1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "compiler/compileBroker.hpp"
  30 #include "compiler/compileLog.hpp"
  31 #include "memory/resourceArea.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "opto/addnode.hpp"
  34 #include "opto/arraycopynode.hpp"
  35 #include "opto/c2compiler.hpp"
  36 #include "opto/callGenerator.hpp"
  37 #include "opto/castnode.hpp"
  38 #include "opto/cfgnode.hpp"
  39 #include "opto/convertnode.hpp"
  40 #include "opto/countbitsnode.hpp"
  41 #include "opto/intrinsicnode.hpp"
  42 #include "opto/idealKit.hpp"
  43 #include "opto/mathexactnode.hpp"
  44 #include "opto/movenode.hpp"
  45 #include "opto/mulnode.hpp"
  46 #include "opto/narrowptrnode.hpp"
  47 #include "opto/opaquenode.hpp"
  48 #include "opto/parse.hpp"
  49 #include "opto/runtime.hpp"
  50 #include "opto/subnode.hpp"
  51 #include "prims/nativeLookup.hpp"
  52 #include "prims/unsafe.hpp"
  53 #include "runtime/sharedRuntime.hpp"
  54 #ifdef TRACE_HAVE_INTRINSICS
  55 #include "trace/traceMacros.hpp"
  56 #endif
  57 
  58 class LibraryIntrinsic : public InlineCallGenerator {
  59   // Extend the set of intrinsics known to the runtime:
  60  public:
  61  private:
  62   bool             _is_virtual;
  63   bool             _does_virtual_dispatch;
  64   int8_t           _predicates_count;  // Intrinsic is predicated by several conditions
  65   int8_t           _last_predicate; // Last generated predicate
  66   vmIntrinsics::ID _intrinsic_id;
  67 
  68  public:
  69   LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
  70     : InlineCallGenerator(m),
  71       _is_virtual(is_virtual),
  72       _does_virtual_dispatch(does_virtual_dispatch),
  73       _predicates_count((int8_t)predicates_count),
  74       _last_predicate((int8_t)-1),
  75       _intrinsic_id(id)
  76   {
  77   }
  78   virtual bool is_intrinsic() const { return true; }
  79   virtual bool is_virtual()   const { return _is_virtual; }
  80   virtual bool is_predicated() const { return _predicates_count > 0; }
  81   virtual int  predicates_count() const { return _predicates_count; }
  82   virtual bool does_virtual_dispatch()   const { return _does_virtual_dispatch; }
  83   virtual JVMState* generate(JVMState* jvms);
  84   virtual Node* generate_predicate(JVMState* jvms, int predicate);
  85   vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
  86 };
  87 
  88 
  89 // Local helper class for LibraryIntrinsic:
  90 class LibraryCallKit : public GraphKit {
  91  private:
  92   LibraryIntrinsic* _intrinsic;     // the library intrinsic being called
  93   Node*             _result;        // the result node, if any
  94   int               _reexecute_sp;  // the stack pointer when bytecode needs to be reexecuted
  95 
  96   const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);
  97 
  98  public:
  99   LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
 100     : GraphKit(jvms),
 101       _intrinsic(intrinsic),
 102       _result(NULL)
 103   {
 104     // Check if this is a root compile.  In that case we don't have a caller.
 105     if (!jvms->has_method()) {
 106       _reexecute_sp = sp();
 107     } else {
 108       // Find out how many arguments the interpreter needs when deoptimizing
 109       // and save the stack pointer value so it can used by uncommon_trap.
 110       // We find the argument count by looking at the declared signature.
 111       bool ignored_will_link;
 112       ciSignature* declared_signature = NULL;
 113       ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
 114       const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
 115       _reexecute_sp = sp() + nargs;  // "push" arguments back on stack
 116     }
 117   }
 118 
 119   virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
 120 
 121   ciMethod*         caller()    const    { return jvms()->method(); }
 122   int               bci()       const    { return jvms()->bci(); }
 123   LibraryIntrinsic* intrinsic() const    { return _intrinsic; }
 124   vmIntrinsics::ID  intrinsic_id() const { return _intrinsic->intrinsic_id(); }
 125   ciMethod*         callee()    const    { return _intrinsic->method(); }
 126 
 127   bool  try_to_inline(int predicate);
 128   Node* try_to_predicate(int predicate);
 129 
 130   void push_result() {
 131     // Push the result onto the stack.
 132     if (!stopped() && result() != NULL) {
 133       BasicType bt = result()->bottom_type()->basic_type();
 134       push_node(bt, result());
 135     }
 136   }
 137 
 138  private:
 139   void fatal_unexpected_iid(vmIntrinsics::ID iid) {
 140     fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
 141   }
 142 
 143   void  set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
 144   void  set_result(RegionNode* region, PhiNode* value);
 145   Node*     result() { return _result; }
 146 
 147   virtual int reexecute_sp() { return _reexecute_sp; }
 148 
 149   // Helper functions to inline natives
 150   Node* generate_guard(Node* test, RegionNode* region, float true_prob);
 151   Node* generate_slow_guard(Node* test, RegionNode* region);
 152   Node* generate_fair_guard(Node* test, RegionNode* region);
 153   Node* generate_negative_guard(Node* index, RegionNode* region,
 154                                 // resulting CastII of index:
 155                                 Node* *pos_index = NULL);
 156   Node* generate_limit_guard(Node* offset, Node* subseq_length,
 157                              Node* array_length,
 158                              RegionNode* region);
 159   void  generate_string_range_check(Node* array, Node* offset,
 160                                     Node* length, bool char_count);
 161   Node* generate_current_thread(Node* &tls_output);
 162   Node* load_mirror_from_klass(Node* klass);
 163   Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
 164                                       RegionNode* region, int null_path,
 165                                       int offset);
 166   Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
 167                                RegionNode* region, int null_path) {
 168     int offset = java_lang_Class::klass_offset_in_bytes();
 169     return load_klass_from_mirror_common(mirror, never_see_null,
 170                                          region, null_path,
 171                                          offset);
 172   }
 173   Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
 174                                      RegionNode* region, int null_path) {
 175     int offset = java_lang_Class::array_klass_offset_in_bytes();
 176     return load_klass_from_mirror_common(mirror, never_see_null,
 177                                          region, null_path,
 178                                          offset);
 179   }
 180   Node* generate_access_flags_guard(Node* kls,
 181                                     int modifier_mask, int modifier_bits,
 182                                     RegionNode* region);
 183   Node* generate_interface_guard(Node* kls, RegionNode* region);
 184   Node* generate_array_guard(Node* kls, RegionNode* region) {
 185     return generate_array_guard_common(kls, region, false, false);
 186   }
 187   Node* generate_non_array_guard(Node* kls, RegionNode* region) {
 188     return generate_array_guard_common(kls, region, false, true);
 189   }
 190   Node* generate_objArray_guard(Node* kls, RegionNode* region) {
 191     return generate_array_guard_common(kls, region, true, false);
 192   }
 193   Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
 194     return generate_array_guard_common(kls, region, true, true);
 195   }
 196   Node* generate_array_guard_common(Node* kls, RegionNode* region,
 197                                     bool obj_array, bool not_array);
 198   Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
 199   CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
 200                                      bool is_virtual = false, bool is_static = false);
 201   CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
 202     return generate_method_call(method_id, false, true);
 203   }
 204   CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
 205     return generate_method_call(method_id, true, false);
 206   }
 207   Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
 208   Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
 209 
 210   Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
 211   bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
 212   bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
 213   bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
 214   Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
 215                           RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
 216   bool inline_string_indexOfChar();
 217   bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
 218   bool inline_string_toBytesU();
 219   bool inline_string_getCharsU();
 220   bool inline_string_copy(bool compress);
 221   bool inline_string_char_access(bool is_store);
 222   Node* round_double_node(Node* n);
 223   bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
 224   bool inline_math_native(vmIntrinsics::ID id);
 225   bool inline_trig(vmIntrinsics::ID id);
 226   bool inline_math(vmIntrinsics::ID id);
 227   template <typename OverflowOp>
 228   bool inline_math_overflow(Node* arg1, Node* arg2);
 229   void inline_math_mathExact(Node* math, Node* test);
 230   bool inline_math_addExactI(bool is_increment);
 231   bool inline_math_addExactL(bool is_increment);
 232   bool inline_math_multiplyExactI();
 233   bool inline_math_multiplyExactL();
 234   bool inline_math_negateExactI();
 235   bool inline_math_negateExactL();
 236   bool inline_math_subtractExactI(bool is_decrement);
 237   bool inline_math_subtractExactL(bool is_decrement);
 238   bool inline_min_max(vmIntrinsics::ID id);
 239   bool inline_notify(vmIntrinsics::ID id);
 240   Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
 241   // This returns Type::AnyPtr, RawPtr, or OopPtr.
 242   int classify_unsafe_addr(Node* &base, Node* &offset);
 243   Node* make_unsafe_address(Node* base, Node* offset);
 244   // Helper for inline_unsafe_access.
 245   // Generates the guards that check whether the result of
 246   // Unsafe.getObject should be recorded in an SATB log buffer.
 247   void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
 248 
 249   typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
 250   bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
 251   static bool klass_needs_init_guard(Node* kls);
 252   bool inline_unsafe_allocate();
 253   bool inline_unsafe_newArray(bool uninitialized);
 254   bool inline_unsafe_copyMemory();
 255   bool inline_native_currentThread();
 256 
 257   bool inline_native_time_funcs(address method, const char* funcName);
 258   bool inline_native_isInterrupted();
 259   bool inline_native_Class_query(vmIntrinsics::ID id);
 260   bool inline_native_subtype_check();
 261   bool inline_native_getLength();
 262   bool inline_array_copyOf(bool is_copyOfRange);
 263   bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
 264   bool inline_objects_checkIndex();
 265   void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
 266   bool inline_native_clone(bool is_virtual);
 267   bool inline_native_Reflection_getCallerClass();
 268   // Helper function for inlining native object hash method
 269   bool inline_native_hashcode(bool is_virtual, bool is_static);
 270   bool inline_native_getClass();
 271 
 272   // Helper functions for inlining arraycopy
 273   bool inline_arraycopy();
 274   AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
 275                                                 RegionNode* slow_region);
 276   JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
 277   void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
 278 
 279   typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
 280   MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
 281   MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
 282   bool inline_unsafe_load_store(BasicType type,  LoadStoreKind kind, AccessKind access_kind);
 283   bool inline_unsafe_fence(vmIntrinsics::ID id);
 284   bool inline_onspinwait();
 285   bool inline_fp_conversions(vmIntrinsics::ID id);
 286   bool inline_number_methods(vmIntrinsics::ID id);
 287   bool inline_reference_get();
 288   bool inline_Class_cast();
 289   bool inline_aescrypt_Block(vmIntrinsics::ID id);
 290   bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
 291   bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
 292   Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
 293   Node* inline_counterMode_AESCrypt_predicate();
 294   Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
 295   Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
 296   bool inline_ghash_processBlocks();
 297   bool inline_sha_implCompress(vmIntrinsics::ID id);
 298   bool inline_digestBase_implCompressMB(int predicate);
 299   bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
 300                                  bool long_state, address stubAddr, const char *stubName,
 301                                  Node* src_start, Node* ofs, Node* limit);
 302   Node* get_state_from_sha_object(Node *sha_object);
 303   Node* get_state_from_sha5_object(Node *sha_object);
 304   Node* inline_digestBase_implCompressMB_predicate(int predicate);
 305   bool inline_encodeISOArray();
 306   bool inline_updateCRC32();
 307   bool inline_updateBytesCRC32();
 308   bool inline_updateByteBufferCRC32();
 309   Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
 310   bool inline_updateBytesCRC32C();
 311   bool inline_updateDirectByteBufferCRC32C();
 312   bool inline_updateBytesAdler32();
 313   bool inline_updateByteBufferAdler32();
 314   bool inline_multiplyToLen();
 315   bool inline_hasNegatives();
 316   bool inline_squareToLen();
 317   bool inline_mulAdd();
 318   bool inline_montgomeryMultiply();
 319   bool inline_montgomerySquare();
 320   bool inline_vectorizedMismatch();
 321 
 322   bool inline_profileBoolean();
 323   bool inline_isCompileConstant();
 324 };
 325 
 326 //---------------------------make_vm_intrinsic----------------------------
 327 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
 328   vmIntrinsics::ID id = m->intrinsic_id();
 329   assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
 330 
 331   if (!m->is_loaded()) {
 332     // Do not attempt to inline unloaded methods.
 333     return NULL;
 334   }
 335 
 336   C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
 337   bool is_available = false;
 338 
 339   {
 340     // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
 341     // the compiler must transition to '_thread_in_vm' state because both
 342     // methods access VM-internal data.
 343     VM_ENTRY_MARK;
 344     methodHandle mh(THREAD, m->get_Method());
 345     is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
 346                    !C->directive()->is_intrinsic_disabled(mh) &&
 347                    !vmIntrinsics::is_disabled_by_flags(mh);
 348 
 349   }
 350 
 351   if (is_available) {
 352     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
 353     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
 354     return new LibraryIntrinsic(m, is_virtual,
 355                                 vmIntrinsics::predicates_needed(id),
 356                                 vmIntrinsics::does_virtual_dispatch(id),
 357                                 (vmIntrinsics::ID) id);
 358   } else {
 359     return NULL;
 360   }
 361 }
 362 
 363 //----------------------register_library_intrinsics-----------------------
 364 // Initialize this file's data structures, for each Compile instance.
 365 void Compile::register_library_intrinsics() {
 366   // Nothing to do here.
 367 }
 368 
 369 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
 370   LibraryCallKit kit(jvms, this);
 371   Compile* C = kit.C;
 372   int nodes = C->unique();
 373 #ifndef PRODUCT
 374   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 375     char buf[1000];
 376     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 377     tty->print_cr("Intrinsic %s", str);
 378   }
 379 #endif
 380   ciMethod* callee = kit.callee();
 381   const int bci    = kit.bci();
 382 
 383   // Try to inline the intrinsic.
 384   if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
 385       kit.try_to_inline(_last_predicate)) {
 386     if (C->print_intrinsics() || C->print_inlining()) {
 387       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
 388     }
 389     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 390     if (C->log()) {
 391       C->log()->elem("intrinsic id='%s'%s nodes='%d'",
 392                      vmIntrinsics::name_at(intrinsic_id()),
 393                      (is_virtual() ? " virtual='1'" : ""),
 394                      C->unique() - nodes);
 395     }
 396     // Push the result from the inlined method onto the stack.
 397     kit.push_result();
 398     C->print_inlining_update(this);
 399     return kit.transfer_exceptions_into_jvms();
 400   }
 401 
 402   // The intrinsic bailed out
 403   if (C->print_intrinsics() || C->print_inlining()) {
 404     if (jvms->has_method()) {
 405       // Not a root compile.
 406       const char* msg;
 407       if (callee->intrinsic_candidate()) {
 408         msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
 409       } else {
 410         msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
 411                            : "failed to inline (intrinsic), method not annotated";
 412       }
 413       C->print_inlining(callee, jvms->depth() - 1, bci, msg);
 414     } else {
 415       // Root compile
 416       tty->print("Did not generate intrinsic %s%s at bci:%d in",
 417                vmIntrinsics::name_at(intrinsic_id()),
 418                (is_virtual() ? " (virtual)" : ""), bci);
 419     }
 420   }
 421   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 422   C->print_inlining_update(this);
 423   return NULL;
 424 }
 425 
 426 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
 427   LibraryCallKit kit(jvms, this);
 428   Compile* C = kit.C;
 429   int nodes = C->unique();
 430   _last_predicate = predicate;
 431 #ifndef PRODUCT
 432   assert(is_predicated() && predicate < predicates_count(), "sanity");
 433   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 434     char buf[1000];
 435     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 436     tty->print_cr("Predicate for intrinsic %s", str);
 437   }
 438 #endif
 439   ciMethod* callee = kit.callee();
 440   const int bci    = kit.bci();
 441 
 442   Node* slow_ctl = kit.try_to_predicate(predicate);
 443   if (!kit.failing()) {
 444     if (C->print_intrinsics() || C->print_inlining()) {
 445       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
 446     }
 447     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 448     if (C->log()) {
 449       C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
 450                      vmIntrinsics::name_at(intrinsic_id()),
 451                      (is_virtual() ? " virtual='1'" : ""),
 452                      C->unique() - nodes);
 453     }
 454     return slow_ctl; // Could be NULL if the check folds.
 455   }
 456 
 457   // The intrinsic bailed out
 458   if (C->print_intrinsics() || C->print_inlining()) {
 459     if (jvms->has_method()) {
 460       // Not a root compile.
 461       const char* msg = "failed to generate predicate for intrinsic";
 462       C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
 463     } else {
 464       // Root compile
 465       C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
 466                                         vmIntrinsics::name_at(intrinsic_id()),
 467                                         (is_virtual() ? " (virtual)" : ""), bci);
 468     }
 469   }
 470   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 471   return NULL;
 472 }
 473 
 474 bool LibraryCallKit::try_to_inline(int predicate) {
 475   // Handle symbolic names for otherwise undistinguished boolean switches:
 476   const bool is_store       = true;
 477   const bool is_compress    = true;
 478   const bool is_native_ptr  = true;
 479   const bool is_static      = true;
 480   const bool is_volatile    = true;
 481 
 482   if (!jvms()->has_method()) {
 483     // Root JVMState has a null method.
 484     assert(map()->memory()->Opcode() == Op_Parm, "");
 485     // Insert the memory aliasing node
 486     set_all_memory(reset_memory());
 487   }
 488   assert(merged_memory(), "");
 489 
 490 
 491   switch (intrinsic_id()) {
 492   case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
 493   case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
 494   case vmIntrinsics::_getClass:                 return inline_native_getClass();
 495 
 496   case vmIntrinsics::_dsin:
 497   case vmIntrinsics::_dcos:
 498   case vmIntrinsics::_dtan:
 499   case vmIntrinsics::_dabs:
 500   case vmIntrinsics::_datan2:
 501   case vmIntrinsics::_dsqrt:
 502   case vmIntrinsics::_dexp:
 503   case vmIntrinsics::_dlog:
 504   case vmIntrinsics::_dlog10:
 505   case vmIntrinsics::_dpow:                     return inline_math_native(intrinsic_id());
 506 
 507   case vmIntrinsics::_min:
 508   case vmIntrinsics::_max:                      return inline_min_max(intrinsic_id());
 509 
 510   case vmIntrinsics::_notify:
 511   case vmIntrinsics::_notifyAll:
 512     if (InlineNotify) {
 513       return inline_notify(intrinsic_id());
 514     }
 515     return false;
 516 
 517   case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
 518   case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
 519   case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
 520   case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
 521   case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
 522   case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
 523   case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
 524   case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
 525   case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
 526   case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
 527   case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
 528   case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
 529 
 530   case vmIntrinsics::_arraycopy:                return inline_arraycopy();
 531 
 532   case vmIntrinsics::_compareToL:               return inline_string_compareTo(StrIntrinsicNode::LL);
 533   case vmIntrinsics::_compareToU:               return inline_string_compareTo(StrIntrinsicNode::UU);
 534   case vmIntrinsics::_compareToLU:              return inline_string_compareTo(StrIntrinsicNode::LU);
 535   case vmIntrinsics::_compareToUL:              return inline_string_compareTo(StrIntrinsicNode::UL);
 536 
 537   case vmIntrinsics::_indexOfL:                 return inline_string_indexOf(StrIntrinsicNode::LL);
 538   case vmIntrinsics::_indexOfU:                 return inline_string_indexOf(StrIntrinsicNode::UU);
 539   case vmIntrinsics::_indexOfUL:                return inline_string_indexOf(StrIntrinsicNode::UL);
 540   case vmIntrinsics::_indexOfIL:                return inline_string_indexOfI(StrIntrinsicNode::LL);
 541   case vmIntrinsics::_indexOfIU:                return inline_string_indexOfI(StrIntrinsicNode::UU);
 542   case vmIntrinsics::_indexOfIUL:               return inline_string_indexOfI(StrIntrinsicNode::UL);
 543   case vmIntrinsics::_indexOfU_char:            return inline_string_indexOfChar();
 544 
 545   case vmIntrinsics::_equalsL:                  return inline_string_equals(StrIntrinsicNode::LL);
 546   case vmIntrinsics::_equalsU:                  return inline_string_equals(StrIntrinsicNode::UU);
 547 
 548   case vmIntrinsics::_toBytesStringU:           return inline_string_toBytesU();
 549   case vmIntrinsics::_getCharsStringU:          return inline_string_getCharsU();
 550   case vmIntrinsics::_getCharStringU:           return inline_string_char_access(!is_store);
 551   case vmIntrinsics::_putCharStringU:           return inline_string_char_access( is_store);
 552 
 553   case vmIntrinsics::_compressStringC:
 554   case vmIntrinsics::_compressStringB:          return inline_string_copy( is_compress);
 555   case vmIntrinsics::_inflateStringC:
 556   case vmIntrinsics::_inflateStringB:           return inline_string_copy(!is_compress);
 557 
 558   case vmIntrinsics::_getObject:                return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   Relaxed, false);
 559   case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  Relaxed, false);
 560   case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     Relaxed, false);
 561   case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    Relaxed, false);
 562   case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     Relaxed, false);
 563   case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      Relaxed, false);
 564   case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     Relaxed, false);
 565   case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    Relaxed, false);
 566   case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   Relaxed, false);
 567 
 568   case vmIntrinsics::_putObject:                return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   Relaxed, false);
 569   case vmIntrinsics::_putBoolean:               return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  Relaxed, false);
 570   case vmIntrinsics::_putByte:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     Relaxed, false);
 571   case vmIntrinsics::_putShort:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    Relaxed, false);
 572   case vmIntrinsics::_putChar:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     Relaxed, false);
 573   case vmIntrinsics::_putInt:                   return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      Relaxed, false);
 574   case vmIntrinsics::_putLong:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     Relaxed, false);
 575   case vmIntrinsics::_putFloat:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    Relaxed, false);
 576   case vmIntrinsics::_putDouble:                return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   Relaxed, false);
 577 
 578   case vmIntrinsics::_getByte_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE,     Relaxed, false);
 579   case vmIntrinsics::_getShort_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT,    Relaxed, false);
 580   case vmIntrinsics::_getChar_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR,     Relaxed, false);
 581   case vmIntrinsics::_getInt_raw:               return inline_unsafe_access( is_native_ptr, !is_store, T_INT,      Relaxed, false);
 582   case vmIntrinsics::_getLong_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_LONG,     Relaxed, false);
 583   case vmIntrinsics::_getFloat_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT,    Relaxed, false);
 584   case vmIntrinsics::_getDouble_raw:            return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE,   Relaxed, false);
 585   case vmIntrinsics::_getAddress_raw:           return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS,  Relaxed, false);
 586 
 587   case vmIntrinsics::_putByte_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_BYTE,     Relaxed, false);
 588   case vmIntrinsics::_putShort_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_SHORT,    Relaxed, false);
 589   case vmIntrinsics::_putChar_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_CHAR,     Relaxed, false);
 590   case vmIntrinsics::_putInt_raw:               return inline_unsafe_access( is_native_ptr,  is_store, T_INT,      Relaxed, false);
 591   case vmIntrinsics::_putLong_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_LONG,     Relaxed, false);
 592   case vmIntrinsics::_putFloat_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_FLOAT,    Relaxed, false);
 593   case vmIntrinsics::_putDouble_raw:            return inline_unsafe_access( is_native_ptr,  is_store, T_DOUBLE,   Relaxed, false);
 594   case vmIntrinsics::_putAddress_raw:           return inline_unsafe_access( is_native_ptr,  is_store, T_ADDRESS,  Relaxed, false);
 595 
 596   case vmIntrinsics::_getObjectVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   Volatile, false);
 597   case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  Volatile, false);
 598   case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     Volatile, false);
 599   case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    Volatile, false);
 600   case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     Volatile, false);
 601   case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      Volatile, false);
 602   case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     Volatile, false);
 603   case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    Volatile, false);
 604   case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   Volatile, false);
 605 
 606   case vmIntrinsics::_putObjectVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   Volatile, false);
 607   case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  Volatile, false);
 608   case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     Volatile, false);
 609   case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    Volatile, false);
 610   case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     Volatile, false);
 611   case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      Volatile, false);
 612   case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     Volatile, false);
 613   case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    Volatile, false);
 614   case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   Volatile, false);
 615 
 616   case vmIntrinsics::_getShortUnaligned:        return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    Relaxed, true);
 617   case vmIntrinsics::_getCharUnaligned:         return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     Relaxed, true);
 618   case vmIntrinsics::_getIntUnaligned:          return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      Relaxed, true);
 619   case vmIntrinsics::_getLongUnaligned:         return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     Relaxed, true);
 620 
 621   case vmIntrinsics::_putShortUnaligned:        return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    Relaxed, true);
 622   case vmIntrinsics::_putCharUnaligned:         return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     Relaxed, true);
 623   case vmIntrinsics::_putIntUnaligned:          return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      Relaxed, true);
 624   case vmIntrinsics::_putLongUnaligned:         return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     Relaxed, true);
 625 
 626   case vmIntrinsics::_getObjectAcquire:         return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   Acquire, false);
 627   case vmIntrinsics::_getBooleanAcquire:        return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  Acquire, false);
 628   case vmIntrinsics::_getByteAcquire:           return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     Acquire, false);
 629   case vmIntrinsics::_getShortAcquire:          return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    Acquire, false);
 630   case vmIntrinsics::_getCharAcquire:           return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     Acquire, false);
 631   case vmIntrinsics::_getIntAcquire:            return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      Acquire, false);
 632   case vmIntrinsics::_getLongAcquire:           return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     Acquire, false);
 633   case vmIntrinsics::_getFloatAcquire:          return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    Acquire, false);
 634   case vmIntrinsics::_getDoubleAcquire:         return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   Acquire, false);
 635 
 636   case vmIntrinsics::_putObjectRelease:         return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   Release, false);
 637   case vmIntrinsics::_putBooleanRelease:        return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  Release, false);
 638   case vmIntrinsics::_putByteRelease:           return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     Release, false);
 639   case vmIntrinsics::_putShortRelease:          return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    Release, false);
 640   case vmIntrinsics::_putCharRelease:           return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     Release, false);
 641   case vmIntrinsics::_putIntRelease:            return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      Release, false);
 642   case vmIntrinsics::_putLongRelease:           return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     Release, false);
 643   case vmIntrinsics::_putFloatRelease:          return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    Release, false);
 644   case vmIntrinsics::_putDoubleRelease:         return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   Release, false);
 645 
 646   case vmIntrinsics::_getObjectOpaque:          return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   Opaque, false);
 647   case vmIntrinsics::_getBooleanOpaque:         return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  Opaque, false);
 648   case vmIntrinsics::_getByteOpaque:            return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     Opaque, false);
 649   case vmIntrinsics::_getShortOpaque:           return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    Opaque, false);
 650   case vmIntrinsics::_getCharOpaque:            return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     Opaque, false);
 651   case vmIntrinsics::_getIntOpaque:             return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      Opaque, false);
 652   case vmIntrinsics::_getLongOpaque:            return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     Opaque, false);
 653   case vmIntrinsics::_getFloatOpaque:           return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    Opaque, false);
 654   case vmIntrinsics::_getDoubleOpaque:          return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   Opaque, false);
 655 
 656   case vmIntrinsics::_putObjectOpaque:          return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   Opaque, false);
 657   case vmIntrinsics::_putBooleanOpaque:         return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  Opaque, false);
 658   case vmIntrinsics::_putByteOpaque:            return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     Opaque, false);
 659   case vmIntrinsics::_putShortOpaque:           return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    Opaque, false);
 660   case vmIntrinsics::_putCharOpaque:            return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     Opaque, false);
 661   case vmIntrinsics::_putIntOpaque:             return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      Opaque, false);
 662   case vmIntrinsics::_putLongOpaque:            return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     Opaque, false);
 663   case vmIntrinsics::_putFloatOpaque:           return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    Opaque, false);
 664   case vmIntrinsics::_putDoubleOpaque:          return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   Opaque, false);
 665 
 666   case vmIntrinsics::_compareAndSwapObject:             return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap,      Volatile);
 667   case vmIntrinsics::_compareAndSwapInt:                return inline_unsafe_load_store(T_INT,    LS_cmp_swap,      Volatile);
 668   case vmIntrinsics::_compareAndSwapLong:               return inline_unsafe_load_store(T_LONG,   LS_cmp_swap,      Volatile);
 669 
 670   case vmIntrinsics::_weakCompareAndSwapObject:         return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
 671   case vmIntrinsics::_weakCompareAndSwapObjectAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
 672   case vmIntrinsics::_weakCompareAndSwapObjectRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
 673   case vmIntrinsics::_weakCompareAndSwapInt:            return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Relaxed);
 674   case vmIntrinsics::_weakCompareAndSwapIntAcquire:     return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Acquire);
 675   case vmIntrinsics::_weakCompareAndSwapIntRelease:     return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Release);
 676   case vmIntrinsics::_weakCompareAndSwapLong:           return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Relaxed);
 677   case vmIntrinsics::_weakCompareAndSwapLongAcquire:    return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Acquire);
 678   case vmIntrinsics::_weakCompareAndSwapLongRelease:    return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Release);
 679 
 680   case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Volatile);
 681   case vmIntrinsics::_compareAndExchangeObjectAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Acquire);
 682   case vmIntrinsics::_compareAndExchangeObjectRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Release);
 683   case vmIntrinsics::_compareAndExchangeIntVolatile:    return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Volatile);
 684   case vmIntrinsics::_compareAndExchangeIntAcquire:     return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Acquire);
 685   case vmIntrinsics::_compareAndExchangeIntRelease:     return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Release);
 686   case vmIntrinsics::_compareAndExchangeLongVolatile:   return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Volatile);
 687   case vmIntrinsics::_compareAndExchangeLongAcquire:    return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Acquire);
 688   case vmIntrinsics::_compareAndExchangeLongRelease:    return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Release);
 689 
 690   case vmIntrinsics::_getAndAddInt:                     return inline_unsafe_load_store(T_INT,    LS_get_add,       Volatile);
 691   case vmIntrinsics::_getAndAddLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_add,       Volatile);
 692   case vmIntrinsics::_getAndSetInt:                     return inline_unsafe_load_store(T_INT,    LS_get_set,       Volatile);
 693   case vmIntrinsics::_getAndSetLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_set,       Volatile);
 694   case vmIntrinsics::_getAndSetObject:                  return inline_unsafe_load_store(T_OBJECT, LS_get_set,       Volatile);
 695 
 696   case vmIntrinsics::_loadFence:
 697   case vmIntrinsics::_storeFence:
 698   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 699 
 700   case vmIntrinsics::_onSpinWait:               return inline_onspinwait();
 701 
 702   case vmIntrinsics::_currentThread:            return inline_native_currentThread();
 703   case vmIntrinsics::_isInterrupted:            return inline_native_isInterrupted();
 704 
 705 #ifdef TRACE_HAVE_INTRINSICS
 706   case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
 707 #endif
 708   case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
 709   case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
 710   case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
 711   case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
 712   case vmIntrinsics::_getLength:                return inline_native_getLength();
 713   case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
 714   case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
 715   case vmIntrinsics::_equalsB:                  return inline_array_equals(StrIntrinsicNode::LL);
 716   case vmIntrinsics::_equalsC:                  return inline_array_equals(StrIntrinsicNode::UU);
 717   case vmIntrinsics::_Objects_checkIndex:       return inline_objects_checkIndex();
 718   case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());
 719 
 720   case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
 721   case vmIntrinsics::_newArray:                   return inline_unsafe_newArray(false);
 722 
 723   case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
 724 
 725   case vmIntrinsics::_isInstance:
 726   case vmIntrinsics::_getModifiers:
 727   case vmIntrinsics::_isInterface:
 728   case vmIntrinsics::_isArray:
 729   case vmIntrinsics::_isPrimitive:
 730   case vmIntrinsics::_getSuperclass:
 731   case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
 732 
 733   case vmIntrinsics::_floatToRawIntBits:
 734   case vmIntrinsics::_floatToIntBits:
 735   case vmIntrinsics::_intBitsToFloat:
 736   case vmIntrinsics::_doubleToRawLongBits:
 737   case vmIntrinsics::_doubleToLongBits:
 738   case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
 739 
 740   case vmIntrinsics::_numberOfLeadingZeros_i:
 741   case vmIntrinsics::_numberOfLeadingZeros_l:
 742   case vmIntrinsics::_numberOfTrailingZeros_i:
 743   case vmIntrinsics::_numberOfTrailingZeros_l:
 744   case vmIntrinsics::_bitCount_i:
 745   case vmIntrinsics::_bitCount_l:
 746   case vmIntrinsics::_reverseBytes_i:
 747   case vmIntrinsics::_reverseBytes_l:
 748   case vmIntrinsics::_reverseBytes_s:
 749   case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
 750 
 751   case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
 752 
 753   case vmIntrinsics::_Reference_get:            return inline_reference_get();
 754 
 755   case vmIntrinsics::_Class_cast:               return inline_Class_cast();
 756 
 757   case vmIntrinsics::_aescrypt_encryptBlock:
 758   case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
 759 
 760   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 761   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 762     return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
 763 
 764   case vmIntrinsics::_counterMode_AESCrypt:
 765     return inline_counterMode_AESCrypt(intrinsic_id());
 766 
 767   case vmIntrinsics::_sha_implCompress:
 768   case vmIntrinsics::_sha2_implCompress:
 769   case vmIntrinsics::_sha5_implCompress:
 770     return inline_sha_implCompress(intrinsic_id());
 771 
 772   case vmIntrinsics::_digestBase_implCompressMB:
 773     return inline_digestBase_implCompressMB(predicate);
 774 
 775   case vmIntrinsics::_multiplyToLen:
 776     return inline_multiplyToLen();
 777 
 778   case vmIntrinsics::_squareToLen:
 779     return inline_squareToLen();
 780 
 781   case vmIntrinsics::_mulAdd:
 782     return inline_mulAdd();
 783 
 784   case vmIntrinsics::_montgomeryMultiply:
 785     return inline_montgomeryMultiply();
 786   case vmIntrinsics::_montgomerySquare:
 787     return inline_montgomerySquare();
 788 
 789   case vmIntrinsics::_vectorizedMismatch:
 790     return inline_vectorizedMismatch();
 791 
 792   case vmIntrinsics::_ghash_processBlocks:
 793     return inline_ghash_processBlocks();
 794 
 795   case vmIntrinsics::_encodeISOArray:
 796   case vmIntrinsics::_encodeByteISOArray:
 797     return inline_encodeISOArray();
 798 
 799   case vmIntrinsics::_updateCRC32:
 800     return inline_updateCRC32();
 801   case vmIntrinsics::_updateBytesCRC32:
 802     return inline_updateBytesCRC32();
 803   case vmIntrinsics::_updateByteBufferCRC32:
 804     return inline_updateByteBufferCRC32();
 805 
 806   case vmIntrinsics::_updateBytesCRC32C:
 807     return inline_updateBytesCRC32C();
 808   case vmIntrinsics::_updateDirectByteBufferCRC32C:
 809     return inline_updateDirectByteBufferCRC32C();
 810 
 811   case vmIntrinsics::_updateBytesAdler32:
 812     return inline_updateBytesAdler32();
 813   case vmIntrinsics::_updateByteBufferAdler32:
 814     return inline_updateByteBufferAdler32();
 815 
 816   case vmIntrinsics::_profileBoolean:
 817     return inline_profileBoolean();
 818   case vmIntrinsics::_isCompileConstant:
 819     return inline_isCompileConstant();
 820 
 821   case vmIntrinsics::_hasNegatives:
 822     return inline_hasNegatives();
 823 
 824   default:
 825     // If you get here, it may be that someone has added a new intrinsic
 826     // to the list in vmSymbols.hpp without implementing it here.
 827 #ifndef PRODUCT
 828     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 829       tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
 830                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 831     }
 832 #endif
 833     return false;
 834   }
 835 }
 836 
 837 Node* LibraryCallKit::try_to_predicate(int predicate) {
 838   if (!jvms()->has_method()) {
 839     // Root JVMState has a null method.
 840     assert(map()->memory()->Opcode() == Op_Parm, "");
 841     // Insert the memory aliasing node
 842     set_all_memory(reset_memory());
 843   }
 844   assert(merged_memory(), "");
 845 
 846   switch (intrinsic_id()) {
 847   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 848     return inline_cipherBlockChaining_AESCrypt_predicate(false);
 849   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 850     return inline_cipherBlockChaining_AESCrypt_predicate(true);
 851   case vmIntrinsics::_counterMode_AESCrypt:
 852     return inline_counterMode_AESCrypt_predicate();
 853   case vmIntrinsics::_digestBase_implCompressMB:
 854     return inline_digestBase_implCompressMB_predicate(predicate);
 855 
 856   default:
 857     // If you get here, it may be that someone has added a new intrinsic
 858     // to the list in vmSymbols.hpp without implementing it here.
 859 #ifndef PRODUCT
 860     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 861       tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
 862                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 863     }
 864 #endif
 865     Node* slow_ctl = control();
 866     set_control(top()); // No fast path instrinsic
 867     return slow_ctl;
 868   }
 869 }
 870 
 871 //------------------------------set_result-------------------------------
 872 // Helper function for finishing intrinsics.
 873 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
 874   record_for_igvn(region);
 875   set_control(_gvn.transform(region));
 876   set_result( _gvn.transform(value));
 877   assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
 878 }
 879 
 880 //------------------------------generate_guard---------------------------
 881 // Helper function for generating guarded fast-slow graph structures.
 882 // The given 'test', if true, guards a slow path.  If the test fails
 883 // then a fast path can be taken.  (We generally hope it fails.)
 884 // In all cases, GraphKit::control() is updated to the fast path.
 885 // The returned value represents the control for the slow path.
 886 // The return value is never 'top'; it is either a valid control
 887 // or NULL if it is obvious that the slow path can never be taken.
 888 // Also, if region and the slow control are not NULL, the slow edge
 889 // is appended to the region.
 890 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
 891   if (stopped()) {
 892     // Already short circuited.
 893     return NULL;
 894   }
 895 
 896   // Build an if node and its projections.
 897   // If test is true we take the slow path, which we assume is uncommon.
 898   if (_gvn.type(test) == TypeInt::ZERO) {
 899     // The slow branch is never taken.  No need to build this guard.
 900     return NULL;
 901   }
 902 
 903   IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
 904 
 905   Node* if_slow = _gvn.transform(new IfTrueNode(iff));
 906   if (if_slow == top()) {
 907     // The slow branch is never taken.  No need to build this guard.
 908     return NULL;
 909   }
 910 
 911   if (region != NULL)
 912     region->add_req(if_slow);
 913 
 914   Node* if_fast = _gvn.transform(new IfFalseNode(iff));
 915   set_control(if_fast);
 916 
 917   return if_slow;
 918 }
 919 
 920 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
 921   return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
 922 }
 923 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
 924   return generate_guard(test, region, PROB_FAIR);
 925 }
 926 
 927 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
 928                                                      Node* *pos_index) {
 929   if (stopped())
 930     return NULL;                // already stopped
 931   if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
 932     return NULL;                // index is already adequately typed
 933   Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
 934   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 935   Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
 936   if (is_neg != NULL && pos_index != NULL) {
 937     // Emulate effect of Parse::adjust_map_after_if.
 938     Node* ccast = new CastIINode(index, TypeInt::POS);
 939     ccast->set_req(0, control());
 940     (*pos_index) = _gvn.transform(ccast);
 941   }
 942   return is_neg;
 943 }
 944 
 945 // Make sure that 'position' is a valid limit index, in [0..length].
 946 // There are two equivalent plans for checking this:
 947 //   A. (offset + copyLength)  unsigned<=  arrayLength
 948 //   B. offset  <=  (arrayLength - copyLength)
 949 // We require that all of the values above, except for the sum and
 950 // difference, are already known to be non-negative.
 951 // Plan A is robust in the face of overflow, if offset and copyLength
 952 // are both hugely positive.
 953 //
 954 // Plan B is less direct and intuitive, but it does not overflow at
 955 // all, since the difference of two non-negatives is always
 956 // representable.  Whenever Java methods must perform the equivalent
 957 // check they generally use Plan B instead of Plan A.
 958 // For the moment we use Plan A.
 959 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
 960                                                   Node* subseq_length,
 961                                                   Node* array_length,
 962                                                   RegionNode* region) {
 963   if (stopped())
 964     return NULL;                // already stopped
 965   bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
 966   if (zero_offset && subseq_length->eqv_uncast(array_length))
 967     return NULL;                // common case of whole-array copy
 968   Node* last = subseq_length;
 969   if (!zero_offset)             // last += offset
 970     last = _gvn.transform(new AddINode(last, offset));
 971   Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
 972   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 973   Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
 974   return is_over;
 975 }
 976 
 977 // Emit range checks for the given String.value byte array
 978 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
 979   if (stopped()) {
 980     return; // already stopped
 981   }
 982   RegionNode* bailout = new RegionNode(1);
 983   record_for_igvn(bailout);
 984   if (char_count) {
 985     // Convert char count to byte count
 986     count = _gvn.transform(new LShiftINode(count, intcon(1)));
 987   }
 988 
 989   // Offset and count must not be negative
 990   generate_negative_guard(offset, bailout);
 991   generate_negative_guard(count, bailout);
 992   // Offset + count must not exceed length of array
 993   generate_limit_guard(offset, count, load_array_length(array), bailout);
 994 
 995   if (bailout->req() > 1) {
 996     PreserveJVMState pjvms(this);
 997     set_control(_gvn.transform(bailout));
 998     uncommon_trap(Deoptimization::Reason_intrinsic,
 999                   Deoptimization::Action_maybe_recompile);
1000   }
1001 }
1002 
1003 //--------------------------generate_current_thread--------------------
1004 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1005   ciKlass*    thread_klass = env()->Thread_klass();
1006   const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1007   Node* thread = _gvn.transform(new ThreadLocalNode());
1008   Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1009   Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1010   tls_output = thread;
1011   return threadObj;
1012 }
1013 
1014 
1015 //------------------------------make_string_method_node------------------------
1016 // Helper method for String intrinsic functions. This version is called with
1017 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1018 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1019 // containing the lengths of str1 and str2.
1020 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1021   Node* result = NULL;
1022   switch (opcode) {
1023   case Op_StrIndexOf:
1024     result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1025                                 str1_start, cnt1, str2_start, cnt2, ae);
1026     break;
1027   case Op_StrComp:
1028     result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1029                              str1_start, cnt1, str2_start, cnt2, ae);
1030     break;
1031   case Op_StrEquals:
1032     // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1033     // Use the constant length if there is one because optimized match rule may exist.
1034     result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1035                                str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1036     break;
1037   default:
1038     ShouldNotReachHere();
1039     return NULL;
1040   }
1041 
1042   // All these intrinsics have checks.
1043   C->set_has_split_ifs(true); // Has chance for split-if optimization
1044 
1045   return _gvn.transform(result);
1046 }
1047 
1048 //------------------------------inline_string_compareTo------------------------
1049 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1050   Node* arg1 = argument(0);
1051   Node* arg2 = argument(1);
1052 
1053   // Get start addr and length of first argument
1054   Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1055   Node* arg1_cnt    = load_array_length(arg1);
1056 
1057   // Get start addr and length of second argument
1058   Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1059   Node* arg2_cnt    = load_array_length(arg2);
1060 
1061   Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1062   set_result(result);
1063   return true;
1064 }
1065 
1066 //------------------------------inline_string_equals------------------------
1067 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1068   Node* arg1 = argument(0);
1069   Node* arg2 = argument(1);
1070 
1071   // paths (plus control) merge
1072   RegionNode* region = new RegionNode(3);
1073   Node* phi = new PhiNode(region, TypeInt::BOOL);
1074 
1075   if (!stopped()) {
1076     // Get start addr and length of first argument
1077     Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1078     Node* arg1_cnt    = load_array_length(arg1);
1079 
1080     // Get start addr and length of second argument
1081     Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1082     Node* arg2_cnt    = load_array_length(arg2);
1083 
1084     // Check for arg1_cnt != arg2_cnt
1085     Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1086     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1087     Node* if_ne = generate_slow_guard(bol, NULL);
1088     if (if_ne != NULL) {
1089       phi->init_req(2, intcon(0));
1090       region->init_req(2, if_ne);
1091     }
1092 
1093     // Check for count == 0 is done by assembler code for StrEquals.
1094 
1095     if (!stopped()) {
1096       Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1097       phi->init_req(1, equals);
1098       region->init_req(1, control());
1099     }
1100   }
1101 
1102   // post merge
1103   set_control(_gvn.transform(region));
1104   record_for_igvn(region);
1105 
1106   set_result(_gvn.transform(phi));
1107   return true;
1108 }
1109 
1110 //------------------------------inline_array_equals----------------------------
1111 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1112   assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1113   Node* arg1 = argument(0);
1114   Node* arg2 = argument(1);
1115 
1116   const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1117   set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1118   return true;
1119 }
1120 
1121 //------------------------------inline_hasNegatives------------------------------
1122 bool LibraryCallKit::inline_hasNegatives() {
1123   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1124     return false;
1125   }
1126 
1127   assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1128   // no receiver since it is static method
1129   Node* ba         = argument(0);
1130   Node* offset     = argument(1);
1131   Node* len        = argument(2);
1132 
1133   // Range checks
1134   generate_string_range_check(ba, offset, len, false);
1135   if (stopped()) {
1136     return true;
1137   }
1138   Node* ba_start = array_element_address(ba, offset, T_BYTE);
1139   Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1140   set_result(_gvn.transform(result));
1141   return true;
1142 }
1143 
1144 bool LibraryCallKit::inline_objects_checkIndex() {
1145   Node* index = argument(0);
1146   Node* length = argument(1);
1147   if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1148     return false;
1149   }
1150 
1151   Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1152   Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1153 
1154   {
1155     BuildCutout unless(this, len_pos_bol, PROB_MAX);
1156     uncommon_trap(Deoptimization::Reason_intrinsic,
1157                   Deoptimization::Action_make_not_entrant);
1158   }
1159 
1160   if (stopped()) {
1161     return false;
1162   }
1163 
1164   Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1165   BoolTest::mask btest = BoolTest::lt;
1166   Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1167   RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1168   _gvn.set_type(rc, rc->Value(&_gvn));
1169   if (!rc_bool->is_Con()) {
1170     record_for_igvn(rc);
1171   }
1172   set_control(_gvn.transform(new IfTrueNode(rc)));
1173   {
1174     PreserveJVMState pjvms(this);
1175     set_control(_gvn.transform(new IfFalseNode(rc)));
1176     uncommon_trap(Deoptimization::Reason_range_check,
1177                   Deoptimization::Action_make_not_entrant);
1178   }
1179 
1180   if (stopped()) {
1181     return false;
1182   }
1183 
1184   Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1185   result->set_req(0, control());
1186   result = _gvn.transform(result);
1187   set_result(result);
1188   replace_in_map(index, result);
1189   return true;
1190 }
1191 
1192 //------------------------------inline_string_indexOf------------------------
1193 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1194   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1195     return false;
1196   }
1197   Node* src = argument(0);
1198   Node* tgt = argument(1);
1199 
1200   // Make the merge point
1201   RegionNode* result_rgn = new RegionNode(4);
1202   Node*       result_phi = new PhiNode(result_rgn, TypeInt::INT);
1203 
1204   // Get start addr and length of source string
1205   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1206   Node* src_count = load_array_length(src);
1207 
1208   // Get start addr and length of substring
1209   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1210   Node* tgt_count = load_array_length(tgt);
1211 
1212   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1213     // Divide src size by 2 if String is UTF16 encoded
1214     src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1215   }
1216   if (ae == StrIntrinsicNode::UU) {
1217     // Divide substring size by 2 if String is UTF16 encoded
1218     tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1219   }
1220 
1221   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1222   if (result != NULL) {
1223     result_phi->init_req(3, result);
1224     result_rgn->init_req(3, control());
1225   }
1226   set_control(_gvn.transform(result_rgn));
1227   record_for_igvn(result_rgn);
1228   set_result(_gvn.transform(result_phi));
1229 
1230   return true;
1231 }
1232 
1233 //-----------------------------inline_string_indexOf-----------------------
1234 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1235   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1236     return false;
1237   }
1238   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1239     return false;
1240   }
1241   assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1242   Node* src         = argument(0); // byte[]
1243   Node* src_count   = argument(1); // char count
1244   Node* tgt         = argument(2); // byte[]
1245   Node* tgt_count   = argument(3); // char count
1246   Node* from_index  = argument(4); // char index
1247 
1248   // Multiply byte array index by 2 if String is UTF16 encoded
1249   Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1250   src_count = _gvn.transform(new SubINode(src_count, from_index));
1251   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1252   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1253 
1254   // Range checks
1255   generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1256   generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1257   if (stopped()) {
1258     return true;
1259   }
1260 
1261   RegionNode* region = new RegionNode(5);
1262   Node* phi = new PhiNode(region, TypeInt::INT);
1263 
1264   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1265   if (result != NULL) {
1266     // The result is index relative to from_index if substring was found, -1 otherwise.
1267     // Generate code which will fold into cmove.
1268     Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1269     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1270 
1271     Node* if_lt = generate_slow_guard(bol, NULL);
1272     if (if_lt != NULL) {
1273       // result == -1
1274       phi->init_req(3, result);
1275       region->init_req(3, if_lt);
1276     }
1277     if (!stopped()) {
1278       result = _gvn.transform(new AddINode(result, from_index));
1279       phi->init_req(4, result);
1280       region->init_req(4, control());
1281     }
1282   }
1283 
1284   set_control(_gvn.transform(region));
1285   record_for_igvn(region);
1286   set_result(_gvn.transform(phi));
1287 
1288   return true;
1289 }
1290 
1291 // Create StrIndexOfNode with fast path checks
1292 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1293                                         RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1294   // Check for substr count > string count
1295   Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1296   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1297   Node* if_gt = generate_slow_guard(bol, NULL);
1298   if (if_gt != NULL) {
1299     phi->init_req(1, intcon(-1));
1300     region->init_req(1, if_gt);
1301   }
1302   if (!stopped()) {
1303     // Check for substr count == 0
1304     cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1305     bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1306     Node* if_zero = generate_slow_guard(bol, NULL);
1307     if (if_zero != NULL) {
1308       phi->init_req(2, intcon(0));
1309       region->init_req(2, if_zero);
1310     }
1311   }
1312   if (!stopped()) {
1313     return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1314   }
1315   return NULL;
1316 }
1317 
1318 //-----------------------------inline_string_indexOfChar-----------------------
1319 bool LibraryCallKit::inline_string_indexOfChar() {
1320   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1321     return false;
1322   }
1323   if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1324     return false;
1325   }
1326   assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1327   Node* src         = argument(0); // byte[]
1328   Node* tgt         = argument(1); // tgt is int ch
1329   Node* from_index  = argument(2);
1330   Node* max         = argument(3);
1331 
1332   Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1333   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1334   Node* src_count = _gvn.transform(new SubINode(max, from_index));
1335 
1336   // Range checks
1337   generate_string_range_check(src, src_offset, src_count, true);
1338   if (stopped()) {
1339     return true;
1340   }
1341 
1342   RegionNode* region = new RegionNode(3);
1343   Node* phi = new PhiNode(region, TypeInt::INT);
1344 
1345   Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1346   C->set_has_split_ifs(true); // Has chance for split-if optimization
1347   _gvn.transform(result);
1348 
1349   Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1350   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1351 
1352   Node* if_lt = generate_slow_guard(bol, NULL);
1353   if (if_lt != NULL) {
1354     // result == -1
1355     phi->init_req(2, result);
1356     region->init_req(2, if_lt);
1357   }
1358   if (!stopped()) {
1359     result = _gvn.transform(new AddINode(result, from_index));
1360     phi->init_req(1, result);
1361     region->init_req(1, control());
1362   }
1363   set_control(_gvn.transform(region));
1364   record_for_igvn(region);
1365   set_result(_gvn.transform(phi));
1366 
1367   return true;
1368 }
1369 //---------------------------inline_string_copy---------------------
1370 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1371 //   int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1372 //   int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1373 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1374 //   void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1375 //   void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1376 bool LibraryCallKit::inline_string_copy(bool compress) {
1377   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1378     return false;
1379   }
1380   int nargs = 5;  // 2 oops, 3 ints
1381   assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1382 
1383   Node* src         = argument(0);
1384   Node* src_offset  = argument(1);
1385   Node* dst         = argument(2);
1386   Node* dst_offset  = argument(3);
1387   Node* length      = argument(4);
1388 
1389   // Check for allocation before we add nodes that would confuse
1390   // tightly_coupled_allocation()
1391   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1392 
1393   // Figure out the size and type of the elements we will be copying.
1394   const Type* src_type = src->Value(&_gvn);
1395   const Type* dst_type = dst->Value(&_gvn);
1396   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1397   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1398   assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1399          (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1400          "Unsupported array types for inline_string_copy");
1401 
1402   // Range checks
1403   generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1404   generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1405   if (stopped()) {
1406     return true;
1407   }
1408 
1409   // Convert char[] offsets to byte[] offsets
1410   if (compress && src_elem == T_BYTE) {
1411     src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1412   } else if (!compress && dst_elem == T_BYTE) {
1413     dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1414   }
1415 
1416   Node* src_start = array_element_address(src, src_offset, src_elem);
1417   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1418   // 'src_start' points to src array + scaled offset
1419   // 'dst_start' points to dst array + scaled offset
1420   Node* count = NULL;
1421   if (compress) {
1422     count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1423   } else {
1424     inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1425   }
1426 
1427   if (alloc != NULL) {
1428     if (alloc->maybe_set_complete(&_gvn)) {
1429       // "You break it, you buy it."
1430       InitializeNode* init = alloc->initialization();
1431       assert(init->is_complete(), "we just did this");
1432       init->set_complete_with_arraycopy();
1433       assert(dst->is_CheckCastPP(), "sanity");
1434       assert(dst->in(0)->in(0) == init, "dest pinned");
1435     }
1436     // Do not let stores that initialize this object be reordered with
1437     // a subsequent store that would make this object accessible by
1438     // other threads.
1439     // Record what AllocateNode this StoreStore protects so that
1440     // escape analysis can go from the MemBarStoreStoreNode to the
1441     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1442     // based on the escape status of the AllocateNode.
1443     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1444   }
1445   if (compress) {
1446     set_result(_gvn.transform(count));
1447   }
1448   return true;
1449 }
1450 
1451 #ifdef _LP64
1452 #define XTOP ,top() /*additional argument*/
1453 #else  //_LP64
1454 #define XTOP        /*no additional argument*/
1455 #endif //_LP64
1456 
1457 //------------------------inline_string_toBytesU--------------------------
1458 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1459 bool LibraryCallKit::inline_string_toBytesU() {
1460   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1461     return false;
1462   }
1463   // Get the arguments.
1464   Node* value     = argument(0);
1465   Node* offset    = argument(1);
1466   Node* length    = argument(2);
1467 
1468   Node* newcopy = NULL;
1469 
1470   // Set the original stack and the reexecute bit for the interpreter to reexecute
1471   // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1472   { PreserveReexecuteState preexecs(this);
1473     jvms()->set_should_reexecute(true);
1474 
1475     // Check if a null path was taken unconditionally.
1476     value = null_check(value);
1477 
1478     RegionNode* bailout = new RegionNode(1);
1479     record_for_igvn(bailout);
1480 
1481     // Range checks
1482     generate_negative_guard(offset, bailout);
1483     generate_negative_guard(length, bailout);
1484     generate_limit_guard(offset, length, load_array_length(value), bailout);
1485     // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1486     generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1487 
1488     if (bailout->req() > 1) {
1489       PreserveJVMState pjvms(this);
1490       set_control(_gvn.transform(bailout));
1491       uncommon_trap(Deoptimization::Reason_intrinsic,
1492                     Deoptimization::Action_maybe_recompile);
1493     }
1494     if (stopped()) {
1495       return true;
1496     }
1497 
1498     Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1499     Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1500     newcopy = new_array(klass_node, size, 0);  // no arguments to push
1501     AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1502 
1503     // Calculate starting addresses.
1504     Node* src_start = array_element_address(value, offset, T_CHAR);
1505     Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1506 
1507     // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1508     const TypeInt* toffset = gvn().type(offset)->is_int();
1509     bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1510 
1511     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1512     const char* copyfunc_name = "arraycopy";
1513     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1514     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1515                       OptoRuntime::fast_arraycopy_Type(),
1516                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1517                       src_start, dst_start, ConvI2X(length) XTOP);
1518     // Do not let reads from the cloned object float above the arraycopy.
1519     if (alloc != NULL) {
1520       if (alloc->maybe_set_complete(&_gvn)) {
1521         // "You break it, you buy it."
1522         InitializeNode* init = alloc->initialization();
1523         assert(init->is_complete(), "we just did this");
1524         init->set_complete_with_arraycopy();
1525         assert(newcopy->is_CheckCastPP(), "sanity");
1526         assert(newcopy->in(0)->in(0) == init, "dest pinned");
1527       }
1528       // Do not let stores that initialize this object be reordered with
1529       // a subsequent store that would make this object accessible by
1530       // other threads.
1531       // Record what AllocateNode this StoreStore protects so that
1532       // escape analysis can go from the MemBarStoreStoreNode to the
1533       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1534       // based on the escape status of the AllocateNode.
1535       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1536     } else {
1537       insert_mem_bar(Op_MemBarCPUOrder);
1538     }
1539   } // original reexecute is set back here
1540 
1541   C->set_has_split_ifs(true); // Has chance for split-if optimization
1542   if (!stopped()) {
1543     set_result(newcopy);
1544   }
1545   return true;
1546 }
1547 
1548 //------------------------inline_string_getCharsU--------------------------
1549 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1550 bool LibraryCallKit::inline_string_getCharsU() {
1551   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1552     return false;
1553   }
1554 
1555   // Get the arguments.
1556   Node* src       = argument(0);
1557   Node* src_begin = argument(1);
1558   Node* src_end   = argument(2); // exclusive offset (i < src_end)
1559   Node* dst       = argument(3);
1560   Node* dst_begin = argument(4);
1561 
1562   // Check for allocation before we add nodes that would confuse
1563   // tightly_coupled_allocation()
1564   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1565 
1566   // Check if a null path was taken unconditionally.
1567   src = null_check(src);
1568   dst = null_check(dst);
1569   if (stopped()) {
1570     return true;
1571   }
1572 
1573   // Get length and convert char[] offset to byte[] offset
1574   Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1575   src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1576 
1577   // Range checks
1578   generate_string_range_check(src, src_begin, length, true);
1579   generate_string_range_check(dst, dst_begin, length, false);
1580   if (stopped()) {
1581     return true;
1582   }
1583 
1584   if (!stopped()) {
1585     // Calculate starting addresses.
1586     Node* src_start = array_element_address(src, src_begin, T_BYTE);
1587     Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1588 
1589     // Check if array addresses are aligned to HeapWordSize
1590     const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1591     const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1592     bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1593                    tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1594 
1595     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1596     const char* copyfunc_name = "arraycopy";
1597     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1598     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1599                       OptoRuntime::fast_arraycopy_Type(),
1600                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1601                       src_start, dst_start, ConvI2X(length) XTOP);
1602     // Do not let reads from the cloned object float above the arraycopy.
1603     if (alloc != NULL) {
1604       if (alloc->maybe_set_complete(&_gvn)) {
1605         // "You break it, you buy it."
1606         InitializeNode* init = alloc->initialization();
1607         assert(init->is_complete(), "we just did this");
1608         init->set_complete_with_arraycopy();
1609         assert(dst->is_CheckCastPP(), "sanity");
1610         assert(dst->in(0)->in(0) == init, "dest pinned");
1611       }
1612       // Do not let stores that initialize this object be reordered with
1613       // a subsequent store that would make this object accessible by
1614       // other threads.
1615       // Record what AllocateNode this StoreStore protects so that
1616       // escape analysis can go from the MemBarStoreStoreNode to the
1617       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1618       // based on the escape status of the AllocateNode.
1619       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1620     } else {
1621       insert_mem_bar(Op_MemBarCPUOrder);
1622     }
1623   }
1624 
1625   C->set_has_split_ifs(true); // Has chance for split-if optimization
1626   return true;
1627 }
1628 
1629 //----------------------inline_string_char_access----------------------------
1630 // Store/Load char to/from byte[] array.
1631 // static void StringUTF16.putChar(byte[] val, int index, int c)
1632 // static char StringUTF16.getChar(byte[] val, int index)
1633 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1634   Node* value  = argument(0);
1635   Node* index  = argument(1);
1636   Node* ch = is_store ? argument(2) : NULL;
1637 
1638   // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1639   // correctly requires matched array shapes.
1640   assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1641           "sanity: byte[] and char[] bases agree");
1642   assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1643           "sanity: byte[] and char[] scales agree");
1644 
1645   // Bail when getChar over constants is requested: constant folding would
1646   // reject folding mismatched char access over byte[]. A normal inlining for getChar
1647   // Java method would constant fold nicely instead.
1648   if (!is_store && value->is_Con() && index->is_Con()) {
1649     return false;
1650   }
1651 
1652   Node* adr = array_element_address(value, index, T_CHAR);
1653   if (is_store) {
1654     (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1655                            false, false, true /* mismatched */);
1656   } else {
1657     ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1658                    LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1659     set_result(ch);
1660   }
1661   return true;
1662 }
1663 
1664 //--------------------------round_double_node--------------------------------
1665 // Round a double node if necessary.
1666 Node* LibraryCallKit::round_double_node(Node* n) {
1667   if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1668     n = _gvn.transform(new RoundDoubleNode(0, n));
1669   return n;
1670 }
1671 
1672 //------------------------------inline_math-----------------------------------
1673 // public static double Math.abs(double)
1674 // public static double Math.sqrt(double)
1675 // public static double Math.log(double)
1676 // public static double Math.log10(double)
1677 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1678   Node* arg = round_double_node(argument(0));
1679   Node* n = NULL;
1680   switch (id) {
1681   case vmIntrinsics::_dabs:   n = new AbsDNode(                arg);  break;
1682   case vmIntrinsics::_dsqrt:  n = new SqrtDNode(C, control(),  arg);  break;
1683   default:  fatal_unexpected_iid(id);  break;
1684   }
1685   set_result(_gvn.transform(n));
1686   return true;
1687 }
1688 
1689 //------------------------------inline_trig----------------------------------
1690 // Inline sin/cos/tan instructions, if possible.  If rounding is required, do
1691 // argument reduction which will turn into a fast/slow diamond.
1692 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1693   Node* arg = round_double_node(argument(0));
1694   Node* n = NULL;
1695 
1696   n = _gvn.transform(n);
1697 
1698   // Rounding required?  Check for argument reduction!
1699   if (Matcher::strict_fp_requires_explicit_rounding) {
1700     static const double     pi_4 =  0.7853981633974483;
1701     static const double neg_pi_4 = -0.7853981633974483;
1702     // pi/2 in 80-bit extended precision
1703     // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1704     // -pi/2 in 80-bit extended precision
1705     // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00};
1706     // Cutoff value for using this argument reduction technique
1707     //static const double    pi_2_minus_epsilon =  1.564660403643354;
1708     //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1709 
1710     // Pseudocode for sin:
1711     // if (x <= Math.PI / 4.0) {
1712     //   if (x >= -Math.PI / 4.0) return  fsin(x);
1713     //   if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1714     // } else {
1715     //   if (x <=  Math.PI / 2.0) return  fcos(x - Math.PI / 2.0);
1716     // }
1717     // return StrictMath.sin(x);
1718 
1719     // Pseudocode for cos:
1720     // if (x <= Math.PI / 4.0) {
1721     //   if (x >= -Math.PI / 4.0) return  fcos(x);
1722     //   if (x >= -Math.PI / 2.0) return  fsin(x + Math.PI / 2.0);
1723     // } else {
1724     //   if (x <=  Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1725     // }
1726     // return StrictMath.cos(x);
1727 
1728     // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1729     // requires a special machine instruction to load it.  Instead we'll try
1730     // the 'easy' case.  If we really need the extra range +/- PI/2 we'll
1731     // probably do the math inside the SIN encoding.
1732 
1733     // Make the merge point
1734     RegionNode* r = new RegionNode(3);
1735     Node* phi = new PhiNode(r, Type::DOUBLE);
1736 
1737     // Flatten arg so we need only 1 test
1738     Node *abs = _gvn.transform(new AbsDNode(arg));
1739     // Node for PI/4 constant
1740     Node *pi4 = makecon(TypeD::make(pi_4));
1741     // Check PI/4 : abs(arg)
1742     Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1743     // Check: If PI/4 < abs(arg) then go slow
1744     Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1745     // Branch either way
1746     IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1747     set_control(opt_iff(r,iff));
1748 
1749     // Set fast path result
1750     phi->init_req(2, n);
1751 
1752     // Slow path - non-blocking leaf call
1753     Node* call = NULL;
1754     switch (id) {
1755     case vmIntrinsics::_dtan:
1756       call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1757                                CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1758                                "Tan", NULL, arg, top());
1759       break;
1760     }
1761     assert(control()->in(0) == call, "");
1762     Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1763     r->init_req(1, control());
1764     phi->init_req(1, slow_result);
1765 
1766     // Post-merge
1767     set_control(_gvn.transform(r));
1768     record_for_igvn(r);
1769     n = _gvn.transform(phi);
1770 
1771     C->set_has_split_ifs(true); // Has chance for split-if optimization
1772   }
1773   set_result(n);
1774   return true;
1775 }
1776 
1777 //------------------------------runtime_math-----------------------------
1778 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1779   assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1780          "must be (DD)D or (D)D type");
1781 
1782   // Inputs
1783   Node* a = round_double_node(argument(0));
1784   Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1785 
1786   const TypePtr* no_memory_effects = NULL;
1787   Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1788                                  no_memory_effects,
1789                                  a, top(), b, b ? top() : NULL);
1790   Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1791 #ifdef ASSERT
1792   Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1793   assert(value_top == top(), "second value must be top");
1794 #endif
1795 
1796   set_result(value);
1797   return true;
1798 }
1799 
1800 //------------------------------inline_math_native-----------------------------
1801 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1802 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1803   switch (id) {
1804     // These intrinsics are not properly supported on all hardware
1805   case vmIntrinsics::_dsin:
1806     return StubRoutines::dsin() != NULL ?
1807       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1808       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
1809   case vmIntrinsics::_dcos:
1810     return StubRoutines::dcos() != NULL ?
1811       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1812       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
1813   case vmIntrinsics::_dtan:
1814     return StubRoutines::dtan() != NULL ?
1815       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1816       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1817   case vmIntrinsics::_dlog:
1818     return StubRoutines::dlog() != NULL ?
1819       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1820       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
1821   case vmIntrinsics::_dlog10:
1822     return StubRoutines::dlog10() != NULL ?
1823       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1824       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1825 
1826     // These intrinsics are supported on all hardware
1827   case vmIntrinsics::_dsqrt:  return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1828   case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_math(id) : false;
1829 
1830   case vmIntrinsics::_dexp:
1831     return StubRoutines::dexp() != NULL ?
1832       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(),  "dexp") :
1833       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp),  "EXP");
1834   case vmIntrinsics::_dpow:
1835     return StubRoutines::dpow() != NULL ?
1836       runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1837       runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow),  "POW");
1838 #undef FN_PTR
1839 
1840    // These intrinsics are not yet correctly implemented
1841   case vmIntrinsics::_datan2:
1842     return false;
1843 
1844   default:
1845     fatal_unexpected_iid(id);
1846     return false;
1847   }
1848 }
1849 
1850 static bool is_simple_name(Node* n) {
1851   return (n->req() == 1         // constant
1852           || (n->is_Type() && n->as_Type()->type()->singleton())
1853           || n->is_Proj()       // parameter or return value
1854           || n->is_Phi()        // local of some sort
1855           );
1856 }
1857 
1858 //----------------------------inline_notify-----------------------------------*
1859 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1860   const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1861   address func;
1862   if (id == vmIntrinsics::_notify) {
1863     func = OptoRuntime::monitor_notify_Java();
1864   } else {
1865     func = OptoRuntime::monitor_notifyAll_Java();
1866   }
1867   Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1868   make_slow_call_ex(call, env()->Throwable_klass(), false);
1869   return true;
1870 }
1871 
1872 
1873 //----------------------------inline_min_max-----------------------------------
1874 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1875   set_result(generate_min_max(id, argument(0), argument(1)));
1876   return true;
1877 }
1878 
1879 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1880   Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1881   IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1882   Node* fast_path = _gvn.transform( new IfFalseNode(check));
1883   Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1884 
1885   {
1886     PreserveJVMState pjvms(this);
1887     PreserveReexecuteState preexecs(this);
1888     jvms()->set_should_reexecute(true);
1889 
1890     set_control(slow_path);
1891     set_i_o(i_o());
1892 
1893     uncommon_trap(Deoptimization::Reason_intrinsic,
1894                   Deoptimization::Action_none);
1895   }
1896 
1897   set_control(fast_path);
1898   set_result(math);
1899 }
1900 
1901 template <typename OverflowOp>
1902 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1903   typedef typename OverflowOp::MathOp MathOp;
1904 
1905   MathOp* mathOp = new MathOp(arg1, arg2);
1906   Node* operation = _gvn.transform( mathOp );
1907   Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1908   inline_math_mathExact(operation, ofcheck);
1909   return true;
1910 }
1911 
1912 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1913   return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1914 }
1915 
1916 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1917   return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1918 }
1919 
1920 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1921   return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1922 }
1923 
1924 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1925   return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1926 }
1927 
1928 bool LibraryCallKit::inline_math_negateExactI() {
1929   return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1930 }
1931 
1932 bool LibraryCallKit::inline_math_negateExactL() {
1933   return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1934 }
1935 
1936 bool LibraryCallKit::inline_math_multiplyExactI() {
1937   return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1938 }
1939 
1940 bool LibraryCallKit::inline_math_multiplyExactL() {
1941   return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1942 }
1943 
1944 Node*
1945 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1946   // These are the candidate return value:
1947   Node* xvalue = x0;
1948   Node* yvalue = y0;
1949 
1950   if (xvalue == yvalue) {
1951     return xvalue;
1952   }
1953 
1954   bool want_max = (id == vmIntrinsics::_max);
1955 
1956   const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1957   const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1958   if (txvalue == NULL || tyvalue == NULL)  return top();
1959   // This is not really necessary, but it is consistent with a
1960   // hypothetical MaxINode::Value method:
1961   int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1962 
1963   // %%% This folding logic should (ideally) be in a different place.
1964   // Some should be inside IfNode, and there to be a more reliable
1965   // transformation of ?: style patterns into cmoves.  We also want
1966   // more powerful optimizations around cmove and min/max.
1967 
1968   // Try to find a dominating comparison of these guys.
1969   // It can simplify the index computation for Arrays.copyOf
1970   // and similar uses of System.arraycopy.
1971   // First, compute the normalized version of CmpI(x, y).
1972   int   cmp_op = Op_CmpI;
1973   Node* xkey = xvalue;
1974   Node* ykey = yvalue;
1975   Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1976   if (ideal_cmpxy->is_Cmp()) {
1977     // E.g., if we have CmpI(length - offset, count),
1978     // it might idealize to CmpI(length, count + offset)
1979     cmp_op = ideal_cmpxy->Opcode();
1980     xkey = ideal_cmpxy->in(1);
1981     ykey = ideal_cmpxy->in(2);
1982   }
1983 
1984   // Start by locating any relevant comparisons.
1985   Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1986   Node* cmpxy = NULL;
1987   Node* cmpyx = NULL;
1988   for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1989     Node* cmp = start_from->fast_out(k);
1990     if (cmp->outcnt() > 0 &&            // must have prior uses
1991         cmp->in(0) == NULL &&           // must be context-independent
1992         cmp->Opcode() == cmp_op) {      // right kind of compare
1993       if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
1994       if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
1995     }
1996   }
1997 
1998   const int NCMPS = 2;
1999   Node* cmps[NCMPS] = { cmpxy, cmpyx };
2000   int cmpn;
2001   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2002     if (cmps[cmpn] != NULL)  break;     // find a result
2003   }
2004   if (cmpn < NCMPS) {
2005     // Look for a dominating test that tells us the min and max.
2006     int depth = 0;                // Limit search depth for speed
2007     Node* dom = control();
2008     for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2009       if (++depth >= 100)  break;
2010       Node* ifproj = dom;
2011       if (!ifproj->is_Proj())  continue;
2012       Node* iff = ifproj->in(0);
2013       if (!iff->is_If())  continue;
2014       Node* bol = iff->in(1);
2015       if (!bol->is_Bool())  continue;
2016       Node* cmp = bol->in(1);
2017       if (cmp == NULL)  continue;
2018       for (cmpn = 0; cmpn < NCMPS; cmpn++)
2019         if (cmps[cmpn] == cmp)  break;
2020       if (cmpn == NCMPS)  continue;
2021       BoolTest::mask btest = bol->as_Bool()->_test._test;
2022       if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
2023       if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
2024       // At this point, we know that 'x btest y' is true.
2025       switch (btest) {
2026       case BoolTest::eq:
2027         // They are proven equal, so we can collapse the min/max.
2028         // Either value is the answer.  Choose the simpler.
2029         if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2030           return yvalue;
2031         return xvalue;
2032       case BoolTest::lt:          // x < y
2033       case BoolTest::le:          // x <= y
2034         return (want_max ? yvalue : xvalue);
2035       case BoolTest::gt:          // x > y
2036       case BoolTest::ge:          // x >= y
2037         return (want_max ? xvalue : yvalue);
2038       }
2039     }
2040   }
2041 
2042   // We failed to find a dominating test.
2043   // Let's pick a test that might GVN with prior tests.
2044   Node*          best_bol   = NULL;
2045   BoolTest::mask best_btest = BoolTest::illegal;
2046   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2047     Node* cmp = cmps[cmpn];
2048     if (cmp == NULL)  continue;
2049     for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2050       Node* bol = cmp->fast_out(j);
2051       if (!bol->is_Bool())  continue;
2052       BoolTest::mask btest = bol->as_Bool()->_test._test;
2053       if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
2054       if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
2055       if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2056         best_bol   = bol->as_Bool();
2057         best_btest = btest;
2058       }
2059     }
2060   }
2061 
2062   Node* answer_if_true  = NULL;
2063   Node* answer_if_false = NULL;
2064   switch (best_btest) {
2065   default:
2066     if (cmpxy == NULL)
2067       cmpxy = ideal_cmpxy;
2068     best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2069     // and fall through:
2070   case BoolTest::lt:          // x < y
2071   case BoolTest::le:          // x <= y
2072     answer_if_true  = (want_max ? yvalue : xvalue);
2073     answer_if_false = (want_max ? xvalue : yvalue);
2074     break;
2075   case BoolTest::gt:          // x > y
2076   case BoolTest::ge:          // x >= y
2077     answer_if_true  = (want_max ? xvalue : yvalue);
2078     answer_if_false = (want_max ? yvalue : xvalue);
2079     break;
2080   }
2081 
2082   jint hi, lo;
2083   if (want_max) {
2084     // We can sharpen the minimum.
2085     hi = MAX2(txvalue->_hi, tyvalue->_hi);
2086     lo = MAX2(txvalue->_lo, tyvalue->_lo);
2087   } else {
2088     // We can sharpen the maximum.
2089     hi = MIN2(txvalue->_hi, tyvalue->_hi);
2090     lo = MIN2(txvalue->_lo, tyvalue->_lo);
2091   }
2092 
2093   // Use a flow-free graph structure, to avoid creating excess control edges
2094   // which could hinder other optimizations.
2095   // Since Math.min/max is often used with arraycopy, we want
2096   // tightly_coupled_allocation to be able to see beyond min/max expressions.
2097   Node* cmov = CMoveNode::make(NULL, best_bol,
2098                                answer_if_false, answer_if_true,
2099                                TypeInt::make(lo, hi, widen));
2100 
2101   return _gvn.transform(cmov);
2102 
2103   /*
2104   // This is not as desirable as it may seem, since Min and Max
2105   // nodes do not have a full set of optimizations.
2106   // And they would interfere, anyway, with 'if' optimizations
2107   // and with CMoveI canonical forms.
2108   switch (id) {
2109   case vmIntrinsics::_min:
2110     result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2111   case vmIntrinsics::_max:
2112     result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2113   default:
2114     ShouldNotReachHere();
2115   }
2116   */
2117 }
2118 
2119 inline int
2120 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2121   const TypePtr* base_type = TypePtr::NULL_PTR;
2122   if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
2123   if (base_type == NULL) {
2124     // Unknown type.
2125     return Type::AnyPtr;
2126   } else if (base_type == TypePtr::NULL_PTR) {
2127     // Since this is a NULL+long form, we have to switch to a rawptr.
2128     base   = _gvn.transform(new CastX2PNode(offset));
2129     offset = MakeConX(0);
2130     return Type::RawPtr;
2131   } else if (base_type->base() == Type::RawPtr) {
2132     return Type::RawPtr;
2133   } else if (base_type->isa_oopptr()) {
2134     // Base is never null => always a heap address.
2135     if (base_type->ptr() == TypePtr::NotNull) {
2136       return Type::OopPtr;
2137     }
2138     // Offset is small => always a heap address.
2139     const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2140     if (offset_type != NULL &&
2141         base_type->offset() == 0 &&     // (should always be?)
2142         offset_type->_lo >= 0 &&
2143         !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2144       return Type::OopPtr;
2145     }
2146     // Otherwise, it might either be oop+off or NULL+addr.
2147     return Type::AnyPtr;
2148   } else {
2149     // No information:
2150     return Type::AnyPtr;
2151   }
2152 }
2153 
2154 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2155   int kind = classify_unsafe_addr(base, offset);
2156   if (kind == Type::RawPtr) {
2157     return basic_plus_adr(top(), base, offset);
2158   } else {
2159     return basic_plus_adr(base, offset);
2160   }
2161 }
2162 
2163 //--------------------------inline_number_methods-----------------------------
2164 // inline int     Integer.numberOfLeadingZeros(int)
2165 // inline int        Long.numberOfLeadingZeros(long)
2166 //
2167 // inline int     Integer.numberOfTrailingZeros(int)
2168 // inline int        Long.numberOfTrailingZeros(long)
2169 //
2170 // inline int     Integer.bitCount(int)
2171 // inline int        Long.bitCount(long)
2172 //
2173 // inline char  Character.reverseBytes(char)
2174 // inline short     Short.reverseBytes(short)
2175 // inline int     Integer.reverseBytes(int)
2176 // inline long       Long.reverseBytes(long)
2177 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2178   Node* arg = argument(0);
2179   Node* n = NULL;
2180   switch (id) {
2181   case vmIntrinsics::_numberOfLeadingZeros_i:   n = new CountLeadingZerosINode( arg);  break;
2182   case vmIntrinsics::_numberOfLeadingZeros_l:   n = new CountLeadingZerosLNode( arg);  break;
2183   case vmIntrinsics::_numberOfTrailingZeros_i:  n = new CountTrailingZerosINode(arg);  break;
2184   case vmIntrinsics::_numberOfTrailingZeros_l:  n = new CountTrailingZerosLNode(arg);  break;
2185   case vmIntrinsics::_bitCount_i:               n = new PopCountINode(          arg);  break;
2186   case vmIntrinsics::_bitCount_l:               n = new PopCountLNode(          arg);  break;
2187   case vmIntrinsics::_reverseBytes_c:           n = new ReverseBytesUSNode(0,   arg);  break;
2188   case vmIntrinsics::_reverseBytes_s:           n = new ReverseBytesSNode( 0,   arg);  break;
2189   case vmIntrinsics::_reverseBytes_i:           n = new ReverseBytesINode( 0,   arg);  break;
2190   case vmIntrinsics::_reverseBytes_l:           n = new ReverseBytesLNode( 0,   arg);  break;
2191   default:  fatal_unexpected_iid(id);  break;
2192   }
2193   set_result(_gvn.transform(n));
2194   return true;
2195 }
2196 
2197 //----------------------------inline_unsafe_access----------------------------
2198 
2199 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2200 
2201 // Helper that guards and inserts a pre-barrier.
2202 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2203                                         Node* pre_val, bool need_mem_bar) {
2204   // We could be accessing the referent field of a reference object. If so, when G1
2205   // is enabled, we need to log the value in the referent field in an SATB buffer.
2206   // This routine performs some compile time filters and generates suitable
2207   // runtime filters that guard the pre-barrier code.
2208   // Also add memory barrier for non volatile load from the referent field
2209   // to prevent commoning of loads across safepoint.
2210   if (!UseG1GC && !need_mem_bar)
2211     return;
2212 
2213   // Some compile time checks.
2214 
2215   // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2216   const TypeX* otype = offset->find_intptr_t_type();
2217   if (otype != NULL && otype->is_con() &&
2218       otype->get_con() != java_lang_ref_Reference::referent_offset) {
2219     // Constant offset but not the reference_offset so just return
2220     return;
2221   }
2222 
2223   // We only need to generate the runtime guards for instances.
2224   const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2225   if (btype != NULL) {
2226     if (btype->isa_aryptr()) {
2227       // Array type so nothing to do
2228       return;
2229     }
2230 
2231     const TypeInstPtr* itype = btype->isa_instptr();
2232     if (itype != NULL) {
2233       // Can the klass of base_oop be statically determined to be
2234       // _not_ a sub-class of Reference and _not_ Object?
2235       ciKlass* klass = itype->klass();
2236       if ( klass->is_loaded() &&
2237           !klass->is_subtype_of(env()->Reference_klass()) &&
2238           !env()->Object_klass()->is_subtype_of(klass)) {
2239         return;
2240       }
2241     }
2242   }
2243 
2244   // The compile time filters did not reject base_oop/offset so
2245   // we need to generate the following runtime filters
2246   //
2247   // if (offset == java_lang_ref_Reference::_reference_offset) {
2248   //   if (instance_of(base, java.lang.ref.Reference)) {
2249   //     pre_barrier(_, pre_val, ...);
2250   //   }
2251   // }
2252 
2253   float likely   = PROB_LIKELY(  0.999);
2254   float unlikely = PROB_UNLIKELY(0.999);
2255 
2256   IdealKit ideal(this);
2257 #define __ ideal.
2258 
2259   Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2260 
2261   __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2262       // Update graphKit memory and control from IdealKit.
2263       sync_kit(ideal);
2264 
2265       Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2266       Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2267 
2268       // Update IdealKit memory and control from graphKit.
2269       __ sync_kit(this);
2270 
2271       Node* one = __ ConI(1);
2272       // is_instof == 0 if base_oop == NULL
2273       __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2274 
2275         // Update graphKit from IdeakKit.
2276         sync_kit(ideal);
2277 
2278         // Use the pre-barrier to record the value in the referent field
2279         pre_barrier(false /* do_load */,
2280                     __ ctrl(),
2281                     NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2282                     pre_val /* pre_val */,
2283                     T_OBJECT);
2284         if (need_mem_bar) {
2285           // Add memory barrier to prevent commoning reads from this field
2286           // across safepoint since GC can change its value.
2287           insert_mem_bar(Op_MemBarCPUOrder);
2288         }
2289         // Update IdealKit from graphKit.
2290         __ sync_kit(this);
2291 
2292       } __ end_if(); // _ref_type != ref_none
2293   } __ end_if(); // offset == referent_offset
2294 
2295   // Final sync IdealKit and GraphKit.
2296   final_sync(ideal);
2297 #undef __
2298 }
2299 
2300 
2301 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2302   // Attempt to infer a sharper value type from the offset and base type.
2303   ciKlass* sharpened_klass = NULL;
2304 
2305   // See if it is an instance field, with an object type.
2306   if (alias_type->field() != NULL) {
2307     assert(!is_native_ptr, "native pointer op cannot use a java address");
2308     if (alias_type->field()->type()->is_klass()) {
2309       sharpened_klass = alias_type->field()->type()->as_klass();
2310     }
2311   }
2312 
2313   // See if it is a narrow oop array.
2314   if (adr_type->isa_aryptr()) {
2315     if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2316       const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2317       if (elem_type != NULL) {
2318         sharpened_klass = elem_type->klass();
2319       }
2320     }
2321   }
2322 
2323   // The sharpened class might be unloaded if there is no class loader
2324   // contraint in place.
2325   if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2326     const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2327 
2328 #ifndef PRODUCT
2329     if (C->print_intrinsics() || C->print_inlining()) {
2330       tty->print("  from base type: ");  adr_type->dump();
2331       tty->print("  sharpened value: ");  tjp->dump();
2332     }
2333 #endif
2334     // Sharpen the value type.
2335     return tjp;
2336   }
2337   return NULL;
2338 }
2339 
2340 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2341   if (callee()->is_static())  return false;  // caller must have the capability!
2342   guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2343   guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2344   assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2345 
2346 #ifndef PRODUCT
2347   {
2348     ResourceMark rm;
2349     // Check the signatures.
2350     ciSignature* sig = callee()->signature();
2351 #ifdef ASSERT
2352     if (!is_store) {
2353       // Object getObject(Object base, int/long offset), etc.
2354       BasicType rtype = sig->return_type()->basic_type();
2355       if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2356           rtype = T_ADDRESS;  // it is really a C void*
2357       assert(rtype == type, "getter must return the expected value");
2358       if (!is_native_ptr) {
2359         assert(sig->count() == 2, "oop getter has 2 arguments");
2360         assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2361         assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2362       } else {
2363         assert(sig->count() == 1, "native getter has 1 argument");
2364         assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2365       }
2366     } else {
2367       // void putObject(Object base, int/long offset, Object x), etc.
2368       assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2369       if (!is_native_ptr) {
2370         assert(sig->count() == 3, "oop putter has 3 arguments");
2371         assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2372         assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2373       } else {
2374         assert(sig->count() == 2, "native putter has 2 arguments");
2375         assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2376       }
2377       BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2378       if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2379         vtype = T_ADDRESS;  // it is really a C void*
2380       assert(vtype == type, "putter must accept the expected value");
2381     }
2382 #endif // ASSERT
2383  }
2384 #endif //PRODUCT
2385 
2386   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2387 
2388   Node* receiver = argument(0);  // type: oop
2389 
2390   // Build address expression.
2391   Node* adr;
2392   Node* heap_base_oop = top();
2393   Node* offset = top();
2394   Node* val;
2395 
2396   if (!is_native_ptr) {
2397     // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2398     Node* base = argument(1);  // type: oop
2399     // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2400     offset = argument(2);  // type: long
2401     // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2402     // to be plain byte offsets, which are also the same as those accepted
2403     // by oopDesc::field_base.
2404     assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2405            "fieldOffset must be byte-scaled");
2406     // 32-bit machines ignore the high half!
2407     offset = ConvL2X(offset);
2408     adr = make_unsafe_address(base, offset);
2409     heap_base_oop = base;
2410     val = is_store ? argument(4) : NULL;
2411   } else {
2412     Node* ptr = argument(1);  // type: long
2413     ptr = ConvL2X(ptr);  // adjust Java long to machine word
2414     adr = make_unsafe_address(NULL, ptr);
2415     val = is_store ? argument(3) : NULL;
2416   }
2417 
2418   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2419 
2420   // Try to categorize the address.  If it comes up as TypeJavaPtr::BOTTOM,
2421   // there was not enough information to nail it down.
2422   Compile::AliasType* alias_type = C->alias_type(adr_type);
2423   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2424 
2425   assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2426          alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2427   bool mismatched = false;
2428   BasicType bt = alias_type->basic_type();
2429   if (bt != T_ILLEGAL) {
2430     if (bt == T_BYTE && adr_type->isa_aryptr()) {
2431       // Alias type doesn't differentiate between byte[] and boolean[]).
2432       // Use address type to get the element type.
2433       bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2434     }
2435     if (bt == T_ARRAY || bt == T_NARROWOOP) {
2436       // accessing an array field with getObject is not a mismatch
2437       bt = T_OBJECT;
2438     }
2439     if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2440       // Don't intrinsify mismatched object accesses
2441       return false;
2442     }
2443     mismatched = (bt != type);
2444   }
2445 
2446   // First guess at the value type.
2447   const Type *value_type = Type::get_const_basic_type(type);
2448 
2449   // We will need memory barriers unless we can determine a unique
2450   // alias category for this reference.  (Note:  If for some reason
2451   // the barriers get omitted and the unsafe reference begins to "pollute"
2452   // the alias analysis of the rest of the graph, either Compile::can_alias
2453   // or Compile::must_alias will throw a diagnostic assert.)
2454   bool need_mem_bar;
2455   switch (kind) {
2456       case Relaxed:
2457           need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2458           break;
2459       case Opaque:
2460           // Opaque uses CPUOrder membars for protection against code movement.
2461       case Acquire:
2462       case Release:
2463       case Volatile:
2464           need_mem_bar = true;
2465           break;
2466       default:
2467           ShouldNotReachHere();
2468   }
2469 
2470   // Some accesses require access atomicity for all types, notably longs and doubles.
2471   // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2472   bool requires_atomic_access = false;
2473   switch (kind) {
2474       case Relaxed:
2475       case Opaque:
2476           requires_atomic_access = AlwaysAtomicAccesses;
2477           break;
2478       case Acquire:
2479       case Release:
2480       case Volatile:
2481           requires_atomic_access = true;
2482           break;
2483       default:
2484           ShouldNotReachHere();
2485   }
2486 
2487   // Figure out the memory ordering.
2488   // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2489   MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2490 
2491   // If we are reading the value of the referent field of a Reference
2492   // object (either by using Unsafe directly or through reflection)
2493   // then, if G1 is enabled, we need to record the referent in an
2494   // SATB log buffer using the pre-barrier mechanism.
2495   // Also we need to add memory barrier to prevent commoning reads
2496   // from this field across safepoint since GC can change its value.
2497   bool need_read_barrier = !is_native_ptr && !is_store &&
2498                            offset != top() && heap_base_oop != top();
2499 
2500   if (!is_store && type == T_OBJECT) {
2501     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2502     if (tjp != NULL) {
2503       value_type = tjp;
2504     }
2505   }
2506 
2507   receiver = null_check(receiver);
2508   if (stopped()) {
2509     return true;
2510   }
2511   // Heap pointers get a null-check from the interpreter,
2512   // as a courtesy.  However, this is not guaranteed by Unsafe,
2513   // and it is not possible to fully distinguish unintended nulls
2514   // from intended ones in this API.
2515 
2516   // We need to emit leading and trailing CPU membars (see below) in
2517   // addition to memory membars for special access modes. This is a little
2518   // too strong, but avoids the need to insert per-alias-type
2519   // volatile membars (for stores; compare Parse::do_put_xxx), which
2520   // we cannot do effectively here because we probably only have a
2521   // rough approximation of type.
2522 
2523   switch(kind) {
2524     case Relaxed:
2525     case Opaque:
2526     case Acquire:
2527       break;
2528     case Release:
2529     case Volatile:
2530       if (is_store) {
2531         insert_mem_bar(Op_MemBarRelease);
2532       } else {
2533         if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2534           insert_mem_bar(Op_MemBarVolatile);
2535         }
2536       }
2537       break;
2538     default:
2539       ShouldNotReachHere();
2540   }
2541 
2542   // Memory barrier to prevent normal and 'unsafe' accesses from
2543   // bypassing each other.  Happens after null checks, so the
2544   // exception paths do not take memory state from the memory barrier,
2545   // so there's no problems making a strong assert about mixing users
2546   // of safe & unsafe memory.
2547   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2548 
2549   if (!is_store) {
2550     Node* p = NULL;
2551     // Try to constant fold a load from a constant field
2552     ciField* field = alias_type->field();
2553     if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2554       // final or stable field
2555       p = make_constant_from_field(field, heap_base_oop);
2556     }
2557     if (p == NULL) {
2558       // To be valid, unsafe loads may depend on other conditions than
2559       // the one that guards them: pin the Load node
2560       p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2561       // load value
2562       switch (type) {
2563       case T_BOOLEAN:
2564       case T_CHAR:
2565       case T_BYTE:
2566       case T_SHORT:
2567       case T_INT:
2568       case T_LONG:
2569       case T_FLOAT:
2570       case T_DOUBLE:
2571         break;
2572       case T_OBJECT:
2573         if (need_read_barrier) {
2574           // We do not require a mem bar inside pre_barrier if need_mem_bar
2575           // is set: the barriers would be emitted by us.
2576           insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2577         }
2578         break;
2579       case T_ADDRESS:
2580         // Cast to an int type.
2581         p = _gvn.transform(new CastP2XNode(NULL, p));
2582         p = ConvX2UL(p);
2583         break;
2584       default:
2585         fatal("unexpected type %d: %s", type, type2name(type));
2586         break;
2587       }
2588     }
2589     // The load node has the control of the preceding MemBarCPUOrder.  All
2590     // following nodes will have the control of the MemBarCPUOrder inserted at
2591     // the end of this method.  So, pushing the load onto the stack at a later
2592     // point is fine.
2593     set_result(p);
2594   } else {
2595     // place effect of store into memory
2596     switch (type) {
2597     case T_DOUBLE:
2598       val = dstore_rounding(val);
2599       break;
2600     case T_ADDRESS:
2601       // Repackage the long as a pointer.
2602       val = ConvL2X(val);
2603       val = _gvn.transform(new CastX2PNode(val));
2604       break;
2605     }
2606 
2607     if (type != T_OBJECT) {
2608       (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2609     } else {
2610       // Possibly an oop being stored to Java heap or native memory
2611       if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2612         // oop to Java heap.
2613         (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2614       } else {
2615         // We can't tell at compile time if we are storing in the Java heap or outside
2616         // of it. So we need to emit code to conditionally do the proper type of
2617         // store.
2618 
2619         IdealKit ideal(this);
2620 #define __ ideal.
2621         // QQQ who knows what probability is here??
2622         __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2623           // Sync IdealKit and graphKit.
2624           sync_kit(ideal);
2625           Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2626           // Update IdealKit memory.
2627           __ sync_kit(this);
2628         } __ else_(); {
2629           __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2630         } __ end_if();
2631         // Final sync IdealKit and GraphKit.
2632         final_sync(ideal);
2633 #undef __
2634       }
2635     }
2636   }
2637 
2638   switch(kind) {
2639     case Relaxed:
2640     case Opaque:
2641     case Release:
2642       break;
2643     case Acquire:
2644     case Volatile:
2645       if (!is_store) {
2646         insert_mem_bar(Op_MemBarAcquire);
2647       } else {
2648         if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2649           insert_mem_bar(Op_MemBarVolatile);
2650         }
2651       }
2652       break;
2653     default:
2654       ShouldNotReachHere();
2655   }
2656 
2657   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2658 
2659   return true;
2660 }
2661 
2662 //----------------------------inline_unsafe_load_store----------------------------
2663 // This method serves a couple of different customers (depending on LoadStoreKind):
2664 //
2665 // LS_cmp_swap:
2666 //
2667 //   boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2668 //   boolean compareAndSwapInt(   Object o, long offset, int    expected, int    x);
2669 //   boolean compareAndSwapLong(  Object o, long offset, long   expected, long   x);
2670 //
2671 // LS_cmp_swap_weak:
2672 //
2673 //   boolean weakCompareAndSwapObject(       Object o, long offset, Object expected, Object x);
2674 //   boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2675 //   boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2676 //
2677 //   boolean weakCompareAndSwapInt(          Object o, long offset, int    expected, int    x);
2678 //   boolean weakCompareAndSwapIntAcquire(   Object o, long offset, int    expected, int    x);
2679 //   boolean weakCompareAndSwapIntRelease(   Object o, long offset, int    expected, int    x);
2680 //
2681 //   boolean weakCompareAndSwapLong(         Object o, long offset, long   expected, long   x);
2682 //   boolean weakCompareAndSwapLongAcquire(  Object o, long offset, long   expected, long   x);
2683 //   boolean weakCompareAndSwapLongRelease(  Object o, long offset, long   expected, long   x);
2684 //
2685 // LS_cmp_exchange:
2686 //
2687 //   Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2688 //   Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2689 //   Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2690 //
2691 //   Object compareAndExchangeIntVolatile(   Object o, long offset, Object expected, Object x);
2692 //   Object compareAndExchangeIntAcquire(    Object o, long offset, Object expected, Object x);
2693 //   Object compareAndExchangeIntRelease(    Object o, long offset, Object expected, Object x);
2694 //
2695 //   Object compareAndExchangeLongVolatile(  Object o, long offset, Object expected, Object x);
2696 //   Object compareAndExchangeLongAcquire(   Object o, long offset, Object expected, Object x);
2697 //   Object compareAndExchangeLongRelease(   Object o, long offset, Object expected, Object x);
2698 //
2699 // LS_get_add:
2700 //
2701 //   int  getAndAddInt( Object o, long offset, int  delta)
2702 //   long getAndAddLong(Object o, long offset, long delta)
2703 //
2704 // LS_get_set:
2705 //
2706 //   int    getAndSet(Object o, long offset, int    newValue)
2707 //   long   getAndSet(Object o, long offset, long   newValue)
2708 //   Object getAndSet(Object o, long offset, Object newValue)
2709 //
2710 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2711   // This basic scheme here is the same as inline_unsafe_access, but
2712   // differs in enough details that combining them would make the code
2713   // overly confusing.  (This is a true fact! I originally combined
2714   // them, but even I was confused by it!) As much code/comments as
2715   // possible are retained from inline_unsafe_access though to make
2716   // the correspondences clearer. - dl
2717 
2718   if (callee()->is_static())  return false;  // caller must have the capability!
2719 
2720 #ifndef PRODUCT
2721   BasicType rtype;
2722   {
2723     ResourceMark rm;
2724     // Check the signatures.
2725     ciSignature* sig = callee()->signature();
2726     rtype = sig->return_type()->basic_type();
2727     switch(kind) {
2728       case LS_get_add:
2729       case LS_get_set: {
2730       // Check the signatures.
2731 #ifdef ASSERT
2732       assert(rtype == type, "get and set must return the expected type");
2733       assert(sig->count() == 3, "get and set has 3 arguments");
2734       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2735       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2736       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2737 #endif // ASSERT
2738         break;
2739       }
2740       case LS_cmp_swap:
2741       case LS_cmp_swap_weak: {
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         break;
2750       }
2751       case LS_cmp_exchange: {
2752       // Check the signatures.
2753 #ifdef ASSERT
2754       assert(rtype == type, "CAS must return the expected type");
2755       assert(sig->count() == 4, "CAS has 4 arguments");
2756       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2757       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2758 #endif // ASSERT
2759         break;
2760       }
2761       default:
2762         ShouldNotReachHere();
2763     }
2764   }
2765 #endif //PRODUCT
2766 
2767   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2768 
2769   // Get arguments:
2770   Node* receiver = NULL;
2771   Node* base     = NULL;
2772   Node* offset   = NULL;
2773   Node* oldval   = NULL;
2774   Node* newval   = NULL;
2775   switch(kind) {
2776     case LS_cmp_swap:
2777     case LS_cmp_swap_weak:
2778     case LS_cmp_exchange: {
2779       const bool two_slot_type = type2size[type] == 2;
2780       receiver = argument(0);  // type: oop
2781       base     = argument(1);  // type: oop
2782       offset   = argument(2);  // type: long
2783       oldval   = argument(4);  // type: oop, int, or long
2784       newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2785       break;
2786     }
2787     case LS_get_add:
2788     case LS_get_set: {
2789       receiver = argument(0);  // type: oop
2790       base     = argument(1);  // type: oop
2791       offset   = argument(2);  // type: long
2792       oldval   = NULL;
2793       newval   = argument(4);  // type: oop, int, or long
2794       break;
2795     }
2796     default:
2797       ShouldNotReachHere();
2798   }
2799 
2800   // Null check receiver.
2801   receiver = null_check(receiver);
2802   if (stopped()) {
2803     return true;
2804   }
2805 
2806   // Build field offset expression.
2807   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2808   // to be plain byte offsets, which are also the same as those accepted
2809   // by oopDesc::field_base.
2810   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2811   // 32-bit machines ignore the high half of long offsets
2812   offset = ConvL2X(offset);
2813   Node* adr = make_unsafe_address(base, offset);
2814   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2815 
2816   Compile::AliasType* alias_type = C->alias_type(adr_type);
2817   assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2818          alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2819   BasicType bt = alias_type->basic_type();
2820   if (bt != T_ILLEGAL &&
2821       ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2822     // Don't intrinsify mismatched object accesses.
2823     return false;
2824   }
2825 
2826   // For CAS, unlike inline_unsafe_access, there seems no point in
2827   // trying to refine types. Just use the coarse types here.
2828   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2829   const Type *value_type = Type::get_const_basic_type(type);
2830 
2831   switch (kind) {
2832     case LS_get_set:
2833     case LS_cmp_exchange: {
2834       if (type == T_OBJECT) {
2835         const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2836         if (tjp != NULL) {
2837           value_type = tjp;
2838         }
2839       }
2840       break;
2841     }
2842     case LS_cmp_swap:
2843     case LS_cmp_swap_weak:
2844     case LS_get_add:
2845       break;
2846     default:
2847       ShouldNotReachHere();
2848   }
2849 
2850   int alias_idx = C->get_alias_index(adr_type);
2851 
2852   // Memory-model-wise, a LoadStore acts like a little synchronized
2853   // block, so needs barriers on each side.  These don't translate
2854   // into actual barriers on most machines, but we still need rest of
2855   // compiler to respect ordering.
2856 
2857   switch (access_kind) {
2858     case Relaxed:
2859     case Acquire:
2860       break;
2861     case Release:
2862     case Volatile:
2863       insert_mem_bar(Op_MemBarRelease);
2864       break;
2865     default:
2866       ShouldNotReachHere();
2867   }
2868   insert_mem_bar(Op_MemBarCPUOrder);
2869 
2870   // Figure out the memory ordering.
2871   MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2872 
2873   // 4984716: MemBars must be inserted before this
2874   //          memory node in order to avoid a false
2875   //          dependency which will confuse the scheduler.
2876   Node *mem = memory(alias_idx);
2877 
2878   // For now, we handle only those cases that actually exist: ints,
2879   // longs, and Object. Adding others should be straightforward.
2880   Node* load_store = NULL;
2881   switch(type) {
2882   case T_INT:
2883     switch(kind) {
2884       case LS_get_add:
2885         load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2886         break;
2887       case LS_get_set:
2888         load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2889         break;
2890       case LS_cmp_swap_weak:
2891         load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2892         break;
2893       case LS_cmp_swap:
2894         load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2895         break;
2896       case LS_cmp_exchange:
2897         load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2898         break;
2899       default:
2900         ShouldNotReachHere();
2901     }
2902     break;
2903   case T_LONG:
2904     switch(kind) {
2905       case LS_get_add:
2906         load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2907         break;
2908       case LS_get_set:
2909         load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2910         break;
2911       case LS_cmp_swap_weak:
2912         load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2913         break;
2914       case LS_cmp_swap:
2915         load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2916         break;
2917       case LS_cmp_exchange:
2918         load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2919         break;
2920       default:
2921         ShouldNotReachHere();
2922     }
2923     break;
2924   case T_OBJECT:
2925     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2926     // could be delayed during Parse (for example, in adjust_map_after_if()).
2927     // Execute transformation here to avoid barrier generation in such case.
2928     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2929       newval = _gvn.makecon(TypePtr::NULL_PTR);
2930 
2931     // Reference stores need a store barrier.
2932     switch(kind) {
2933       case LS_get_set: {
2934         // If pre-barrier must execute before the oop store, old value will require do_load here.
2935         if (!can_move_pre_barrier()) {
2936           pre_barrier(true /* do_load*/,
2937                       control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2938                       NULL /* pre_val*/,
2939                       T_OBJECT);
2940         } // Else move pre_barrier to use load_store value, see below.
2941         break;
2942       }
2943       case LS_cmp_swap_weak:
2944       case LS_cmp_swap:
2945       case LS_cmp_exchange: {
2946         // Same as for newval above:
2947         if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2948           oldval = _gvn.makecon(TypePtr::NULL_PTR);
2949         }
2950         // The only known value which might get overwritten is oldval.
2951         pre_barrier(false /* do_load */,
2952                     control(), NULL, NULL, max_juint, NULL, NULL,
2953                     oldval /* pre_val */,
2954                     T_OBJECT);
2955         break;
2956       }
2957       default:
2958         ShouldNotReachHere();
2959     }
2960 
2961 #ifdef _LP64
2962     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2963       Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2964 
2965       switch(kind) {
2966         case LS_get_set:
2967           load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2968           break;
2969         case LS_cmp_swap_weak: {
2970           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2971           load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2972           break;
2973         }
2974         case LS_cmp_swap: {
2975           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2976           load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2977           break;
2978         }
2979         case LS_cmp_exchange: {
2980           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2981           load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2982           break;
2983         }
2984         default:
2985           ShouldNotReachHere();
2986       }
2987     } else
2988 #endif
2989     switch (kind) {
2990       case LS_get_set:
2991         load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2992         break;
2993       case LS_cmp_swap_weak:
2994         load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2995         break;
2996       case LS_cmp_swap:
2997         load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2998         break;
2999       case LS_cmp_exchange:
3000         load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
3001         break;
3002       default:
3003         ShouldNotReachHere();
3004     }
3005 
3006     // Emit the post barrier only when the actual store happened. This makes sense
3007     // to check only for LS_cmp_* that can fail to set the value.
3008     // LS_cmp_exchange does not produce any branches by default, so there is no
3009     // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3010     // CompareAndExchange and move branches here, it would make sense to conditionalize
3011     // post_barriers for LS_cmp_exchange as well.
3012     //
3013     // CAS success path is marked more likely since we anticipate this is a performance
3014     // critical path, while CAS failure path can use the penalty for going through unlikely
3015     // path as backoff. Which is still better than doing a store barrier there.
3016     switch (kind) {
3017       case LS_get_set:
3018       case LS_cmp_exchange: {
3019         post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3020         break;
3021       }
3022       case LS_cmp_swap_weak:
3023       case LS_cmp_swap: {
3024         IdealKit ideal(this);
3025         ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3026           sync_kit(ideal);
3027           post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3028           ideal.sync_kit(this);
3029         } ideal.end_if();
3030         final_sync(ideal);
3031         break;
3032       }
3033       default:
3034         ShouldNotReachHere();
3035     }
3036     break;
3037   default:
3038     fatal("unexpected type %d: %s", type, type2name(type));
3039     break;
3040   }
3041 
3042   // SCMemProjNodes represent the memory state of a LoadStore. Their
3043   // main role is to prevent LoadStore nodes from being optimized away
3044   // when their results aren't used.
3045   Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3046   set_memory(proj, alias_idx);
3047 
3048   if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3049 #ifdef _LP64
3050     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3051       load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3052     }
3053 #endif
3054     if (can_move_pre_barrier()) {
3055       // Don't need to load pre_val. The old value is returned by load_store.
3056       // The pre_barrier can execute after the xchg as long as no safepoint
3057       // gets inserted between them.
3058       pre_barrier(false /* do_load */,
3059                   control(), NULL, NULL, max_juint, NULL, NULL,
3060                   load_store /* pre_val */,
3061                   T_OBJECT);
3062     }
3063   }
3064 
3065   // Add the trailing membar surrounding the access
3066   insert_mem_bar(Op_MemBarCPUOrder);
3067 
3068   switch (access_kind) {
3069     case Relaxed:
3070     case Release:
3071       break; // do nothing
3072     case Acquire:
3073     case Volatile:
3074       insert_mem_bar(Op_MemBarAcquire);
3075       break;
3076     default:
3077       ShouldNotReachHere();
3078   }
3079 
3080   assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3081   set_result(load_store);
3082   return true;
3083 }
3084 
3085 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3086   MemNode::MemOrd mo = MemNode::unset;
3087   switch(kind) {
3088     case Opaque:
3089     case Relaxed:  mo = MemNode::unordered; break;
3090     case Acquire:  mo = MemNode::acquire;   break;
3091     case Release:  mo = MemNode::release;   break;
3092     case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3093     default:
3094       ShouldNotReachHere();
3095   }
3096   guarantee(mo != MemNode::unset, "Should select memory ordering");
3097   return mo;
3098 }
3099 
3100 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3101   MemNode::MemOrd mo = MemNode::unset;
3102   switch(kind) {
3103     case Opaque:
3104     case Relaxed:  mo = MemNode::unordered; break;
3105     case Acquire:  mo = MemNode::acquire;   break;
3106     case Release:  mo = MemNode::release;   break;
3107     case Volatile: mo = MemNode::seqcst;    break;
3108     default:
3109       ShouldNotReachHere();
3110   }
3111   guarantee(mo != MemNode::unset, "Should select memory ordering");
3112   return mo;
3113 }
3114 
3115 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3116   // Regardless of form, don't allow previous ld/st to move down,
3117   // then issue acquire, release, or volatile mem_bar.
3118   insert_mem_bar(Op_MemBarCPUOrder);
3119   switch(id) {
3120     case vmIntrinsics::_loadFence:
3121       insert_mem_bar(Op_LoadFence);
3122       return true;
3123     case vmIntrinsics::_storeFence:
3124       insert_mem_bar(Op_StoreFence);
3125       return true;
3126     case vmIntrinsics::_fullFence:
3127       insert_mem_bar(Op_MemBarVolatile);
3128       return true;
3129     default:
3130       fatal_unexpected_iid(id);
3131       return false;
3132   }
3133 }
3134 
3135 bool LibraryCallKit::inline_onspinwait() {
3136   insert_mem_bar(Op_OnSpinWait);
3137   return true;
3138 }
3139 
3140 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3141   if (!kls->is_Con()) {
3142     return true;
3143   }
3144   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3145   if (klsptr == NULL) {
3146     return true;
3147   }
3148   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3149   // don't need a guard for a klass that is already initialized
3150   return !ik->is_initialized();
3151 }
3152 
3153 //----------------------------inline_unsafe_allocate---------------------------
3154 // public native Object Unsafe.allocateInstance(Class<?> cls);
3155 bool LibraryCallKit::inline_unsafe_allocate() {
3156   if (callee()->is_static())  return false;  // caller must have the capability!
3157 
3158   null_check_receiver();  // null-check, then ignore
3159   Node* cls = null_check(argument(1));
3160   if (stopped())  return true;
3161 
3162   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3163   kls = null_check(kls);
3164   if (stopped())  return true;  // argument was like int.class
3165 
3166   Node* test = NULL;
3167   if (LibraryCallKit::klass_needs_init_guard(kls)) {
3168     // Note:  The argument might still be an illegal value like
3169     // Serializable.class or Object[].class.   The runtime will handle it.
3170     // But we must make an explicit check for initialization.
3171     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3172     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3173     // can generate code to load it as unsigned byte.
3174     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3175     Node* bits = intcon(InstanceKlass::fully_initialized);
3176     test = _gvn.transform(new SubINode(inst, bits));
3177     // The 'test' is non-zero if we need to take a slow path.
3178   }
3179 
3180   Node* obj = new_instance(kls, test);
3181   set_result(obj);
3182   return true;
3183 }
3184 
3185 //------------------------inline_native_time_funcs--------------
3186 // inline code for System.currentTimeMillis() and System.nanoTime()
3187 // these have the same type and signature
3188 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3189   const TypeFunc* tf = OptoRuntime::void_long_Type();
3190   const TypePtr* no_memory_effects = NULL;
3191   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3192   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3193 #ifdef ASSERT
3194   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3195   assert(value_top == top(), "second value must be top");
3196 #endif
3197   set_result(value);
3198   return true;
3199 }
3200 
3201 //------------------------inline_native_currentThread------------------
3202 bool LibraryCallKit::inline_native_currentThread() {
3203   Node* junk = NULL;
3204   set_result(generate_current_thread(junk));
3205   return true;
3206 }
3207 
3208 //------------------------inline_native_isInterrupted------------------
3209 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3210 bool LibraryCallKit::inline_native_isInterrupted() {
3211   // Add a fast path to t.isInterrupted(clear_int):
3212   //   (t == Thread.current() &&
3213   //    (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3214   //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3215   // So, in the common case that the interrupt bit is false,
3216   // we avoid making a call into the VM.  Even if the interrupt bit
3217   // is true, if the clear_int argument is false, we avoid the VM call.
3218   // However, if the receiver is not currentThread, we must call the VM,
3219   // because there must be some locking done around the operation.
3220 
3221   // We only go to the fast case code if we pass two guards.
3222   // Paths which do not pass are accumulated in the slow_region.
3223 
3224   enum {
3225     no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
3226     no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
3227     slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
3228     PATH_LIMIT
3229   };
3230 
3231   // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3232   // out of the function.
3233   insert_mem_bar(Op_MemBarCPUOrder);
3234 
3235   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3236   PhiNode*    result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3237 
3238   RegionNode* slow_region = new RegionNode(1);
3239   record_for_igvn(slow_region);
3240 
3241   // (a) Receiving thread must be the current thread.
3242   Node* rec_thr = argument(0);
3243   Node* tls_ptr = NULL;
3244   Node* cur_thr = generate_current_thread(tls_ptr);
3245   Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3246   Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3247 
3248   generate_slow_guard(bol_thr, slow_region);
3249 
3250   // (b) Interrupt bit on TLS must be false.
3251   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3252   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3253   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3254 
3255   // Set the control input on the field _interrupted read to prevent it floating up.
3256   Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3257   Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3258   Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3259 
3260   IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3261 
3262   // First fast path:  if (!TLS._interrupted) return false;
3263   Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3264   result_rgn->init_req(no_int_result_path, false_bit);
3265   result_val->init_req(no_int_result_path, intcon(0));
3266 
3267   // drop through to next case
3268   set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3269 
3270 #ifndef TARGET_OS_FAMILY_windows
3271   // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3272   Node* clr_arg = argument(1);
3273   Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3274   Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3275   IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3276 
3277   // Second fast path:  ... else if (!clear_int) return true;
3278   Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3279   result_rgn->init_req(no_clear_result_path, false_arg);
3280   result_val->init_req(no_clear_result_path, intcon(1));
3281 
3282   // drop through to next case
3283   set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3284 #else
3285   // To return true on Windows you must read the _interrupted field
3286   // and check the event state i.e. take the slow path.
3287 #endif // TARGET_OS_FAMILY_windows
3288 
3289   // (d) Otherwise, go to the slow path.
3290   slow_region->add_req(control());
3291   set_control( _gvn.transform(slow_region));
3292 
3293   if (stopped()) {
3294     // There is no slow path.
3295     result_rgn->init_req(slow_result_path, top());
3296     result_val->init_req(slow_result_path, top());
3297   } else {
3298     // non-virtual because it is a private non-static
3299     CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3300 
3301     Node* slow_val = set_results_for_java_call(slow_call);
3302     // this->control() comes from set_results_for_java_call
3303 
3304     Node* fast_io  = slow_call->in(TypeFunc::I_O);
3305     Node* fast_mem = slow_call->in(TypeFunc::Memory);
3306 
3307     // These two phis are pre-filled with copies of of the fast IO and Memory
3308     PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3309     PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
3310 
3311     result_rgn->init_req(slow_result_path, control());
3312     result_io ->init_req(slow_result_path, i_o());
3313     result_mem->init_req(slow_result_path, reset_memory());
3314     result_val->init_req(slow_result_path, slow_val);
3315 
3316     set_all_memory(_gvn.transform(result_mem));
3317     set_i_o(       _gvn.transform(result_io));
3318   }
3319 
3320   C->set_has_split_ifs(true); // Has chance for split-if optimization
3321   set_result(result_rgn, result_val);
3322   return true;
3323 }
3324 
3325 //---------------------------load_mirror_from_klass----------------------------
3326 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3327 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3328   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3329   return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3330 }
3331 
3332 //-----------------------load_klass_from_mirror_common-------------------------
3333 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3334 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3335 // and branch to the given path on the region.
3336 // If never_see_null, take an uncommon trap on null, so we can optimistically
3337 // compile for the non-null case.
3338 // If the region is NULL, force never_see_null = true.
3339 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3340                                                     bool never_see_null,
3341                                                     RegionNode* region,
3342                                                     int null_path,
3343                                                     int offset) {
3344   if (region == NULL)  never_see_null = true;
3345   Node* p = basic_plus_adr(mirror, offset);
3346   const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3347   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3348   Node* null_ctl = top();
3349   kls = null_check_oop(kls, &null_ctl, never_see_null);
3350   if (region != NULL) {
3351     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3352     region->init_req(null_path, null_ctl);
3353   } else {
3354     assert(null_ctl == top(), "no loose ends");
3355   }
3356   return kls;
3357 }
3358 
3359 //--------------------(inline_native_Class_query helpers)---------------------
3360 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3361 // Fall through if (mods & mask) == bits, take the guard otherwise.
3362 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3363   // Branch around if the given klass has the given modifier bit set.
3364   // Like generate_guard, adds a new path onto the region.
3365   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3366   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3367   Node* mask = intcon(modifier_mask);
3368   Node* bits = intcon(modifier_bits);
3369   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3370   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3371   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3372   return generate_fair_guard(bol, region);
3373 }
3374 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3375   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3376 }
3377 
3378 //-------------------------inline_native_Class_query-------------------
3379 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3380   const Type* return_type = TypeInt::BOOL;
3381   Node* prim_return_value = top();  // what happens if it's a primitive class?
3382   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3383   bool expect_prim = false;     // most of these guys expect to work on refs
3384 
3385   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3386 
3387   Node* mirror = argument(0);
3388   Node* obj    = top();
3389 
3390   switch (id) {
3391   case vmIntrinsics::_isInstance:
3392     // nothing is an instance of a primitive type
3393     prim_return_value = intcon(0);
3394     obj = argument(1);
3395     break;
3396   case vmIntrinsics::_getModifiers:
3397     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3398     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3399     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3400     break;
3401   case vmIntrinsics::_isInterface:
3402     prim_return_value = intcon(0);
3403     break;
3404   case vmIntrinsics::_isArray:
3405     prim_return_value = intcon(0);
3406     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3407     break;
3408   case vmIntrinsics::_isPrimitive:
3409     prim_return_value = intcon(1);
3410     expect_prim = true;  // obviously
3411     break;
3412   case vmIntrinsics::_getSuperclass:
3413     prim_return_value = null();
3414     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3415     break;
3416   case vmIntrinsics::_getClassAccessFlags:
3417     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3418     return_type = TypeInt::INT;  // not bool!  6297094
3419     break;
3420   default:
3421     fatal_unexpected_iid(id);
3422     break;
3423   }
3424 
3425   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3426   if (mirror_con == NULL)  return false;  // cannot happen?
3427 
3428 #ifndef PRODUCT
3429   if (C->print_intrinsics() || C->print_inlining()) {
3430     ciType* k = mirror_con->java_mirror_type();
3431     if (k) {
3432       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3433       k->print_name();
3434       tty->cr();
3435     }
3436   }
3437 #endif
3438 
3439   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3440   RegionNode* region = new RegionNode(PATH_LIMIT);
3441   record_for_igvn(region);
3442   PhiNode* phi = new PhiNode(region, return_type);
3443 
3444   // The mirror will never be null of Reflection.getClassAccessFlags, however
3445   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3446   // if it is. See bug 4774291.
3447 
3448   // For Reflection.getClassAccessFlags(), the null check occurs in
3449   // the wrong place; see inline_unsafe_access(), above, for a similar
3450   // situation.
3451   mirror = null_check(mirror);
3452   // If mirror or obj is dead, only null-path is taken.
3453   if (stopped())  return true;
3454 
3455   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3456 
3457   // Now load the mirror's klass metaobject, and null-check it.
3458   // Side-effects region with the control path if the klass is null.
3459   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3460   // If kls is null, we have a primitive mirror.
3461   phi->init_req(_prim_path, prim_return_value);
3462   if (stopped()) { set_result(region, phi); return true; }
3463   bool safe_for_replace = (region->in(_prim_path) == top());
3464 
3465   Node* p;  // handy temp
3466   Node* null_ctl;
3467 
3468   // Now that we have the non-null klass, we can perform the real query.
3469   // For constant classes, the query will constant-fold in LoadNode::Value.
3470   Node* query_value = top();
3471   switch (id) {
3472   case vmIntrinsics::_isInstance:
3473     // nothing is an instance of a primitive type
3474     query_value = gen_instanceof(obj, kls, safe_for_replace);
3475     break;
3476 
3477   case vmIntrinsics::_getModifiers:
3478     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3479     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3480     break;
3481 
3482   case vmIntrinsics::_isInterface:
3483     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3484     if (generate_interface_guard(kls, region) != NULL)
3485       // A guard was added.  If the guard is taken, it was an interface.
3486       phi->add_req(intcon(1));
3487     // If we fall through, it's a plain class.
3488     query_value = intcon(0);
3489     break;
3490 
3491   case vmIntrinsics::_isArray:
3492     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3493     if (generate_array_guard(kls, region) != NULL)
3494       // A guard was added.  If the guard is taken, it was an array.
3495       phi->add_req(intcon(1));
3496     // If we fall through, it's a plain class.
3497     query_value = intcon(0);
3498     break;
3499 
3500   case vmIntrinsics::_isPrimitive:
3501     query_value = intcon(0); // "normal" path produces false
3502     break;
3503 
3504   case vmIntrinsics::_getSuperclass:
3505     // The rules here are somewhat unfortunate, but we can still do better
3506     // with random logic than with a JNI call.
3507     // Interfaces store null or Object as _super, but must report null.
3508     // Arrays store an intermediate super as _super, but must report Object.
3509     // Other types can report the actual _super.
3510     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3511     if (generate_interface_guard(kls, region) != NULL)
3512       // A guard was added.  If the guard is taken, it was an interface.
3513       phi->add_req(null());
3514     if (generate_array_guard(kls, region) != NULL)
3515       // A guard was added.  If the guard is taken, it was an array.
3516       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3517     // If we fall through, it's a plain class.  Get its _super.
3518     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3519     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3520     null_ctl = top();
3521     kls = null_check_oop(kls, &null_ctl);
3522     if (null_ctl != top()) {
3523       // If the guard is taken, Object.superClass is null (both klass and mirror).
3524       region->add_req(null_ctl);
3525       phi   ->add_req(null());
3526     }
3527     if (!stopped()) {
3528       query_value = load_mirror_from_klass(kls);
3529     }
3530     break;
3531 
3532   case vmIntrinsics::_getClassAccessFlags:
3533     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3534     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3535     break;
3536 
3537   default:
3538     fatal_unexpected_iid(id);
3539     break;
3540   }
3541 
3542   // Fall-through is the normal case of a query to a real class.
3543   phi->init_req(1, query_value);
3544   region->init_req(1, control());
3545 
3546   C->set_has_split_ifs(true); // Has chance for split-if optimization
3547   set_result(region, phi);
3548   return true;
3549 }
3550 
3551 //-------------------------inline_Class_cast-------------------
3552 bool LibraryCallKit::inline_Class_cast() {
3553   Node* mirror = argument(0); // Class
3554   Node* obj    = argument(1);
3555   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3556   if (mirror_con == NULL) {
3557     return false;  // dead path (mirror->is_top()).
3558   }
3559   if (obj == NULL || obj->is_top()) {
3560     return false;  // dead path
3561   }
3562   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3563 
3564   // First, see if Class.cast() can be folded statically.
3565   // java_mirror_type() returns non-null for compile-time Class constants.
3566   ciType* tm = mirror_con->java_mirror_type();
3567   if (tm != NULL && tm->is_klass() &&
3568       tp != NULL && tp->klass() != NULL) {
3569     if (!tp->klass()->is_loaded()) {
3570       // Don't use intrinsic when class is not loaded.
3571       return false;
3572     } else {
3573       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3574       if (static_res == Compile::SSC_always_true) {
3575         // isInstance() is true - fold the code.
3576         set_result(obj);
3577         return true;
3578       } else if (static_res == Compile::SSC_always_false) {
3579         // Don't use intrinsic, have to throw ClassCastException.
3580         // If the reference is null, the non-intrinsic bytecode will
3581         // be optimized appropriately.
3582         return false;
3583       }
3584     }
3585   }
3586 
3587   // Bailout intrinsic and do normal inlining if exception path is frequent.
3588   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3589     return false;
3590   }
3591 
3592   // Generate dynamic checks.
3593   // Class.cast() is java implementation of _checkcast bytecode.
3594   // Do checkcast (Parse::do_checkcast()) optimizations here.
3595 
3596   mirror = null_check(mirror);
3597   // If mirror is dead, only null-path is taken.
3598   if (stopped()) {
3599     return true;
3600   }
3601 
3602   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3603   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3604   RegionNode* region = new RegionNode(PATH_LIMIT);
3605   record_for_igvn(region);
3606 
3607   // Now load the mirror's klass metaobject, and null-check it.
3608   // If kls is null, we have a primitive mirror and
3609   // nothing is an instance of a primitive type.
3610   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3611 
3612   Node* res = top();
3613   if (!stopped()) {
3614     Node* bad_type_ctrl = top();
3615     // Do checkcast optimizations.
3616     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3617     region->init_req(_bad_type_path, bad_type_ctrl);
3618   }
3619   if (region->in(_prim_path) != top() ||
3620       region->in(_bad_type_path) != top()) {
3621     // Let Interpreter throw ClassCastException.
3622     PreserveJVMState pjvms(this);
3623     set_control(_gvn.transform(region));
3624     uncommon_trap(Deoptimization::Reason_intrinsic,
3625                   Deoptimization::Action_maybe_recompile);
3626   }
3627   if (!stopped()) {
3628     set_result(res);
3629   }
3630   return true;
3631 }
3632 
3633 
3634 //--------------------------inline_native_subtype_check------------------------
3635 // This intrinsic takes the JNI calls out of the heart of
3636 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3637 bool LibraryCallKit::inline_native_subtype_check() {
3638   // Pull both arguments off the stack.
3639   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3640   args[0] = argument(0);
3641   args[1] = argument(1);
3642   Node* klasses[2];             // corresponding Klasses: superk, subk
3643   klasses[0] = klasses[1] = top();
3644 
3645   enum {
3646     // A full decision tree on {superc is prim, subc is prim}:
3647     _prim_0_path = 1,           // {P,N} => false
3648                                 // {P,P} & superc!=subc => false
3649     _prim_same_path,            // {P,P} & superc==subc => true
3650     _prim_1_path,               // {N,P} => false
3651     _ref_subtype_path,          // {N,N} & subtype check wins => true
3652     _both_ref_path,             // {N,N} & subtype check loses => false
3653     PATH_LIMIT
3654   };
3655 
3656   RegionNode* region = new RegionNode(PATH_LIMIT);
3657   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3658   record_for_igvn(region);
3659 
3660   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3661   const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3662   int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3663 
3664   // First null-check both mirrors and load each mirror's klass metaobject.
3665   int which_arg;
3666   for (which_arg = 0; which_arg <= 1; which_arg++) {
3667     Node* arg = args[which_arg];
3668     arg = null_check(arg);
3669     if (stopped())  break;
3670     args[which_arg] = arg;
3671 
3672     Node* p = basic_plus_adr(arg, class_klass_offset);
3673     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3674     klasses[which_arg] = _gvn.transform(kls);
3675   }
3676 
3677   // Having loaded both klasses, test each for null.
3678   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3679   for (which_arg = 0; which_arg <= 1; which_arg++) {
3680     Node* kls = klasses[which_arg];
3681     Node* null_ctl = top();
3682     kls = null_check_oop(kls, &null_ctl, never_see_null);
3683     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3684     region->init_req(prim_path, null_ctl);
3685     if (stopped())  break;
3686     klasses[which_arg] = kls;
3687   }
3688 
3689   if (!stopped()) {
3690     // now we have two reference types, in klasses[0..1]
3691     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3692     Node* superk = klasses[0];  // the receiver
3693     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3694     // now we have a successful reference subtype check
3695     region->set_req(_ref_subtype_path, control());
3696   }
3697 
3698   // If both operands are primitive (both klasses null), then
3699   // we must return true when they are identical primitives.
3700   // It is convenient to test this after the first null klass check.
3701   set_control(region->in(_prim_0_path)); // go back to first null check
3702   if (!stopped()) {
3703     // Since superc is primitive, make a guard for the superc==subc case.
3704     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3705     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3706     generate_guard(bol_eq, region, PROB_FAIR);
3707     if (region->req() == PATH_LIMIT+1) {
3708       // A guard was added.  If the added guard is taken, superc==subc.
3709       region->swap_edges(PATH_LIMIT, _prim_same_path);
3710       region->del_req(PATH_LIMIT);
3711     }
3712     region->set_req(_prim_0_path, control()); // Not equal after all.
3713   }
3714 
3715   // these are the only paths that produce 'true':
3716   phi->set_req(_prim_same_path,   intcon(1));
3717   phi->set_req(_ref_subtype_path, intcon(1));
3718 
3719   // pull together the cases:
3720   assert(region->req() == PATH_LIMIT, "sane region");
3721   for (uint i = 1; i < region->req(); i++) {
3722     Node* ctl = region->in(i);
3723     if (ctl == NULL || ctl == top()) {
3724       region->set_req(i, top());
3725       phi   ->set_req(i, top());
3726     } else if (phi->in(i) == NULL) {
3727       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3728     }
3729   }
3730 
3731   set_control(_gvn.transform(region));
3732   set_result(_gvn.transform(phi));
3733   return true;
3734 }
3735 
3736 //---------------------generate_array_guard_common------------------------
3737 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3738                                                   bool obj_array, bool not_array) {
3739 
3740   if (stopped()) {
3741     return NULL;
3742   }
3743 
3744   // If obj_array/non_array==false/false:
3745   // Branch around if the given klass is in fact an array (either obj or prim).
3746   // If obj_array/non_array==false/true:
3747   // Branch around if the given klass is not an array klass of any kind.
3748   // If obj_array/non_array==true/true:
3749   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3750   // If obj_array/non_array==true/false:
3751   // Branch around if the kls is an oop array (Object[] or subtype)
3752   //
3753   // Like generate_guard, adds a new path onto the region.
3754   jint  layout_con = 0;
3755   Node* layout_val = get_layout_helper(kls, layout_con);
3756   if (layout_val == NULL) {
3757     bool query = (obj_array
3758                   ? Klass::layout_helper_is_objArray(layout_con)
3759                   : Klass::layout_helper_is_array(layout_con));
3760     if (query == not_array) {
3761       return NULL;                       // never a branch
3762     } else {                             // always a branch
3763       Node* always_branch = control();
3764       if (region != NULL)
3765         region->add_req(always_branch);
3766       set_control(top());
3767       return always_branch;
3768     }
3769   }
3770   // Now test the correct condition.
3771   jint  nval = (obj_array
3772                 ? ((jint)Klass::_lh_array_tag_type_value
3773                    <<    Klass::_lh_array_tag_shift)
3774                 : Klass::_lh_neutral_value);
3775   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3776   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3777   // invert the test if we are looking for a non-array
3778   if (not_array)  btest = BoolTest(btest).negate();
3779   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3780   return generate_fair_guard(bol, region);
3781 }
3782 
3783 
3784 //-----------------------inline_native_newArray--------------------------
3785 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3786 // private        native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3787 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3788   Node* mirror;
3789   Node* count_val;
3790   if (uninitialized) {
3791     mirror    = argument(1);
3792     count_val = argument(2);
3793   } else {
3794     mirror    = argument(0);
3795     count_val = argument(1);
3796   }
3797 
3798   mirror = null_check(mirror);
3799   // If mirror or obj is dead, only null-path is taken.
3800   if (stopped())  return true;
3801 
3802   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3803   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3804   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3805   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3806   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3807 
3808   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3809   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3810                                                   result_reg, _slow_path);
3811   Node* normal_ctl   = control();
3812   Node* no_array_ctl = result_reg->in(_slow_path);
3813 
3814   // Generate code for the slow case.  We make a call to newArray().
3815   set_control(no_array_ctl);
3816   if (!stopped()) {
3817     // Either the input type is void.class, or else the
3818     // array klass has not yet been cached.  Either the
3819     // ensuing call will throw an exception, or else it
3820     // will cache the array klass for next time.
3821     PreserveJVMState pjvms(this);
3822     CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3823     Node* slow_result = set_results_for_java_call(slow_call);
3824     // this->control() comes from set_results_for_java_call
3825     result_reg->set_req(_slow_path, control());
3826     result_val->set_req(_slow_path, slow_result);
3827     result_io ->set_req(_slow_path, i_o());
3828     result_mem->set_req(_slow_path, reset_memory());
3829   }
3830 
3831   set_control(normal_ctl);
3832   if (!stopped()) {
3833     // Normal case:  The array type has been cached in the java.lang.Class.
3834     // The following call works fine even if the array type is polymorphic.
3835     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3836     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3837     result_reg->init_req(_normal_path, control());
3838     result_val->init_req(_normal_path, obj);
3839     result_io ->init_req(_normal_path, i_o());
3840     result_mem->init_req(_normal_path, reset_memory());
3841 
3842     if (uninitialized) {
3843       // Mark the allocation so that zeroing is skipped
3844       AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3845       alloc->maybe_set_complete(&_gvn);
3846     }
3847   }
3848 
3849   // Return the combined state.
3850   set_i_o(        _gvn.transform(result_io)  );
3851   set_all_memory( _gvn.transform(result_mem));
3852 
3853   C->set_has_split_ifs(true); // Has chance for split-if optimization
3854   set_result(result_reg, result_val);
3855   return true;
3856 }
3857 
3858 //----------------------inline_native_getLength--------------------------
3859 // public static native int java.lang.reflect.Array.getLength(Object array);
3860 bool LibraryCallKit::inline_native_getLength() {
3861   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3862 
3863   Node* array = null_check(argument(0));
3864   // If array is dead, only null-path is taken.
3865   if (stopped())  return true;
3866 
3867   // Deoptimize if it is a non-array.
3868   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3869 
3870   if (non_array != NULL) {
3871     PreserveJVMState pjvms(this);
3872     set_control(non_array);
3873     uncommon_trap(Deoptimization::Reason_intrinsic,
3874                   Deoptimization::Action_maybe_recompile);
3875   }
3876 
3877   // If control is dead, only non-array-path is taken.
3878   if (stopped())  return true;
3879 
3880   // The works fine even if the array type is polymorphic.
3881   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3882   Node* result = load_array_length(array);
3883 
3884   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3885   set_result(result);
3886   return true;
3887 }
3888 
3889 //------------------------inline_array_copyOf----------------------------
3890 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3891 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3892 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3893   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3894 
3895   // Get the arguments.
3896   Node* original          = argument(0);
3897   Node* start             = is_copyOfRange? argument(1): intcon(0);
3898   Node* end               = is_copyOfRange? argument(2): argument(1);
3899   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3900 
3901   Node* newcopy = NULL;
3902 
3903   // Set the original stack and the reexecute bit for the interpreter to reexecute
3904   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3905   { PreserveReexecuteState preexecs(this);
3906     jvms()->set_should_reexecute(true);
3907 
3908     array_type_mirror = null_check(array_type_mirror);
3909     original          = null_check(original);
3910 
3911     // Check if a null path was taken unconditionally.
3912     if (stopped())  return true;
3913 
3914     Node* orig_length = load_array_length(original);
3915 
3916     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3917     klass_node = null_check(klass_node);
3918 
3919     RegionNode* bailout = new RegionNode(1);
3920     record_for_igvn(bailout);
3921 
3922     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3923     // Bail out if that is so.
3924     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3925     if (not_objArray != NULL) {
3926       // Improve the klass node's type from the new optimistic assumption:
3927       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3928       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3929       Node* cast = new CastPPNode(klass_node, akls);
3930       cast->init_req(0, control());
3931       klass_node = _gvn.transform(cast);
3932     }
3933 
3934     // Bail out if either start or end is negative.
3935     generate_negative_guard(start, bailout, &start);
3936     generate_negative_guard(end,   bailout, &end);
3937 
3938     Node* length = end;
3939     if (_gvn.type(start) != TypeInt::ZERO) {
3940       length = _gvn.transform(new SubINode(end, start));
3941     }
3942 
3943     // Bail out if length is negative.
3944     // Without this the new_array would throw
3945     // NegativeArraySizeException but IllegalArgumentException is what
3946     // should be thrown
3947     generate_negative_guard(length, bailout, &length);
3948 
3949     if (bailout->req() > 1) {
3950       PreserveJVMState pjvms(this);
3951       set_control(_gvn.transform(bailout));
3952       uncommon_trap(Deoptimization::Reason_intrinsic,
3953                     Deoptimization::Action_maybe_recompile);
3954     }
3955 
3956     if (!stopped()) {
3957       // How many elements will we copy from the original?
3958       // The answer is MinI(orig_length - start, length).
3959       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3960       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3961 
3962       // Generate a direct call to the right arraycopy function(s).
3963       // We know the copy is disjoint but we might not know if the
3964       // oop stores need checking.
3965       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3966       // This will fail a store-check if x contains any non-nulls.
3967 
3968       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3969       // loads/stores but it is legal only if we're sure the
3970       // Arrays.copyOf would succeed. So we need all input arguments
3971       // to the copyOf to be validated, including that the copy to the
3972       // new array won't trigger an ArrayStoreException. That subtype
3973       // check can be optimized if we know something on the type of
3974       // the input array from type speculation.
3975       if (_gvn.type(klass_node)->singleton()) {
3976         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3977         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3978 
3979         int test = C->static_subtype_check(superk, subk);
3980         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3981           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3982           if (t_original->speculative_type() != NULL) {
3983             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3984           }
3985         }
3986       }
3987 
3988       bool validated = false;
3989       // Reason_class_check rather than Reason_intrinsic because we
3990       // want to intrinsify even if this traps.
3991       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3992         Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3993                                                    klass_node);
3994 
3995         if (not_subtype_ctrl != top()) {
3996           PreserveJVMState pjvms(this);
3997           set_control(not_subtype_ctrl);
3998           uncommon_trap(Deoptimization::Reason_class_check,
3999                         Deoptimization::Action_make_not_entrant);
4000           assert(stopped(), "Should be stopped");
4001         }
4002         validated = true;
4003       }
4004 
4005       if (!stopped()) {
4006         newcopy = new_array(klass_node, length, 0);  // no arguments to push
4007 
4008         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
4009                                                 load_object_klass(original), klass_node);
4010         if (!is_copyOfRange) {
4011           ac->set_copyof(validated);
4012         } else {
4013           ac->set_copyofrange(validated);
4014         }
4015         Node* n = _gvn.transform(ac);
4016         if (n == ac) {
4017           ac->connect_outputs(this);
4018         } else {
4019           assert(validated, "shouldn't transform if all arguments not validated");
4020           set_all_memory(n);
4021         }
4022       }
4023     }
4024   } // original reexecute is set back here
4025 
4026   C->set_has_split_ifs(true); // Has chance for split-if optimization
4027   if (!stopped()) {
4028     set_result(newcopy);
4029   }
4030   return true;
4031 }
4032 
4033 
4034 //----------------------generate_virtual_guard---------------------------
4035 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
4036 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4037                                              RegionNode* slow_region) {
4038   ciMethod* method = callee();
4039   int vtable_index = method->vtable_index();
4040   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4041          "bad index %d", vtable_index);
4042   // Get the Method* out of the appropriate vtable entry.
4043   int entry_offset  = in_bytes(Klass::vtable_start_offset()) +
4044                      vtable_index*vtableEntry::size_in_bytes() +
4045                      vtableEntry::method_offset_in_bytes();
4046   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
4047   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4048 
4049   // Compare the target method with the expected method (e.g., Object.hashCode).
4050   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4051 
4052   Node* native_call = makecon(native_call_addr);
4053   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
4054   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4055 
4056   return generate_slow_guard(test_native, slow_region);
4057 }
4058 
4059 //-----------------------generate_method_call----------------------------
4060 // Use generate_method_call to make a slow-call to the real
4061 // method if the fast path fails.  An alternative would be to
4062 // use a stub like OptoRuntime::slow_arraycopy_Java.
4063 // This only works for expanding the current library call,
4064 // not another intrinsic.  (E.g., don't use this for making an
4065 // arraycopy call inside of the copyOf intrinsic.)
4066 CallJavaNode*
4067 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4068   // When compiling the intrinsic method itself, do not use this technique.
4069   guarantee(callee() != C->method(), "cannot make slow-call to self");
4070 
4071   ciMethod* method = callee();
4072   // ensure the JVMS we have will be correct for this call
4073   guarantee(method_id == method->intrinsic_id(), "must match");
4074 
4075   const TypeFunc* tf = TypeFunc::make(method);
4076   CallJavaNode* slow_call;
4077   if (is_static) {
4078     assert(!is_virtual, "");
4079     slow_call = new CallStaticJavaNode(C, tf,
4080                            SharedRuntime::get_resolve_static_call_stub(),
4081                            method, bci());
4082   } else if (is_virtual) {
4083     null_check_receiver();
4084     int vtable_index = Method::invalid_vtable_index;
4085     if (UseInlineCaches) {
4086       // Suppress the vtable call
4087     } else {
4088       // hashCode and clone are not a miranda methods,
4089       // so the vtable index is fixed.
4090       // No need to use the linkResolver to get it.
4091        vtable_index = method->vtable_index();
4092        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4093               "bad index %d", vtable_index);
4094     }
4095     slow_call = new CallDynamicJavaNode(tf,
4096                           SharedRuntime::get_resolve_virtual_call_stub(),
4097                           method, vtable_index, bci());
4098   } else {  // neither virtual nor static:  opt_virtual
4099     null_check_receiver();
4100     slow_call = new CallStaticJavaNode(C, tf,
4101                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4102                                 method, bci());
4103     slow_call->set_optimized_virtual(true);
4104   }
4105   set_arguments_for_java_call(slow_call);
4106   set_edges_for_java_call(slow_call);
4107   return slow_call;
4108 }
4109 
4110 
4111 /**
4112  * Build special case code for calls to hashCode on an object. This call may
4113  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4114  * slightly different code.
4115  */
4116 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4117   assert(is_static == callee()->is_static(), "correct intrinsic selection");
4118   assert(!(is_virtual && is_static), "either virtual, special, or static");
4119 
4120   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4121 
4122   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4123   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
4124   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
4125   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4126   Node* obj = NULL;
4127   if (!is_static) {
4128     // Check for hashing null object
4129     obj = null_check_receiver();
4130     if (stopped())  return true;        // unconditionally null
4131     result_reg->init_req(_null_path, top());
4132     result_val->init_req(_null_path, top());
4133   } else {
4134     // Do a null check, and return zero if null.
4135     // System.identityHashCode(null) == 0
4136     obj = argument(0);
4137     Node* null_ctl = top();
4138     obj = null_check_oop(obj, &null_ctl);
4139     result_reg->init_req(_null_path, null_ctl);
4140     result_val->init_req(_null_path, _gvn.intcon(0));
4141   }
4142 
4143   // Unconditionally null?  Then return right away.
4144   if (stopped()) {
4145     set_control( result_reg->in(_null_path));
4146     if (!stopped())
4147       set_result(result_val->in(_null_path));
4148     return true;
4149   }
4150 
4151   // We only go to the fast case code if we pass a number of guards.  The
4152   // paths which do not pass are accumulated in the slow_region.
4153   RegionNode* slow_region = new RegionNode(1);
4154   record_for_igvn(slow_region);
4155 
4156   // If this is a virtual call, we generate a funny guard.  We pull out
4157   // the vtable entry corresponding to hashCode() from the target object.
4158   // If the target method which we are calling happens to be the native
4159   // Object hashCode() method, we pass the guard.  We do not need this
4160   // guard for non-virtual calls -- the caller is known to be the native
4161   // Object hashCode().
4162   if (is_virtual) {
4163     // After null check, get the object's klass.
4164     Node* obj_klass = load_object_klass(obj);
4165     generate_virtual_guard(obj_klass, slow_region);
4166   }
4167 
4168   // Get the header out of the object, use LoadMarkNode when available
4169   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4170   // The control of the load must be NULL. Otherwise, the load can move before
4171   // the null check after castPP removal.
4172   Node* no_ctrl = NULL;
4173   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4174 
4175   // Test the header to see if it is unlocked.
4176   Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4177   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4178   Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
4179   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4180   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4181 
4182   generate_slow_guard(test_unlocked, slow_region);
4183 
4184   // Get the hash value and check to see that it has been properly assigned.
4185   // We depend on hash_mask being at most 32 bits and avoid the use of
4186   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4187   // vm: see markOop.hpp.
4188   Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
4189   Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
4190   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4191   // This hack lets the hash bits live anywhere in the mark object now, as long
4192   // as the shift drops the relevant bits into the low 32 bits.  Note that
4193   // Java spec says that HashCode is an int so there's no point in capturing
4194   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4195   hshifted_header      = ConvX2I(hshifted_header);
4196   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4197 
4198   Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
4199   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4200   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4201 
4202   generate_slow_guard(test_assigned, slow_region);
4203 
4204   Node* init_mem = reset_memory();
4205   // fill in the rest of the null path:
4206   result_io ->init_req(_null_path, i_o());
4207   result_mem->init_req(_null_path, init_mem);
4208 
4209   result_val->init_req(_fast_path, hash_val);
4210   result_reg->init_req(_fast_path, control());
4211   result_io ->init_req(_fast_path, i_o());
4212   result_mem->init_req(_fast_path, init_mem);
4213 
4214   // Generate code for the slow case.  We make a call to hashCode().
4215   set_control(_gvn.transform(slow_region));
4216   if (!stopped()) {
4217     // No need for PreserveJVMState, because we're using up the present state.
4218     set_all_memory(init_mem);
4219     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4220     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4221     Node* slow_result = set_results_for_java_call(slow_call);
4222     // this->control() comes from set_results_for_java_call
4223     result_reg->init_req(_slow_path, control());
4224     result_val->init_req(_slow_path, slow_result);
4225     result_io  ->set_req(_slow_path, i_o());
4226     result_mem ->set_req(_slow_path, reset_memory());
4227   }
4228 
4229   // Return the combined state.
4230   set_i_o(        _gvn.transform(result_io)  );
4231   set_all_memory( _gvn.transform(result_mem));
4232 
4233   set_result(result_reg, result_val);
4234   return true;
4235 }
4236 
4237 //---------------------------inline_native_getClass----------------------------
4238 // public final native Class<?> java.lang.Object.getClass();
4239 //
4240 // Build special case code for calls to getClass on an object.
4241 bool LibraryCallKit::inline_native_getClass() {
4242   Node* obj = null_check_receiver();
4243   if (stopped())  return true;
4244   set_result(load_mirror_from_klass(load_object_klass(obj)));
4245   return true;
4246 }
4247 
4248 //-----------------inline_native_Reflection_getCallerClass---------------------
4249 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4250 //
4251 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4252 //
4253 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4254 // in that it must skip particular security frames and checks for
4255 // caller sensitive methods.
4256 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4257 #ifndef PRODUCT
4258   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4259     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4260   }
4261 #endif
4262 
4263   if (!jvms()->has_method()) {
4264 #ifndef PRODUCT
4265     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4266       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
4267     }
4268 #endif
4269     return false;
4270   }
4271 
4272   // Walk back up the JVM state to find the caller at the required
4273   // depth.
4274   JVMState* caller_jvms = jvms();
4275 
4276   // Cf. JVM_GetCallerClass
4277   // NOTE: Start the loop at depth 1 because the current JVM state does
4278   // not include the Reflection.getCallerClass() frame.
4279   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4280     ciMethod* m = caller_jvms->method();
4281     switch (n) {
4282     case 0:
4283       fatal("current JVM state does not include the Reflection.getCallerClass frame");
4284       break;
4285     case 1:
4286       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4287       if (!m->caller_sensitive()) {
4288 #ifndef PRODUCT
4289         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4290           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
4291         }
4292 #endif
4293         return false;  // bail-out; let JVM_GetCallerClass do the work
4294       }
4295       break;
4296     default:
4297       if (!m->is_ignored_by_security_stack_walk()) {
4298         // We have reached the desired frame; return the holder class.
4299         // Acquire method holder as java.lang.Class and push as constant.
4300         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4301         ciInstance* caller_mirror = caller_klass->java_mirror();
4302         set_result(makecon(TypeInstPtr::make(caller_mirror)));
4303 
4304 #ifndef PRODUCT
4305         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4306           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());
4307           tty->print_cr("  JVM state at this point:");
4308           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4309             ciMethod* m = jvms()->of_depth(i)->method();
4310             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4311           }
4312         }
4313 #endif
4314         return true;
4315       }
4316       break;
4317     }
4318   }
4319 
4320 #ifndef PRODUCT
4321   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4322     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4323     tty->print_cr("  JVM state at this point:");
4324     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4325       ciMethod* m = jvms()->of_depth(i)->method();
4326       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4327     }
4328   }
4329 #endif
4330 
4331   return false;  // bail-out; let JVM_GetCallerClass do the work
4332 }
4333 
4334 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4335   Node* arg = argument(0);
4336   Node* result = NULL;
4337 
4338   switch (id) {
4339   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4340   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4341   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4342   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4343 
4344   case vmIntrinsics::_doubleToLongBits: {
4345     // two paths (plus control) merge in a wood
4346     RegionNode *r = new RegionNode(3);
4347     Node *phi = new PhiNode(r, TypeLong::LONG);
4348 
4349     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4350     // Build the boolean node
4351     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4352 
4353     // Branch either way.
4354     // NaN case is less traveled, which makes all the difference.
4355     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4356     Node *opt_isnan = _gvn.transform(ifisnan);
4357     assert( opt_isnan->is_If(), "Expect an IfNode");
4358     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4359     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4360 
4361     set_control(iftrue);
4362 
4363     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4364     Node *slow_result = longcon(nan_bits); // return NaN
4365     phi->init_req(1, _gvn.transform( slow_result ));
4366     r->init_req(1, iftrue);
4367 
4368     // Else fall through
4369     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4370     set_control(iffalse);
4371 
4372     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4373     r->init_req(2, iffalse);
4374 
4375     // Post merge
4376     set_control(_gvn.transform(r));
4377     record_for_igvn(r);
4378 
4379     C->set_has_split_ifs(true); // Has chance for split-if optimization
4380     result = phi;
4381     assert(result->bottom_type()->isa_long(), "must be");
4382     break;
4383   }
4384 
4385   case vmIntrinsics::_floatToIntBits: {
4386     // two paths (plus control) merge in a wood
4387     RegionNode *r = new RegionNode(3);
4388     Node *phi = new PhiNode(r, TypeInt::INT);
4389 
4390     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4391     // Build the boolean node
4392     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4393 
4394     // Branch either way.
4395     // NaN case is less traveled, which makes all the difference.
4396     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4397     Node *opt_isnan = _gvn.transform(ifisnan);
4398     assert( opt_isnan->is_If(), "Expect an IfNode");
4399     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4400     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4401 
4402     set_control(iftrue);
4403 
4404     static const jint nan_bits = 0x7fc00000;
4405     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4406     phi->init_req(1, _gvn.transform( slow_result ));
4407     r->init_req(1, iftrue);
4408 
4409     // Else fall through
4410     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4411     set_control(iffalse);
4412 
4413     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4414     r->init_req(2, iffalse);
4415 
4416     // Post merge
4417     set_control(_gvn.transform(r));
4418     record_for_igvn(r);
4419 
4420     C->set_has_split_ifs(true); // Has chance for split-if optimization
4421     result = phi;
4422     assert(result->bottom_type()->isa_int(), "must be");
4423     break;
4424   }
4425 
4426   default:
4427     fatal_unexpected_iid(id);
4428     break;
4429   }
4430   set_result(_gvn.transform(result));
4431   return true;
4432 }
4433 
4434 //----------------------inline_unsafe_copyMemory-------------------------
4435 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4436 bool LibraryCallKit::inline_unsafe_copyMemory() {
4437   if (callee()->is_static())  return false;  // caller must have the capability!
4438   null_check_receiver();  // null-check receiver
4439   if (stopped())  return true;
4440 
4441   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4442 
4443   Node* src_ptr =         argument(1);   // type: oop
4444   Node* src_off = ConvL2X(argument(2));  // type: long
4445   Node* dst_ptr =         argument(4);   // type: oop
4446   Node* dst_off = ConvL2X(argument(5));  // type: long
4447   Node* size    = ConvL2X(argument(7));  // type: long
4448 
4449   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4450          "fieldOffset must be byte-scaled");
4451 
4452   Node* src = make_unsafe_address(src_ptr, src_off);
4453   Node* dst = make_unsafe_address(dst_ptr, dst_off);
4454 
4455   // Conservatively insert a memory barrier on all memory slices.
4456   // Do not let writes of the copy source or destination float below the copy.
4457   insert_mem_bar(Op_MemBarCPUOrder);
4458 
4459   // Call it.  Note that the length argument is not scaled.
4460   make_runtime_call(RC_LEAF|RC_NO_FP,
4461                     OptoRuntime::fast_arraycopy_Type(),
4462                     StubRoutines::unsafe_arraycopy(),
4463                     "unsafe_arraycopy",
4464                     TypeRawPtr::BOTTOM,
4465                     src, dst, size XTOP);
4466 
4467   // Do not let reads of the copy destination float above the copy.
4468   insert_mem_bar(Op_MemBarCPUOrder);
4469 
4470   return true;
4471 }
4472 
4473 //------------------------clone_coping-----------------------------------
4474 // Helper function for inline_native_clone.
4475 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4476   assert(obj_size != NULL, "");
4477   Node* raw_obj = alloc_obj->in(1);
4478   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4479 
4480   AllocateNode* alloc = NULL;
4481   if (ReduceBulkZeroing) {
4482     // We will be completely responsible for initializing this object -
4483     // mark Initialize node as complete.
4484     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4485     // The object was just allocated - there should be no any stores!
4486     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4487     // Mark as complete_with_arraycopy so that on AllocateNode
4488     // expansion, we know this AllocateNode is initialized by an array
4489     // copy and a StoreStore barrier exists after the array copy.
4490     alloc->initialization()->set_complete_with_arraycopy();
4491   }
4492 
4493   // Copy the fastest available way.
4494   // TODO: generate fields copies for small objects instead.
4495   Node* src  = obj;
4496   Node* dest = alloc_obj;
4497   Node* size = _gvn.transform(obj_size);
4498 
4499   // Exclude the header but include array length to copy by 8 bytes words.
4500   // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4501   int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4502                             instanceOopDesc::base_offset_in_bytes();
4503   // base_off:
4504   // 8  - 32-bit VM
4505   // 12 - 64-bit VM, compressed klass
4506   // 16 - 64-bit VM, normal klass
4507   if (base_off % BytesPerLong != 0) {
4508     assert(UseCompressedClassPointers, "");
4509     if (is_array) {
4510       // Exclude length to copy by 8 bytes words.
4511       base_off += sizeof(int);
4512     } else {
4513       // Include klass to copy by 8 bytes words.
4514       base_off = instanceOopDesc::klass_offset_in_bytes();
4515     }
4516     assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4517   }
4518   src  = basic_plus_adr(src,  base_off);
4519   dest = basic_plus_adr(dest, base_off);
4520 
4521   // Compute the length also, if needed:
4522   Node* countx = size;
4523   countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4524   countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4525 
4526   const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4527 
4528   ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4529   ac->set_clonebasic();
4530   Node* n = _gvn.transform(ac);
4531   if (n == ac) {
4532     set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4533   } else {
4534     set_all_memory(n);
4535   }
4536 
4537   // If necessary, emit some card marks afterwards.  (Non-arrays only.)
4538   if (card_mark) {
4539     assert(!is_array, "");
4540     // Put in store barrier for any and all oops we are sticking
4541     // into this object.  (We could avoid this if we could prove
4542     // that the object type contains no oop fields at all.)
4543     Node* no_particular_value = NULL;
4544     Node* no_particular_field = NULL;
4545     int raw_adr_idx = Compile::AliasIdxRaw;
4546     post_barrier(control(),
4547                  memory(raw_adr_type),
4548                  alloc_obj,
4549                  no_particular_field,
4550                  raw_adr_idx,
4551                  no_particular_value,
4552                  T_OBJECT,
4553                  false);
4554   }
4555 
4556   // Do not let reads from the cloned object float above the arraycopy.
4557   if (alloc != NULL) {
4558     // Do not let stores that initialize this object be reordered with
4559     // a subsequent store that would make this object accessible by
4560     // other threads.
4561     // Record what AllocateNode this StoreStore protects so that
4562     // escape analysis can go from the MemBarStoreStoreNode to the
4563     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4564     // based on the escape status of the AllocateNode.
4565     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4566   } else {
4567     insert_mem_bar(Op_MemBarCPUOrder);
4568   }
4569 }
4570 
4571 //------------------------inline_native_clone----------------------------
4572 // protected native Object java.lang.Object.clone();
4573 //
4574 // Here are the simple edge cases:
4575 //  null receiver => normal trap
4576 //  virtual and clone was overridden => slow path to out-of-line clone
4577 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4578 //
4579 // The general case has two steps, allocation and copying.
4580 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4581 //
4582 // Copying also has two cases, oop arrays and everything else.
4583 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4584 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4585 //
4586 // These steps fold up nicely if and when the cloned object's klass
4587 // can be sharply typed as an object array, a type array, or an instance.
4588 //
4589 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4590   PhiNode* result_val;
4591 
4592   // Set the reexecute bit for the interpreter to reexecute
4593   // the bytecode that invokes Object.clone if deoptimization happens.
4594   { PreserveReexecuteState preexecs(this);
4595     jvms()->set_should_reexecute(true);
4596 
4597     Node* obj = null_check_receiver();
4598     if (stopped())  return true;
4599 
4600     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4601 
4602     // If we are going to clone an instance, we need its exact type to
4603     // know the number and types of fields to convert the clone to
4604     // loads/stores. Maybe a speculative type can help us.
4605     if (!obj_type->klass_is_exact() &&
4606         obj_type->speculative_type() != NULL &&
4607         obj_type->speculative_type()->is_instance_klass()) {
4608       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4609       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4610           !spec_ik->has_injected_fields()) {
4611         ciKlass* k = obj_type->klass();
4612         if (!k->is_instance_klass() ||
4613             k->as_instance_klass()->is_interface() ||
4614             k->as_instance_klass()->has_subklass()) {
4615           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4616         }
4617       }
4618     }
4619 
4620     Node* obj_klass = load_object_klass(obj);
4621     const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4622     const TypeOopPtr*   toop   = ((tklass != NULL)
4623                                 ? tklass->as_instance_type()
4624                                 : TypeInstPtr::NOTNULL);
4625 
4626     // Conservatively insert a memory barrier on all memory slices.
4627     // Do not let writes into the original float below the clone.
4628     insert_mem_bar(Op_MemBarCPUOrder);
4629 
4630     // paths into result_reg:
4631     enum {
4632       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4633       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4634       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4635       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4636       PATH_LIMIT
4637     };
4638     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4639     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4640     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4641     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4642     record_for_igvn(result_reg);
4643 
4644     const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4645     int raw_adr_idx = Compile::AliasIdxRaw;
4646 
4647     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4648     if (array_ctl != NULL) {
4649       // It's an array.
4650       PreserveJVMState pjvms(this);
4651       set_control(array_ctl);
4652       Node* obj_length = load_array_length(obj);
4653       Node* obj_size  = NULL;
4654       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4655 
4656       if (!use_ReduceInitialCardMarks()) {
4657         // If it is an oop array, it requires very special treatment,
4658         // because card marking is required on each card of the array.
4659         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4660         if (is_obja != NULL) {
4661           PreserveJVMState pjvms2(this);
4662           set_control(is_obja);
4663           // Generate a direct call to the right arraycopy function(s).
4664           Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4665           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4666           ac->set_cloneoop();
4667           Node* n = _gvn.transform(ac);
4668           assert(n == ac, "cannot disappear");
4669           ac->connect_outputs(this);
4670 
4671           result_reg->init_req(_objArray_path, control());
4672           result_val->init_req(_objArray_path, alloc_obj);
4673           result_i_o ->set_req(_objArray_path, i_o());
4674           result_mem ->set_req(_objArray_path, reset_memory());
4675         }
4676       }
4677       // Otherwise, there are no card marks to worry about.
4678       // (We can dispense with card marks if we know the allocation
4679       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4680       //  causes the non-eden paths to take compensating steps to
4681       //  simulate a fresh allocation, so that no further
4682       //  card marks are required in compiled code to initialize
4683       //  the object.)
4684 
4685       if (!stopped()) {
4686         copy_to_clone(obj, alloc_obj, obj_size, true, false);
4687 
4688         // Present the results of the copy.
4689         result_reg->init_req(_array_path, control());
4690         result_val->init_req(_array_path, alloc_obj);
4691         result_i_o ->set_req(_array_path, i_o());
4692         result_mem ->set_req(_array_path, reset_memory());
4693       }
4694     }
4695 
4696     // We only go to the instance fast case code if we pass a number of guards.
4697     // The paths which do not pass are accumulated in the slow_region.
4698     RegionNode* slow_region = new RegionNode(1);
4699     record_for_igvn(slow_region);
4700     if (!stopped()) {
4701       // It's an instance (we did array above).  Make the slow-path tests.
4702       // If this is a virtual call, we generate a funny guard.  We grab
4703       // the vtable entry corresponding to clone() from the target object.
4704       // If the target method which we are calling happens to be the
4705       // Object clone() method, we pass the guard.  We do not need this
4706       // guard for non-virtual calls; the caller is known to be the native
4707       // Object clone().
4708       if (is_virtual) {
4709         generate_virtual_guard(obj_klass, slow_region);
4710       }
4711 
4712       // The object must be easily cloneable and must not have a finalizer.
4713       // Both of these conditions may be checked in a single test.
4714       // We could optimize the test further, but we don't care.
4715       generate_access_flags_guard(obj_klass,
4716                                   // Test both conditions:
4717                                   JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4718                                   // Must be cloneable but not finalizer:
4719                                   JVM_ACC_IS_CLONEABLE_FAST,
4720                                   slow_region);
4721     }
4722 
4723     if (!stopped()) {
4724       // It's an instance, and it passed the slow-path tests.
4725       PreserveJVMState pjvms(this);
4726       Node* obj_size  = NULL;
4727       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4728       // is reexecuted if deoptimization occurs and there could be problems when merging
4729       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4730       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4731 
4732       copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4733 
4734       // Present the results of the slow call.
4735       result_reg->init_req(_instance_path, control());
4736       result_val->init_req(_instance_path, alloc_obj);
4737       result_i_o ->set_req(_instance_path, i_o());
4738       result_mem ->set_req(_instance_path, reset_memory());
4739     }
4740 
4741     // Generate code for the slow case.  We make a call to clone().
4742     set_control(_gvn.transform(slow_region));
4743     if (!stopped()) {
4744       PreserveJVMState pjvms(this);
4745       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4746       Node* slow_result = set_results_for_java_call(slow_call);
4747       // this->control() comes from set_results_for_java_call
4748       result_reg->init_req(_slow_path, control());
4749       result_val->init_req(_slow_path, slow_result);
4750       result_i_o ->set_req(_slow_path, i_o());
4751       result_mem ->set_req(_slow_path, reset_memory());
4752     }
4753 
4754     // Return the combined state.
4755     set_control(    _gvn.transform(result_reg));
4756     set_i_o(        _gvn.transform(result_i_o));
4757     set_all_memory( _gvn.transform(result_mem));
4758   } // original reexecute is set back here
4759 
4760   set_result(_gvn.transform(result_val));
4761   return true;
4762 }
4763 
4764 // If we have a tighly coupled allocation, the arraycopy may take care
4765 // of the array initialization. If one of the guards we insert between
4766 // the allocation and the arraycopy causes a deoptimization, an
4767 // unitialized array will escape the compiled method. To prevent that
4768 // we set the JVM state for uncommon traps between the allocation and
4769 // the arraycopy to the state before the allocation so, in case of
4770 // deoptimization, we'll reexecute the allocation and the
4771 // initialization.
4772 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4773   if (alloc != NULL) {
4774     ciMethod* trap_method = alloc->jvms()->method();
4775     int trap_bci = alloc->jvms()->bci();
4776 
4777     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4778           !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4779       // Make sure there's no store between the allocation and the
4780       // arraycopy otherwise visible side effects could be rexecuted
4781       // in case of deoptimization and cause incorrect execution.
4782       bool no_interfering_store = true;
4783       Node* mem = alloc->in(TypeFunc::Memory);
4784       if (mem->is_MergeMem()) {
4785         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4786           Node* n = mms.memory();
4787           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4788             assert(n->is_Store(), "what else?");
4789             no_interfering_store = false;
4790             break;
4791           }
4792         }
4793       } else {
4794         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4795           Node* n = mms.memory();
4796           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4797             assert(n->is_Store(), "what else?");
4798             no_interfering_store = false;
4799             break;
4800           }
4801         }
4802       }
4803 
4804       if (no_interfering_store) {
4805         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4806         uint size = alloc->req();
4807         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4808         old_jvms->set_map(sfpt);
4809         for (uint i = 0; i < size; i++) {
4810           sfpt->init_req(i, alloc->in(i));
4811         }
4812         // re-push array length for deoptimization
4813         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4814         old_jvms->set_sp(old_jvms->sp()+1);
4815         old_jvms->set_monoff(old_jvms->monoff()+1);
4816         old_jvms->set_scloff(old_jvms->scloff()+1);
4817         old_jvms->set_endoff(old_jvms->endoff()+1);
4818         old_jvms->set_should_reexecute(true);
4819 
4820         sfpt->set_i_o(map()->i_o());
4821         sfpt->set_memory(map()->memory());
4822         sfpt->set_control(map()->control());
4823 
4824         JVMState* saved_jvms = jvms();
4825         saved_reexecute_sp = _reexecute_sp;
4826 
4827         set_jvms(sfpt->jvms());
4828         _reexecute_sp = jvms()->sp();
4829 
4830         return saved_jvms;
4831       }
4832     }
4833   }
4834   return NULL;
4835 }
4836 
4837 // In case of a deoptimization, we restart execution at the
4838 // allocation, allocating a new array. We would leave an uninitialized
4839 // array in the heap that GCs wouldn't expect. Move the allocation
4840 // after the traps so we don't allocate the array if we
4841 // deoptimize. This is possible because tightly_coupled_allocation()
4842 // guarantees there's no observer of the allocated array at this point
4843 // and the control flow is simple enough.
4844 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4845   if (saved_jvms != NULL && !stopped()) {
4846     assert(alloc != NULL, "only with a tightly coupled allocation");
4847     // restore JVM state to the state at the arraycopy
4848     saved_jvms->map()->set_control(map()->control());
4849     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4850     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4851     // If we've improved the types of some nodes (null check) while
4852     // emitting the guards, propagate them to the current state
4853     map()->replaced_nodes().apply(saved_jvms->map());
4854     set_jvms(saved_jvms);
4855     _reexecute_sp = saved_reexecute_sp;
4856 
4857     // Remove the allocation from above the guards
4858     CallProjections callprojs;
4859     alloc->extract_projections(&callprojs, true);
4860     InitializeNode* init = alloc->initialization();
4861     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4862     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4863     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4864     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4865 
4866     // move the allocation here (after the guards)
4867     _gvn.hash_delete(alloc);
4868     alloc->set_req(TypeFunc::Control, control());
4869     alloc->set_req(TypeFunc::I_O, i_o());
4870     Node *mem = reset_memory();
4871     set_all_memory(mem);
4872     alloc->set_req(TypeFunc::Memory, mem);
4873     set_control(init->proj_out(TypeFunc::Control));
4874     set_i_o(callprojs.fallthrough_ioproj);
4875 
4876     // Update memory as done in GraphKit::set_output_for_allocation()
4877     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4878     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4879     if (ary_type->isa_aryptr() && length_type != NULL) {
4880       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4881     }
4882     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4883     int            elemidx  = C->get_alias_index(telemref);
4884     set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4885     set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4886 
4887     Node* allocx = _gvn.transform(alloc);
4888     assert(allocx == alloc, "where has the allocation gone?");
4889     assert(dest->is_CheckCastPP(), "not an allocation result?");
4890 
4891     _gvn.hash_delete(dest);
4892     dest->set_req(0, control());
4893     Node* destx = _gvn.transform(dest);
4894     assert(destx == dest, "where has the allocation result gone?");
4895   }
4896 }
4897 
4898 
4899 //------------------------------inline_arraycopy-----------------------
4900 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4901 //                                                      Object dest, int destPos,
4902 //                                                      int length);
4903 bool LibraryCallKit::inline_arraycopy() {
4904   // Get the arguments.
4905   Node* src         = argument(0);  // type: oop
4906   Node* src_offset  = argument(1);  // type: int
4907   Node* dest        = argument(2);  // type: oop
4908   Node* dest_offset = argument(3);  // type: int
4909   Node* length      = argument(4);  // type: int
4910 
4911 
4912   // Check for allocation before we add nodes that would confuse
4913   // tightly_coupled_allocation()
4914   AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4915 
4916   int saved_reexecute_sp = -1;
4917   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4918   // See arraycopy_restore_alloc_state() comment
4919   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4920   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4921   // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4922   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4923 
4924   // The following tests must be performed
4925   // (1) src and dest are arrays.
4926   // (2) src and dest arrays must have elements of the same BasicType
4927   // (3) src and dest must not be null.
4928   // (4) src_offset must not be negative.
4929   // (5) dest_offset must not be negative.
4930   // (6) length must not be negative.
4931   // (7) src_offset + length must not exceed length of src.
4932   // (8) dest_offset + length must not exceed length of dest.
4933   // (9) each element of an oop array must be assignable
4934 
4935   // (3) src and dest must not be null.
4936   // always do this here because we need the JVM state for uncommon traps
4937   Node* null_ctl = top();
4938   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4939   assert(null_ctl->is_top(), "no null control here");
4940   dest = null_check(dest, T_ARRAY);
4941 
4942   if (!can_emit_guards) {
4943     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4944     // guards but the arraycopy node could still take advantage of a
4945     // tightly allocated allocation. tightly_coupled_allocation() is
4946     // called again to make sure it takes the null check above into
4947     // account: the null check is mandatory and if it caused an
4948     // uncommon trap to be emitted then the allocation can't be
4949     // considered tightly coupled in this context.
4950     alloc = tightly_coupled_allocation(dest, NULL);
4951   }
4952 
4953   bool validated = false;
4954 
4955   const Type* src_type  = _gvn.type(src);
4956   const Type* dest_type = _gvn.type(dest);
4957   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4958   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4959 
4960   // Do we have the type of src?
4961   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4962   // Do we have the type of dest?
4963   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4964   // Is the type for src from speculation?
4965   bool src_spec = false;
4966   // Is the type for dest from speculation?
4967   bool dest_spec = false;
4968 
4969   if ((!has_src || !has_dest) && can_emit_guards) {
4970     // We don't have sufficient type information, let's see if
4971     // speculative types can help. We need to have types for both src
4972     // and dest so that it pays off.
4973 
4974     // Do we already have or could we have type information for src
4975     bool could_have_src = has_src;
4976     // Do we already have or could we have type information for dest
4977     bool could_have_dest = has_dest;
4978 
4979     ciKlass* src_k = NULL;
4980     if (!has_src) {
4981       src_k = src_type->speculative_type_not_null();
4982       if (src_k != NULL && src_k->is_array_klass()) {
4983         could_have_src = true;
4984       }
4985     }
4986 
4987     ciKlass* dest_k = NULL;
4988     if (!has_dest) {
4989       dest_k = dest_type->speculative_type_not_null();
4990       if (dest_k != NULL && dest_k->is_array_klass()) {
4991         could_have_dest = true;
4992       }
4993     }
4994 
4995     if (could_have_src && could_have_dest) {
4996       // This is going to pay off so emit the required guards
4997       if (!has_src) {
4998         src = maybe_cast_profiled_obj(src, src_k, true);
4999         src_type  = _gvn.type(src);
5000         top_src  = src_type->isa_aryptr();
5001         has_src = (top_src != NULL && top_src->klass() != NULL);
5002         src_spec = true;
5003       }
5004       if (!has_dest) {
5005         dest = maybe_cast_profiled_obj(dest, dest_k, true);
5006         dest_type  = _gvn.type(dest);
5007         top_dest  = dest_type->isa_aryptr();
5008         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5009         dest_spec = true;
5010       }
5011     }
5012   }
5013 
5014   if (has_src && has_dest && can_emit_guards) {
5015     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
5016     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5017     if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
5018     if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;
5019 
5020     if (src_elem == dest_elem && src_elem == T_OBJECT) {
5021       // If both arrays are object arrays then having the exact types
5022       // for both will remove the need for a subtype check at runtime
5023       // before the call and may make it possible to pick a faster copy
5024       // routine (without a subtype check on every element)
5025       // Do we have the exact type of src?
5026       bool could_have_src = src_spec;
5027       // Do we have the exact type of dest?
5028       bool could_have_dest = dest_spec;
5029       ciKlass* src_k = top_src->klass();
5030       ciKlass* dest_k = top_dest->klass();
5031       if (!src_spec) {
5032         src_k = src_type->speculative_type_not_null();
5033         if (src_k != NULL && src_k->is_array_klass()) {
5034           could_have_src = true;
5035         }
5036       }
5037       if (!dest_spec) {
5038         dest_k = dest_type->speculative_type_not_null();
5039         if (dest_k != NULL && dest_k->is_array_klass()) {
5040           could_have_dest = true;
5041         }
5042       }
5043       if (could_have_src && could_have_dest) {
5044         // If we can have both exact types, emit the missing guards
5045         if (could_have_src && !src_spec) {
5046           src = maybe_cast_profiled_obj(src, src_k, true);
5047         }
5048         if (could_have_dest && !dest_spec) {
5049           dest = maybe_cast_profiled_obj(dest, dest_k, true);
5050         }
5051       }
5052     }
5053   }
5054 
5055   ciMethod* trap_method = method();
5056   int trap_bci = bci();
5057   if (saved_jvms != NULL) {
5058     trap_method = alloc->jvms()->method();
5059     trap_bci = alloc->jvms()->bci();
5060   }
5061 
5062   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5063       can_emit_guards &&
5064       !src->is_top() && !dest->is_top()) {
5065     // validate arguments: enables transformation the ArrayCopyNode
5066     validated = true;
5067 
5068     RegionNode* slow_region = new RegionNode(1);
5069     record_for_igvn(slow_region);
5070 
5071     // (1) src and dest are arrays.
5072     generate_non_array_guard(load_object_klass(src), slow_region);
5073     generate_non_array_guard(load_object_klass(dest), slow_region);
5074 
5075     // (2) src and dest arrays must have elements of the same BasicType
5076     // done at macro expansion or at Ideal transformation time
5077 
5078     // (4) src_offset must not be negative.
5079     generate_negative_guard(src_offset, slow_region);
5080 
5081     // (5) dest_offset must not be negative.
5082     generate_negative_guard(dest_offset, slow_region);
5083 
5084     // (7) src_offset + length must not exceed length of src.
5085     generate_limit_guard(src_offset, length,
5086                          load_array_length(src),
5087                          slow_region);
5088 
5089     // (8) dest_offset + length must not exceed length of dest.
5090     generate_limit_guard(dest_offset, length,
5091                          load_array_length(dest),
5092                          slow_region);
5093 
5094     // (9) each element of an oop array must be assignable
5095     Node* src_klass  = load_object_klass(src);
5096     Node* dest_klass = load_object_klass(dest);
5097     Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5098 
5099     if (not_subtype_ctrl != top()) {
5100       PreserveJVMState pjvms(this);
5101       set_control(not_subtype_ctrl);
5102       uncommon_trap(Deoptimization::Reason_intrinsic,
5103                     Deoptimization::Action_make_not_entrant);
5104       assert(stopped(), "Should be stopped");
5105     }
5106     {
5107       PreserveJVMState pjvms(this);
5108       set_control(_gvn.transform(slow_region));
5109       uncommon_trap(Deoptimization::Reason_intrinsic,
5110                     Deoptimization::Action_make_not_entrant);
5111       assert(stopped(), "Should be stopped");
5112     }
5113   }
5114 
5115   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5116 
5117   if (stopped()) {
5118     return true;
5119   }
5120 
5121   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5122                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
5123                                           // so the compiler has a chance to eliminate them: during macro expansion,
5124                                           // we have to set their control (CastPP nodes are eliminated).
5125                                           load_object_klass(src), load_object_klass(dest),
5126                                           load_array_length(src), load_array_length(dest));
5127 
5128   ac->set_arraycopy(validated);
5129 
5130   Node* n = _gvn.transform(ac);
5131   if (n == ac) {
5132     ac->connect_outputs(this);
5133   } else {
5134     assert(validated, "shouldn't transform if all arguments not validated");
5135     set_all_memory(n);
5136   }
5137 
5138   return true;
5139 }
5140 
5141 
5142 // Helper function which determines if an arraycopy immediately follows
5143 // an allocation, with no intervening tests or other escapes for the object.
5144 AllocateArrayNode*
5145 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5146                                            RegionNode* slow_region) {
5147   if (stopped())             return NULL;  // no fast path
5148   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
5149 
5150   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5151   if (alloc == NULL)  return NULL;
5152 
5153   Node* rawmem = memory(Compile::AliasIdxRaw);
5154   // Is the allocation's memory state untouched?
5155   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5156     // Bail out if there have been raw-memory effects since the allocation.
5157     // (Example:  There might have been a call or safepoint.)
5158     return NULL;
5159   }
5160   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5161   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5162     return NULL;
5163   }
5164 
5165   // There must be no unexpected observers of this allocation.
5166   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5167     Node* obs = ptr->fast_out(i);
5168     if (obs != this->map()) {
5169       return NULL;
5170     }
5171   }
5172 
5173   // This arraycopy must unconditionally follow the allocation of the ptr.
5174   Node* alloc_ctl = ptr->in(0);
5175   assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5176 
5177   Node* ctl = control();
5178   while (ctl != alloc_ctl) {
5179     // There may be guards which feed into the slow_region.
5180     // Any other control flow means that we might not get a chance
5181     // to finish initializing the allocated object.
5182     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5183       IfNode* iff = ctl->in(0)->as_If();
5184       Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5185       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5186       if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5187         ctl = iff->in(0);       // This test feeds the known slow_region.
5188         continue;
5189       }
5190       // One more try:  Various low-level checks bottom out in
5191       // uncommon traps.  If the debug-info of the trap omits
5192       // any reference to the allocation, as we've already
5193       // observed, then there can be no objection to the trap.
5194       bool found_trap = false;
5195       for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5196         Node* obs = not_ctl->fast_out(j);
5197         if (obs->in(0) == not_ctl && obs->is_Call() &&
5198             (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5199           found_trap = true; break;
5200         }
5201       }
5202       if (found_trap) {
5203         ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
5204         continue;
5205       }
5206     }
5207     return NULL;
5208   }
5209 
5210   // If we get this far, we have an allocation which immediately
5211   // precedes the arraycopy, and we can take over zeroing the new object.
5212   // The arraycopy will finish the initialization, and provide
5213   // a new control state to which we will anchor the destination pointer.
5214 
5215   return alloc;
5216 }
5217 
5218 //-------------inline_encodeISOArray-----------------------------------
5219 // encode char[] to byte[] in ISO_8859_1
5220 bool LibraryCallKit::inline_encodeISOArray() {
5221   assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5222   // no receiver since it is static method
5223   Node *src         = argument(0);
5224   Node *src_offset  = argument(1);
5225   Node *dst         = argument(2);
5226   Node *dst_offset  = argument(3);
5227   Node *length      = argument(4);
5228 
5229   const Type* src_type = src->Value(&_gvn);
5230   const Type* dst_type = dst->Value(&_gvn);
5231   const TypeAryPtr* top_src = src_type->isa_aryptr();
5232   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5233   if (top_src  == NULL || top_src->klass()  == NULL ||
5234       top_dest == NULL || top_dest->klass() == NULL) {
5235     // failed array check
5236     return false;
5237   }
5238 
5239   // Figure out the size and type of the elements we will be copying.
5240   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5241   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5242   if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5243     return false;
5244   }
5245 
5246   Node* src_start = array_element_address(src, src_offset, T_CHAR);
5247   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5248   // 'src_start' points to src array + scaled offset
5249   // 'dst_start' points to dst array + scaled offset
5250 
5251   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5252   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5253   enc = _gvn.transform(enc);
5254   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5255   set_memory(res_mem, mtype);
5256   set_result(enc);
5257   return true;
5258 }
5259 
5260 //-------------inline_multiplyToLen-----------------------------------
5261 bool LibraryCallKit::inline_multiplyToLen() {
5262   assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5263 
5264   address stubAddr = StubRoutines::multiplyToLen();
5265   if (stubAddr == NULL) {
5266     return false; // Intrinsic's stub is not implemented on this platform
5267   }
5268   const char* stubName = "multiplyToLen";
5269 
5270   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5271 
5272   // no receiver because it is a static method
5273   Node* x    = argument(0);
5274   Node* xlen = argument(1);
5275   Node* y    = argument(2);
5276   Node* ylen = argument(3);
5277   Node* z    = argument(4);
5278 
5279   const Type* x_type = x->Value(&_gvn);
5280   const Type* y_type = y->Value(&_gvn);
5281   const TypeAryPtr* top_x = x_type->isa_aryptr();
5282   const TypeAryPtr* top_y = y_type->isa_aryptr();
5283   if (top_x  == NULL || top_x->klass()  == NULL ||
5284       top_y == NULL || top_y->klass() == NULL) {
5285     // failed array check
5286     return false;
5287   }
5288 
5289   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5290   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5291   if (x_elem != T_INT || y_elem != T_INT) {
5292     return false;
5293   }
5294 
5295   // Set the original stack and the reexecute bit for the interpreter to reexecute
5296   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5297   // on the return from z array allocation in runtime.
5298   { PreserveReexecuteState preexecs(this);
5299     jvms()->set_should_reexecute(true);
5300 
5301     Node* x_start = array_element_address(x, intcon(0), x_elem);
5302     Node* y_start = array_element_address(y, intcon(0), y_elem);
5303     // 'x_start' points to x array + scaled xlen
5304     // 'y_start' points to y array + scaled ylen
5305 
5306     // Allocate the result array
5307     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5308     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5309     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5310 
5311     IdealKit ideal(this);
5312 
5313 #define __ ideal.
5314      Node* one = __ ConI(1);
5315      Node* zero = __ ConI(0);
5316      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5317      __ set(need_alloc, zero);
5318      __ set(z_alloc, z);
5319      __ if_then(z, BoolTest::eq, null()); {
5320        __ increment (need_alloc, one);
5321      } __ else_(); {
5322        // Update graphKit memory and control from IdealKit.
5323        sync_kit(ideal);
5324        Node* zlen_arg = load_array_length(z);
5325        // Update IdealKit memory and control from graphKit.
5326        __ sync_kit(this);
5327        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5328          __ increment (need_alloc, one);
5329        } __ end_if();
5330      } __ end_if();
5331 
5332      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5333        // Update graphKit memory and control from IdealKit.
5334        sync_kit(ideal);
5335        Node * narr = new_array(klass_node, zlen, 1);
5336        // Update IdealKit memory and control from graphKit.
5337        __ sync_kit(this);
5338        __ set(z_alloc, narr);
5339      } __ end_if();
5340 
5341      sync_kit(ideal);
5342      z = __ value(z_alloc);
5343      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5344      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5345      // Final sync IdealKit and GraphKit.
5346      final_sync(ideal);
5347 #undef __
5348 
5349     Node* z_start = array_element_address(z, intcon(0), T_INT);
5350 
5351     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5352                                    OptoRuntime::multiplyToLen_Type(),
5353                                    stubAddr, stubName, TypePtr::BOTTOM,
5354                                    x_start, xlen, y_start, ylen, z_start, zlen);
5355   } // original reexecute is set back here
5356 
5357   C->set_has_split_ifs(true); // Has chance for split-if optimization
5358   set_result(z);
5359   return true;
5360 }
5361 
5362 //-------------inline_squareToLen------------------------------------
5363 bool LibraryCallKit::inline_squareToLen() {
5364   assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5365 
5366   address stubAddr = StubRoutines::squareToLen();
5367   if (stubAddr == NULL) {
5368     return false; // Intrinsic's stub is not implemented on this platform
5369   }
5370   const char* stubName = "squareToLen";
5371 
5372   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5373 
5374   Node* x    = argument(0);
5375   Node* len  = argument(1);
5376   Node* z    = argument(2);
5377   Node* zlen = argument(3);
5378 
5379   const Type* x_type = x->Value(&_gvn);
5380   const Type* z_type = z->Value(&_gvn);
5381   const TypeAryPtr* top_x = x_type->isa_aryptr();
5382   const TypeAryPtr* top_z = z_type->isa_aryptr();
5383   if (top_x  == NULL || top_x->klass()  == NULL ||
5384       top_z  == NULL || top_z->klass()  == NULL) {
5385     // failed array check
5386     return false;
5387   }
5388 
5389   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5390   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5391   if (x_elem != T_INT || z_elem != T_INT) {
5392     return false;
5393   }
5394 
5395 
5396   Node* x_start = array_element_address(x, intcon(0), x_elem);
5397   Node* z_start = array_element_address(z, intcon(0), z_elem);
5398 
5399   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5400                                   OptoRuntime::squareToLen_Type(),
5401                                   stubAddr, stubName, TypePtr::BOTTOM,
5402                                   x_start, len, z_start, zlen);
5403 
5404   set_result(z);
5405   return true;
5406 }
5407 
5408 //-------------inline_mulAdd------------------------------------------
5409 bool LibraryCallKit::inline_mulAdd() {
5410   assert(UseMulAddIntrinsic, "not implemented on this platform");
5411 
5412   address stubAddr = StubRoutines::mulAdd();
5413   if (stubAddr == NULL) {
5414     return false; // Intrinsic's stub is not implemented on this platform
5415   }
5416   const char* stubName = "mulAdd";
5417 
5418   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5419 
5420   Node* out      = argument(0);
5421   Node* in       = argument(1);
5422   Node* offset   = argument(2);
5423   Node* len      = argument(3);
5424   Node* k        = argument(4);
5425 
5426   const Type* out_type = out->Value(&_gvn);
5427   const Type* in_type = in->Value(&_gvn);
5428   const TypeAryPtr* top_out = out_type->isa_aryptr();
5429   const TypeAryPtr* top_in = in_type->isa_aryptr();
5430   if (top_out  == NULL || top_out->klass()  == NULL ||
5431       top_in == NULL || top_in->klass() == NULL) {
5432     // failed array check
5433     return false;
5434   }
5435 
5436   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5437   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5438   if (out_elem != T_INT || in_elem != T_INT) {
5439     return false;
5440   }
5441 
5442   Node* outlen = load_array_length(out);
5443   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5444   Node* out_start = array_element_address(out, intcon(0), out_elem);
5445   Node* in_start = array_element_address(in, intcon(0), in_elem);
5446 
5447   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5448                                   OptoRuntime::mulAdd_Type(),
5449                                   stubAddr, stubName, TypePtr::BOTTOM,
5450                                   out_start,in_start, new_offset, len, k);
5451   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5452   set_result(result);
5453   return true;
5454 }
5455 
5456 //-------------inline_montgomeryMultiply-----------------------------------
5457 bool LibraryCallKit::inline_montgomeryMultiply() {
5458   address stubAddr = StubRoutines::montgomeryMultiply();
5459   if (stubAddr == NULL) {
5460     return false; // Intrinsic's stub is not implemented on this platform
5461   }
5462 
5463   assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5464   const char* stubName = "montgomery_square";
5465 
5466   assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5467 
5468   Node* a    = argument(0);
5469   Node* b    = argument(1);
5470   Node* n    = argument(2);
5471   Node* len  = argument(3);
5472   Node* inv  = argument(4);
5473   Node* m    = argument(6);
5474 
5475   const Type* a_type = a->Value(&_gvn);
5476   const TypeAryPtr* top_a = a_type->isa_aryptr();
5477   const Type* b_type = b->Value(&_gvn);
5478   const TypeAryPtr* top_b = b_type->isa_aryptr();
5479   const Type* n_type = a->Value(&_gvn);
5480   const TypeAryPtr* top_n = n_type->isa_aryptr();
5481   const Type* m_type = a->Value(&_gvn);
5482   const TypeAryPtr* top_m = m_type->isa_aryptr();
5483   if (top_a  == NULL || top_a->klass()  == NULL ||
5484       top_b == NULL || top_b->klass()  == NULL ||
5485       top_n == NULL || top_n->klass()  == NULL ||
5486       top_m == NULL || top_m->klass()  == NULL) {
5487     // failed array check
5488     return false;
5489   }
5490 
5491   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5492   BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5493   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5494   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5495   if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5496     return false;
5497   }
5498 
5499   // Make the call
5500   {
5501     Node* a_start = array_element_address(a, intcon(0), a_elem);
5502     Node* b_start = array_element_address(b, intcon(0), b_elem);
5503     Node* n_start = array_element_address(n, intcon(0), n_elem);
5504     Node* m_start = array_element_address(m, intcon(0), m_elem);
5505 
5506     Node* call = make_runtime_call(RC_LEAF,
5507                                    OptoRuntime::montgomeryMultiply_Type(),
5508                                    stubAddr, stubName, TypePtr::BOTTOM,
5509                                    a_start, b_start, n_start, len, inv, top(),
5510                                    m_start);
5511     set_result(m);
5512   }
5513 
5514   return true;
5515 }
5516 
5517 bool LibraryCallKit::inline_montgomerySquare() {
5518   address stubAddr = StubRoutines::montgomerySquare();
5519   if (stubAddr == NULL) {
5520     return false; // Intrinsic's stub is not implemented on this platform
5521   }
5522 
5523   assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5524   const char* stubName = "montgomery_square";
5525 
5526   assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5527 
5528   Node* a    = argument(0);
5529   Node* n    = argument(1);
5530   Node* len  = argument(2);
5531   Node* inv  = argument(3);
5532   Node* m    = argument(5);
5533 
5534   const Type* a_type = a->Value(&_gvn);
5535   const TypeAryPtr* top_a = a_type->isa_aryptr();
5536   const Type* n_type = a->Value(&_gvn);
5537   const TypeAryPtr* top_n = n_type->isa_aryptr();
5538   const Type* m_type = a->Value(&_gvn);
5539   const TypeAryPtr* top_m = m_type->isa_aryptr();
5540   if (top_a  == NULL || top_a->klass()  == NULL ||
5541       top_n == NULL || top_n->klass()  == NULL ||
5542       top_m == NULL || top_m->klass()  == NULL) {
5543     // failed array check
5544     return false;
5545   }
5546 
5547   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5548   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5549   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5550   if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5551     return false;
5552   }
5553 
5554   // Make the call
5555   {
5556     Node* a_start = array_element_address(a, intcon(0), a_elem);
5557     Node* n_start = array_element_address(n, intcon(0), n_elem);
5558     Node* m_start = array_element_address(m, intcon(0), m_elem);
5559 
5560     Node* call = make_runtime_call(RC_LEAF,
5561                                    OptoRuntime::montgomerySquare_Type(),
5562                                    stubAddr, stubName, TypePtr::BOTTOM,
5563                                    a_start, n_start, len, inv, top(),
5564                                    m_start);
5565     set_result(m);
5566   }
5567 
5568   return true;
5569 }
5570 
5571 //-------------inline_vectorizedMismatch------------------------------
5572 bool LibraryCallKit::inline_vectorizedMismatch() {
5573   assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5574 
5575   address stubAddr = StubRoutines::vectorizedMismatch();
5576   if (stubAddr == NULL) {
5577     return false; // Intrinsic's stub is not implemented on this platform
5578   }
5579   const char* stubName = "vectorizedMismatch";
5580   int size_l = callee()->signature()->size();
5581   assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5582 
5583   Node* obja = argument(0);
5584   Node* aoffset = argument(1);
5585   Node* objb = argument(3);
5586   Node* boffset = argument(4);
5587   Node* length = argument(6);
5588   Node* scale = argument(7);
5589 
5590   const Type* a_type = obja->Value(&_gvn);
5591   const Type* b_type = objb->Value(&_gvn);
5592   const TypeAryPtr* top_a = a_type->isa_aryptr();
5593   const TypeAryPtr* top_b = b_type->isa_aryptr();
5594   if (top_a == NULL || top_a->klass() == NULL ||
5595     top_b == NULL || top_b->klass() == NULL) {
5596     // failed array check
5597     return false;
5598   }
5599 
5600   Node* call;
5601   jvms()->set_should_reexecute(true);
5602 
5603   Node* obja_adr = make_unsafe_address(obja, aoffset);
5604   Node* objb_adr = make_unsafe_address(objb, boffset);
5605 
5606   call = make_runtime_call(RC_LEAF,
5607     OptoRuntime::vectorizedMismatch_Type(),
5608     stubAddr, stubName, TypePtr::BOTTOM,
5609     obja_adr, objb_adr, length, scale);
5610 
5611   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5612   set_result(result);
5613   return true;
5614 }
5615 
5616 /**
5617  * Calculate CRC32 for byte.
5618  * int java.util.zip.CRC32.update(int crc, int b)
5619  */
5620 bool LibraryCallKit::inline_updateCRC32() {
5621   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5622   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5623   // no receiver since it is static method
5624   Node* crc  = argument(0); // type: int
5625   Node* b    = argument(1); // type: int
5626 
5627   /*
5628    *    int c = ~ crc;
5629    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5630    *    b = b ^ (c >>> 8);
5631    *    crc = ~b;
5632    */
5633 
5634   Node* M1 = intcon(-1);
5635   crc = _gvn.transform(new XorINode(crc, M1));
5636   Node* result = _gvn.transform(new XorINode(crc, b));
5637   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5638 
5639   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5640   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5641   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5642   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5643 
5644   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5645   result = _gvn.transform(new XorINode(crc, result));
5646   result = _gvn.transform(new XorINode(result, M1));
5647   set_result(result);
5648   return true;
5649 }
5650 
5651 /**
5652  * Calculate CRC32 for byte[] array.
5653  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5654  */
5655 bool LibraryCallKit::inline_updateBytesCRC32() {
5656   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5657   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5658   // no receiver since it is static method
5659   Node* crc     = argument(0); // type: int
5660   Node* src     = argument(1); // type: oop
5661   Node* offset  = argument(2); // type: int
5662   Node* length  = argument(3); // type: int
5663 
5664   const Type* src_type = src->Value(&_gvn);
5665   const TypeAryPtr* top_src = src_type->isa_aryptr();
5666   if (top_src  == NULL || top_src->klass()  == NULL) {
5667     // failed array check
5668     return false;
5669   }
5670 
5671   // Figure out the size and type of the elements we will be copying.
5672   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5673   if (src_elem != T_BYTE) {
5674     return false;
5675   }
5676 
5677   // 'src_start' points to src array + scaled offset
5678   Node* src_start = array_element_address(src, offset, src_elem);
5679 
5680   // We assume that range check is done by caller.
5681   // TODO: generate range check (offset+length < src.length) in debug VM.
5682 
5683   // Call the stub.
5684   address stubAddr = StubRoutines::updateBytesCRC32();
5685   const char *stubName = "updateBytesCRC32";
5686 
5687   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5688                                  stubAddr, stubName, TypePtr::BOTTOM,
5689                                  crc, src_start, length);
5690   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5691   set_result(result);
5692   return true;
5693 }
5694 
5695 /**
5696  * Calculate CRC32 for ByteBuffer.
5697  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5698  */
5699 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5700   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5701   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5702   // no receiver since it is static method
5703   Node* crc     = argument(0); // type: int
5704   Node* src     = argument(1); // type: long
5705   Node* offset  = argument(3); // type: int
5706   Node* length  = argument(4); // type: int
5707 
5708   src = ConvL2X(src);  // adjust Java long to machine word
5709   Node* base = _gvn.transform(new CastX2PNode(src));
5710   offset = ConvI2X(offset);
5711 
5712   // 'src_start' points to src array + scaled offset
5713   Node* src_start = basic_plus_adr(top(), base, offset);
5714 
5715   // Call the stub.
5716   address stubAddr = StubRoutines::updateBytesCRC32();
5717   const char *stubName = "updateBytesCRC32";
5718 
5719   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5720                                  stubAddr, stubName, TypePtr::BOTTOM,
5721                                  crc, src_start, length);
5722   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5723   set_result(result);
5724   return true;
5725 }
5726 
5727 //------------------------------get_table_from_crc32c_class-----------------------
5728 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5729   Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5730   assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5731 
5732   return table;
5733 }
5734 
5735 //------------------------------inline_updateBytesCRC32C-----------------------
5736 //
5737 // Calculate CRC32C for byte[] array.
5738 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5739 //
5740 bool LibraryCallKit::inline_updateBytesCRC32C() {
5741   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5742   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5743   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5744   // no receiver since it is a static method
5745   Node* crc     = argument(0); // type: int
5746   Node* src     = argument(1); // type: oop
5747   Node* offset  = argument(2); // type: int
5748   Node* end     = argument(3); // type: int
5749 
5750   Node* length = _gvn.transform(new SubINode(end, offset));
5751 
5752   const Type* src_type = src->Value(&_gvn);
5753   const TypeAryPtr* top_src = src_type->isa_aryptr();
5754   if (top_src  == NULL || top_src->klass()  == NULL) {
5755     // failed array check
5756     return false;
5757   }
5758 
5759   // Figure out the size and type of the elements we will be copying.
5760   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5761   if (src_elem != T_BYTE) {
5762     return false;
5763   }
5764 
5765   // 'src_start' points to src array + scaled offset
5766   Node* src_start = array_element_address(src, offset, src_elem);
5767 
5768   // static final int[] byteTable in class CRC32C
5769   Node* table = get_table_from_crc32c_class(callee()->holder());
5770   Node* table_start = array_element_address(table, intcon(0), T_INT);
5771 
5772   // We assume that range check is done by caller.
5773   // TODO: generate range check (offset+length < src.length) in debug VM.
5774 
5775   // Call the stub.
5776   address stubAddr = StubRoutines::updateBytesCRC32C();
5777   const char *stubName = "updateBytesCRC32C";
5778 
5779   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5780                                  stubAddr, stubName, TypePtr::BOTTOM,
5781                                  crc, src_start, length, table_start);
5782   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5783   set_result(result);
5784   return true;
5785 }
5786 
5787 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5788 //
5789 // Calculate CRC32C for DirectByteBuffer.
5790 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5791 //
5792 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5793   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5794   assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5795   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5796   // no receiver since it is a static method
5797   Node* crc     = argument(0); // type: int
5798   Node* src     = argument(1); // type: long
5799   Node* offset  = argument(3); // type: int
5800   Node* end     = argument(4); // type: int
5801 
5802   Node* length = _gvn.transform(new SubINode(end, offset));
5803 
5804   src = ConvL2X(src);  // adjust Java long to machine word
5805   Node* base = _gvn.transform(new CastX2PNode(src));
5806   offset = ConvI2X(offset);
5807 
5808   // 'src_start' points to src array + scaled offset
5809   Node* src_start = basic_plus_adr(top(), base, offset);
5810 
5811   // static final int[] byteTable in class CRC32C
5812   Node* table = get_table_from_crc32c_class(callee()->holder());
5813   Node* table_start = array_element_address(table, intcon(0), T_INT);
5814 
5815   // Call the stub.
5816   address stubAddr = StubRoutines::updateBytesCRC32C();
5817   const char *stubName = "updateBytesCRC32C";
5818 
5819   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5820                                  stubAddr, stubName, TypePtr::BOTTOM,
5821                                  crc, src_start, length, table_start);
5822   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5823   set_result(result);
5824   return true;
5825 }
5826 
5827 //------------------------------inline_updateBytesAdler32----------------------
5828 //
5829 // Calculate Adler32 checksum for byte[] array.
5830 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5831 //
5832 bool LibraryCallKit::inline_updateBytesAdler32() {
5833   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5834   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5835   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5836   // no receiver since it is static method
5837   Node* crc     = argument(0); // type: int
5838   Node* src     = argument(1); // type: oop
5839   Node* offset  = argument(2); // type: int
5840   Node* length  = argument(3); // type: int
5841 
5842   const Type* src_type = src->Value(&_gvn);
5843   const TypeAryPtr* top_src = src_type->isa_aryptr();
5844   if (top_src  == NULL || top_src->klass()  == NULL) {
5845     // failed array check
5846     return false;
5847   }
5848 
5849   // Figure out the size and type of the elements we will be copying.
5850   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5851   if (src_elem != T_BYTE) {
5852     return false;
5853   }
5854 
5855   // 'src_start' points to src array + scaled offset
5856   Node* src_start = array_element_address(src, offset, src_elem);
5857 
5858   // We assume that range check is done by caller.
5859   // TODO: generate range check (offset+length < src.length) in debug VM.
5860 
5861   // Call the stub.
5862   address stubAddr = StubRoutines::updateBytesAdler32();
5863   const char *stubName = "updateBytesAdler32";
5864 
5865   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5866                                  stubAddr, stubName, TypePtr::BOTTOM,
5867                                  crc, src_start, length);
5868   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5869   set_result(result);
5870   return true;
5871 }
5872 
5873 //------------------------------inline_updateByteBufferAdler32---------------
5874 //
5875 // Calculate Adler32 checksum for DirectByteBuffer.
5876 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5877 //
5878 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5879   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5880   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5881   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5882   // no receiver since it is static method
5883   Node* crc     = argument(0); // type: int
5884   Node* src     = argument(1); // type: long
5885   Node* offset  = argument(3); // type: int
5886   Node* length  = argument(4); // type: int
5887 
5888   src = ConvL2X(src);  // adjust Java long to machine word
5889   Node* base = _gvn.transform(new CastX2PNode(src));
5890   offset = ConvI2X(offset);
5891 
5892   // 'src_start' points to src array + scaled offset
5893   Node* src_start = basic_plus_adr(top(), base, offset);
5894 
5895   // Call the stub.
5896   address stubAddr = StubRoutines::updateBytesAdler32();
5897   const char *stubName = "updateBytesAdler32";
5898 
5899   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5900                                  stubAddr, stubName, TypePtr::BOTTOM,
5901                                  crc, src_start, length);
5902 
5903   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5904   set_result(result);
5905   return true;
5906 }
5907 
5908 //----------------------------inline_reference_get----------------------------
5909 // public T java.lang.ref.Reference.get();
5910 bool LibraryCallKit::inline_reference_get() {
5911   const int referent_offset = java_lang_ref_Reference::referent_offset;
5912   guarantee(referent_offset > 0, "should have already been set");
5913 
5914   // Get the argument:
5915   Node* reference_obj = null_check_receiver();
5916   if (stopped()) return true;
5917 
5918   Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5919 
5920   ciInstanceKlass* klass = env()->Object_klass();
5921   const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5922 
5923   Node* no_ctrl = NULL;
5924   Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5925 
5926   // Use the pre-barrier to record the value in the referent field
5927   pre_barrier(false /* do_load */,
5928               control(),
5929               NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5930               result /* pre_val */,
5931               T_OBJECT);
5932 
5933   // Add memory barrier to prevent commoning reads from this field
5934   // across safepoint since GC can change its value.
5935   insert_mem_bar(Op_MemBarCPUOrder);
5936 
5937   set_result(result);
5938   return true;
5939 }
5940 
5941 
5942 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5943                                               bool is_exact=true, bool is_static=false,
5944                                               ciInstanceKlass * fromKls=NULL) {
5945   if (fromKls == NULL) {
5946     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5947     assert(tinst != NULL, "obj is null");
5948     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5949     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5950     fromKls = tinst->klass()->as_instance_klass();
5951   } else {
5952     assert(is_static, "only for static field access");
5953   }
5954   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5955                                               ciSymbol::make(fieldTypeString),
5956                                               is_static);
5957 
5958   assert (field != NULL, "undefined field");
5959   if (field == NULL) return (Node *) NULL;
5960 
5961   if (is_static) {
5962     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5963     fromObj = makecon(tip);
5964   }
5965 
5966   // Next code  copied from Parse::do_get_xxx():
5967 
5968   // Compute address and memory type.
5969   int offset  = field->offset_in_bytes();
5970   bool is_vol = field->is_volatile();
5971   ciType* field_klass = field->type();
5972   assert(field_klass->is_loaded(), "should be loaded");
5973   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5974   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5975   BasicType bt = field->layout_type();
5976 
5977   // Build the resultant type of the load
5978   const Type *type;
5979   if (bt == T_OBJECT) {
5980     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5981   } else {
5982     type = Type::get_const_basic_type(bt);
5983   }
5984 
5985   if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5986     insert_mem_bar(Op_MemBarVolatile);   // StoreLoad barrier
5987   }
5988   // Build the load.
5989   MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5990   Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5991   // If reference is volatile, prevent following memory ops from
5992   // floating up past the volatile read.  Also prevents commoning
5993   // another volatile read.
5994   if (is_vol) {
5995     // Memory barrier includes bogus read of value to force load BEFORE membar
5996     insert_mem_bar(Op_MemBarAcquire, loadedField);
5997   }
5998   return loadedField;
5999 }
6000 
6001 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6002                                                  bool is_exact = true, bool is_static = false,
6003                                                  ciInstanceKlass * fromKls = NULL) {
6004   if (fromKls == NULL) {
6005     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6006     assert(tinst != NULL, "obj is null");
6007     assert(tinst->klass()->is_loaded(), "obj is not loaded");
6008     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6009     fromKls = tinst->klass()->as_instance_klass();
6010   }
6011   else {
6012     assert(is_static, "only for static field access");
6013   }
6014   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6015     ciSymbol::make(fieldTypeString),
6016     is_static);
6017 
6018   assert(field != NULL, "undefined field");
6019   assert(!field->is_volatile(), "not defined for volatile fields");
6020 
6021   if (is_static) {
6022     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6023     fromObj = makecon(tip);
6024   }
6025 
6026   // Next code  copied from Parse::do_get_xxx():
6027 
6028   // Compute address and memory type.
6029   int offset = field->offset_in_bytes();
6030   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6031 
6032   return adr;
6033 }
6034 
6035 //------------------------------inline_aescrypt_Block-----------------------
6036 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6037   address stubAddr = NULL;
6038   const char *stubName;
6039   assert(UseAES, "need AES instruction support");
6040 
6041   switch(id) {
6042   case vmIntrinsics::_aescrypt_encryptBlock:
6043     stubAddr = StubRoutines::aescrypt_encryptBlock();
6044     stubName = "aescrypt_encryptBlock";
6045     break;
6046   case vmIntrinsics::_aescrypt_decryptBlock:
6047     stubAddr = StubRoutines::aescrypt_decryptBlock();
6048     stubName = "aescrypt_decryptBlock";
6049     break;
6050   }
6051   if (stubAddr == NULL) return false;
6052 
6053   Node* aescrypt_object = argument(0);
6054   Node* src             = argument(1);
6055   Node* src_offset      = argument(2);
6056   Node* dest            = argument(3);
6057   Node* dest_offset     = argument(4);
6058 
6059   // (1) src and dest are arrays.
6060   const Type* src_type = src->Value(&_gvn);
6061   const Type* dest_type = dest->Value(&_gvn);
6062   const TypeAryPtr* top_src = src_type->isa_aryptr();
6063   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6064   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6065 
6066   // for the quick and dirty code we will skip all the checks.
6067   // we are just trying to get the call to be generated.
6068   Node* src_start  = src;
6069   Node* dest_start = dest;
6070   if (src_offset != NULL || dest_offset != NULL) {
6071     assert(src_offset != NULL && dest_offset != NULL, "");
6072     src_start  = array_element_address(src,  src_offset,  T_BYTE);
6073     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6074   }
6075 
6076   // now need to get the start of its expanded key array
6077   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6078   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6079   if (k_start == NULL) return false;
6080 
6081   if (Matcher::pass_original_key_for_aes()) {
6082     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6083     // compatibility issues between Java key expansion and SPARC crypto instructions
6084     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6085     if (original_k_start == NULL) return false;
6086 
6087     // Call the stub.
6088     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6089                       stubAddr, stubName, TypePtr::BOTTOM,
6090                       src_start, dest_start, k_start, original_k_start);
6091   } else {
6092     // Call the stub.
6093     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6094                       stubAddr, stubName, TypePtr::BOTTOM,
6095                       src_start, dest_start, k_start);
6096   }
6097 
6098   return true;
6099 }
6100 
6101 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6102 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6103   address stubAddr = NULL;
6104   const char *stubName = NULL;
6105 
6106   assert(UseAES, "need AES instruction support");
6107 
6108   switch(id) {
6109   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6110     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6111     stubName = "cipherBlockChaining_encryptAESCrypt";
6112     break;
6113   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6114     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6115     stubName = "cipherBlockChaining_decryptAESCrypt";
6116     break;
6117   }
6118   if (stubAddr == NULL) return false;
6119 
6120   Node* cipherBlockChaining_object = argument(0);
6121   Node* src                        = argument(1);
6122   Node* src_offset                 = argument(2);
6123   Node* len                        = argument(3);
6124   Node* dest                       = argument(4);
6125   Node* dest_offset                = argument(5);
6126 
6127   // (1) src and dest are arrays.
6128   const Type* src_type = src->Value(&_gvn);
6129   const Type* dest_type = dest->Value(&_gvn);
6130   const TypeAryPtr* top_src = src_type->isa_aryptr();
6131   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6132   assert (top_src  != NULL && top_src->klass()  != NULL
6133           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6134 
6135   // checks are the responsibility of the caller
6136   Node* src_start  = src;
6137   Node* dest_start = dest;
6138   if (src_offset != NULL || dest_offset != NULL) {
6139     assert(src_offset != NULL && dest_offset != NULL, "");
6140     src_start  = array_element_address(src,  src_offset,  T_BYTE);
6141     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6142   }
6143 
6144   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6145   // (because of the predicated logic executed earlier).
6146   // so we cast it here safely.
6147   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6148 
6149   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6150   if (embeddedCipherObj == NULL) return false;
6151 
6152   // cast it to what we know it will be at runtime
6153   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6154   assert(tinst != NULL, "CBC obj is null");
6155   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6156   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6157   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6158 
6159   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6160   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6161   const TypeOopPtr* xtype = aklass->as_instance_type();
6162   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6163   aescrypt_object = _gvn.transform(aescrypt_object);
6164 
6165   // we need to get the start of the aescrypt_object's expanded key array
6166   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6167   if (k_start == NULL) return false;
6168 
6169   // similarly, get the start address of the r vector
6170   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6171   if (objRvec == NULL) return false;
6172   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6173 
6174   Node* cbcCrypt;
6175   if (Matcher::pass_original_key_for_aes()) {
6176     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6177     // compatibility issues between Java key expansion and SPARC crypto instructions
6178     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6179     if (original_k_start == NULL) return false;
6180 
6181     // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6182     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6183                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6184                                  stubAddr, stubName, TypePtr::BOTTOM,
6185                                  src_start, dest_start, k_start, r_start, len, original_k_start);
6186   } else {
6187     // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6188     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6189                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6190                                  stubAddr, stubName, TypePtr::BOTTOM,
6191                                  src_start, dest_start, k_start, r_start, len);
6192   }
6193 
6194   // return cipher length (int)
6195   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6196   set_result(retvalue);
6197   return true;
6198 }
6199 
6200 //------------------------------inline_counterMode_AESCrypt-----------------------
6201 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6202   assert(UseAES, "need AES instruction support");
6203   if (!UseAESCTRIntrinsics) return false;
6204 
6205   address stubAddr = NULL;
6206   const char *stubName = NULL;
6207   if (id == vmIntrinsics::_counterMode_AESCrypt) {
6208     stubAddr = StubRoutines::counterMode_AESCrypt();
6209     stubName = "counterMode_AESCrypt";
6210   }
6211   if (stubAddr == NULL) return false;
6212 
6213   Node* counterMode_object = argument(0);
6214   Node* src = argument(1);
6215   Node* src_offset = argument(2);
6216   Node* len = argument(3);
6217   Node* dest = argument(4);
6218   Node* dest_offset = argument(5);
6219 
6220   // (1) src and dest are arrays.
6221   const Type* src_type = src->Value(&_gvn);
6222   const Type* dest_type = dest->Value(&_gvn);
6223   const TypeAryPtr* top_src = src_type->isa_aryptr();
6224   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6225   assert(top_src != NULL && top_src->klass() != NULL &&
6226          top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6227 
6228   // checks are the responsibility of the caller
6229   Node* src_start = src;
6230   Node* dest_start = dest;
6231   if (src_offset != NULL || dest_offset != NULL) {
6232     assert(src_offset != NULL && dest_offset != NULL, "");
6233     src_start = array_element_address(src, src_offset, T_BYTE);
6234     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6235   }
6236 
6237   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6238   // (because of the predicated logic executed earlier).
6239   // so we cast it here safely.
6240   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6241   Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6242   if (embeddedCipherObj == NULL) return false;
6243   // cast it to what we know it will be at runtime
6244   const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6245   assert(tinst != NULL, "CTR obj is null");
6246   assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6247   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6248   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6249   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6250   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6251   const TypeOopPtr* xtype = aklass->as_instance_type();
6252   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6253   aescrypt_object = _gvn.transform(aescrypt_object);
6254   // we need to get the start of the aescrypt_object's expanded key array
6255   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6256   if (k_start == NULL) return false;
6257   // similarly, get the start address of the r vector
6258   Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6259   if (obj_counter == NULL) return false;
6260   Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6261 
6262   Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6263   if (saved_encCounter == NULL) return false;
6264   Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6265   Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6266 
6267   Node* ctrCrypt;
6268   if (Matcher::pass_original_key_for_aes()) {
6269     // no SPARC version for AES/CTR intrinsics now.
6270     return false;
6271   }
6272   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6273   ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6274                                OptoRuntime::counterMode_aescrypt_Type(),
6275                                stubAddr, stubName, TypePtr::BOTTOM,
6276                                src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6277 
6278   // return cipher length (int)
6279   Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6280   set_result(retvalue);
6281   return true;
6282 }
6283 
6284 //------------------------------get_key_start_from_aescrypt_object-----------------------
6285 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6286 #ifdef PPC64
6287   // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6288   // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6289   // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6290   // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6291   Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6292   assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6293   if (objSessionK == NULL) {
6294     return (Node *) NULL;
6295   }
6296   Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6297 #else
6298   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6299 #endif // PPC64
6300   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6301   if (objAESCryptKey == NULL) return (Node *) NULL;
6302 
6303   // now have the array, need to get the start address of the K array
6304   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6305   return k_start;
6306 }
6307 
6308 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6309 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6310   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6311   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6312   if (objAESCryptKey == NULL) return (Node *) NULL;
6313 
6314   // now have the array, need to get the start address of the lastKey array
6315   Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6316   return original_k_start;
6317 }
6318 
6319 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6320 // Return node representing slow path of predicate check.
6321 // the pseudo code we want to emulate with this predicate is:
6322 // for encryption:
6323 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6324 // for decryption:
6325 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6326 //    note cipher==plain is more conservative than the original java code but that's OK
6327 //
6328 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6329   // The receiver was checked for NULL already.
6330   Node* objCBC = argument(0);
6331 
6332   // Load embeddedCipher field of CipherBlockChaining object.
6333   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6334 
6335   // get AESCrypt klass for instanceOf check
6336   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6337   // will have same classloader as CipherBlockChaining object
6338   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6339   assert(tinst != NULL, "CBCobj is null");
6340   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6341 
6342   // we want to do an instanceof comparison against the AESCrypt class
6343   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6344   if (!klass_AESCrypt->is_loaded()) {
6345     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6346     Node* ctrl = control();
6347     set_control(top()); // no regular fast path
6348     return ctrl;
6349   }
6350   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6351 
6352   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6353   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
6354   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6355 
6356   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6357 
6358   // for encryption, we are done
6359   if (!decrypting)
6360     return instof_false;  // even if it is NULL
6361 
6362   // for decryption, we need to add a further check to avoid
6363   // taking the intrinsic path when cipher and plain are the same
6364   // see the original java code for why.
6365   RegionNode* region = new RegionNode(3);
6366   region->init_req(1, instof_false);
6367   Node* src = argument(1);
6368   Node* dest = argument(4);
6369   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6370   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6371   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6372   region->init_req(2, src_dest_conjoint);
6373 
6374   record_for_igvn(region);
6375   return _gvn.transform(region);
6376 }
6377 
6378 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6379 // Return node representing slow path of predicate check.
6380 // the pseudo code we want to emulate with this predicate is:
6381 // for encryption:
6382 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6383 // for decryption:
6384 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6385 //    note cipher==plain is more conservative than the original java code but that's OK
6386 //
6387 
6388 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6389   // The receiver was checked for NULL already.
6390   Node* objCTR = argument(0);
6391 
6392   // Load embeddedCipher field of CipherBlockChaining object.
6393   Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6394 
6395   // get AESCrypt klass for instanceOf check
6396   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6397   // will have same classloader as CipherBlockChaining object
6398   const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6399   assert(tinst != NULL, "CTRobj is null");
6400   assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6401 
6402   // we want to do an instanceof comparison against the AESCrypt class
6403   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6404   if (!klass_AESCrypt->is_loaded()) {
6405     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6406     Node* ctrl = control();
6407     set_control(top()); // no regular fast path
6408     return ctrl;
6409   }
6410 
6411   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6412   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6413   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6414   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6415   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6416 
6417   return instof_false; // even if it is NULL
6418 }
6419 
6420 //------------------------------inline_ghash_processBlocks
6421 bool LibraryCallKit::inline_ghash_processBlocks() {
6422   address stubAddr;
6423   const char *stubName;
6424   assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6425 
6426   stubAddr = StubRoutines::ghash_processBlocks();
6427   stubName = "ghash_processBlocks";
6428 
6429   Node* data           = argument(0);
6430   Node* offset         = argument(1);
6431   Node* len            = argument(2);
6432   Node* state          = argument(3);
6433   Node* subkeyH        = argument(4);
6434 
6435   Node* state_start  = array_element_address(state, intcon(0), T_LONG);
6436   assert(state_start, "state is NULL");
6437   Node* subkeyH_start  = array_element_address(subkeyH, intcon(0), T_LONG);
6438   assert(subkeyH_start, "subkeyH is NULL");
6439   Node* data_start  = array_element_address(data, offset, T_BYTE);
6440   assert(data_start, "data is NULL");
6441 
6442   Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6443                                   OptoRuntime::ghash_processBlocks_Type(),
6444                                   stubAddr, stubName, TypePtr::BOTTOM,
6445                                   state_start, subkeyH_start, data_start, len);
6446   return true;
6447 }
6448 
6449 //------------------------------inline_sha_implCompress-----------------------
6450 //
6451 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6452 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6453 //
6454 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6455 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6456 //
6457 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6458 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6459 //
6460 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6461   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6462 
6463   Node* sha_obj = argument(0);
6464   Node* src     = argument(1); // type oop
6465   Node* ofs     = argument(2); // type int
6466 
6467   const Type* src_type = src->Value(&_gvn);
6468   const TypeAryPtr* top_src = src_type->isa_aryptr();
6469   if (top_src  == NULL || top_src->klass()  == NULL) {
6470     // failed array check
6471     return false;
6472   }
6473   // Figure out the size and type of the elements we will be copying.
6474   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6475   if (src_elem != T_BYTE) {
6476     return false;
6477   }
6478   // 'src_start' points to src array + offset
6479   Node* src_start = array_element_address(src, ofs, src_elem);
6480   Node* state = NULL;
6481   address stubAddr;
6482   const char *stubName;
6483 
6484   switch(id) {
6485   case vmIntrinsics::_sha_implCompress:
6486     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6487     state = get_state_from_sha_object(sha_obj);
6488     stubAddr = StubRoutines::sha1_implCompress();
6489     stubName = "sha1_implCompress";
6490     break;
6491   case vmIntrinsics::_sha2_implCompress:
6492     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6493     state = get_state_from_sha_object(sha_obj);
6494     stubAddr = StubRoutines::sha256_implCompress();
6495     stubName = "sha256_implCompress";
6496     break;
6497   case vmIntrinsics::_sha5_implCompress:
6498     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6499     state = get_state_from_sha5_object(sha_obj);
6500     stubAddr = StubRoutines::sha512_implCompress();
6501     stubName = "sha512_implCompress";
6502     break;
6503   default:
6504     fatal_unexpected_iid(id);
6505     return false;
6506   }
6507   if (state == NULL) return false;
6508 
6509   // Call the stub.
6510   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6511                                  stubAddr, stubName, TypePtr::BOTTOM,
6512                                  src_start, state);
6513 
6514   return true;
6515 }
6516 
6517 //------------------------------inline_digestBase_implCompressMB-----------------------
6518 //
6519 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6520 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6521 //
6522 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6523   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6524          "need SHA1/SHA256/SHA512 instruction support");
6525   assert((uint)predicate < 3, "sanity");
6526   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6527 
6528   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6529   Node* src            = argument(1); // byte[] array
6530   Node* ofs            = argument(2); // type int
6531   Node* limit          = argument(3); // type int
6532 
6533   const Type* src_type = src->Value(&_gvn);
6534   const TypeAryPtr* top_src = src_type->isa_aryptr();
6535   if (top_src  == NULL || top_src->klass()  == NULL) {
6536     // failed array check
6537     return false;
6538   }
6539   // Figure out the size and type of the elements we will be copying.
6540   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6541   if (src_elem != T_BYTE) {
6542     return false;
6543   }
6544   // 'src_start' points to src array + offset
6545   Node* src_start = array_element_address(src, ofs, src_elem);
6546 
6547   const char* klass_SHA_name = NULL;
6548   const char* stub_name = NULL;
6549   address     stub_addr = NULL;
6550   bool        long_state = false;
6551 
6552   switch (predicate) {
6553   case 0:
6554     if (UseSHA1Intrinsics) {
6555       klass_SHA_name = "sun/security/provider/SHA";
6556       stub_name = "sha1_implCompressMB";
6557       stub_addr = StubRoutines::sha1_implCompressMB();
6558     }
6559     break;
6560   case 1:
6561     if (UseSHA256Intrinsics) {
6562       klass_SHA_name = "sun/security/provider/SHA2";
6563       stub_name = "sha256_implCompressMB";
6564       stub_addr = StubRoutines::sha256_implCompressMB();
6565     }
6566     break;
6567   case 2:
6568     if (UseSHA512Intrinsics) {
6569       klass_SHA_name = "sun/security/provider/SHA5";
6570       stub_name = "sha512_implCompressMB";
6571       stub_addr = StubRoutines::sha512_implCompressMB();
6572       long_state = true;
6573     }
6574     break;
6575   default:
6576     fatal("unknown SHA intrinsic predicate: %d", predicate);
6577   }
6578   if (klass_SHA_name != NULL) {
6579     // get DigestBase klass to lookup for SHA klass
6580     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6581     assert(tinst != NULL, "digestBase_obj is not instance???");
6582     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6583 
6584     ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6585     assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6586     ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6587     return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6588   }
6589   return false;
6590 }
6591 //------------------------------inline_sha_implCompressMB-----------------------
6592 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6593                                                bool long_state, address stubAddr, const char *stubName,
6594                                                Node* src_start, Node* ofs, Node* limit) {
6595   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6596   const TypeOopPtr* xtype = aklass->as_instance_type();
6597   Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6598   sha_obj = _gvn.transform(sha_obj);
6599 
6600   Node* state;
6601   if (long_state) {
6602     state = get_state_from_sha5_object(sha_obj);
6603   } else {
6604     state = get_state_from_sha_object(sha_obj);
6605   }
6606   if (state == NULL) return false;
6607 
6608   // Call the stub.
6609   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6610                                  OptoRuntime::digestBase_implCompressMB_Type(),
6611                                  stubAddr, stubName, TypePtr::BOTTOM,
6612                                  src_start, state, ofs, limit);
6613   // return ofs (int)
6614   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6615   set_result(result);
6616 
6617   return true;
6618 }
6619 
6620 //------------------------------get_state_from_sha_object-----------------------
6621 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6622   Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6623   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6624   if (sha_state == NULL) return (Node *) NULL;
6625 
6626   // now have the array, need to get the start address of the state array
6627   Node* state = array_element_address(sha_state, intcon(0), T_INT);
6628   return state;
6629 }
6630 
6631 //------------------------------get_state_from_sha5_object-----------------------
6632 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6633   Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6634   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6635   if (sha_state == NULL) return (Node *) NULL;
6636 
6637   // now have the array, need to get the start address of the state array
6638   Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6639   return state;
6640 }
6641 
6642 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6643 // Return node representing slow path of predicate check.
6644 // the pseudo code we want to emulate with this predicate is:
6645 //    if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6646 //
6647 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6648   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6649          "need SHA1/SHA256/SHA512 instruction support");
6650   assert((uint)predicate < 3, "sanity");
6651 
6652   // The receiver was checked for NULL already.
6653   Node* digestBaseObj = argument(0);
6654 
6655   // get DigestBase klass for instanceOf check
6656   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6657   assert(tinst != NULL, "digestBaseObj is null");
6658   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6659 
6660   const char* klass_SHA_name = NULL;
6661   switch (predicate) {
6662   case 0:
6663     if (UseSHA1Intrinsics) {
6664       // we want to do an instanceof comparison against the SHA class
6665       klass_SHA_name = "sun/security/provider/SHA";
6666     }
6667     break;
6668   case 1:
6669     if (UseSHA256Intrinsics) {
6670       // we want to do an instanceof comparison against the SHA2 class
6671       klass_SHA_name = "sun/security/provider/SHA2";
6672     }
6673     break;
6674   case 2:
6675     if (UseSHA512Intrinsics) {
6676       // we want to do an instanceof comparison against the SHA5 class
6677       klass_SHA_name = "sun/security/provider/SHA5";
6678     }
6679     break;
6680   default:
6681     fatal("unknown SHA intrinsic predicate: %d", predicate);
6682   }
6683 
6684   ciKlass* klass_SHA = NULL;
6685   if (klass_SHA_name != NULL) {
6686     klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6687   }
6688   if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6689     // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6690     Node* ctrl = control();
6691     set_control(top()); // no intrinsic path
6692     return ctrl;
6693   }
6694   ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6695 
6696   Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6697   Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6698   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6699   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6700 
6701   return instof_false;  // even if it is NULL
6702 }
6703 
6704 bool LibraryCallKit::inline_profileBoolean() {
6705   Node* counts = argument(1);
6706   const TypeAryPtr* ary = NULL;
6707   ciArray* aobj = NULL;
6708   if (counts->is_Con()
6709       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6710       && (aobj = ary->const_oop()->as_array()) != NULL
6711       && (aobj->length() == 2)) {
6712     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6713     jint false_cnt = aobj->element_value(0).as_int();
6714     jint  true_cnt = aobj->element_value(1).as_int();
6715 
6716     if (C->log() != NULL) {
6717       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6718                      false_cnt, true_cnt);
6719     }
6720 
6721     if (false_cnt + true_cnt == 0) {
6722       // According to profile, never executed.
6723       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6724                           Deoptimization::Action_reinterpret);
6725       return true;
6726     }
6727 
6728     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6729     // is a number of each value occurrences.
6730     Node* result = argument(0);
6731     if (false_cnt == 0 || true_cnt == 0) {
6732       // According to profile, one value has been never seen.
6733       int expected_val = (false_cnt == 0) ? 1 : 0;
6734 
6735       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6736       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6737 
6738       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6739       Node* fast_path = _gvn.transform(new IfTrueNode(check));
6740       Node* slow_path = _gvn.transform(new IfFalseNode(check));
6741 
6742       { // Slow path: uncommon trap for never seen value and then reexecute
6743         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6744         // the value has been seen at least once.
6745         PreserveJVMState pjvms(this);
6746         PreserveReexecuteState preexecs(this);
6747         jvms()->set_should_reexecute(true);
6748 
6749         set_control(slow_path);
6750         set_i_o(i_o());
6751 
6752         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6753                             Deoptimization::Action_reinterpret);
6754       }
6755       // The guard for never seen value enables sharpening of the result and
6756       // returning a constant. It allows to eliminate branches on the same value
6757       // later on.
6758       set_control(fast_path);
6759       result = intcon(expected_val);
6760     }
6761     // Stop profiling.
6762     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6763     // By replacing method body with profile data (represented as ProfileBooleanNode
6764     // on IR level) we effectively disable profiling.
6765     // It enables full speed execution once optimized code is generated.
6766     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6767     C->record_for_igvn(profile);
6768     set_result(profile);
6769     return true;
6770   } else {
6771     // Continue profiling.
6772     // Profile data isn't available at the moment. So, execute method's bytecode version.
6773     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6774     // is compiled and counters aren't available since corresponding MethodHandle
6775     // isn't a compile-time constant.
6776     return false;
6777   }
6778 }
6779 
6780 bool LibraryCallKit::inline_isCompileConstant() {
6781   Node* n = argument(0);
6782   set_result(n->is_Con() ? intcon(1) : intcon(0));
6783   return true;
6784 }