiver();  // null-check receiver
4404   if (stopped())  return true;
4405 
4406   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4407 
4408   Node* src_ptr =         argument(1);   // type: oop
4409   Node* src_off = ConvL2X(argument(2));  // type: long
4410   Node* dst_ptr =         argument(4);   // type: oop
4411   Node* dst_off = ConvL2X(argument(5));  // type: long
4412   Node* size    = ConvL2X(argument(7));  // type: long
4413 
4414   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4415          "fieldOffset must be byte-scaled");
4416 
4417   Node* src = make_unsafe_address(src_ptr, src_off);
4418   Node* dst = make_unsafe_address(dst_ptr, dst_off);
4419 
4420   // Conservatively insert a memory barrier on all memory slices.
4421   // Do not let writes of the copy source or destination float below the copy.
4422   insert_mem_bar(Op_MemBarCPUOrder);
4423 
4424   // Call it.  Note that the length argument is not scaled.
4425   make_runtime_call(RC_LEAF|RC_NO_FP,
4426                     OptoRuntime::fast_arraycopy_Type(),
4427                     StubRoutines::unsafe_arraycopy(),
4428                     "unsafe_arraycopy",
4429                     TypeRawPtr::BOTTOM,
4430                     src, dst, size XTOP);
4431 
4432   // Do not let reads of the copy destination float above the copy.
4433   insert_mem_bar(Op_MemBarCPUOrder);
4434 
4435   return true;
4436 }
4437 
4438 //------------------------clone_coping-----------------------------------
4439 // Helper function for inline_native_clone.
4440 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4441   assert(obj_size != NULL, "");
4442   Node* raw_obj = alloc_obj->in(1);
4443   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4444 
4445   AllocateNode* alloc = NULL;
4446   if (ReduceBulkZeroing) {
4447     // We will be completely responsible for initializing this object -
4448     // mark Initialize node as complete.
4449     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4450     // The object was just allocated - there should be no any stores!
4451     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4452     // Mark as complete_with_arraycopy so that on AllocateNode
4453     // expansion, we know this AllocateNode is initialized by an array
4454     // copy and a StoreStore barrier exists after the array copy.
4455     alloc->initialization()->set_complete_with_arraycopy();
4456   }
4457 
4458   // Copy the fastest available way.
4459   // TODO: generate fields copies for small objects instead.
4460   Node* src  = obj;
4461   Node* dest = alloc_obj;
4462   Node* size = _gvn.transform(obj_size);
4463 
4464   // Exclude the header but include array length to copy by 8 bytes words.
4465   // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4466   int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4467                             instanceOopDesc::base_offset_in_bytes();
4468   // base_off:
4469   // 8  - 32-bit VM
4470   // 12 - 64-bit VM, compressed klass
4471   // 16 - 64-bit VM, normal klass
4472   if (base_off % BytesPerLong != 0) {
4473     assert(UseCompressedClassPointers, "");
4474     if (is_array) {
4475       // Exclude length to copy by 8 bytes words.
4476       base_off += sizeof(int);
4477     } else {
4478       // Include klass to copy by 8 bytes words.
4479       base_off = instanceOopDesc::klass_offset_in_bytes();
4480     }
4481     assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4482   }
4483   src  = basic_plus_adr(src,  base_off);
4484   dest = basic_plus_adr(dest, base_off);
4485 
4486   // Compute the length also, if needed:
4487   Node* countx = size;
4488   countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4489   countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4490 
4491   const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4492 
4493   ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4494   ac->set_clonebasic();
4495   Node* n = _gvn.transform(ac);
4496   if (n == ac) {
4497     set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4498   } else {
4499     set_all_memory(n);
4500   }
4501 
4502   // If necessary, emit some card marks afterwards.  (Non-arrays only.)
4503   if (card_mark) {
4504     assert(!is_array, "");
4505     // Put in store barrier for any and all oops we are sticking
4506     // into this object.  (We could avoid this if we could prove
4507     // that the object type contains no oop fields at all.)
4508     Node* no_particular_value = NULL;
4509     Node* no_particular_field = NULL;
4510     int raw_adr_idx = Compile::AliasIdxRaw;
4511     post_barrier(control(),
4512                  memory(raw_adr_type),
4513                  alloc_obj,
4514                  no_particular_field,
4515                  raw_adr_idx,
4516                  no_particular_value,
4517                  T_OBJECT,
4518                  false);
4519   }
4520 
4521   // Do not let reads from the cloned object float above the arraycopy.
4522   if (alloc != NULL) {
4523     // Do not let stores that initialize this object be reordered with
4524     // a subsequent store that would make this object accessible by
4525     // other threads.
4526     // Record what AllocateNode this StoreStore protects so that
4527     // escape analysis can go from the MemBarStoreStoreNode to the
4528     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4529     // based on the escape status of the AllocateNode.
4530     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4531   } else {
4532     insert_mem_bar(Op_MemBarCPUOrder);
4533   }
4534 }
4535 
4536 //------------------------inline_native_clone----------------------------
4537 // protected native Object java.lang.Object.clone();
4538 //
4539 // Here are the simple edge cases:
4540 //  null receiver => normal trap
4541 //  virtual and clone was overridden => slow path to out-of-line clone
4542 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4543 //
4544 // The general case has two steps, allocation and copying.
4545 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4546 //
4547 // Copying also has two cases, oop arrays and everything else.
4548 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4549 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4550 //
4551 // These steps fold up nicely if and when the cloned object's klass
4552 // can be sharply typed as an object array, a type array, or an instance.
4553 //
4554 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4555   PhiNode* result_val;
4556 
4557   // Set the reexecute bit for the interpreter to reexecute
4558   // the bytecode that invokes Object.clone if deoptimization happens.
4559   { PreserveReexecuteState preexecs(this);
4560     jvms()->set_should_reexecute(true);
4561 
4562     Node* obj = null_check_receiver();
4563     if (stopped())  return true;
4564 
4565     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4566 
4567     // If we are going to clone an instance, we need its exact type to
4568     // know the number and types of fields to convert the clone to
4569     // loads/stores. Maybe a speculative type can help us.
4570     if (!obj_type->klass_is_exact() &&
4571         obj_type->speculative_type() != NULL &&
4572         obj_type->speculative_type()->is_instance_klass()) {
4573       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4574       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4575           !spec_ik->has_injected_fields()) {
4576         ciKlass* k = obj_type->klass();
4577         if (!k->is_instance_klass() ||
4578             k->as_instance_klass()->is_interface() ||
4579             k->as_instance_klass()->has_subklass()) {
4580           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4581         }
4582       }
4583     }
4584 
4585     Node* obj_klass = load_object_klass(obj);
4586     const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4587     const TypeOopPtr*   toop   = ((tklass != NULL)
4588                                 ? tklass->as_instance_type()
4589                                 : TypeInstPtr::NOTNULL);
4590 
4591     // Conservatively insert a memory barrier on all memory slices.
4592     // Do not let writes into the original float below the clone.
4593     insert_mem_bar(Op_MemBarCPUOrder);
4594 
4595     // paths into result_reg:
4596     enum {
4597       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4598       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4599       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4600       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4601       PATH_LIMIT
4602     };
4603     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4604     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4605     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4606     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4607     record_for_igvn(result_reg);
4608 
4609     const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4610     int raw_adr_idx = Compile::AliasIdxRaw;
4611 
4612     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4613     if (array_ctl != NULL) {
4614       // It's an array.
4615       PreserveJVMState pjvms(this);
4616       set_control(array_ctl);
4617       Node* obj_length = load_array_length(obj);
4618       Node* obj_size  = NULL;
4619       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4620 
4621       if (!use_ReduceInitialCardMarks()) {
4622         // If it is an oop array, it requires very special treatment,
4623         // because card marking is required on each card of the array.
4624         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4625         if (is_obja != NULL) {
4626           PreserveJVMState pjvms2(this);
4627           set_control(is_obja);
4628           // Generate a direct call to the right arraycopy function(s).
4629           Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4630           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4631           ac->set_cloneoop();
4632           Node* n = _gvn.transform(ac);
4633           assert(n == ac, "cannot disappear");
4634           ac->connect_outputs(this);
4635 
4636           result_reg->init_req(_objArray_path, control());
4637           result_val->init_req(_objArray_path, alloc_obj);
4638           result_i_o ->set_req(_objArray_path, i_o());
4639           result_mem ->set_req(_objArray_path, reset_memory());
4640         }
4641       }
4642       // Otherwise, there are no card marks to worry about.
4643       // (We can dispense with card marks if we know the allocation
4644       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4645       //  causes the non-eden paths to take compensating steps to
4646       //  simulate a fresh allocation, so that no further
4647       //  card marks are required in compiled code to initialize
4648       //  the object.)
4649 
4650       if (!stopped()) {
4651         copy_to_clone(obj, alloc_obj, obj_size, true, false);
4652 
4653         // Present the results of the copy.
4654         result_reg->init_req(_array_path, control());
4655         result_val->init_req(_array_path, alloc_obj);
4656         result_i_o ->set_req(_array_path, i_o());
4657         result_mem ->set_req(_array_path, reset_memory());
4658       }
4659     }
4660 
4661     // We only go to the instance fast case code if we pass a number of guards.
4662     // The paths which do not pass are accumulated in the slow_region.
4663     RegionNode* slow_region = new RegionNode(1);
4664     record_for_igvn(slow_region);
4665     if (!stopped()) {
4666       // It's an instance (we did array above).  Make the slow-path tests.
4667       // If this is a virtual call, we generate a funny guard.  We grab
4668       // the vtable entry corresponding to clone() from the target object.
4669       // If the target method which we are calling happens to be the
4670       // Object clone() method, we pass the guard.  We do not need this
4671       // guard for non-virtual calls; the caller is known to be the native
4672       // Object clone().
4673       if (is_virtual) {
4674         generate_virtual_guard(obj_klass, slow_region);
4675       }
4676 
4677       // The object must be cloneable and must not have a finalizer.
4678       // Both of these conditions may be checked in a single test.
4679       // We could optimize the cloneable test further, but we don't care.
4680       generate_access_flags_guard(obj_klass,
4681                                   // Test both conditions:
4682                                   JVM_ACC_IS_CLONEABLE | JVM_ACC_HAS_FINALIZER,
4683                                   // Must be cloneable but not finalizer:
4684                                   JVM_ACC_IS_CLONEABLE,
4685                                   slow_region);
4686     }
4687 
4688     if (!stopped()) {
4689       // It's an instance, and it passed the slow-path tests.
4690       PreserveJVMState pjvms(this);
4691       Node* obj_size  = NULL;
4692       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4693       // is reexecuted if deoptimization occurs and there could be problems when merging
4694       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4695       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4696 
4697       copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4698 
4699       // Present the results of the slow call.
4700       result_reg->init_req(_instance_path, control());
4701       result_val->init_req(_instance_path, alloc_obj);
4702       result_i_o ->set_req(_instance_path, i_o());
4703       result_mem ->set_req(_instance_path, reset_memory());
4704     }
4705 
4706     // Generate code for the slow case.  We make a call to clone().
4707     set_control(_gvn.transform(slow_region));
4708     if (!stopped()) {
4709       PreserveJVMState pjvms(this);
4710       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4711       Node* slow_result = set_results_for_java_call(slow_call);
4712       // this->control() comes from set_results_for_java_call
4713       result_reg->init_req(_slow_path, control());
4714       result_val->init_req(_slow_path, slow_result);
4715       result_i_o ->set_req(_slow_path, i_o());
4716       result_mem ->set_req(_slow_path, reset_memory());
4717     }
4718 
4719     // Return the combined state.
4720     set_control(    _gvn.transform(result_reg));
4721     set_i_o(        _gvn.transform(result_i_o));
4722     set_all_memory( _gvn.transform(result_mem));
4723   } // original reexecute is set back here
4724 
4725   set_result(_gvn.transform(result_val));
4726   return true;
4727 }
4728 
4729 // If we have a tighly coupled allocation, the arraycopy may take care
4730 // of the array initialization. If one of the guards we insert between
4731 // the allocation and the arraycopy causes a deoptimization, an
4732 // unitialized array will escape the compiled method. To prevent that
4733 // we set the JVM state for uncommon traps between the allocation and
4734 // the arraycopy to the state before the allocation so, in case of
4735 // deoptimization, we'll reexecute the allocation and the
4736 // initialization.
4737 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4738   if (alloc != NULL) {
4739     ciMethod* trap_method = alloc->jvms()->method();
4740     int trap_bci = alloc->jvms()->bci();
4741 
4742     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4743           !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4744       // Make sure there's no store between the allocation and the
4745       // arraycopy otherwise visible side effects could be rexecuted
4746       // in case of deoptimization and cause incorrect execution.
4747       bool no_interfering_store = true;
4748       Node* mem = alloc->in(TypeFunc::Memory);
4749       if (mem->is_MergeMem()) {
4750         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4751           Node* n = mms.memory();
4752           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4753             assert(n->is_Store(), "what else?");
4754             no_interfering_store = false;
4755             break;
4756           }
4757         }
4758       } else {
4759         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4760           Node* n = mms.memory();
4761           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4762             assert(n->is_Store(), "what else?");
4763             no_interfering_store = false;
4764             break;
4765           }
4766         }
4767       }
4768 
4769       if (no_interfering_store) {
4770         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4771         uint size = alloc->req();
4772         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4773         old_jvms->set_map(sfpt);
4774         for (uint i = 0; i < size; i++) {
4775           sfpt->init_req(i, alloc->in(i));
4776         }
4777         // re-push array length for deoptimization
4778         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4779         old_jvms->set_sp(old_jvms->sp()+1);
4780         old_jvms->set_monoff(old_jvms->monoff()+1);
4781         old_jvms->set_scloff(old_jvms->scloff()+1);
4782         old_jvms->set_endoff(old_jvms->endoff()+1);
4783         old_jvms->set_should_reexecute(true);
4784 
4785         sfpt->set_i_o(map()->i_o());
4786         sfpt->set_memory(map()->memory());
4787         sfpt->set_control(map()->control());
4788 
4789         JVMState* saved_jvms = jvms();
4790         saved_reexecute_sp = _reexecute_sp;
4791 
4792         set_jvms(sfpt->jvms());
4793         _reexecute_sp = jvms()->sp();
4794 
4795         return saved_jvms;
4796       }
4797     }
4798   }
4799   return NULL;
4800 }
4801 
4802 // In case of a deoptimization, we restart execution at the
4803 // allocation, allocating a new array. We would leave an uninitialized
4804 // array in the heap that GCs wouldn't expect. Move the allocation
4805 // after the traps so we don't allocate the array if we
4806 // deoptimize. This is possible because tightly_coupled_allocation()
4807 // guarantees there's no observer of the allocated array at this point
4808 // and the control flow is simple enough.
4809 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4810   if (saved_jvms != NULL && !stopped()) {
4811     assert(alloc != NULL, "only with a tightly coupled allocation");
4812     // restore JVM state to the state at the arraycopy
4813     saved_jvms->map()->set_control(map()->control());
4814     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4815     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4816     // If we've improved the types of some nodes (null check) while
4817     // emitting the guards, propagate them to the current state
4818     map()->replaced_nodes().apply(saved_jvms->map());
4819     set_jvms(saved_jvms);
4820     _reexecute_sp = saved_reexecute_sp;
4821 
4822     // Remove the allocation from above the guards
4823     CallProjections callprojs;
4824     alloc->extract_projections(&callprojs, true);
4825     InitializeNode* init = alloc->initialization();
4826     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4827     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4828     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4829     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4830 
4831     // move the allocation here (after the guards)
4832     _gvn.hash_delete(alloc);
4833     alloc->set_req(TypeFunc::Control, control());
4834     alloc->set_req(TypeFunc::I_O, i_o());
4835     Node *mem = reset_memory();
4836     set_all_memory(mem);
4837     alloc->set_req(TypeFunc::Memory, mem);
4838     set_control(init->proj_out(TypeFunc::Control));
4839     set_i_o(callprojs.fallthrough_ioproj);
4840 
4841     // Update memory as done in GraphKit::set_output_for_allocation()
4842     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4843     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4844     if (ary_type->isa_aryptr() && length_type != NULL) {
4845       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4846     }
4847     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4848     int            elemidx  = C->get_alias_index(telemref);
4849     set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4850     set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4851 
4852     Node* allocx = _gvn.transform(alloc);
4853     assert(allocx == alloc, "where has the allocation gone?");
4854     assert(dest->is_CheckCastPP(), "not an allocation result?");
4855 
4856     _gvn.hash_delete(dest);
4857     dest->set_req(0, control());
4858     Node* destx = _gvn.transform(dest);
4859     assert(destx == dest, "where has the allocation result gone?");
4860   }
4861 }
4862 
4863 
4864 //------------------------------inline_arraycopy-----------------------
4865 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4866 //                                                      Object dest, int destPos,
4867 //                                                      int length);
4868 bool LibraryCallKit::inline_arraycopy() {
4869   // Get the arguments.
4870   Node* src         = argument(0);  // type: oop
4871   Node* src_offset  = argument(1);  // type: int
4872   Node* dest        = argument(2);  // type: oop
4873   Node* dest_offset = argument(3);  // type: int
4874   Node* length      = argument(4);  // type: int
4875 
4876 
4877   // Check for allocation before we add nodes that would confuse
4878   // tightly_coupled_allocation()
4879   AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4880 
4881   int saved_reexecute_sp = -1;
4882   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4883   // See arraycopy_restore_alloc_state() comment
4884   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4885   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4886   // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4887   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4888 
4889   // The following tests must be performed
4890   // (1) src and dest are arrays.
4891   // (2) src and dest arrays must have elements of the same BasicType
4892   // (3) src and dest must not be null.
4893   // (4) src_offset must not be negative.
4894   // (5) dest_offset must not be negative.
4895   // (6) length must not be negative.
4896   // (7) src_offset + length must not exceed length of src.
4897   // (8) dest_offset + length must not exceed length of dest.
4898   // (9) each element of an oop array must be assignable
4899 
4900   // (3) src and dest must not be null.
4901   // always do this here because we need the JVM state for uncommon traps
4902   Node* null_ctl = top();
4903   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4904   assert(null_ctl->is_top(), "no null control here");
4905   dest = null_check(dest, T_ARRAY);
4906 
4907   if (!can_emit_guards) {
4908     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4909     // guards but the arraycopy node could still take advantage of a
4910     // tightly allocated allocation. tightly_coupled_allocation() is
4911     // called again to make sure it takes the null check above into
4912     // account: the null check is mandatory and if it caused an
4913     // uncommon trap to be emitted then the allocation can't be
4914     // considered tightly coupled in this context.
4915     alloc = tightly_coupled_allocation(dest, NULL);
4916   }
4917 
4918   bool validated = false;
4919 
4920   const Type* src_type  = _gvn.type(src);
4921   const Type* dest_type = _gvn.type(dest);
4922   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4923   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4924 
4925   // Do we have the type of src?
4926   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4927   // Do we have the type of dest?
4928   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4929   // Is the type for src from speculation?
4930   bool src_spec = false;
4931   // Is the type for dest from speculation?
4932   bool dest_spec = false;
4933 
4934   if ((!has_src || !has_dest) && can_emit_guards) {
4935     // We don't have sufficient type information, let's see if
4936     // speculative types can help. We need to have types for both src
4937     // and dest so that it pays off.
4938 
4939     // Do we already have or could we have type information for src
4940     bool could_have_src = has_src;
4941     // Do we already have or could we have type information for dest
4942     bool could_have_dest = has_dest;
4943 
4944     ciKlass* src_k = NULL;
4945     if (!has_src) {
4946       src_k = src_type->speculative_type_not_null();
4947       if (src_k != NULL && src_k->is_array_klass()) {
4948         could_have_src = true;
4949       }
4950     }
4951 
4952     ciKlass* dest_k = NULL;
4953     if (!has_dest) {
4954       dest_k = dest_type->speculative_type_not_null();
4955       if (dest_k != NULL && dest_k->is_array_klass()) {
4956         could_have_dest = true;
4957       }
4958     }
4959 
4960     if (could_have_src && could_have_dest) {
4961       // This is going to pay off so emit the required guards
4962       if (!has_src) {
4963         src = maybe_cast_profiled_obj(src, src_k, true);
4964         src_type  = _gvn.type(src);
4965         top_src  = src_type->isa_aryptr();
4966         has_src = (top_src != NULL && top_src->klass() != NULL);
4967         src_spec = true;
4968       }
4969       if (!has_dest) {
4970         dest = maybe_cast_profiled_obj(dest, dest_k, true);
4971         dest_type  = _gvn.type(dest);
4972         top_dest  = dest_type->isa_aryptr();
4973         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4974         dest_spec = true;
4975       }
4976     }
4977   }
4978 
4979   if (has_src && has_dest && can_emit_guards) {
4980     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
4981     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4982     if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
4983     if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;
4984 
4985     if (src_elem == dest_elem && src_elem == T_OBJECT) {
4986       // If both arrays are object arrays then having the exact types
4987       // for both will remove the need for a subtype check at runtime
4988       // before the call and may make it possible to pick a faster copy
4989       // routine (without a subtype check on every element)
4990       // Do we have the exact type of src?
4991       bool could_have_src = src_spec;
4992       // Do we have the exact type of dest?
4993       bool could_have_dest = dest_spec;
4994       ciKlass* src_k = top_src->klass();
4995       ciKlass* dest_k = top_dest->klass();
4996       if (!src_spec) {
4997         src_k = src_type->speculative_type_not_null();
4998         if (src_k != NULL && src_k->is_array_klass()) {
4999           could_have_src = true;
5000         }
5001       }
5002       if (!dest_spec) {
5003         dest_k = dest_type->speculative_type_not_null();
5004         if (dest_k != NULL && dest_k->is_array_klass()) {
5005           could_have_dest = true;
5006         }
5007       }
5008       if (could_have_src && could_have_dest) {
5009         // If we can have both exact types, emit the missing guards
5010         if (could_have_src && !src_spec) {
5011           src = maybe_cast_profiled_obj(src, src_k, true);
5012         }
5013         if (could_have_dest && !dest_spec) {
5014           dest = maybe_cast_profiled_obj(dest, dest_k, true);
5015         }
5016       }
5017     }
5018   }
5019 
5020   ciMethod* trap_method = method();
5021   int trap_bci = bci();
5022   if (saved_jvms != NULL) {
5023     trap_method = alloc->jvms()->method();
5024     trap_bci = alloc->jvms()->bci();
5025   }
5026 
5027   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5028       can_emit_guards &&
5029       !src->is_top() && !dest->is_top()) {
5030     // validate arguments: enables transformation the ArrayCopyNode
5031     validated = true;
5032 
5033     RegionNode* slow_region = new RegionNode(1);
5034     record_for_igvn(slow_region);
5035 
5036     // (1) src and dest are arrays.
5037     generate_non_array_guard(load_object_klass(src), slow_region);
5038     generate_non_array_guard(load_object_klass(dest), slow_region);
5039 
5040     // (2) src and dest arrays must have elements of the same BasicType
5041     // done at macro expansion or at Ideal transformation time
5042 
5043     // (4) src_offset must not be negative.
5044     generate_negative_guard(src_offset, slow_region);
5045 
5046     // (5) dest_offset must not be negative.
5047     generate_negative_guard(dest_offset, slow_region);
5048 
5049     // (7) src_offset + length must not exceed length of src.
5050     generate_limit_guard(src_offset, length,
5051                          load_array_length(src),
5052                          slow_region);
5053 
5054     // (8) dest_offset + length must not exceed length of dest.
5055     generate_limit_guard(dest_offset, length,
5056                          load_array_length(dest),
5057                          slow_region);
5058 
5059     // (9) each element of an oop array must be assignable
5060     Node* src_klass  = load_object_klass(src);
5061     Node* dest_klass = load_object_klass(dest);
5062     Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5063 
5064     if (not_subtype_ctrl != top()) {
5065       PreserveJVMState pjvms(this);
5066       set_control(not_subtype_ctrl);
5067       uncommon_trap(Deoptimization::Reason_intrinsic,
5068                     Deoptimization::Action_make_not_entrant);
5069       assert(stopped(), "Should be stopped");
5070     }
5071     {
5072       PreserveJVMState pjvms(this);
5073       set_control(_gvn.transform(slow_region));
5074       uncommon_trap(Deoptimization::Reason_intrinsic,
5075                     Deoptimization::Action_make_not_entrant);
5076       assert(stopped(), "Should be stopped");
5077     }
5078   }
5079 
5080   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5081 
5082   if (stopped()) {
5083     return true;
5084   }
5085 
5086   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5087                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
5088                                           // so the compiler has a chance to eliminate them: during macro expansion,
5089                                           // we have to set their control (CastPP nodes are eliminated).
5090                                           load_object_klass(src), load_object_klass(dest),
5091                                           load_array_length(src), load_array_length(dest));
5092 
5093   ac->set_arraycopy(validated);
5094 
5095   Node* n = _gvn.transform(ac);
5096   if (n == ac) {
5097     ac->connect_outputs(this);
5098   } else {
5099     assert(validated, "shouldn't transform if all arguments not validated");
5100     set_all_memory(n);
5101   }
5102 
5103   return true;
5104 }
5105 
5106 
5107 // Helper function which determines if an arraycopy immediately follows
5108 // an allocation, with no intervening tests or other escapes for the object.
5109 AllocateArrayNode*
5110 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5111                                            RegionNode* slow_region) {
5112   if (stopped())             return NULL;  // no fast path
5113   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
5114 
5115   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5116   if (alloc == NULL)  return NULL;
5117 
5118   Node* rawmem = memory(Compile::AliasIdxRaw);
5119   // Is the allocation's memory state untouched?
5120   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5121     // Bail out if there have been raw-memory effects since the allocation.
5122     // (Example:  There might have been a call or safepoint.)
5123     return NULL;
5124   }
5125   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5126   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5127     return NULL;
5128   }
5129 
5130   // There must be no unexpected observers of this allocation.
5131   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5132     Node* obs = ptr->fast_out(i);
5133     if (obs != this->map()) {
5134       return NULL;
5135     }
5136   }
5137 
5138   // This arraycopy must unconditionally follow the allocation of the ptr.
5139   Node* alloc_ctl = ptr->in(0);
5140   assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5141 
5142   Node* ctl = control();
5143   while (ctl != alloc_ctl) {
5144     // There may be guards which feed into the slow_region.
5145     // Any other control flow means that we might not get a chance
5146     // to finish initializing the allocated object.
5147     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5148       IfNode* iff = ctl->in(0)->as_If();
5149       Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5150       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5151       if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5152         ctl = iff->in(0);       // This test feeds the known slow_region.
5153         continue;
5154       }
5155       // One more try:  Various low-level checks bottom out in
5156       // uncommon traps.  If the debug-info of the trap omits
5157       // any reference to the allocation, as we've already
5158       // observed, then there can be no objection to the trap.
5159       bool found_trap = false;
5160       for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5161         Node* obs = not_ctl->fast_out(j);
5162         if (obs->in(0) == not_ctl && obs->is_Call() &&
5163             (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5164           found_trap = true; break;
5165         }
5166       }
5167       if (found_trap) {
5168         ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
5169         continue;
5170       }
5171     }
5172     return NULL;
5173   }
5174 
5175   // If we get this far, we have an allocation which immediately
5176   // precedes the arraycopy, and we can take over zeroing the new object.
5177   // The arraycopy will finish the initialization, and provide
5178   // a new control state to which we will anchor the destination pointer.
5179 
5180   return alloc;
5181 }
5182 
5183 //-------------inline_encodeISOArray-----------------------------------
5184 // encode char[] to byte[] in ISO_8859_1
5185 bool LibraryCallKit::inline_encodeISOArray() {
5186   assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5187   // no receiver since it is static method
5188   Node *src         = argument(0);
5189   Node *src_offset  = argument(1);
5190   Node *dst         = argument(2);
5191   Node *dst_offset  = argument(3);
5192   Node *length      = argument(4);
5193 
5194   const Type* src_type = src->Value(&_gvn);
5195   const Type* dst_type = dst->Value(&_gvn);
5196   const TypeAryPtr* top_src = src_type->isa_aryptr();
5197   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5198   if (top_src  == NULL || top_src->klass()  == NULL ||
5199       top_dest == NULL || top_dest->klass() == NULL) {
5200     // failed array check
5201     return false;
5202   }
5203 
5204   // Figure out the size and type of the elements we will be copying.
5205   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5206   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5207   if (src_elem != T_CHAR || dst_elem != T_BYTE) {
5208     return false;
5209   }
5210   Node* src_start = array_element_address(src, src_offset, src_elem);
5211   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5212   // 'src_start' points to src array + scaled offset
5213   // 'dst_start' points to dst array + scaled offset
5214 
5215   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5216   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5217   enc = _gvn.transform(enc);
5218   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5219   set_memory(res_mem, mtype);
5220   set_result(enc);
5221   return true;
5222 }
5223 
5224 //-------------inline_multiplyToLen-----------------------------------
5225 bool LibraryCallKit::inline_multiplyToLen() {
5226   assert(UseMultiplyToLenIntrinsic, "not implementated on this platform");
5227 
5228   address stubAddr = StubRoutines::multiplyToLen();
5229   if (stubAddr == NULL) {
5230     return false; // Intrinsic's stub is not implemented on this platform
5231   }
5232   const char* stubName = "multiplyToLen";
5233 
5234   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5235 
5236   Node* x    = argument(1);
5237   Node* xlen = argument(2);
5238   Node* y    = argument(3);
5239   Node* ylen = argument(4);
5240   Node* z    = argument(5);
5241 
5242   const Type* x_type = x->Value(&_gvn);
5243   const Type* y_type = y->Value(&_gvn);
5244   const TypeAryPtr* top_x = x_type->isa_aryptr();
5245   const TypeAryPtr* top_y = y_type->isa_aryptr();
5246   if (top_x  == NULL || top_x->klass()  == NULL ||
5247       top_y == NULL || top_y->klass() == NULL) {
5248     // failed array check
5249     return false;
5250   }
5251 
5252   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5253   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5254   if (x_elem != T_INT || y_elem != T_INT) {
5255     return false;
5256   }
5257 
5258   // Set the original stack and the reexecute bit for the interpreter to reexecute
5259   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5260   // on the return from z array allocation in runtime.
5261   { PreserveReexecuteState preexecs(this);
5262     jvms()->set_should_reexecute(true);
5263 
5264     Node* x_start = array_element_address(x, intcon(0), x_elem);
5265     Node* y_start = array_element_address(y, intcon(0), y_elem);
5266     // 'x_start' points to x array + scaled xlen
5267     // 'y_start' points to y array + scaled ylen
5268 
5269     // Allocate the result array
5270     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5271     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5272     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5273 
5274     IdealKit ideal(this);
5275 
5276 #define __ ideal.
5277      Node* one = __ ConI(1);
5278      Node* zero = __ ConI(0);
5279      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5280      __ set(need_alloc, zero);
5281      __ set(z_alloc, z);
5282      __ if_then(z, BoolTest::eq, null()); {
5283        __ increment (need_alloc, one);
5284      } __ else_(); {
5285        // Update graphKit memory and control from IdealKit.
5286        sync_kit(ideal);
5287        Node* zlen_arg = load_array_length(z);
5288        // Update IdealKit memory and control from graphKit.
5289        __ sync_kit(this);
5290        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5291          __ increment (need_alloc, one);
5292        } __ end_if();
5293      } __ end_if();
5294 
5295      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5296        // Update graphKit memory and control from IdealKit.
5297        sync_kit(ideal);
5298        Node * narr = new_array(klass_node, zlen, 1);
5299        // Update IdealKit memory and control from graphKit.
5300        __ sync_kit(this);
5301        __ set(z_alloc, narr);
5302      } __ end_if();
5303 
5304      sync_kit(ideal);
5305      z = __ value(z_alloc);
5306      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5307      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5308      // Final sync IdealKit and GraphKit.
5309      final_sync(ideal);
5310 #undef __
5311 
5312     Node* z_start = array_element_address(z, intcon(0), T_INT);
5313 
5314     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5315                                    OptoRuntime::multiplyToLen_Type(),
5316                                    stubAddr, stubName, TypePtr::BOTTOM,
5317                                    x_start, xlen, y_start, ylen, z_start, zlen);
5318   } // original reexecute is set back here
5319 
5320   C->set_has_split_ifs(true); // Has chance for split-if optimization
5321   set_result(z);
5322   return true;
5323 }
5324 
5325 //-------------inline_squareToLen------------------------------------
5326 bool LibraryCallKit::inline_squareToLen() {
5327   assert(UseSquareToLenIntrinsic, "not implementated on this platform");
5328 
5329   address stubAddr = StubRoutines::squareToLen();
5330   if (stubAddr == NULL) {
5331     return false; // Intrinsic's stub is not implemented on this platform
5332   }
5333   const char* stubName = "squareToLen";
5334 
5335   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5336 
5337   Node* x    = argument(0);
5338   Node* len  = argument(1);
5339   Node* z    = argument(2);
5340   Node* zlen = argument(3);
5341 
5342   const Type* x_type = x->Value(&_gvn);
5343   const Type* z_type = z->Value(&_gvn);
5344   const TypeAryPtr* top_x = x_type->isa_aryptr();
5345   const TypeAryPtr* top_z = z_type->isa_aryptr();
5346   if (top_x  == NULL || top_x->klass()  == NULL ||
5347       top_z  == NULL || top_z->klass()  == NULL) {
5348     // failed array check
5349     return false;
5350   }
5351 
5352   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5353   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5354   if (x_elem != T_INT || z_elem != T_INT) {
5355     return false;
5356   }
5357 
5358 
5359   Node* x_start = array_element_address(x, intcon(0), x_elem);
5360   Node* z_start = array_element_address(z, intcon(0), z_elem);
5361 
5362   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5363                                   OptoRuntime::squareToLen_Type(),
5364                                   stubAddr, stubName, TypePtr::BOTTOM,
5365                                   x_start, len, z_start, zlen);
5366 
5367   set_result(z);
5368   return true;
5369 }
5370 
5371 //-------------inline_mulAdd------------------------------------------
5372 bool LibraryCallKit::inline_mulAdd() {
5373   assert(UseMulAddIntrinsic, "not implementated on this platform");
5374 
5375   address stubAddr = StubRoutines::mulAdd();
5376   if (stubAddr == NULL) {
5377     return false; // Intrinsic's stub is not implemented on this platform
5378   }
5379   const char* stubName = "mulAdd";
5380 
5381   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5382 
5383   Node* out      = argument(0);
5384   Node* in       = argument(1);
5385   Node* offset   = argument(2);
5386   Node* len      = argument(3);
5387   Node* k        = argument(4);
5388 
5389   const Type* out_type = out->Value(&_gvn);
5390   const Type* in_type = in->Value(&_gvn);
5391   const TypeAryPtr* top_out = out_type->isa_aryptr();
5392   const TypeAryPtr* top_in = in_type->isa_aryptr();
5393   if (top_out  == NULL || top_out->klass()  == NULL ||
5394       top_in == NULL || top_in->klass() == NULL) {
5395     // failed array check
5396     return false;
5397   }
5398 
5399   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5400   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5401   if (out_elem != T_INT || in_elem != T_INT) {
5402     return false;
5403   }
5404 
5405   Node* outlen = load_array_length(out);
5406   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5407   Node* out_start = array_element_address(out, intcon(0), out_elem);
5408   Node* in_start = array_element_address(in, intcon(0), in_elem);
5409 
5410   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5411                                   OptoRuntime::mulAdd_Type(),
5412                                   stubAddr, stubName, TypePtr::BOTTOM,
5413                                   out_start,in_start, new_offset, len, k);
5414   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5415   set_result(result);
5416   return true;
5417 }
5418 
5419 
5420 /**
5421  * Calculate CRC32 for byte.
5422  * int java.util.zip.CRC32.update(int crc, int b)
5423  */
5424 bool LibraryCallKit::inline_updateCRC32() {
5425   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5426   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5427   // no receiver since it is static method
5428   Node* crc  = argument(0); // type: int
5429   Node* b    = argument(1); // type: int
5430 
5431   /*
5432    *    int c = ~ crc;
5433    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5434    *    b = b ^ (c >>> 8);
5435    *    crc = ~b;
5436    */
5437 
5438   Node* M1 = intcon(-1);
5439   crc = _gvn.transform(new XorINode(crc, M1));
5440   Node* result = _gvn.transform(new XorINode(crc, b));
5441   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5442 
5443   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5444   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5445   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5446   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5447 
5448   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5449   result = _gvn.transform(new XorINode(crc, result));
5450   result = _gvn.transform(new XorINode(result, M1));
5451   set_result(result);
5452   return true;
5453 }
5454 
5455 /**
5456  * Calculate CRC32 for byte[] array.
5457  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5458  */
5459 bool LibraryCallKit::inline_updateBytesCRC32() {
5460   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5461   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5462   // no receiver since it is static method
5463   Node* crc     = argument(0); // type: int
5464   Node* src     = argument(1); // type: oop
5465   Node* offset  = argument(2); // type: int
5466   Node* length  = argument(3); // type: int
5467 
5468   const Type* src_type = src->Value(&_gvn);
5469   const TypeAryPtr* top_src = src_type->isa_aryptr();
5470   if (top_src  == NULL || top_src->klass()  == NULL) {
5471     // failed array check
5472     return false;
5473   }
5474 
5475   // Figure out the size and type of the elements we will be copying.
5476   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5477   if (src_elem != T_BYTE) {
5478     return false;
5479   }
5480 
5481   // 'src_start' points to src array + scaled offset
5482   Node* src_start = array_element_address(src, offset, src_elem);
5483 
5484   // We assume that range check is done by caller.
5485   // TODO: generate range check (offset+length < src.length) in debug VM.
5486 
5487   // Call the stub.
5488   address stubAddr = StubRoutines::updateBytesCRC32();
5489   const char *stubName = "updateBytesCRC32";
5490 
5491   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5492                                  stubAddr, stubName, TypePtr::BOTTOM,
5493                                  crc, src_start, length);
5494   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5495   set_result(result);
5496   return true;
5497 }
5498 
5499 /**
5500  * Calculate CRC32 for ByteBuffer.
5501  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5502  */
5503 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5504   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5505   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5506   // no receiver since it is static method
5507   Node* crc     = argument(0); // type: int
5508   Node* src     = argument(1); // type: long
5509   Node* offset  = argument(3); // type: int
5510   Node* length  = argument(4); // type: int
5511 
5512   src = ConvL2X(src);  // adjust Java long to machine word
5513   Node* base = _gvn.transform(new CastX2PNode(src));
5514   offset = ConvI2X(offset);
5515 
5516   // 'src_start' points to src array + scaled offset
5517   Node* src_start = basic_plus_adr(top(), base, offset);
5518 
5519   // Call the stub.
5520   address stubAddr = StubRoutines::updateBytesCRC32();
5521   const char *stubName = "updateBytesCRC32";
5522 
5523   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5524                                  stubAddr, stubName, TypePtr::BOTTOM,
5525                                  crc, src_start, length);
5526   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5527   set_result(result);
5528   return true;
5529 }
5530 
5531 //----------------------------inline_reference_get----------------------------
5532 // public T java.lang.ref.Reference.get();
5533 bool LibraryCallKit::inline_reference_get() {
5534   const int referent_offset = java_lang_ref_Reference::referent_offset;
5535   guarantee(referent_offset > 0, "should have already been set");
5536 
5537   // Get the argument:
5538   Node* reference_obj = null_check_receiver();
5539   if (stopped()) return true;
5540 
5541   Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5542 
5543   ciInstanceKlass* klass = env()->Object_klass();
5544   const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5545 
5546   Node* no_ctrl = NULL;
5547   Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5548 
5549   // Use the pre-barrier to record the value in the referent field
5550   pre_barrier(false /* do_load */,
5551               control(),
5552               NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5553               result /* pre_val */,
5554               T_OBJECT);
5555 
5556   // Add memory barrier to prevent commoning reads from this field
5557   // across safepoint since GC can change its value.
5558   insert_mem_bar(Op_MemBarCPUOrder);
5559 
5560   set_result(result);
5561   return true;
5562 }
5563 
5564 
5565 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5566                                               bool is_exact=true, bool is_static=false) {
5567 
5568   const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5569   assert(tinst != NULL, "obj is null");
5570   assert(tinst->klass()->is_loaded(), "obj is not loaded");
5571   assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5572 
5573   ciField* field = tinst->klass()->as_instance_klass()->get_field_by_name(ciSymbol::make(fieldName),
5574                                                                           ciSymbol::make(fieldTypeString),
5575                                                                           is_static);
5576   if (field == NULL) return (Node *) NULL;
5577   assert (field != NULL, "undefined field");
5578 
5579   // Next code  copied from Parse::do_get_xxx():
5580 
5581   // Compute address and memory type.
5582   int offset  = field->offset_in_bytes();
5583   bool is_vol = field->is_volatile();
5584   ciType* field_klass = field->type();
5585   assert(field_klass->is_loaded(), "should be loaded");
5586   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5587   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5588   BasicType bt = field->layout_type();
5589 
5590   // Build the resultant type of the load
5591   const Type *type;
5592   if (bt == T_OBJECT) {
5593     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5594   } else {
5595     type = Type::get_const_basic_type(bt);
5596   }
5597 
5598   if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5599     insert_mem_bar(Op_MemBarVolatile);   // StoreLoad barrier
5600   }
5601   // Build the load.
5602   MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5603   Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5604   // If reference is volatile, prevent following memory ops from
5605   // floating up past the volatile read.  Also prevents commoning
5606   // another volatile read.
5607   if (is_vol) {
5608     // Memory barrier includes bogus read of value to force load BEFORE membar
5609     insert_mem_bar(Op_MemBarAcquire, loadedField);
5610   }
5611   return loadedField;
5612 }
5613 
5614 
5615 //------------------------------inline_aescrypt_Block-----------------------
5616 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5617   address stubAddr;
5618   const char *stubName;
5619   assert(UseAES, "need AES instruction support");
5620 
5621   switch(id) {
5622   case vmIntrinsics::_aescrypt_encryptBlock:
5623     stubAddr = StubRoutines::aescrypt_encryptBlock();
5624     stubName = "aescrypt_encryptBlock";
5625     break;
5626   case vmIntrinsics::_aescrypt_decryptBlock:
5627     stubAddr = StubRoutines::aescrypt_decryptBlock();
5628     stubName = "aescrypt_decryptBlock";
5629     break;
5630   }
5631   if (stubAddr == NULL) return false;
5632 
5633   Node* aescrypt_object = argument(0);
5634   Node* src             = argument(1);
5635   Node* src_offset      = argument(2);
5636   Node* dest            = argument(3);
5637   Node* dest_offset     = argument(4);
5638 
5639   // (1) src and dest are arrays.
5640   const Type* src_type = src->Value(&_gvn);
5641   const Type* dest_type = dest->Value(&_gvn);
5642   const TypeAryPtr* top_src = src_type->isa_aryptr();
5643   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5644   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5645 
5646   // for the quick and dirty code we will skip all the checks.
5647   // we are just trying to get the call to be generated.
5648   Node* src_start  = src;
5649   Node* dest_start = dest;
5650   if (src_offset != NULL || dest_offset != NULL) {
5651     assert(src_offset != NULL && dest_offset != NULL, "");
5652     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5653     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5654   }
5655 
5656   // now need to get the start of its expanded key array
5657   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5658   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5659   if (k_start == NULL) return false;
5660 
5661   if (Matcher::pass_original_key_for_aes()) {
5662     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5663     // compatibility issues between Java key expansion and SPARC crypto instructions
5664     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5665     if (original_k_start == NULL) return false;
5666 
5667     // Call the stub.
5668     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5669                       stubAddr, stubName, TypePtr::BOTTOM,
5670                       src_start, dest_start, k_start, original_k_start);
5671   } else {
5672     // Call the stub.
5673     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5674                       stubAddr, stubName, TypePtr::BOTTOM,
5675                       src_start, dest_start, k_start);
5676   }
5677 
5678   return true;
5679 }
5680 
5681 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5682 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5683   address stubAddr;
5684   const char *stubName;
5685 
5686   assert(UseAES, "need AES instruction support");
5687 
5688   switch(id) {
5689   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5690     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5691     stubName = "cipherBlockChaining_encryptAESCrypt";
5692     break;
5693   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5694     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5695     stubName = "cipherBlockChaining_decryptAESCrypt";
5696     break;
5697   }
5698   if (stubAddr == NULL) return false;
5699 
5700   Node* cipherBlockChaining_object = argument(0);
5701   Node* src                        = argument(1);
5702   Node* src_offset                 = argument(2);
5703   Node* len                        = argument(3);
5704   Node* dest                       = argument(4);
5705   Node* dest_offset                = argument(5);
5706 
5707   // (1) src and dest are arrays.
5708   const Type* src_type = src->Value(&_gvn);
5709   const Type* dest_type = dest->Value(&_gvn);
5710   const TypeAryPtr* top_src = src_type->isa_aryptr();
5711   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5712   assert (top_src  != NULL && top_src->klass()  != NULL
5713           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5714 
5715   // checks are the responsibility of the caller
5716   Node* src_start  = src;
5717   Node* dest_start = dest;
5718   if (src_offset != NULL || dest_offset != NULL) {
5719     assert(src_offset != NULL && dest_offset != NULL, "");
5720     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5721     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5722   }
5723 
5724   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5725   // (because of the predicated logic executed earlier).
5726   // so we cast it here safely.
5727   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5728 
5729   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5730   if (embeddedCipherObj == NULL) return false;
5731 
5732   // cast it to what we know it will be at runtime
5733   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5734   assert(tinst != NULL, "CBC obj is null");
5735   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5736   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5737   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5738 
5739   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5740   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5741   const TypeOopPtr* xtype = aklass->as_instance_type();
5742   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5743   aescrypt_object = _gvn.transform(aescrypt_object);
5744 
5745   // we need to get the start of the aescrypt_object's expanded key array
5746   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5747   if (k_start == NULL) return false;
5748 
5749   // similarly, get the start address of the r vector
5750   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5751   if (objRvec == NULL) return false;
5752   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5753 
5754   Node* cbcCrypt;
5755   if (Matcher::pass_original_key_for_aes()) {
5756     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5757     // compatibility issues between Java key expansion and SPARC crypto instructions
5758     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5759     if (original_k_start == NULL) return false;
5760 
5761     // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5762     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5763                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5764                                  stubAddr, stubName, TypePtr::BOTTOM,
5765                                  src_start, dest_start, k_start, r_start, len, original_k_start);
5766   } else {
5767     // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5768     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5769                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5770                                  stubAddr, stubName, TypePtr::BOTTOM,
5771                                  src_start, dest_start, k_start, r_start, len);
5772   }
5773 
5774   // return cipher length (int)
5775   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5776   set_result(retvalue);
5777   return true;
5778 }
5779 
5780 //------------------------------get_key_start_from_aescrypt_object-----------------------
5781 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
5782   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
5783   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5784   if (objAESCryptKey == NULL) return (Node *) NULL;
5785 
5786   // now have the array, need to get the start address of the K array
5787   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
5788   return k_start;
5789 }
5790 
5791 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
5792 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
5793   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
5794   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5795   if (objAESCryptKey == NULL) return (Node *) NULL;
5796 
5797   // now have the array, need to get the start address of the lastKey array
5798   Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
5799   return original_k_start;
5800 }
5801 
5802 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
5803 // Return node representing slow path of predicate check.
5804 // the pseudo code we want to emulate with this predicate is:
5805 // for encryption:
5806 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
5807 // for decryption:
5808 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
5809 //    note cipher==plain is more conservative than the original java code but that's OK
5810 //
5811 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
5812   // The receiver was checked for NULL already.
5813   Node* objCBC = argument(0);
5814 
5815   // Load embeddedCipher field of CipherBlockChaining object.
5816   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5817 
5818   // get AESCrypt klass for instanceOf check
5819   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
5820   // will have same classloader as CipherBlockChaining object
5821   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
5822   assert(tinst != NULL, "CBCobj is null");
5823   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
5824 
5825   // we want to do an instanceof comparison against the AESCrypt class
5826   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5827   if (!klass_AESCrypt->is_loaded()) {
5828     // if AESCrypt is not even loaded, we never take the intrinsic fast path
5829     Node* ctrl = control();
5830     set_control(top()); // no regular fast path
5831     return ctrl;
5832   }
5833   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5834 
5835   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
5836   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
5837   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
5838 
5839   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
5840 
5841   // for encryption, we are done
5842   if (!decrypting)
5843     return instof_false;  // even if it is NULL
5844 
5845   // for decryption, we need to add a further check to avoid
5846   // taking the intrinsic path when cipher and plain are the same
5847   // see the original java code for why.
5848   RegionNode* region = new RegionNode(3);
5849   region->init_req(1, instof_false);
5850   Node* src = argument(1);
5851   Node* dest = argument(4);
5852   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
5853   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
5854   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
5855   region->init_req(2, src_dest_conjoint);
5856 
5857   record_for_igvn(region);
5858   return _gvn.transform(region);
5859 }
5860 
5861 //------------------------------inline_sha_implCompress-----------------------
5862 //
5863 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
5864 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
5865 //
5866 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
5867 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
5868 //
5869 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
5870 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
5871 //
5872 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
5873   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
5874 
5875   Node* sha_obj = argument(0);
5876   Node* src     = argument(1); // type oop
5877   Node* ofs     = argument(2); // type int
5878 
5879   const Type* src_type = src->Value(&_gvn);
5880   const TypeAryPtr* top_src = src_type->isa_aryptr();
5881   if (top_src  == NULL || top_src->klass()  == NULL) {
5882     // failed array check
5883     return false;
5884   }
5885   // Figure out the size and type of the elements we will be copying.
5886   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5887   if (src_elem != T_BYTE) {
5888     return false;
5889   }
5890   // 'src_start' points to src array + offset
5891   Node* src_start = array_element_address(src, ofs, src_elem);
5892   Node* state = NULL;
5893   address stubAddr;
5894   const char *stubName;
5895 
5896   switch(id) {
5897   case vmIntrinsics::_sha_implCompress:
5898     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
5899     state = get_state_from_sha_object(sha_obj);
5900     stubAddr = StubRoutines::sha1_implCompress();
5901     stubName = "sha1_implCompress";
5902     break;
5903   case vmIntrinsics::_sha2_implCompress:
5904     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
5905     state = get_state_from_sha_object(sha_obj);
5906     stubAddr = StubRoutines::sha256_implCompress();
5907     stubName = "sha256_implCompress";
5908     break;
5909   case vmIntrinsics::_sha5_implCompress:
5910     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
5911     state = get_state_from_sha5_object(sha_obj);
5912     stubAddr = StubRoutines::sha512_implCompress();
5913     stubName = "sha512_implCompress";
5914     break;
5915   default:
5916     fatal_unexpected_iid(id);
5917     return false;
5918   }
5919   if (state == NULL) return false;
5920 
5921   // Call the stub.
5922   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
5923                                  stubAddr, stubName, TypePtr::BOTTOM,
5924                                  src_start, state);
5925 
5926   return true;
5927 }
5928 
5929 //------------------------------inline_digestBase_implCompressMB-----------------------
5930 //
5931 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
5932 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
5933 //
5934 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
5935   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
5936          "need SHA1/SHA256/SHA512 instruction support");
5937   assert((uint)predicate < 3, "sanity");
5938   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
5939 
5940   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
5941   Node* src            = argument(1); // byte[] array
5942   Node* ofs            = argument(2); // type int
5943   Node* limit          = argument(3); // type int
5944 
5945   const Type* src_type = src->Value(&_gvn);
5946   const TypeAryPtr* top_src = src_type->isa_aryptr();
5947   if (top_src  == NULL || top_src->klass()  == NULL) {
5948     // failed array check
5949     return false;
5950   }
5951   // Figure out the size and type of the elements we will be copying.
5952   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5953   if (src_elem != T_BYTE) {
5954     return false;
5955   }
5956   // 'src_start' points to src array + offset
5957   Node* src_start = array_element_address(src, ofs, src_elem);
5958 
5959   const char* klass_SHA_name = NULL;
5960   const char* stub_name = NULL;
5961   address     stub_addr = NULL;
5962   bool        long_state = false;
5963 
5964   switch (predicate) {
5965   case 0:
5966     if (UseSHA1Intrinsics) {
5967       klass_SHA_name = "sun/security/provider/SHA";
5968       stub_name = "sha1_implCompressMB";
5969       stub_addr = StubRoutines::sha1_implCompressMB();
5970     }
5971     break;
5972   case 1:
5973     if (UseSHA256Intrinsics) {
5974       klass_SHA_name = "sun/security/provider/SHA2";
5975       stub_name = "sha256_implCompressMB";
5976       stub_addr = StubRoutines::sha256_implCompressMB();
5977     }
5978     break;
5979   case 2:
5980     if (UseSHA512Intrinsics) {
5981       klass_SHA_name = "sun/security/provider/SHA5";
5982       stub_name = "sha512_implCompressMB";
5983       stub_addr = StubRoutines::sha512_implCompressMB();
5984       long_state = true;
5985     }
5986     break;
5987   default:
5988     fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
5989   }
5990   if (klass_SHA_name != NULL) {
5991     // get DigestBase klass to lookup for SHA klass
5992     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
5993     assert(tinst != NULL, "digestBase_obj is not instance???");
5994     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
5995 
5996     ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
5997     assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
5998     ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
5999     return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6000   }
6001   return false;
6002 }
6003 //------------------------------inline_sha_implCompressMB-----------------------
6004 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6005                                                bool long_state, address stubAddr, const char *stubName,
6006                                                Node* src_start, Node* ofs, Node* limit) {
6007   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6008   const TypeOopPtr* xtype = aklass->as_instance_type();
6009   Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6010   sha_obj = _gvn.transform(sha_obj);
6011 
6012   Node* state;
6013   if (long_state) {
6014     state = get_state_from_sha5_object(sha_obj);
6015   } else {
6016     state = get_state_from_sha_object(sha_obj);
6017   }
6018   if (state == NULL) return false;
6019 
6020   // Call the stub.
6021   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6022                                  OptoRuntime::digestBase_implCompressMB_Type(),
6023                                  stubAddr, stubName, TypePtr::BOTTOM,
6024                                  src_start, state, ofs, limit);
6025   // return ofs (int)
6026   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6027   set_result(result);
6028 
6029   return true;
6030 }
6031 
6032 //------------------------------get_state_from_sha_object-----------------------
6033 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6034   Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6035   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6036   if (sha_state == NULL) return (Node *) NULL;
6037 
6038   // now have the array, need to get the start address of the state array
6039   Node* state = array_element_address(sha_state, intcon(0), T_INT);
6040   return state;
6041 }
6042 
6043 //------------------------------get_state_from_sha5_object-----------------------
6044 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6045   Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6046   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6047   if (sha_state == NULL) return (Node *) NULL;
6048 
6049   // now have the array, need to get the start address of the state array
6050   Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6051   return state;
6052 }
6053 
6054 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6055 // Return node representing slow path of predicate check.
6056 // the pseudo code we want to emulate with this predicate is:
6057 //    if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6058 //
6059 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6060   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6061          "need SHA1/SHA256/SHA512 instruction support");
6062   assert((uint)predicate < 3, "sanity");
6063 
6064   // The receiver was checked for NULL already.
6065   Node* digestBaseObj = argument(0);
6066 
6067   // get DigestBase klass for instanceOf check
6068   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6069   assert(tinst != NULL, "digestBaseObj is null");
6070   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6071 
6072   const char* klass_SHA_name = NULL;
6073   switch (predicate) {
6074   case 0:
6075     if (UseSHA1Intrinsics) {
6076       // we want to do an instanceof comparison against the SHA class
6077       klass_SHA_name = "sun/security/provider/SHA";
6078     }
6079     break;
6080   case 1:
6081     if (UseSHA256Intrinsics) {
6082       // we want to do an instanceof comparison against the SHA2 class
6083       klass_SHA_name = "sun/security/provider/SHA2";
6084     }
6085     break;
6086   case 2:
6087     if (UseSHA512Intrinsics) {
6088       // we want to do an instanceof comparison against the SHA5 class
6089       klass_SHA_name = "sun/security/provider/SHA5";
6090     }
6091     break;
6092   default:
6093     fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
6094   }
6095 
6096   ciKlass* klass_SHA = NULL;
6097   if (klass_SHA_name != NULL) {
6098     klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6099   }
6100   if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6101     // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6102     Node* ctrl = control();
6103     set_control(top()); // no intrinsic path
6104     return ctrl;
6105   }
6106   ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6107 
6108   Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6109   Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6110   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6111   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6112 
6113   return instof_false;  // even if it is NULL
6114 }
6115 
6116 bool LibraryCallKit::inline_profileBoolean() {
6117   Node* counts = argument(1);
6118   const TypeAryPtr* ary = NULL;
6119   ciArray* aobj = NULL;
6120   if (counts->is_Con()
6121       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6122       && (aobj = ary->const_oop()->as_array()) != NULL
6123       && (aobj->length() == 2)) {
6124     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6125     jint false_cnt = aobj->element_value(0).as_int();
6126     jint  true_cnt = aobj->element_value(1).as_int();
6127 
6128     method()->set_injected_profile(true);
6129 
6130     if (C->log() != NULL) {
6131       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6132                      false_cnt, true_cnt);
6133     }
6134 
6135     if (false_cnt + true_cnt == 0) {
6136       // According to profile, never executed.
6137       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6138                           Deoptimization::Action_reinterpret);
6139       return true;
6140     }
6141 
6142     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6143     // is a number of each value occurrences.
6144     Node* result = argument(0);
6145     if (false_cnt == 0 || true_cnt == 0) {
6146       // According to profile, one value has been never seen.
6147       int expected_val = (false_cnt == 0) ? 1 : 0;
6148 
6149       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6150       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6151 
6152       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6153       Node* fast_path = _gvn.transform(new IfTrueNode(check));
6154       Node* slow_path = _gvn.transform(new IfFalseNode(check));
6155 
6156       { // Slow path: uncommon trap for never seen value and then reexecute
6157         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6158         // the value has been seen at least once.
6159         PreserveJVMState pjvms(this);
6160         PreserveReexecuteState preexecs(this);
6161         jvms()->set_should_reexecute(true);
6162 
6163         set_control(slow_path);
6164         set_i_o(i_o());
6165 
6166         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6167                             Deoptimization::Action_reinterpret);
6168       }
6169       // The guard for never seen value enables sharpening of the result and
6170       // returning a constant. It allows to eliminate branches on the same value
6171       // later on.
6172       set_control(fast_path);
6173       result = intcon(expected_val);
6174     }
6175     // Stop profiling.
6176     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6177     // By replacing method body with profile data (represented as ProfileBooleanNode
6178     // on IR level) we effectively disable profiling.
6179     // It enables full speed execution once optimized code is generated.
6180     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6181     C->record_for_igvn(profile);
6182     set_result(profile);
6183     return true;
6184   } else {
6185     // Continue profiling.
6186     // Profile data isn't available at the moment. So, execute method's bytecode version.
6187     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6188     // is compiled and counters aren't available since corresponding MethodHandle
6189     // isn't a compile-time constant.
6190     return false;
6191   }
6192 }
6193 
6194 bool LibraryCallKit::inline_isCompileConstant() {
6195   Node* n = argument(0);
6196   set_result(n->is_Con() ? intcon(1) : intcon(0));
6197   return true;
6198 }