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