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);
  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_math(vmIntrinsics::ID id);
 226   template <typename OverflowOp>
 227   bool inline_math_overflow(Node* arg1, Node* arg2);
 228   void inline_math_mathExact(Node* math, Node* test);
 229   bool inline_math_addExactI(bool is_increment);
 230   bool inline_math_addExactL(bool is_increment);
 231   bool inline_math_multiplyExactI();
 232   bool inline_math_multiplyExactL();
 233   bool inline_math_negateExactI();
 234   bool inline_math_negateExactL();
 235   bool inline_math_subtractExactI(bool is_decrement);
 236   bool inline_math_subtractExactL(bool is_decrement);
 237   bool inline_min_max(vmIntrinsics::ID id);
 238   bool inline_notify(vmIntrinsics::ID id);
 239   Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
 240   // This returns Type::AnyPtr, RawPtr, or OopPtr.
 241   int classify_unsafe_addr(Node* &base, Node* &offset);
 242   Node* make_unsafe_address(Node* base, Node* offset);
 243   // Helper for inline_unsafe_access.
 244   // Generates the guards that check whether the result of
 245   // Unsafe.getObject should be recorded in an SATB log buffer.
 246   void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
 247 
 248   typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
 249   bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
 250   static bool klass_needs_init_guard(Node* kls);
 251   bool inline_unsafe_allocate();
 252   bool inline_unsafe_newArray(bool uninitialized);
 253   bool inline_unsafe_copyMemory();
 254   bool inline_native_currentThread();
 255 
 256   bool inline_native_time_funcs(address method, const char* funcName);
 257   bool inline_native_isInterrupted();
 258   bool inline_native_Class_query(vmIntrinsics::ID id);
 259   bool inline_native_subtype_check();
 260   bool inline_native_getLength();
 261   bool inline_array_copyOf(bool is_copyOfRange);
 262   bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
 263   bool inline_preconditions_checkIndex();
 264   void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
 265   bool inline_native_clone(bool is_virtual);
 266   bool inline_native_Reflection_getCallerClass();
 267   // Helper function for inlining native object hash method
 268   bool inline_native_hashcode(bool is_virtual, bool is_static);
 269   bool inline_native_getClass();
 270 
 271   // Helper functions for inlining arraycopy
 272   bool inline_arraycopy();
 273   AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
 274                                                 RegionNode* slow_region);
 275   JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
 276   void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
 277 
 278   typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
 279   MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
 280   MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
 281   bool inline_unsafe_load_store(BasicType type,  LoadStoreKind kind, AccessKind access_kind);
 282   bool inline_unsafe_fence(vmIntrinsics::ID id);
 283   bool inline_onspinwait();
 284   bool inline_fp_conversions(vmIntrinsics::ID id);
 285   bool inline_number_methods(vmIntrinsics::ID id);
 286   bool inline_reference_get();
 287   bool inline_Class_cast();
 288   bool inline_aescrypt_Block(vmIntrinsics::ID id);
 289   bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
 290   bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
 291   Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
 292   Node* inline_counterMode_AESCrypt_predicate();
 293   Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
 294   Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
 295   bool inline_ghash_processBlocks();
 296   bool inline_sha_implCompress(vmIntrinsics::ID id);
 297   bool inline_digestBase_implCompressMB(int predicate);
 298   bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
 299                                  bool long_state, address stubAddr, const char *stubName,
 300                                  Node* src_start, Node* ofs, Node* limit);
 301   Node* get_state_from_sha_object(Node *sha_object);
 302   Node* get_state_from_sha5_object(Node *sha_object);
 303   Node* inline_digestBase_implCompressMB_predicate(int predicate);
 304   bool inline_encodeISOArray();
 305   bool inline_updateCRC32();
 306   bool inline_updateBytesCRC32();
 307   bool inline_updateByteBufferCRC32();
 308   Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
 309   bool inline_updateBytesCRC32C();
 310   bool inline_updateDirectByteBufferCRC32C();
 311   bool inline_updateBytesAdler32();
 312   bool inline_updateByteBufferAdler32();
 313   bool inline_multiplyToLen();
 314   bool inline_hasNegatives();
 315   bool inline_squareToLen();
 316   bool inline_mulAdd();
 317   bool inline_montgomeryMultiply();
 318   bool inline_montgomerySquare();
 319   bool inline_vectorizedMismatch();
 320 
 321   bool inline_profileBoolean();
 322   bool inline_isCompileConstant();
 323 };
 324 
 325 //---------------------------make_vm_intrinsic----------------------------
 326 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
 327   vmIntrinsics::ID id = m->intrinsic_id();
 328   assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
 329 
 330   if (!m->is_loaded()) {
 331     // Do not attempt to inline unloaded methods.
 332     return NULL;
 333   }
 334 
 335   C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
 336   bool is_available = false;
 337 
 338   {
 339     // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
 340     // the compiler must transition to '_thread_in_vm' state because both
 341     // methods access VM-internal data.
 342     VM_ENTRY_MARK;
 343     methodHandle mh(THREAD, m->get_Method());
 344     is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
 345                    !C->directive()->is_intrinsic_disabled(mh) &&
 346                    !vmIntrinsics::is_disabled_by_flags(mh);
 347 
 348   }
 349 
 350   if (is_available) {
 351     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
 352     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
 353     return new LibraryIntrinsic(m, is_virtual,
 354                                 vmIntrinsics::predicates_needed(id),
 355                                 vmIntrinsics::does_virtual_dispatch(id),
 356                                 (vmIntrinsics::ID) id);
 357   } else {
 358     return NULL;
 359   }
 360 }
 361 
 362 //----------------------register_library_intrinsics-----------------------
 363 // Initialize this file's data structures, for each Compile instance.
 364 void Compile::register_library_intrinsics() {
 365   // Nothing to do here.
 366 }
 367 
 368 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
 369   LibraryCallKit kit(jvms, this);
 370   Compile* C = kit.C;
 371   int nodes = C->unique();
 372 #ifndef PRODUCT
 373   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 374     char buf[1000];
 375     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 376     tty->print_cr("Intrinsic %s", str);
 377   }
 378 #endif
 379   ciMethod* callee = kit.callee();
 380   const int bci    = kit.bci();
 381 
 382   // Try to inline the intrinsic.
 383   if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
 384       kit.try_to_inline(_last_predicate)) {
 385     if (C->print_intrinsics() || C->print_inlining()) {
 386       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
 387     }
 388     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 389     if (C->log()) {
 390       C->log()->elem("intrinsic id='%s'%s nodes='%d'",
 391                      vmIntrinsics::name_at(intrinsic_id()),
 392                      (is_virtual() ? " virtual='1'" : ""),
 393                      C->unique() - nodes);
 394     }
 395     // Push the result from the inlined method onto the stack.
 396     kit.push_result();
 397     C->print_inlining_update(this);
 398     return kit.transfer_exceptions_into_jvms();
 399   }
 400 
 401   // The intrinsic bailed out
 402   if (C->print_intrinsics() || C->print_inlining()) {
 403     if (jvms->has_method()) {
 404       // Not a root compile.
 405       const char* msg;
 406       if (callee->intrinsic_candidate()) {
 407         msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
 408       } else {
 409         msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
 410                            : "failed to inline (intrinsic), method not annotated";
 411       }
 412       C->print_inlining(callee, jvms->depth() - 1, bci, msg);
 413     } else {
 414       // Root compile
 415       tty->print("Did not generate intrinsic %s%s at bci:%d in",
 416                vmIntrinsics::name_at(intrinsic_id()),
 417                (is_virtual() ? " (virtual)" : ""), bci);
 418     }
 419   }
 420   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 421   C->print_inlining_update(this);
 422   return NULL;
 423 }
 424 
 425 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
 426   LibraryCallKit kit(jvms, this);
 427   Compile* C = kit.C;
 428   int nodes = C->unique();
 429   _last_predicate = predicate;
 430 #ifndef PRODUCT
 431   assert(is_predicated() && predicate < predicates_count(), "sanity");
 432   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 433     char buf[1000];
 434     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 435     tty->print_cr("Predicate for intrinsic %s", str);
 436   }
 437 #endif
 438   ciMethod* callee = kit.callee();
 439   const int bci    = kit.bci();
 440 
 441   Node* slow_ctl = kit.try_to_predicate(predicate);
 442   if (!kit.failing()) {
 443     if (C->print_intrinsics() || C->print_inlining()) {
 444       C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
 445     }
 446     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 447     if (C->log()) {
 448       C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
 449                      vmIntrinsics::name_at(intrinsic_id()),
 450                      (is_virtual() ? " virtual='1'" : ""),
 451                      C->unique() - nodes);
 452     }
 453     return slow_ctl; // Could be NULL if the check folds.
 454   }
 455 
 456   // The intrinsic bailed out
 457   if (C->print_intrinsics() || C->print_inlining()) {
 458     if (jvms->has_method()) {
 459       // Not a root compile.
 460       const char* msg = "failed to generate predicate for intrinsic";
 461       C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
 462     } else {
 463       // Root compile
 464       C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
 465                                         vmIntrinsics::name_at(intrinsic_id()),
 466                                         (is_virtual() ? " (virtual)" : ""), bci);
 467     }
 468   }
 469   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 470   return NULL;
 471 }
 472 
 473 bool LibraryCallKit::try_to_inline(int predicate) {
 474   // Handle symbolic names for otherwise undistinguished boolean switches:
 475   const bool is_store       = true;
 476   const bool is_compress    = true;
 477   const bool is_static      = true;
 478   const bool is_volatile    = true;
 479 
 480   if (!jvms()->has_method()) {
 481     // Root JVMState has a null method.
 482     assert(map()->memory()->Opcode() == Op_Parm, "");
 483     // Insert the memory aliasing node
 484     set_all_memory(reset_memory());
 485   }
 486   assert(merged_memory(), "");
 487 
 488 
 489   switch (intrinsic_id()) {
 490   case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
 491   case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
 492   case vmIntrinsics::_getClass:                 return inline_native_getClass();
 493 
 494   case vmIntrinsics::_dsin:
 495   case vmIntrinsics::_dcos:
 496   case vmIntrinsics::_dtan:
 497   case vmIntrinsics::_dabs:
 498   case vmIntrinsics::_datan2:
 499   case vmIntrinsics::_dsqrt:
 500   case vmIntrinsics::_dexp:
 501   case vmIntrinsics::_dlog:
 502   case vmIntrinsics::_dlog10:
 503   case vmIntrinsics::_dpow:                     return inline_math_native(intrinsic_id());
 504 
 505   case vmIntrinsics::_min:
 506   case vmIntrinsics::_max:                      return inline_min_max(intrinsic_id());
 507 
 508   case vmIntrinsics::_notify:
 509   case vmIntrinsics::_notifyAll:
 510     if (InlineNotify) {
 511       return inline_notify(intrinsic_id());
 512     }
 513     return false;
 514 
 515   case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
 516   case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
 517   case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
 518   case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
 519   case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
 520   case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
 521   case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
 522   case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
 523   case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
 524   case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
 525   case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
 526   case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
 527 
 528   case vmIntrinsics::_arraycopy:                return inline_arraycopy();
 529 
 530   case vmIntrinsics::_compareToL:               return inline_string_compareTo(StrIntrinsicNode::LL);
 531   case vmIntrinsics::_compareToU:               return inline_string_compareTo(StrIntrinsicNode::UU);
 532   case vmIntrinsics::_compareToLU:              return inline_string_compareTo(StrIntrinsicNode::LU);
 533   case vmIntrinsics::_compareToUL:              return inline_string_compareTo(StrIntrinsicNode::UL);
 534 
 535   case vmIntrinsics::_indexOfL:                 return inline_string_indexOf(StrIntrinsicNode::LL);
 536   case vmIntrinsics::_indexOfU:                 return inline_string_indexOf(StrIntrinsicNode::UU);
 537   case vmIntrinsics::_indexOfUL:                return inline_string_indexOf(StrIntrinsicNode::UL);
 538   case vmIntrinsics::_indexOfIL:                return inline_string_indexOfI(StrIntrinsicNode::LL);
 539   case vmIntrinsics::_indexOfIU:                return inline_string_indexOfI(StrIntrinsicNode::UU);
 540   case vmIntrinsics::_indexOfIUL:               return inline_string_indexOfI(StrIntrinsicNode::UL);
 541   case vmIntrinsics::_indexOfU_char:            return inline_string_indexOfChar();
 542 
 543   case vmIntrinsics::_equalsL:                  return inline_string_equals(StrIntrinsicNode::LL);
 544   case vmIntrinsics::_equalsU:                  return inline_string_equals(StrIntrinsicNode::UU);
 545 
 546   case vmIntrinsics::_toBytesStringU:           return inline_string_toBytesU();
 547   case vmIntrinsics::_getCharsStringU:          return inline_string_getCharsU();
 548   case vmIntrinsics::_getCharStringU:           return inline_string_char_access(!is_store);
 549   case vmIntrinsics::_putCharStringU:           return inline_string_char_access( is_store);
 550 
 551   case vmIntrinsics::_compressStringC:
 552   case vmIntrinsics::_compressStringB:          return inline_string_copy( is_compress);
 553   case vmIntrinsics::_inflateStringC:
 554   case vmIntrinsics::_inflateStringB:           return inline_string_copy(!is_compress);
 555 
 556   case vmIntrinsics::_getObject:                return inline_unsafe_access(!is_store, T_OBJECT,   Relaxed, false);
 557   case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_store, T_BOOLEAN,  Relaxed, false);
 558   case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_store, T_BYTE,     Relaxed, false);
 559   case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, false);
 560   case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, false);
 561   case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_store, T_INT,      Relaxed, false);
 562   case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_store, T_LONG,     Relaxed, false);
 563   case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_store, T_FLOAT,    Relaxed, false);
 564   case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_store, T_DOUBLE,   Relaxed, false);
 565 
 566   case vmIntrinsics::_putObject:                return inline_unsafe_access( is_store, T_OBJECT,   Relaxed, false);
 567   case vmIntrinsics::_putBoolean:               return inline_unsafe_access( is_store, T_BOOLEAN,  Relaxed, false);
 568   case vmIntrinsics::_putByte:                  return inline_unsafe_access( is_store, T_BYTE,     Relaxed, false);
 569   case vmIntrinsics::_putShort:                 return inline_unsafe_access( is_store, T_SHORT,    Relaxed, false);
 570   case vmIntrinsics::_putChar:                  return inline_unsafe_access( is_store, T_CHAR,     Relaxed, false);
 571   case vmIntrinsics::_putInt:                   return inline_unsafe_access( is_store, T_INT,      Relaxed, false);
 572   case vmIntrinsics::_putLong:                  return inline_unsafe_access( is_store, T_LONG,     Relaxed, false);
 573   case vmIntrinsics::_putFloat:                 return inline_unsafe_access( is_store, T_FLOAT,    Relaxed, false);
 574   case vmIntrinsics::_putDouble:                return inline_unsafe_access( is_store, T_DOUBLE,   Relaxed, false);
 575 
 576   case vmIntrinsics::_getObjectVolatile:        return inline_unsafe_access(!is_store, T_OBJECT,   Volatile, false);
 577   case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_store, T_BOOLEAN,  Volatile, false);
 578   case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_store, T_BYTE,     Volatile, false);
 579   case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_store, T_SHORT,    Volatile, false);
 580   case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_store, T_CHAR,     Volatile, false);
 581   case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_store, T_INT,      Volatile, false);
 582   case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_store, T_LONG,     Volatile, false);
 583   case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_store, T_FLOAT,    Volatile, false);
 584   case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_store, T_DOUBLE,   Volatile, false);
 585 
 586   case vmIntrinsics::_putObjectVolatile:        return inline_unsafe_access( is_store, T_OBJECT,   Volatile, false);
 587   case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access( is_store, T_BOOLEAN,  Volatile, false);
 588   case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access( is_store, T_BYTE,     Volatile, false);
 589   case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access( is_store, T_SHORT,    Volatile, false);
 590   case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access( is_store, T_CHAR,     Volatile, false);
 591   case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access( is_store, T_INT,      Volatile, false);
 592   case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access( is_store, T_LONG,     Volatile, false);
 593   case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access( is_store, T_FLOAT,    Volatile, false);
 594   case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access( is_store, T_DOUBLE,   Volatile, false);
 595 
 596   case vmIntrinsics::_getShortUnaligned:        return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, true);
 597   case vmIntrinsics::_getCharUnaligned:         return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, true);
 598   case vmIntrinsics::_getIntUnaligned:          return inline_unsafe_access(!is_store, T_INT,      Relaxed, true);
 599   case vmIntrinsics::_getLongUnaligned:         return inline_unsafe_access(!is_store, T_LONG,     Relaxed, true);
 600 
 601   case vmIntrinsics::_putShortUnaligned:        return inline_unsafe_access( is_store, T_SHORT,    Relaxed, true);
 602   case vmIntrinsics::_putCharUnaligned:         return inline_unsafe_access( is_store, T_CHAR,     Relaxed, true);
 603   case vmIntrinsics::_putIntUnaligned:          return inline_unsafe_access( is_store, T_INT,      Relaxed, true);
 604   case vmIntrinsics::_putLongUnaligned:         return inline_unsafe_access( is_store, T_LONG,     Relaxed, true);
 605 
 606   case vmIntrinsics::_getObjectAcquire:         return inline_unsafe_access(!is_store, T_OBJECT,   Acquire, false);
 607   case vmIntrinsics::_getBooleanAcquire:        return inline_unsafe_access(!is_store, T_BOOLEAN,  Acquire, false);
 608   case vmIntrinsics::_getByteAcquire:           return inline_unsafe_access(!is_store, T_BYTE,     Acquire, false);
 609   case vmIntrinsics::_getShortAcquire:          return inline_unsafe_access(!is_store, T_SHORT,    Acquire, false);
 610   case vmIntrinsics::_getCharAcquire:           return inline_unsafe_access(!is_store, T_CHAR,     Acquire, false);
 611   case vmIntrinsics::_getIntAcquire:            return inline_unsafe_access(!is_store, T_INT,      Acquire, false);
 612   case vmIntrinsics::_getLongAcquire:           return inline_unsafe_access(!is_store, T_LONG,     Acquire, false);
 613   case vmIntrinsics::_getFloatAcquire:          return inline_unsafe_access(!is_store, T_FLOAT,    Acquire, false);
 614   case vmIntrinsics::_getDoubleAcquire:         return inline_unsafe_access(!is_store, T_DOUBLE,   Acquire, false);
 615 
 616   case vmIntrinsics::_putObjectRelease:         return inline_unsafe_access( is_store, T_OBJECT,   Release, false);
 617   case vmIntrinsics::_putBooleanRelease:        return inline_unsafe_access( is_store, T_BOOLEAN,  Release, false);
 618   case vmIntrinsics::_putByteRelease:           return inline_unsafe_access( is_store, T_BYTE,     Release, false);
 619   case vmIntrinsics::_putShortRelease:          return inline_unsafe_access( is_store, T_SHORT,    Release, false);
 620   case vmIntrinsics::_putCharRelease:           return inline_unsafe_access( is_store, T_CHAR,     Release, false);
 621   case vmIntrinsics::_putIntRelease:            return inline_unsafe_access( is_store, T_INT,      Release, false);
 622   case vmIntrinsics::_putLongRelease:           return inline_unsafe_access( is_store, T_LONG,     Release, false);
 623   case vmIntrinsics::_putFloatRelease:          return inline_unsafe_access( is_store, T_FLOAT,    Release, false);
 624   case vmIntrinsics::_putDoubleRelease:         return inline_unsafe_access( is_store, T_DOUBLE,   Release, false);
 625 
 626   case vmIntrinsics::_getObjectOpaque:          return inline_unsafe_access(!is_store, T_OBJECT,   Opaque, false);
 627   case vmIntrinsics::_getBooleanOpaque:         return inline_unsafe_access(!is_store, T_BOOLEAN,  Opaque, false);
 628   case vmIntrinsics::_getByteOpaque:            return inline_unsafe_access(!is_store, T_BYTE,     Opaque, false);
 629   case vmIntrinsics::_getShortOpaque:           return inline_unsafe_access(!is_store, T_SHORT,    Opaque, false);
 630   case vmIntrinsics::_getCharOpaque:            return inline_unsafe_access(!is_store, T_CHAR,     Opaque, false);
 631   case vmIntrinsics::_getIntOpaque:             return inline_unsafe_access(!is_store, T_INT,      Opaque, false);
 632   case vmIntrinsics::_getLongOpaque:            return inline_unsafe_access(!is_store, T_LONG,     Opaque, false);
 633   case vmIntrinsics::_getFloatOpaque:           return inline_unsafe_access(!is_store, T_FLOAT,    Opaque, false);
 634   case vmIntrinsics::_getDoubleOpaque:          return inline_unsafe_access(!is_store, T_DOUBLE,   Opaque, false);
 635 
 636   case vmIntrinsics::_putObjectOpaque:          return inline_unsafe_access( is_store, T_OBJECT,   Opaque, false);
 637   case vmIntrinsics::_putBooleanOpaque:         return inline_unsafe_access( is_store, T_BOOLEAN,  Opaque, false);
 638   case vmIntrinsics::_putByteOpaque:            return inline_unsafe_access( is_store, T_BYTE,     Opaque, false);
 639   case vmIntrinsics::_putShortOpaque:           return inline_unsafe_access( is_store, T_SHORT,    Opaque, false);
 640   case vmIntrinsics::_putCharOpaque:            return inline_unsafe_access( is_store, T_CHAR,     Opaque, false);
 641   case vmIntrinsics::_putIntOpaque:             return inline_unsafe_access( is_store, T_INT,      Opaque, false);
 642   case vmIntrinsics::_putLongOpaque:            return inline_unsafe_access( is_store, T_LONG,     Opaque, false);
 643   case vmIntrinsics::_putFloatOpaque:           return inline_unsafe_access( is_store, T_FLOAT,    Opaque, false);
 644   case vmIntrinsics::_putDoubleOpaque:          return inline_unsafe_access( is_store, T_DOUBLE,   Opaque, false);
 645 
 646   case vmIntrinsics::_compareAndSwapObject:             return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap,      Volatile);
 647   case vmIntrinsics::_compareAndSwapByte:               return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap,      Volatile);
 648   case vmIntrinsics::_compareAndSwapShort:              return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap,      Volatile);
 649   case vmIntrinsics::_compareAndSwapInt:                return inline_unsafe_load_store(T_INT,    LS_cmp_swap,      Volatile);
 650   case vmIntrinsics::_compareAndSwapLong:               return inline_unsafe_load_store(T_LONG,   LS_cmp_swap,      Volatile);
 651 
 652   case vmIntrinsics::_weakCompareAndSwapObject:         return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
 653   case vmIntrinsics::_weakCompareAndSwapObjectAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
 654   case vmIntrinsics::_weakCompareAndSwapObjectRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
 655   case vmIntrinsics::_weakCompareAndSwapObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
 656   case vmIntrinsics::_weakCompareAndSwapByte:           return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Relaxed);
 657   case vmIntrinsics::_weakCompareAndSwapByteAcquire:    return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Acquire);
 658   case vmIntrinsics::_weakCompareAndSwapByteRelease:    return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Release);
 659   case vmIntrinsics::_weakCompareAndSwapByteVolatile:   return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Volatile);
 660   case vmIntrinsics::_weakCompareAndSwapShort:          return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Relaxed);
 661   case vmIntrinsics::_weakCompareAndSwapShortAcquire:   return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Acquire);
 662   case vmIntrinsics::_weakCompareAndSwapShortRelease:   return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Release);
 663   case vmIntrinsics::_weakCompareAndSwapShortVolatile:  return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Volatile);
 664   case vmIntrinsics::_weakCompareAndSwapInt:            return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Relaxed);
 665   case vmIntrinsics::_weakCompareAndSwapIntAcquire:     return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Acquire);
 666   case vmIntrinsics::_weakCompareAndSwapIntRelease:     return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Release);
 667   case vmIntrinsics::_weakCompareAndSwapIntVolatile:    return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Volatile);
 668   case vmIntrinsics::_weakCompareAndSwapLong:           return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Relaxed);
 669   case vmIntrinsics::_weakCompareAndSwapLongAcquire:    return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Acquire);
 670   case vmIntrinsics::_weakCompareAndSwapLongRelease:    return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Release);
 671   case vmIntrinsics::_weakCompareAndSwapLongVolatile:   return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Volatile);
 672 
 673   case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Volatile);
 674   case vmIntrinsics::_compareAndExchangeObjectAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Acquire);
 675   case vmIntrinsics::_compareAndExchangeObjectRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Release);
 676   case vmIntrinsics::_compareAndExchangeByteVolatile:   return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Volatile);
 677   case vmIntrinsics::_compareAndExchangeByteAcquire:    return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Acquire);
 678   case vmIntrinsics::_compareAndExchangeByteRelease:    return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Release);
 679   case vmIntrinsics::_compareAndExchangeShortVolatile:  return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Volatile);
 680   case vmIntrinsics::_compareAndExchangeShortAcquire:   return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Acquire);
 681   case vmIntrinsics::_compareAndExchangeShortRelease:   return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Release);
 682   case vmIntrinsics::_compareAndExchangeIntVolatile:    return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Volatile);
 683   case vmIntrinsics::_compareAndExchangeIntAcquire:     return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Acquire);
 684   case vmIntrinsics::_compareAndExchangeIntRelease:     return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Release);
 685   case vmIntrinsics::_compareAndExchangeLongVolatile:   return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Volatile);
 686   case vmIntrinsics::_compareAndExchangeLongAcquire:    return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Acquire);
 687   case vmIntrinsics::_compareAndExchangeLongRelease:    return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Release);
 688 
 689   case vmIntrinsics::_getAndAddByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_add,       Volatile);
 690   case vmIntrinsics::_getAndAddShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_add,       Volatile);
 691   case vmIntrinsics::_getAndAddInt:                     return inline_unsafe_load_store(T_INT,    LS_get_add,       Volatile);
 692   case vmIntrinsics::_getAndAddLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_add,       Volatile);
 693 
 694   case vmIntrinsics::_getAndSetByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_set,       Volatile);
 695   case vmIntrinsics::_getAndSetShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_set,       Volatile);
 696   case vmIntrinsics::_getAndSetInt:                     return inline_unsafe_load_store(T_INT,    LS_get_set,       Volatile);
 697   case vmIntrinsics::_getAndSetLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_set,       Volatile);
 698   case vmIntrinsics::_getAndSetObject:                  return inline_unsafe_load_store(T_OBJECT, LS_get_set,       Volatile);
 699 
 700   case vmIntrinsics::_loadFence:
 701   case vmIntrinsics::_storeFence:
 702   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 703 
 704   case vmIntrinsics::_onSpinWait:               return inline_onspinwait();
 705 
 706   case vmIntrinsics::_currentThread:            return inline_native_currentThread();
 707   case vmIntrinsics::_isInterrupted:            return inline_native_isInterrupted();
 708 
 709 #ifdef TRACE_HAVE_INTRINSICS
 710   case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
 711 #endif
 712   case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
 713   case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
 714   case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
 715   case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
 716   case vmIntrinsics::_getLength:                return inline_native_getLength();
 717   case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
 718   case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
 719   case vmIntrinsics::_equalsB:                  return inline_array_equals(StrIntrinsicNode::LL);
 720   case vmIntrinsics::_equalsC:                  return inline_array_equals(StrIntrinsicNode::UU);
 721   case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
 722   case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());
 723 
 724   case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
 725   case vmIntrinsics::_newArray:                   return inline_unsafe_newArray(false);
 726 
 727   case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
 728 
 729   case vmIntrinsics::_isInstance:
 730   case vmIntrinsics::_getModifiers:
 731   case vmIntrinsics::_isInterface:
 732   case vmIntrinsics::_isArray:
 733   case vmIntrinsics::_isPrimitive:
 734   case vmIntrinsics::_getSuperclass:
 735   case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
 736 
 737   case vmIntrinsics::_floatToRawIntBits:
 738   case vmIntrinsics::_floatToIntBits:
 739   case vmIntrinsics::_intBitsToFloat:
 740   case vmIntrinsics::_doubleToRawLongBits:
 741   case vmIntrinsics::_doubleToLongBits:
 742   case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
 743 
 744   case vmIntrinsics::_numberOfLeadingZeros_i:
 745   case vmIntrinsics::_numberOfLeadingZeros_l:
 746   case vmIntrinsics::_numberOfTrailingZeros_i:
 747   case vmIntrinsics::_numberOfTrailingZeros_l:
 748   case vmIntrinsics::_bitCount_i:
 749   case vmIntrinsics::_bitCount_l:
 750   case vmIntrinsics::_reverseBytes_i:
 751   case vmIntrinsics::_reverseBytes_l:
 752   case vmIntrinsics::_reverseBytes_s:
 753   case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
 754 
 755   case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
 756 
 757   case vmIntrinsics::_Reference_get:            return inline_reference_get();
 758 
 759   case vmIntrinsics::_Class_cast:               return inline_Class_cast();
 760 
 761   case vmIntrinsics::_aescrypt_encryptBlock:
 762   case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
 763 
 764   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 765   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 766     return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
 767 
 768   case vmIntrinsics::_counterMode_AESCrypt:
 769     return inline_counterMode_AESCrypt(intrinsic_id());
 770 
 771   case vmIntrinsics::_sha_implCompress:
 772   case vmIntrinsics::_sha2_implCompress:
 773   case vmIntrinsics::_sha5_implCompress:
 774     return inline_sha_implCompress(intrinsic_id());
 775 
 776   case vmIntrinsics::_digestBase_implCompressMB:
 777     return inline_digestBase_implCompressMB(predicate);
 778 
 779   case vmIntrinsics::_multiplyToLen:
 780     return inline_multiplyToLen();
 781 
 782   case vmIntrinsics::_squareToLen:
 783     return inline_squareToLen();
 784 
 785   case vmIntrinsics::_mulAdd:
 786     return inline_mulAdd();
 787 
 788   case vmIntrinsics::_montgomeryMultiply:
 789     return inline_montgomeryMultiply();
 790   case vmIntrinsics::_montgomerySquare:
 791     return inline_montgomerySquare();
 792 
 793   case vmIntrinsics::_vectorizedMismatch:
 794     return inline_vectorizedMismatch();
 795 
 796   case vmIntrinsics::_ghash_processBlocks:
 797     return inline_ghash_processBlocks();
 798 
 799   case vmIntrinsics::_encodeISOArray:
 800   case vmIntrinsics::_encodeByteISOArray:
 801     return inline_encodeISOArray();
 802 
 803   case vmIntrinsics::_updateCRC32:
 804     return inline_updateCRC32();
 805   case vmIntrinsics::_updateBytesCRC32:
 806     return inline_updateBytesCRC32();
 807   case vmIntrinsics::_updateByteBufferCRC32:
 808     return inline_updateByteBufferCRC32();
 809 
 810   case vmIntrinsics::_updateBytesCRC32C:
 811     return inline_updateBytesCRC32C();
 812   case vmIntrinsics::_updateDirectByteBufferCRC32C:
 813     return inline_updateDirectByteBufferCRC32C();
 814 
 815   case vmIntrinsics::_updateBytesAdler32:
 816     return inline_updateBytesAdler32();
 817   case vmIntrinsics::_updateByteBufferAdler32:
 818     return inline_updateByteBufferAdler32();
 819 
 820   case vmIntrinsics::_profileBoolean:
 821     return inline_profileBoolean();
 822   case vmIntrinsics::_isCompileConstant:
 823     return inline_isCompileConstant();
 824 
 825   case vmIntrinsics::_hasNegatives:
 826     return inline_hasNegatives();
 827 
 828   default:
 829     // If you get here, it may be that someone has added a new intrinsic
 830     // to the list in vmSymbols.hpp without implementing it here.
 831 #ifndef PRODUCT
 832     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 833       tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
 834                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 835     }
 836 #endif
 837     return false;
 838   }
 839 }
 840 
 841 Node* LibraryCallKit::try_to_predicate(int predicate) {
 842   if (!jvms()->has_method()) {
 843     // Root JVMState has a null method.
 844     assert(map()->memory()->Opcode() == Op_Parm, "");
 845     // Insert the memory aliasing node
 846     set_all_memory(reset_memory());
 847   }
 848   assert(merged_memory(), "");
 849 
 850   switch (intrinsic_id()) {
 851   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 852     return inline_cipherBlockChaining_AESCrypt_predicate(false);
 853   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 854     return inline_cipherBlockChaining_AESCrypt_predicate(true);
 855   case vmIntrinsics::_counterMode_AESCrypt:
 856     return inline_counterMode_AESCrypt_predicate();
 857   case vmIntrinsics::_digestBase_implCompressMB:
 858     return inline_digestBase_implCompressMB_predicate(predicate);
 859 
 860   default:
 861     // If you get here, it may be that someone has added a new intrinsic
 862     // to the list in vmSymbols.hpp without implementing it here.
 863 #ifndef PRODUCT
 864     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 865       tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
 866                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 867     }
 868 #endif
 869     Node* slow_ctl = control();
 870     set_control(top()); // No fast path instrinsic
 871     return slow_ctl;
 872   }
 873 }
 874 
 875 //------------------------------set_result-------------------------------
 876 // Helper function for finishing intrinsics.
 877 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
 878   record_for_igvn(region);
 879   set_control(_gvn.transform(region));
 880   set_result( _gvn.transform(value));
 881   assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
 882 }
 883 
 884 //------------------------------generate_guard---------------------------
 885 // Helper function for generating guarded fast-slow graph structures.
 886 // The given 'test', if true, guards a slow path.  If the test fails
 887 // then a fast path can be taken.  (We generally hope it fails.)
 888 // In all cases, GraphKit::control() is updated to the fast path.
 889 // The returned value represents the control for the slow path.
 890 // The return value is never 'top'; it is either a valid control
 891 // or NULL if it is obvious that the slow path can never be taken.
 892 // Also, if region and the slow control are not NULL, the slow edge
 893 // is appended to the region.
 894 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
 895   if (stopped()) {
 896     // Already short circuited.
 897     return NULL;
 898   }
 899 
 900   // Build an if node and its projections.
 901   // If test is true we take the slow path, which we assume is uncommon.
 902   if (_gvn.type(test) == TypeInt::ZERO) {
 903     // The slow branch is never taken.  No need to build this guard.
 904     return NULL;
 905   }
 906 
 907   IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
 908 
 909   Node* if_slow = _gvn.transform(new IfTrueNode(iff));
 910   if (if_slow == top()) {
 911     // The slow branch is never taken.  No need to build this guard.
 912     return NULL;
 913   }
 914 
 915   if (region != NULL)
 916     region->add_req(if_slow);
 917 
 918   Node* if_fast = _gvn.transform(new IfFalseNode(iff));
 919   set_control(if_fast);
 920 
 921   return if_slow;
 922 }
 923 
 924 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
 925   return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
 926 }
 927 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
 928   return generate_guard(test, region, PROB_FAIR);
 929 }
 930 
 931 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
 932                                                      Node* *pos_index) {
 933   if (stopped())
 934     return NULL;                // already stopped
 935   if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
 936     return NULL;                // index is already adequately typed
 937   Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
 938   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 939   Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
 940   if (is_neg != NULL && pos_index != NULL) {
 941     // Emulate effect of Parse::adjust_map_after_if.
 942     Node* ccast = new CastIINode(index, TypeInt::POS);
 943     ccast->set_req(0, control());
 944     (*pos_index) = _gvn.transform(ccast);
 945   }
 946   return is_neg;
 947 }
 948 
 949 // Make sure that 'position' is a valid limit index, in [0..length].
 950 // There are two equivalent plans for checking this:
 951 //   A. (offset + copyLength)  unsigned<=  arrayLength
 952 //   B. offset  <=  (arrayLength - copyLength)
 953 // We require that all of the values above, except for the sum and
 954 // difference, are already known to be non-negative.
 955 // Plan A is robust in the face of overflow, if offset and copyLength
 956 // are both hugely positive.
 957 //
 958 // Plan B is less direct and intuitive, but it does not overflow at
 959 // all, since the difference of two non-negatives is always
 960 // representable.  Whenever Java methods must perform the equivalent
 961 // check they generally use Plan B instead of Plan A.
 962 // For the moment we use Plan A.
 963 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
 964                                                   Node* subseq_length,
 965                                                   Node* array_length,
 966                                                   RegionNode* region) {
 967   if (stopped())
 968     return NULL;                // already stopped
 969   bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
 970   if (zero_offset && subseq_length->eqv_uncast(array_length))
 971     return NULL;                // common case of whole-array copy
 972   Node* last = subseq_length;
 973   if (!zero_offset)             // last += offset
 974     last = _gvn.transform(new AddINode(last, offset));
 975   Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
 976   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 977   Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
 978   return is_over;
 979 }
 980 
 981 // Emit range checks for the given String.value byte array
 982 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
 983   if (stopped()) {
 984     return; // already stopped
 985   }
 986   RegionNode* bailout = new RegionNode(1);
 987   record_for_igvn(bailout);
 988   if (char_count) {
 989     // Convert char count to byte count
 990     count = _gvn.transform(new LShiftINode(count, intcon(1)));
 991   }
 992 
 993   // Offset and count must not be negative
 994   generate_negative_guard(offset, bailout);
 995   generate_negative_guard(count, bailout);
 996   // Offset + count must not exceed length of array
 997   generate_limit_guard(offset, count, load_array_length(array), bailout);
 998 
 999   if (bailout->req() > 1) {
1000     PreserveJVMState pjvms(this);
1001     set_control(_gvn.transform(bailout));
1002     uncommon_trap(Deoptimization::Reason_intrinsic,
1003                   Deoptimization::Action_maybe_recompile);
1004   }
1005 }
1006 
1007 //--------------------------generate_current_thread--------------------
1008 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1009   ciKlass*    thread_klass = env()->Thread_klass();
1010   const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1011   Node* thread = _gvn.transform(new ThreadLocalNode());
1012   Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1013   Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1014   tls_output = thread;
1015   return threadObj;
1016 }
1017 
1018 
1019 //------------------------------make_string_method_node------------------------
1020 // Helper method for String intrinsic functions. This version is called with
1021 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1022 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1023 // containing the lengths of str1 and str2.
1024 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1025   Node* result = NULL;
1026   switch (opcode) {
1027   case Op_StrIndexOf:
1028     result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1029                                 str1_start, cnt1, str2_start, cnt2, ae);
1030     break;
1031   case Op_StrComp:
1032     result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1033                              str1_start, cnt1, str2_start, cnt2, ae);
1034     break;
1035   case Op_StrEquals:
1036     // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1037     // Use the constant length if there is one because optimized match rule may exist.
1038     result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1039                                str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1040     break;
1041   default:
1042     ShouldNotReachHere();
1043     return NULL;
1044   }
1045 
1046   // All these intrinsics have checks.
1047   C->set_has_split_ifs(true); // Has chance for split-if optimization
1048 
1049   return _gvn.transform(result);
1050 }
1051 
1052 //------------------------------inline_string_compareTo------------------------
1053 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1054   Node* arg1 = argument(0);
1055   Node* arg2 = argument(1);
1056 
1057   // Get start addr and length of first argument
1058   Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1059   Node* arg1_cnt    = load_array_length(arg1);
1060 
1061   // Get start addr and length of second argument
1062   Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1063   Node* arg2_cnt    = load_array_length(arg2);
1064 
1065   Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1066   set_result(result);
1067   return true;
1068 }
1069 
1070 //------------------------------inline_string_equals------------------------
1071 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1072   Node* arg1 = argument(0);
1073   Node* arg2 = argument(1);
1074 
1075   // paths (plus control) merge
1076   RegionNode* region = new RegionNode(3);
1077   Node* phi = new PhiNode(region, TypeInt::BOOL);
1078 
1079   if (!stopped()) {
1080     // Get start addr and length of first argument
1081     Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1082     Node* arg1_cnt    = load_array_length(arg1);
1083 
1084     // Get start addr and length of second argument
1085     Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1086     Node* arg2_cnt    = load_array_length(arg2);
1087 
1088     // Check for arg1_cnt != arg2_cnt
1089     Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1090     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1091     Node* if_ne = generate_slow_guard(bol, NULL);
1092     if (if_ne != NULL) {
1093       phi->init_req(2, intcon(0));
1094       region->init_req(2, if_ne);
1095     }
1096 
1097     // Check for count == 0 is done by assembler code for StrEquals.
1098 
1099     if (!stopped()) {
1100       Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1101       phi->init_req(1, equals);
1102       region->init_req(1, control());
1103     }
1104   }
1105 
1106   // post merge
1107   set_control(_gvn.transform(region));
1108   record_for_igvn(region);
1109 
1110   set_result(_gvn.transform(phi));
1111   return true;
1112 }
1113 
1114 //------------------------------inline_array_equals----------------------------
1115 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1116   assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1117   Node* arg1 = argument(0);
1118   Node* arg2 = argument(1);
1119 
1120   const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1121   set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1122   return true;
1123 }
1124 
1125 //------------------------------inline_hasNegatives------------------------------
1126 bool LibraryCallKit::inline_hasNegatives() {
1127   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1128     return false;
1129   }
1130 
1131   assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1132   // no receiver since it is static method
1133   Node* ba         = argument(0);
1134   Node* offset     = argument(1);
1135   Node* len        = argument(2);
1136 
1137   // Range checks
1138   generate_string_range_check(ba, offset, len, false);
1139   if (stopped()) {
1140     return true;
1141   }
1142   Node* ba_start = array_element_address(ba, offset, T_BYTE);
1143   Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1144   set_result(_gvn.transform(result));
1145   return true;
1146 }
1147 
1148 bool LibraryCallKit::inline_preconditions_checkIndex() {
1149   Node* index = argument(0);
1150   Node* length = argument(1);
1151   if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1152     return false;
1153   }
1154 
1155   Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1156   Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1157 
1158   {
1159     BuildCutout unless(this, len_pos_bol, PROB_MAX);
1160     uncommon_trap(Deoptimization::Reason_intrinsic,
1161                   Deoptimization::Action_make_not_entrant);
1162   }
1163 
1164   if (stopped()) {
1165     return false;
1166   }
1167 
1168   Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1169   BoolTest::mask btest = BoolTest::lt;
1170   Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1171   RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1172   _gvn.set_type(rc, rc->Value(&_gvn));
1173   if (!rc_bool->is_Con()) {
1174     record_for_igvn(rc);
1175   }
1176   set_control(_gvn.transform(new IfTrueNode(rc)));
1177   {
1178     PreserveJVMState pjvms(this);
1179     set_control(_gvn.transform(new IfFalseNode(rc)));
1180     uncommon_trap(Deoptimization::Reason_range_check,
1181                   Deoptimization::Action_make_not_entrant);
1182   }
1183 
1184   if (stopped()) {
1185     return false;
1186   }
1187 
1188   Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1189   result->set_req(0, control());
1190   result = _gvn.transform(result);
1191   set_result(result);
1192   replace_in_map(index, result);
1193   return true;
1194 }
1195 
1196 //------------------------------inline_string_indexOf------------------------
1197 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1198   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1199     return false;
1200   }
1201   Node* src = argument(0);
1202   Node* tgt = argument(1);
1203 
1204   // Make the merge point
1205   RegionNode* result_rgn = new RegionNode(4);
1206   Node*       result_phi = new PhiNode(result_rgn, TypeInt::INT);
1207 
1208   // Get start addr and length of source string
1209   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1210   Node* src_count = load_array_length(src);
1211 
1212   // Get start addr and length of substring
1213   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1214   Node* tgt_count = load_array_length(tgt);
1215 
1216   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1217     // Divide src size by 2 if String is UTF16 encoded
1218     src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1219   }
1220   if (ae == StrIntrinsicNode::UU) {
1221     // Divide substring size by 2 if String is UTF16 encoded
1222     tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1223   }
1224 
1225   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1226   if (result != NULL) {
1227     result_phi->init_req(3, result);
1228     result_rgn->init_req(3, control());
1229   }
1230   set_control(_gvn.transform(result_rgn));
1231   record_for_igvn(result_rgn);
1232   set_result(_gvn.transform(result_phi));
1233 
1234   return true;
1235 }
1236 
1237 //-----------------------------inline_string_indexOf-----------------------
1238 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1239   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1240     return false;
1241   }
1242   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1243     return false;
1244   }
1245   assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1246   Node* src         = argument(0); // byte[]
1247   Node* src_count   = argument(1); // char count
1248   Node* tgt         = argument(2); // byte[]
1249   Node* tgt_count   = argument(3); // char count
1250   Node* from_index  = argument(4); // char index
1251 
1252   // Multiply byte array index by 2 if String is UTF16 encoded
1253   Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1254   src_count = _gvn.transform(new SubINode(src_count, from_index));
1255   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1256   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1257 
1258   // Range checks
1259   generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1260   generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1261   if (stopped()) {
1262     return true;
1263   }
1264 
1265   RegionNode* region = new RegionNode(5);
1266   Node* phi = new PhiNode(region, TypeInt::INT);
1267 
1268   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1269   if (result != NULL) {
1270     // The result is index relative to from_index if substring was found, -1 otherwise.
1271     // Generate code which will fold into cmove.
1272     Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1273     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1274 
1275     Node* if_lt = generate_slow_guard(bol, NULL);
1276     if (if_lt != NULL) {
1277       // result == -1
1278       phi->init_req(3, result);
1279       region->init_req(3, if_lt);
1280     }
1281     if (!stopped()) {
1282       result = _gvn.transform(new AddINode(result, from_index));
1283       phi->init_req(4, result);
1284       region->init_req(4, control());
1285     }
1286   }
1287 
1288   set_control(_gvn.transform(region));
1289   record_for_igvn(region);
1290   set_result(_gvn.transform(phi));
1291 
1292   return true;
1293 }
1294 
1295 // Create StrIndexOfNode with fast path checks
1296 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1297                                         RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1298   // Check for substr count > string count
1299   Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1300   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1301   Node* if_gt = generate_slow_guard(bol, NULL);
1302   if (if_gt != NULL) {
1303     phi->init_req(1, intcon(-1));
1304     region->init_req(1, if_gt);
1305   }
1306   if (!stopped()) {
1307     // Check for substr count == 0
1308     cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1309     bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1310     Node* if_zero = generate_slow_guard(bol, NULL);
1311     if (if_zero != NULL) {
1312       phi->init_req(2, intcon(0));
1313       region->init_req(2, if_zero);
1314     }
1315   }
1316   if (!stopped()) {
1317     return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1318   }
1319   return NULL;
1320 }
1321 
1322 //-----------------------------inline_string_indexOfChar-----------------------
1323 bool LibraryCallKit::inline_string_indexOfChar() {
1324   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1325     return false;
1326   }
1327   if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1328     return false;
1329   }
1330   assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1331   Node* src         = argument(0); // byte[]
1332   Node* tgt         = argument(1); // tgt is int ch
1333   Node* from_index  = argument(2);
1334   Node* max         = argument(3);
1335 
1336   Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1337   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1338   Node* src_count = _gvn.transform(new SubINode(max, from_index));
1339 
1340   // Range checks
1341   generate_string_range_check(src, src_offset, src_count, true);
1342   if (stopped()) {
1343     return true;
1344   }
1345 
1346   RegionNode* region = new RegionNode(3);
1347   Node* phi = new PhiNode(region, TypeInt::INT);
1348 
1349   Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1350   C->set_has_split_ifs(true); // Has chance for split-if optimization
1351   _gvn.transform(result);
1352 
1353   Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1354   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1355 
1356   Node* if_lt = generate_slow_guard(bol, NULL);
1357   if (if_lt != NULL) {
1358     // result == -1
1359     phi->init_req(2, result);
1360     region->init_req(2, if_lt);
1361   }
1362   if (!stopped()) {
1363     result = _gvn.transform(new AddINode(result, from_index));
1364     phi->init_req(1, result);
1365     region->init_req(1, control());
1366   }
1367   set_control(_gvn.transform(region));
1368   record_for_igvn(region);
1369   set_result(_gvn.transform(phi));
1370 
1371   return true;
1372 }
1373 //---------------------------inline_string_copy---------------------
1374 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1375 //   int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1376 //   int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1377 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1378 //   void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1379 //   void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1380 bool LibraryCallKit::inline_string_copy(bool compress) {
1381   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1382     return false;
1383   }
1384   int nargs = 5;  // 2 oops, 3 ints
1385   assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1386 
1387   Node* src         = argument(0);
1388   Node* src_offset  = argument(1);
1389   Node* dst         = argument(2);
1390   Node* dst_offset  = argument(3);
1391   Node* length      = argument(4);
1392 
1393   // Check for allocation before we add nodes that would confuse
1394   // tightly_coupled_allocation()
1395   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1396 
1397   // Figure out the size and type of the elements we will be copying.
1398   const Type* src_type = src->Value(&_gvn);
1399   const Type* dst_type = dst->Value(&_gvn);
1400   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1401   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1402   assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1403          (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1404          "Unsupported array types for inline_string_copy");
1405 
1406   // Range checks
1407   generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1408   generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1409   if (stopped()) {
1410     return true;
1411   }
1412 
1413   // Convert char[] offsets to byte[] offsets
1414   if (compress && src_elem == T_BYTE) {
1415     src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1416   } else if (!compress && dst_elem == T_BYTE) {
1417     dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1418   }
1419 
1420   Node* src_start = array_element_address(src, src_offset, src_elem);
1421   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1422   // 'src_start' points to src array + scaled offset
1423   // 'dst_start' points to dst array + scaled offset
1424   Node* count = NULL;
1425   if (compress) {
1426     count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1427   } else {
1428     inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1429   }
1430 
1431   if (alloc != NULL) {
1432     if (alloc->maybe_set_complete(&_gvn)) {
1433       // "You break it, you buy it."
1434       InitializeNode* init = alloc->initialization();
1435       assert(init->is_complete(), "we just did this");
1436       init->set_complete_with_arraycopy();
1437       assert(dst->is_CheckCastPP(), "sanity");
1438       assert(dst->in(0)->in(0) == init, "dest pinned");
1439     }
1440     // Do not let stores that initialize this object be reordered with
1441     // a subsequent store that would make this object accessible by
1442     // other threads.
1443     // Record what AllocateNode this StoreStore protects so that
1444     // escape analysis can go from the MemBarStoreStoreNode to the
1445     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1446     // based on the escape status of the AllocateNode.
1447     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1448   }
1449   if (compress) {
1450     set_result(_gvn.transform(count));
1451   }
1452   return true;
1453 }
1454 
1455 #ifdef _LP64
1456 #define XTOP ,top() /*additional argument*/
1457 #else  //_LP64
1458 #define XTOP        /*no additional argument*/
1459 #endif //_LP64
1460 
1461 //------------------------inline_string_toBytesU--------------------------
1462 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1463 bool LibraryCallKit::inline_string_toBytesU() {
1464   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1465     return false;
1466   }
1467   // Get the arguments.
1468   Node* value     = argument(0);
1469   Node* offset    = argument(1);
1470   Node* length    = argument(2);
1471 
1472   Node* newcopy = NULL;
1473 
1474   // Set the original stack and the reexecute bit for the interpreter to reexecute
1475   // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1476   { PreserveReexecuteState preexecs(this);
1477     jvms()->set_should_reexecute(true);
1478 
1479     // Check if a null path was taken unconditionally.
1480     value = null_check(value);
1481 
1482     RegionNode* bailout = new RegionNode(1);
1483     record_for_igvn(bailout);
1484 
1485     // Range checks
1486     generate_negative_guard(offset, bailout);
1487     generate_negative_guard(length, bailout);
1488     generate_limit_guard(offset, length, load_array_length(value), bailout);
1489     // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1490     generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1491 
1492     if (bailout->req() > 1) {
1493       PreserveJVMState pjvms(this);
1494       set_control(_gvn.transform(bailout));
1495       uncommon_trap(Deoptimization::Reason_intrinsic,
1496                     Deoptimization::Action_maybe_recompile);
1497     }
1498     if (stopped()) {
1499       return true;
1500     }
1501 
1502     Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1503     Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1504     newcopy = new_array(klass_node, size, 0);  // no arguments to push
1505     AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1506 
1507     // Calculate starting addresses.
1508     Node* src_start = array_element_address(value, offset, T_CHAR);
1509     Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1510 
1511     // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1512     const TypeInt* toffset = gvn().type(offset)->is_int();
1513     bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1514 
1515     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1516     const char* copyfunc_name = "arraycopy";
1517     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1518     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1519                       OptoRuntime::fast_arraycopy_Type(),
1520                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1521                       src_start, dst_start, ConvI2X(length) XTOP);
1522     // Do not let reads from the cloned object float above the arraycopy.
1523     if (alloc != NULL) {
1524       if (alloc->maybe_set_complete(&_gvn)) {
1525         // "You break it, you buy it."
1526         InitializeNode* init = alloc->initialization();
1527         assert(init->is_complete(), "we just did this");
1528         init->set_complete_with_arraycopy();
1529         assert(newcopy->is_CheckCastPP(), "sanity");
1530         assert(newcopy->in(0)->in(0) == init, "dest pinned");
1531       }
1532       // Do not let stores that initialize this object be reordered with
1533       // a subsequent store that would make this object accessible by
1534       // other threads.
1535       // Record what AllocateNode this StoreStore protects so that
1536       // escape analysis can go from the MemBarStoreStoreNode to the
1537       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1538       // based on the escape status of the AllocateNode.
1539       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1540     } else {
1541       insert_mem_bar(Op_MemBarCPUOrder);
1542     }
1543   } // original reexecute is set back here
1544 
1545   C->set_has_split_ifs(true); // Has chance for split-if optimization
1546   if (!stopped()) {
1547     set_result(newcopy);
1548   }
1549   return true;
1550 }
1551 
1552 //------------------------inline_string_getCharsU--------------------------
1553 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1554 bool LibraryCallKit::inline_string_getCharsU() {
1555   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1556     return false;
1557   }
1558 
1559   // Get the arguments.
1560   Node* src       = argument(0);
1561   Node* src_begin = argument(1);
1562   Node* src_end   = argument(2); // exclusive offset (i < src_end)
1563   Node* dst       = argument(3);
1564   Node* dst_begin = argument(4);
1565 
1566   // Check for allocation before we add nodes that would confuse
1567   // tightly_coupled_allocation()
1568   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1569 
1570   // Check if a null path was taken unconditionally.
1571   src = null_check(src);
1572   dst = null_check(dst);
1573   if (stopped()) {
1574     return true;
1575   }
1576 
1577   // Get length and convert char[] offset to byte[] offset
1578   Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1579   src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1580 
1581   // Range checks
1582   generate_string_range_check(src, src_begin, length, true);
1583   generate_string_range_check(dst, dst_begin, length, false);
1584   if (stopped()) {
1585     return true;
1586   }
1587 
1588   if (!stopped()) {
1589     // Calculate starting addresses.
1590     Node* src_start = array_element_address(src, src_begin, T_BYTE);
1591     Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1592 
1593     // Check if array addresses are aligned to HeapWordSize
1594     const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1595     const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1596     bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1597                    tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1598 
1599     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1600     const char* copyfunc_name = "arraycopy";
1601     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1602     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1603                       OptoRuntime::fast_arraycopy_Type(),
1604                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1605                       src_start, dst_start, ConvI2X(length) XTOP);
1606     // Do not let reads from the cloned object float above the arraycopy.
1607     if (alloc != NULL) {
1608       if (alloc->maybe_set_complete(&_gvn)) {
1609         // "You break it, you buy it."
1610         InitializeNode* init = alloc->initialization();
1611         assert(init->is_complete(), "we just did this");
1612         init->set_complete_with_arraycopy();
1613         assert(dst->is_CheckCastPP(), "sanity");
1614         assert(dst->in(0)->in(0) == init, "dest pinned");
1615       }
1616       // Do not let stores that initialize this object be reordered with
1617       // a subsequent store that would make this object accessible by
1618       // other threads.
1619       // Record what AllocateNode this StoreStore protects so that
1620       // escape analysis can go from the MemBarStoreStoreNode to the
1621       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1622       // based on the escape status of the AllocateNode.
1623       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1624     } else {
1625       insert_mem_bar(Op_MemBarCPUOrder);
1626     }
1627   }
1628 
1629   C->set_has_split_ifs(true); // Has chance for split-if optimization
1630   return true;
1631 }
1632 
1633 //----------------------inline_string_char_access----------------------------
1634 // Store/Load char to/from byte[] array.
1635 // static void StringUTF16.putChar(byte[] val, int index, int c)
1636 // static char StringUTF16.getChar(byte[] val, int index)
1637 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1638   Node* value  = argument(0);
1639   Node* index  = argument(1);
1640   Node* ch = is_store ? argument(2) : NULL;
1641 
1642   // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1643   // correctly requires matched array shapes.
1644   assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1645           "sanity: byte[] and char[] bases agree");
1646   assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1647           "sanity: byte[] and char[] scales agree");
1648 
1649   // Bail when getChar over constants is requested: constant folding would
1650   // reject folding mismatched char access over byte[]. A normal inlining for getChar
1651   // Java method would constant fold nicely instead.
1652   if (!is_store && value->is_Con() && index->is_Con()) {
1653     return false;
1654   }
1655 
1656   Node* adr = array_element_address(value, index, T_CHAR);
1657   if (is_store) {
1658     (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1659                            false, false, true /* mismatched */);
1660   } else {
1661     ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1662                    LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1663     set_result(ch);
1664   }
1665   return true;
1666 }
1667 
1668 //--------------------------round_double_node--------------------------------
1669 // Round a double node if necessary.
1670 Node* LibraryCallKit::round_double_node(Node* n) {
1671   if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1672     n = _gvn.transform(new RoundDoubleNode(0, n));
1673   return n;
1674 }
1675 
1676 //------------------------------inline_math-----------------------------------
1677 // public static double Math.abs(double)
1678 // public static double Math.sqrt(double)
1679 // public static double Math.log(double)
1680 // public static double Math.log10(double)
1681 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1682   Node* arg = round_double_node(argument(0));
1683   Node* n = NULL;
1684   switch (id) {
1685   case vmIntrinsics::_dabs:   n = new AbsDNode(                arg);  break;
1686   case vmIntrinsics::_dsqrt:  n = new SqrtDNode(C, control(),  arg);  break;
1687   default:  fatal_unexpected_iid(id);  break;
1688   }
1689   set_result(_gvn.transform(n));
























































































1690   return true;
1691 }
1692 
1693 //------------------------------runtime_math-----------------------------
1694 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1695   assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1696          "must be (DD)D or (D)D type");
1697 
1698   // Inputs
1699   Node* a = round_double_node(argument(0));
1700   Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1701 
1702   const TypePtr* no_memory_effects = NULL;
1703   Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1704                                  no_memory_effects,
1705                                  a, top(), b, b ? top() : NULL);
1706   Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1707 #ifdef ASSERT
1708   Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1709   assert(value_top == top(), "second value must be top");
1710 #endif
1711 
1712   set_result(value);
1713   return true;
1714 }
1715 
1716 //------------------------------inline_math_native-----------------------------
1717 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1718 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1719   switch (id) {
1720     // These intrinsics are not properly supported on all hardware
1721   case vmIntrinsics::_dsin:
1722     return StubRoutines::dsin() != NULL ?
1723       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1724       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
1725   case vmIntrinsics::_dcos:
1726     return StubRoutines::dcos() != NULL ?
1727       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1728       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
1729   case vmIntrinsics::_dtan:
1730     return StubRoutines::dtan() != NULL ?
1731       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1732       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1733   case vmIntrinsics::_dlog:
1734     return StubRoutines::dlog() != NULL ?
1735       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1736       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
1737   case vmIntrinsics::_dlog10:
1738     return StubRoutines::dlog10() != NULL ?
1739       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1740       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1741 
1742     // These intrinsics are supported on all hardware
1743   case vmIntrinsics::_dsqrt:  return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1744   case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_math(id) : false;
1745 
1746   case vmIntrinsics::_dexp:
1747     return StubRoutines::dexp() != NULL ?
1748       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(),  "dexp") :
1749       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp),  "EXP");
1750   case vmIntrinsics::_dpow:
1751     return StubRoutines::dpow() != NULL ?
1752       runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1753       runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow),  "POW");
1754 #undef FN_PTR
1755 
1756    // These intrinsics are not yet correctly implemented
1757   case vmIntrinsics::_datan2:
1758     return false;
1759 
1760   default:
1761     fatal_unexpected_iid(id);
1762     return false;
1763   }
1764 }
1765 
1766 static bool is_simple_name(Node* n) {
1767   return (n->req() == 1         // constant
1768           || (n->is_Type() && n->as_Type()->type()->singleton())
1769           || n->is_Proj()       // parameter or return value
1770           || n->is_Phi()        // local of some sort
1771           );
1772 }
1773 
1774 //----------------------------inline_notify-----------------------------------*
1775 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1776   const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1777   address func;
1778   if (id == vmIntrinsics::_notify) {
1779     func = OptoRuntime::monitor_notify_Java();
1780   } else {
1781     func = OptoRuntime::monitor_notifyAll_Java();
1782   }
1783   Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1784   make_slow_call_ex(call, env()->Throwable_klass(), false);
1785   return true;
1786 }
1787 
1788 
1789 //----------------------------inline_min_max-----------------------------------
1790 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1791   set_result(generate_min_max(id, argument(0), argument(1)));
1792   return true;
1793 }
1794 
1795 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1796   Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1797   IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1798   Node* fast_path = _gvn.transform( new IfFalseNode(check));
1799   Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1800 
1801   {
1802     PreserveJVMState pjvms(this);
1803     PreserveReexecuteState preexecs(this);
1804     jvms()->set_should_reexecute(true);
1805 
1806     set_control(slow_path);
1807     set_i_o(i_o());
1808 
1809     uncommon_trap(Deoptimization::Reason_intrinsic,
1810                   Deoptimization::Action_none);
1811   }
1812 
1813   set_control(fast_path);
1814   set_result(math);
1815 }
1816 
1817 template <typename OverflowOp>
1818 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1819   typedef typename OverflowOp::MathOp MathOp;
1820 
1821   MathOp* mathOp = new MathOp(arg1, arg2);
1822   Node* operation = _gvn.transform( mathOp );
1823   Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1824   inline_math_mathExact(operation, ofcheck);
1825   return true;
1826 }
1827 
1828 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1829   return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1830 }
1831 
1832 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1833   return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1834 }
1835 
1836 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1837   return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1838 }
1839 
1840 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1841   return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1842 }
1843 
1844 bool LibraryCallKit::inline_math_negateExactI() {
1845   return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1846 }
1847 
1848 bool LibraryCallKit::inline_math_negateExactL() {
1849   return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1850 }
1851 
1852 bool LibraryCallKit::inline_math_multiplyExactI() {
1853   return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1854 }
1855 
1856 bool LibraryCallKit::inline_math_multiplyExactL() {
1857   return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1858 }
1859 
1860 Node*
1861 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1862   // These are the candidate return value:
1863   Node* xvalue = x0;
1864   Node* yvalue = y0;
1865 
1866   if (xvalue == yvalue) {
1867     return xvalue;
1868   }
1869 
1870   bool want_max = (id == vmIntrinsics::_max);
1871 
1872   const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1873   const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1874   if (txvalue == NULL || tyvalue == NULL)  return top();
1875   // This is not really necessary, but it is consistent with a
1876   // hypothetical MaxINode::Value method:
1877   int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1878 
1879   // %%% This folding logic should (ideally) be in a different place.
1880   // Some should be inside IfNode, and there to be a more reliable
1881   // transformation of ?: style patterns into cmoves.  We also want
1882   // more powerful optimizations around cmove and min/max.
1883 
1884   // Try to find a dominating comparison of these guys.
1885   // It can simplify the index computation for Arrays.copyOf
1886   // and similar uses of System.arraycopy.
1887   // First, compute the normalized version of CmpI(x, y).
1888   int   cmp_op = Op_CmpI;
1889   Node* xkey = xvalue;
1890   Node* ykey = yvalue;
1891   Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1892   if (ideal_cmpxy->is_Cmp()) {
1893     // E.g., if we have CmpI(length - offset, count),
1894     // it might idealize to CmpI(length, count + offset)
1895     cmp_op = ideal_cmpxy->Opcode();
1896     xkey = ideal_cmpxy->in(1);
1897     ykey = ideal_cmpxy->in(2);
1898   }
1899 
1900   // Start by locating any relevant comparisons.
1901   Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1902   Node* cmpxy = NULL;
1903   Node* cmpyx = NULL;
1904   for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1905     Node* cmp = start_from->fast_out(k);
1906     if (cmp->outcnt() > 0 &&            // must have prior uses
1907         cmp->in(0) == NULL &&           // must be context-independent
1908         cmp->Opcode() == cmp_op) {      // right kind of compare
1909       if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
1910       if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
1911     }
1912   }
1913 
1914   const int NCMPS = 2;
1915   Node* cmps[NCMPS] = { cmpxy, cmpyx };
1916   int cmpn;
1917   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1918     if (cmps[cmpn] != NULL)  break;     // find a result
1919   }
1920   if (cmpn < NCMPS) {
1921     // Look for a dominating test that tells us the min and max.
1922     int depth = 0;                // Limit search depth for speed
1923     Node* dom = control();
1924     for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1925       if (++depth >= 100)  break;
1926       Node* ifproj = dom;
1927       if (!ifproj->is_Proj())  continue;
1928       Node* iff = ifproj->in(0);
1929       if (!iff->is_If())  continue;
1930       Node* bol = iff->in(1);
1931       if (!bol->is_Bool())  continue;
1932       Node* cmp = bol->in(1);
1933       if (cmp == NULL)  continue;
1934       for (cmpn = 0; cmpn < NCMPS; cmpn++)
1935         if (cmps[cmpn] == cmp)  break;
1936       if (cmpn == NCMPS)  continue;
1937       BoolTest::mask btest = bol->as_Bool()->_test._test;
1938       if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
1939       if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
1940       // At this point, we know that 'x btest y' is true.
1941       switch (btest) {
1942       case BoolTest::eq:
1943         // They are proven equal, so we can collapse the min/max.
1944         // Either value is the answer.  Choose the simpler.
1945         if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1946           return yvalue;
1947         return xvalue;
1948       case BoolTest::lt:          // x < y
1949       case BoolTest::le:          // x <= y
1950         return (want_max ? yvalue : xvalue);
1951       case BoolTest::gt:          // x > y
1952       case BoolTest::ge:          // x >= y
1953         return (want_max ? xvalue : yvalue);
1954       }
1955     }
1956   }
1957 
1958   // We failed to find a dominating test.
1959   // Let's pick a test that might GVN with prior tests.
1960   Node*          best_bol   = NULL;
1961   BoolTest::mask best_btest = BoolTest::illegal;
1962   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1963     Node* cmp = cmps[cmpn];
1964     if (cmp == NULL)  continue;
1965     for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1966       Node* bol = cmp->fast_out(j);
1967       if (!bol->is_Bool())  continue;
1968       BoolTest::mask btest = bol->as_Bool()->_test._test;
1969       if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
1970       if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
1971       if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1972         best_bol   = bol->as_Bool();
1973         best_btest = btest;
1974       }
1975     }
1976   }
1977 
1978   Node* answer_if_true  = NULL;
1979   Node* answer_if_false = NULL;
1980   switch (best_btest) {
1981   default:
1982     if (cmpxy == NULL)
1983       cmpxy = ideal_cmpxy;
1984     best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
1985     // and fall through:
1986   case BoolTest::lt:          // x < y
1987   case BoolTest::le:          // x <= y
1988     answer_if_true  = (want_max ? yvalue : xvalue);
1989     answer_if_false = (want_max ? xvalue : yvalue);
1990     break;
1991   case BoolTest::gt:          // x > y
1992   case BoolTest::ge:          // x >= y
1993     answer_if_true  = (want_max ? xvalue : yvalue);
1994     answer_if_false = (want_max ? yvalue : xvalue);
1995     break;
1996   }
1997 
1998   jint hi, lo;
1999   if (want_max) {
2000     // We can sharpen the minimum.
2001     hi = MAX2(txvalue->_hi, tyvalue->_hi);
2002     lo = MAX2(txvalue->_lo, tyvalue->_lo);
2003   } else {
2004     // We can sharpen the maximum.
2005     hi = MIN2(txvalue->_hi, tyvalue->_hi);
2006     lo = MIN2(txvalue->_lo, tyvalue->_lo);
2007   }
2008 
2009   // Use a flow-free graph structure, to avoid creating excess control edges
2010   // which could hinder other optimizations.
2011   // Since Math.min/max is often used with arraycopy, we want
2012   // tightly_coupled_allocation to be able to see beyond min/max expressions.
2013   Node* cmov = CMoveNode::make(NULL, best_bol,
2014                                answer_if_false, answer_if_true,
2015                                TypeInt::make(lo, hi, widen));
2016 
2017   return _gvn.transform(cmov);
2018 
2019   /*
2020   // This is not as desirable as it may seem, since Min and Max
2021   // nodes do not have a full set of optimizations.
2022   // And they would interfere, anyway, with 'if' optimizations
2023   // and with CMoveI canonical forms.
2024   switch (id) {
2025   case vmIntrinsics::_min:
2026     result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2027   case vmIntrinsics::_max:
2028     result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2029   default:
2030     ShouldNotReachHere();
2031   }
2032   */
2033 }
2034 
2035 inline int
2036 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2037   const TypePtr* base_type = TypePtr::NULL_PTR;
2038   if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
2039   if (base_type == NULL) {
2040     // Unknown type.
2041     return Type::AnyPtr;
2042   } else if (base_type == TypePtr::NULL_PTR) {
2043     // Since this is a NULL+long form, we have to switch to a rawptr.
2044     base   = _gvn.transform(new CastX2PNode(offset));
2045     offset = MakeConX(0);
2046     return Type::RawPtr;
2047   } else if (base_type->base() == Type::RawPtr) {
2048     return Type::RawPtr;
2049   } else if (base_type->isa_oopptr()) {
2050     // Base is never null => always a heap address.
2051     if (base_type->ptr() == TypePtr::NotNull) {
2052       return Type::OopPtr;
2053     }
2054     // Offset is small => always a heap address.
2055     const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2056     if (offset_type != NULL &&
2057         base_type->offset() == 0 &&     // (should always be?)
2058         offset_type->_lo >= 0 &&
2059         !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2060       return Type::OopPtr;
2061     }
2062     // Otherwise, it might either be oop+off or NULL+addr.
2063     return Type::AnyPtr;
2064   } else {
2065     // No information:
2066     return Type::AnyPtr;
2067   }
2068 }
2069 
2070 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2071   int kind = classify_unsafe_addr(base, offset);
2072   if (kind == Type::RawPtr) {
2073     return basic_plus_adr(top(), base, offset);
2074   } else {
2075     return basic_plus_adr(base, offset);
2076   }
2077 }
2078 
2079 //--------------------------inline_number_methods-----------------------------
2080 // inline int     Integer.numberOfLeadingZeros(int)
2081 // inline int        Long.numberOfLeadingZeros(long)
2082 //
2083 // inline int     Integer.numberOfTrailingZeros(int)
2084 // inline int        Long.numberOfTrailingZeros(long)
2085 //
2086 // inline int     Integer.bitCount(int)
2087 // inline int        Long.bitCount(long)
2088 //
2089 // inline char  Character.reverseBytes(char)
2090 // inline short     Short.reverseBytes(short)
2091 // inline int     Integer.reverseBytes(int)
2092 // inline long       Long.reverseBytes(long)
2093 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2094   Node* arg = argument(0);
2095   Node* n = NULL;
2096   switch (id) {
2097   case vmIntrinsics::_numberOfLeadingZeros_i:   n = new CountLeadingZerosINode( arg);  break;
2098   case vmIntrinsics::_numberOfLeadingZeros_l:   n = new CountLeadingZerosLNode( arg);  break;
2099   case vmIntrinsics::_numberOfTrailingZeros_i:  n = new CountTrailingZerosINode(arg);  break;
2100   case vmIntrinsics::_numberOfTrailingZeros_l:  n = new CountTrailingZerosLNode(arg);  break;
2101   case vmIntrinsics::_bitCount_i:               n = new PopCountINode(          arg);  break;
2102   case vmIntrinsics::_bitCount_l:               n = new PopCountLNode(          arg);  break;
2103   case vmIntrinsics::_reverseBytes_c:           n = new ReverseBytesUSNode(0,   arg);  break;
2104   case vmIntrinsics::_reverseBytes_s:           n = new ReverseBytesSNode( 0,   arg);  break;
2105   case vmIntrinsics::_reverseBytes_i:           n = new ReverseBytesINode( 0,   arg);  break;
2106   case vmIntrinsics::_reverseBytes_l:           n = new ReverseBytesLNode( 0,   arg);  break;
2107   default:  fatal_unexpected_iid(id);  break;
2108   }
2109   set_result(_gvn.transform(n));
2110   return true;
2111 }
2112 
2113 //----------------------------inline_unsafe_access----------------------------
2114 
2115 // Helper that guards and inserts a pre-barrier.
2116 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2117                                         Node* pre_val, bool need_mem_bar) {
2118   // We could be accessing the referent field of a reference object. If so, when G1
2119   // is enabled, we need to log the value in the referent field in an SATB buffer.
2120   // This routine performs some compile time filters and generates suitable
2121   // runtime filters that guard the pre-barrier code.
2122   // Also add memory barrier for non volatile load from the referent field
2123   // to prevent commoning of loads across safepoint.
2124   if (!UseG1GC && !need_mem_bar)
2125     return;
2126 
2127   // Some compile time checks.
2128 
2129   // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2130   const TypeX* otype = offset->find_intptr_t_type();
2131   if (otype != NULL && otype->is_con() &&
2132       otype->get_con() != java_lang_ref_Reference::referent_offset) {
2133     // Constant offset but not the reference_offset so just return
2134     return;
2135   }
2136 
2137   // We only need to generate the runtime guards for instances.
2138   const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2139   if (btype != NULL) {
2140     if (btype->isa_aryptr()) {
2141       // Array type so nothing to do
2142       return;
2143     }
2144 
2145     const TypeInstPtr* itype = btype->isa_instptr();
2146     if (itype != NULL) {
2147       // Can the klass of base_oop be statically determined to be
2148       // _not_ a sub-class of Reference and _not_ Object?
2149       ciKlass* klass = itype->klass();
2150       if ( klass->is_loaded() &&
2151           !klass->is_subtype_of(env()->Reference_klass()) &&
2152           !env()->Object_klass()->is_subtype_of(klass)) {
2153         return;
2154       }
2155     }
2156   }
2157 
2158   // The compile time filters did not reject base_oop/offset so
2159   // we need to generate the following runtime filters
2160   //
2161   // if (offset == java_lang_ref_Reference::_reference_offset) {
2162   //   if (instance_of(base, java.lang.ref.Reference)) {
2163   //     pre_barrier(_, pre_val, ...);
2164   //   }
2165   // }
2166 
2167   float likely   = PROB_LIKELY(  0.999);
2168   float unlikely = PROB_UNLIKELY(0.999);
2169 
2170   IdealKit ideal(this);
2171 #define __ ideal.
2172 
2173   Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2174 
2175   __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2176       // Update graphKit memory and control from IdealKit.
2177       sync_kit(ideal);
2178 
2179       Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2180       Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2181 
2182       // Update IdealKit memory and control from graphKit.
2183       __ sync_kit(this);
2184 
2185       Node* one = __ ConI(1);
2186       // is_instof == 0 if base_oop == NULL
2187       __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2188 
2189         // Update graphKit from IdeakKit.
2190         sync_kit(ideal);
2191 
2192         // Use the pre-barrier to record the value in the referent field
2193         pre_barrier(false /* do_load */,
2194                     __ ctrl(),
2195                     NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2196                     pre_val /* pre_val */,
2197                     T_OBJECT);
2198         if (need_mem_bar) {
2199           // Add memory barrier to prevent commoning reads from this field
2200           // across safepoint since GC can change its value.
2201           insert_mem_bar(Op_MemBarCPUOrder);
2202         }
2203         // Update IdealKit from graphKit.
2204         __ sync_kit(this);
2205 
2206       } __ end_if(); // _ref_type != ref_none
2207   } __ end_if(); // offset == referent_offset
2208 
2209   // Final sync IdealKit and GraphKit.
2210   final_sync(ideal);
2211 #undef __
2212 }
2213 
2214 
2215 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2216   // Attempt to infer a sharper value type from the offset and base type.
2217   ciKlass* sharpened_klass = NULL;
2218 
2219   // See if it is an instance field, with an object type.
2220   if (alias_type->field() != NULL) {
2221     if (alias_type->field()->type()->is_klass()) {
2222       sharpened_klass = alias_type->field()->type()->as_klass();
2223     }
2224   }
2225 
2226   // See if it is a narrow oop array.
2227   if (adr_type->isa_aryptr()) {
2228     if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2229       const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2230       if (elem_type != NULL) {
2231         sharpened_klass = elem_type->klass();
2232       }
2233     }
2234   }
2235 
2236   // The sharpened class might be unloaded if there is no class loader
2237   // contraint in place.
2238   if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2239     const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2240 
2241 #ifndef PRODUCT
2242     if (C->print_intrinsics() || C->print_inlining()) {
2243       tty->print("  from base type: ");  adr_type->dump();
2244       tty->print("  sharpened value: ");  tjp->dump();
2245     }
2246 #endif
2247     // Sharpen the value type.
2248     return tjp;
2249   }
2250   return NULL;
2251 }
2252 
2253 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2254   if (callee()->is_static())  return false;  // caller must have the capability!
2255   guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2256   guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2257   assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2258 
2259 #ifndef PRODUCT
2260   {
2261     ResourceMark rm;
2262     // Check the signatures.
2263     ciSignature* sig = callee()->signature();
2264 #ifdef ASSERT
2265     if (!is_store) {
2266       // Object getObject(Object base, int/long offset), etc.
2267       BasicType rtype = sig->return_type()->basic_type();
2268       assert(rtype == type, "getter must return the expected value");
2269       assert(sig->count() == 2, "oop getter has 2 arguments");
2270       assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2271       assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2272     } else {
2273       // void putObject(Object base, int/long offset, Object x), etc.
2274       assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2275       assert(sig->count() == 3, "oop putter has 3 arguments");
2276       assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2277       assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2278       BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2279       assert(vtype == type, "putter must accept the expected value");
2280     }
2281 #endif // ASSERT
2282  }
2283 #endif //PRODUCT
2284 
2285   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2286 
2287   Node* receiver = argument(0);  // type: oop
2288 
2289   // Build address expression.
2290   Node* adr;
2291   Node* heap_base_oop = top();
2292   Node* offset = top();
2293   Node* val;
2294 
2295   // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2296   Node* base = argument(1);  // type: oop
2297   // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2298   offset = argument(2);  // type: long
2299   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2300   // to be plain byte offsets, which are also the same as those accepted
2301   // by oopDesc::field_base.
2302   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2303          "fieldOffset must be byte-scaled");
2304   // 32-bit machines ignore the high half!
2305   offset = ConvL2X(offset);
2306   adr = make_unsafe_address(base, offset);
2307   if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2308     heap_base_oop = base;
2309   }
2310   val = is_store ? argument(4) : NULL;
2311 
2312   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2313 
2314   // Try to categorize the address.  If it comes up as TypeJavaPtr::BOTTOM,
2315   // there was not enough information to nail it down.
2316   Compile::AliasType* alias_type = C->alias_type(adr_type);
2317   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2318 
2319   assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2320          alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2321   bool mismatched = false;
2322   BasicType bt = alias_type->basic_type();
2323   if (bt != T_ILLEGAL) {
2324     if (bt == T_BYTE && adr_type->isa_aryptr()) {
2325       // Alias type doesn't differentiate between byte[] and boolean[]).
2326       // Use address type to get the element type.
2327       bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2328     }
2329     if (bt == T_ARRAY || bt == T_NARROWOOP) {
2330       // accessing an array field with getObject is not a mismatch
2331       bt = T_OBJECT;
2332     }
2333     if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2334       // Don't intrinsify mismatched object accesses
2335       return false;
2336     }
2337     mismatched = (bt != type);
2338   }
2339 
2340   // First guess at the value type.
2341   const Type *value_type = Type::get_const_basic_type(type);
2342 
2343   // We will need memory barriers unless we can determine a unique
2344   // alias category for this reference.  (Note:  If for some reason
2345   // the barriers get omitted and the unsafe reference begins to "pollute"
2346   // the alias analysis of the rest of the graph, either Compile::can_alias
2347   // or Compile::must_alias will throw a diagnostic assert.)
2348   bool need_mem_bar;
2349   switch (kind) {
2350       case Relaxed:
2351           need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2352           break;
2353       case Opaque:
2354           // Opaque uses CPUOrder membars for protection against code movement.
2355       case Acquire:
2356       case Release:
2357       case Volatile:
2358           need_mem_bar = true;
2359           break;
2360       default:
2361           ShouldNotReachHere();
2362   }
2363 
2364   // Some accesses require access atomicity for all types, notably longs and doubles.
2365   // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2366   bool requires_atomic_access = false;
2367   switch (kind) {
2368       case Relaxed:
2369           requires_atomic_access = AlwaysAtomicAccesses;
2370           break;
2371       case Opaque:
2372           // Opaque accesses are atomic.
2373       case Acquire:
2374       case Release:
2375       case Volatile:
2376           requires_atomic_access = true;
2377           break;
2378       default:
2379           ShouldNotReachHere();
2380   }
2381 
2382   // Figure out the memory ordering.
2383   // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2384   MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2385 
2386   // If we are reading the value of the referent field of a Reference
2387   // object (either by using Unsafe directly or through reflection)
2388   // then, if G1 is enabled, we need to record the referent in an
2389   // SATB log buffer using the pre-barrier mechanism.
2390   // Also we need to add memory barrier to prevent commoning reads
2391   // from this field across safepoint since GC can change its value.
2392   bool need_read_barrier = !is_store &&
2393                            offset != top() && heap_base_oop != top();
2394 
2395   if (!is_store && type == T_OBJECT) {
2396     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2397     if (tjp != NULL) {
2398       value_type = tjp;
2399     }
2400   }
2401 
2402   receiver = null_check(receiver);
2403   if (stopped()) {
2404     return true;
2405   }
2406   // Heap pointers get a null-check from the interpreter,
2407   // as a courtesy.  However, this is not guaranteed by Unsafe,
2408   // and it is not possible to fully distinguish unintended nulls
2409   // from intended ones in this API.
2410 
2411   // We need to emit leading and trailing CPU membars (see below) in
2412   // addition to memory membars for special access modes. This is a little
2413   // too strong, but avoids the need to insert per-alias-type
2414   // volatile membars (for stores; compare Parse::do_put_xxx), which
2415   // we cannot do effectively here because we probably only have a
2416   // rough approximation of type.
2417 
2418   switch(kind) {
2419     case Relaxed:
2420     case Opaque:
2421     case Acquire:
2422       break;
2423     case Release:
2424     case Volatile:
2425       if (is_store) {
2426         insert_mem_bar(Op_MemBarRelease);
2427       } else {
2428         if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2429           insert_mem_bar(Op_MemBarVolatile);
2430         }
2431       }
2432       break;
2433     default:
2434       ShouldNotReachHere();
2435   }
2436 
2437   // Memory barrier to prevent normal and 'unsafe' accesses from
2438   // bypassing each other.  Happens after null checks, so the
2439   // exception paths do not take memory state from the memory barrier,
2440   // so there's no problems making a strong assert about mixing users
2441   // of safe & unsafe memory.
2442   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2443 
2444   if (!is_store) {
2445     Node* p = NULL;
2446     // Try to constant fold a load from a constant field
2447     ciField* field = alias_type->field();
2448     if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2449       // final or stable field
2450       p = make_constant_from_field(field, heap_base_oop);
2451     }
2452     if (p == NULL) {
2453       // To be valid, unsafe loads may depend on other conditions than
2454       // the one that guards them: pin the Load node
2455       p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2456       // load value
2457       switch (type) {
2458       case T_BOOLEAN:
2459       case T_CHAR:
2460       case T_BYTE:
2461       case T_SHORT:
2462       case T_INT:
2463       case T_LONG:
2464       case T_FLOAT:
2465       case T_DOUBLE:
2466         break;
2467       case T_OBJECT:
2468         if (need_read_barrier) {
2469           // We do not require a mem bar inside pre_barrier if need_mem_bar
2470           // is set: the barriers would be emitted by us.
2471           insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2472         }
2473         break;
2474       case T_ADDRESS:
2475         // Cast to an int type.
2476         p = _gvn.transform(new CastP2XNode(NULL, p));
2477         p = ConvX2UL(p);
2478         break;
2479       default:
2480         fatal("unexpected type %d: %s", type, type2name(type));
2481         break;
2482       }
2483     }
2484     // The load node has the control of the preceding MemBarCPUOrder.  All
2485     // following nodes will have the control of the MemBarCPUOrder inserted at
2486     // the end of this method.  So, pushing the load onto the stack at a later
2487     // point is fine.
2488     set_result(p);
2489   } else {
2490     // place effect of store into memory
2491     switch (type) {
2492     case T_DOUBLE:
2493       val = dstore_rounding(val);
2494       break;
2495     case T_ADDRESS:
2496       // Repackage the long as a pointer.
2497       val = ConvL2X(val);
2498       val = _gvn.transform(new CastX2PNode(val));
2499       break;
2500     }
2501 
2502     if (type != T_OBJECT) {
2503       (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2504     } else {
2505       // Possibly an oop being stored to Java heap or native memory
2506       if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2507         // oop to Java heap.
2508         (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2509       } else {
2510         // We can't tell at compile time if we are storing in the Java heap or outside
2511         // of it. So we need to emit code to conditionally do the proper type of
2512         // store.
2513 
2514         IdealKit ideal(this);
2515 #define __ ideal.
2516         // QQQ who knows what probability is here??
2517         __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2518           // Sync IdealKit and graphKit.
2519           sync_kit(ideal);
2520           Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2521           // Update IdealKit memory.
2522           __ sync_kit(this);
2523         } __ else_(); {
2524           __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2525         } __ end_if();
2526         // Final sync IdealKit and GraphKit.
2527         final_sync(ideal);
2528 #undef __
2529       }
2530     }
2531   }
2532 
2533   switch(kind) {
2534     case Relaxed:
2535     case Opaque:
2536     case Release:
2537       break;
2538     case Acquire:
2539     case Volatile:
2540       if (!is_store) {
2541         insert_mem_bar(Op_MemBarAcquire);
2542       } else {
2543         if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2544           insert_mem_bar(Op_MemBarVolatile);
2545         }
2546       }
2547       break;
2548     default:
2549       ShouldNotReachHere();
2550   }
2551 
2552   if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2553 
2554   return true;
2555 }
2556 
2557 //----------------------------inline_unsafe_load_store----------------------------
2558 // This method serves a couple of different customers (depending on LoadStoreKind):
2559 //
2560 // LS_cmp_swap:
2561 //
2562 //   boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2563 //   boolean compareAndSwapInt(   Object o, long offset, int    expected, int    x);
2564 //   boolean compareAndSwapLong(  Object o, long offset, long   expected, long   x);
2565 //
2566 // LS_cmp_swap_weak:
2567 //
2568 //   boolean weakCompareAndSwapObject(       Object o, long offset, Object expected, Object x);
2569 //   boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2570 //   boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2571 //
2572 //   boolean weakCompareAndSwapInt(          Object o, long offset, int    expected, int    x);
2573 //   boolean weakCompareAndSwapIntAcquire(   Object o, long offset, int    expected, int    x);
2574 //   boolean weakCompareAndSwapIntRelease(   Object o, long offset, int    expected, int    x);
2575 //
2576 //   boolean weakCompareAndSwapLong(         Object o, long offset, long   expected, long   x);
2577 //   boolean weakCompareAndSwapLongAcquire(  Object o, long offset, long   expected, long   x);
2578 //   boolean weakCompareAndSwapLongRelease(  Object o, long offset, long   expected, long   x);
2579 //
2580 // LS_cmp_exchange:
2581 //
2582 //   Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2583 //   Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2584 //   Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2585 //
2586 //   Object compareAndExchangeIntVolatile(   Object o, long offset, Object expected, Object x);
2587 //   Object compareAndExchangeIntAcquire(    Object o, long offset, Object expected, Object x);
2588 //   Object compareAndExchangeIntRelease(    Object o, long offset, Object expected, Object x);
2589 //
2590 //   Object compareAndExchangeLongVolatile(  Object o, long offset, Object expected, Object x);
2591 //   Object compareAndExchangeLongAcquire(   Object o, long offset, Object expected, Object x);
2592 //   Object compareAndExchangeLongRelease(   Object o, long offset, Object expected, Object x);
2593 //
2594 // LS_get_add:
2595 //
2596 //   int  getAndAddInt( Object o, long offset, int  delta)
2597 //   long getAndAddLong(Object o, long offset, long delta)
2598 //
2599 // LS_get_set:
2600 //
2601 //   int    getAndSet(Object o, long offset, int    newValue)
2602 //   long   getAndSet(Object o, long offset, long   newValue)
2603 //   Object getAndSet(Object o, long offset, Object newValue)
2604 //
2605 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2606   // This basic scheme here is the same as inline_unsafe_access, but
2607   // differs in enough details that combining them would make the code
2608   // overly confusing.  (This is a true fact! I originally combined
2609   // them, but even I was confused by it!) As much code/comments as
2610   // possible are retained from inline_unsafe_access though to make
2611   // the correspondences clearer. - dl
2612 
2613   if (callee()->is_static())  return false;  // caller must have the capability!
2614 
2615 #ifndef PRODUCT
2616   BasicType rtype;
2617   {
2618     ResourceMark rm;
2619     // Check the signatures.
2620     ciSignature* sig = callee()->signature();
2621     rtype = sig->return_type()->basic_type();
2622     switch(kind) {
2623       case LS_get_add:
2624       case LS_get_set: {
2625       // Check the signatures.
2626 #ifdef ASSERT
2627       assert(rtype == type, "get and set must return the expected type");
2628       assert(sig->count() == 3, "get and set has 3 arguments");
2629       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2630       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2631       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2632       assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2633 #endif // ASSERT
2634         break;
2635       }
2636       case LS_cmp_swap:
2637       case LS_cmp_swap_weak: {
2638       // Check the signatures.
2639 #ifdef ASSERT
2640       assert(rtype == T_BOOLEAN, "CAS must return boolean");
2641       assert(sig->count() == 4, "CAS has 4 arguments");
2642       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2643       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2644 #endif // ASSERT
2645         break;
2646       }
2647       case LS_cmp_exchange: {
2648       // Check the signatures.
2649 #ifdef ASSERT
2650       assert(rtype == type, "CAS must return the expected type");
2651       assert(sig->count() == 4, "CAS has 4 arguments");
2652       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2653       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2654 #endif // ASSERT
2655         break;
2656       }
2657       default:
2658         ShouldNotReachHere();
2659     }
2660   }
2661 #endif //PRODUCT
2662 
2663   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2664 
2665   // Get arguments:
2666   Node* receiver = NULL;
2667   Node* base     = NULL;
2668   Node* offset   = NULL;
2669   Node* oldval   = NULL;
2670   Node* newval   = NULL;
2671   switch(kind) {
2672     case LS_cmp_swap:
2673     case LS_cmp_swap_weak:
2674     case LS_cmp_exchange: {
2675       const bool two_slot_type = type2size[type] == 2;
2676       receiver = argument(0);  // type: oop
2677       base     = argument(1);  // type: oop
2678       offset   = argument(2);  // type: long
2679       oldval   = argument(4);  // type: oop, int, or long
2680       newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2681       break;
2682     }
2683     case LS_get_add:
2684     case LS_get_set: {
2685       receiver = argument(0);  // type: oop
2686       base     = argument(1);  // type: oop
2687       offset   = argument(2);  // type: long
2688       oldval   = NULL;
2689       newval   = argument(4);  // type: oop, int, or long
2690       break;
2691     }
2692     default:
2693       ShouldNotReachHere();
2694   }
2695 
2696   // Null check receiver.
2697   receiver = null_check(receiver);
2698   if (stopped()) {
2699     return true;
2700   }
2701 
2702   // Build field offset expression.
2703   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2704   // to be plain byte offsets, which are also the same as those accepted
2705   // by oopDesc::field_base.
2706   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2707   // 32-bit machines ignore the high half of long offsets
2708   offset = ConvL2X(offset);
2709   Node* adr = make_unsafe_address(base, offset);
2710   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2711 
2712   Compile::AliasType* alias_type = C->alias_type(adr_type);
2713   assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2714          alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2715   BasicType bt = alias_type->basic_type();
2716   if (bt != T_ILLEGAL &&
2717       ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2718     // Don't intrinsify mismatched object accesses.
2719     return false;
2720   }
2721 
2722   // For CAS, unlike inline_unsafe_access, there seems no point in
2723   // trying to refine types. Just use the coarse types here.
2724   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2725   const Type *value_type = Type::get_const_basic_type(type);
2726 
2727   switch (kind) {
2728     case LS_get_set:
2729     case LS_cmp_exchange: {
2730       if (type == T_OBJECT) {
2731         const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2732         if (tjp != NULL) {
2733           value_type = tjp;
2734         }
2735       }
2736       break;
2737     }
2738     case LS_cmp_swap:
2739     case LS_cmp_swap_weak:
2740     case LS_get_add:
2741       break;
2742     default:
2743       ShouldNotReachHere();
2744   }
2745 
2746   int alias_idx = C->get_alias_index(adr_type);
2747 
2748   // Memory-model-wise, a LoadStore acts like a little synchronized
2749   // block, so needs barriers on each side.  These don't translate
2750   // into actual barriers on most machines, but we still need rest of
2751   // compiler to respect ordering.
2752 
2753   switch (access_kind) {
2754     case Relaxed:
2755     case Acquire:
2756       break;
2757     case Release:
2758       insert_mem_bar(Op_MemBarRelease);
2759       break;
2760     case Volatile:
2761       if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2762         insert_mem_bar(Op_MemBarVolatile);
2763       } else {
2764         insert_mem_bar(Op_MemBarRelease);
2765       }
2766       break;
2767     default:
2768       ShouldNotReachHere();
2769   }
2770   insert_mem_bar(Op_MemBarCPUOrder);
2771 
2772   // Figure out the memory ordering.
2773   MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2774 
2775   // 4984716: MemBars must be inserted before this
2776   //          memory node in order to avoid a false
2777   //          dependency which will confuse the scheduler.
2778   Node *mem = memory(alias_idx);
2779 
2780   // For now, we handle only those cases that actually exist: ints,
2781   // longs, and Object. Adding others should be straightforward.
2782   Node* load_store = NULL;
2783   switch(type) {
2784   case T_BYTE:
2785     switch(kind) {
2786       case LS_get_add:
2787         load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2788         break;
2789       case LS_get_set:
2790         load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2791         break;
2792       case LS_cmp_swap_weak:
2793         load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2794         break;
2795       case LS_cmp_swap:
2796         load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2797         break;
2798       case LS_cmp_exchange:
2799         load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2800         break;
2801       default:
2802         ShouldNotReachHere();
2803     }
2804     break;
2805   case T_SHORT:
2806     switch(kind) {
2807       case LS_get_add:
2808         load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2809         break;
2810       case LS_get_set:
2811         load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2812         break;
2813       case LS_cmp_swap_weak:
2814         load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2815         break;
2816       case LS_cmp_swap:
2817         load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2818         break;
2819       case LS_cmp_exchange:
2820         load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2821         break;
2822       default:
2823         ShouldNotReachHere();
2824     }
2825     break;
2826   case T_INT:
2827     switch(kind) {
2828       case LS_get_add:
2829         load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2830         break;
2831       case LS_get_set:
2832         load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2833         break;
2834       case LS_cmp_swap_weak:
2835         load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2836         break;
2837       case LS_cmp_swap:
2838         load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2839         break;
2840       case LS_cmp_exchange:
2841         load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2842         break;
2843       default:
2844         ShouldNotReachHere();
2845     }
2846     break;
2847   case T_LONG:
2848     switch(kind) {
2849       case LS_get_add:
2850         load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2851         break;
2852       case LS_get_set:
2853         load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2854         break;
2855       case LS_cmp_swap_weak:
2856         load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2857         break;
2858       case LS_cmp_swap:
2859         load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2860         break;
2861       case LS_cmp_exchange:
2862         load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2863         break;
2864       default:
2865         ShouldNotReachHere();
2866     }
2867     break;
2868   case T_OBJECT:
2869     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2870     // could be delayed during Parse (for example, in adjust_map_after_if()).
2871     // Execute transformation here to avoid barrier generation in such case.
2872     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2873       newval = _gvn.makecon(TypePtr::NULL_PTR);
2874 
2875     // Reference stores need a store barrier.
2876     switch(kind) {
2877       case LS_get_set: {
2878         // If pre-barrier must execute before the oop store, old value will require do_load here.
2879         if (!can_move_pre_barrier()) {
2880           pre_barrier(true /* do_load*/,
2881                       control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2882                       NULL /* pre_val*/,
2883                       T_OBJECT);
2884         } // Else move pre_barrier to use load_store value, see below.
2885         break;
2886       }
2887       case LS_cmp_swap_weak:
2888       case LS_cmp_swap:
2889       case LS_cmp_exchange: {
2890         // Same as for newval above:
2891         if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2892           oldval = _gvn.makecon(TypePtr::NULL_PTR);
2893         }
2894         // The only known value which might get overwritten is oldval.
2895         pre_barrier(false /* do_load */,
2896                     control(), NULL, NULL, max_juint, NULL, NULL,
2897                     oldval /* pre_val */,
2898                     T_OBJECT);
2899         break;
2900       }
2901       default:
2902         ShouldNotReachHere();
2903     }
2904 
2905 #ifdef _LP64
2906     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2907       Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2908 
2909       switch(kind) {
2910         case LS_get_set:
2911           load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2912           break;
2913         case LS_cmp_swap_weak: {
2914           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2915           load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2916           break;
2917         }
2918         case LS_cmp_swap: {
2919           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2920           load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2921           break;
2922         }
2923         case LS_cmp_exchange: {
2924           Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2925           load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2926           break;
2927         }
2928         default:
2929           ShouldNotReachHere();
2930       }
2931     } else
2932 #endif
2933     switch (kind) {
2934       case LS_get_set:
2935         load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2936         break;
2937       case LS_cmp_swap_weak:
2938         load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2939         break;
2940       case LS_cmp_swap:
2941         load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2942         break;
2943       case LS_cmp_exchange:
2944         load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2945         break;
2946       default:
2947         ShouldNotReachHere();
2948     }
2949 
2950     // Emit the post barrier only when the actual store happened. This makes sense
2951     // to check only for LS_cmp_* that can fail to set the value.
2952     // LS_cmp_exchange does not produce any branches by default, so there is no
2953     // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
2954     // CompareAndExchange and move branches here, it would make sense to conditionalize
2955     // post_barriers for LS_cmp_exchange as well.
2956     //
2957     // CAS success path is marked more likely since we anticipate this is a performance
2958     // critical path, while CAS failure path can use the penalty for going through unlikely
2959     // path as backoff. Which is still better than doing a store barrier there.
2960     switch (kind) {
2961       case LS_get_set:
2962       case LS_cmp_exchange: {
2963         post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2964         break;
2965       }
2966       case LS_cmp_swap_weak:
2967       case LS_cmp_swap: {
2968         IdealKit ideal(this);
2969         ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
2970           sync_kit(ideal);
2971           post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2972           ideal.sync_kit(this);
2973         } ideal.end_if();
2974         final_sync(ideal);
2975         break;
2976       }
2977       default:
2978         ShouldNotReachHere();
2979     }
2980     break;
2981   default:
2982     fatal("unexpected type %d: %s", type, type2name(type));
2983     break;
2984   }
2985 
2986   // SCMemProjNodes represent the memory state of a LoadStore. Their
2987   // main role is to prevent LoadStore nodes from being optimized away
2988   // when their results aren't used.
2989   Node* proj = _gvn.transform(new SCMemProjNode(load_store));
2990   set_memory(proj, alias_idx);
2991 
2992   if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
2993 #ifdef _LP64
2994     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2995       load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
2996     }
2997 #endif
2998     if (can_move_pre_barrier()) {
2999       // Don't need to load pre_val. The old value is returned by load_store.
3000       // The pre_barrier can execute after the xchg as long as no safepoint
3001       // gets inserted between them.
3002       pre_barrier(false /* do_load */,
3003                   control(), NULL, NULL, max_juint, NULL, NULL,
3004                   load_store /* pre_val */,
3005                   T_OBJECT);
3006     }
3007   }
3008 
3009   // Add the trailing membar surrounding the access
3010   insert_mem_bar(Op_MemBarCPUOrder);
3011 
3012   switch (access_kind) {
3013     case Relaxed:
3014     case Release:
3015       break; // do nothing
3016     case Acquire:
3017     case Volatile:
3018       insert_mem_bar(Op_MemBarAcquire);
3019       // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3020       break;
3021     default:
3022       ShouldNotReachHere();
3023   }
3024 
3025   assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3026   set_result(load_store);
3027   return true;
3028 }
3029 
3030 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3031   MemNode::MemOrd mo = MemNode::unset;
3032   switch(kind) {
3033     case Opaque:
3034     case Relaxed:  mo = MemNode::unordered; break;
3035     case Acquire:  mo = MemNode::acquire;   break;
3036     case Release:  mo = MemNode::release;   break;
3037     case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3038     default:
3039       ShouldNotReachHere();
3040   }
3041   guarantee(mo != MemNode::unset, "Should select memory ordering");
3042   return mo;
3043 }
3044 
3045 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3046   MemNode::MemOrd mo = MemNode::unset;
3047   switch(kind) {
3048     case Opaque:
3049     case Relaxed:  mo = MemNode::unordered; break;
3050     case Acquire:  mo = MemNode::acquire;   break;
3051     case Release:  mo = MemNode::release;   break;
3052     case Volatile: mo = MemNode::seqcst;    break;
3053     default:
3054       ShouldNotReachHere();
3055   }
3056   guarantee(mo != MemNode::unset, "Should select memory ordering");
3057   return mo;
3058 }
3059 
3060 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3061   // Regardless of form, don't allow previous ld/st to move down,
3062   // then issue acquire, release, or volatile mem_bar.
3063   insert_mem_bar(Op_MemBarCPUOrder);
3064   switch(id) {
3065     case vmIntrinsics::_loadFence:
3066       insert_mem_bar(Op_LoadFence);
3067       return true;
3068     case vmIntrinsics::_storeFence:
3069       insert_mem_bar(Op_StoreFence);
3070       return true;
3071     case vmIntrinsics::_fullFence:
3072       insert_mem_bar(Op_MemBarVolatile);
3073       return true;
3074     default:
3075       fatal_unexpected_iid(id);
3076       return false;
3077   }
3078 }
3079 
3080 bool LibraryCallKit::inline_onspinwait() {
3081   insert_mem_bar(Op_OnSpinWait);
3082   return true;
3083 }
3084 
3085 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3086   if (!kls->is_Con()) {
3087     return true;
3088   }
3089   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3090   if (klsptr == NULL) {
3091     return true;
3092   }
3093   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3094   // don't need a guard for a klass that is already initialized
3095   return !ik->is_initialized();
3096 }
3097 
3098 //----------------------------inline_unsafe_allocate---------------------------
3099 // public native Object Unsafe.allocateInstance(Class<?> cls);
3100 bool LibraryCallKit::inline_unsafe_allocate() {
3101   if (callee()->is_static())  return false;  // caller must have the capability!
3102 
3103   null_check_receiver();  // null-check, then ignore
3104   Node* cls = null_check(argument(1));
3105   if (stopped())  return true;
3106 
3107   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3108   kls = null_check(kls);
3109   if (stopped())  return true;  // argument was like int.class
3110 
3111   Node* test = NULL;
3112   if (LibraryCallKit::klass_needs_init_guard(kls)) {
3113     // Note:  The argument might still be an illegal value like
3114     // Serializable.class or Object[].class.   The runtime will handle it.
3115     // But we must make an explicit check for initialization.
3116     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3117     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3118     // can generate code to load it as unsigned byte.
3119     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3120     Node* bits = intcon(InstanceKlass::fully_initialized);
3121     test = _gvn.transform(new SubINode(inst, bits));
3122     // The 'test' is non-zero if we need to take a slow path.
3123   }
3124 
3125   Node* obj = new_instance(kls, test);
3126   set_result(obj);
3127   return true;
3128 }
3129 
3130 //------------------------inline_native_time_funcs--------------
3131 // inline code for System.currentTimeMillis() and System.nanoTime()
3132 // these have the same type and signature
3133 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3134   const TypeFunc* tf = OptoRuntime::void_long_Type();
3135   const TypePtr* no_memory_effects = NULL;
3136   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3137   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3138 #ifdef ASSERT
3139   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3140   assert(value_top == top(), "second value must be top");
3141 #endif
3142   set_result(value);
3143   return true;
3144 }
3145 
3146 //------------------------inline_native_currentThread------------------
3147 bool LibraryCallKit::inline_native_currentThread() {
3148   Node* junk = NULL;
3149   set_result(generate_current_thread(junk));
3150   return true;
3151 }
3152 
3153 //------------------------inline_native_isInterrupted------------------
3154 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3155 bool LibraryCallKit::inline_native_isInterrupted() {
3156   // Add a fast path to t.isInterrupted(clear_int):
3157   //   (t == Thread.current() &&
3158   //    (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3159   //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3160   // So, in the common case that the interrupt bit is false,
3161   // we avoid making a call into the VM.  Even if the interrupt bit
3162   // is true, if the clear_int argument is false, we avoid the VM call.
3163   // However, if the receiver is not currentThread, we must call the VM,
3164   // because there must be some locking done around the operation.
3165 
3166   // We only go to the fast case code if we pass two guards.
3167   // Paths which do not pass are accumulated in the slow_region.
3168 
3169   enum {
3170     no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
3171     no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
3172     slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
3173     PATH_LIMIT
3174   };
3175 
3176   // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3177   // out of the function.
3178   insert_mem_bar(Op_MemBarCPUOrder);
3179 
3180   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3181   PhiNode*    result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3182 
3183   RegionNode* slow_region = new RegionNode(1);
3184   record_for_igvn(slow_region);
3185 
3186   // (a) Receiving thread must be the current thread.
3187   Node* rec_thr = argument(0);
3188   Node* tls_ptr = NULL;
3189   Node* cur_thr = generate_current_thread(tls_ptr);
3190   Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3191   Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3192 
3193   generate_slow_guard(bol_thr, slow_region);
3194 
3195   // (b) Interrupt bit on TLS must be false.
3196   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3197   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3198   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3199 
3200   // Set the control input on the field _interrupted read to prevent it floating up.
3201   Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3202   Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3203   Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3204 
3205   IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3206 
3207   // First fast path:  if (!TLS._interrupted) return false;
3208   Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3209   result_rgn->init_req(no_int_result_path, false_bit);
3210   result_val->init_req(no_int_result_path, intcon(0));
3211 
3212   // drop through to next case
3213   set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3214 
3215 #ifndef TARGET_OS_FAMILY_windows
3216   // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3217   Node* clr_arg = argument(1);
3218   Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3219   Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3220   IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3221 
3222   // Second fast path:  ... else if (!clear_int) return true;
3223   Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3224   result_rgn->init_req(no_clear_result_path, false_arg);
3225   result_val->init_req(no_clear_result_path, intcon(1));
3226 
3227   // drop through to next case
3228   set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3229 #else
3230   // To return true on Windows you must read the _interrupted field
3231   // and check the event state i.e. take the slow path.
3232 #endif // TARGET_OS_FAMILY_windows
3233 
3234   // (d) Otherwise, go to the slow path.
3235   slow_region->add_req(control());
3236   set_control( _gvn.transform(slow_region));
3237 
3238   if (stopped()) {
3239     // There is no slow path.
3240     result_rgn->init_req(slow_result_path, top());
3241     result_val->init_req(slow_result_path, top());
3242   } else {
3243     // non-virtual because it is a private non-static
3244     CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3245 
3246     Node* slow_val = set_results_for_java_call(slow_call);
3247     // this->control() comes from set_results_for_java_call
3248 
3249     Node* fast_io  = slow_call->in(TypeFunc::I_O);
3250     Node* fast_mem = slow_call->in(TypeFunc::Memory);
3251 
3252     // These two phis are pre-filled with copies of of the fast IO and Memory
3253     PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3254     PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
3255 
3256     result_rgn->init_req(slow_result_path, control());
3257     result_io ->init_req(slow_result_path, i_o());
3258     result_mem->init_req(slow_result_path, reset_memory());
3259     result_val->init_req(slow_result_path, slow_val);
3260 
3261     set_all_memory(_gvn.transform(result_mem));
3262     set_i_o(       _gvn.transform(result_io));
3263   }
3264 
3265   C->set_has_split_ifs(true); // Has chance for split-if optimization
3266   set_result(result_rgn, result_val);
3267   return true;
3268 }
3269 
3270 //---------------------------load_mirror_from_klass----------------------------
3271 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3272 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3273   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3274   return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3275 }
3276 
3277 //-----------------------load_klass_from_mirror_common-------------------------
3278 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3279 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3280 // and branch to the given path on the region.
3281 // If never_see_null, take an uncommon trap on null, so we can optimistically
3282 // compile for the non-null case.
3283 // If the region is NULL, force never_see_null = true.
3284 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3285                                                     bool never_see_null,
3286                                                     RegionNode* region,
3287                                                     int null_path,
3288                                                     int offset) {
3289   if (region == NULL)  never_see_null = true;
3290   Node* p = basic_plus_adr(mirror, offset);
3291   const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3292   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3293   Node* null_ctl = top();
3294   kls = null_check_oop(kls, &null_ctl, never_see_null);
3295   if (region != NULL) {
3296     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3297     region->init_req(null_path, null_ctl);
3298   } else {
3299     assert(null_ctl == top(), "no loose ends");
3300   }
3301   return kls;
3302 }
3303 
3304 //--------------------(inline_native_Class_query helpers)---------------------
3305 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3306 // Fall through if (mods & mask) == bits, take the guard otherwise.
3307 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3308   // Branch around if the given klass has the given modifier bit set.
3309   // Like generate_guard, adds a new path onto the region.
3310   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3311   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3312   Node* mask = intcon(modifier_mask);
3313   Node* bits = intcon(modifier_bits);
3314   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3315   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3316   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3317   return generate_fair_guard(bol, region);
3318 }
3319 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3320   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3321 }
3322 
3323 //-------------------------inline_native_Class_query-------------------
3324 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3325   const Type* return_type = TypeInt::BOOL;
3326   Node* prim_return_value = top();  // what happens if it's a primitive class?
3327   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3328   bool expect_prim = false;     // most of these guys expect to work on refs
3329 
3330   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3331 
3332   Node* mirror = argument(0);
3333   Node* obj    = top();
3334 
3335   switch (id) {
3336   case vmIntrinsics::_isInstance:
3337     // nothing is an instance of a primitive type
3338     prim_return_value = intcon(0);
3339     obj = argument(1);
3340     break;
3341   case vmIntrinsics::_getModifiers:
3342     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3343     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3344     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3345     break;
3346   case vmIntrinsics::_isInterface:
3347     prim_return_value = intcon(0);
3348     break;
3349   case vmIntrinsics::_isArray:
3350     prim_return_value = intcon(0);
3351     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3352     break;
3353   case vmIntrinsics::_isPrimitive:
3354     prim_return_value = intcon(1);
3355     expect_prim = true;  // obviously
3356     break;
3357   case vmIntrinsics::_getSuperclass:
3358     prim_return_value = null();
3359     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3360     break;
3361   case vmIntrinsics::_getClassAccessFlags:
3362     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3363     return_type = TypeInt::INT;  // not bool!  6297094
3364     break;
3365   default:
3366     fatal_unexpected_iid(id);
3367     break;
3368   }
3369 
3370   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3371   if (mirror_con == NULL)  return false;  // cannot happen?
3372 
3373 #ifndef PRODUCT
3374   if (C->print_intrinsics() || C->print_inlining()) {
3375     ciType* k = mirror_con->java_mirror_type();
3376     if (k) {
3377       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3378       k->print_name();
3379       tty->cr();
3380     }
3381   }
3382 #endif
3383 
3384   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3385   RegionNode* region = new RegionNode(PATH_LIMIT);
3386   record_for_igvn(region);
3387   PhiNode* phi = new PhiNode(region, return_type);
3388 
3389   // The mirror will never be null of Reflection.getClassAccessFlags, however
3390   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3391   // if it is. See bug 4774291.
3392 
3393   // For Reflection.getClassAccessFlags(), the null check occurs in
3394   // the wrong place; see inline_unsafe_access(), above, for a similar
3395   // situation.
3396   mirror = null_check(mirror);
3397   // If mirror or obj is dead, only null-path is taken.
3398   if (stopped())  return true;
3399 
3400   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3401 
3402   // Now load the mirror's klass metaobject, and null-check it.
3403   // Side-effects region with the control path if the klass is null.
3404   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3405   // If kls is null, we have a primitive mirror.
3406   phi->init_req(_prim_path, prim_return_value);
3407   if (stopped()) { set_result(region, phi); return true; }
3408   bool safe_for_replace = (region->in(_prim_path) == top());
3409 
3410   Node* p;  // handy temp
3411   Node* null_ctl;
3412 
3413   // Now that we have the non-null klass, we can perform the real query.
3414   // For constant classes, the query will constant-fold in LoadNode::Value.
3415   Node* query_value = top();
3416   switch (id) {
3417   case vmIntrinsics::_isInstance:
3418     // nothing is an instance of a primitive type
3419     query_value = gen_instanceof(obj, kls, safe_for_replace);
3420     break;
3421 
3422   case vmIntrinsics::_getModifiers:
3423     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3424     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3425     break;
3426 
3427   case vmIntrinsics::_isInterface:
3428     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3429     if (generate_interface_guard(kls, region) != NULL)
3430       // A guard was added.  If the guard is taken, it was an interface.
3431       phi->add_req(intcon(1));
3432     // If we fall through, it's a plain class.
3433     query_value = intcon(0);
3434     break;
3435 
3436   case vmIntrinsics::_isArray:
3437     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3438     if (generate_array_guard(kls, region) != NULL)
3439       // A guard was added.  If the guard is taken, it was an array.
3440       phi->add_req(intcon(1));
3441     // If we fall through, it's a plain class.
3442     query_value = intcon(0);
3443     break;
3444 
3445   case vmIntrinsics::_isPrimitive:
3446     query_value = intcon(0); // "normal" path produces false
3447     break;
3448 
3449   case vmIntrinsics::_getSuperclass:
3450     // The rules here are somewhat unfortunate, but we can still do better
3451     // with random logic than with a JNI call.
3452     // Interfaces store null or Object as _super, but must report null.
3453     // Arrays store an intermediate super as _super, but must report Object.
3454     // Other types can report the actual _super.
3455     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3456     if (generate_interface_guard(kls, region) != NULL)
3457       // A guard was added.  If the guard is taken, it was an interface.
3458       phi->add_req(null());
3459     if (generate_array_guard(kls, region) != NULL)
3460       // A guard was added.  If the guard is taken, it was an array.
3461       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3462     // If we fall through, it's a plain class.  Get its _super.
3463     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3464     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3465     null_ctl = top();
3466     kls = null_check_oop(kls, &null_ctl);
3467     if (null_ctl != top()) {
3468       // If the guard is taken, Object.superClass is null (both klass and mirror).
3469       region->add_req(null_ctl);
3470       phi   ->add_req(null());
3471     }
3472     if (!stopped()) {
3473       query_value = load_mirror_from_klass(kls);
3474     }
3475     break;
3476 
3477   case vmIntrinsics::_getClassAccessFlags:
3478     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3479     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3480     break;
3481 
3482   default:
3483     fatal_unexpected_iid(id);
3484     break;
3485   }
3486 
3487   // Fall-through is the normal case of a query to a real class.
3488   phi->init_req(1, query_value);
3489   region->init_req(1, control());
3490 
3491   C->set_has_split_ifs(true); // Has chance for split-if optimization
3492   set_result(region, phi);
3493   return true;
3494 }
3495 
3496 //-------------------------inline_Class_cast-------------------
3497 bool LibraryCallKit::inline_Class_cast() {
3498   Node* mirror = argument(0); // Class
3499   Node* obj    = argument(1);
3500   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3501   if (mirror_con == NULL) {
3502     return false;  // dead path (mirror->is_top()).
3503   }
3504   if (obj == NULL || obj->is_top()) {
3505     return false;  // dead path
3506   }
3507   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3508 
3509   // First, see if Class.cast() can be folded statically.
3510   // java_mirror_type() returns non-null for compile-time Class constants.
3511   ciType* tm = mirror_con->java_mirror_type();
3512   if (tm != NULL && tm->is_klass() &&
3513       tp != NULL && tp->klass() != NULL) {
3514     if (!tp->klass()->is_loaded()) {
3515       // Don't use intrinsic when class is not loaded.
3516       return false;
3517     } else {
3518       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3519       if (static_res == Compile::SSC_always_true) {
3520         // isInstance() is true - fold the code.
3521         set_result(obj);
3522         return true;
3523       } else if (static_res == Compile::SSC_always_false) {
3524         // Don't use intrinsic, have to throw ClassCastException.
3525         // If the reference is null, the non-intrinsic bytecode will
3526         // be optimized appropriately.
3527         return false;
3528       }
3529     }
3530   }
3531 
3532   // Bailout intrinsic and do normal inlining if exception path is frequent.
3533   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3534     return false;
3535   }
3536 
3537   // Generate dynamic checks.
3538   // Class.cast() is java implementation of _checkcast bytecode.
3539   // Do checkcast (Parse::do_checkcast()) optimizations here.
3540 
3541   mirror = null_check(mirror);
3542   // If mirror is dead, only null-path is taken.
3543   if (stopped()) {
3544     return true;
3545   }
3546 
3547   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3548   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3549   RegionNode* region = new RegionNode(PATH_LIMIT);
3550   record_for_igvn(region);
3551 
3552   // Now load the mirror's klass metaobject, and null-check it.
3553   // If kls is null, we have a primitive mirror and
3554   // nothing is an instance of a primitive type.
3555   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3556 
3557   Node* res = top();
3558   if (!stopped()) {
3559     Node* bad_type_ctrl = top();
3560     // Do checkcast optimizations.
3561     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3562     region->init_req(_bad_type_path, bad_type_ctrl);
3563   }
3564   if (region->in(_prim_path) != top() ||
3565       region->in(_bad_type_path) != top()) {
3566     // Let Interpreter throw ClassCastException.
3567     PreserveJVMState pjvms(this);
3568     set_control(_gvn.transform(region));
3569     uncommon_trap(Deoptimization::Reason_intrinsic,
3570                   Deoptimization::Action_maybe_recompile);
3571   }
3572   if (!stopped()) {
3573     set_result(res);
3574   }
3575   return true;
3576 }
3577 
3578 
3579 //--------------------------inline_native_subtype_check------------------------
3580 // This intrinsic takes the JNI calls out of the heart of
3581 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3582 bool LibraryCallKit::inline_native_subtype_check() {
3583   // Pull both arguments off the stack.
3584   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3585   args[0] = argument(0);
3586   args[1] = argument(1);
3587   Node* klasses[2];             // corresponding Klasses: superk, subk
3588   klasses[0] = klasses[1] = top();
3589 
3590   enum {
3591     // A full decision tree on {superc is prim, subc is prim}:
3592     _prim_0_path = 1,           // {P,N} => false
3593                                 // {P,P} & superc!=subc => false
3594     _prim_same_path,            // {P,P} & superc==subc => true
3595     _prim_1_path,               // {N,P} => false
3596     _ref_subtype_path,          // {N,N} & subtype check wins => true
3597     _both_ref_path,             // {N,N} & subtype check loses => false
3598     PATH_LIMIT
3599   };
3600 
3601   RegionNode* region = new RegionNode(PATH_LIMIT);
3602   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3603   record_for_igvn(region);
3604 
3605   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3606   const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3607   int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3608 
3609   // First null-check both mirrors and load each mirror's klass metaobject.
3610   int which_arg;
3611   for (which_arg = 0; which_arg <= 1; which_arg++) {
3612     Node* arg = args[which_arg];
3613     arg = null_check(arg);
3614     if (stopped())  break;
3615     args[which_arg] = arg;
3616 
3617     Node* p = basic_plus_adr(arg, class_klass_offset);
3618     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3619     klasses[which_arg] = _gvn.transform(kls);
3620   }
3621 
3622   // Having loaded both klasses, test each for null.
3623   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3624   for (which_arg = 0; which_arg <= 1; which_arg++) {
3625     Node* kls = klasses[which_arg];
3626     Node* null_ctl = top();
3627     kls = null_check_oop(kls, &null_ctl, never_see_null);
3628     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3629     region->init_req(prim_path, null_ctl);
3630     if (stopped())  break;
3631     klasses[which_arg] = kls;
3632   }
3633 
3634   if (!stopped()) {
3635     // now we have two reference types, in klasses[0..1]
3636     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3637     Node* superk = klasses[0];  // the receiver
3638     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3639     // now we have a successful reference subtype check
3640     region->set_req(_ref_subtype_path, control());
3641   }
3642 
3643   // If both operands are primitive (both klasses null), then
3644   // we must return true when they are identical primitives.
3645   // It is convenient to test this after the first null klass check.
3646   set_control(region->in(_prim_0_path)); // go back to first null check
3647   if (!stopped()) {
3648     // Since superc is primitive, make a guard for the superc==subc case.
3649     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3650     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3651     generate_guard(bol_eq, region, PROB_FAIR);
3652     if (region->req() == PATH_LIMIT+1) {
3653       // A guard was added.  If the added guard is taken, superc==subc.
3654       region->swap_edges(PATH_LIMIT, _prim_same_path);
3655       region->del_req(PATH_LIMIT);
3656     }
3657     region->set_req(_prim_0_path, control()); // Not equal after all.
3658   }
3659 
3660   // these are the only paths that produce 'true':
3661   phi->set_req(_prim_same_path,   intcon(1));
3662   phi->set_req(_ref_subtype_path, intcon(1));
3663 
3664   // pull together the cases:
3665   assert(region->req() == PATH_LIMIT, "sane region");
3666   for (uint i = 1; i < region->req(); i++) {
3667     Node* ctl = region->in(i);
3668     if (ctl == NULL || ctl == top()) {
3669       region->set_req(i, top());
3670       phi   ->set_req(i, top());
3671     } else if (phi->in(i) == NULL) {
3672       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3673     }
3674   }
3675 
3676   set_control(_gvn.transform(region));
3677   set_result(_gvn.transform(phi));
3678   return true;
3679 }
3680 
3681 //---------------------generate_array_guard_common------------------------
3682 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3683                                                   bool obj_array, bool not_array) {
3684 
3685   if (stopped()) {
3686     return NULL;
3687   }
3688 
3689   // If obj_array/non_array==false/false:
3690   // Branch around if the given klass is in fact an array (either obj or prim).
3691   // If obj_array/non_array==false/true:
3692   // Branch around if the given klass is not an array klass of any kind.
3693   // If obj_array/non_array==true/true:
3694   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3695   // If obj_array/non_array==true/false:
3696   // Branch around if the kls is an oop array (Object[] or subtype)
3697   //
3698   // Like generate_guard, adds a new path onto the region.
3699   jint  layout_con = 0;
3700   Node* layout_val = get_layout_helper(kls, layout_con);
3701   if (layout_val == NULL) {
3702     bool query = (obj_array
3703                   ? Klass::layout_helper_is_objArray(layout_con)
3704                   : Klass::layout_helper_is_array(layout_con));
3705     if (query == not_array) {
3706       return NULL;                       // never a branch
3707     } else {                             // always a branch
3708       Node* always_branch = control();
3709       if (region != NULL)
3710         region->add_req(always_branch);
3711       set_control(top());
3712       return always_branch;
3713     }
3714   }
3715   // Now test the correct condition.
3716   jint  nval = (obj_array
3717                 ? (jint)(Klass::_lh_array_tag_type_value
3718                    <<    Klass::_lh_array_tag_shift)
3719                 : Klass::_lh_neutral_value);
3720   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3721   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3722   // invert the test if we are looking for a non-array
3723   if (not_array)  btest = BoolTest(btest).negate();
3724   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3725   return generate_fair_guard(bol, region);
3726 }
3727 
3728 
3729 //-----------------------inline_native_newArray--------------------------
3730 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3731 // private        native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3732 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3733   Node* mirror;
3734   Node* count_val;
3735   if (uninitialized) {
3736     mirror    = argument(1);
3737     count_val = argument(2);
3738   } else {
3739     mirror    = argument(0);
3740     count_val = argument(1);
3741   }
3742 
3743   mirror = null_check(mirror);
3744   // If mirror or obj is dead, only null-path is taken.
3745   if (stopped())  return true;
3746 
3747   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3748   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3749   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3750   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3751   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3752 
3753   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3754   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3755                                                   result_reg, _slow_path);
3756   Node* normal_ctl   = control();
3757   Node* no_array_ctl = result_reg->in(_slow_path);
3758 
3759   // Generate code for the slow case.  We make a call to newArray().
3760   set_control(no_array_ctl);
3761   if (!stopped()) {
3762     // Either the input type is void.class, or else the
3763     // array klass has not yet been cached.  Either the
3764     // ensuing call will throw an exception, or else it
3765     // will cache the array klass for next time.
3766     PreserveJVMState pjvms(this);
3767     CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3768     Node* slow_result = set_results_for_java_call(slow_call);
3769     // this->control() comes from set_results_for_java_call
3770     result_reg->set_req(_slow_path, control());
3771     result_val->set_req(_slow_path, slow_result);
3772     result_io ->set_req(_slow_path, i_o());
3773     result_mem->set_req(_slow_path, reset_memory());
3774   }
3775 
3776   set_control(normal_ctl);
3777   if (!stopped()) {
3778     // Normal case:  The array type has been cached in the java.lang.Class.
3779     // The following call works fine even if the array type is polymorphic.
3780     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3781     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3782     result_reg->init_req(_normal_path, control());
3783     result_val->init_req(_normal_path, obj);
3784     result_io ->init_req(_normal_path, i_o());
3785     result_mem->init_req(_normal_path, reset_memory());
3786 
3787     if (uninitialized) {
3788       // Mark the allocation so that zeroing is skipped
3789       AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3790       alloc->maybe_set_complete(&_gvn);
3791     }
3792   }
3793 
3794   // Return the combined state.
3795   set_i_o(        _gvn.transform(result_io)  );
3796   set_all_memory( _gvn.transform(result_mem));
3797 
3798   C->set_has_split_ifs(true); // Has chance for split-if optimization
3799   set_result(result_reg, result_val);
3800   return true;
3801 }
3802 
3803 //----------------------inline_native_getLength--------------------------
3804 // public static native int java.lang.reflect.Array.getLength(Object array);
3805 bool LibraryCallKit::inline_native_getLength() {
3806   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3807 
3808   Node* array = null_check(argument(0));
3809   // If array is dead, only null-path is taken.
3810   if (stopped())  return true;
3811 
3812   // Deoptimize if it is a non-array.
3813   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3814 
3815   if (non_array != NULL) {
3816     PreserveJVMState pjvms(this);
3817     set_control(non_array);
3818     uncommon_trap(Deoptimization::Reason_intrinsic,
3819                   Deoptimization::Action_maybe_recompile);
3820   }
3821 
3822   // If control is dead, only non-array-path is taken.
3823   if (stopped())  return true;
3824 
3825   // The works fine even if the array type is polymorphic.
3826   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3827   Node* result = load_array_length(array);
3828 
3829   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3830   set_result(result);
3831   return true;
3832 }
3833 
3834 //------------------------inline_array_copyOf----------------------------
3835 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3836 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3837 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3838   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3839 
3840   // Get the arguments.
3841   Node* original          = argument(0);
3842   Node* start             = is_copyOfRange? argument(1): intcon(0);
3843   Node* end               = is_copyOfRange? argument(2): argument(1);
3844   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3845 
3846   Node* newcopy = NULL;
3847 
3848   // Set the original stack and the reexecute bit for the interpreter to reexecute
3849   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3850   { PreserveReexecuteState preexecs(this);
3851     jvms()->set_should_reexecute(true);
3852 
3853     array_type_mirror = null_check(array_type_mirror);
3854     original          = null_check(original);
3855 
3856     // Check if a null path was taken unconditionally.
3857     if (stopped())  return true;
3858 
3859     Node* orig_length = load_array_length(original);
3860 
3861     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3862     klass_node = null_check(klass_node);
3863 
3864     RegionNode* bailout = new RegionNode(1);
3865     record_for_igvn(bailout);
3866 
3867     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3868     // Bail out if that is so.
3869     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3870     if (not_objArray != NULL) {
3871       // Improve the klass node's type from the new optimistic assumption:
3872       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3873       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3874       Node* cast = new CastPPNode(klass_node, akls);
3875       cast->init_req(0, control());
3876       klass_node = _gvn.transform(cast);
3877     }
3878 
3879     // Bail out if either start or end is negative.
3880     generate_negative_guard(start, bailout, &start);
3881     generate_negative_guard(end,   bailout, &end);
3882 
3883     Node* length = end;
3884     if (_gvn.type(start) != TypeInt::ZERO) {
3885       length = _gvn.transform(new SubINode(end, start));
3886     }
3887 
3888     // Bail out if length is negative.
3889     // Without this the new_array would throw
3890     // NegativeArraySizeException but IllegalArgumentException is what
3891     // should be thrown
3892     generate_negative_guard(length, bailout, &length);
3893 
3894     if (bailout->req() > 1) {
3895       PreserveJVMState pjvms(this);
3896       set_control(_gvn.transform(bailout));
3897       uncommon_trap(Deoptimization::Reason_intrinsic,
3898                     Deoptimization::Action_maybe_recompile);
3899     }
3900 
3901     if (!stopped()) {
3902       // How many elements will we copy from the original?
3903       // The answer is MinI(orig_length - start, length).
3904       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3905       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3906 
3907       // Generate a direct call to the right arraycopy function(s).
3908       // We know the copy is disjoint but we might not know if the
3909       // oop stores need checking.
3910       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3911       // This will fail a store-check if x contains any non-nulls.
3912 
3913       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3914       // loads/stores but it is legal only if we're sure the
3915       // Arrays.copyOf would succeed. So we need all input arguments
3916       // to the copyOf to be validated, including that the copy to the
3917       // new array won't trigger an ArrayStoreException. That subtype
3918       // check can be optimized if we know something on the type of
3919       // the input array from type speculation.
3920       if (_gvn.type(klass_node)->singleton()) {
3921         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3922         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3923 
3924         int test = C->static_subtype_check(superk, subk);
3925         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3926           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3927           if (t_original->speculative_type() != NULL) {
3928             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3929           }
3930         }
3931       }
3932 
3933       bool validated = false;
3934       // Reason_class_check rather than Reason_intrinsic because we
3935       // want to intrinsify even if this traps.
3936       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3937         Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3938                                                    klass_node);
3939 
3940         if (not_subtype_ctrl != top()) {
3941           PreserveJVMState pjvms(this);
3942           set_control(not_subtype_ctrl);
3943           uncommon_trap(Deoptimization::Reason_class_check,
3944                         Deoptimization::Action_make_not_entrant);
3945           assert(stopped(), "Should be stopped");
3946         }
3947         validated = true;
3948       }
3949 
3950       if (!stopped()) {
3951         newcopy = new_array(klass_node, length, 0);  // no arguments to push
3952 
3953         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
3954                                                 load_object_klass(original), klass_node);
3955         if (!is_copyOfRange) {
3956           ac->set_copyof(validated);
3957         } else {
3958           ac->set_copyofrange(validated);
3959         }
3960         Node* n = _gvn.transform(ac);
3961         if (n == ac) {
3962           ac->connect_outputs(this);
3963         } else {
3964           assert(validated, "shouldn't transform if all arguments not validated");
3965           set_all_memory(n);
3966         }
3967       }
3968     }
3969   } // original reexecute is set back here
3970 
3971   C->set_has_split_ifs(true); // Has chance for split-if optimization
3972   if (!stopped()) {
3973     set_result(newcopy);
3974   }
3975   return true;
3976 }
3977 
3978 
3979 //----------------------generate_virtual_guard---------------------------
3980 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
3981 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3982                                              RegionNode* slow_region) {
3983   ciMethod* method = callee();
3984   int vtable_index = method->vtable_index();
3985   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3986          "bad index %d", vtable_index);
3987   // Get the Method* out of the appropriate vtable entry.
3988   int entry_offset  = in_bytes(Klass::vtable_start_offset()) +
3989                      vtable_index*vtableEntry::size_in_bytes() +
3990                      vtableEntry::method_offset_in_bytes();
3991   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
3992   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3993 
3994   // Compare the target method with the expected method (e.g., Object.hashCode).
3995   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3996 
3997   Node* native_call = makecon(native_call_addr);
3998   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
3999   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4000 
4001   return generate_slow_guard(test_native, slow_region);
4002 }
4003 
4004 //-----------------------generate_method_call----------------------------
4005 // Use generate_method_call to make a slow-call to the real
4006 // method if the fast path fails.  An alternative would be to
4007 // use a stub like OptoRuntime::slow_arraycopy_Java.
4008 // This only works for expanding the current library call,
4009 // not another intrinsic.  (E.g., don't use this for making an
4010 // arraycopy call inside of the copyOf intrinsic.)
4011 CallJavaNode*
4012 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4013   // When compiling the intrinsic method itself, do not use this technique.
4014   guarantee(callee() != C->method(), "cannot make slow-call to self");
4015 
4016   ciMethod* method = callee();
4017   // ensure the JVMS we have will be correct for this call
4018   guarantee(method_id == method->intrinsic_id(), "must match");
4019 
4020   const TypeFunc* tf = TypeFunc::make(method);
4021   CallJavaNode* slow_call;
4022   if (is_static) {
4023     assert(!is_virtual, "");
4024     slow_call = new CallStaticJavaNode(C, tf,
4025                            SharedRuntime::get_resolve_static_call_stub(),
4026                            method, bci());
4027   } else if (is_virtual) {
4028     null_check_receiver();
4029     int vtable_index = Method::invalid_vtable_index;
4030     if (UseInlineCaches) {
4031       // Suppress the vtable call
4032     } else {
4033       // hashCode and clone are not a miranda methods,
4034       // so the vtable index is fixed.
4035       // No need to use the linkResolver to get it.
4036        vtable_index = method->vtable_index();
4037        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4038               "bad index %d", vtable_index);
4039     }
4040     slow_call = new CallDynamicJavaNode(tf,
4041                           SharedRuntime::get_resolve_virtual_call_stub(),
4042                           method, vtable_index, bci());
4043   } else {  // neither virtual nor static:  opt_virtual
4044     null_check_receiver();
4045     slow_call = new CallStaticJavaNode(C, tf,
4046                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4047                                 method, bci());
4048     slow_call->set_optimized_virtual(true);
4049   }
4050   set_arguments_for_java_call(slow_call);
4051   set_edges_for_java_call(slow_call);
4052   return slow_call;
4053 }
4054 
4055 
4056 /**
4057  * Build special case code for calls to hashCode on an object. This call may
4058  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4059  * slightly different code.
4060  */
4061 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4062   assert(is_static == callee()->is_static(), "correct intrinsic selection");
4063   assert(!(is_virtual && is_static), "either virtual, special, or static");
4064 
4065   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4066 
4067   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4068   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
4069   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
4070   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4071   Node* obj = NULL;
4072   if (!is_static) {
4073     // Check for hashing null object
4074     obj = null_check_receiver();
4075     if (stopped())  return true;        // unconditionally null
4076     result_reg->init_req(_null_path, top());
4077     result_val->init_req(_null_path, top());
4078   } else {
4079     // Do a null check, and return zero if null.
4080     // System.identityHashCode(null) == 0
4081     obj = argument(0);
4082     Node* null_ctl = top();
4083     obj = null_check_oop(obj, &null_ctl);
4084     result_reg->init_req(_null_path, null_ctl);
4085     result_val->init_req(_null_path, _gvn.intcon(0));
4086   }
4087 
4088   // Unconditionally null?  Then return right away.
4089   if (stopped()) {
4090     set_control( result_reg->in(_null_path));
4091     if (!stopped())
4092       set_result(result_val->in(_null_path));
4093     return true;
4094   }
4095 
4096   // We only go to the fast case code if we pass a number of guards.  The
4097   // paths which do not pass are accumulated in the slow_region.
4098   RegionNode* slow_region = new RegionNode(1);
4099   record_for_igvn(slow_region);
4100 
4101   // If this is a virtual call, we generate a funny guard.  We pull out
4102   // the vtable entry corresponding to hashCode() from the target object.
4103   // If the target method which we are calling happens to be the native
4104   // Object hashCode() method, we pass the guard.  We do not need this
4105   // guard for non-virtual calls -- the caller is known to be the native
4106   // Object hashCode().
4107   if (is_virtual) {
4108     // After null check, get the object's klass.
4109     Node* obj_klass = load_object_klass(obj);
4110     generate_virtual_guard(obj_klass, slow_region);
4111   }
4112 
4113   // Get the header out of the object, use LoadMarkNode when available
4114   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4115   // The control of the load must be NULL. Otherwise, the load can move before
4116   // the null check after castPP removal.
4117   Node* no_ctrl = NULL;
4118   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4119 
4120   // Test the header to see if it is unlocked.
4121   Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4122   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4123   Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
4124   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4125   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4126 
4127   generate_slow_guard(test_unlocked, slow_region);
4128 
4129   // Get the hash value and check to see that it has been properly assigned.
4130   // We depend on hash_mask being at most 32 bits and avoid the use of
4131   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4132   // vm: see markOop.hpp.
4133   Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
4134   Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
4135   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4136   // This hack lets the hash bits live anywhere in the mark object now, as long
4137   // as the shift drops the relevant bits into the low 32 bits.  Note that
4138   // Java spec says that HashCode is an int so there's no point in capturing
4139   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4140   hshifted_header      = ConvX2I(hshifted_header);
4141   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4142 
4143   Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
4144   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4145   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4146 
4147   generate_slow_guard(test_assigned, slow_region);
4148 
4149   Node* init_mem = reset_memory();
4150   // fill in the rest of the null path:
4151   result_io ->init_req(_null_path, i_o());
4152   result_mem->init_req(_null_path, init_mem);
4153 
4154   result_val->init_req(_fast_path, hash_val);
4155   result_reg->init_req(_fast_path, control());
4156   result_io ->init_req(_fast_path, i_o());
4157   result_mem->init_req(_fast_path, init_mem);
4158 
4159   // Generate code for the slow case.  We make a call to hashCode().
4160   set_control(_gvn.transform(slow_region));
4161   if (!stopped()) {
4162     // No need for PreserveJVMState, because we're using up the present state.
4163     set_all_memory(init_mem);
4164     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4165     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4166     Node* slow_result = set_results_for_java_call(slow_call);
4167     // this->control() comes from set_results_for_java_call
4168     result_reg->init_req(_slow_path, control());
4169     result_val->init_req(_slow_path, slow_result);
4170     result_io  ->set_req(_slow_path, i_o());
4171     result_mem ->set_req(_slow_path, reset_memory());
4172   }
4173 
4174   // Return the combined state.
4175   set_i_o(        _gvn.transform(result_io)  );
4176   set_all_memory( _gvn.transform(result_mem));
4177 
4178   set_result(result_reg, result_val);
4179   return true;
4180 }
4181 
4182 //---------------------------inline_native_getClass----------------------------
4183 // public final native Class<?> java.lang.Object.getClass();
4184 //
4185 // Build special case code for calls to getClass on an object.
4186 bool LibraryCallKit::inline_native_getClass() {
4187   Node* obj = null_check_receiver();
4188   if (stopped())  return true;
4189   set_result(load_mirror_from_klass(load_object_klass(obj)));
4190   return true;
4191 }
4192 
4193 //-----------------inline_native_Reflection_getCallerClass---------------------
4194 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4195 //
4196 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4197 //
4198 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4199 // in that it must skip particular security frames and checks for
4200 // caller sensitive methods.
4201 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4202 #ifndef PRODUCT
4203   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4204     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4205   }
4206 #endif
4207 
4208   if (!jvms()->has_method()) {
4209 #ifndef PRODUCT
4210     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4211       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
4212     }
4213 #endif
4214     return false;
4215   }
4216 
4217   // Walk back up the JVM state to find the caller at the required
4218   // depth.
4219   JVMState* caller_jvms = jvms();
4220 
4221   // Cf. JVM_GetCallerClass
4222   // NOTE: Start the loop at depth 1 because the current JVM state does
4223   // not include the Reflection.getCallerClass() frame.
4224   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4225     ciMethod* m = caller_jvms->method();
4226     switch (n) {
4227     case 0:
4228       fatal("current JVM state does not include the Reflection.getCallerClass frame");
4229       break;
4230     case 1:
4231       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4232       if (!m->caller_sensitive()) {
4233 #ifndef PRODUCT
4234         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4235           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
4236         }
4237 #endif
4238         return false;  // bail-out; let JVM_GetCallerClass do the work
4239       }
4240       break;
4241     default:
4242       if (!m->is_ignored_by_security_stack_walk()) {
4243         // We have reached the desired frame; return the holder class.
4244         // Acquire method holder as java.lang.Class and push as constant.
4245         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4246         ciInstance* caller_mirror = caller_klass->java_mirror();
4247         set_result(makecon(TypeInstPtr::make(caller_mirror)));
4248 
4249 #ifndef PRODUCT
4250         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4251           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());
4252           tty->print_cr("  JVM state at this point:");
4253           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4254             ciMethod* m = jvms()->of_depth(i)->method();
4255             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4256           }
4257         }
4258 #endif
4259         return true;
4260       }
4261       break;
4262     }
4263   }
4264 
4265 #ifndef PRODUCT
4266   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4267     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4268     tty->print_cr("  JVM state at this point:");
4269     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4270       ciMethod* m = jvms()->of_depth(i)->method();
4271       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4272     }
4273   }
4274 #endif
4275 
4276   return false;  // bail-out; let JVM_GetCallerClass do the work
4277 }
4278 
4279 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4280   Node* arg = argument(0);
4281   Node* result = NULL;
4282 
4283   switch (id) {
4284   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4285   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4286   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4287   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4288 
4289   case vmIntrinsics::_doubleToLongBits: {
4290     // two paths (plus control) merge in a wood
4291     RegionNode *r = new RegionNode(3);
4292     Node *phi = new PhiNode(r, TypeLong::LONG);
4293 
4294     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4295     // Build the boolean node
4296     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4297 
4298     // Branch either way.
4299     // NaN case is less traveled, which makes all the difference.
4300     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4301     Node *opt_isnan = _gvn.transform(ifisnan);
4302     assert( opt_isnan->is_If(), "Expect an IfNode");
4303     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4304     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4305 
4306     set_control(iftrue);
4307 
4308     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4309     Node *slow_result = longcon(nan_bits); // return NaN
4310     phi->init_req(1, _gvn.transform( slow_result ));
4311     r->init_req(1, iftrue);
4312 
4313     // Else fall through
4314     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4315     set_control(iffalse);
4316 
4317     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4318     r->init_req(2, iffalse);
4319 
4320     // Post merge
4321     set_control(_gvn.transform(r));
4322     record_for_igvn(r);
4323 
4324     C->set_has_split_ifs(true); // Has chance for split-if optimization
4325     result = phi;
4326     assert(result->bottom_type()->isa_long(), "must be");
4327     break;
4328   }
4329 
4330   case vmIntrinsics::_floatToIntBits: {
4331     // two paths (plus control) merge in a wood
4332     RegionNode *r = new RegionNode(3);
4333     Node *phi = new PhiNode(r, TypeInt::INT);
4334 
4335     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4336     // Build the boolean node
4337     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4338 
4339     // Branch either way.
4340     // NaN case is less traveled, which makes all the difference.
4341     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4342     Node *opt_isnan = _gvn.transform(ifisnan);
4343     assert( opt_isnan->is_If(), "Expect an IfNode");
4344     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4345     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4346 
4347     set_control(iftrue);
4348 
4349     static const jint nan_bits = 0x7fc00000;
4350     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4351     phi->init_req(1, _gvn.transform( slow_result ));
4352     r->init_req(1, iftrue);
4353 
4354     // Else fall through
4355     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4356     set_control(iffalse);
4357 
4358     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4359     r->init_req(2, iffalse);
4360 
4361     // Post merge
4362     set_control(_gvn.transform(r));
4363     record_for_igvn(r);
4364 
4365     C->set_has_split_ifs(true); // Has chance for split-if optimization
4366     result = phi;
4367     assert(result->bottom_type()->isa_int(), "must be");
4368     break;
4369   }
4370 
4371   default:
4372     fatal_unexpected_iid(id);
4373     break;
4374   }
4375   set_result(_gvn.transform(result));
4376   return true;
4377 }
4378 
4379 //----------------------inline_unsafe_copyMemory-------------------------
4380 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4381 bool LibraryCallKit::inline_unsafe_copyMemory() {
4382   if (callee()->is_static())  return false;  // caller must have the capability!
4383   null_check_receiver();  // null-check receiver
4384   if (stopped())  return true;
4385 
4386   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4387 
4388   Node* src_ptr =         argument(1);   // type: oop
4389   Node* src_off = ConvL2X(argument(2));  // type: long
4390   Node* dst_ptr =         argument(4);   // type: oop
4391   Node* dst_off = ConvL2X(argument(5));  // type: long
4392   Node* size    = ConvL2X(argument(7));  // type: long
4393 
4394   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4395          "fieldOffset must be byte-scaled");
4396 
4397   Node* src = make_unsafe_address(src_ptr, src_off);
4398   Node* dst = make_unsafe_address(dst_ptr, dst_off);
4399 
4400   // Conservatively insert a memory barrier on all memory slices.
4401   // Do not let writes of the copy source or destination float below the copy.
4402   insert_mem_bar(Op_MemBarCPUOrder);
4403 
4404   // Call it.  Note that the length argument is not scaled.
4405   make_runtime_call(RC_LEAF|RC_NO_FP,
4406                     OptoRuntime::fast_arraycopy_Type(),
4407                     StubRoutines::unsafe_arraycopy(),
4408                     "unsafe_arraycopy",
4409                     TypeRawPtr::BOTTOM,
4410                     src, dst, size XTOP);
4411 
4412   // Do not let reads of the copy destination float above the copy.
4413   insert_mem_bar(Op_MemBarCPUOrder);
4414 
4415   return true;
4416 }
4417 
4418 //------------------------clone_coping-----------------------------------
4419 // Helper function for inline_native_clone.
4420 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4421   assert(obj_size != NULL, "");
4422   Node* raw_obj = alloc_obj->in(1);
4423   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4424 
4425   AllocateNode* alloc = NULL;
4426   if (ReduceBulkZeroing) {
4427     // We will be completely responsible for initializing this object -
4428     // mark Initialize node as complete.
4429     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4430     // The object was just allocated - there should be no any stores!
4431     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4432     // Mark as complete_with_arraycopy so that on AllocateNode
4433     // expansion, we know this AllocateNode is initialized by an array
4434     // copy and a StoreStore barrier exists after the array copy.
4435     alloc->initialization()->set_complete_with_arraycopy();
4436   }
4437 
4438   // Copy the fastest available way.
4439   // TODO: generate fields copies for small objects instead.
4440   Node* src  = obj;
4441   Node* dest = alloc_obj;
4442   Node* size = _gvn.transform(obj_size);
4443 
4444   // Exclude the header but include array length to copy by 8 bytes words.
4445   // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4446   int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4447                             instanceOopDesc::base_offset_in_bytes();
4448   // base_off:
4449   // 8  - 32-bit VM
4450   // 12 - 64-bit VM, compressed klass
4451   // 16 - 64-bit VM, normal klass
4452   if (base_off % BytesPerLong != 0) {
4453     assert(UseCompressedClassPointers, "");
4454     if (is_array) {
4455       // Exclude length to copy by 8 bytes words.
4456       base_off += sizeof(int);
4457     } else {
4458       // Include klass to copy by 8 bytes words.
4459       base_off = instanceOopDesc::klass_offset_in_bytes();
4460     }
4461     assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4462   }
4463   src  = basic_plus_adr(src,  base_off);
4464   dest = basic_plus_adr(dest, base_off);
4465 
4466   // Compute the length also, if needed:
4467   Node* countx = size;
4468   countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4469   countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4470 
4471   const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4472 
4473   ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4474   ac->set_clonebasic();
4475   Node* n = _gvn.transform(ac);
4476   if (n == ac) {
4477     set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4478   } else {
4479     set_all_memory(n);
4480   }
4481 
4482   // If necessary, emit some card marks afterwards.  (Non-arrays only.)
4483   if (card_mark) {
4484     assert(!is_array, "");
4485     // Put in store barrier for any and all oops we are sticking
4486     // into this object.  (We could avoid this if we could prove
4487     // that the object type contains no oop fields at all.)
4488     Node* no_particular_value = NULL;
4489     Node* no_particular_field = NULL;
4490     int raw_adr_idx = Compile::AliasIdxRaw;
4491     post_barrier(control(),
4492                  memory(raw_adr_type),
4493                  alloc_obj,
4494                  no_particular_field,
4495                  raw_adr_idx,
4496                  no_particular_value,
4497                  T_OBJECT,
4498                  false);
4499   }
4500 
4501   // Do not let reads from the cloned object float above the arraycopy.
4502   if (alloc != NULL) {
4503     // Do not let stores that initialize this object be reordered with
4504     // a subsequent store that would make this object accessible by
4505     // other threads.
4506     // Record what AllocateNode this StoreStore protects so that
4507     // escape analysis can go from the MemBarStoreStoreNode to the
4508     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4509     // based on the escape status of the AllocateNode.
4510     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4511   } else {
4512     insert_mem_bar(Op_MemBarCPUOrder);
4513   }
4514 }
4515 
4516 //------------------------inline_native_clone----------------------------
4517 // protected native Object java.lang.Object.clone();
4518 //
4519 // Here are the simple edge cases:
4520 //  null receiver => normal trap
4521 //  virtual and clone was overridden => slow path to out-of-line clone
4522 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4523 //
4524 // The general case has two steps, allocation and copying.
4525 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4526 //
4527 // Copying also has two cases, oop arrays and everything else.
4528 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4529 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4530 //
4531 // These steps fold up nicely if and when the cloned object's klass
4532 // can be sharply typed as an object array, a type array, or an instance.
4533 //
4534 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4535   PhiNode* result_val;
4536 
4537   // Set the reexecute bit for the interpreter to reexecute
4538   // the bytecode that invokes Object.clone if deoptimization happens.
4539   { PreserveReexecuteState preexecs(this);
4540     jvms()->set_should_reexecute(true);
4541 
4542     Node* obj = null_check_receiver();
4543     if (stopped())  return true;
4544 
4545     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4546 
4547     // If we are going to clone an instance, we need its exact type to
4548     // know the number and types of fields to convert the clone to
4549     // loads/stores. Maybe a speculative type can help us.
4550     if (!obj_type->klass_is_exact() &&
4551         obj_type->speculative_type() != NULL &&
4552         obj_type->speculative_type()->is_instance_klass()) {
4553       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4554       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4555           !spec_ik->has_injected_fields()) {
4556         ciKlass* k = obj_type->klass();
4557         if (!k->is_instance_klass() ||
4558             k->as_instance_klass()->is_interface() ||
4559             k->as_instance_klass()->has_subklass()) {
4560           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4561         }
4562       }
4563     }
4564 
4565     Node* obj_klass = load_object_klass(obj);
4566     const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4567     const TypeOopPtr*   toop   = ((tklass != NULL)
4568                                 ? tklass->as_instance_type()
4569                                 : TypeInstPtr::NOTNULL);
4570 
4571     // Conservatively insert a memory barrier on all memory slices.
4572     // Do not let writes into the original float below the clone.
4573     insert_mem_bar(Op_MemBarCPUOrder);
4574 
4575     // paths into result_reg:
4576     enum {
4577       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4578       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4579       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4580       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4581       PATH_LIMIT
4582     };
4583     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4584     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4585     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4586     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4587     record_for_igvn(result_reg);
4588 
4589     const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4590     int raw_adr_idx = Compile::AliasIdxRaw;
4591 
4592     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4593     if (array_ctl != NULL) {
4594       // It's an array.
4595       PreserveJVMState pjvms(this);
4596       set_control(array_ctl);
4597       Node* obj_length = load_array_length(obj);
4598       Node* obj_size  = NULL;
4599       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4600 
4601       if (!use_ReduceInitialCardMarks()) {
4602         // If it is an oop array, it requires very special treatment,
4603         // because card marking is required on each card of the array.
4604         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4605         if (is_obja != NULL) {
4606           PreserveJVMState pjvms2(this);
4607           set_control(is_obja);
4608           // Generate a direct call to the right arraycopy function(s).
4609           Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4610           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4611           ac->set_cloneoop();
4612           Node* n = _gvn.transform(ac);
4613           assert(n == ac, "cannot disappear");
4614           ac->connect_outputs(this);
4615 
4616           result_reg->init_req(_objArray_path, control());
4617           result_val->init_req(_objArray_path, alloc_obj);
4618           result_i_o ->set_req(_objArray_path, i_o());
4619           result_mem ->set_req(_objArray_path, reset_memory());
4620         }
4621       }
4622       // Otherwise, there are no card marks to worry about.
4623       // (We can dispense with card marks if we know the allocation
4624       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4625       //  causes the non-eden paths to take compensating steps to
4626       //  simulate a fresh allocation, so that no further
4627       //  card marks are required in compiled code to initialize
4628       //  the object.)
4629 
4630       if (!stopped()) {
4631         copy_to_clone(obj, alloc_obj, obj_size, true, false);
4632 
4633         // Present the results of the copy.
4634         result_reg->init_req(_array_path, control());
4635         result_val->init_req(_array_path, alloc_obj);
4636         result_i_o ->set_req(_array_path, i_o());
4637         result_mem ->set_req(_array_path, reset_memory());
4638       }
4639     }
4640 
4641     // We only go to the instance fast case code if we pass a number of guards.
4642     // The paths which do not pass are accumulated in the slow_region.
4643     RegionNode* slow_region = new RegionNode(1);
4644     record_for_igvn(slow_region);
4645     if (!stopped()) {
4646       // It's an instance (we did array above).  Make the slow-path tests.
4647       // If this is a virtual call, we generate a funny guard.  We grab
4648       // the vtable entry corresponding to clone() from the target object.
4649       // If the target method which we are calling happens to be the
4650       // Object clone() method, we pass the guard.  We do not need this
4651       // guard for non-virtual calls; the caller is known to be the native
4652       // Object clone().
4653       if (is_virtual) {
4654         generate_virtual_guard(obj_klass, slow_region);
4655       }
4656 
4657       // The object must be easily cloneable and must not have a finalizer.
4658       // Both of these conditions may be checked in a single test.
4659       // We could optimize the test further, but we don't care.
4660       generate_access_flags_guard(obj_klass,
4661                                   // Test both conditions:
4662                                   JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4663                                   // Must be cloneable but not finalizer:
4664                                   JVM_ACC_IS_CLONEABLE_FAST,
4665                                   slow_region);
4666     }
4667 
4668     if (!stopped()) {
4669       // It's an instance, and it passed the slow-path tests.
4670       PreserveJVMState pjvms(this);
4671       Node* obj_size  = NULL;
4672       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4673       // is reexecuted if deoptimization occurs and there could be problems when merging
4674       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4675       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4676 
4677       copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4678 
4679       // Present the results of the slow call.
4680       result_reg->init_req(_instance_path, control());
4681       result_val->init_req(_instance_path, alloc_obj);
4682       result_i_o ->set_req(_instance_path, i_o());
4683       result_mem ->set_req(_instance_path, reset_memory());
4684     }
4685 
4686     // Generate code for the slow case.  We make a call to clone().
4687     set_control(_gvn.transform(slow_region));
4688     if (!stopped()) {
4689       PreserveJVMState pjvms(this);
4690       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4691       Node* slow_result = set_results_for_java_call(slow_call);
4692       // this->control() comes from set_results_for_java_call
4693       result_reg->init_req(_slow_path, control());
4694       result_val->init_req(_slow_path, slow_result);
4695       result_i_o ->set_req(_slow_path, i_o());
4696       result_mem ->set_req(_slow_path, reset_memory());
4697     }
4698 
4699     // Return the combined state.
4700     set_control(    _gvn.transform(result_reg));
4701     set_i_o(        _gvn.transform(result_i_o));
4702     set_all_memory( _gvn.transform(result_mem));
4703   } // original reexecute is set back here
4704 
4705   set_result(_gvn.transform(result_val));
4706   return true;
4707 }
4708 
4709 // If we have a tighly coupled allocation, the arraycopy may take care
4710 // of the array initialization. If one of the guards we insert between
4711 // the allocation and the arraycopy causes a deoptimization, an
4712 // unitialized array will escape the compiled method. To prevent that
4713 // we set the JVM state for uncommon traps between the allocation and
4714 // the arraycopy to the state before the allocation so, in case of
4715 // deoptimization, we'll reexecute the allocation and the
4716 // initialization.
4717 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4718   if (alloc != NULL) {
4719     ciMethod* trap_method = alloc->jvms()->method();
4720     int trap_bci = alloc->jvms()->bci();
4721 
4722     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4723           !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4724       // Make sure there's no store between the allocation and the
4725       // arraycopy otherwise visible side effects could be rexecuted
4726       // in case of deoptimization and cause incorrect execution.
4727       bool no_interfering_store = true;
4728       Node* mem = alloc->in(TypeFunc::Memory);
4729       if (mem->is_MergeMem()) {
4730         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4731           Node* n = mms.memory();
4732           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4733             assert(n->is_Store(), "what else?");
4734             no_interfering_store = false;
4735             break;
4736           }
4737         }
4738       } else {
4739         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4740           Node* n = mms.memory();
4741           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4742             assert(n->is_Store(), "what else?");
4743             no_interfering_store = false;
4744             break;
4745           }
4746         }
4747       }
4748 
4749       if (no_interfering_store) {
4750         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4751         uint size = alloc->req();
4752         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4753         old_jvms->set_map(sfpt);
4754         for (uint i = 0; i < size; i++) {
4755           sfpt->init_req(i, alloc->in(i));
4756         }
4757         // re-push array length for deoptimization
4758         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4759         old_jvms->set_sp(old_jvms->sp()+1);
4760         old_jvms->set_monoff(old_jvms->monoff()+1);
4761         old_jvms->set_scloff(old_jvms->scloff()+1);
4762         old_jvms->set_endoff(old_jvms->endoff()+1);
4763         old_jvms->set_should_reexecute(true);
4764 
4765         sfpt->set_i_o(map()->i_o());
4766         sfpt->set_memory(map()->memory());
4767         sfpt->set_control(map()->control());
4768 
4769         JVMState* saved_jvms = jvms();
4770         saved_reexecute_sp = _reexecute_sp;
4771 
4772         set_jvms(sfpt->jvms());
4773         _reexecute_sp = jvms()->sp();
4774 
4775         return saved_jvms;
4776       }
4777     }
4778   }
4779   return NULL;
4780 }
4781 
4782 // In case of a deoptimization, we restart execution at the
4783 // allocation, allocating a new array. We would leave an uninitialized
4784 // array in the heap that GCs wouldn't expect. Move the allocation
4785 // after the traps so we don't allocate the array if we
4786 // deoptimize. This is possible because tightly_coupled_allocation()
4787 // guarantees there's no observer of the allocated array at this point
4788 // and the control flow is simple enough.
4789 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4790   if (saved_jvms != NULL && !stopped()) {
4791     assert(alloc != NULL, "only with a tightly coupled allocation");
4792     // restore JVM state to the state at the arraycopy
4793     saved_jvms->map()->set_control(map()->control());
4794     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4795     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4796     // If we've improved the types of some nodes (null check) while
4797     // emitting the guards, propagate them to the current state
4798     map()->replaced_nodes().apply(saved_jvms->map());
4799     set_jvms(saved_jvms);
4800     _reexecute_sp = saved_reexecute_sp;
4801 
4802     // Remove the allocation from above the guards
4803     CallProjections callprojs;
4804     alloc->extract_projections(&callprojs, true);
4805     InitializeNode* init = alloc->initialization();
4806     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4807     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4808     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4809     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4810 
4811     // move the allocation here (after the guards)
4812     _gvn.hash_delete(alloc);
4813     alloc->set_req(TypeFunc::Control, control());
4814     alloc->set_req(TypeFunc::I_O, i_o());
4815     Node *mem = reset_memory();
4816     set_all_memory(mem);
4817     alloc->set_req(TypeFunc::Memory, mem);
4818     set_control(init->proj_out(TypeFunc::Control));
4819     set_i_o(callprojs.fallthrough_ioproj);
4820 
4821     // Update memory as done in GraphKit::set_output_for_allocation()
4822     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4823     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4824     if (ary_type->isa_aryptr() && length_type != NULL) {
4825       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4826     }
4827     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4828     int            elemidx  = C->get_alias_index(telemref);
4829     set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4830     set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4831 
4832     Node* allocx = _gvn.transform(alloc);
4833     assert(allocx == alloc, "where has the allocation gone?");
4834     assert(dest->is_CheckCastPP(), "not an allocation result?");
4835 
4836     _gvn.hash_delete(dest);
4837     dest->set_req(0, control());
4838     Node* destx = _gvn.transform(dest);
4839     assert(destx == dest, "where has the allocation result gone?");
4840   }
4841 }
4842 
4843 
4844 //------------------------------inline_arraycopy-----------------------
4845 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4846 //                                                      Object dest, int destPos,
4847 //                                                      int length);
4848 bool LibraryCallKit::inline_arraycopy() {
4849   // Get the arguments.
4850   Node* src         = argument(0);  // type: oop
4851   Node* src_offset  = argument(1);  // type: int
4852   Node* dest        = argument(2);  // type: oop
4853   Node* dest_offset = argument(3);  // type: int
4854   Node* length      = argument(4);  // type: int
4855 
4856 
4857   // Check for allocation before we add nodes that would confuse
4858   // tightly_coupled_allocation()
4859   AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4860 
4861   int saved_reexecute_sp = -1;
4862   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4863   // See arraycopy_restore_alloc_state() comment
4864   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4865   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4866   // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4867   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4868 
4869   // The following tests must be performed
4870   // (1) src and dest are arrays.
4871   // (2) src and dest arrays must have elements of the same BasicType
4872   // (3) src and dest must not be null.
4873   // (4) src_offset must not be negative.
4874   // (5) dest_offset must not be negative.
4875   // (6) length must not be negative.
4876   // (7) src_offset + length must not exceed length of src.
4877   // (8) dest_offset + length must not exceed length of dest.
4878   // (9) each element of an oop array must be assignable
4879 
4880   // (3) src and dest must not be null.
4881   // always do this here because we need the JVM state for uncommon traps
4882   Node* null_ctl = top();
4883   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4884   assert(null_ctl->is_top(), "no null control here");
4885   dest = null_check(dest, T_ARRAY);
4886 
4887   if (!can_emit_guards) {
4888     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4889     // guards but the arraycopy node could still take advantage of a
4890     // tightly allocated allocation. tightly_coupled_allocation() is
4891     // called again to make sure it takes the null check above into
4892     // account: the null check is mandatory and if it caused an
4893     // uncommon trap to be emitted then the allocation can't be
4894     // considered tightly coupled in this context.
4895     alloc = tightly_coupled_allocation(dest, NULL);
4896   }
4897 
4898   bool validated = false;
4899 
4900   const Type* src_type  = _gvn.type(src);
4901   const Type* dest_type = _gvn.type(dest);
4902   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4903   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4904 
4905   // Do we have the type of src?
4906   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4907   // Do we have the type of dest?
4908   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4909   // Is the type for src from speculation?
4910   bool src_spec = false;
4911   // Is the type for dest from speculation?
4912   bool dest_spec = false;
4913 
4914   if ((!has_src || !has_dest) && can_emit_guards) {
4915     // We don't have sufficient type information, let's see if
4916     // speculative types can help. We need to have types for both src
4917     // and dest so that it pays off.
4918 
4919     // Do we already have or could we have type information for src
4920     bool could_have_src = has_src;
4921     // Do we already have or could we have type information for dest
4922     bool could_have_dest = has_dest;
4923 
4924     ciKlass* src_k = NULL;
4925     if (!has_src) {
4926       src_k = src_type->speculative_type_not_null();
4927       if (src_k != NULL && src_k->is_array_klass()) {
4928         could_have_src = true;
4929       }
4930     }
4931 
4932     ciKlass* dest_k = NULL;
4933     if (!has_dest) {
4934       dest_k = dest_type->speculative_type_not_null();
4935       if (dest_k != NULL && dest_k->is_array_klass()) {
4936         could_have_dest = true;
4937       }
4938     }
4939 
4940     if (could_have_src && could_have_dest) {
4941       // This is going to pay off so emit the required guards
4942       if (!has_src) {
4943         src = maybe_cast_profiled_obj(src, src_k, true);
4944         src_type  = _gvn.type(src);
4945         top_src  = src_type->isa_aryptr();
4946         has_src = (top_src != NULL && top_src->klass() != NULL);
4947         src_spec = true;
4948       }
4949       if (!has_dest) {
4950         dest = maybe_cast_profiled_obj(dest, dest_k, true);
4951         dest_type  = _gvn.type(dest);
4952         top_dest  = dest_type->isa_aryptr();
4953         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4954         dest_spec = true;
4955       }
4956     }
4957   }
4958 
4959   if (has_src && has_dest && can_emit_guards) {
4960     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
4961     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4962     if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
4963     if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;
4964 
4965     if (src_elem == dest_elem && src_elem == T_OBJECT) {
4966       // If both arrays are object arrays then having the exact types
4967       // for both will remove the need for a subtype check at runtime
4968       // before the call and may make it possible to pick a faster copy
4969       // routine (without a subtype check on every element)
4970       // Do we have the exact type of src?
4971       bool could_have_src = src_spec;
4972       // Do we have the exact type of dest?
4973       bool could_have_dest = dest_spec;
4974       ciKlass* src_k = top_src->klass();
4975       ciKlass* dest_k = top_dest->klass();
4976       if (!src_spec) {
4977         src_k = src_type->speculative_type_not_null();
4978         if (src_k != NULL && src_k->is_array_klass()) {
4979           could_have_src = true;
4980         }
4981       }
4982       if (!dest_spec) {
4983         dest_k = dest_type->speculative_type_not_null();
4984         if (dest_k != NULL && dest_k->is_array_klass()) {
4985           could_have_dest = true;
4986         }
4987       }
4988       if (could_have_src && could_have_dest) {
4989         // If we can have both exact types, emit the missing guards
4990         if (could_have_src && !src_spec) {
4991           src = maybe_cast_profiled_obj(src, src_k, true);
4992         }
4993         if (could_have_dest && !dest_spec) {
4994           dest = maybe_cast_profiled_obj(dest, dest_k, true);
4995         }
4996       }
4997     }
4998   }
4999 
5000   ciMethod* trap_method = method();
5001   int trap_bci = bci();
5002   if (saved_jvms != NULL) {
5003     trap_method = alloc->jvms()->method();
5004     trap_bci = alloc->jvms()->bci();
5005   }
5006 
5007   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5008       can_emit_guards &&
5009       !src->is_top() && !dest->is_top()) {
5010     // validate arguments: enables transformation the ArrayCopyNode
5011     validated = true;
5012 
5013     RegionNode* slow_region = new RegionNode(1);
5014     record_for_igvn(slow_region);
5015 
5016     // (1) src and dest are arrays.
5017     generate_non_array_guard(load_object_klass(src), slow_region);
5018     generate_non_array_guard(load_object_klass(dest), slow_region);
5019 
5020     // (2) src and dest arrays must have elements of the same BasicType
5021     // done at macro expansion or at Ideal transformation time
5022 
5023     // (4) src_offset must not be negative.
5024     generate_negative_guard(src_offset, slow_region);
5025 
5026     // (5) dest_offset must not be negative.
5027     generate_negative_guard(dest_offset, slow_region);
5028 
5029     // (7) src_offset + length must not exceed length of src.
5030     generate_limit_guard(src_offset, length,
5031                          load_array_length(src),
5032                          slow_region);
5033 
5034     // (8) dest_offset + length must not exceed length of dest.
5035     generate_limit_guard(dest_offset, length,
5036                          load_array_length(dest),
5037                          slow_region);
5038 
5039     // (9) each element of an oop array must be assignable
5040     Node* src_klass  = load_object_klass(src);
5041     Node* dest_klass = load_object_klass(dest);
5042     Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5043 
5044     if (not_subtype_ctrl != top()) {
5045       PreserveJVMState pjvms(this);
5046       set_control(not_subtype_ctrl);
5047       uncommon_trap(Deoptimization::Reason_intrinsic,
5048                     Deoptimization::Action_make_not_entrant);
5049       assert(stopped(), "Should be stopped");
5050     }
5051     {
5052       PreserveJVMState pjvms(this);
5053       set_control(_gvn.transform(slow_region));
5054       uncommon_trap(Deoptimization::Reason_intrinsic,
5055                     Deoptimization::Action_make_not_entrant);
5056       assert(stopped(), "Should be stopped");
5057     }
5058   }
5059 
5060   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5061 
5062   if (stopped()) {
5063     return true;
5064   }
5065 
5066   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5067                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
5068                                           // so the compiler has a chance to eliminate them: during macro expansion,
5069                                           // we have to set their control (CastPP nodes are eliminated).
5070                                           load_object_klass(src), load_object_klass(dest),
5071                                           load_array_length(src), load_array_length(dest));
5072 
5073   ac->set_arraycopy(validated);
5074 
5075   Node* n = _gvn.transform(ac);
5076   if (n == ac) {
5077     ac->connect_outputs(this);
5078   } else {
5079     assert(validated, "shouldn't transform if all arguments not validated");
5080     set_all_memory(n);
5081   }
5082 
5083   return true;
5084 }
5085 
5086 
5087 // Helper function which determines if an arraycopy immediately follows
5088 // an allocation, with no intervening tests or other escapes for the object.
5089 AllocateArrayNode*
5090 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5091                                            RegionNode* slow_region) {
5092   if (stopped())             return NULL;  // no fast path
5093   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
5094 
5095   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5096   if (alloc == NULL)  return NULL;
5097 
5098   Node* rawmem = memory(Compile::AliasIdxRaw);
5099   // Is the allocation's memory state untouched?
5100   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5101     // Bail out if there have been raw-memory effects since the allocation.
5102     // (Example:  There might have been a call or safepoint.)
5103     return NULL;
5104   }
5105   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5106   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5107     return NULL;
5108   }
5109 
5110   // There must be no unexpected observers of this allocation.
5111   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5112     Node* obs = ptr->fast_out(i);
5113     if (obs != this->map()) {
5114       return NULL;
5115     }
5116   }
5117 
5118   // This arraycopy must unconditionally follow the allocation of the ptr.
5119   Node* alloc_ctl = ptr->in(0);
5120   assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5121 
5122   Node* ctl = control();
5123   while (ctl != alloc_ctl) {
5124     // There may be guards which feed into the slow_region.
5125     // Any other control flow means that we might not get a chance
5126     // to finish initializing the allocated object.
5127     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5128       IfNode* iff = ctl->in(0)->as_If();
5129       Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5130       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5131       if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5132         ctl = iff->in(0);       // This test feeds the known slow_region.
5133         continue;
5134       }
5135       // One more try:  Various low-level checks bottom out in
5136       // uncommon traps.  If the debug-info of the trap omits
5137       // any reference to the allocation, as we've already
5138       // observed, then there can be no objection to the trap.
5139       bool found_trap = false;
5140       for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5141         Node* obs = not_ctl->fast_out(j);
5142         if (obs->in(0) == not_ctl && obs->is_Call() &&
5143             (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5144           found_trap = true; break;
5145         }
5146       }
5147       if (found_trap) {
5148         ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
5149         continue;
5150       }
5151     }
5152     return NULL;
5153   }
5154 
5155   // If we get this far, we have an allocation which immediately
5156   // precedes the arraycopy, and we can take over zeroing the new object.
5157   // The arraycopy will finish the initialization, and provide
5158   // a new control state to which we will anchor the destination pointer.
5159 
5160   return alloc;
5161 }
5162 
5163 //-------------inline_encodeISOArray-----------------------------------
5164 // encode char[] to byte[] in ISO_8859_1
5165 bool LibraryCallKit::inline_encodeISOArray() {
5166   assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5167   // no receiver since it is static method
5168   Node *src         = argument(0);
5169   Node *src_offset  = argument(1);
5170   Node *dst         = argument(2);
5171   Node *dst_offset  = argument(3);
5172   Node *length      = argument(4);
5173 
5174   const Type* src_type = src->Value(&_gvn);
5175   const Type* dst_type = dst->Value(&_gvn);
5176   const TypeAryPtr* top_src = src_type->isa_aryptr();
5177   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5178   if (top_src  == NULL || top_src->klass()  == NULL ||
5179       top_dest == NULL || top_dest->klass() == NULL) {
5180     // failed array check
5181     return false;
5182   }
5183 
5184   // Figure out the size and type of the elements we will be copying.
5185   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5186   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5187   if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5188     return false;
5189   }
5190 
5191   Node* src_start = array_element_address(src, src_offset, T_CHAR);
5192   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5193   // 'src_start' points to src array + scaled offset
5194   // 'dst_start' points to dst array + scaled offset
5195 
5196   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5197   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5198   enc = _gvn.transform(enc);
5199   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5200   set_memory(res_mem, mtype);
5201   set_result(enc);
5202   return true;
5203 }
5204 
5205 //-------------inline_multiplyToLen-----------------------------------
5206 bool LibraryCallKit::inline_multiplyToLen() {
5207   assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5208 
5209   address stubAddr = StubRoutines::multiplyToLen();
5210   if (stubAddr == NULL) {
5211     return false; // Intrinsic's stub is not implemented on this platform
5212   }
5213   const char* stubName = "multiplyToLen";
5214 
5215   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5216 
5217   // no receiver because it is a static method
5218   Node* x    = argument(0);
5219   Node* xlen = argument(1);
5220   Node* y    = argument(2);
5221   Node* ylen = argument(3);
5222   Node* z    = argument(4);
5223 
5224   const Type* x_type = x->Value(&_gvn);
5225   const Type* y_type = y->Value(&_gvn);
5226   const TypeAryPtr* top_x = x_type->isa_aryptr();
5227   const TypeAryPtr* top_y = y_type->isa_aryptr();
5228   if (top_x  == NULL || top_x->klass()  == NULL ||
5229       top_y == NULL || top_y->klass() == NULL) {
5230     // failed array check
5231     return false;
5232   }
5233 
5234   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5235   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5236   if (x_elem != T_INT || y_elem != T_INT) {
5237     return false;
5238   }
5239 
5240   // Set the original stack and the reexecute bit for the interpreter to reexecute
5241   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5242   // on the return from z array allocation in runtime.
5243   { PreserveReexecuteState preexecs(this);
5244     jvms()->set_should_reexecute(true);
5245 
5246     Node* x_start = array_element_address(x, intcon(0), x_elem);
5247     Node* y_start = array_element_address(y, intcon(0), y_elem);
5248     // 'x_start' points to x array + scaled xlen
5249     // 'y_start' points to y array + scaled ylen
5250 
5251     // Allocate the result array
5252     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5253     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5254     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5255 
5256     IdealKit ideal(this);
5257 
5258 #define __ ideal.
5259      Node* one = __ ConI(1);
5260      Node* zero = __ ConI(0);
5261      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5262      __ set(need_alloc, zero);
5263      __ set(z_alloc, z);
5264      __ if_then(z, BoolTest::eq, null()); {
5265        __ increment (need_alloc, one);
5266      } __ else_(); {
5267        // Update graphKit memory and control from IdealKit.
5268        sync_kit(ideal);
5269        Node* zlen_arg = load_array_length(z);
5270        // Update IdealKit memory and control from graphKit.
5271        __ sync_kit(this);
5272        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5273          __ increment (need_alloc, one);
5274        } __ end_if();
5275      } __ end_if();
5276 
5277      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5278        // Update graphKit memory and control from IdealKit.
5279        sync_kit(ideal);
5280        Node * narr = new_array(klass_node, zlen, 1);
5281        // Update IdealKit memory and control from graphKit.
5282        __ sync_kit(this);
5283        __ set(z_alloc, narr);
5284      } __ end_if();
5285 
5286      sync_kit(ideal);
5287      z = __ value(z_alloc);
5288      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5289      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5290      // Final sync IdealKit and GraphKit.
5291      final_sync(ideal);
5292 #undef __
5293 
5294     Node* z_start = array_element_address(z, intcon(0), T_INT);
5295 
5296     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5297                                    OptoRuntime::multiplyToLen_Type(),
5298                                    stubAddr, stubName, TypePtr::BOTTOM,
5299                                    x_start, xlen, y_start, ylen, z_start, zlen);
5300   } // original reexecute is set back here
5301 
5302   C->set_has_split_ifs(true); // Has chance for split-if optimization
5303   set_result(z);
5304   return true;
5305 }
5306 
5307 //-------------inline_squareToLen------------------------------------
5308 bool LibraryCallKit::inline_squareToLen() {
5309   assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5310 
5311   address stubAddr = StubRoutines::squareToLen();
5312   if (stubAddr == NULL) {
5313     return false; // Intrinsic's stub is not implemented on this platform
5314   }
5315   const char* stubName = "squareToLen";
5316 
5317   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5318 
5319   Node* x    = argument(0);
5320   Node* len  = argument(1);
5321   Node* z    = argument(2);
5322   Node* zlen = argument(3);
5323 
5324   const Type* x_type = x->Value(&_gvn);
5325   const Type* z_type = z->Value(&_gvn);
5326   const TypeAryPtr* top_x = x_type->isa_aryptr();
5327   const TypeAryPtr* top_z = z_type->isa_aryptr();
5328   if (top_x  == NULL || top_x->klass()  == NULL ||
5329       top_z  == NULL || top_z->klass()  == NULL) {
5330     // failed array check
5331     return false;
5332   }
5333 
5334   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5335   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5336   if (x_elem != T_INT || z_elem != T_INT) {
5337     return false;
5338   }
5339 
5340 
5341   Node* x_start = array_element_address(x, intcon(0), x_elem);
5342   Node* z_start = array_element_address(z, intcon(0), z_elem);
5343 
5344   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5345                                   OptoRuntime::squareToLen_Type(),
5346                                   stubAddr, stubName, TypePtr::BOTTOM,
5347                                   x_start, len, z_start, zlen);
5348 
5349   set_result(z);
5350   return true;
5351 }
5352 
5353 //-------------inline_mulAdd------------------------------------------
5354 bool LibraryCallKit::inline_mulAdd() {
5355   assert(UseMulAddIntrinsic, "not implemented on this platform");
5356 
5357   address stubAddr = StubRoutines::mulAdd();
5358   if (stubAddr == NULL) {
5359     return false; // Intrinsic's stub is not implemented on this platform
5360   }
5361   const char* stubName = "mulAdd";
5362 
5363   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5364 
5365   Node* out      = argument(0);
5366   Node* in       = argument(1);
5367   Node* offset   = argument(2);
5368   Node* len      = argument(3);
5369   Node* k        = argument(4);
5370 
5371   const Type* out_type = out->Value(&_gvn);
5372   const Type* in_type = in->Value(&_gvn);
5373   const TypeAryPtr* top_out = out_type->isa_aryptr();
5374   const TypeAryPtr* top_in = in_type->isa_aryptr();
5375   if (top_out  == NULL || top_out->klass()  == NULL ||
5376       top_in == NULL || top_in->klass() == NULL) {
5377     // failed array check
5378     return false;
5379   }
5380 
5381   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5382   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5383   if (out_elem != T_INT || in_elem != T_INT) {
5384     return false;
5385   }
5386 
5387   Node* outlen = load_array_length(out);
5388   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5389   Node* out_start = array_element_address(out, intcon(0), out_elem);
5390   Node* in_start = array_element_address(in, intcon(0), in_elem);
5391 
5392   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5393                                   OptoRuntime::mulAdd_Type(),
5394                                   stubAddr, stubName, TypePtr::BOTTOM,
5395                                   out_start,in_start, new_offset, len, k);
5396   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5397   set_result(result);
5398   return true;
5399 }
5400 
5401 //-------------inline_montgomeryMultiply-----------------------------------
5402 bool LibraryCallKit::inline_montgomeryMultiply() {
5403   address stubAddr = StubRoutines::montgomeryMultiply();
5404   if (stubAddr == NULL) {
5405     return false; // Intrinsic's stub is not implemented on this platform
5406   }
5407 
5408   assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5409   const char* stubName = "montgomery_square";
5410 
5411   assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5412 
5413   Node* a    = argument(0);
5414   Node* b    = argument(1);
5415   Node* n    = argument(2);
5416   Node* len  = argument(3);
5417   Node* inv  = argument(4);
5418   Node* m    = argument(6);
5419 
5420   const Type* a_type = a->Value(&_gvn);
5421   const TypeAryPtr* top_a = a_type->isa_aryptr();
5422   const Type* b_type = b->Value(&_gvn);
5423   const TypeAryPtr* top_b = b_type->isa_aryptr();
5424   const Type* n_type = a->Value(&_gvn);
5425   const TypeAryPtr* top_n = n_type->isa_aryptr();
5426   const Type* m_type = a->Value(&_gvn);
5427   const TypeAryPtr* top_m = m_type->isa_aryptr();
5428   if (top_a  == NULL || top_a->klass()  == NULL ||
5429       top_b == NULL || top_b->klass()  == NULL ||
5430       top_n == NULL || top_n->klass()  == NULL ||
5431       top_m == NULL || top_m->klass()  == NULL) {
5432     // failed array check
5433     return false;
5434   }
5435 
5436   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5437   BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5438   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5439   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5440   if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5441     return false;
5442   }
5443 
5444   // Make the call
5445   {
5446     Node* a_start = array_element_address(a, intcon(0), a_elem);
5447     Node* b_start = array_element_address(b, intcon(0), b_elem);
5448     Node* n_start = array_element_address(n, intcon(0), n_elem);
5449     Node* m_start = array_element_address(m, intcon(0), m_elem);
5450 
5451     Node* call = make_runtime_call(RC_LEAF,
5452                                    OptoRuntime::montgomeryMultiply_Type(),
5453                                    stubAddr, stubName, TypePtr::BOTTOM,
5454                                    a_start, b_start, n_start, len, inv, top(),
5455                                    m_start);
5456     set_result(m);
5457   }
5458 
5459   return true;
5460 }
5461 
5462 bool LibraryCallKit::inline_montgomerySquare() {
5463   address stubAddr = StubRoutines::montgomerySquare();
5464   if (stubAddr == NULL) {
5465     return false; // Intrinsic's stub is not implemented on this platform
5466   }
5467 
5468   assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5469   const char* stubName = "montgomery_square";
5470 
5471   assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5472 
5473   Node* a    = argument(0);
5474   Node* n    = argument(1);
5475   Node* len  = argument(2);
5476   Node* inv  = argument(3);
5477   Node* m    = argument(5);
5478 
5479   const Type* a_type = a->Value(&_gvn);
5480   const TypeAryPtr* top_a = a_type->isa_aryptr();
5481   const Type* n_type = a->Value(&_gvn);
5482   const TypeAryPtr* top_n = n_type->isa_aryptr();
5483   const Type* m_type = a->Value(&_gvn);
5484   const TypeAryPtr* top_m = m_type->isa_aryptr();
5485   if (top_a  == NULL || top_a->klass()  == NULL ||
5486       top_n == NULL || top_n->klass()  == NULL ||
5487       top_m == NULL || top_m->klass()  == NULL) {
5488     // failed array check
5489     return false;
5490   }
5491 
5492   BasicType a_elem = a_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 || 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* n_start = array_element_address(n, intcon(0), n_elem);
5503     Node* m_start = array_element_address(m, intcon(0), m_elem);
5504 
5505     Node* call = make_runtime_call(RC_LEAF,
5506                                    OptoRuntime::montgomerySquare_Type(),
5507                                    stubAddr, stubName, TypePtr::BOTTOM,
5508                                    a_start, n_start, len, inv, top(),
5509                                    m_start);
5510     set_result(m);
5511   }
5512 
5513   return true;
5514 }
5515 
5516 //-------------inline_vectorizedMismatch------------------------------
5517 bool LibraryCallKit::inline_vectorizedMismatch() {
5518   assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5519 
5520   address stubAddr = StubRoutines::vectorizedMismatch();
5521   if (stubAddr == NULL) {
5522     return false; // Intrinsic's stub is not implemented on this platform
5523   }
5524   const char* stubName = "vectorizedMismatch";
5525   int size_l = callee()->signature()->size();
5526   assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5527 
5528   Node* obja = argument(0);
5529   Node* aoffset = argument(1);
5530   Node* objb = argument(3);
5531   Node* boffset = argument(4);
5532   Node* length = argument(6);
5533   Node* scale = argument(7);
5534 
5535   const Type* a_type = obja->Value(&_gvn);
5536   const Type* b_type = objb->Value(&_gvn);
5537   const TypeAryPtr* top_a = a_type->isa_aryptr();
5538   const TypeAryPtr* top_b = b_type->isa_aryptr();
5539   if (top_a == NULL || top_a->klass() == NULL ||
5540     top_b == NULL || top_b->klass() == NULL) {
5541     // failed array check
5542     return false;
5543   }
5544 
5545   Node* call;
5546   jvms()->set_should_reexecute(true);
5547 
5548   Node* obja_adr = make_unsafe_address(obja, aoffset);
5549   Node* objb_adr = make_unsafe_address(objb, boffset);
5550 
5551   call = make_runtime_call(RC_LEAF,
5552     OptoRuntime::vectorizedMismatch_Type(),
5553     stubAddr, stubName, TypePtr::BOTTOM,
5554     obja_adr, objb_adr, length, scale);
5555 
5556   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5557   set_result(result);
5558   return true;
5559 }
5560 
5561 /**
5562  * Calculate CRC32 for byte.
5563  * int java.util.zip.CRC32.update(int crc, int b)
5564  */
5565 bool LibraryCallKit::inline_updateCRC32() {
5566   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5567   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5568   // no receiver since it is static method
5569   Node* crc  = argument(0); // type: int
5570   Node* b    = argument(1); // type: int
5571 
5572   /*
5573    *    int c = ~ crc;
5574    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5575    *    b = b ^ (c >>> 8);
5576    *    crc = ~b;
5577    */
5578 
5579   Node* M1 = intcon(-1);
5580   crc = _gvn.transform(new XorINode(crc, M1));
5581   Node* result = _gvn.transform(new XorINode(crc, b));
5582   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5583 
5584   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5585   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5586   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5587   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5588 
5589   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5590   result = _gvn.transform(new XorINode(crc, result));
5591   result = _gvn.transform(new XorINode(result, M1));
5592   set_result(result);
5593   return true;
5594 }
5595 
5596 /**
5597  * Calculate CRC32 for byte[] array.
5598  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5599  */
5600 bool LibraryCallKit::inline_updateBytesCRC32() {
5601   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5602   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5603   // no receiver since it is static method
5604   Node* crc     = argument(0); // type: int
5605   Node* src     = argument(1); // type: oop
5606   Node* offset  = argument(2); // type: int
5607   Node* length  = argument(3); // type: int
5608 
5609   const Type* src_type = src->Value(&_gvn);
5610   const TypeAryPtr* top_src = src_type->isa_aryptr();
5611   if (top_src  == NULL || top_src->klass()  == NULL) {
5612     // failed array check
5613     return false;
5614   }
5615 
5616   // Figure out the size and type of the elements we will be copying.
5617   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5618   if (src_elem != T_BYTE) {
5619     return false;
5620   }
5621 
5622   // 'src_start' points to src array + scaled offset
5623   Node* src_start = array_element_address(src, offset, src_elem);
5624 
5625   // We assume that range check is done by caller.
5626   // TODO: generate range check (offset+length < src.length) in debug VM.
5627 
5628   // Call the stub.
5629   address stubAddr = StubRoutines::updateBytesCRC32();
5630   const char *stubName = "updateBytesCRC32";
5631 
5632   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5633                                  stubAddr, stubName, TypePtr::BOTTOM,
5634                                  crc, src_start, length);
5635   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5636   set_result(result);
5637   return true;
5638 }
5639 
5640 /**
5641  * Calculate CRC32 for ByteBuffer.
5642  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5643  */
5644 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5645   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5646   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5647   // no receiver since it is static method
5648   Node* crc     = argument(0); // type: int
5649   Node* src     = argument(1); // type: long
5650   Node* offset  = argument(3); // type: int
5651   Node* length  = argument(4); // type: int
5652 
5653   src = ConvL2X(src);  // adjust Java long to machine word
5654   Node* base = _gvn.transform(new CastX2PNode(src));
5655   offset = ConvI2X(offset);
5656 
5657   // 'src_start' points to src array + scaled offset
5658   Node* src_start = basic_plus_adr(top(), base, offset);
5659 
5660   // Call the stub.
5661   address stubAddr = StubRoutines::updateBytesCRC32();
5662   const char *stubName = "updateBytesCRC32";
5663 
5664   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5665                                  stubAddr, stubName, TypePtr::BOTTOM,
5666                                  crc, src_start, length);
5667   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5668   set_result(result);
5669   return true;
5670 }
5671 
5672 //------------------------------get_table_from_crc32c_class-----------------------
5673 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5674   Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5675   assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5676 
5677   return table;
5678 }
5679 
5680 //------------------------------inline_updateBytesCRC32C-----------------------
5681 //
5682 // Calculate CRC32C for byte[] array.
5683 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5684 //
5685 bool LibraryCallKit::inline_updateBytesCRC32C() {
5686   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5687   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5688   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5689   // no receiver since it is a static method
5690   Node* crc     = argument(0); // type: int
5691   Node* src     = argument(1); // type: oop
5692   Node* offset  = argument(2); // type: int
5693   Node* end     = argument(3); // type: int
5694 
5695   Node* length = _gvn.transform(new SubINode(end, offset));
5696 
5697   const Type* src_type = src->Value(&_gvn);
5698   const TypeAryPtr* top_src = src_type->isa_aryptr();
5699   if (top_src  == NULL || top_src->klass()  == NULL) {
5700     // failed array check
5701     return false;
5702   }
5703 
5704   // Figure out the size and type of the elements we will be copying.
5705   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5706   if (src_elem != T_BYTE) {
5707     return false;
5708   }
5709 
5710   // 'src_start' points to src array + scaled offset
5711   Node* src_start = array_element_address(src, offset, src_elem);
5712 
5713   // static final int[] byteTable in class CRC32C
5714   Node* table = get_table_from_crc32c_class(callee()->holder());
5715   Node* table_start = array_element_address(table, intcon(0), T_INT);
5716 
5717   // We assume that range check is done by caller.
5718   // TODO: generate range check (offset+length < src.length) in debug VM.
5719 
5720   // Call the stub.
5721   address stubAddr = StubRoutines::updateBytesCRC32C();
5722   const char *stubName = "updateBytesCRC32C";
5723 
5724   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5725                                  stubAddr, stubName, TypePtr::BOTTOM,
5726                                  crc, src_start, length, table_start);
5727   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5728   set_result(result);
5729   return true;
5730 }
5731 
5732 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5733 //
5734 // Calculate CRC32C for DirectByteBuffer.
5735 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5736 //
5737 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5738   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5739   assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5740   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5741   // no receiver since it is a static method
5742   Node* crc     = argument(0); // type: int
5743   Node* src     = argument(1); // type: long
5744   Node* offset  = argument(3); // type: int
5745   Node* end     = argument(4); // type: int
5746 
5747   Node* length = _gvn.transform(new SubINode(end, offset));
5748 
5749   src = ConvL2X(src);  // adjust Java long to machine word
5750   Node* base = _gvn.transform(new CastX2PNode(src));
5751   offset = ConvI2X(offset);
5752 
5753   // 'src_start' points to src array + scaled offset
5754   Node* src_start = basic_plus_adr(top(), base, offset);
5755 
5756   // static final int[] byteTable in class CRC32C
5757   Node* table = get_table_from_crc32c_class(callee()->holder());
5758   Node* table_start = array_element_address(table, intcon(0), T_INT);
5759 
5760   // Call the stub.
5761   address stubAddr = StubRoutines::updateBytesCRC32C();
5762   const char *stubName = "updateBytesCRC32C";
5763 
5764   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5765                                  stubAddr, stubName, TypePtr::BOTTOM,
5766                                  crc, src_start, length, table_start);
5767   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5768   set_result(result);
5769   return true;
5770 }
5771 
5772 //------------------------------inline_updateBytesAdler32----------------------
5773 //
5774 // Calculate Adler32 checksum for byte[] array.
5775 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5776 //
5777 bool LibraryCallKit::inline_updateBytesAdler32() {
5778   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5779   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5780   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5781   // no receiver since it is static method
5782   Node* crc     = argument(0); // type: int
5783   Node* src     = argument(1); // type: oop
5784   Node* offset  = argument(2); // type: int
5785   Node* length  = argument(3); // type: int
5786 
5787   const Type* src_type = src->Value(&_gvn);
5788   const TypeAryPtr* top_src = src_type->isa_aryptr();
5789   if (top_src  == NULL || top_src->klass()  == NULL) {
5790     // failed array check
5791     return false;
5792   }
5793 
5794   // Figure out the size and type of the elements we will be copying.
5795   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5796   if (src_elem != T_BYTE) {
5797     return false;
5798   }
5799 
5800   // 'src_start' points to src array + scaled offset
5801   Node* src_start = array_element_address(src, offset, src_elem);
5802 
5803   // We assume that range check is done by caller.
5804   // TODO: generate range check (offset+length < src.length) in debug VM.
5805 
5806   // Call the stub.
5807   address stubAddr = StubRoutines::updateBytesAdler32();
5808   const char *stubName = "updateBytesAdler32";
5809 
5810   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5811                                  stubAddr, stubName, TypePtr::BOTTOM,
5812                                  crc, src_start, length);
5813   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5814   set_result(result);
5815   return true;
5816 }
5817 
5818 //------------------------------inline_updateByteBufferAdler32---------------
5819 //
5820 // Calculate Adler32 checksum for DirectByteBuffer.
5821 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5822 //
5823 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5824   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5825   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5826   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5827   // no receiver since it is static method
5828   Node* crc     = argument(0); // type: int
5829   Node* src     = argument(1); // type: long
5830   Node* offset  = argument(3); // type: int
5831   Node* length  = argument(4); // type: int
5832 
5833   src = ConvL2X(src);  // adjust Java long to machine word
5834   Node* base = _gvn.transform(new CastX2PNode(src));
5835   offset = ConvI2X(offset);
5836 
5837   // 'src_start' points to src array + scaled offset
5838   Node* src_start = basic_plus_adr(top(), base, offset);
5839 
5840   // Call the stub.
5841   address stubAddr = StubRoutines::updateBytesAdler32();
5842   const char *stubName = "updateBytesAdler32";
5843 
5844   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5845                                  stubAddr, stubName, TypePtr::BOTTOM,
5846                                  crc, src_start, length);
5847 
5848   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5849   set_result(result);
5850   return true;
5851 }
5852 
5853 //----------------------------inline_reference_get----------------------------
5854 // public T java.lang.ref.Reference.get();
5855 bool LibraryCallKit::inline_reference_get() {
5856   const int referent_offset = java_lang_ref_Reference::referent_offset;
5857   guarantee(referent_offset > 0, "should have already been set");
5858 
5859   // Get the argument:
5860   Node* reference_obj = null_check_receiver();
5861   if (stopped()) return true;
5862 
5863   Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5864 
5865   ciInstanceKlass* klass = env()->Object_klass();
5866   const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5867 
5868   Node* no_ctrl = NULL;
5869   Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5870 
5871   // Use the pre-barrier to record the value in the referent field
5872   pre_barrier(false /* do_load */,
5873               control(),
5874               NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5875               result /* pre_val */,
5876               T_OBJECT);
5877 
5878   // Add memory barrier to prevent commoning reads from this field
5879   // across safepoint since GC can change its value.
5880   insert_mem_bar(Op_MemBarCPUOrder);
5881 
5882   set_result(result);
5883   return true;
5884 }
5885 
5886 
5887 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5888                                               bool is_exact=true, bool is_static=false,
5889                                               ciInstanceKlass * fromKls=NULL) {
5890   if (fromKls == NULL) {
5891     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5892     assert(tinst != NULL, "obj is null");
5893     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5894     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5895     fromKls = tinst->klass()->as_instance_klass();
5896   } else {
5897     assert(is_static, "only for static field access");
5898   }
5899   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5900                                               ciSymbol::make(fieldTypeString),
5901                                               is_static);
5902 
5903   assert (field != NULL, "undefined field");
5904   if (field == NULL) return (Node *) NULL;
5905 
5906   if (is_static) {
5907     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5908     fromObj = makecon(tip);
5909   }
5910 
5911   // Next code  copied from Parse::do_get_xxx():
5912 
5913   // Compute address and memory type.
5914   int offset  = field->offset_in_bytes();
5915   bool is_vol = field->is_volatile();
5916   ciType* field_klass = field->type();
5917   assert(field_klass->is_loaded(), "should be loaded");
5918   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5919   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5920   BasicType bt = field->layout_type();
5921 
5922   // Build the resultant type of the load
5923   const Type *type;
5924   if (bt == T_OBJECT) {
5925     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5926   } else {
5927     type = Type::get_const_basic_type(bt);
5928   }
5929 
5930   if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5931     insert_mem_bar(Op_MemBarVolatile);   // StoreLoad barrier
5932   }
5933   // Build the load.
5934   MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5935   Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5936   // If reference is volatile, prevent following memory ops from
5937   // floating up past the volatile read.  Also prevents commoning
5938   // another volatile read.
5939   if (is_vol) {
5940     // Memory barrier includes bogus read of value to force load BEFORE membar
5941     insert_mem_bar(Op_MemBarAcquire, loadedField);
5942   }
5943   return loadedField;
5944 }
5945 
5946 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5947                                                  bool is_exact = true, bool is_static = false,
5948                                                  ciInstanceKlass * fromKls = NULL) {
5949   if (fromKls == NULL) {
5950     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5951     assert(tinst != NULL, "obj is null");
5952     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5953     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5954     fromKls = tinst->klass()->as_instance_klass();
5955   }
5956   else {
5957     assert(is_static, "only for static field access");
5958   }
5959   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5960     ciSymbol::make(fieldTypeString),
5961     is_static);
5962 
5963   assert(field != NULL, "undefined field");
5964   assert(!field->is_volatile(), "not defined for volatile fields");
5965 
5966   if (is_static) {
5967     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5968     fromObj = makecon(tip);
5969   }
5970 
5971   // Next code  copied from Parse::do_get_xxx():
5972 
5973   // Compute address and memory type.
5974   int offset = field->offset_in_bytes();
5975   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5976 
5977   return adr;
5978 }
5979 
5980 //------------------------------inline_aescrypt_Block-----------------------
5981 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5982   address stubAddr = NULL;
5983   const char *stubName;
5984   assert(UseAES, "need AES instruction support");
5985 
5986   switch(id) {
5987   case vmIntrinsics::_aescrypt_encryptBlock:
5988     stubAddr = StubRoutines::aescrypt_encryptBlock();
5989     stubName = "aescrypt_encryptBlock";
5990     break;
5991   case vmIntrinsics::_aescrypt_decryptBlock:
5992     stubAddr = StubRoutines::aescrypt_decryptBlock();
5993     stubName = "aescrypt_decryptBlock";
5994     break;
5995   }
5996   if (stubAddr == NULL) return false;
5997 
5998   Node* aescrypt_object = argument(0);
5999   Node* src             = argument(1);
6000   Node* src_offset      = argument(2);
6001   Node* dest            = argument(3);
6002   Node* dest_offset     = argument(4);
6003 
6004   // (1) src and dest are arrays.
6005   const Type* src_type = src->Value(&_gvn);
6006   const Type* dest_type = dest->Value(&_gvn);
6007   const TypeAryPtr* top_src = src_type->isa_aryptr();
6008   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6009   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6010 
6011   // for the quick and dirty code we will skip all the checks.
6012   // we are just trying to get the call to be generated.
6013   Node* src_start  = src;
6014   Node* dest_start = dest;
6015   if (src_offset != NULL || dest_offset != NULL) {
6016     assert(src_offset != NULL && dest_offset != NULL, "");
6017     src_start  = array_element_address(src,  src_offset,  T_BYTE);
6018     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6019   }
6020 
6021   // now need to get the start of its expanded key array
6022   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6023   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6024   if (k_start == NULL) return false;
6025 
6026   if (Matcher::pass_original_key_for_aes()) {
6027     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6028     // compatibility issues between Java key expansion and SPARC crypto instructions
6029     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6030     if (original_k_start == NULL) return false;
6031 
6032     // Call the stub.
6033     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6034                       stubAddr, stubName, TypePtr::BOTTOM,
6035                       src_start, dest_start, k_start, original_k_start);
6036   } else {
6037     // Call the stub.
6038     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6039                       stubAddr, stubName, TypePtr::BOTTOM,
6040                       src_start, dest_start, k_start);
6041   }
6042 
6043   return true;
6044 }
6045 
6046 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6047 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6048   address stubAddr = NULL;
6049   const char *stubName = NULL;
6050 
6051   assert(UseAES, "need AES instruction support");
6052 
6053   switch(id) {
6054   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6055     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6056     stubName = "cipherBlockChaining_encryptAESCrypt";
6057     break;
6058   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6059     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6060     stubName = "cipherBlockChaining_decryptAESCrypt";
6061     break;
6062   }
6063   if (stubAddr == NULL) return false;
6064 
6065   Node* cipherBlockChaining_object = argument(0);
6066   Node* src                        = argument(1);
6067   Node* src_offset                 = argument(2);
6068   Node* len                        = argument(3);
6069   Node* dest                       = argument(4);
6070   Node* dest_offset                = argument(5);
6071 
6072   // (1) src and dest are arrays.
6073   const Type* src_type = src->Value(&_gvn);
6074   const Type* dest_type = dest->Value(&_gvn);
6075   const TypeAryPtr* top_src = src_type->isa_aryptr();
6076   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6077   assert (top_src  != NULL && top_src->klass()  != NULL
6078           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6079 
6080   // checks are the responsibility of the caller
6081   Node* src_start  = src;
6082   Node* dest_start = dest;
6083   if (src_offset != NULL || dest_offset != NULL) {
6084     assert(src_offset != NULL && dest_offset != NULL, "");
6085     src_start  = array_element_address(src,  src_offset,  T_BYTE);
6086     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6087   }
6088 
6089   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6090   // (because of the predicated logic executed earlier).
6091   // so we cast it here safely.
6092   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6093 
6094   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6095   if (embeddedCipherObj == NULL) return false;
6096 
6097   // cast it to what we know it will be at runtime
6098   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6099   assert(tinst != NULL, "CBC obj is null");
6100   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6101   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6102   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6103 
6104   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6105   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6106   const TypeOopPtr* xtype = aklass->as_instance_type();
6107   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6108   aescrypt_object = _gvn.transform(aescrypt_object);
6109 
6110   // we need to get the start of the aescrypt_object's expanded key array
6111   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6112   if (k_start == NULL) return false;
6113 
6114   // similarly, get the start address of the r vector
6115   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6116   if (objRvec == NULL) return false;
6117   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6118 
6119   Node* cbcCrypt;
6120   if (Matcher::pass_original_key_for_aes()) {
6121     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6122     // compatibility issues between Java key expansion and SPARC crypto instructions
6123     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6124     if (original_k_start == NULL) return false;
6125 
6126     // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6127     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6128                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6129                                  stubAddr, stubName, TypePtr::BOTTOM,
6130                                  src_start, dest_start, k_start, r_start, len, original_k_start);
6131   } else {
6132     // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6133     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6134                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6135                                  stubAddr, stubName, TypePtr::BOTTOM,
6136                                  src_start, dest_start, k_start, r_start, len);
6137   }
6138 
6139   // return cipher length (int)
6140   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6141   set_result(retvalue);
6142   return true;
6143 }
6144 
6145 //------------------------------inline_counterMode_AESCrypt-----------------------
6146 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6147   assert(UseAES, "need AES instruction support");
6148   if (!UseAESCTRIntrinsics) return false;
6149 
6150   address stubAddr = NULL;
6151   const char *stubName = NULL;
6152   if (id == vmIntrinsics::_counterMode_AESCrypt) {
6153     stubAddr = StubRoutines::counterMode_AESCrypt();
6154     stubName = "counterMode_AESCrypt";
6155   }
6156   if (stubAddr == NULL) return false;
6157 
6158   Node* counterMode_object = argument(0);
6159   Node* src = argument(1);
6160   Node* src_offset = argument(2);
6161   Node* len = argument(3);
6162   Node* dest = argument(4);
6163   Node* dest_offset = argument(5);
6164 
6165   // (1) src and dest are arrays.
6166   const Type* src_type = src->Value(&_gvn);
6167   const Type* dest_type = dest->Value(&_gvn);
6168   const TypeAryPtr* top_src = src_type->isa_aryptr();
6169   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6170   assert(top_src != NULL && top_src->klass() != NULL &&
6171          top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6172 
6173   // checks are the responsibility of the caller
6174   Node* src_start = src;
6175   Node* dest_start = dest;
6176   if (src_offset != NULL || dest_offset != NULL) {
6177     assert(src_offset != NULL && dest_offset != NULL, "");
6178     src_start = array_element_address(src, src_offset, T_BYTE);
6179     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6180   }
6181 
6182   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6183   // (because of the predicated logic executed earlier).
6184   // so we cast it here safely.
6185   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6186   Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6187   if (embeddedCipherObj == NULL) return false;
6188   // cast it to what we know it will be at runtime
6189   const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6190   assert(tinst != NULL, "CTR obj is null");
6191   assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6192   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6193   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6194   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6195   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6196   const TypeOopPtr* xtype = aklass->as_instance_type();
6197   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6198   aescrypt_object = _gvn.transform(aescrypt_object);
6199   // we need to get the start of the aescrypt_object's expanded key array
6200   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6201   if (k_start == NULL) return false;
6202   // similarly, get the start address of the r vector
6203   Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6204   if (obj_counter == NULL) return false;
6205   Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6206 
6207   Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6208   if (saved_encCounter == NULL) return false;
6209   Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6210   Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6211 
6212   Node* ctrCrypt;
6213   if (Matcher::pass_original_key_for_aes()) {
6214     // no SPARC version for AES/CTR intrinsics now.
6215     return false;
6216   }
6217   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6218   ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6219                                OptoRuntime::counterMode_aescrypt_Type(),
6220                                stubAddr, stubName, TypePtr::BOTTOM,
6221                                src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6222 
6223   // return cipher length (int)
6224   Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6225   set_result(retvalue);
6226   return true;
6227 }
6228 
6229 //------------------------------get_key_start_from_aescrypt_object-----------------------
6230 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6231 #ifdef PPC64
6232   // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6233   // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6234   // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6235   // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6236   Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6237   assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6238   if (objSessionK == NULL) {
6239     return (Node *) NULL;
6240   }
6241   Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6242 #else
6243   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6244 #endif // PPC64
6245   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6246   if (objAESCryptKey == NULL) return (Node *) NULL;
6247 
6248   // now have the array, need to get the start address of the K array
6249   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6250   return k_start;
6251 }
6252 
6253 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6254 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6255   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6256   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6257   if (objAESCryptKey == NULL) return (Node *) NULL;
6258 
6259   // now have the array, need to get the start address of the lastKey array
6260   Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6261   return original_k_start;
6262 }
6263 
6264 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6265 // Return node representing slow path of predicate check.
6266 // the pseudo code we want to emulate with this predicate is:
6267 // for encryption:
6268 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6269 // for decryption:
6270 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6271 //    note cipher==plain is more conservative than the original java code but that's OK
6272 //
6273 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6274   // The receiver was checked for NULL already.
6275   Node* objCBC = argument(0);
6276 
6277   // Load embeddedCipher field of CipherBlockChaining object.
6278   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6279 
6280   // get AESCrypt klass for instanceOf check
6281   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6282   // will have same classloader as CipherBlockChaining object
6283   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6284   assert(tinst != NULL, "CBCobj is null");
6285   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6286 
6287   // we want to do an instanceof comparison against the AESCrypt class
6288   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6289   if (!klass_AESCrypt->is_loaded()) {
6290     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6291     Node* ctrl = control();
6292     set_control(top()); // no regular fast path
6293     return ctrl;
6294   }
6295   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6296 
6297   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6298   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
6299   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6300 
6301   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6302 
6303   // for encryption, we are done
6304   if (!decrypting)
6305     return instof_false;  // even if it is NULL
6306 
6307   // for decryption, we need to add a further check to avoid
6308   // taking the intrinsic path when cipher and plain are the same
6309   // see the original java code for why.
6310   RegionNode* region = new RegionNode(3);
6311   region->init_req(1, instof_false);
6312   Node* src = argument(1);
6313   Node* dest = argument(4);
6314   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6315   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6316   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6317   region->init_req(2, src_dest_conjoint);
6318 
6319   record_for_igvn(region);
6320   return _gvn.transform(region);
6321 }
6322 
6323 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6324 // Return node representing slow path of predicate check.
6325 // the pseudo code we want to emulate with this predicate is:
6326 // for encryption:
6327 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6328 // for decryption:
6329 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6330 //    note cipher==plain is more conservative than the original java code but that's OK
6331 //
6332 
6333 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6334   // The receiver was checked for NULL already.
6335   Node* objCTR = argument(0);
6336 
6337   // Load embeddedCipher field of CipherBlockChaining object.
6338   Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6339 
6340   // get AESCrypt klass for instanceOf check
6341   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6342   // will have same classloader as CipherBlockChaining object
6343   const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6344   assert(tinst != NULL, "CTRobj is null");
6345   assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6346 
6347   // we want to do an instanceof comparison against the AESCrypt class
6348   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6349   if (!klass_AESCrypt->is_loaded()) {
6350     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6351     Node* ctrl = control();
6352     set_control(top()); // no regular fast path
6353     return ctrl;
6354   }
6355 
6356   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6357   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6358   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6359   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6360   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6361 
6362   return instof_false; // even if it is NULL
6363 }
6364 
6365 //------------------------------inline_ghash_processBlocks
6366 bool LibraryCallKit::inline_ghash_processBlocks() {
6367   address stubAddr;
6368   const char *stubName;
6369   assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6370 
6371   stubAddr = StubRoutines::ghash_processBlocks();
6372   stubName = "ghash_processBlocks";
6373 
6374   Node* data           = argument(0);
6375   Node* offset         = argument(1);
6376   Node* len            = argument(2);
6377   Node* state          = argument(3);
6378   Node* subkeyH        = argument(4);
6379 
6380   Node* state_start  = array_element_address(state, intcon(0), T_LONG);
6381   assert(state_start, "state is NULL");
6382   Node* subkeyH_start  = array_element_address(subkeyH, intcon(0), T_LONG);
6383   assert(subkeyH_start, "subkeyH is NULL");
6384   Node* data_start  = array_element_address(data, offset, T_BYTE);
6385   assert(data_start, "data is NULL");
6386 
6387   Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6388                                   OptoRuntime::ghash_processBlocks_Type(),
6389                                   stubAddr, stubName, TypePtr::BOTTOM,
6390                                   state_start, subkeyH_start, data_start, len);
6391   return true;
6392 }
6393 
6394 //------------------------------inline_sha_implCompress-----------------------
6395 //
6396 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6397 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6398 //
6399 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6400 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6401 //
6402 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6403 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6404 //
6405 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6406   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6407 
6408   Node* sha_obj = argument(0);
6409   Node* src     = argument(1); // type oop
6410   Node* ofs     = argument(2); // type int
6411 
6412   const Type* src_type = src->Value(&_gvn);
6413   const TypeAryPtr* top_src = src_type->isa_aryptr();
6414   if (top_src  == NULL || top_src->klass()  == NULL) {
6415     // failed array check
6416     return false;
6417   }
6418   // Figure out the size and type of the elements we will be copying.
6419   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6420   if (src_elem != T_BYTE) {
6421     return false;
6422   }
6423   // 'src_start' points to src array + offset
6424   Node* src_start = array_element_address(src, ofs, src_elem);
6425   Node* state = NULL;
6426   address stubAddr;
6427   const char *stubName;
6428 
6429   switch(id) {
6430   case vmIntrinsics::_sha_implCompress:
6431     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6432     state = get_state_from_sha_object(sha_obj);
6433     stubAddr = StubRoutines::sha1_implCompress();
6434     stubName = "sha1_implCompress";
6435     break;
6436   case vmIntrinsics::_sha2_implCompress:
6437     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6438     state = get_state_from_sha_object(sha_obj);
6439     stubAddr = StubRoutines::sha256_implCompress();
6440     stubName = "sha256_implCompress";
6441     break;
6442   case vmIntrinsics::_sha5_implCompress:
6443     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6444     state = get_state_from_sha5_object(sha_obj);
6445     stubAddr = StubRoutines::sha512_implCompress();
6446     stubName = "sha512_implCompress";
6447     break;
6448   default:
6449     fatal_unexpected_iid(id);
6450     return false;
6451   }
6452   if (state == NULL) return false;
6453 
6454   // Call the stub.
6455   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6456                                  stubAddr, stubName, TypePtr::BOTTOM,
6457                                  src_start, state);
6458 
6459   return true;
6460 }
6461 
6462 //------------------------------inline_digestBase_implCompressMB-----------------------
6463 //
6464 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6465 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6466 //
6467 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6468   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6469          "need SHA1/SHA256/SHA512 instruction support");
6470   assert((uint)predicate < 3, "sanity");
6471   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6472 
6473   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6474   Node* src            = argument(1); // byte[] array
6475   Node* ofs            = argument(2); // type int
6476   Node* limit          = argument(3); // type int
6477 
6478   const Type* src_type = src->Value(&_gvn);
6479   const TypeAryPtr* top_src = src_type->isa_aryptr();
6480   if (top_src  == NULL || top_src->klass()  == NULL) {
6481     // failed array check
6482     return false;
6483   }
6484   // Figure out the size and type of the elements we will be copying.
6485   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6486   if (src_elem != T_BYTE) {
6487     return false;
6488   }
6489   // 'src_start' points to src array + offset
6490   Node* src_start = array_element_address(src, ofs, src_elem);
6491 
6492   const char* klass_SHA_name = NULL;
6493   const char* stub_name = NULL;
6494   address     stub_addr = NULL;
6495   bool        long_state = false;
6496 
6497   switch (predicate) {
6498   case 0:
6499     if (UseSHA1Intrinsics) {
6500       klass_SHA_name = "sun/security/provider/SHA";
6501       stub_name = "sha1_implCompressMB";
6502       stub_addr = StubRoutines::sha1_implCompressMB();
6503     }
6504     break;
6505   case 1:
6506     if (UseSHA256Intrinsics) {
6507       klass_SHA_name = "sun/security/provider/SHA2";
6508       stub_name = "sha256_implCompressMB";
6509       stub_addr = StubRoutines::sha256_implCompressMB();
6510     }
6511     break;
6512   case 2:
6513     if (UseSHA512Intrinsics) {
6514       klass_SHA_name = "sun/security/provider/SHA5";
6515       stub_name = "sha512_implCompressMB";
6516       stub_addr = StubRoutines::sha512_implCompressMB();
6517       long_state = true;
6518     }
6519     break;
6520   default:
6521     fatal("unknown SHA intrinsic predicate: %d", predicate);
6522   }
6523   if (klass_SHA_name != NULL) {
6524     // get DigestBase klass to lookup for SHA klass
6525     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6526     assert(tinst != NULL, "digestBase_obj is not instance???");
6527     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6528 
6529     ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6530     assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6531     ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6532     return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6533   }
6534   return false;
6535 }
6536 //------------------------------inline_sha_implCompressMB-----------------------
6537 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6538                                                bool long_state, address stubAddr, const char *stubName,
6539                                                Node* src_start, Node* ofs, Node* limit) {
6540   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6541   const TypeOopPtr* xtype = aklass->as_instance_type();
6542   Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6543   sha_obj = _gvn.transform(sha_obj);
6544 
6545   Node* state;
6546   if (long_state) {
6547     state = get_state_from_sha5_object(sha_obj);
6548   } else {
6549     state = get_state_from_sha_object(sha_obj);
6550   }
6551   if (state == NULL) return false;
6552 
6553   // Call the stub.
6554   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6555                                  OptoRuntime::digestBase_implCompressMB_Type(),
6556                                  stubAddr, stubName, TypePtr::BOTTOM,
6557                                  src_start, state, ofs, limit);
6558   // return ofs (int)
6559   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6560   set_result(result);
6561 
6562   return true;
6563 }
6564 
6565 //------------------------------get_state_from_sha_object-----------------------
6566 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6567   Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6568   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6569   if (sha_state == NULL) return (Node *) NULL;
6570 
6571   // now have the array, need to get the start address of the state array
6572   Node* state = array_element_address(sha_state, intcon(0), T_INT);
6573   return state;
6574 }
6575 
6576 //------------------------------get_state_from_sha5_object-----------------------
6577 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6578   Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6579   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6580   if (sha_state == NULL) return (Node *) NULL;
6581 
6582   // now have the array, need to get the start address of the state array
6583   Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6584   return state;
6585 }
6586 
6587 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6588 // Return node representing slow path of predicate check.
6589 // the pseudo code we want to emulate with this predicate is:
6590 //    if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6591 //
6592 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6593   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6594          "need SHA1/SHA256/SHA512 instruction support");
6595   assert((uint)predicate < 3, "sanity");
6596 
6597   // The receiver was checked for NULL already.
6598   Node* digestBaseObj = argument(0);
6599 
6600   // get DigestBase klass for instanceOf check
6601   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6602   assert(tinst != NULL, "digestBaseObj is null");
6603   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6604 
6605   const char* klass_SHA_name = NULL;
6606   switch (predicate) {
6607   case 0:
6608     if (UseSHA1Intrinsics) {
6609       // we want to do an instanceof comparison against the SHA class
6610       klass_SHA_name = "sun/security/provider/SHA";
6611     }
6612     break;
6613   case 1:
6614     if (UseSHA256Intrinsics) {
6615       // we want to do an instanceof comparison against the SHA2 class
6616       klass_SHA_name = "sun/security/provider/SHA2";
6617     }
6618     break;
6619   case 2:
6620     if (UseSHA512Intrinsics) {
6621       // we want to do an instanceof comparison against the SHA5 class
6622       klass_SHA_name = "sun/security/provider/SHA5";
6623     }
6624     break;
6625   default:
6626     fatal("unknown SHA intrinsic predicate: %d", predicate);
6627   }
6628 
6629   ciKlass* klass_SHA = NULL;
6630   if (klass_SHA_name != NULL) {
6631     klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6632   }
6633   if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6634     // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6635     Node* ctrl = control();
6636     set_control(top()); // no intrinsic path
6637     return ctrl;
6638   }
6639   ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6640 
6641   Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6642   Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6643   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6644   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6645 
6646   return instof_false;  // even if it is NULL
6647 }
6648 
6649 bool LibraryCallKit::inline_profileBoolean() {
6650   Node* counts = argument(1);
6651   const TypeAryPtr* ary = NULL;
6652   ciArray* aobj = NULL;
6653   if (counts->is_Con()
6654       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6655       && (aobj = ary->const_oop()->as_array()) != NULL
6656       && (aobj->length() == 2)) {
6657     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6658     jint false_cnt = aobj->element_value(0).as_int();
6659     jint  true_cnt = aobj->element_value(1).as_int();
6660 
6661     if (C->log() != NULL) {
6662       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6663                      false_cnt, true_cnt);
6664     }
6665 
6666     if (false_cnt + true_cnt == 0) {
6667       // According to profile, never executed.
6668       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6669                           Deoptimization::Action_reinterpret);
6670       return true;
6671     }
6672 
6673     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6674     // is a number of each value occurrences.
6675     Node* result = argument(0);
6676     if (false_cnt == 0 || true_cnt == 0) {
6677       // According to profile, one value has been never seen.
6678       int expected_val = (false_cnt == 0) ? 1 : 0;
6679 
6680       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6681       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6682 
6683       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6684       Node* fast_path = _gvn.transform(new IfTrueNode(check));
6685       Node* slow_path = _gvn.transform(new IfFalseNode(check));
6686 
6687       { // Slow path: uncommon trap for never seen value and then reexecute
6688         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6689         // the value has been seen at least once.
6690         PreserveJVMState pjvms(this);
6691         PreserveReexecuteState preexecs(this);
6692         jvms()->set_should_reexecute(true);
6693 
6694         set_control(slow_path);
6695         set_i_o(i_o());
6696 
6697         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6698                             Deoptimization::Action_reinterpret);
6699       }
6700       // The guard for never seen value enables sharpening of the result and
6701       // returning a constant. It allows to eliminate branches on the same value
6702       // later on.
6703       set_control(fast_path);
6704       result = intcon(expected_val);
6705     }
6706     // Stop profiling.
6707     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6708     // By replacing method body with profile data (represented as ProfileBooleanNode
6709     // on IR level) we effectively disable profiling.
6710     // It enables full speed execution once optimized code is generated.
6711     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6712     C->record_for_igvn(profile);
6713     set_result(profile);
6714     return true;
6715   } else {
6716     // Continue profiling.
6717     // Profile data isn't available at the moment. So, execute method's bytecode version.
6718     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6719     // is compiled and counters aren't available since corresponding MethodHandle
6720     // isn't a compile-time constant.
6721     return false;
6722   }
6723 }
6724 
6725 bool LibraryCallKit::inline_isCompileConstant() {
6726   Node* n = argument(0);
6727   set_result(n->is_Con() ? intcon(1) : intcon(0));
6728   return true;
6729 }
--- EOF ---