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     Node* p = NULL;
2650     // Try to constant fold a load from a constant field
2651     ciField* field = alias_type->field();
2652     if (heap_base_oop != top() &&
2653         field != NULL && field->is_constant() && field->layout_type() == type) {
2654       // final or stable field
2655       const Type* con_type = Type::make_constant(alias_type->field(), heap_base_oop);

2656       if (con_type != NULL) {






2657         p = makecon(con_type);
2658       }
2659     }
2660     if (p == NULL) {
2661       MemNode::MemOrd mo = is_volatile ? MemNode::acquire : MemNode::unordered;
2662       // To be valid, unsafe loads may depend on other conditions than
2663       // the one that guards them: pin the Load node
2664       p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, is_volatile);
2665       // load value
2666       switch (type) {
2667       case T_BOOLEAN:
2668       case T_CHAR:
2669       case T_BYTE:
2670       case T_SHORT:
2671       case T_INT:
2672       case T_LONG:
2673       case T_FLOAT:
2674       case T_DOUBLE:
2675         break;
2676       case T_OBJECT:
2677         if (need_read_barrier) {
2678           insert_pre_barrier(heap_base_oop, offset, p, !(is_volatile || need_mem_bar));
2679         }
2680         break;
2681       case T_ADDRESS:
2682         // Cast to an int type.
2683         p = _gvn.transform(new CastP2XNode(NULL, p));
2684         p = ConvX2UL(p);
2685         break;
2686       default:
2687         fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
2688         break;
2689       }
2690     }
2691     // The load node has the control of the preceding MemBarCPUOrder.  All
2692     // following nodes will have the control of the MemBarCPUOrder inserted at
2693     // the end of this method.  So, pushing the load onto the stack at a later
2694     // point is fine.
2695     set_result(p);
2696   } else {
2697     // place effect of store into memory
2698     switch (type) {
2699     case T_DOUBLE:
2700       val = dstore_rounding(val);
2701       break;
2702     case T_ADDRESS:
2703       // Repackage the long as a pointer.
2704       val = ConvL2X(val);
2705       val = _gvn.transform(new CastX2PNode(val));
2706       break;
2707     }
2708 











2709     MemNode::MemOrd mo = is_volatile ? MemNode::release : MemNode::unordered;
2710     if (type != T_OBJECT ) {
2711       (void) store_to_memory(control(), adr, val, type, adr_type, mo, is_volatile);
2712     } else {
2713       // Possibly an oop being stored to Java heap or native memory
2714       if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2715         // oop to Java heap.
2716         (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo);
2717       } else {
2718         // We can't tell at compile time if we are storing in the Java heap or outside
2719         // of it. So we need to emit code to conditionally do the proper type of
2720         // store.
2721 
2722         IdealKit ideal(this);
2723 #define __ ideal.
2724         // QQQ who knows what probability is here??
2725         __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2726           // Sync IdealKit and graphKit.
2727           sync_kit(ideal);
2728           Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo);
2729           // Update IdealKit memory.
2730           __ sync_kit(this);
2731         } __ else_(); {
2732           __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, is_volatile);
2733         } __ end_if();
2734         // Final sync IdealKit and GraphKit.
2735         final_sync(ideal);
2736 #undef __
2737       }
2738     }
2739   }
2740 
2741   if (is_volatile) {
2742     if (!is_store) {
2743       insert_mem_bar(Op_MemBarAcquire);
2744     } else {
2745       if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2746         insert_mem_bar(Op_MemBarVolatile);
2747       }
2748     }
2749   }
2750 
2751   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2752 
2753   return true;
2754 }
2755 
2756 //----------------------------inline_unsafe_load_store----------------------------
2757 // This method serves a couple of different customers (depending on LoadStoreKind):
2758 //
2759 // LS_cmpxchg:
2760 //   public final native boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2761 //   public final native boolean compareAndSwapInt(   Object o, long offset, int    expected, int    x);
2762 //   public final native boolean compareAndSwapLong(  Object o, long offset, long   expected, long   x);
2763 //
2764 // LS_xadd:
2765 //   public int  getAndAddInt( Object o, long offset, int  delta)
2766 //   public long getAndAddLong(Object o, long offset, long delta)
2767 //
2768 // LS_xchg:
2769 //   int    getAndSet(Object o, long offset, int    newValue)
2770 //   long   getAndSet(Object o, long offset, long   newValue)
2771 //   Object getAndSet(Object o, long offset, Object newValue)
2772 //
2773 bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind) {
2774   // This basic scheme here is the same as inline_unsafe_access, but
2775   // differs in enough details that combining them would make the code
2776   // overly confusing.  (This is a true fact! I originally combined
2777   // them, but even I was confused by it!) As much code/comments as
2778   // possible are retained from inline_unsafe_access though to make
2779   // the correspondences clearer. - dl
2780 
2781   if (callee()->is_static())  return false;  // caller must have the capability!
2782 
2783 #ifndef PRODUCT
2784   BasicType rtype;
2785   {
2786     ResourceMark rm;
2787     // Check the signatures.
2788     ciSignature* sig = callee()->signature();
2789     rtype = sig->return_type()->basic_type();
2790     if (kind == LS_xadd || kind == LS_xchg) {
2791       // Check the signatures.
2792 #ifdef ASSERT
2793       assert(rtype == type, "get and set must return the expected type");
2794       assert(sig->count() == 3, "get and set has 3 arguments");
2795       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2796       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2797       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2798 #endif // ASSERT
2799     } else if (kind == LS_cmpxchg) {
2800       // Check the signatures.
2801 #ifdef ASSERT
2802       assert(rtype == T_BOOLEAN, "CAS must return boolean");
2803       assert(sig->count() == 4, "CAS has 4 arguments");
2804       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2805       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2806 #endif // ASSERT
2807     } else {
2808       ShouldNotReachHere();
2809     }
2810   }
2811 #endif //PRODUCT
2812 
2813   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2814 
2815   // Get arguments:
2816   Node* receiver = NULL;
2817   Node* base     = NULL;
2818   Node* offset   = NULL;
2819   Node* oldval   = NULL;
2820   Node* newval   = NULL;
2821   if (kind == LS_cmpxchg) {
2822     const bool two_slot_type = type2size[type] == 2;
2823     receiver = argument(0);  // type: oop
2824     base     = argument(1);  // type: oop
2825     offset   = argument(2);  // type: long
2826     oldval   = argument(4);  // type: oop, int, or long
2827     newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2828   } else if (kind == LS_xadd || kind == LS_xchg){
2829     receiver = argument(0);  // type: oop
2830     base     = argument(1);  // type: oop
2831     offset   = argument(2);  // type: long
2832     oldval   = NULL;
2833     newval   = argument(4);  // type: oop, int, or long
2834   }
2835 
2836   // Null check receiver.
2837   receiver = null_check(receiver);
2838   if (stopped()) {
2839     return true;
2840   }
2841 
2842   // Build field offset expression.
2843   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2844   // to be plain byte offsets, which are also the same as those accepted
2845   // by oopDesc::field_base.
2846   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2847   // 32-bit machines ignore the high half of long offsets
2848   offset = ConvL2X(offset);
2849   Node* adr = make_unsafe_address(base, offset);
2850   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2851 
2852   // For CAS, unlike inline_unsafe_access, there seems no point in
2853   // trying to refine types. Just use the coarse types here.
2854   const Type *value_type = Type::get_const_basic_type(type);
2855   Compile::AliasType* alias_type = C->alias_type(adr_type);
2856   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2857 
2858   if (kind == LS_xchg && type == T_OBJECT) {
2859     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2860     if (tjp != NULL) {
2861       value_type = tjp;
2862     }
2863   }
2864 
2865   int alias_idx = C->get_alias_index(adr_type);
2866 
2867   // Memory-model-wise, a LoadStore acts like a little synchronized
2868   // block, so needs barriers on each side.  These don't translate
2869   // into actual barriers on most machines, but we still need rest of
2870   // compiler to respect ordering.
2871 
2872   insert_mem_bar(Op_MemBarRelease);
2873   insert_mem_bar(Op_MemBarCPUOrder);
2874 
2875   // 4984716: MemBars must be inserted before this
2876   //          memory node in order to avoid a false
2877   //          dependency which will confuse the scheduler.
2878   Node *mem = memory(alias_idx);
2879 
2880   // For now, we handle only those cases that actually exist: ints,
2881   // longs, and Object. Adding others should be straightforward.
2882   Node* load_store;
2883   switch(type) {
2884   case T_INT:
2885     if (kind == LS_xadd) {
2886       load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2887     } else if (kind == LS_xchg) {
2888       load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2889     } else if (kind == LS_cmpxchg) {
2890       load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval));
2891     } else {
2892       ShouldNotReachHere();
2893     }
2894     break;
2895   case T_LONG:
2896     if (kind == LS_xadd) {
2897       load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2898     } else if (kind == LS_xchg) {
2899       load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2900     } else if (kind == LS_cmpxchg) {
2901       load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval));
2902     } else {
2903       ShouldNotReachHere();
2904     }
2905     break;
2906   case T_OBJECT:
2907     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2908     // could be delayed during Parse (for example, in adjust_map_after_if()).
2909     // Execute transformation here to avoid barrier generation in such case.
2910     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2911       newval = _gvn.makecon(TypePtr::NULL_PTR);
2912 
2913     // Reference stores need a store barrier.
2914     if (kind == LS_xchg) {
2915       // If pre-barrier must execute before the oop store, old value will require do_load here.
2916       if (!can_move_pre_barrier()) {
2917         pre_barrier(true /* do_load*/,
2918                     control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2919                     NULL /* pre_val*/,
2920                     T_OBJECT);
2921       } // Else move pre_barrier to use load_store value, see below.
2922     } else if (kind == LS_cmpxchg) {
2923       // Same as for newval above:
2924       if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2925         oldval = _gvn.makecon(TypePtr::NULL_PTR);
2926       }
2927       // The only known value which might get overwritten is oldval.
2928       pre_barrier(false /* do_load */,
2929                   control(), NULL, NULL, max_juint, NULL, NULL,
2930                   oldval /* pre_val */,
2931                   T_OBJECT);
2932     } else {
2933       ShouldNotReachHere();
2934     }
2935 
2936 #ifdef _LP64
2937     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2938       Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2939       if (kind == LS_xchg) {
2940         load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr,
2941                                                        newval_enc, adr_type, value_type->make_narrowoop()));
2942       } else {
2943         assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2944         Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2945         load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr,
2946                                                                 newval_enc, oldval_enc));
2947       }
2948     } else
2949 #endif
2950     {
2951       if (kind == LS_xchg) {
2952         load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2953       } else {
2954         assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2955         load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval));
2956       }
2957     }
2958     post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2959     break;
2960   default:
2961     fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
2962     break;
2963   }
2964 
2965   // SCMemProjNodes represent the memory state of a LoadStore. Their
2966   // main role is to prevent LoadStore nodes from being optimized away
2967   // when their results aren't used.
2968   Node* proj = _gvn.transform(new SCMemProjNode(load_store));
2969   set_memory(proj, alias_idx);
2970 
2971   if (type == T_OBJECT && kind == LS_xchg) {
2972 #ifdef _LP64
2973     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2974       load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
2975     }
2976 #endif
2977     if (can_move_pre_barrier()) {
2978       // Don't need to load pre_val. The old value is returned by load_store.
2979       // The pre_barrier can execute after the xchg as long as no safepoint
2980       // gets inserted between them.
2981       pre_barrier(false /* do_load */,
2982                   control(), NULL, NULL, max_juint, NULL, NULL,
2983                   load_store /* pre_val */,
2984                   T_OBJECT);
2985     }
2986   }
2987 
2988   // Add the trailing membar surrounding the access
2989   insert_mem_bar(Op_MemBarCPUOrder);
2990   insert_mem_bar(Op_MemBarAcquire);
2991 
2992   assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2993   set_result(load_store);
2994   return true;
2995 }
2996 
2997 //----------------------------inline_unsafe_ordered_store----------------------
2998 // public native void sun.misc.Unsafe.putOrderedObject(Object o, long offset, Object x);
2999 // public native void sun.misc.Unsafe.putOrderedInt(Object o, long offset, int x);
3000 // public native void sun.misc.Unsafe.putOrderedLong(Object o, long offset, long x);
3001 bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
3002   // This is another variant of inline_unsafe_access, differing in
3003   // that it always issues store-store ("release") barrier and ensures
3004   // store-atomicity (which only matters for "long").
3005 
3006   if (callee()->is_static())  return false;  // caller must have the capability!
3007 
3008 #ifndef PRODUCT
3009   {
3010     ResourceMark rm;
3011     // Check the signatures.
3012     ciSignature* sig = callee()->signature();
3013 #ifdef ASSERT
3014     BasicType rtype = sig->return_type()->basic_type();
3015     assert(rtype == T_VOID, "must return void");
3016     assert(sig->count() == 3, "has 3 arguments");
3017     assert(sig->type_at(0)->basic_type() == T_OBJECT, "base is object");
3018     assert(sig->type_at(1)->basic_type() == T_LONG, "offset is long");
3019 #endif // ASSERT
3020   }
3021 #endif //PRODUCT
3022 
3023   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
3024 
3025   // Get arguments:
3026   Node* receiver = argument(0);  // type: oop
3027   Node* base     = argument(1);  // type: oop
3028   Node* offset   = argument(2);  // type: long
3029   Node* val      = argument(4);  // type: oop, int, or long
3030 
3031   // Null check receiver.
3032   receiver = null_check(receiver);
3033   if (stopped()) {
3034     return true;
3035   }
3036 
3037   // Build field offset expression.
3038   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3039   // 32-bit machines ignore the high half of long offsets
3040   offset = ConvL2X(offset);
3041   Node* adr = make_unsafe_address(base, offset);
3042   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3043   const Type *value_type = Type::get_const_basic_type(type);
3044   Compile::AliasType* alias_type = C->alias_type(adr_type);
3045 
3046   insert_mem_bar(Op_MemBarRelease);
3047   insert_mem_bar(Op_MemBarCPUOrder);
3048   // Ensure that the store is atomic for longs:
3049   const bool require_atomic_access = true;
3050   Node* store;
3051   if (type == T_OBJECT) // reference stores need a store barrier.
3052     store = store_oop_to_unknown(control(), base, adr, adr_type, val, type, MemNode::release);
3053   else {
3054     store = store_to_memory(control(), adr, val, type, adr_type, MemNode::release, require_atomic_access);
3055   }
3056   insert_mem_bar(Op_MemBarCPUOrder);
3057   return true;
3058 }
3059 
3060 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3061   // Regardless of form, don't allow previous ld/st to move down,
3062   // then issue acquire, release, or volatile mem_bar.
3063   insert_mem_bar(Op_MemBarCPUOrder);
3064   switch(id) {
3065     case vmIntrinsics::_loadFence:
3066       insert_mem_bar(Op_LoadFence);
3067       return true;
3068     case vmIntrinsics::_storeFence:
3069       insert_mem_bar(Op_StoreFence);
3070       return true;
3071     case vmIntrinsics::_fullFence:
3072       insert_mem_bar(Op_MemBarVolatile);
3073       return true;
3074     default:
3075       fatal_unexpected_iid(id);
3076       return false;
3077   }
3078 }
3079 
3080 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3081   if (!kls->is_Con()) {
3082     return true;
3083   }
3084   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3085   if (klsptr == NULL) {
3086     return true;
3087   }
3088   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3089   // don't need a guard for a klass that is already initialized
3090   return !ik->is_initialized();
3091 }
3092 
3093 //----------------------------inline_unsafe_allocate---------------------------
3094 // public native Object sun.misc.Unsafe.allocateInstance(Class<?> cls);
3095 bool LibraryCallKit::inline_unsafe_allocate() {
3096   if (callee()->is_static())  return false;  // caller must have the capability!
3097 
3098   null_check_receiver();  // null-check, then ignore
3099   Node* cls = null_check(argument(1));
3100   if (stopped())  return true;
3101 
3102   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3103   kls = null_check(kls);
3104   if (stopped())  return true;  // argument was like int.class
3105 
3106   Node* test = NULL;
3107   if (LibraryCallKit::klass_needs_init_guard(kls)) {
3108     // Note:  The argument might still be an illegal value like
3109     // Serializable.class or Object[].class.   The runtime will handle it.
3110     // But we must make an explicit check for initialization.
3111     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3112     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3113     // can generate code to load it as unsigned byte.
3114     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3115     Node* bits = intcon(InstanceKlass::fully_initialized);
3116     test = _gvn.transform(new SubINode(inst, bits));
3117     // The 'test' is non-zero if we need to take a slow path.
3118   }
3119 
3120   Node* obj = new_instance(kls, test);
3121   set_result(obj);
3122   return true;
3123 }
3124 
3125 #ifdef TRACE_HAVE_INTRINSICS
3126 /*
3127  * oop -> myklass
3128  * myklass->trace_id |= USED
3129  * return myklass->trace_id & ~0x3
3130  */
3131 bool LibraryCallKit::inline_native_classID() {
3132   null_check_receiver();  // null-check, then ignore
3133   Node* cls = null_check(argument(1), T_OBJECT);
3134   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3135   kls = null_check(kls, T_OBJECT);
3136   ByteSize offset = TRACE_ID_OFFSET;
3137   Node* insp = basic_plus_adr(kls, in_bytes(offset));
3138   Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3139   Node* bits = longcon(~0x03l); // ignore bit 0 & 1
3140   Node* andl = _gvn.transform(new AndLNode(tvalue, bits));
3141   Node* clsused = longcon(0x01l); // set the class bit
3142   Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3143 
3144   const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3145   store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3146   set_result(andl);
3147   return true;
3148 }
3149 
3150 bool LibraryCallKit::inline_native_threadID() {
3151   Node* tls_ptr = NULL;
3152   Node* cur_thr = generate_current_thread(tls_ptr);
3153   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3154   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3155   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::thread_id_offset()));
3156 
3157   Node* threadid = NULL;
3158   size_t thread_id_size = OSThread::thread_id_size();
3159   if (thread_id_size == (size_t) BytesPerLong) {
3160     threadid = ConvL2I(make_load(control(), p, TypeLong::LONG, T_LONG, MemNode::unordered));
3161   } else if (thread_id_size == (size_t) BytesPerInt) {
3162     threadid = make_load(control(), p, TypeInt::INT, T_INT, MemNode::unordered);
3163   } else {
3164     ShouldNotReachHere();
3165   }
3166   set_result(threadid);
3167   return true;
3168 }
3169 #endif
3170 
3171 //------------------------inline_native_time_funcs--------------
3172 // inline code for System.currentTimeMillis() and System.nanoTime()
3173 // these have the same type and signature
3174 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3175   const TypeFunc* tf = OptoRuntime::void_long_Type();
3176   const TypePtr* no_memory_effects = NULL;
3177   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3178   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3179 #ifdef ASSERT
3180   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3181   assert(value_top == top(), "second value must be top");
3182 #endif
3183   set_result(value);
3184   return true;
3185 }
3186 
3187 //------------------------inline_native_currentThread------------------
3188 bool LibraryCallKit::inline_native_currentThread() {
3189   Node* junk = NULL;
3190   set_result(generate_current_thread(junk));
3191   return true;
3192 }
3193 
3194 //------------------------inline_native_isInterrupted------------------
3195 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3196 bool LibraryCallKit::inline_native_isInterrupted() {
3197   // Add a fast path to t.isInterrupted(clear_int):
3198   //   (t == Thread.current() &&
3199   //    (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3200   //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3201   // So, in the common case that the interrupt bit is false,
3202   // we avoid making a call into the VM.  Even if the interrupt bit
3203   // is true, if the clear_int argument is false, we avoid the VM call.
3204   // However, if the receiver is not currentThread, we must call the VM,
3205   // because there must be some locking done around the operation.
3206 
3207   // We only go to the fast case code if we pass two guards.
3208   // Paths which do not pass are accumulated in the slow_region.
3209 
3210   enum {
3211     no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
3212     no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
3213     slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
3214     PATH_LIMIT
3215   };
3216 
3217   // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3218   // out of the function.
3219   insert_mem_bar(Op_MemBarCPUOrder);
3220 
3221   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3222   PhiNode*    result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3223 
3224   RegionNode* slow_region = new RegionNode(1);
3225   record_for_igvn(slow_region);
3226 
3227   // (a) Receiving thread must be the current thread.
3228   Node* rec_thr = argument(0);
3229   Node* tls_ptr = NULL;
3230   Node* cur_thr = generate_current_thread(tls_ptr);
3231   Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3232   Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3233 
3234   generate_slow_guard(bol_thr, slow_region);
3235 
3236   // (b) Interrupt bit on TLS must be false.
3237   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3238   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3239   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3240 
3241   // Set the control input on the field _interrupted read to prevent it floating up.
3242   Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3243   Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3244   Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3245 
3246   IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3247 
3248   // First fast path:  if (!TLS._interrupted) return false;
3249   Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3250   result_rgn->init_req(no_int_result_path, false_bit);
3251   result_val->init_req(no_int_result_path, intcon(0));
3252 
3253   // drop through to next case
3254   set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3255 
3256 #ifndef TARGET_OS_FAMILY_windows
3257   // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3258   Node* clr_arg = argument(1);
3259   Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3260   Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3261   IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3262 
3263   // Second fast path:  ... else if (!clear_int) return true;
3264   Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3265   result_rgn->init_req(no_clear_result_path, false_arg);
3266   result_val->init_req(no_clear_result_path, intcon(1));
3267 
3268   // drop through to next case
3269   set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3270 #else
3271   // To return true on Windows you must read the _interrupted field
3272   // and check the the event state i.e. take the slow path.
3273 #endif // TARGET_OS_FAMILY_windows
3274 
3275   // (d) Otherwise, go to the slow path.
3276   slow_region->add_req(control());
3277   set_control( _gvn.transform(slow_region));
3278 
3279   if (stopped()) {
3280     // There is no slow path.
3281     result_rgn->init_req(slow_result_path, top());
3282     result_val->init_req(slow_result_path, top());
3283   } else {
3284     // non-virtual because it is a private non-static
3285     CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3286 
3287     Node* slow_val = set_results_for_java_call(slow_call);
3288     // this->control() comes from set_results_for_java_call
3289 
3290     Node* fast_io  = slow_call->in(TypeFunc::I_O);
3291     Node* fast_mem = slow_call->in(TypeFunc::Memory);
3292 
3293     // These two phis are pre-filled with copies of of the fast IO and Memory
3294     PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3295     PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
3296 
3297     result_rgn->init_req(slow_result_path, control());
3298     result_io ->init_req(slow_result_path, i_o());
3299     result_mem->init_req(slow_result_path, reset_memory());
3300     result_val->init_req(slow_result_path, slow_val);
3301 
3302     set_all_memory(_gvn.transform(result_mem));
3303     set_i_o(       _gvn.transform(result_io));
3304   }
3305 
3306   C->set_has_split_ifs(true); // Has chance for split-if optimization
3307   set_result(result_rgn, result_val);
3308   return true;
3309 }
3310 
3311 //---------------------------load_mirror_from_klass----------------------------
3312 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3313 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3314   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3315   return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3316 }
3317 
3318 //-----------------------load_klass_from_mirror_common-------------------------
3319 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3320 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3321 // and branch to the given path on the region.
3322 // If never_see_null, take an uncommon trap on null, so we can optimistically
3323 // compile for the non-null case.
3324 // If the region is NULL, force never_see_null = true.
3325 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3326                                                     bool never_see_null,
3327                                                     RegionNode* region,
3328                                                     int null_path,
3329                                                     int offset) {
3330   if (region == NULL)  never_see_null = true;
3331   Node* p = basic_plus_adr(mirror, offset);
3332   const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3333   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3334   Node* null_ctl = top();
3335   kls = null_check_oop(kls, &null_ctl, never_see_null);
3336   if (region != NULL) {
3337     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3338     region->init_req(null_path, null_ctl);
3339   } else {
3340     assert(null_ctl == top(), "no loose ends");
3341   }
3342   return kls;
3343 }
3344 
3345 //--------------------(inline_native_Class_query helpers)---------------------
3346 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE, JVM_ACC_HAS_FINALIZER.
3347 // Fall through if (mods & mask) == bits, take the guard otherwise.
3348 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3349   // Branch around if the given klass has the given modifier bit set.
3350   // Like generate_guard, adds a new path onto the region.
3351   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3352   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3353   Node* mask = intcon(modifier_mask);
3354   Node* bits = intcon(modifier_bits);
3355   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3356   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3357   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3358   return generate_fair_guard(bol, region);
3359 }
3360 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3361   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3362 }
3363 
3364 //-------------------------inline_native_Class_query-------------------
3365 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3366   const Type* return_type = TypeInt::BOOL;
3367   Node* prim_return_value = top();  // what happens if it's a primitive class?
3368   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3369   bool expect_prim = false;     // most of these guys expect to work on refs
3370 
3371   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3372 
3373   Node* mirror = argument(0);
3374   Node* obj    = top();
3375 
3376   switch (id) {
3377   case vmIntrinsics::_isInstance:
3378     // nothing is an instance of a primitive type
3379     prim_return_value = intcon(0);
3380     obj = argument(1);
3381     break;
3382   case vmIntrinsics::_getModifiers:
3383     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3384     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3385     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3386     break;
3387   case vmIntrinsics::_isInterface:
3388     prim_return_value = intcon(0);
3389     break;
3390   case vmIntrinsics::_isArray:
3391     prim_return_value = intcon(0);
3392     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3393     break;
3394   case vmIntrinsics::_isPrimitive:
3395     prim_return_value = intcon(1);
3396     expect_prim = true;  // obviously
3397     break;
3398   case vmIntrinsics::_getSuperclass:
3399     prim_return_value = null();
3400     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3401     break;
3402   case vmIntrinsics::_getClassAccessFlags:
3403     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3404     return_type = TypeInt::INT;  // not bool!  6297094
3405     break;
3406   default:
3407     fatal_unexpected_iid(id);
3408     break;
3409   }
3410 
3411   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3412   if (mirror_con == NULL)  return false;  // cannot happen?
3413 
3414 #ifndef PRODUCT
3415   if (C->print_intrinsics() || C->print_inlining()) {
3416     ciType* k = mirror_con->java_mirror_type();
3417     if (k) {
3418       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3419       k->print_name();
3420       tty->cr();
3421     }
3422   }
3423 #endif
3424 
3425   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3426   RegionNode* region = new RegionNode(PATH_LIMIT);
3427   record_for_igvn(region);
3428   PhiNode* phi = new PhiNode(region, return_type);
3429 
3430   // The mirror will never be null of Reflection.getClassAccessFlags, however
3431   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3432   // if it is. See bug 4774291.
3433 
3434   // For Reflection.getClassAccessFlags(), the null check occurs in
3435   // the wrong place; see inline_unsafe_access(), above, for a similar
3436   // situation.
3437   mirror = null_check(mirror);
3438   // If mirror or obj is dead, only null-path is taken.
3439   if (stopped())  return true;
3440 
3441   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3442 
3443   // Now load the mirror's klass metaobject, and null-check it.
3444   // Side-effects region with the control path if the klass is null.
3445   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3446   // If kls is null, we have a primitive mirror.
3447   phi->init_req(_prim_path, prim_return_value);
3448   if (stopped()) { set_result(region, phi); return true; }
3449   bool safe_for_replace = (region->in(_prim_path) == top());
3450 
3451   Node* p;  // handy temp
3452   Node* null_ctl;
3453 
3454   // Now that we have the non-null klass, we can perform the real query.
3455   // For constant classes, the query will constant-fold in LoadNode::Value.
3456   Node* query_value = top();
3457   switch (id) {
3458   case vmIntrinsics::_isInstance:
3459     // nothing is an instance of a primitive type
3460     query_value = gen_instanceof(obj, kls, safe_for_replace);
3461     break;
3462 
3463   case vmIntrinsics::_getModifiers:
3464     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3465     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3466     break;
3467 
3468   case vmIntrinsics::_isInterface:
3469     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3470     if (generate_interface_guard(kls, region) != NULL)
3471       // A guard was added.  If the guard is taken, it was an interface.
3472       phi->add_req(intcon(1));
3473     // If we fall through, it's a plain class.
3474     query_value = intcon(0);
3475     break;
3476 
3477   case vmIntrinsics::_isArray:
3478     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3479     if (generate_array_guard(kls, region) != NULL)
3480       // A guard was added.  If the guard is taken, it was an array.
3481       phi->add_req(intcon(1));
3482     // If we fall through, it's a plain class.
3483     query_value = intcon(0);
3484     break;
3485 
3486   case vmIntrinsics::_isPrimitive:
3487     query_value = intcon(0); // "normal" path produces false
3488     break;
3489 
3490   case vmIntrinsics::_getSuperclass:
3491     // The rules here are somewhat unfortunate, but we can still do better
3492     // with random logic than with a JNI call.
3493     // Interfaces store null or Object as _super, but must report null.
3494     // Arrays store an intermediate super as _super, but must report Object.
3495     // Other types can report the actual _super.
3496     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3497     if (generate_interface_guard(kls, region) != NULL)
3498       // A guard was added.  If the guard is taken, it was an interface.
3499       phi->add_req(null());
3500     if (generate_array_guard(kls, region) != NULL)
3501       // A guard was added.  If the guard is taken, it was an array.
3502       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3503     // If we fall through, it's a plain class.  Get its _super.
3504     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3505     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3506     null_ctl = top();
3507     kls = null_check_oop(kls, &null_ctl);
3508     if (null_ctl != top()) {
3509       // If the guard is taken, Object.superClass is null (both klass and mirror).
3510       region->add_req(null_ctl);
3511       phi   ->add_req(null());
3512     }
3513     if (!stopped()) {
3514       query_value = load_mirror_from_klass(kls);
3515     }
3516     break;
3517 
3518   case vmIntrinsics::_getClassAccessFlags:
3519     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3520     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3521     break;
3522 
3523   default:
3524     fatal_unexpected_iid(id);
3525     break;
3526   }
3527 
3528   // Fall-through is the normal case of a query to a real class.
3529   phi->init_req(1, query_value);
3530   region->init_req(1, control());
3531 
3532   C->set_has_split_ifs(true); // Has chance for split-if optimization
3533   set_result(region, phi);
3534   return true;
3535 }
3536 
3537 //-------------------------inline_Class_cast-------------------
3538 bool LibraryCallKit::inline_Class_cast() {
3539   Node* mirror = argument(0); // Class
3540   Node* obj    = argument(1);
3541   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3542   if (mirror_con == NULL) {
3543     return false;  // dead path (mirror->is_top()).
3544   }
3545   if (obj == NULL || obj->is_top()) {
3546     return false;  // dead path
3547   }
3548   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3549 
3550   // First, see if Class.cast() can be folded statically.
3551   // java_mirror_type() returns non-null for compile-time Class constants.
3552   ciType* tm = mirror_con->java_mirror_type();
3553   if (tm != NULL && tm->is_klass() &&
3554       tp != NULL && tp->klass() != NULL) {
3555     if (!tp->klass()->is_loaded()) {
3556       // Don't use intrinsic when class is not loaded.
3557       return false;
3558     } else {
3559       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3560       if (static_res == Compile::SSC_always_true) {
3561         // isInstance() is true - fold the code.
3562         set_result(obj);
3563         return true;
3564       } else if (static_res == Compile::SSC_always_false) {
3565         // Don't use intrinsic, have to throw ClassCastException.
3566         // If the reference is null, the non-intrinsic bytecode will
3567         // be optimized appropriately.
3568         return false;
3569       }
3570     }
3571   }
3572 
3573   // Bailout intrinsic and do normal inlining if exception path is frequent.
3574   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3575     return false;
3576   }
3577 
3578   // Generate dynamic checks.
3579   // Class.cast() is java implementation of _checkcast bytecode.
3580   // Do checkcast (Parse::do_checkcast()) optimizations here.
3581 
3582   mirror = null_check(mirror);
3583   // If mirror is dead, only null-path is taken.
3584   if (stopped()) {
3585     return true;
3586   }
3587 
3588   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3589   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3590   RegionNode* region = new RegionNode(PATH_LIMIT);
3591   record_for_igvn(region);
3592 
3593   // Now load the mirror's klass metaobject, and null-check it.
3594   // If kls is null, we have a primitive mirror and
3595   // nothing is an instance of a primitive type.
3596   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3597 
3598   Node* res = top();
3599   if (!stopped()) {
3600     Node* bad_type_ctrl = top();
3601     // Do checkcast optimizations.
3602     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3603     region->init_req(_bad_type_path, bad_type_ctrl);
3604   }
3605   if (region->in(_prim_path) != top() ||
3606       region->in(_bad_type_path) != top()) {
3607     // Let Interpreter throw ClassCastException.
3608     PreserveJVMState pjvms(this);
3609     set_control(_gvn.transform(region));
3610     uncommon_trap(Deoptimization::Reason_intrinsic,
3611                   Deoptimization::Action_maybe_recompile);
3612   }
3613   if (!stopped()) {
3614     set_result(res);
3615   }
3616   return true;
3617 }
3618 
3619 
3620 //--------------------------inline_native_subtype_check------------------------
3621 // This intrinsic takes the JNI calls out of the heart of
3622 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3623 bool LibraryCallKit::inline_native_subtype_check() {
3624   // Pull both arguments off the stack.
3625   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3626   args[0] = argument(0);
3627   args[1] = argument(1);
3628   Node* klasses[2];             // corresponding Klasses: superk, subk
3629   klasses[0] = klasses[1] = top();
3630 
3631   enum {
3632     // A full decision tree on {superc is prim, subc is prim}:
3633     _prim_0_path = 1,           // {P,N} => false
3634                                 // {P,P} & superc!=subc => false
3635     _prim_same_path,            // {P,P} & superc==subc => true
3636     _prim_1_path,               // {N,P} => false
3637     _ref_subtype_path,          // {N,N} & subtype check wins => true
3638     _both_ref_path,             // {N,N} & subtype check loses => false
3639     PATH_LIMIT
3640   };
3641 
3642   RegionNode* region = new RegionNode(PATH_LIMIT);
3643   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3644   record_for_igvn(region);
3645 
3646   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3647   const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3648   int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3649 
3650   // First null-check both mirrors and load each mirror's klass metaobject.
3651   int which_arg;
3652   for (which_arg = 0; which_arg <= 1; which_arg++) {
3653     Node* arg = args[which_arg];
3654     arg = null_check(arg);
3655     if (stopped())  break;
3656     args[which_arg] = arg;
3657 
3658     Node* p = basic_plus_adr(arg, class_klass_offset);
3659     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3660     klasses[which_arg] = _gvn.transform(kls);
3661   }
3662 
3663   // Having loaded both klasses, test each for null.
3664   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3665   for (which_arg = 0; which_arg <= 1; which_arg++) {
3666     Node* kls = klasses[which_arg];
3667     Node* null_ctl = top();
3668     kls = null_check_oop(kls, &null_ctl, never_see_null);
3669     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3670     region->init_req(prim_path, null_ctl);
3671     if (stopped())  break;
3672     klasses[which_arg] = kls;
3673   }
3674 
3675   if (!stopped()) {
3676     // now we have two reference types, in klasses[0..1]
3677     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3678     Node* superk = klasses[0];  // the receiver
3679     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3680     // now we have a successful reference subtype check
3681     region->set_req(_ref_subtype_path, control());
3682   }
3683 
3684   // If both operands are primitive (both klasses null), then
3685   // we must return true when they are identical primitives.
3686   // It is convenient to test this after the first null klass check.
3687   set_control(region->in(_prim_0_path)); // go back to first null check
3688   if (!stopped()) {
3689     // Since superc is primitive, make a guard for the superc==subc case.
3690     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3691     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3692     generate_guard(bol_eq, region, PROB_FAIR);
3693     if (region->req() == PATH_LIMIT+1) {
3694       // A guard was added.  If the added guard is taken, superc==subc.
3695       region->swap_edges(PATH_LIMIT, _prim_same_path);
3696       region->del_req(PATH_LIMIT);
3697     }
3698     region->set_req(_prim_0_path, control()); // Not equal after all.
3699   }
3700 
3701   // these are the only paths that produce 'true':
3702   phi->set_req(_prim_same_path,   intcon(1));
3703   phi->set_req(_ref_subtype_path, intcon(1));
3704 
3705   // pull together the cases:
3706   assert(region->req() == PATH_LIMIT, "sane region");
3707   for (uint i = 1; i < region->req(); i++) {
3708     Node* ctl = region->in(i);
3709     if (ctl == NULL || ctl == top()) {
3710       region->set_req(i, top());
3711       phi   ->set_req(i, top());
3712     } else if (phi->in(i) == NULL) {
3713       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3714     }
3715   }
3716 
3717   set_control(_gvn.transform(region));
3718   set_result(_gvn.transform(phi));
3719   return true;
3720 }
3721 
3722 //---------------------generate_array_guard_common------------------------
3723 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3724                                                   bool obj_array, bool not_array) {
3725 
3726   if (stopped()) {
3727     return NULL;
3728   }
3729 
3730   // If obj_array/non_array==false/false:
3731   // Branch around if the given klass is in fact an array (either obj or prim).
3732   // If obj_array/non_array==false/true:
3733   // Branch around if the given klass is not an array klass of any kind.
3734   // If obj_array/non_array==true/true:
3735   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3736   // If obj_array/non_array==true/false:
3737   // Branch around if the kls is an oop array (Object[] or subtype)
3738   //
3739   // Like generate_guard, adds a new path onto the region.
3740   jint  layout_con = 0;
3741   Node* layout_val = get_layout_helper(kls, layout_con);
3742   if (layout_val == NULL) {
3743     bool query = (obj_array
3744                   ? Klass::layout_helper_is_objArray(layout_con)
3745                   : Klass::layout_helper_is_array(layout_con));
3746     if (query == not_array) {
3747       return NULL;                       // never a branch
3748     } else {                             // always a branch
3749       Node* always_branch = control();
3750       if (region != NULL)
3751         region->add_req(always_branch);
3752       set_control(top());
3753       return always_branch;
3754     }
3755   }
3756   // Now test the correct condition.
3757   jint  nval = (obj_array
3758                 ? ((jint)Klass::_lh_array_tag_type_value
3759                    <<    Klass::_lh_array_tag_shift)
3760                 : Klass::_lh_neutral_value);
3761   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3762   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3763   // invert the test if we are looking for a non-array
3764   if (not_array)  btest = BoolTest(btest).negate();
3765   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3766   return generate_fair_guard(bol, region);
3767 }
3768 
3769 
3770 //-----------------------inline_native_newArray--------------------------
3771 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3772 bool LibraryCallKit::inline_native_newArray() {
3773   Node* mirror    = argument(0);
3774   Node* count_val = argument(1);
3775 
3776   mirror = null_check(mirror);
3777   // If mirror or obj is dead, only null-path is taken.
3778   if (stopped())  return true;
3779 
3780   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3781   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3782   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3783   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3784   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3785 
3786   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3787   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3788                                                   result_reg, _slow_path);
3789   Node* normal_ctl   = control();
3790   Node* no_array_ctl = result_reg->in(_slow_path);
3791 
3792   // Generate code for the slow case.  We make a call to newArray().
3793   set_control(no_array_ctl);
3794   if (!stopped()) {
3795     // Either the input type is void.class, or else the
3796     // array klass has not yet been cached.  Either the
3797     // ensuing call will throw an exception, or else it
3798     // will cache the array klass for next time.
3799     PreserveJVMState pjvms(this);
3800     CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3801     Node* slow_result = set_results_for_java_call(slow_call);
3802     // this->control() comes from set_results_for_java_call
3803     result_reg->set_req(_slow_path, control());
3804     result_val->set_req(_slow_path, slow_result);
3805     result_io ->set_req(_slow_path, i_o());
3806     result_mem->set_req(_slow_path, reset_memory());
3807   }
3808 
3809   set_control(normal_ctl);
3810   if (!stopped()) {
3811     // Normal case:  The array type has been cached in the java.lang.Class.
3812     // The following call works fine even if the array type is polymorphic.
3813     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3814     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3815     result_reg->init_req(_normal_path, control());
3816     result_val->init_req(_normal_path, obj);
3817     result_io ->init_req(_normal_path, i_o());
3818     result_mem->init_req(_normal_path, reset_memory());
3819   }
3820 
3821   // Return the combined state.
3822   set_i_o(        _gvn.transform(result_io)  );
3823   set_all_memory( _gvn.transform(result_mem));
3824 
3825   C->set_has_split_ifs(true); // Has chance for split-if optimization
3826   set_result(result_reg, result_val);
3827   return true;
3828 }
3829 
3830 //----------------------inline_native_getLength--------------------------
3831 // public static native int java.lang.reflect.Array.getLength(Object array);
3832 bool LibraryCallKit::inline_native_getLength() {
3833   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3834 
3835   Node* array = null_check(argument(0));
3836   // If array is dead, only null-path is taken.
3837   if (stopped())  return true;
3838 
3839   // Deoptimize if it is a non-array.
3840   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3841 
3842   if (non_array != NULL) {
3843     PreserveJVMState pjvms(this);
3844     set_control(non_array);
3845     uncommon_trap(Deoptimization::Reason_intrinsic,
3846                   Deoptimization::Action_maybe_recompile);
3847   }
3848 
3849   // If control is dead, only non-array-path is taken.
3850   if (stopped())  return true;
3851 
3852   // The works fine even if the array type is polymorphic.
3853   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3854   Node* result = load_array_length(array);
3855 
3856   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3857   set_result(result);
3858   return true;
3859 }
3860 
3861 //------------------------inline_array_copyOf----------------------------
3862 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3863 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3864 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3865   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3866 
3867   // Get the arguments.
3868   Node* original          = argument(0);
3869   Node* start             = is_copyOfRange? argument(1): intcon(0);
3870   Node* end               = is_copyOfRange? argument(2): argument(1);
3871   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3872 
3873   Node* newcopy;
3874 
3875   // Set the original stack and the reexecute bit for the interpreter to reexecute
3876   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3877   { PreserveReexecuteState preexecs(this);
3878     jvms()->set_should_reexecute(true);
3879 
3880     array_type_mirror = null_check(array_type_mirror);
3881     original          = null_check(original);
3882 
3883     // Check if a null path was taken unconditionally.
3884     if (stopped())  return true;
3885 
3886     Node* orig_length = load_array_length(original);
3887 
3888     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3889     klass_node = null_check(klass_node);
3890 
3891     RegionNode* bailout = new RegionNode(1);
3892     record_for_igvn(bailout);
3893 
3894     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3895     // Bail out if that is so.
3896     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3897     if (not_objArray != NULL) {
3898       // Improve the klass node's type from the new optimistic assumption:
3899       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3900       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3901       Node* cast = new CastPPNode(klass_node, akls);
3902       cast->init_req(0, control());
3903       klass_node = _gvn.transform(cast);
3904     }
3905 
3906     // Bail out if either start or end is negative.
3907     generate_negative_guard(start, bailout, &start);
3908     generate_negative_guard(end,   bailout, &end);
3909 
3910     Node* length = end;
3911     if (_gvn.type(start) != TypeInt::ZERO) {
3912       length = _gvn.transform(new SubINode(end, start));
3913     }
3914 
3915     // Bail out if length is negative.
3916     // Without this the new_array would throw
3917     // NegativeArraySizeException but IllegalArgumentException is what
3918     // should be thrown
3919     generate_negative_guard(length, bailout, &length);
3920 
3921     if (bailout->req() > 1) {
3922       PreserveJVMState pjvms(this);
3923       set_control(_gvn.transform(bailout));
3924       uncommon_trap(Deoptimization::Reason_intrinsic,
3925                     Deoptimization::Action_maybe_recompile);
3926     }
3927 
3928     if (!stopped()) {
3929       // How many elements will we copy from the original?
3930       // The answer is MinI(orig_length - start, length).
3931       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3932       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3933 
3934       // Generate a direct call to the right arraycopy function(s).
3935       // We know the copy is disjoint but we might not know if the
3936       // oop stores need checking.
3937       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3938       // This will fail a store-check if x contains any non-nulls.
3939 
3940       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3941       // loads/stores but it is legal only if we're sure the
3942       // Arrays.copyOf would succeed. So we need all input arguments
3943       // to the copyOf to be validated, including that the copy to the
3944       // new array won't trigger an ArrayStoreException. That subtype
3945       // check can be optimized if we know something on the type of
3946       // the input array from type speculation.
3947       if (_gvn.type(klass_node)->singleton()) {
3948         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3949         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3950 
3951         int test = C->static_subtype_check(superk, subk);
3952         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3953           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3954           if (t_original->speculative_type() != NULL) {
3955             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3956           }
3957         }
3958       }
3959 
3960       bool validated = false;
3961       // Reason_class_check rather than Reason_intrinsic because we
3962       // want to intrinsify even if this traps.
3963       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3964         Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3965                                                    klass_node);
3966 
3967         if (not_subtype_ctrl != top()) {
3968           PreserveJVMState pjvms(this);
3969           set_control(not_subtype_ctrl);
3970           uncommon_trap(Deoptimization::Reason_class_check,
3971                         Deoptimization::Action_make_not_entrant);
3972           assert(stopped(), "Should be stopped");
3973         }
3974         validated = true;
3975       }
3976 
3977       if (!stopped()) {
3978         newcopy = new_array(klass_node, length, 0);  // no arguments to push
3979 
3980         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
3981                                                 load_object_klass(original), klass_node);
3982         if (!is_copyOfRange) {
3983           ac->set_copyof(validated);
3984         } else {
3985           ac->set_copyofrange(validated);
3986         }
3987         Node* n = _gvn.transform(ac);
3988         if (n == ac) {
3989           ac->connect_outputs(this);
3990         } else {
3991           assert(validated, "shouldn't transform if all arguments not validated");
3992           set_all_memory(n);
3993         }
3994       }
3995     }
3996   } // original reexecute is set back here
3997 
3998   C->set_has_split_ifs(true); // Has chance for split-if optimization
3999   if (!stopped()) {
4000     set_result(newcopy);
4001   }
4002   return true;
4003 }
4004 
4005 
4006 //----------------------generate_virtual_guard---------------------------
4007 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
4008 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4009                                              RegionNode* slow_region) {
4010   ciMethod* method = callee();
4011   int vtable_index = method->vtable_index();
4012   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4013          err_msg_res("bad index %d", vtable_index));
4014   // Get the Method* out of the appropriate vtable entry.
4015   int entry_offset  = (InstanceKlass::vtable_start_offset() +
4016                      vtable_index*vtableEntry::size()) * wordSize +
4017                      vtableEntry::method_offset_in_bytes();
4018   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
4019   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4020 
4021   // Compare the target method with the expected method (e.g., Object.hashCode).
4022   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4023 
4024   Node* native_call = makecon(native_call_addr);
4025   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
4026   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4027 
4028   return generate_slow_guard(test_native, slow_region);
4029 }
4030 
4031 //-----------------------generate_method_call----------------------------
4032 // Use generate_method_call to make a slow-call to the real
4033 // method if the fast path fails.  An alternative would be to
4034 // use a stub like OptoRuntime::slow_arraycopy_Java.
4035 // This only works for expanding the current library call,
4036 // not another intrinsic.  (E.g., don't use this for making an
4037 // arraycopy call inside of the copyOf intrinsic.)
4038 CallJavaNode*
4039 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4040   // When compiling the intrinsic method itself, do not use this technique.
4041   guarantee(callee() != C->method(), "cannot make slow-call to self");
4042 
4043   ciMethod* method = callee();
4044   // ensure the JVMS we have will be correct for this call
4045   guarantee(method_id == method->intrinsic_id(), "must match");
4046 
4047   const TypeFunc* tf = TypeFunc::make(method);
4048   CallJavaNode* slow_call;
4049   if (is_static) {
4050     assert(!is_virtual, "");
4051     slow_call = new CallStaticJavaNode(C, tf,
4052                            SharedRuntime::get_resolve_static_call_stub(),
4053                            method, bci());
4054   } else if (is_virtual) {
4055     null_check_receiver();
4056     int vtable_index = Method::invalid_vtable_index;
4057     if (UseInlineCaches) {
4058       // Suppress the vtable call
4059     } else {
4060       // hashCode and clone are not a miranda methods,
4061       // so the vtable index is fixed.
4062       // No need to use the linkResolver to get it.
4063        vtable_index = method->vtable_index();
4064        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4065               err_msg_res("bad index %d", vtable_index));
4066     }
4067     slow_call = new CallDynamicJavaNode(tf,
4068                           SharedRuntime::get_resolve_virtual_call_stub(),
4069                           method, vtable_index, bci());
4070   } else {  // neither virtual nor static:  opt_virtual
4071     null_check_receiver();
4072     slow_call = new CallStaticJavaNode(C, tf,
4073                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4074                                 method, bci());
4075     slow_call->set_optimized_virtual(true);
4076   }
4077   set_arguments_for_java_call(slow_call);
4078   set_edges_for_java_call(slow_call);
4079   return slow_call;
4080 }
4081 
4082 
4083 /**
4084  * Build special case code for calls to hashCode on an object. This call may
4085  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4086  * slightly different code.
4087  */
4088 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4089   assert(is_static == callee()->is_static(), "correct intrinsic selection");
4090   assert(!(is_virtual && is_static), "either virtual, special, or static");
4091 
4092   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4093 
4094   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4095   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
4096   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
4097   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4098   Node* obj = NULL;
4099   if (!is_static) {
4100     // Check for hashing null object
4101     obj = null_check_receiver();
4102     if (stopped())  return true;        // unconditionally null
4103     result_reg->init_req(_null_path, top());
4104     result_val->init_req(_null_path, top());
4105   } else {
4106     // Do a null check, and return zero if null.
4107     // System.identityHashCode(null) == 0
4108     obj = argument(0);
4109     Node* null_ctl = top();
4110     obj = null_check_oop(obj, &null_ctl);
4111     result_reg->init_req(_null_path, null_ctl);
4112     result_val->init_req(_null_path, _gvn.intcon(0));
4113   }
4114 
4115   // Unconditionally null?  Then return right away.
4116   if (stopped()) {
4117     set_control( result_reg->in(_null_path));
4118     if (!stopped())
4119       set_result(result_val->in(_null_path));
4120     return true;
4121   }
4122 
4123   // We only go to the fast case code if we pass a number of guards.  The
4124   // paths which do not pass are accumulated in the slow_region.
4125   RegionNode* slow_region = new RegionNode(1);
4126   record_for_igvn(slow_region);
4127 
4128   // If this is a virtual call, we generate a funny guard.  We pull out
4129   // the vtable entry corresponding to hashCode() from the target object.
4130   // If the target method which we are calling happens to be the native
4131   // Object hashCode() method, we pass the guard.  We do not need this
4132   // guard for non-virtual calls -- the caller is known to be the native
4133   // Object hashCode().
4134   if (is_virtual) {
4135     // After null check, get the object's klass.
4136     Node* obj_klass = load_object_klass(obj);
4137     generate_virtual_guard(obj_klass, slow_region);
4138   }
4139 
4140   // Get the header out of the object, use LoadMarkNode when available
4141   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4142   // The control of the load must be NULL. Otherwise, the load can move before
4143   // the null check after castPP removal.
4144   Node* no_ctrl = NULL;
4145   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4146 
4147   // Test the header to see if it is unlocked.
4148   Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4149   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4150   Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
4151   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4152   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4153 
4154   generate_slow_guard(test_unlocked, slow_region);
4155 
4156   // Get the hash value and check to see that it has been properly assigned.
4157   // We depend on hash_mask being at most 32 bits and avoid the use of
4158   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4159   // vm: see markOop.hpp.
4160   Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
4161   Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
4162   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4163   // This hack lets the hash bits live anywhere in the mark object now, as long
4164   // as the shift drops the relevant bits into the low 32 bits.  Note that
4165   // Java spec says that HashCode is an int so there's no point in capturing
4166   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4167   hshifted_header      = ConvX2I(hshifted_header);
4168   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4169 
4170   Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
4171   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4172   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4173 
4174   generate_slow_guard(test_assigned, slow_region);
4175 
4176   Node* init_mem = reset_memory();
4177   // fill in the rest of the null path:
4178   result_io ->init_req(_null_path, i_o());
4179   result_mem->init_req(_null_path, init_mem);
4180 
4181   result_val->init_req(_fast_path, hash_val);
4182   result_reg->init_req(_fast_path, control());
4183   result_io ->init_req(_fast_path, i_o());
4184   result_mem->init_req(_fast_path, init_mem);
4185 
4186   // Generate code for the slow case.  We make a call to hashCode().
4187   set_control(_gvn.transform(slow_region));
4188   if (!stopped()) {
4189     // No need for PreserveJVMState, because we're using up the present state.
4190     set_all_memory(init_mem);
4191     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4192     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4193     Node* slow_result = set_results_for_java_call(slow_call);
4194     // this->control() comes from set_results_for_java_call
4195     result_reg->init_req(_slow_path, control());
4196     result_val->init_req(_slow_path, slow_result);
4197     result_io  ->set_req(_slow_path, i_o());
4198     result_mem ->set_req(_slow_path, reset_memory());
4199   }
4200 
4201   // Return the combined state.
4202   set_i_o(        _gvn.transform(result_io)  );
4203   set_all_memory( _gvn.transform(result_mem));
4204 
4205   set_result(result_reg, result_val);
4206   return true;
4207 }
4208 
4209 //---------------------------inline_native_getClass----------------------------
4210 // public final native Class<?> java.lang.Object.getClass();
4211 //
4212 // Build special case code for calls to getClass on an object.
4213 bool LibraryCallKit::inline_native_getClass() {
4214   Node* obj = null_check_receiver();
4215   if (stopped())  return true;
4216   set_result(load_mirror_from_klass(load_object_klass(obj)));
4217   return true;
4218 }
4219 
4220 //-----------------inline_native_Reflection_getCallerClass---------------------
4221 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4222 //
4223 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4224 //
4225 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4226 // in that it must skip particular security frames and checks for
4227 // caller sensitive methods.
4228 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4229 #ifndef PRODUCT
4230   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4231     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4232   }
4233 #endif
4234 
4235   if (!jvms()->has_method()) {
4236 #ifndef PRODUCT
4237     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4238       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
4239     }
4240 #endif
4241     return false;
4242   }
4243 
4244   // Walk back up the JVM state to find the caller at the required
4245   // depth.
4246   JVMState* caller_jvms = jvms();
4247 
4248   // Cf. JVM_GetCallerClass
4249   // NOTE: Start the loop at depth 1 because the current JVM state does
4250   // not include the Reflection.getCallerClass() frame.
4251   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4252     ciMethod* m = caller_jvms->method();
4253     switch (n) {
4254     case 0:
4255       fatal("current JVM state does not include the Reflection.getCallerClass frame");
4256       break;
4257     case 1:
4258       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4259       if (!m->caller_sensitive()) {
4260 #ifndef PRODUCT
4261         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4262           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
4263         }
4264 #endif
4265         return false;  // bail-out; let JVM_GetCallerClass do the work
4266       }
4267       break;
4268     default:
4269       if (!m->is_ignored_by_security_stack_walk()) {
4270         // We have reached the desired frame; return the holder class.
4271         // Acquire method holder as java.lang.Class and push as constant.
4272         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4273         ciInstance* caller_mirror = caller_klass->java_mirror();
4274         set_result(makecon(TypeInstPtr::make(caller_mirror)));
4275 
4276 #ifndef PRODUCT
4277         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4278           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());
4279           tty->print_cr("  JVM state at this point:");
4280           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4281             ciMethod* m = jvms()->of_depth(i)->method();
4282             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4283           }
4284         }
4285 #endif
4286         return true;
4287       }
4288       break;
4289     }
4290   }
4291 
4292 #ifndef PRODUCT
4293   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4294     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4295     tty->print_cr("  JVM state at this point:");
4296     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4297       ciMethod* m = jvms()->of_depth(i)->method();
4298       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4299     }
4300   }
4301 #endif
4302 
4303   return false;  // bail-out; let JVM_GetCallerClass do the work
4304 }
4305 
4306 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4307   Node* arg = argument(0);
4308   Node* result;
4309 
4310   switch (id) {
4311   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4312   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4313   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4314   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4315 
4316   case vmIntrinsics::_doubleToLongBits: {
4317     // two paths (plus control) merge in a wood
4318     RegionNode *r = new RegionNode(3);
4319     Node *phi = new PhiNode(r, TypeLong::LONG);
4320 
4321     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4322     // Build the boolean node
4323     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4324 
4325     // Branch either way.
4326     // NaN case is less traveled, which makes all the difference.
4327     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4328     Node *opt_isnan = _gvn.transform(ifisnan);
4329     assert( opt_isnan->is_If(), "Expect an IfNode");
4330     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4331     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4332 
4333     set_control(iftrue);
4334 
4335     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4336     Node *slow_result = longcon(nan_bits); // return NaN
4337     phi->init_req(1, _gvn.transform( slow_result ));
4338     r->init_req(1, iftrue);
4339 
4340     // Else fall through
4341     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4342     set_control(iffalse);
4343 
4344     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4345     r->init_req(2, iffalse);
4346 
4347     // Post merge
4348     set_control(_gvn.transform(r));
4349     record_for_igvn(r);
4350 
4351     C->set_has_split_ifs(true); // Has chance for split-if optimization
4352     result = phi;
4353     assert(result->bottom_type()->isa_long(), "must be");
4354     break;
4355   }
4356 
4357   case vmIntrinsics::_floatToIntBits: {
4358     // two paths (plus control) merge in a wood
4359     RegionNode *r = new RegionNode(3);
4360     Node *phi = new PhiNode(r, TypeInt::INT);
4361 
4362     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4363     // Build the boolean node
4364     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4365 
4366     // Branch either way.
4367     // NaN case is less traveled, which makes all the difference.
4368     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4369     Node *opt_isnan = _gvn.transform(ifisnan);
4370     assert( opt_isnan->is_If(), "Expect an IfNode");
4371     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4372     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4373 
4374     set_control(iftrue);
4375 
4376     static const jint nan_bits = 0x7fc00000;
4377     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4378     phi->init_req(1, _gvn.transform( slow_result ));
4379     r->init_req(1, iftrue);
4380 
4381     // Else fall through
4382     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4383     set_control(iffalse);
4384 
4385     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4386     r->init_req(2, iffalse);
4387 
4388     // Post merge
4389     set_control(_gvn.transform(r));
4390     record_for_igvn(r);
4391 
4392     C->set_has_split_ifs(true); // Has chance for split-if optimization
4393     result = phi;
4394     assert(result->bottom_type()->isa_int(), "must be");
4395     break;
4396   }
4397 
4398   default:
4399     fatal_unexpected_iid(id);
4400     break;
4401   }
4402   set_result(_gvn.transform(result));
4403   return true;
4404 }
4405 
4406 #ifdef _LP64
4407 #define XTOP ,top() /*additional argument*/
4408 #else  //_LP64
4409 #define XTOP        /*no additional argument*/
4410 #endif //_LP64
4411 
4412 //----------------------inline_unsafe_copyMemory-------------------------
4413 // public native void sun.misc.Unsafe.copyMemory(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4414 bool LibraryCallKit::inline_unsafe_copyMemory() {
4415   if (callee()->is_static())  return false;  // caller must have the capability!
4416   null_check_receiver();  // null-check receiver
4417   if (stopped())  return true;
4418 
4419   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4420 
4421   Node* src_ptr =         argument(1);   // type: oop
4422   Node* src_off = ConvL2X(argument(2));  // type: long
4423   Node* dst_ptr =         argument(4);   // type: oop
4424   Node* dst_off = ConvL2X(argument(5));  // type: long
4425   Node* size    = ConvL2X(argument(7));  // type: long
4426 
4427   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4428          "fieldOffset must be byte-scaled");
4429 
4430   Node* src = make_unsafe_address(src_ptr, src_off);
4431   Node* dst = make_unsafe_address(dst_ptr, dst_off);
4432 
4433   // Conservatively insert a memory barrier on all memory slices.
4434   // Do not let writes of the copy source or destination float below the copy.
4435   insert_mem_bar(Op_MemBarCPUOrder);
4436 
4437   // Call it.  Note that the length argument is not scaled.
4438   make_runtime_call(RC_LEAF|RC_NO_FP,
4439                     OptoRuntime::fast_arraycopy_Type(),
4440                     StubRoutines::unsafe_arraycopy(),
4441                     "unsafe_arraycopy",
4442                     TypeRawPtr::BOTTOM,
4443                     src, dst, size XTOP);
4444 
4445   // Do not let reads of the copy destination float above the copy.
4446   insert_mem_bar(Op_MemBarCPUOrder);
4447 
4448   return true;
4449 }
4450 
4451 //------------------------clone_coping-----------------------------------
4452 // Helper function for inline_native_clone.
4453 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4454   assert(obj_size != NULL, "");
4455   Node* raw_obj = alloc_obj->in(1);
4456   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4457 
4458   AllocateNode* alloc = NULL;
4459   if (ReduceBulkZeroing) {
4460     // We will be completely responsible for initializing this object -
4461     // mark Initialize node as complete.
4462     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4463     // The object was just allocated - there should be no any stores!
4464     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4465     // Mark as complete_with_arraycopy so that on AllocateNode
4466     // expansion, we know this AllocateNode is initialized by an array
4467     // copy and a StoreStore barrier exists after the array copy.
4468     alloc->initialization()->set_complete_with_arraycopy();
4469   }
4470 
4471   // Copy the fastest available way.
4472   // TODO: generate fields copies for small objects instead.
4473   Node* src  = obj;
4474   Node* dest = alloc_obj;
4475   Node* size = _gvn.transform(obj_size);
4476 
4477   // Exclude the header but include array length to copy by 8 bytes words.
4478   // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4479   int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4480                             instanceOopDesc::base_offset_in_bytes();
4481   // base_off:
4482   // 8  - 32-bit VM
4483   // 12 - 64-bit VM, compressed klass
4484   // 16 - 64-bit VM, normal klass
4485   if (base_off % BytesPerLong != 0) {
4486     assert(UseCompressedClassPointers, "");
4487     if (is_array) {
4488       // Exclude length to copy by 8 bytes words.
4489       base_off += sizeof(int);
4490     } else {
4491       // Include klass to copy by 8 bytes words.
4492       base_off = instanceOopDesc::klass_offset_in_bytes();
4493     }
4494     assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4495   }
4496   src  = basic_plus_adr(src,  base_off);
4497   dest = basic_plus_adr(dest, base_off);
4498 
4499   // Compute the length also, if needed:
4500   Node* countx = size;
4501   countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4502   countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4503 
4504   const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4505 
4506   ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4507   ac->set_clonebasic();
4508   Node* n = _gvn.transform(ac);
4509   if (n == ac) {
4510     set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4511   } else {
4512     set_all_memory(n);
4513   }
4514 
4515   // If necessary, emit some card marks afterwards.  (Non-arrays only.)
4516   if (card_mark) {
4517     assert(!is_array, "");
4518     // Put in store barrier for any and all oops we are sticking
4519     // into this object.  (We could avoid this if we could prove
4520     // that the object type contains no oop fields at all.)
4521     Node* no_particular_value = NULL;
4522     Node* no_particular_field = NULL;
4523     int raw_adr_idx = Compile::AliasIdxRaw;
4524     post_barrier(control(),
4525                  memory(raw_adr_type),
4526                  alloc_obj,
4527                  no_particular_field,
4528                  raw_adr_idx,
4529                  no_particular_value,
4530                  T_OBJECT,
4531                  false);
4532   }
4533 
4534   // Do not let reads from the cloned object float above the arraycopy.
4535   if (alloc != NULL) {
4536     // Do not let stores that initialize this object be reordered with
4537     // a subsequent store that would make this object accessible by
4538     // other threads.
4539     // Record what AllocateNode this StoreStore protects so that
4540     // escape analysis can go from the MemBarStoreStoreNode to the
4541     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4542     // based on the escape status of the AllocateNode.
4543     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4544   } else {
4545     insert_mem_bar(Op_MemBarCPUOrder);
4546   }
4547 }
4548 
4549 //------------------------inline_native_clone----------------------------
4550 // protected native Object java.lang.Object.clone();
4551 //
4552 // Here are the simple edge cases:
4553 //  null receiver => normal trap
4554 //  virtual and clone was overridden => slow path to out-of-line clone
4555 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4556 //
4557 // The general case has two steps, allocation and copying.
4558 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4559 //
4560 // Copying also has two cases, oop arrays and everything else.
4561 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4562 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4563 //
4564 // These steps fold up nicely if and when the cloned object's klass
4565 // can be sharply typed as an object array, a type array, or an instance.
4566 //
4567 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4568   PhiNode* result_val;
4569 
4570   // Set the reexecute bit for the interpreter to reexecute
4571   // the bytecode that invokes Object.clone if deoptimization happens.
4572   { PreserveReexecuteState preexecs(this);
4573     jvms()->set_should_reexecute(true);
4574 
4575     Node* obj = null_check_receiver();
4576     if (stopped())  return true;
4577 
4578     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4579 
4580     // If we are going to clone an instance, we need its exact type to
4581     // know the number and types of fields to convert the clone to
4582     // loads/stores. Maybe a speculative type can help us.
4583     if (!obj_type->klass_is_exact() &&
4584         obj_type->speculative_type() != NULL &&
4585         obj_type->speculative_type()->is_instance_klass()) {
4586       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4587       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4588           !spec_ik->has_injected_fields()) {
4589         ciKlass* k = obj_type->klass();
4590         if (!k->is_instance_klass() ||
4591             k->as_instance_klass()->is_interface() ||
4592             k->as_instance_klass()->has_subklass()) {
4593           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4594         }
4595       }
4596     }
4597 
4598     Node* obj_klass = load_object_klass(obj);
4599     const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4600     const TypeOopPtr*   toop   = ((tklass != NULL)
4601                                 ? tklass->as_instance_type()
4602                                 : TypeInstPtr::NOTNULL);
4603 
4604     // Conservatively insert a memory barrier on all memory slices.
4605     // Do not let writes into the original float below the clone.
4606     insert_mem_bar(Op_MemBarCPUOrder);
4607 
4608     // paths into result_reg:
4609     enum {
4610       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4611       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4612       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4613       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4614       PATH_LIMIT
4615     };
4616     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4617     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4618     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4619     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4620     record_for_igvn(result_reg);
4621 
4622     const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4623     int raw_adr_idx = Compile::AliasIdxRaw;
4624 
4625     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4626     if (array_ctl != NULL) {
4627       // It's an array.
4628       PreserveJVMState pjvms(this);
4629       set_control(array_ctl);
4630       Node* obj_length = load_array_length(obj);
4631       Node* obj_size  = NULL;
4632       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4633 
4634       if (!use_ReduceInitialCardMarks()) {
4635         // If it is an oop array, it requires very special treatment,
4636         // because card marking is required on each card of the array.
4637         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4638         if (is_obja != NULL) {
4639           PreserveJVMState pjvms2(this);
4640           set_control(is_obja);
4641           // Generate a direct call to the right arraycopy function(s).
4642           Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4643           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4644           ac->set_cloneoop();
4645           Node* n = _gvn.transform(ac);
4646           assert(n == ac, "cannot disappear");
4647           ac->connect_outputs(this);
4648 
4649           result_reg->init_req(_objArray_path, control());
4650           result_val->init_req(_objArray_path, alloc_obj);
4651           result_i_o ->set_req(_objArray_path, i_o());
4652           result_mem ->set_req(_objArray_path, reset_memory());
4653         }
4654       }
4655       // Otherwise, there are no card marks to worry about.
4656       // (We can dispense with card marks if we know the allocation
4657       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4658       //  causes the non-eden paths to take compensating steps to
4659       //  simulate a fresh allocation, so that no further
4660       //  card marks are required in compiled code to initialize
4661       //  the object.)
4662 
4663       if (!stopped()) {
4664         copy_to_clone(obj, alloc_obj, obj_size, true, false);
4665 
4666         // Present the results of the copy.
4667         result_reg->init_req(_array_path, control());
4668         result_val->init_req(_array_path, alloc_obj);
4669         result_i_o ->set_req(_array_path, i_o());
4670         result_mem ->set_req(_array_path, reset_memory());
4671       }
4672     }
4673 
4674     // We only go to the instance fast case code if we pass a number of guards.
4675     // The paths which do not pass are accumulated in the slow_region.
4676     RegionNode* slow_region = new RegionNode(1);
4677     record_for_igvn(slow_region);
4678     if (!stopped()) {
4679       // It's an instance (we did array above).  Make the slow-path tests.
4680       // If this is a virtual call, we generate a funny guard.  We grab
4681       // the vtable entry corresponding to clone() from the target object.
4682       // If the target method which we are calling happens to be the
4683       // Object clone() method, we pass the guard.  We do not need this
4684       // guard for non-virtual calls; the caller is known to be the native
4685       // Object clone().
4686       if (is_virtual) {
4687         generate_virtual_guard(obj_klass, slow_region);
4688       }
4689 
4690       // The object must be cloneable and must not have a finalizer.
4691       // Both of these conditions may be checked in a single test.
4692       // We could optimize the cloneable test further, but we don't care.
4693       generate_access_flags_guard(obj_klass,
4694                                   // Test both conditions:
4695                                   JVM_ACC_IS_CLONEABLE | JVM_ACC_HAS_FINALIZER,
4696                                   // Must be cloneable but not finalizer:
4697                                   JVM_ACC_IS_CLONEABLE,
4698                                   slow_region);
4699     }
4700 
4701     if (!stopped()) {
4702       // It's an instance, and it passed the slow-path tests.
4703       PreserveJVMState pjvms(this);
4704       Node* obj_size  = NULL;
4705       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4706       // is reexecuted if deoptimization occurs and there could be problems when merging
4707       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4708       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4709 
4710       copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4711 
4712       // Present the results of the slow call.
4713       result_reg->init_req(_instance_path, control());
4714       result_val->init_req(_instance_path, alloc_obj);
4715       result_i_o ->set_req(_instance_path, i_o());
4716       result_mem ->set_req(_instance_path, reset_memory());
4717     }
4718 
4719     // Generate code for the slow case.  We make a call to clone().
4720     set_control(_gvn.transform(slow_region));
4721     if (!stopped()) {
4722       PreserveJVMState pjvms(this);
4723       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4724       Node* slow_result = set_results_for_java_call(slow_call);
4725       // this->control() comes from set_results_for_java_call
4726       result_reg->init_req(_slow_path, control());
4727       result_val->init_req(_slow_path, slow_result);
4728       result_i_o ->set_req(_slow_path, i_o());
4729       result_mem ->set_req(_slow_path, reset_memory());
4730     }
4731 
4732     // Return the combined state.
4733     set_control(    _gvn.transform(result_reg));
4734     set_i_o(        _gvn.transform(result_i_o));
4735     set_all_memory( _gvn.transform(result_mem));
4736   } // original reexecute is set back here
4737 
4738   set_result(_gvn.transform(result_val));
4739   return true;
4740 }
4741 
4742 // If we have a tighly coupled allocation, the arraycopy may take care
4743 // of the array initialization. If one of the guards we insert between
4744 // the allocation and the arraycopy causes a deoptimization, an
4745 // unitialized array will escape the compiled method. To prevent that
4746 // we set the JVM state for uncommon traps between the allocation and
4747 // the arraycopy to the state before the allocation so, in case of
4748 // deoptimization, we'll reexecute the allocation and the
4749 // initialization.
4750 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4751   if (alloc != NULL) {
4752     ciMethod* trap_method = alloc->jvms()->method();
4753     int trap_bci = alloc->jvms()->bci();
4754 
4755     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4756           !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4757       // Make sure there's no store between the allocation and the
4758       // arraycopy otherwise visible side effects could be rexecuted
4759       // in case of deoptimization and cause incorrect execution.
4760       bool no_interfering_store = true;
4761       Node* mem = alloc->in(TypeFunc::Memory);
4762       if (mem->is_MergeMem()) {
4763         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4764           Node* n = mms.memory();
4765           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4766             assert(n->is_Store(), "what else?");
4767             no_interfering_store = false;
4768             break;
4769           }
4770         }
4771       } else {
4772         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4773           Node* n = mms.memory();
4774           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4775             assert(n->is_Store(), "what else?");
4776             no_interfering_store = false;
4777             break;
4778           }
4779         }
4780       }
4781 
4782       if (no_interfering_store) {
4783         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4784         uint size = alloc->req();
4785         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4786         old_jvms->set_map(sfpt);
4787         for (uint i = 0; i < size; i++) {
4788           sfpt->init_req(i, alloc->in(i));
4789         }
4790         // re-push array length for deoptimization
4791         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4792         old_jvms->set_sp(old_jvms->sp()+1);
4793         old_jvms->set_monoff(old_jvms->monoff()+1);
4794         old_jvms->set_scloff(old_jvms->scloff()+1);
4795         old_jvms->set_endoff(old_jvms->endoff()+1);
4796         old_jvms->set_should_reexecute(true);
4797 
4798         sfpt->set_i_o(map()->i_o());
4799         sfpt->set_memory(map()->memory());
4800         sfpt->set_control(map()->control());
4801 
4802         JVMState* saved_jvms = jvms();
4803         saved_reexecute_sp = _reexecute_sp;
4804 
4805         set_jvms(sfpt->jvms());
4806         _reexecute_sp = jvms()->sp();
4807 
4808         return saved_jvms;
4809       }
4810     }
4811   }
4812   return NULL;
4813 }
4814 
4815 // In case of a deoptimization, we restart execution at the
4816 // allocation, allocating a new array. We would leave an uninitialized
4817 // array in the heap that GCs wouldn't expect. Move the allocation
4818 // after the traps so we don't allocate the array if we
4819 // deoptimize. This is possible because tightly_coupled_allocation()
4820 // guarantees there's no observer of the allocated array at this point
4821 // and the control flow is simple enough.
4822 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4823   if (saved_jvms != NULL && !stopped()) {
4824     assert(alloc != NULL, "only with a tightly coupled allocation");
4825     // restore JVM state to the state at the arraycopy
4826     saved_jvms->map()->set_control(map()->control());
4827     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4828     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4829     // If we've improved the types of some nodes (null check) while
4830     // emitting the guards, propagate them to the current state
4831     map()->replaced_nodes().apply(saved_jvms->map());
4832     set_jvms(saved_jvms);
4833     _reexecute_sp = saved_reexecute_sp;
4834 
4835     // Remove the allocation from above the guards
4836     CallProjections callprojs;
4837     alloc->extract_projections(&callprojs, true);
4838     InitializeNode* init = alloc->initialization();
4839     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4840     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4841     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4842     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4843 
4844     // move the allocation here (after the guards)
4845     _gvn.hash_delete(alloc);
4846     alloc->set_req(TypeFunc::Control, control());
4847     alloc->set_req(TypeFunc::I_O, i_o());
4848     Node *mem = reset_memory();
4849     set_all_memory(mem);
4850     alloc->set_req(TypeFunc::Memory, mem);
4851     set_control(init->proj_out(TypeFunc::Control));
4852     set_i_o(callprojs.fallthrough_ioproj);
4853 
4854     // Update memory as done in GraphKit::set_output_for_allocation()
4855     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4856     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4857     if (ary_type->isa_aryptr() && length_type != NULL) {
4858       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4859     }
4860     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4861     int            elemidx  = C->get_alias_index(telemref);
4862     set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4863     set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4864 
4865     Node* allocx = _gvn.transform(alloc);
4866     assert(allocx == alloc, "where has the allocation gone?");
4867     assert(dest->is_CheckCastPP(), "not an allocation result?");
4868 
4869     _gvn.hash_delete(dest);
4870     dest->set_req(0, control());
4871     Node* destx = _gvn.transform(dest);
4872     assert(destx == dest, "where has the allocation result gone?");
4873   }
4874 }
4875 
4876 
4877 //------------------------------inline_arraycopy-----------------------
4878 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4879 //                                                      Object dest, int destPos,
4880 //                                                      int length);
4881 bool LibraryCallKit::inline_arraycopy() {
4882   // Get the arguments.
4883   Node* src         = argument(0);  // type: oop
4884   Node* src_offset  = argument(1);  // type: int
4885   Node* dest        = argument(2);  // type: oop
4886   Node* dest_offset = argument(3);  // type: int
4887   Node* length      = argument(4);  // type: int
4888 
4889 
4890   // Check for allocation before we add nodes that would confuse
4891   // tightly_coupled_allocation()
4892   AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4893 
4894   int saved_reexecute_sp = -1;
4895   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4896   // See arraycopy_restore_alloc_state() comment
4897   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4898   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4899   // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4900   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4901 
4902   // The following tests must be performed
4903   // (1) src and dest are arrays.
4904   // (2) src and dest arrays must have elements of the same BasicType
4905   // (3) src and dest must not be null.
4906   // (4) src_offset must not be negative.
4907   // (5) dest_offset must not be negative.
4908   // (6) length must not be negative.
4909   // (7) src_offset + length must not exceed length of src.
4910   // (8) dest_offset + length must not exceed length of dest.
4911   // (9) each element of an oop array must be assignable
4912 
4913   // (3) src and dest must not be null.
4914   // always do this here because we need the JVM state for uncommon traps
4915   Node* null_ctl = top();
4916   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4917   assert(null_ctl->is_top(), "no null control here");
4918   dest = null_check(dest, T_ARRAY);
4919 
4920   if (!can_emit_guards) {
4921     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4922     // guards but the arraycopy node could still take advantage of a
4923     // tightly allocated allocation. tightly_coupled_allocation() is
4924     // called again to make sure it takes the null check above into
4925     // account: the null check is mandatory and if it caused an
4926     // uncommon trap to be emitted then the allocation can't be
4927     // considered tightly coupled in this context.
4928     alloc = tightly_coupled_allocation(dest, NULL);
4929   }
4930 
4931   bool validated = false;
4932 
4933   const Type* src_type  = _gvn.type(src);
4934   const Type* dest_type = _gvn.type(dest);
4935   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4936   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4937 
4938   // Do we have the type of src?
4939   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4940   // Do we have the type of dest?
4941   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4942   // Is the type for src from speculation?
4943   bool src_spec = false;
4944   // Is the type for dest from speculation?
4945   bool dest_spec = false;
4946 
4947   if ((!has_src || !has_dest) && can_emit_guards) {
4948     // We don't have sufficient type information, let's see if
4949     // speculative types can help. We need to have types for both src
4950     // and dest so that it pays off.
4951 
4952     // Do we already have or could we have type information for src
4953     bool could_have_src = has_src;
4954     // Do we already have or could we have type information for dest
4955     bool could_have_dest = has_dest;
4956 
4957     ciKlass* src_k = NULL;
4958     if (!has_src) {
4959       src_k = src_type->speculative_type_not_null();
4960       if (src_k != NULL && src_k->is_array_klass()) {
4961         could_have_src = true;
4962       }
4963     }
4964 
4965     ciKlass* dest_k = NULL;
4966     if (!has_dest) {
4967       dest_k = dest_type->speculative_type_not_null();
4968       if (dest_k != NULL && dest_k->is_array_klass()) {
4969         could_have_dest = true;
4970       }
4971     }
4972 
4973     if (could_have_src && could_have_dest) {
4974       // This is going to pay off so emit the required guards
4975       if (!has_src) {
4976         src = maybe_cast_profiled_obj(src, src_k, true);
4977         src_type  = _gvn.type(src);
4978         top_src  = src_type->isa_aryptr();
4979         has_src = (top_src != NULL && top_src->klass() != NULL);
4980         src_spec = true;
4981       }
4982       if (!has_dest) {
4983         dest = maybe_cast_profiled_obj(dest, dest_k, true);
4984         dest_type  = _gvn.type(dest);
4985         top_dest  = dest_type->isa_aryptr();
4986         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4987         dest_spec = true;
4988       }
4989     }
4990   }
4991 
4992   if (has_src && has_dest && can_emit_guards) {
4993     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
4994     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4995     if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
4996     if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;
4997 
4998     if (src_elem == dest_elem && src_elem == T_OBJECT) {
4999       // If both arrays are object arrays then having the exact types
5000       // for both will remove the need for a subtype check at runtime
5001       // before the call and may make it possible to pick a faster copy
5002       // routine (without a subtype check on every element)
5003       // Do we have the exact type of src?
5004       bool could_have_src = src_spec;
5005       // Do we have the exact type of dest?
5006       bool could_have_dest = dest_spec;
5007       ciKlass* src_k = top_src->klass();
5008       ciKlass* dest_k = top_dest->klass();
5009       if (!src_spec) {
5010         src_k = src_type->speculative_type_not_null();
5011         if (src_k != NULL && src_k->is_array_klass()) {
5012           could_have_src = true;
5013         }
5014       }
5015       if (!dest_spec) {
5016         dest_k = dest_type->speculative_type_not_null();
5017         if (dest_k != NULL && dest_k->is_array_klass()) {
5018           could_have_dest = true;
5019         }
5020       }
5021       if (could_have_src && could_have_dest) {
5022         // If we can have both exact types, emit the missing guards
5023         if (could_have_src && !src_spec) {
5024           src = maybe_cast_profiled_obj(src, src_k, true);
5025         }
5026         if (could_have_dest && !dest_spec) {
5027           dest = maybe_cast_profiled_obj(dest, dest_k, true);
5028         }
5029       }
5030     }
5031   }
5032 
5033   ciMethod* trap_method = method();
5034   int trap_bci = bci();
5035   if (saved_jvms != NULL) {
5036     trap_method = alloc->jvms()->method();
5037     trap_bci = alloc->jvms()->bci();
5038   }
5039 
5040   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5041       can_emit_guards &&
5042       !src->is_top() && !dest->is_top()) {
5043     // validate arguments: enables transformation the ArrayCopyNode
5044     validated = true;
5045 
5046     RegionNode* slow_region = new RegionNode(1);
5047     record_for_igvn(slow_region);
5048 
5049     // (1) src and dest are arrays.
5050     generate_non_array_guard(load_object_klass(src), slow_region);
5051     generate_non_array_guard(load_object_klass(dest), slow_region);
5052 
5053     // (2) src and dest arrays must have elements of the same BasicType
5054     // done at macro expansion or at Ideal transformation time
5055 
5056     // (4) src_offset must not be negative.
5057     generate_negative_guard(src_offset, slow_region);
5058 
5059     // (5) dest_offset must not be negative.
5060     generate_negative_guard(dest_offset, slow_region);
5061 
5062     // (7) src_offset + length must not exceed length of src.
5063     generate_limit_guard(src_offset, length,
5064                          load_array_length(src),
5065                          slow_region);
5066 
5067     // (8) dest_offset + length must not exceed length of dest.
5068     generate_limit_guard(dest_offset, length,
5069                          load_array_length(dest),
5070                          slow_region);
5071 
5072     // (9) each element of an oop array must be assignable
5073     Node* src_klass  = load_object_klass(src);
5074     Node* dest_klass = load_object_klass(dest);
5075     Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5076 
5077     if (not_subtype_ctrl != top()) {
5078       PreserveJVMState pjvms(this);
5079       set_control(not_subtype_ctrl);
5080       uncommon_trap(Deoptimization::Reason_intrinsic,
5081                     Deoptimization::Action_make_not_entrant);
5082       assert(stopped(), "Should be stopped");
5083     }
5084     {
5085       PreserveJVMState pjvms(this);
5086       set_control(_gvn.transform(slow_region));
5087       uncommon_trap(Deoptimization::Reason_intrinsic,
5088                     Deoptimization::Action_make_not_entrant);
5089       assert(stopped(), "Should be stopped");
5090     }
5091   }
5092 
5093   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5094 
5095   if (stopped()) {
5096     return true;
5097   }
5098 
5099   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5100                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
5101                                           // so the compiler has a chance to eliminate them: during macro expansion,
5102                                           // we have to set their control (CastPP nodes are eliminated).
5103                                           load_object_klass(src), load_object_klass(dest),
5104                                           load_array_length(src), load_array_length(dest));
5105 
5106   ac->set_arraycopy(validated);
5107 
5108   Node* n = _gvn.transform(ac);
5109   if (n == ac) {
5110     ac->connect_outputs(this);
5111   } else {
5112     assert(validated, "shouldn't transform if all arguments not validated");
5113     set_all_memory(n);
5114   }
5115 
5116   return true;
5117 }
5118 
5119 
5120 // Helper function which determines if an arraycopy immediately follows
5121 // an allocation, with no intervening tests or other escapes for the object.
5122 AllocateArrayNode*
5123 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5124                                            RegionNode* slow_region) {
5125   if (stopped())             return NULL;  // no fast path
5126   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
5127 
5128   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5129   if (alloc == NULL)  return NULL;
5130 
5131   Node* rawmem = memory(Compile::AliasIdxRaw);
5132   // Is the allocation's memory state untouched?
5133   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5134     // Bail out if there have been raw-memory effects since the allocation.
5135     // (Example:  There might have been a call or safepoint.)
5136     return NULL;
5137   }
5138   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5139   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5140     return NULL;
5141   }
5142 
5143   // There must be no unexpected observers of this allocation.
5144   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5145     Node* obs = ptr->fast_out(i);
5146     if (obs != this->map()) {
5147       return NULL;
5148     }
5149   }
5150 
5151   // This arraycopy must unconditionally follow the allocation of the ptr.
5152   Node* alloc_ctl = ptr->in(0);
5153   assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5154 
5155   Node* ctl = control();
5156   while (ctl != alloc_ctl) {
5157     // There may be guards which feed into the slow_region.
5158     // Any other control flow means that we might not get a chance
5159     // to finish initializing the allocated object.
5160     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5161       IfNode* iff = ctl->in(0)->as_If();
5162       Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5163       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5164       if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5165         ctl = iff->in(0);       // This test feeds the known slow_region.
5166         continue;
5167       }
5168       // One more try:  Various low-level checks bottom out in
5169       // uncommon traps.  If the debug-info of the trap omits
5170       // any reference to the allocation, as we've already
5171       // observed, then there can be no objection to the trap.
5172       bool found_trap = false;
5173       for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5174         Node* obs = not_ctl->fast_out(j);
5175         if (obs->in(0) == not_ctl && obs->is_Call() &&
5176             (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5177           found_trap = true; break;
5178         }
5179       }
5180       if (found_trap) {
5181         ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
5182         continue;
5183       }
5184     }
5185     return NULL;
5186   }
5187 
5188   // If we get this far, we have an allocation which immediately
5189   // precedes the arraycopy, and we can take over zeroing the new object.
5190   // The arraycopy will finish the initialization, and provide
5191   // a new control state to which we will anchor the destination pointer.
5192 
5193   return alloc;
5194 }
5195 
5196 //-------------inline_encodeISOArray-----------------------------------
5197 // encode char[] to byte[] in ISO_8859_1
5198 bool LibraryCallKit::inline_encodeISOArray() {
5199   assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5200   // no receiver since it is static method
5201   Node *src         = argument(0);
5202   Node *src_offset  = argument(1);
5203   Node *dst         = argument(2);
5204   Node *dst_offset  = argument(3);
5205   Node *length      = argument(4);
5206 
5207   const Type* src_type = src->Value(&_gvn);
5208   const Type* dst_type = dst->Value(&_gvn);
5209   const TypeAryPtr* top_src = src_type->isa_aryptr();
5210   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5211   if (top_src  == NULL || top_src->klass()  == NULL ||
5212       top_dest == NULL || top_dest->klass() == NULL) {
5213     // failed array check
5214     return false;
5215   }
5216 
5217   // Figure out the size and type of the elements we will be copying.
5218   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5219   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5220   if (src_elem != T_CHAR || dst_elem != T_BYTE) {
5221     return false;
5222   }
5223   Node* src_start = array_element_address(src, src_offset, src_elem);
5224   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5225   // 'src_start' points to src array + scaled offset
5226   // 'dst_start' points to dst array + scaled offset
5227 
5228   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5229   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5230   enc = _gvn.transform(enc);
5231   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5232   set_memory(res_mem, mtype);
5233   set_result(enc);
5234   return true;
5235 }
5236 
5237 //-------------inline_multiplyToLen-----------------------------------
5238 bool LibraryCallKit::inline_multiplyToLen() {
5239   assert(UseMultiplyToLenIntrinsic, "not implementated on this platform");
5240 
5241   address stubAddr = StubRoutines::multiplyToLen();
5242   if (stubAddr == NULL) {
5243     return false; // Intrinsic's stub is not implemented on this platform
5244   }
5245   const char* stubName = "multiplyToLen";
5246 
5247   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5248 
5249   Node* x    = argument(1);
5250   Node* xlen = argument(2);
5251   Node* y    = argument(3);
5252   Node* ylen = argument(4);
5253   Node* z    = argument(5);
5254 
5255   const Type* x_type = x->Value(&_gvn);
5256   const Type* y_type = y->Value(&_gvn);
5257   const TypeAryPtr* top_x = x_type->isa_aryptr();
5258   const TypeAryPtr* top_y = y_type->isa_aryptr();
5259   if (top_x  == NULL || top_x->klass()  == NULL ||
5260       top_y == NULL || top_y->klass() == NULL) {
5261     // failed array check
5262     return false;
5263   }
5264 
5265   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5266   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5267   if (x_elem != T_INT || y_elem != T_INT) {
5268     return false;
5269   }
5270 
5271   // Set the original stack and the reexecute bit for the interpreter to reexecute
5272   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5273   // on the return from z array allocation in runtime.
5274   { PreserveReexecuteState preexecs(this);
5275     jvms()->set_should_reexecute(true);
5276 
5277     Node* x_start = array_element_address(x, intcon(0), x_elem);
5278     Node* y_start = array_element_address(y, intcon(0), y_elem);
5279     // 'x_start' points to x array + scaled xlen
5280     // 'y_start' points to y array + scaled ylen
5281 
5282     // Allocate the result array
5283     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5284     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5285     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5286 
5287     IdealKit ideal(this);
5288 
5289 #define __ ideal.
5290      Node* one = __ ConI(1);
5291      Node* zero = __ ConI(0);
5292      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5293      __ set(need_alloc, zero);
5294      __ set(z_alloc, z);
5295      __ if_then(z, BoolTest::eq, null()); {
5296        __ increment (need_alloc, one);
5297      } __ else_(); {
5298        // Update graphKit memory and control from IdealKit.
5299        sync_kit(ideal);
5300        Node* zlen_arg = load_array_length(z);
5301        // Update IdealKit memory and control from graphKit.
5302        __ sync_kit(this);
5303        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5304          __ increment (need_alloc, one);
5305        } __ end_if();
5306      } __ end_if();
5307 
5308      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5309        // Update graphKit memory and control from IdealKit.
5310        sync_kit(ideal);
5311        Node * narr = new_array(klass_node, zlen, 1);
5312        // Update IdealKit memory and control from graphKit.
5313        __ sync_kit(this);
5314        __ set(z_alloc, narr);
5315      } __ end_if();
5316 
5317      sync_kit(ideal);
5318      z = __ value(z_alloc);
5319      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5320      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5321      // Final sync IdealKit and GraphKit.
5322      final_sync(ideal);
5323 #undef __
5324 
5325     Node* z_start = array_element_address(z, intcon(0), T_INT);
5326 
5327     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5328                                    OptoRuntime::multiplyToLen_Type(),
5329                                    stubAddr, stubName, TypePtr::BOTTOM,
5330                                    x_start, xlen, y_start, ylen, z_start, zlen);
5331   } // original reexecute is set back here
5332 
5333   C->set_has_split_ifs(true); // Has chance for split-if optimization
5334   set_result(z);
5335   return true;
5336 }
5337 
5338 //-------------inline_squareToLen------------------------------------
5339 bool LibraryCallKit::inline_squareToLen() {
5340   assert(UseSquareToLenIntrinsic, "not implementated on this platform");
5341 
5342   address stubAddr = StubRoutines::squareToLen();
5343   if (stubAddr == NULL) {
5344     return false; // Intrinsic's stub is not implemented on this platform
5345   }
5346   const char* stubName = "squareToLen";
5347 
5348   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5349 
5350   Node* x    = argument(0);
5351   Node* len  = argument(1);
5352   Node* z    = argument(2);
5353   Node* zlen = argument(3);
5354 
5355   const Type* x_type = x->Value(&_gvn);
5356   const Type* z_type = z->Value(&_gvn);
5357   const TypeAryPtr* top_x = x_type->isa_aryptr();
5358   const TypeAryPtr* top_z = z_type->isa_aryptr();
5359   if (top_x  == NULL || top_x->klass()  == NULL ||
5360       top_z  == NULL || top_z->klass()  == NULL) {
5361     // failed array check
5362     return false;
5363   }
5364 
5365   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5366   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5367   if (x_elem != T_INT || z_elem != T_INT) {
5368     return false;
5369   }
5370 
5371 
5372   Node* x_start = array_element_address(x, intcon(0), x_elem);
5373   Node* z_start = array_element_address(z, intcon(0), z_elem);
5374 
5375   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5376                                   OptoRuntime::squareToLen_Type(),
5377                                   stubAddr, stubName, TypePtr::BOTTOM,
5378                                   x_start, len, z_start, zlen);
5379 
5380   set_result(z);
5381   return true;
5382 }
5383 
5384 //-------------inline_mulAdd------------------------------------------
5385 bool LibraryCallKit::inline_mulAdd() {
5386   assert(UseMulAddIntrinsic, "not implementated on this platform");
5387 
5388   address stubAddr = StubRoutines::mulAdd();
5389   if (stubAddr == NULL) {
5390     return false; // Intrinsic's stub is not implemented on this platform
5391   }
5392   const char* stubName = "mulAdd";
5393 
5394   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5395 
5396   Node* out      = argument(0);
5397   Node* in       = argument(1);
5398   Node* offset   = argument(2);
5399   Node* len      = argument(3);
5400   Node* k        = argument(4);
5401 
5402   const Type* out_type = out->Value(&_gvn);
5403   const Type* in_type = in->Value(&_gvn);
5404   const TypeAryPtr* top_out = out_type->isa_aryptr();
5405   const TypeAryPtr* top_in = in_type->isa_aryptr();
5406   if (top_out  == NULL || top_out->klass()  == NULL ||
5407       top_in == NULL || top_in->klass() == NULL) {
5408     // failed array check
5409     return false;
5410   }
5411 
5412   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5413   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5414   if (out_elem != T_INT || in_elem != T_INT) {
5415     return false;
5416   }
5417 
5418   Node* outlen = load_array_length(out);
5419   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5420   Node* out_start = array_element_address(out, intcon(0), out_elem);
5421   Node* in_start = array_element_address(in, intcon(0), in_elem);
5422 
5423   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5424                                   OptoRuntime::mulAdd_Type(),
5425                                   stubAddr, stubName, TypePtr::BOTTOM,
5426                                   out_start,in_start, new_offset, len, k);
5427   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5428   set_result(result);
5429   return true;
5430 }
5431 
5432 
5433 /**
5434  * Calculate CRC32 for byte.
5435  * int java.util.zip.CRC32.update(int crc, int b)
5436  */
5437 bool LibraryCallKit::inline_updateCRC32() {
5438   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5439   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5440   // no receiver since it is static method
5441   Node* crc  = argument(0); // type: int
5442   Node* b    = argument(1); // type: int
5443 
5444   /*
5445    *    int c = ~ crc;
5446    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5447    *    b = b ^ (c >>> 8);
5448    *    crc = ~b;
5449    */
5450 
5451   Node* M1 = intcon(-1);
5452   crc = _gvn.transform(new XorINode(crc, M1));
5453   Node* result = _gvn.transform(new XorINode(crc, b));
5454   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5455 
5456   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5457   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5458   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5459   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5460 
5461   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5462   result = _gvn.transform(new XorINode(crc, result));
5463   result = _gvn.transform(new XorINode(result, M1));
5464   set_result(result);
5465   return true;
5466 }
5467 
5468 /**
5469  * Calculate CRC32 for byte[] array.
5470  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5471  */
5472 bool LibraryCallKit::inline_updateBytesCRC32() {
5473   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5474   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5475   // no receiver since it is static method
5476   Node* crc     = argument(0); // type: int
5477   Node* src     = argument(1); // type: oop
5478   Node* offset  = argument(2); // type: int
5479   Node* length  = argument(3); // type: int
5480 
5481   const Type* src_type = src->Value(&_gvn);
5482   const TypeAryPtr* top_src = src_type->isa_aryptr();
5483   if (top_src  == NULL || top_src->klass()  == NULL) {
5484     // failed array check
5485     return false;
5486   }
5487 
5488   // Figure out the size and type of the elements we will be copying.
5489   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5490   if (src_elem != T_BYTE) {
5491     return false;
5492   }
5493 
5494   // 'src_start' points to src array + scaled offset
5495   Node* src_start = array_element_address(src, offset, src_elem);
5496 
5497   // We assume that range check is done by caller.
5498   // TODO: generate range check (offset+length < src.length) in debug VM.
5499 
5500   // Call the stub.
5501   address stubAddr = StubRoutines::updateBytesCRC32();
5502   const char *stubName = "updateBytesCRC32";
5503 
5504   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5505                                  stubAddr, stubName, TypePtr::BOTTOM,
5506                                  crc, src_start, length);
5507   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5508   set_result(result);
5509   return true;
5510 }
5511 
5512 /**
5513  * Calculate CRC32 for ByteBuffer.
5514  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5515  */
5516 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5517   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5518   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5519   // no receiver since it is static method
5520   Node* crc     = argument(0); // type: int
5521   Node* src     = argument(1); // type: long
5522   Node* offset  = argument(3); // type: int
5523   Node* length  = argument(4); // type: int
5524 
5525   src = ConvL2X(src);  // adjust Java long to machine word
5526   Node* base = _gvn.transform(new CastX2PNode(src));
5527   offset = ConvI2X(offset);
5528 
5529   // 'src_start' points to src array + scaled offset
5530   Node* src_start = basic_plus_adr(top(), base, offset);
5531 
5532   // Call the stub.
5533   address stubAddr = StubRoutines::updateBytesCRC32();
5534   const char *stubName = "updateBytesCRC32";
5535 
5536   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5537                                  stubAddr, stubName, TypePtr::BOTTOM,
5538                                  crc, src_start, length);
5539   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5540   set_result(result);
5541   return true;
5542 }
5543 
5544 //----------------------------inline_reference_get----------------------------
5545 // public T java.lang.ref.Reference.get();
5546 bool LibraryCallKit::inline_reference_get() {
5547   const int referent_offset = java_lang_ref_Reference::referent_offset;
5548   guarantee(referent_offset > 0, "should have already been set");
5549 
5550   // Get the argument:
5551   Node* reference_obj = null_check_receiver();
5552   if (stopped()) return true;
5553 
5554   Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5555 
5556   ciInstanceKlass* klass = env()->Object_klass();
5557   const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5558 
5559   Node* no_ctrl = NULL;
5560   Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5561 
5562   // Use the pre-barrier to record the value in the referent field
5563   pre_barrier(false /* do_load */,
5564               control(),
5565               NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5566               result /* pre_val */,
5567               T_OBJECT);
5568 
5569   // Add memory barrier to prevent commoning reads from this field
5570   // across safepoint since GC can change its value.
5571   insert_mem_bar(Op_MemBarCPUOrder);
5572 
5573   set_result(result);
5574   return true;
5575 }
5576 
5577 
5578 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5579                                               bool is_exact=true, bool is_static=false) {
5580 
5581   const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5582   assert(tinst != NULL, "obj is null");
5583   assert(tinst->klass()->is_loaded(), "obj is not loaded");
5584   assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5585 
5586   ciField* field = tinst->klass()->as_instance_klass()->get_field_by_name(ciSymbol::make(fieldName),
5587                                                                           ciSymbol::make(fieldTypeString),
5588                                                                           is_static);
5589   if (field == NULL) return (Node *) NULL;
5590   assert (field != NULL, "undefined field");
5591 
5592   // Next code  copied from Parse::do_get_xxx():
5593 
5594   // Compute address and memory type.
5595   int offset  = field->offset_in_bytes();
5596   bool is_vol = field->is_volatile();
5597   ciType* field_klass = field->type();
5598   assert(field_klass->is_loaded(), "should be loaded");
5599   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5600   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5601   BasicType bt = field->layout_type();
5602 
5603   // Build the resultant type of the load
5604   const Type *type;
5605   if (bt == T_OBJECT) {
5606     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5607   } else {
5608     type = Type::get_const_basic_type(bt);
5609   }
5610 
5611   if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5612     insert_mem_bar(Op_MemBarVolatile);   // StoreLoad barrier
5613   }
5614   // Build the load.
5615   MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5616   Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5617   // If reference is volatile, prevent following memory ops from
5618   // floating up past the volatile read.  Also prevents commoning
5619   // another volatile read.
5620   if (is_vol) {
5621     // Memory barrier includes bogus read of value to force load BEFORE membar
5622     insert_mem_bar(Op_MemBarAcquire, loadedField);
5623   }
5624   return loadedField;
5625 }
5626 
5627 
5628 //------------------------------inline_aescrypt_Block-----------------------
5629 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5630   address stubAddr;
5631   const char *stubName;
5632   assert(UseAES, "need AES instruction support");
5633 
5634   switch(id) {
5635   case vmIntrinsics::_aescrypt_encryptBlock:
5636     stubAddr = StubRoutines::aescrypt_encryptBlock();
5637     stubName = "aescrypt_encryptBlock";
5638     break;
5639   case vmIntrinsics::_aescrypt_decryptBlock:
5640     stubAddr = StubRoutines::aescrypt_decryptBlock();
5641     stubName = "aescrypt_decryptBlock";
5642     break;
5643   }
5644   if (stubAddr == NULL) return false;
5645 
5646   Node* aescrypt_object = argument(0);
5647   Node* src             = argument(1);
5648   Node* src_offset      = argument(2);
5649   Node* dest            = argument(3);
5650   Node* dest_offset     = argument(4);
5651 
5652   // (1) src and dest are arrays.
5653   const Type* src_type = src->Value(&_gvn);
5654   const Type* dest_type = dest->Value(&_gvn);
5655   const TypeAryPtr* top_src = src_type->isa_aryptr();
5656   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5657   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5658 
5659   // for the quick and dirty code we will skip all the checks.
5660   // we are just trying to get the call to be generated.
5661   Node* src_start  = src;
5662   Node* dest_start = dest;
5663   if (src_offset != NULL || dest_offset != NULL) {
5664     assert(src_offset != NULL && dest_offset != NULL, "");
5665     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5666     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5667   }
5668 
5669   // now need to get the start of its expanded key array
5670   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5671   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5672   if (k_start == NULL) return false;
5673 
5674   if (Matcher::pass_original_key_for_aes()) {
5675     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5676     // compatibility issues between Java key expansion and SPARC crypto instructions
5677     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5678     if (original_k_start == NULL) return false;
5679 
5680     // Call the stub.
5681     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5682                       stubAddr, stubName, TypePtr::BOTTOM,
5683                       src_start, dest_start, k_start, original_k_start);
5684   } else {
5685     // Call the stub.
5686     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5687                       stubAddr, stubName, TypePtr::BOTTOM,
5688                       src_start, dest_start, k_start);
5689   }
5690 
5691   return true;
5692 }
5693 
5694 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5695 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5696   address stubAddr;
5697   const char *stubName;
5698 
5699   assert(UseAES, "need AES instruction support");
5700 
5701   switch(id) {
5702   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5703     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5704     stubName = "cipherBlockChaining_encryptAESCrypt";
5705     break;
5706   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5707     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5708     stubName = "cipherBlockChaining_decryptAESCrypt";
5709     break;
5710   }
5711   if (stubAddr == NULL) return false;
5712 
5713   Node* cipherBlockChaining_object = argument(0);
5714   Node* src                        = argument(1);
5715   Node* src_offset                 = argument(2);
5716   Node* len                        = argument(3);
5717   Node* dest                       = argument(4);
5718   Node* dest_offset                = argument(5);
5719 
5720   // (1) src and dest are arrays.
5721   const Type* src_type = src->Value(&_gvn);
5722   const Type* dest_type = dest->Value(&_gvn);
5723   const TypeAryPtr* top_src = src_type->isa_aryptr();
5724   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5725   assert (top_src  != NULL && top_src->klass()  != NULL
5726           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5727 
5728   // checks are the responsibility of the caller
5729   Node* src_start  = src;
5730   Node* dest_start = dest;
5731   if (src_offset != NULL || dest_offset != NULL) {
5732     assert(src_offset != NULL && dest_offset != NULL, "");
5733     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5734     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5735   }
5736 
5737   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5738   // (because of the predicated logic executed earlier).
5739   // so we cast it here safely.
5740   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5741 
5742   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5743   if (embeddedCipherObj == NULL) return false;
5744 
5745   // cast it to what we know it will be at runtime
5746   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5747   assert(tinst != NULL, "CBC obj is null");
5748   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5749   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5750   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5751 
5752   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5753   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5754   const TypeOopPtr* xtype = aklass->as_instance_type();
5755   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5756   aescrypt_object = _gvn.transform(aescrypt_object);
5757 
5758   // we need to get the start of the aescrypt_object's expanded key array
5759   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5760   if (k_start == NULL) return false;
5761 
5762   // similarly, get the start address of the r vector
5763   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5764   if (objRvec == NULL) return false;
5765   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5766 
5767   Node* cbcCrypt;
5768   if (Matcher::pass_original_key_for_aes()) {
5769     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5770     // compatibility issues between Java key expansion and SPARC crypto instructions
5771     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5772     if (original_k_start == NULL) return false;
5773 
5774     // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5775     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5776                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5777                                  stubAddr, stubName, TypePtr::BOTTOM,
5778                                  src_start, dest_start, k_start, r_start, len, original_k_start);
5779   } else {
5780     // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5781     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5782                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5783                                  stubAddr, stubName, TypePtr::BOTTOM,
5784                                  src_start, dest_start, k_start, r_start, len);
5785   }
5786 
5787   // return cipher length (int)
5788   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5789   set_result(retvalue);
5790   return true;
5791 }
5792 
5793 //------------------------------get_key_start_from_aescrypt_object-----------------------
5794 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
5795   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
5796   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5797   if (objAESCryptKey == NULL) return (Node *) NULL;
5798 
5799   // now have the array, need to get the start address of the K array
5800   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
5801   return k_start;
5802 }
5803 
5804 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
5805 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
5806   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
5807   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5808   if (objAESCryptKey == NULL) return (Node *) NULL;
5809 
5810   // now have the array, need to get the start address of the lastKey array
5811   Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
5812   return original_k_start;
5813 }
5814 
5815 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
5816 // Return node representing slow path of predicate check.
5817 // the pseudo code we want to emulate with this predicate is:
5818 // for encryption:
5819 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
5820 // for decryption:
5821 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
5822 //    note cipher==plain is more conservative than the original java code but that's OK
5823 //
5824 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
5825   // The receiver was checked for NULL already.
5826   Node* objCBC = argument(0);
5827 
5828   // Load embeddedCipher field of CipherBlockChaining object.
5829   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5830 
5831   // get AESCrypt klass for instanceOf check
5832   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
5833   // will have same classloader as CipherBlockChaining object
5834   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
5835   assert(tinst != NULL, "CBCobj is null");
5836   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
5837 
5838   // we want to do an instanceof comparison against the AESCrypt class
5839   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5840   if (!klass_AESCrypt->is_loaded()) {
5841     // if AESCrypt is not even loaded, we never take the intrinsic fast path
5842     Node* ctrl = control();
5843     set_control(top()); // no regular fast path
5844     return ctrl;
5845   }
5846   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5847 
5848   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
5849   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
5850   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
5851 
5852   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
5853 
5854   // for encryption, we are done
5855   if (!decrypting)
5856     return instof_false;  // even if it is NULL
5857 
5858   // for decryption, we need to add a further check to avoid
5859   // taking the intrinsic path when cipher and plain are the same
5860   // see the original java code for why.
5861   RegionNode* region = new RegionNode(3);
5862   region->init_req(1, instof_false);
5863   Node* src = argument(1);
5864   Node* dest = argument(4);
5865   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
5866   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
5867   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
5868   region->init_req(2, src_dest_conjoint);
5869 
5870   record_for_igvn(region);
5871   return _gvn.transform(region);
5872 }
5873 
5874 //------------------------------inline_sha_implCompress-----------------------
5875 //
5876 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
5877 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
5878 //
5879 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
5880 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
5881 //
5882 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
5883 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
5884 //
5885 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
5886   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
5887 
5888   Node* sha_obj = argument(0);
5889   Node* src     = argument(1); // type oop
5890   Node* ofs     = argument(2); // type int
5891 
5892   const Type* src_type = src->Value(&_gvn);
5893   const TypeAryPtr* top_src = src_type->isa_aryptr();
5894   if (top_src  == NULL || top_src->klass()  == NULL) {
5895     // failed array check
5896     return false;
5897   }
5898   // Figure out the size and type of the elements we will be copying.
5899   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5900   if (src_elem != T_BYTE) {
5901     return false;
5902   }
5903   // 'src_start' points to src array + offset
5904   Node* src_start = array_element_address(src, ofs, src_elem);
5905   Node* state = NULL;
5906   address stubAddr;
5907   const char *stubName;
5908 
5909   switch(id) {
5910   case vmIntrinsics::_sha_implCompress:
5911     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
5912     state = get_state_from_sha_object(sha_obj);
5913     stubAddr = StubRoutines::sha1_implCompress();
5914     stubName = "sha1_implCompress";
5915     break;
5916   case vmIntrinsics::_sha2_implCompress:
5917     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
5918     state = get_state_from_sha_object(sha_obj);
5919     stubAddr = StubRoutines::sha256_implCompress();
5920     stubName = "sha256_implCompress";
5921     break;
5922   case vmIntrinsics::_sha5_implCompress:
5923     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
5924     state = get_state_from_sha5_object(sha_obj);
5925     stubAddr = StubRoutines::sha512_implCompress();
5926     stubName = "sha512_implCompress";
5927     break;
5928   default:
5929     fatal_unexpected_iid(id);
5930     return false;
5931   }
5932   if (state == NULL) return false;
5933 
5934   // Call the stub.
5935   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
5936                                  stubAddr, stubName, TypePtr::BOTTOM,
5937                                  src_start, state);
5938 
5939   return true;
5940 }
5941 
5942 //------------------------------inline_digestBase_implCompressMB-----------------------
5943 //
5944 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
5945 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
5946 //
5947 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
5948   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
5949          "need SHA1/SHA256/SHA512 instruction support");
5950   assert((uint)predicate < 3, "sanity");
5951   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
5952 
5953   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
5954   Node* src            = argument(1); // byte[] array
5955   Node* ofs            = argument(2); // type int
5956   Node* limit          = argument(3); // type int
5957 
5958   const Type* src_type = src->Value(&_gvn);
5959   const TypeAryPtr* top_src = src_type->isa_aryptr();
5960   if (top_src  == NULL || top_src->klass()  == NULL) {
5961     // failed array check
5962     return false;
5963   }
5964   // Figure out the size and type of the elements we will be copying.
5965   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5966   if (src_elem != T_BYTE) {
5967     return false;
5968   }
5969   // 'src_start' points to src array + offset
5970   Node* src_start = array_element_address(src, ofs, src_elem);
5971 
5972   const char* klass_SHA_name = NULL;
5973   const char* stub_name = NULL;
5974   address     stub_addr = NULL;
5975   bool        long_state = false;
5976 
5977   switch (predicate) {
5978   case 0:
5979     if (UseSHA1Intrinsics) {
5980       klass_SHA_name = "sun/security/provider/SHA";
5981       stub_name = "sha1_implCompressMB";
5982       stub_addr = StubRoutines::sha1_implCompressMB();
5983     }
5984     break;
5985   case 1:
5986     if (UseSHA256Intrinsics) {
5987       klass_SHA_name = "sun/security/provider/SHA2";
5988       stub_name = "sha256_implCompressMB";
5989       stub_addr = StubRoutines::sha256_implCompressMB();
5990     }
5991     break;
5992   case 2:
5993     if (UseSHA512Intrinsics) {
5994       klass_SHA_name = "sun/security/provider/SHA5";
5995       stub_name = "sha512_implCompressMB";
5996       stub_addr = StubRoutines::sha512_implCompressMB();
5997       long_state = true;
5998     }
5999     break;
6000   default:
6001     fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
6002   }
6003   if (klass_SHA_name != NULL) {
6004     // get DigestBase klass to lookup for SHA klass
6005     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6006     assert(tinst != NULL, "digestBase_obj is not instance???");
6007     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6008 
6009     ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6010     assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6011     ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6012     return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6013   }
6014   return false;
6015 }
6016 //------------------------------inline_sha_implCompressMB-----------------------
6017 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6018                                                bool long_state, address stubAddr, const char *stubName,
6019                                                Node* src_start, Node* ofs, Node* limit) {
6020   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6021   const TypeOopPtr* xtype = aklass->as_instance_type();
6022   Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6023   sha_obj = _gvn.transform(sha_obj);
6024 
6025   Node* state;
6026   if (long_state) {
6027     state = get_state_from_sha5_object(sha_obj);
6028   } else {
6029     state = get_state_from_sha_object(sha_obj);
6030   }
6031   if (state == NULL) return false;
6032 
6033   // Call the stub.
6034   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6035                                  OptoRuntime::digestBase_implCompressMB_Type(),
6036                                  stubAddr, stubName, TypePtr::BOTTOM,
6037                                  src_start, state, ofs, limit);
6038   // return ofs (int)
6039   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6040   set_result(result);
6041 
6042   return true;
6043 }
6044 
6045 //------------------------------get_state_from_sha_object-----------------------
6046 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6047   Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6048   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6049   if (sha_state == NULL) return (Node *) NULL;
6050 
6051   // now have the array, need to get the start address of the state array
6052   Node* state = array_element_address(sha_state, intcon(0), T_INT);
6053   return state;
6054 }
6055 
6056 //------------------------------get_state_from_sha5_object-----------------------
6057 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6058   Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6059   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6060   if (sha_state == NULL) return (Node *) NULL;
6061 
6062   // now have the array, need to get the start address of the state array
6063   Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6064   return state;
6065 }
6066 
6067 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6068 // Return node representing slow path of predicate check.
6069 // the pseudo code we want to emulate with this predicate is:
6070 //    if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6071 //
6072 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6073   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6074          "need SHA1/SHA256/SHA512 instruction support");
6075   assert((uint)predicate < 3, "sanity");
6076 
6077   // The receiver was checked for NULL already.
6078   Node* digestBaseObj = argument(0);
6079 
6080   // get DigestBase klass for instanceOf check
6081   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6082   assert(tinst != NULL, "digestBaseObj is null");
6083   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6084 
6085   const char* klass_SHA_name = NULL;
6086   switch (predicate) {
6087   case 0:
6088     if (UseSHA1Intrinsics) {
6089       // we want to do an instanceof comparison against the SHA class
6090       klass_SHA_name = "sun/security/provider/SHA";
6091     }
6092     break;
6093   case 1:
6094     if (UseSHA256Intrinsics) {
6095       // we want to do an instanceof comparison against the SHA2 class
6096       klass_SHA_name = "sun/security/provider/SHA2";
6097     }
6098     break;
6099   case 2:
6100     if (UseSHA512Intrinsics) {
6101       // we want to do an instanceof comparison against the SHA5 class
6102       klass_SHA_name = "sun/security/provider/SHA5";
6103     }
6104     break;
6105   default:
6106     fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
6107   }
6108 
6109   ciKlass* klass_SHA = NULL;
6110   if (klass_SHA_name != NULL) {
6111     klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6112   }
6113   if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6114     // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6115     Node* ctrl = control();
6116     set_control(top()); // no intrinsic path
6117     return ctrl;
6118   }
6119   ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6120 
6121   Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6122   Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6123   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6124   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6125 
6126   return instof_false;  // even if it is NULL
6127 }
6128 
6129 bool LibraryCallKit::inline_profileBoolean() {
6130   Node* counts = argument(1);
6131   const TypeAryPtr* ary = NULL;
6132   ciArray* aobj = NULL;
6133   if (counts->is_Con()
6134       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6135       && (aobj = ary->const_oop()->as_array()) != NULL
6136       && (aobj->length() == 2)) {
6137     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6138     jint false_cnt = aobj->element_value(0).as_int();
6139     jint  true_cnt = aobj->element_value(1).as_int();
6140 
6141     if (C->log() != NULL) {
6142       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6143                      false_cnt, true_cnt);
6144     }
6145 
6146     if (false_cnt + true_cnt == 0) {
6147       // According to profile, never executed.
6148       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6149                           Deoptimization::Action_reinterpret);
6150       return true;
6151     }
6152 
6153     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6154     // is a number of each value occurrences.
6155     Node* result = argument(0);
6156     if (false_cnt == 0 || true_cnt == 0) {
6157       // According to profile, one value has been never seen.
6158       int expected_val = (false_cnt == 0) ? 1 : 0;
6159 
6160       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6161       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6162 
6163       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6164       Node* fast_path = _gvn.transform(new IfTrueNode(check));
6165       Node* slow_path = _gvn.transform(new IfFalseNode(check));
6166 
6167       { // Slow path: uncommon trap for never seen value and then reexecute
6168         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6169         // the value has been seen at least once.
6170         PreserveJVMState pjvms(this);
6171         PreserveReexecuteState preexecs(this);
6172         jvms()->set_should_reexecute(true);
6173 
6174         set_control(slow_path);
6175         set_i_o(i_o());
6176 
6177         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6178                             Deoptimization::Action_reinterpret);
6179       }
6180       // The guard for never seen value enables sharpening of the result and
6181       // returning a constant. It allows to eliminate branches on the same value
6182       // later on.
6183       set_control(fast_path);
6184       result = intcon(expected_val);
6185     }
6186     // Stop profiling.
6187     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6188     // By replacing method body with profile data (represented as ProfileBooleanNode
6189     // on IR level) we effectively disable profiling.
6190     // It enables full speed execution once optimized code is generated.
6191     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6192     C->record_for_igvn(profile);
6193     set_result(profile);
6194     return true;
6195   } else {
6196     // Continue profiling.
6197     // Profile data isn't available at the moment. So, execute method's bytecode version.
6198     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6199     // is compiled and counters aren't available since corresponding MethodHandle
6200     // isn't a compile-time constant.
6201     return false;
6202   }
6203 }
6204 
6205 bool LibraryCallKit::inline_isCompileConstant() {
6206   Node* n = argument(0);
6207   set_result(n->is_Con() ? intcon(1) : intcon(0));
6208   return true;
6209 }
--- EOF ---