1 /*
   2  * Copyright (c) 1999, 2018, 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 "ci/ciUtilities.inline.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "compiler/compileBroker.hpp"
  31 #include "compiler/compileLog.hpp"
  32 #include "gc/shared/barrierSet.hpp"
  33 #include "jfr/support/jfrIntrinsics.hpp"
  34 #include "memory/resourceArea.hpp"
  35 #include "oops/objArrayKlass.hpp"
  36 #include "opto/addnode.hpp"
  37 #include "opto/arraycopynode.hpp"
  38 #include "opto/c2compiler.hpp"
  39 #include "opto/callGenerator.hpp"
  40 #include "opto/castnode.hpp"
  41 #include "opto/cfgnode.hpp"
  42 #include "opto/convertnode.hpp"
  43 #include "opto/countbitsnode.hpp"
  44 #include "opto/intrinsicnode.hpp"
  45 #include "opto/idealKit.hpp"
  46 #include "opto/mathexactnode.hpp"
  47 #include "opto/movenode.hpp"
  48 #include "opto/mulnode.hpp"
  49 #include "opto/narrowptrnode.hpp"
  50 #include "opto/opaquenode.hpp"
  51 #include "opto/parse.hpp"
  52 #include "opto/runtime.hpp"
  53 #include "opto/rootnode.hpp"
  54 #include "opto/subnode.hpp"
  55 #include "prims/nativeLookup.hpp"
  56 #include "prims/unsafe.hpp"
  57 #include "runtime/objectMonitor.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "utilities/macros.hpp"
  60 
  61 
  62 class LibraryIntrinsic : public InlineCallGenerator {
  63   // Extend the set of intrinsics known to the runtime:
  64  public:
  65  private:
  66   bool             _is_virtual;
  67   bool             _does_virtual_dispatch;
  68   int8_t           _predicates_count;  // Intrinsic is predicated by several conditions
  69   int8_t           _last_predicate; // Last generated predicate
  70   vmIntrinsics::ID _intrinsic_id;
  71 
  72  public:
  73   LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
  74     : InlineCallGenerator(m),
  75       _is_virtual(is_virtual),
  76       _does_virtual_dispatch(does_virtual_dispatch),
  77       _predicates_count((int8_t)predicates_count),
  78       _last_predicate((int8_t)-1),
  79       _intrinsic_id(id)
  80   {
  81   }
  82   virtual bool is_intrinsic() const { return true; }
  83   virtual bool is_virtual()   const { return _is_virtual; }
  84   virtual bool is_predicated() const { return _predicates_count > 0; }
  85   virtual int  predicates_count() const { return _predicates_count; }
  86   virtual bool does_virtual_dispatch()   const { return _does_virtual_dispatch; }
  87   virtual JVMState* generate(JVMState* jvms);
  88   virtual Node* generate_predicate(JVMState* jvms, int predicate);
  89   vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
  90 };
  91 
  92 
  93 // Local helper class for LibraryIntrinsic:
  94 class LibraryCallKit : public GraphKit {
  95  private:
  96   LibraryIntrinsic* _intrinsic;     // the library intrinsic being called
  97   Node*             _result;        // the result node, if any
  98   int               _reexecute_sp;  // the stack pointer when bytecode needs to be reexecuted
  99 
 100   const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
 101 
 102  public:
 103   LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
 104     : GraphKit(jvms),
 105       _intrinsic(intrinsic),
 106       _result(NULL)
 107   {
 108     // Check if this is a root compile.  In that case we don't have a caller.
 109     if (!jvms->has_method()) {
 110       _reexecute_sp = sp();
 111     } else {
 112       // Find out how many arguments the interpreter needs when deoptimizing
 113       // and save the stack pointer value so it can used by uncommon_trap.
 114       // We find the argument count by looking at the declared signature.
 115       bool ignored_will_link;
 116       ciSignature* declared_signature = NULL;
 117       ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
 118       const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
 119       _reexecute_sp = sp() + nargs;  // "push" arguments back on stack
 120     }
 121   }
 122 
 123   virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
 124 
 125   ciMethod*         caller()    const    { return jvms()->method(); }
 126   int               bci()       const    { return jvms()->bci(); }
 127   LibraryIntrinsic* intrinsic() const    { return _intrinsic; }
 128   vmIntrinsics::ID  intrinsic_id() const { return _intrinsic->intrinsic_id(); }
 129   ciMethod*         callee()    const    { return _intrinsic->method(); }
 130 
 131   bool  try_to_inline(int predicate);
 132   Node* try_to_predicate(int predicate);
 133 
 134   void push_result() {
 135     // Push the result onto the stack.
 136     if (!stopped() && result() != NULL) {
 137       BasicType bt = result()->bottom_type()->basic_type();
 138       push_node(bt, result());
 139     }
 140   }
 141 
 142  private:
 143   void fatal_unexpected_iid(vmIntrinsics::ID iid) {
 144     fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
 145   }
 146 
 147   void  set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
 148   void  set_result(RegionNode* region, PhiNode* value);
 149   Node*     result() { return _result; }
 150 
 151   virtual int reexecute_sp() { return _reexecute_sp; }
 152 
 153   // Helper functions to inline natives
 154   Node* generate_guard(Node* test, RegionNode* region, float true_prob);
 155   Node* generate_slow_guard(Node* test, RegionNode* region);
 156   Node* generate_fair_guard(Node* test, RegionNode* region);
 157   Node* generate_negative_guard(Node* index, RegionNode* region,
 158                                 // resulting CastII of index:
 159                                 Node* *pos_index = NULL);
 160   Node* generate_limit_guard(Node* offset, Node* subseq_length,
 161                              Node* array_length,
 162                              RegionNode* region);
 163   void  generate_string_range_check(Node* array, Node* offset,
 164                                     Node* length, bool char_count);
 165   Node* generate_current_thread(Node* &tls_output);
 166   Node* load_mirror_from_klass(Node* klass);
 167   Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
 168                                       RegionNode* region, int null_path,
 169                                       int offset);
 170   Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
 171                                RegionNode* region, int null_path) {
 172     int offset = java_lang_Class::klass_offset_in_bytes();
 173     return load_klass_from_mirror_common(mirror, never_see_null,
 174                                          region, null_path,
 175                                          offset);
 176   }
 177   Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
 178                                      RegionNode* region, int null_path) {
 179     int offset = java_lang_Class::array_klass_offset_in_bytes();
 180     return load_klass_from_mirror_common(mirror, never_see_null,
 181                                          region, null_path,
 182                                          offset);
 183   }
 184   Node* generate_access_flags_guard(Node* kls,
 185                                     int modifier_mask, int modifier_bits,
 186                                     RegionNode* region);
 187   Node* generate_interface_guard(Node* kls, RegionNode* region);
 188   Node* generate_array_guard(Node* kls, RegionNode* region) {
 189     return generate_array_guard_common(kls, region, false, false);
 190   }
 191   Node* generate_non_array_guard(Node* kls, RegionNode* region) {
 192     return generate_array_guard_common(kls, region, false, true);
 193   }
 194   Node* generate_objArray_guard(Node* kls, RegionNode* region) {
 195     return generate_array_guard_common(kls, region, true, false);
 196   }
 197   Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
 198     return generate_array_guard_common(kls, region, true, true);
 199   }
 200   Node* generate_array_guard_common(Node* kls, RegionNode* region,
 201                                     bool obj_array, bool not_array);
 202   Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
 203   CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
 204                                      bool is_virtual = false, bool is_static = false);
 205   CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
 206     return generate_method_call(method_id, false, true);
 207   }
 208   CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
 209     return generate_method_call(method_id, true, false);
 210   }
 211   Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
 212   Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
 213 
 214   Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
 215   bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
 216   bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
 217   bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
 218   Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
 219                           RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
 220   bool inline_string_indexOfChar();
 221   bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
 222   bool inline_string_toBytesU();
 223   bool inline_string_getCharsU();
 224   bool inline_string_copy(bool compress);
 225   bool inline_string_char_access(bool is_store);
 226   Node* round_double_node(Node* n);
 227   bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
 228   bool inline_math_native(vmIntrinsics::ID id);
 229   bool inline_math(vmIntrinsics::ID id);
 230   template <typename OverflowOp>
 231   bool inline_math_overflow(Node* arg1, Node* arg2);
 232   void inline_math_mathExact(Node* math, Node* test);
 233   bool inline_math_addExactI(bool is_increment);
 234   bool inline_math_addExactL(bool is_increment);
 235   bool inline_math_multiplyExactI();
 236   bool inline_math_multiplyExactL();
 237   bool inline_math_multiplyHigh();
 238   bool inline_math_negateExactI();
 239   bool inline_math_negateExactL();
 240   bool inline_math_subtractExactI(bool is_decrement);
 241   bool inline_math_subtractExactL(bool is_decrement);
 242   bool inline_min_max(vmIntrinsics::ID id);
 243   bool inline_notify(vmIntrinsics::ID id);
 244   Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
 245   // This returns Type::AnyPtr, RawPtr, or OopPtr.
 246   int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
 247   Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
 248 
 249   typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
 250   DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
 251   bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
 252   static bool klass_needs_init_guard(Node* kls);
 253   bool inline_unsafe_allocate();
 254   bool inline_unsafe_newArray(bool uninitialized);
 255   bool inline_unsafe_copyMemory();
 256   bool inline_native_currentThread();
 257 
 258   bool inline_native_time_funcs(address method, const char* funcName);
 259 #ifdef JFR_HAVE_INTRINSICS
 260   bool inline_native_classID();
 261   bool inline_native_getEventWriter();
 262 #endif
 263   bool inline_native_isInterrupted();
 264   bool inline_native_Class_query(vmIntrinsics::ID id);
 265   bool inline_native_subtype_check();
 266   bool inline_native_getLength();
 267   bool inline_array_copyOf(bool is_copyOfRange);
 268   bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
 269   bool inline_preconditions_checkIndex();
 270   void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
 271   bool inline_native_clone(bool is_virtual);
 272   bool inline_native_Reflection_getCallerClass();
 273   // Helper function for inlining native object hash method
 274   bool inline_native_hashcode(bool is_virtual, bool is_static);
 275   bool inline_native_getClass();
 276 
 277   // Helper functions for inlining arraycopy
 278   bool inline_arraycopy();
 279   AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
 280                                                 RegionNode* slow_region);
 281   JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
 282   void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
 283                                       uint new_idx);
 284 
 285   typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
 286   bool inline_unsafe_load_store(BasicType type,  LoadStoreKind kind, AccessKind access_kind);
 287   bool inline_unsafe_fence(vmIntrinsics::ID id);
 288   bool inline_onspinwait();
 289   bool inline_fp_conversions(vmIntrinsics::ID id);
 290   bool inline_number_methods(vmIntrinsics::ID id);
 291   bool inline_reference_get();
 292   bool inline_Class_cast();
 293   bool inline_aescrypt_Block(vmIntrinsics::ID id);
 294   bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
 295   bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
 296   Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
 297   Node* inline_counterMode_AESCrypt_predicate();
 298   Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
 299   Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
 300   bool inline_ghash_processBlocks();
 301   bool inline_base64_encodeBlock();
 302   bool inline_sha_implCompress(vmIntrinsics::ID id);
 303   bool inline_digestBase_implCompressMB(int predicate);
 304   bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
 305                                  bool long_state, address stubAddr, const char *stubName,
 306                                  Node* src_start, Node* ofs, Node* limit);
 307   Node* get_state_from_sha_object(Node *sha_object);
 308   Node* get_state_from_sha5_object(Node *sha_object);
 309   Node* inline_digestBase_implCompressMB_predicate(int predicate);
 310   bool inline_encodeISOArray();
 311   bool inline_updateCRC32();
 312   bool inline_updateBytesCRC32();
 313   bool inline_updateByteBufferCRC32();
 314   Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
 315   bool inline_updateBytesCRC32C();
 316   bool inline_updateDirectByteBufferCRC32C();
 317   bool inline_updateBytesAdler32();
 318   bool inline_updateByteBufferAdler32();
 319   bool inline_multiplyToLen();
 320   bool inline_hasNegatives();
 321   bool inline_squareToLen();
 322   bool inline_mulAdd();
 323   bool inline_montgomeryMultiply();
 324   bool inline_montgomerySquare();
 325   bool inline_vectorizedMismatch();
 326   bool inline_fma(vmIntrinsics::ID id);
 327 
 328   bool inline_profileBoolean();
 329   bool inline_isCompileConstant();
 330   void clear_upper_avx() {
 331 #ifdef X86
 332     if (UseAVX >= 2) {
 333       C->set_clear_upper_avx(true);
 334     }
 335 #endif
 336   }
 337 };
 338 
 339 //---------------------------make_vm_intrinsic----------------------------
 340 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
 341   vmIntrinsics::ID id = m->intrinsic_id();
 342   assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
 343 
 344   if (!m->is_loaded()) {
 345     // Do not attempt to inline unloaded methods.
 346     return NULL;
 347   }
 348 
 349   C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
 350   bool is_available = false;
 351 
 352   {
 353     // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
 354     // the compiler must transition to '_thread_in_vm' state because both
 355     // methods access VM-internal data.
 356     VM_ENTRY_MARK;
 357     methodHandle mh(THREAD, m->get_Method());
 358     is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
 359                    !C->directive()->is_intrinsic_disabled(mh) &&
 360                    !vmIntrinsics::is_disabled_by_flags(mh);
 361 
 362   }
 363 
 364   if (is_available) {
 365     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
 366     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
 367     return new LibraryIntrinsic(m, is_virtual,
 368                                 vmIntrinsics::predicates_needed(id),
 369                                 vmIntrinsics::does_virtual_dispatch(id),
 370                                 (vmIntrinsics::ID) id);
 371   } else {
 372     return NULL;
 373   }
 374 }
 375 
 376 //----------------------register_library_intrinsics-----------------------
 377 // Initialize this file's data structures, for each Compile instance.
 378 void Compile::register_library_intrinsics() {
 379   // Nothing to do here.
 380 }
 381 
 382 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
 383   LibraryCallKit kit(jvms, this);
 384   Compile* C = kit.C;
 385   int nodes = C->unique();
 386 #ifndef PRODUCT
 387   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 388     char buf[1000];
 389     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 390     tty->print_cr("Intrinsic %s", str);
 391   }
 392 #endif
 393   ciMethod* callee = kit.callee();
 394   const int bci    = kit.bci();
 395 
 396   // Try to inline the intrinsic.
 397   if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
 398       kit.try_to_inline(_last_predicate)) {
 399     const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
 400                                           : "(intrinsic)";
 401     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
 402     if (C->print_intrinsics() || C->print_inlining()) {
 403       C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
 404     }
 405     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 406     if (C->log()) {
 407       C->log()->elem("intrinsic id='%s'%s nodes='%d'",
 408                      vmIntrinsics::name_at(intrinsic_id()),
 409                      (is_virtual() ? " virtual='1'" : ""),
 410                      C->unique() - nodes);
 411     }
 412     // Push the result from the inlined method onto the stack.
 413     kit.push_result();
 414     C->print_inlining_update(this);
 415     return kit.transfer_exceptions_into_jvms();
 416   }
 417 
 418   // The intrinsic bailed out
 419   if (jvms->has_method()) {
 420     // Not a root compile.
 421     const char* msg;
 422     if (callee->intrinsic_candidate()) {
 423       msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
 424     } else {
 425       msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
 426                          : "failed to inline (intrinsic), method not annotated";
 427     }
 428     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
 429     if (C->print_intrinsics() || C->print_inlining()) {
 430       C->print_inlining(callee, jvms->depth() - 1, bci, msg);
 431     }
 432   } else {
 433     // Root compile
 434     ResourceMark rm;
 435     stringStream msg_stream;
 436     msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
 437                      vmIntrinsics::name_at(intrinsic_id()),
 438                      is_virtual() ? " (virtual)" : "", bci);
 439     const char *msg = msg_stream.as_string();
 440     log_debug(jit, inlining)("%s", msg);
 441     if (C->print_intrinsics() || C->print_inlining()) {
 442       tty->print("%s", msg);
 443     }
 444   }
 445   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 446   C->print_inlining_update(this);
 447   return NULL;
 448 }
 449 
 450 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
 451   LibraryCallKit kit(jvms, this);
 452   Compile* C = kit.C;
 453   int nodes = C->unique();
 454   _last_predicate = predicate;
 455 #ifndef PRODUCT
 456   assert(is_predicated() && predicate < predicates_count(), "sanity");
 457   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 458     char buf[1000];
 459     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 460     tty->print_cr("Predicate for intrinsic %s", str);
 461   }
 462 #endif
 463   ciMethod* callee = kit.callee();
 464   const int bci    = kit.bci();
 465 
 466   Node* slow_ctl = kit.try_to_predicate(predicate);
 467   if (!kit.failing()) {
 468     const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
 469                                           : "(intrinsic, predicate)";
 470     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
 471     if (C->print_intrinsics() || C->print_inlining()) {
 472       C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
 473     }
 474     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 475     if (C->log()) {
 476       C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
 477                      vmIntrinsics::name_at(intrinsic_id()),
 478                      (is_virtual() ? " virtual='1'" : ""),
 479                      C->unique() - nodes);
 480     }
 481     return slow_ctl; // Could be NULL if the check folds.
 482   }
 483 
 484   // The intrinsic bailed out
 485   if (jvms->has_method()) {
 486     // Not a root compile.
 487     const char* msg = "failed to generate predicate for intrinsic";
 488     CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
 489     if (C->print_intrinsics() || C->print_inlining()) {
 490       C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
 491     }
 492   } else {
 493     // Root compile
 494     ResourceMark rm;
 495     stringStream msg_stream;
 496     msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
 497                      vmIntrinsics::name_at(intrinsic_id()),
 498                      is_virtual() ? " (virtual)" : "", bci);
 499     const char *msg = msg_stream.as_string();
 500     log_debug(jit, inlining)("%s", msg);
 501     if (C->print_intrinsics() || C->print_inlining()) {
 502       C->print_inlining_stream()->print("%s", msg);
 503     }
 504   }
 505   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 506   return NULL;
 507 }
 508 
 509 bool LibraryCallKit::try_to_inline(int predicate) {
 510   // Handle symbolic names for otherwise undistinguished boolean switches:
 511   const bool is_store       = true;
 512   const bool is_compress    = true;
 513   const bool is_static      = true;
 514   const bool is_volatile    = true;
 515 
 516   if (!jvms()->has_method()) {
 517     // Root JVMState has a null method.
 518     assert(map()->memory()->Opcode() == Op_Parm, "");
 519     // Insert the memory aliasing node
 520     set_all_memory(reset_memory());
 521   }
 522   assert(merged_memory(), "");
 523 
 524 
 525   switch (intrinsic_id()) {
 526   case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
 527   case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
 528   case vmIntrinsics::_getClass:                 return inline_native_getClass();
 529 
 530   case vmIntrinsics::_dsin:
 531   case vmIntrinsics::_dcos:
 532   case vmIntrinsics::_dtan:
 533   case vmIntrinsics::_dabs:
 534   case vmIntrinsics::_datan2:
 535   case vmIntrinsics::_dsqrt:
 536   case vmIntrinsics::_dexp:
 537   case vmIntrinsics::_dlog:
 538   case vmIntrinsics::_dlog10:
 539   case vmIntrinsics::_dpow:                     return inline_math_native(intrinsic_id());
 540 
 541   case vmIntrinsics::_min:
 542   case vmIntrinsics::_max:                      return inline_min_max(intrinsic_id());
 543 
 544   case vmIntrinsics::_notify:
 545   case vmIntrinsics::_notifyAll:
 546     return inline_notify(intrinsic_id());
 547 
 548   case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
 549   case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
 550   case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
 551   case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
 552   case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
 553   case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
 554   case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
 555   case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
 556   case vmIntrinsics::_multiplyHigh:             return inline_math_multiplyHigh();
 557   case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
 558   case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
 559   case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
 560   case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
 561 
 562   case vmIntrinsics::_arraycopy:                return inline_arraycopy();
 563 
 564   case vmIntrinsics::_compareToL:               return inline_string_compareTo(StrIntrinsicNode::LL);
 565   case vmIntrinsics::_compareToU:               return inline_string_compareTo(StrIntrinsicNode::UU);
 566   case vmIntrinsics::_compareToLU:              return inline_string_compareTo(StrIntrinsicNode::LU);
 567   case vmIntrinsics::_compareToUL:              return inline_string_compareTo(StrIntrinsicNode::UL);
 568 
 569   case vmIntrinsics::_indexOfL:                 return inline_string_indexOf(StrIntrinsicNode::LL);
 570   case vmIntrinsics::_indexOfU:                 return inline_string_indexOf(StrIntrinsicNode::UU);
 571   case vmIntrinsics::_indexOfUL:                return inline_string_indexOf(StrIntrinsicNode::UL);
 572   case vmIntrinsics::_indexOfIL:                return inline_string_indexOfI(StrIntrinsicNode::LL);
 573   case vmIntrinsics::_indexOfIU:                return inline_string_indexOfI(StrIntrinsicNode::UU);
 574   case vmIntrinsics::_indexOfIUL:               return inline_string_indexOfI(StrIntrinsicNode::UL);
 575   case vmIntrinsics::_indexOfU_char:            return inline_string_indexOfChar();
 576 
 577   case vmIntrinsics::_equalsL:                  return inline_string_equals(StrIntrinsicNode::LL);
 578   case vmIntrinsics::_equalsU:                  return inline_string_equals(StrIntrinsicNode::UU);
 579 
 580   case vmIntrinsics::_toBytesStringU:           return inline_string_toBytesU();
 581   case vmIntrinsics::_getCharsStringU:          return inline_string_getCharsU();
 582   case vmIntrinsics::_getCharStringU:           return inline_string_char_access(!is_store);
 583   case vmIntrinsics::_putCharStringU:           return inline_string_char_access( is_store);
 584 
 585   case vmIntrinsics::_compressStringC:
 586   case vmIntrinsics::_compressStringB:          return inline_string_copy( is_compress);
 587   case vmIntrinsics::_inflateStringC:
 588   case vmIntrinsics::_inflateStringB:           return inline_string_copy(!is_compress);
 589 
 590   case vmIntrinsics::_getReference:             return inline_unsafe_access(!is_store, T_OBJECT,   Relaxed, false);
 591   case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_store, T_BOOLEAN,  Relaxed, false);
 592   case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_store, T_BYTE,     Relaxed, false);
 593   case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, false);
 594   case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, false);
 595   case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_store, T_INT,      Relaxed, false);
 596   case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_store, T_LONG,     Relaxed, false);
 597   case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_store, T_FLOAT,    Relaxed, false);
 598   case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_store, T_DOUBLE,   Relaxed, false);
 599 
 600   case vmIntrinsics::_putReference:             return inline_unsafe_access( is_store, T_OBJECT,   Relaxed, false);
 601   case vmIntrinsics::_putBoolean:               return inline_unsafe_access( is_store, T_BOOLEAN,  Relaxed, false);
 602   case vmIntrinsics::_putByte:                  return inline_unsafe_access( is_store, T_BYTE,     Relaxed, false);
 603   case vmIntrinsics::_putShort:                 return inline_unsafe_access( is_store, T_SHORT,    Relaxed, false);
 604   case vmIntrinsics::_putChar:                  return inline_unsafe_access( is_store, T_CHAR,     Relaxed, false);
 605   case vmIntrinsics::_putInt:                   return inline_unsafe_access( is_store, T_INT,      Relaxed, false);
 606   case vmIntrinsics::_putLong:                  return inline_unsafe_access( is_store, T_LONG,     Relaxed, false);
 607   case vmIntrinsics::_putFloat:                 return inline_unsafe_access( is_store, T_FLOAT,    Relaxed, false);
 608   case vmIntrinsics::_putDouble:                return inline_unsafe_access( is_store, T_DOUBLE,   Relaxed, false);
 609 
 610   case vmIntrinsics::_getReferenceVolatile:     return inline_unsafe_access(!is_store, T_OBJECT,   Volatile, false);
 611   case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_store, T_BOOLEAN,  Volatile, false);
 612   case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_store, T_BYTE,     Volatile, false);
 613   case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_store, T_SHORT,    Volatile, false);
 614   case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_store, T_CHAR,     Volatile, false);
 615   case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_store, T_INT,      Volatile, false);
 616   case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_store, T_LONG,     Volatile, false);
 617   case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_store, T_FLOAT,    Volatile, false);
 618   case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_store, T_DOUBLE,   Volatile, false);
 619 
 620   case vmIntrinsics::_putReferenceVolatile:     return inline_unsafe_access( is_store, T_OBJECT,   Volatile, false);
 621   case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access( is_store, T_BOOLEAN,  Volatile, false);
 622   case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access( is_store, T_BYTE,     Volatile, false);
 623   case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access( is_store, T_SHORT,    Volatile, false);
 624   case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access( is_store, T_CHAR,     Volatile, false);
 625   case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access( is_store, T_INT,      Volatile, false);
 626   case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access( is_store, T_LONG,     Volatile, false);
 627   case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access( is_store, T_FLOAT,    Volatile, false);
 628   case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access( is_store, T_DOUBLE,   Volatile, false);
 629 
 630   case vmIntrinsics::_getShortUnaligned:        return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, true);
 631   case vmIntrinsics::_getCharUnaligned:         return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, true);
 632   case vmIntrinsics::_getIntUnaligned:          return inline_unsafe_access(!is_store, T_INT,      Relaxed, true);
 633   case vmIntrinsics::_getLongUnaligned:         return inline_unsafe_access(!is_store, T_LONG,     Relaxed, true);
 634 
 635   case vmIntrinsics::_putShortUnaligned:        return inline_unsafe_access( is_store, T_SHORT,    Relaxed, true);
 636   case vmIntrinsics::_putCharUnaligned:         return inline_unsafe_access( is_store, T_CHAR,     Relaxed, true);
 637   case vmIntrinsics::_putIntUnaligned:          return inline_unsafe_access( is_store, T_INT,      Relaxed, true);
 638   case vmIntrinsics::_putLongUnaligned:         return inline_unsafe_access( is_store, T_LONG,     Relaxed, true);
 639 
 640   case vmIntrinsics::_getReferenceAcquire:      return inline_unsafe_access(!is_store, T_OBJECT,   Acquire, false);
 641   case vmIntrinsics::_getBooleanAcquire:        return inline_unsafe_access(!is_store, T_BOOLEAN,  Acquire, false);
 642   case vmIntrinsics::_getByteAcquire:           return inline_unsafe_access(!is_store, T_BYTE,     Acquire, false);
 643   case vmIntrinsics::_getShortAcquire:          return inline_unsafe_access(!is_store, T_SHORT,    Acquire, false);
 644   case vmIntrinsics::_getCharAcquire:           return inline_unsafe_access(!is_store, T_CHAR,     Acquire, false);
 645   case vmIntrinsics::_getIntAcquire:            return inline_unsafe_access(!is_store, T_INT,      Acquire, false);
 646   case vmIntrinsics::_getLongAcquire:           return inline_unsafe_access(!is_store, T_LONG,     Acquire, false);
 647   case vmIntrinsics::_getFloatAcquire:          return inline_unsafe_access(!is_store, T_FLOAT,    Acquire, false);
 648   case vmIntrinsics::_getDoubleAcquire:         return inline_unsafe_access(!is_store, T_DOUBLE,   Acquire, false);
 649 
 650   case vmIntrinsics::_putReferenceRelease:      return inline_unsafe_access( is_store, T_OBJECT,   Release, false);
 651   case vmIntrinsics::_putBooleanRelease:        return inline_unsafe_access( is_store, T_BOOLEAN,  Release, false);
 652   case vmIntrinsics::_putByteRelease:           return inline_unsafe_access( is_store, T_BYTE,     Release, false);
 653   case vmIntrinsics::_putShortRelease:          return inline_unsafe_access( is_store, T_SHORT,    Release, false);
 654   case vmIntrinsics::_putCharRelease:           return inline_unsafe_access( is_store, T_CHAR,     Release, false);
 655   case vmIntrinsics::_putIntRelease:            return inline_unsafe_access( is_store, T_INT,      Release, false);
 656   case vmIntrinsics::_putLongRelease:           return inline_unsafe_access( is_store, T_LONG,     Release, false);
 657   case vmIntrinsics::_putFloatRelease:          return inline_unsafe_access( is_store, T_FLOAT,    Release, false);
 658   case vmIntrinsics::_putDoubleRelease:         return inline_unsafe_access( is_store, T_DOUBLE,   Release, false);
 659 
 660   case vmIntrinsics::_getReferenceOpaque:       return inline_unsafe_access(!is_store, T_OBJECT,   Opaque, false);
 661   case vmIntrinsics::_getBooleanOpaque:         return inline_unsafe_access(!is_store, T_BOOLEAN,  Opaque, false);
 662   case vmIntrinsics::_getByteOpaque:            return inline_unsafe_access(!is_store, T_BYTE,     Opaque, false);
 663   case vmIntrinsics::_getShortOpaque:           return inline_unsafe_access(!is_store, T_SHORT,    Opaque, false);
 664   case vmIntrinsics::_getCharOpaque:            return inline_unsafe_access(!is_store, T_CHAR,     Opaque, false);
 665   case vmIntrinsics::_getIntOpaque:             return inline_unsafe_access(!is_store, T_INT,      Opaque, false);
 666   case vmIntrinsics::_getLongOpaque:            return inline_unsafe_access(!is_store, T_LONG,     Opaque, false);
 667   case vmIntrinsics::_getFloatOpaque:           return inline_unsafe_access(!is_store, T_FLOAT,    Opaque, false);
 668   case vmIntrinsics::_getDoubleOpaque:          return inline_unsafe_access(!is_store, T_DOUBLE,   Opaque, false);
 669 
 670   case vmIntrinsics::_putReferenceOpaque:       return inline_unsafe_access( is_store, T_OBJECT,   Opaque, false);
 671   case vmIntrinsics::_putBooleanOpaque:         return inline_unsafe_access( is_store, T_BOOLEAN,  Opaque, false);
 672   case vmIntrinsics::_putByteOpaque:            return inline_unsafe_access( is_store, T_BYTE,     Opaque, false);
 673   case vmIntrinsics::_putShortOpaque:           return inline_unsafe_access( is_store, T_SHORT,    Opaque, false);
 674   case vmIntrinsics::_putCharOpaque:            return inline_unsafe_access( is_store, T_CHAR,     Opaque, false);
 675   case vmIntrinsics::_putIntOpaque:             return inline_unsafe_access( is_store, T_INT,      Opaque, false);
 676   case vmIntrinsics::_putLongOpaque:            return inline_unsafe_access( is_store, T_LONG,     Opaque, false);
 677   case vmIntrinsics::_putFloatOpaque:           return inline_unsafe_access( is_store, T_FLOAT,    Opaque, false);
 678   case vmIntrinsics::_putDoubleOpaque:          return inline_unsafe_access( is_store, T_DOUBLE,   Opaque, false);
 679 
 680   case vmIntrinsics::_compareAndSetReference:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap,      Volatile);
 681   case vmIntrinsics::_compareAndSetByte:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap,      Volatile);
 682   case vmIntrinsics::_compareAndSetShort:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap,      Volatile);
 683   case vmIntrinsics::_compareAndSetInt:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap,      Volatile);
 684   case vmIntrinsics::_compareAndSetLong:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap,      Volatile);
 685 
 686   case vmIntrinsics::_weakCompareAndSetReferencePlain:     return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
 687   case vmIntrinsics::_weakCompareAndSetReferenceAcquire:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
 688   case vmIntrinsics::_weakCompareAndSetReferenceRelease:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
 689   case vmIntrinsics::_weakCompareAndSetReference:          return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
 690   case vmIntrinsics::_weakCompareAndSetBytePlain:          return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Relaxed);
 691   case vmIntrinsics::_weakCompareAndSetByteAcquire:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Acquire);
 692   case vmIntrinsics::_weakCompareAndSetByteRelease:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Release);
 693   case vmIntrinsics::_weakCompareAndSetByte:               return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Volatile);
 694   case vmIntrinsics::_weakCompareAndSetShortPlain:         return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Relaxed);
 695   case vmIntrinsics::_weakCompareAndSetShortAcquire:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Acquire);
 696   case vmIntrinsics::_weakCompareAndSetShortRelease:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Release);
 697   case vmIntrinsics::_weakCompareAndSetShort:              return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Volatile);
 698   case vmIntrinsics::_weakCompareAndSetIntPlain:           return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Relaxed);
 699   case vmIntrinsics::_weakCompareAndSetIntAcquire:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Acquire);
 700   case vmIntrinsics::_weakCompareAndSetIntRelease:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Release);
 701   case vmIntrinsics::_weakCompareAndSetInt:                return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Volatile);
 702   case vmIntrinsics::_weakCompareAndSetLongPlain:          return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Relaxed);
 703   case vmIntrinsics::_weakCompareAndSetLongAcquire:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Acquire);
 704   case vmIntrinsics::_weakCompareAndSetLongRelease:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Release);
 705   case vmIntrinsics::_weakCompareAndSetLong:               return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Volatile);
 706 
 707   case vmIntrinsics::_compareAndExchangeReference:         return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Volatile);
 708   case vmIntrinsics::_compareAndExchangeReferenceAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Acquire);
 709   case vmIntrinsics::_compareAndExchangeReferenceRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Release);
 710   case vmIntrinsics::_compareAndExchangeByte:              return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Volatile);
 711   case vmIntrinsics::_compareAndExchangeByteAcquire:       return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Acquire);
 712   case vmIntrinsics::_compareAndExchangeByteRelease:       return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Release);
 713   case vmIntrinsics::_compareAndExchangeShort:             return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Volatile);
 714   case vmIntrinsics::_compareAndExchangeShortAcquire:      return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Acquire);
 715   case vmIntrinsics::_compareAndExchangeShortRelease:      return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Release);
 716   case vmIntrinsics::_compareAndExchangeInt:               return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Volatile);
 717   case vmIntrinsics::_compareAndExchangeIntAcquire:        return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Acquire);
 718   case vmIntrinsics::_compareAndExchangeIntRelease:        return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Release);
 719   case vmIntrinsics::_compareAndExchangeLong:              return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Volatile);
 720   case vmIntrinsics::_compareAndExchangeLongAcquire:       return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Acquire);
 721   case vmIntrinsics::_compareAndExchangeLongRelease:       return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Release);
 722 
 723   case vmIntrinsics::_getAndAddByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_add,       Volatile);
 724   case vmIntrinsics::_getAndAddShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_add,       Volatile);
 725   case vmIntrinsics::_getAndAddInt:                     return inline_unsafe_load_store(T_INT,    LS_get_add,       Volatile);
 726   case vmIntrinsics::_getAndAddLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_add,       Volatile);
 727 
 728   case vmIntrinsics::_getAndSetByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_set,       Volatile);
 729   case vmIntrinsics::_getAndSetShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_set,       Volatile);
 730   case vmIntrinsics::_getAndSetInt:                     return inline_unsafe_load_store(T_INT,    LS_get_set,       Volatile);
 731   case vmIntrinsics::_getAndSetLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_set,       Volatile);
 732   case vmIntrinsics::_getAndSetReference:               return inline_unsafe_load_store(T_OBJECT, LS_get_set,       Volatile);
 733 
 734   case vmIntrinsics::_loadFence:
 735   case vmIntrinsics::_storeFence:
 736   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 737 
 738   case vmIntrinsics::_onSpinWait:               return inline_onspinwait();
 739 
 740   case vmIntrinsics::_currentThread:            return inline_native_currentThread();
 741   case vmIntrinsics::_isInterrupted:            return inline_native_isInterrupted();
 742 
 743 #ifdef JFR_HAVE_INTRINSICS
 744   case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
 745   case vmIntrinsics::_getClassId:               return inline_native_classID();
 746   case vmIntrinsics::_getEventWriter:           return inline_native_getEventWriter();
 747 #endif
 748   case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
 749   case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
 750   case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
 751   case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
 752   case vmIntrinsics::_getLength:                return inline_native_getLength();
 753   case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
 754   case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
 755   case vmIntrinsics::_equalsB:                  return inline_array_equals(StrIntrinsicNode::LL);
 756   case vmIntrinsics::_equalsC:                  return inline_array_equals(StrIntrinsicNode::UU);
 757   case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
 758   case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());
 759 
 760   case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
 761   case vmIntrinsics::_newArray:                   return inline_unsafe_newArray(false);
 762 
 763   case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
 764 
 765   case vmIntrinsics::_isInstance:
 766   case vmIntrinsics::_getModifiers:
 767   case vmIntrinsics::_isInterface:
 768   case vmIntrinsics::_isArray:
 769   case vmIntrinsics::_isPrimitive:
 770   case vmIntrinsics::_getSuperclass:
 771   case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
 772 
 773   case vmIntrinsics::_floatToRawIntBits:
 774   case vmIntrinsics::_floatToIntBits:
 775   case vmIntrinsics::_intBitsToFloat:
 776   case vmIntrinsics::_doubleToRawLongBits:
 777   case vmIntrinsics::_doubleToLongBits:
 778   case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
 779 
 780   case vmIntrinsics::_numberOfLeadingZeros_i:
 781   case vmIntrinsics::_numberOfLeadingZeros_l:
 782   case vmIntrinsics::_numberOfTrailingZeros_i:
 783   case vmIntrinsics::_numberOfTrailingZeros_l:
 784   case vmIntrinsics::_bitCount_i:
 785   case vmIntrinsics::_bitCount_l:
 786   case vmIntrinsics::_reverseBytes_i:
 787   case vmIntrinsics::_reverseBytes_l:
 788   case vmIntrinsics::_reverseBytes_s:
 789   case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
 790 
 791   case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
 792 
 793   case vmIntrinsics::_Reference_get:            return inline_reference_get();
 794 
 795   case vmIntrinsics::_Class_cast:               return inline_Class_cast();
 796 
 797   case vmIntrinsics::_aescrypt_encryptBlock:
 798   case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
 799 
 800   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 801   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 802     return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
 803 
 804   case vmIntrinsics::_counterMode_AESCrypt:
 805     return inline_counterMode_AESCrypt(intrinsic_id());
 806 
 807   case vmIntrinsics::_sha_implCompress:
 808   case vmIntrinsics::_sha2_implCompress:
 809   case vmIntrinsics::_sha5_implCompress:
 810     return inline_sha_implCompress(intrinsic_id());
 811 
 812   case vmIntrinsics::_digestBase_implCompressMB:
 813     return inline_digestBase_implCompressMB(predicate);
 814 
 815   case vmIntrinsics::_multiplyToLen:
 816     return inline_multiplyToLen();
 817 
 818   case vmIntrinsics::_squareToLen:
 819     return inline_squareToLen();
 820 
 821   case vmIntrinsics::_mulAdd:
 822     return inline_mulAdd();
 823 
 824   case vmIntrinsics::_montgomeryMultiply:
 825     return inline_montgomeryMultiply();
 826   case vmIntrinsics::_montgomerySquare:
 827     return inline_montgomerySquare();
 828 
 829   case vmIntrinsics::_vectorizedMismatch:
 830     return inline_vectorizedMismatch();
 831 
 832   case vmIntrinsics::_ghash_processBlocks:
 833     return inline_ghash_processBlocks();
 834   case vmIntrinsics::_base64_encodeBlock:
 835     return inline_base64_encodeBlock();
 836 
 837   case vmIntrinsics::_encodeISOArray:
 838   case vmIntrinsics::_encodeByteISOArray:
 839     return inline_encodeISOArray();
 840 
 841   case vmIntrinsics::_updateCRC32:
 842     return inline_updateCRC32();
 843   case vmIntrinsics::_updateBytesCRC32:
 844     return inline_updateBytesCRC32();
 845   case vmIntrinsics::_updateByteBufferCRC32:
 846     return inline_updateByteBufferCRC32();
 847 
 848   case vmIntrinsics::_updateBytesCRC32C:
 849     return inline_updateBytesCRC32C();
 850   case vmIntrinsics::_updateDirectByteBufferCRC32C:
 851     return inline_updateDirectByteBufferCRC32C();
 852 
 853   case vmIntrinsics::_updateBytesAdler32:
 854     return inline_updateBytesAdler32();
 855   case vmIntrinsics::_updateByteBufferAdler32:
 856     return inline_updateByteBufferAdler32();
 857 
 858   case vmIntrinsics::_profileBoolean:
 859     return inline_profileBoolean();
 860   case vmIntrinsics::_isCompileConstant:
 861     return inline_isCompileConstant();
 862 
 863   case vmIntrinsics::_hasNegatives:
 864     return inline_hasNegatives();
 865 
 866   case vmIntrinsics::_fmaD:
 867   case vmIntrinsics::_fmaF:
 868     return inline_fma(intrinsic_id());
 869 
 870   default:
 871     // If you get here, it may be that someone has added a new intrinsic
 872     // to the list in vmSymbols.hpp without implementing it here.
 873 #ifndef PRODUCT
 874     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 875       tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
 876                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 877     }
 878 #endif
 879     return false;
 880   }
 881 }
 882 
 883 Node* LibraryCallKit::try_to_predicate(int predicate) {
 884   if (!jvms()->has_method()) {
 885     // Root JVMState has a null method.
 886     assert(map()->memory()->Opcode() == Op_Parm, "");
 887     // Insert the memory aliasing node
 888     set_all_memory(reset_memory());
 889   }
 890   assert(merged_memory(), "");
 891 
 892   switch (intrinsic_id()) {
 893   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 894     return inline_cipherBlockChaining_AESCrypt_predicate(false);
 895   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 896     return inline_cipherBlockChaining_AESCrypt_predicate(true);
 897   case vmIntrinsics::_counterMode_AESCrypt:
 898     return inline_counterMode_AESCrypt_predicate();
 899   case vmIntrinsics::_digestBase_implCompressMB:
 900     return inline_digestBase_implCompressMB_predicate(predicate);
 901 
 902   default:
 903     // If you get here, it may be that someone has added a new intrinsic
 904     // to the list in vmSymbols.hpp without implementing it here.
 905 #ifndef PRODUCT
 906     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 907       tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
 908                     vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
 909     }
 910 #endif
 911     Node* slow_ctl = control();
 912     set_control(top()); // No fast path instrinsic
 913     return slow_ctl;
 914   }
 915 }
 916 
 917 //------------------------------set_result-------------------------------
 918 // Helper function for finishing intrinsics.
 919 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
 920   record_for_igvn(region);
 921   set_control(_gvn.transform(region));
 922   set_result( _gvn.transform(value));
 923   assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
 924 }
 925 
 926 //------------------------------generate_guard---------------------------
 927 // Helper function for generating guarded fast-slow graph structures.
 928 // The given 'test', if true, guards a slow path.  If the test fails
 929 // then a fast path can be taken.  (We generally hope it fails.)
 930 // In all cases, GraphKit::control() is updated to the fast path.
 931 // The returned value represents the control for the slow path.
 932 // The return value is never 'top'; it is either a valid control
 933 // or NULL if it is obvious that the slow path can never be taken.
 934 // Also, if region and the slow control are not NULL, the slow edge
 935 // is appended to the region.
 936 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
 937   if (stopped()) {
 938     // Already short circuited.
 939     return NULL;
 940   }
 941 
 942   // Build an if node and its projections.
 943   // If test is true we take the slow path, which we assume is uncommon.
 944   if (_gvn.type(test) == TypeInt::ZERO) {
 945     // The slow branch is never taken.  No need to build this guard.
 946     return NULL;
 947   }
 948 
 949   IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
 950 
 951   Node* if_slow = _gvn.transform(new IfTrueNode(iff));
 952   if (if_slow == top()) {
 953     // The slow branch is never taken.  No need to build this guard.
 954     return NULL;
 955   }
 956 
 957   if (region != NULL)
 958     region->add_req(if_slow);
 959 
 960   Node* if_fast = _gvn.transform(new IfFalseNode(iff));
 961   set_control(if_fast);
 962 
 963   return if_slow;
 964 }
 965 
 966 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
 967   return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
 968 }
 969 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
 970   return generate_guard(test, region, PROB_FAIR);
 971 }
 972 
 973 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
 974                                                      Node* *pos_index) {
 975   if (stopped())
 976     return NULL;                // already stopped
 977   if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
 978     return NULL;                // index is already adequately typed
 979   Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
 980   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 981   Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
 982   if (is_neg != NULL && pos_index != NULL) {
 983     // Emulate effect of Parse::adjust_map_after_if.
 984     Node* ccast = new CastIINode(index, TypeInt::POS);
 985     ccast->set_req(0, control());
 986     (*pos_index) = _gvn.transform(ccast);
 987   }
 988   return is_neg;
 989 }
 990 
 991 // Make sure that 'position' is a valid limit index, in [0..length].
 992 // There are two equivalent plans for checking this:
 993 //   A. (offset + copyLength)  unsigned<=  arrayLength
 994 //   B. offset  <=  (arrayLength - copyLength)
 995 // We require that all of the values above, except for the sum and
 996 // difference, are already known to be non-negative.
 997 // Plan A is robust in the face of overflow, if offset and copyLength
 998 // are both hugely positive.
 999 //
1000 // Plan B is less direct and intuitive, but it does not overflow at
1001 // all, since the difference of two non-negatives is always
1002 // representable.  Whenever Java methods must perform the equivalent
1003 // check they generally use Plan B instead of Plan A.
1004 // For the moment we use Plan A.
1005 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1006                                                   Node* subseq_length,
1007                                                   Node* array_length,
1008                                                   RegionNode* region) {
1009   if (stopped())
1010     return NULL;                // already stopped
1011   bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1012   if (zero_offset && subseq_length->eqv_uncast(array_length))
1013     return NULL;                // common case of whole-array copy
1014   Node* last = subseq_length;
1015   if (!zero_offset)             // last += offset
1016     last = _gvn.transform(new AddINode(last, offset));
1017   Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1018   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1019   Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1020   return is_over;
1021 }
1022 
1023 // Emit range checks for the given String.value byte array
1024 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1025   if (stopped()) {
1026     return; // already stopped
1027   }
1028   RegionNode* bailout = new RegionNode(1);
1029   record_for_igvn(bailout);
1030   if (char_count) {
1031     // Convert char count to byte count
1032     count = _gvn.transform(new LShiftINode(count, intcon(1)));
1033   }
1034 
1035   // Offset and count must not be negative
1036   generate_negative_guard(offset, bailout);
1037   generate_negative_guard(count, bailout);
1038   // Offset + count must not exceed length of array
1039   generate_limit_guard(offset, count, load_array_length(array), bailout);
1040 
1041   if (bailout->req() > 1) {
1042     PreserveJVMState pjvms(this);
1043     set_control(_gvn.transform(bailout));
1044     uncommon_trap(Deoptimization::Reason_intrinsic,
1045                   Deoptimization::Action_maybe_recompile);
1046   }
1047 }
1048 
1049 //--------------------------generate_current_thread--------------------
1050 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1051   ciKlass*    thread_klass = env()->Thread_klass();
1052   const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1053   Node* thread = _gvn.transform(new ThreadLocalNode());
1054   Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1055   Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1056   tls_output = thread;
1057   return threadObj;
1058 }
1059 
1060 
1061 //------------------------------make_string_method_node------------------------
1062 // Helper method for String intrinsic functions. This version is called with
1063 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1064 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1065 // containing the lengths of str1 and str2.
1066 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1067   Node* result = NULL;
1068   switch (opcode) {
1069   case Op_StrIndexOf:
1070     result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1071                                 str1_start, cnt1, str2_start, cnt2, ae);
1072     break;
1073   case Op_StrComp:
1074     result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1075                              str1_start, cnt1, str2_start, cnt2, ae);
1076     break;
1077   case Op_StrEquals:
1078     // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1079     // Use the constant length if there is one because optimized match rule may exist.
1080     result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1081                                str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1082     break;
1083   default:
1084     ShouldNotReachHere();
1085     return NULL;
1086   }
1087 
1088   // All these intrinsics have checks.
1089   C->set_has_split_ifs(true); // Has chance for split-if optimization
1090   clear_upper_avx();
1091 
1092   return _gvn.transform(result);
1093 }
1094 
1095 //------------------------------inline_string_compareTo------------------------
1096 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1097   Node* arg1 = argument(0);
1098   Node* arg2 = argument(1);
1099 
1100   arg1 = must_be_not_null(arg1, true);
1101   arg2 = must_be_not_null(arg2, true);
1102 
1103   arg1 = access_resolve(arg1, ACCESS_READ);
1104   arg2 = access_resolve(arg2, ACCESS_READ);
1105 
1106   // Get start addr and length of first argument
1107   Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1108   Node* arg1_cnt    = load_array_length(arg1);
1109 
1110   // Get start addr and length of second argument
1111   Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1112   Node* arg2_cnt    = load_array_length(arg2);
1113 
1114   Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1115   set_result(result);
1116   return true;
1117 }
1118 
1119 //------------------------------inline_string_equals------------------------
1120 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1121   Node* arg1 = argument(0);
1122   Node* arg2 = argument(1);
1123 
1124   // paths (plus control) merge
1125   RegionNode* region = new RegionNode(3);
1126   Node* phi = new PhiNode(region, TypeInt::BOOL);
1127 
1128   if (!stopped()) {
1129 
1130     arg1 = must_be_not_null(arg1, true);
1131     arg2 = must_be_not_null(arg2, true);
1132 
1133     arg1 = access_resolve(arg1, ACCESS_READ);
1134     arg2 = access_resolve(arg2, ACCESS_READ);
1135 
1136     // Get start addr and length of first argument
1137     Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
1138     Node* arg1_cnt    = load_array_length(arg1);
1139 
1140     // Get start addr and length of second argument
1141     Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
1142     Node* arg2_cnt    = load_array_length(arg2);
1143 
1144     // Check for arg1_cnt != arg2_cnt
1145     Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1146     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1147     Node* if_ne = generate_slow_guard(bol, NULL);
1148     if (if_ne != NULL) {
1149       phi->init_req(2, intcon(0));
1150       region->init_req(2, if_ne);
1151     }
1152 
1153     // Check for count == 0 is done by assembler code for StrEquals.
1154 
1155     if (!stopped()) {
1156       Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1157       phi->init_req(1, equals);
1158       region->init_req(1, control());
1159     }
1160   }
1161 
1162   // post merge
1163   set_control(_gvn.transform(region));
1164   record_for_igvn(region);
1165 
1166   set_result(_gvn.transform(phi));
1167   return true;
1168 }
1169 
1170 //------------------------------inline_array_equals----------------------------
1171 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1172   assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1173   Node* arg1 = argument(0);
1174   Node* arg2 = argument(1);
1175 
1176   arg1 = access_resolve(arg1, ACCESS_READ);
1177   arg2 = access_resolve(arg2, ACCESS_READ);
1178 
1179   const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1180   set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1181   clear_upper_avx();
1182 
1183   return true;
1184 }
1185 
1186 //------------------------------inline_hasNegatives------------------------------
1187 bool LibraryCallKit::inline_hasNegatives() {
1188   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1189     return false;
1190   }
1191 
1192   assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1193   // no receiver since it is static method
1194   Node* ba         = argument(0);
1195   Node* offset     = argument(1);
1196   Node* len        = argument(2);
1197 
1198   ba = must_be_not_null(ba, true);
1199 
1200   // Range checks
1201   generate_string_range_check(ba, offset, len, false);
1202   if (stopped()) {
1203     return true;
1204   }
1205   ba = access_resolve(ba, ACCESS_READ);
1206   Node* ba_start = array_element_address(ba, offset, T_BYTE);
1207   Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1208   set_result(_gvn.transform(result));
1209   return true;
1210 }
1211 
1212 bool LibraryCallKit::inline_preconditions_checkIndex() {
1213   Node* index = argument(0);
1214   Node* length = argument(1);
1215   if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1216     return false;
1217   }
1218 
1219   Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1220   Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1221 
1222   {
1223     BuildCutout unless(this, len_pos_bol, PROB_MAX);
1224     uncommon_trap(Deoptimization::Reason_intrinsic,
1225                   Deoptimization::Action_make_not_entrant);
1226   }
1227 
1228   if (stopped()) {
1229     return false;
1230   }
1231 
1232   Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1233   BoolTest::mask btest = BoolTest::lt;
1234   Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1235   RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1236   _gvn.set_type(rc, rc->Value(&_gvn));
1237   if (!rc_bool->is_Con()) {
1238     record_for_igvn(rc);
1239   }
1240   set_control(_gvn.transform(new IfTrueNode(rc)));
1241   {
1242     PreserveJVMState pjvms(this);
1243     set_control(_gvn.transform(new IfFalseNode(rc)));
1244     uncommon_trap(Deoptimization::Reason_range_check,
1245                   Deoptimization::Action_make_not_entrant);
1246   }
1247 
1248   if (stopped()) {
1249     return false;
1250   }
1251 
1252   Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1253   result->set_req(0, control());
1254   result = _gvn.transform(result);
1255   set_result(result);
1256   replace_in_map(index, result);
1257   clear_upper_avx();
1258   return true;
1259 }
1260 
1261 //------------------------------inline_string_indexOf------------------------
1262 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1263   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1264     return false;
1265   }
1266   Node* src = argument(0);
1267   Node* tgt = argument(1);
1268 
1269   // Make the merge point
1270   RegionNode* result_rgn = new RegionNode(4);
1271   Node*       result_phi = new PhiNode(result_rgn, TypeInt::INT);
1272 
1273   src = must_be_not_null(src, true);
1274   tgt = must_be_not_null(tgt, true);
1275 
1276   src = access_resolve(src, ACCESS_READ);
1277   tgt = access_resolve(tgt, ACCESS_READ);
1278 
1279   // Get start addr and length of source string
1280   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1281   Node* src_count = load_array_length(src);
1282 
1283   // Get start addr and length of substring
1284   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1285   Node* tgt_count = load_array_length(tgt);
1286 
1287   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1288     // Divide src size by 2 if String is UTF16 encoded
1289     src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1290   }
1291   if (ae == StrIntrinsicNode::UU) {
1292     // Divide substring size by 2 if String is UTF16 encoded
1293     tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1294   }
1295 
1296   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1297   if (result != NULL) {
1298     result_phi->init_req(3, result);
1299     result_rgn->init_req(3, control());
1300   }
1301   set_control(_gvn.transform(result_rgn));
1302   record_for_igvn(result_rgn);
1303   set_result(_gvn.transform(result_phi));
1304 
1305   return true;
1306 }
1307 
1308 //-----------------------------inline_string_indexOf-----------------------
1309 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1310   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1311     return false;
1312   }
1313   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1314     return false;
1315   }
1316   assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1317   Node* src         = argument(0); // byte[]
1318   Node* src_count   = argument(1); // char count
1319   Node* tgt         = argument(2); // byte[]
1320   Node* tgt_count   = argument(3); // char count
1321   Node* from_index  = argument(4); // char index
1322 
1323   src = must_be_not_null(src, true);
1324   tgt = must_be_not_null(tgt, true);
1325 
1326   src = access_resolve(src, ACCESS_READ);
1327   tgt = access_resolve(tgt, ACCESS_READ);
1328 
1329   // Multiply byte array index by 2 if String is UTF16 encoded
1330   Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1331   src_count = _gvn.transform(new SubINode(src_count, from_index));
1332   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1333   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1334 
1335   // Range checks
1336   generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1337   generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1338   if (stopped()) {
1339     return true;
1340   }
1341 
1342   RegionNode* region = new RegionNode(5);
1343   Node* phi = new PhiNode(region, TypeInt::INT);
1344 
1345   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1346   if (result != NULL) {
1347     // The result is index relative to from_index if substring was found, -1 otherwise.
1348     // Generate code which will fold into cmove.
1349     Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1350     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1351 
1352     Node* if_lt = generate_slow_guard(bol, NULL);
1353     if (if_lt != NULL) {
1354       // result == -1
1355       phi->init_req(3, result);
1356       region->init_req(3, if_lt);
1357     }
1358     if (!stopped()) {
1359       result = _gvn.transform(new AddINode(result, from_index));
1360       phi->init_req(4, result);
1361       region->init_req(4, control());
1362     }
1363   }
1364 
1365   set_control(_gvn.transform(region));
1366   record_for_igvn(region);
1367   set_result(_gvn.transform(phi));
1368   clear_upper_avx();
1369 
1370   return true;
1371 }
1372 
1373 // Create StrIndexOfNode with fast path checks
1374 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1375                                         RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1376   // Check for substr count > string count
1377   Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1378   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1379   Node* if_gt = generate_slow_guard(bol, NULL);
1380   if (if_gt != NULL) {
1381     phi->init_req(1, intcon(-1));
1382     region->init_req(1, if_gt);
1383   }
1384   if (!stopped()) {
1385     // Check for substr count == 0
1386     cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1387     bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1388     Node* if_zero = generate_slow_guard(bol, NULL);
1389     if (if_zero != NULL) {
1390       phi->init_req(2, intcon(0));
1391       region->init_req(2, if_zero);
1392     }
1393   }
1394   if (!stopped()) {
1395     return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1396   }
1397   return NULL;
1398 }
1399 
1400 //-----------------------------inline_string_indexOfChar-----------------------
1401 bool LibraryCallKit::inline_string_indexOfChar() {
1402   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1403     return false;
1404   }
1405   if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1406     return false;
1407   }
1408   assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1409   Node* src         = argument(0); // byte[]
1410   Node* tgt         = argument(1); // tgt is int ch
1411   Node* from_index  = argument(2);
1412   Node* max         = argument(3);
1413 
1414   src = must_be_not_null(src, true);
1415   src = access_resolve(src, ACCESS_READ);
1416 
1417   Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1418   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1419   Node* src_count = _gvn.transform(new SubINode(max, from_index));
1420 
1421   // Range checks
1422   generate_string_range_check(src, src_offset, src_count, true);
1423   if (stopped()) {
1424     return true;
1425   }
1426 
1427   RegionNode* region = new RegionNode(3);
1428   Node* phi = new PhiNode(region, TypeInt::INT);
1429 
1430   Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1431   C->set_has_split_ifs(true); // Has chance for split-if optimization
1432   _gvn.transform(result);
1433 
1434   Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1435   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1436 
1437   Node* if_lt = generate_slow_guard(bol, NULL);
1438   if (if_lt != NULL) {
1439     // result == -1
1440     phi->init_req(2, result);
1441     region->init_req(2, if_lt);
1442   }
1443   if (!stopped()) {
1444     result = _gvn.transform(new AddINode(result, from_index));
1445     phi->init_req(1, result);
1446     region->init_req(1, control());
1447   }
1448   set_control(_gvn.transform(region));
1449   record_for_igvn(region);
1450   set_result(_gvn.transform(phi));
1451 
1452   return true;
1453 }
1454 //---------------------------inline_string_copy---------------------
1455 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1456 //   int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1457 //   int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1458 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1459 //   void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1460 //   void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1461 bool LibraryCallKit::inline_string_copy(bool compress) {
1462   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1463     return false;
1464   }
1465   int nargs = 5;  // 2 oops, 3 ints
1466   assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1467 
1468   Node* src         = argument(0);
1469   Node* src_offset  = argument(1);
1470   Node* dst         = argument(2);
1471   Node* dst_offset  = argument(3);
1472   Node* length      = argument(4);
1473 
1474   // Check for allocation before we add nodes that would confuse
1475   // tightly_coupled_allocation()
1476   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1477 
1478   // Figure out the size and type of the elements we will be copying.
1479   const Type* src_type = src->Value(&_gvn);
1480   const Type* dst_type = dst->Value(&_gvn);
1481   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1482   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1483   assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1484          (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1485          "Unsupported array types for inline_string_copy");
1486 
1487   src = must_be_not_null(src, true);
1488   dst = must_be_not_null(dst, true);
1489 
1490   // Convert char[] offsets to byte[] offsets
1491   bool convert_src = (compress && src_elem == T_BYTE);
1492   bool convert_dst = (!compress && dst_elem == T_BYTE);
1493   if (convert_src) {
1494     src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1495   } else if (convert_dst) {
1496     dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1497   }
1498 
1499   // Range checks
1500   generate_string_range_check(src, src_offset, length, convert_src);
1501   generate_string_range_check(dst, dst_offset, length, convert_dst);
1502   if (stopped()) {
1503     return true;
1504   }
1505 
1506   src = access_resolve(src, ACCESS_READ);
1507   dst = access_resolve(dst, ACCESS_WRITE);
1508 
1509   Node* src_start = array_element_address(src, src_offset, src_elem);
1510   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1511   // 'src_start' points to src array + scaled offset
1512   // 'dst_start' points to dst array + scaled offset
1513   Node* count = NULL;
1514   if (compress) {
1515     count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1516   } else {
1517     inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1518   }
1519 
1520   if (alloc != NULL) {
1521     if (alloc->maybe_set_complete(&_gvn)) {
1522       // "You break it, you buy it."
1523       InitializeNode* init = alloc->initialization();
1524       assert(init->is_complete(), "we just did this");
1525       init->set_complete_with_arraycopy();
1526       assert(dst->is_CheckCastPP(), "sanity");
1527       assert(dst->in(0)->in(0) == init, "dest pinned");
1528     }
1529     // Do not let stores that initialize this object be reordered with
1530     // a subsequent store that would make this object accessible by
1531     // other threads.
1532     // Record what AllocateNode this StoreStore protects so that
1533     // escape analysis can go from the MemBarStoreStoreNode to the
1534     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1535     // based on the escape status of the AllocateNode.
1536     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1537   }
1538   if (compress) {
1539     set_result(_gvn.transform(count));
1540   }
1541   clear_upper_avx();
1542 
1543   return true;
1544 }
1545 
1546 #ifdef _LP64
1547 #define XTOP ,top() /*additional argument*/
1548 #else  //_LP64
1549 #define XTOP        /*no additional argument*/
1550 #endif //_LP64
1551 
1552 //------------------------inline_string_toBytesU--------------------------
1553 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1554 bool LibraryCallKit::inline_string_toBytesU() {
1555   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1556     return false;
1557   }
1558   // Get the arguments.
1559   Node* value     = argument(0);
1560   Node* offset    = argument(1);
1561   Node* length    = argument(2);
1562 
1563   Node* newcopy = NULL;
1564 
1565   // Set the original stack and the reexecute bit for the interpreter to reexecute
1566   // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1567   { PreserveReexecuteState preexecs(this);
1568     jvms()->set_should_reexecute(true);
1569 
1570     // Check if a null path was taken unconditionally.
1571     value = null_check(value);
1572 
1573     RegionNode* bailout = new RegionNode(1);
1574     record_for_igvn(bailout);
1575 
1576     // Range checks
1577     generate_negative_guard(offset, bailout);
1578     generate_negative_guard(length, bailout);
1579     generate_limit_guard(offset, length, load_array_length(value), bailout);
1580     // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1581     generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1582 
1583     if (bailout->req() > 1) {
1584       PreserveJVMState pjvms(this);
1585       set_control(_gvn.transform(bailout));
1586       uncommon_trap(Deoptimization::Reason_intrinsic,
1587                     Deoptimization::Action_maybe_recompile);
1588     }
1589     if (stopped()) {
1590       return true;
1591     }
1592 
1593     Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1594     Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1595     newcopy = new_array(klass_node, size, 0);  // no arguments to push
1596     AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1597 
1598     // Calculate starting addresses.
1599     value = access_resolve(value, ACCESS_READ);
1600     Node* src_start = array_element_address(value, offset, T_CHAR);
1601     Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1602 
1603     // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1604     const TypeInt* toffset = gvn().type(offset)->is_int();
1605     bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1606 
1607     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1608     const char* copyfunc_name = "arraycopy";
1609     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1610     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1611                       OptoRuntime::fast_arraycopy_Type(),
1612                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1613                       src_start, dst_start, ConvI2X(length) XTOP);
1614     // Do not let reads from the cloned object float above the arraycopy.
1615     if (alloc != NULL) {
1616       if (alloc->maybe_set_complete(&_gvn)) {
1617         // "You break it, you buy it."
1618         InitializeNode* init = alloc->initialization();
1619         assert(init->is_complete(), "we just did this");
1620         init->set_complete_with_arraycopy();
1621         assert(newcopy->is_CheckCastPP(), "sanity");
1622         assert(newcopy->in(0)->in(0) == init, "dest pinned");
1623       }
1624       // Do not let stores that initialize this object be reordered with
1625       // a subsequent store that would make this object accessible by
1626       // other threads.
1627       // Record what AllocateNode this StoreStore protects so that
1628       // escape analysis can go from the MemBarStoreStoreNode to the
1629       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1630       // based on the escape status of the AllocateNode.
1631       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1632     } else {
1633       insert_mem_bar(Op_MemBarCPUOrder);
1634     }
1635   } // original reexecute is set back here
1636 
1637   C->set_has_split_ifs(true); // Has chance for split-if optimization
1638   if (!stopped()) {
1639     set_result(newcopy);
1640   }
1641   clear_upper_avx();
1642 
1643   return true;
1644 }
1645 
1646 //------------------------inline_string_getCharsU--------------------------
1647 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1648 bool LibraryCallKit::inline_string_getCharsU() {
1649   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1650     return false;
1651   }
1652 
1653   // Get the arguments.
1654   Node* src       = argument(0);
1655   Node* src_begin = argument(1);
1656   Node* src_end   = argument(2); // exclusive offset (i < src_end)
1657   Node* dst       = argument(3);
1658   Node* dst_begin = argument(4);
1659 
1660   // Check for allocation before we add nodes that would confuse
1661   // tightly_coupled_allocation()
1662   AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1663 
1664   // Check if a null path was taken unconditionally.
1665   src = null_check(src);
1666   dst = null_check(dst);
1667   if (stopped()) {
1668     return true;
1669   }
1670 
1671   // Get length and convert char[] offset to byte[] offset
1672   Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1673   src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1674 
1675   // Range checks
1676   generate_string_range_check(src, src_begin, length, true);
1677   generate_string_range_check(dst, dst_begin, length, false);
1678   if (stopped()) {
1679     return true;
1680   }
1681 
1682   if (!stopped()) {
1683     src = access_resolve(src, ACCESS_READ);
1684     dst = access_resolve(dst, ACCESS_WRITE);
1685 
1686     // Calculate starting addresses.
1687     Node* src_start = array_element_address(src, src_begin, T_BYTE);
1688     Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1689 
1690     // Check if array addresses are aligned to HeapWordSize
1691     const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1692     const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1693     bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1694                    tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1695 
1696     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1697     const char* copyfunc_name = "arraycopy";
1698     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1699     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1700                       OptoRuntime::fast_arraycopy_Type(),
1701                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1702                       src_start, dst_start, ConvI2X(length) XTOP);
1703     // Do not let reads from the cloned object float above the arraycopy.
1704     if (alloc != NULL) {
1705       if (alloc->maybe_set_complete(&_gvn)) {
1706         // "You break it, you buy it."
1707         InitializeNode* init = alloc->initialization();
1708         assert(init->is_complete(), "we just did this");
1709         init->set_complete_with_arraycopy();
1710         assert(dst->is_CheckCastPP(), "sanity");
1711         assert(dst->in(0)->in(0) == init, "dest pinned");
1712       }
1713       // Do not let stores that initialize this object be reordered with
1714       // a subsequent store that would make this object accessible by
1715       // other threads.
1716       // Record what AllocateNode this StoreStore protects so that
1717       // escape analysis can go from the MemBarStoreStoreNode to the
1718       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1719       // based on the escape status of the AllocateNode.
1720       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1721     } else {
1722       insert_mem_bar(Op_MemBarCPUOrder);
1723     }
1724   }
1725 
1726   C->set_has_split_ifs(true); // Has chance for split-if optimization
1727   return true;
1728 }
1729 
1730 //----------------------inline_string_char_access----------------------------
1731 // Store/Load char to/from byte[] array.
1732 // static void StringUTF16.putChar(byte[] val, int index, int c)
1733 // static char StringUTF16.getChar(byte[] val, int index)
1734 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1735   Node* value  = argument(0);
1736   Node* index  = argument(1);
1737   Node* ch = is_store ? argument(2) : NULL;
1738 
1739   // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1740   // correctly requires matched array shapes.
1741   assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1742           "sanity: byte[] and char[] bases agree");
1743   assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1744           "sanity: byte[] and char[] scales agree");
1745 
1746   // Bail when getChar over constants is requested: constant folding would
1747   // reject folding mismatched char access over byte[]. A normal inlining for getChar
1748   // Java method would constant fold nicely instead.
1749   if (!is_store && value->is_Con() && index->is_Con()) {
1750     return false;
1751   }
1752 
1753   value = must_be_not_null(value, true);
1754   value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1755 
1756   Node* adr = array_element_address(value, index, T_CHAR);
1757   if (adr->is_top()) {
1758     return false;
1759   }
1760   if (is_store) {
1761     access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1762   } else {
1763     ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1764     set_result(ch);
1765   }
1766   return true;
1767 }
1768 
1769 //--------------------------round_double_node--------------------------------
1770 // Round a double node if necessary.
1771 Node* LibraryCallKit::round_double_node(Node* n) {
1772   if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1773     n = _gvn.transform(new RoundDoubleNode(0, n));
1774   return n;
1775 }
1776 
1777 //------------------------------inline_math-----------------------------------
1778 // public static double Math.abs(double)
1779 // public static double Math.sqrt(double)
1780 // public static double Math.log(double)
1781 // public static double Math.log10(double)
1782 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1783   Node* arg = round_double_node(argument(0));
1784   Node* n = NULL;
1785   switch (id) {
1786   case vmIntrinsics::_dabs:   n = new AbsDNode(                arg);  break;
1787   case vmIntrinsics::_dsqrt:  n = new SqrtDNode(C, control(),  arg);  break;
1788   default:  fatal_unexpected_iid(id);  break;
1789   }
1790   set_result(_gvn.transform(n));
1791   return true;
1792 }
1793 
1794 //------------------------------runtime_math-----------------------------
1795 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1796   assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1797          "must be (DD)D or (D)D type");
1798 
1799   // Inputs
1800   Node* a = round_double_node(argument(0));
1801   Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1802 
1803   const TypePtr* no_memory_effects = NULL;
1804   Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1805                                  no_memory_effects,
1806                                  a, top(), b, b ? top() : NULL);
1807   Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1808 #ifdef ASSERT
1809   Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1810   assert(value_top == top(), "second value must be top");
1811 #endif
1812 
1813   set_result(value);
1814   return true;
1815 }
1816 
1817 //------------------------------inline_math_native-----------------------------
1818 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1819 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1820   switch (id) {
1821     // These intrinsics are not properly supported on all hardware
1822   case vmIntrinsics::_dsin:
1823     return StubRoutines::dsin() != NULL ?
1824       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1825       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
1826   case vmIntrinsics::_dcos:
1827     return StubRoutines::dcos() != NULL ?
1828       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1829       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
1830   case vmIntrinsics::_dtan:
1831     return StubRoutines::dtan() != NULL ?
1832       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1833       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1834   case vmIntrinsics::_dlog:
1835     return StubRoutines::dlog() != NULL ?
1836       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1837       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
1838   case vmIntrinsics::_dlog10:
1839     return StubRoutines::dlog10() != NULL ?
1840       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1841       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1842 
1843     // These intrinsics are supported on all hardware
1844   case vmIntrinsics::_dsqrt:  return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1845   case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_math(id) : false;
1846 
1847   case vmIntrinsics::_dexp:
1848     return StubRoutines::dexp() != NULL ?
1849       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(),  "dexp") :
1850       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp),  "EXP");
1851   case vmIntrinsics::_dpow: {
1852     Node* exp = round_double_node(argument(2));
1853     const TypeD* d = _gvn.type(exp)->isa_double_constant();
1854     if (d != NULL && d->getd() == 2.0) {
1855       // Special case: pow(x, 2.0) => x * x
1856       Node* base = round_double_node(argument(0));
1857       set_result(_gvn.transform(new MulDNode(base, base)));
1858       return true;
1859     }
1860     return StubRoutines::dpow() != NULL ?
1861       runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(),  "dpow") :
1862       runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow),  "POW");
1863   }
1864 #undef FN_PTR
1865 
1866    // These intrinsics are not yet correctly implemented
1867   case vmIntrinsics::_datan2:
1868     return false;
1869 
1870   default:
1871     fatal_unexpected_iid(id);
1872     return false;
1873   }
1874 }
1875 
1876 static bool is_simple_name(Node* n) {
1877   return (n->req() == 1         // constant
1878           || (n->is_Type() && n->as_Type()->type()->singleton())
1879           || n->is_Proj()       // parameter or return value
1880           || n->is_Phi()        // local of some sort
1881           );
1882 }
1883 
1884 //----------------------------inline_notify-----------------------------------*
1885 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1886   const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1887   address func;
1888   if (id == vmIntrinsics::_notify) {
1889     func = OptoRuntime::monitor_notify_Java();
1890   } else {
1891     func = OptoRuntime::monitor_notifyAll_Java();
1892   }
1893   Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1894   make_slow_call_ex(call, env()->Throwable_klass(), false);
1895   return true;
1896 }
1897 
1898 
1899 //----------------------------inline_min_max-----------------------------------
1900 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1901   set_result(generate_min_max(id, argument(0), argument(1)));
1902   return true;
1903 }
1904 
1905 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1906   Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1907   IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1908   Node* fast_path = _gvn.transform( new IfFalseNode(check));
1909   Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1910 
1911   {
1912     PreserveJVMState pjvms(this);
1913     PreserveReexecuteState preexecs(this);
1914     jvms()->set_should_reexecute(true);
1915 
1916     set_control(slow_path);
1917     set_i_o(i_o());
1918 
1919     uncommon_trap(Deoptimization::Reason_intrinsic,
1920                   Deoptimization::Action_none);
1921   }
1922 
1923   set_control(fast_path);
1924   set_result(math);
1925 }
1926 
1927 template <typename OverflowOp>
1928 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1929   typedef typename OverflowOp::MathOp MathOp;
1930 
1931   MathOp* mathOp = new MathOp(arg1, arg2);
1932   Node* operation = _gvn.transform( mathOp );
1933   Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1934   inline_math_mathExact(operation, ofcheck);
1935   return true;
1936 }
1937 
1938 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1939   return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1940 }
1941 
1942 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1943   return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1944 }
1945 
1946 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1947   return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1948 }
1949 
1950 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1951   return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1952 }
1953 
1954 bool LibraryCallKit::inline_math_negateExactI() {
1955   return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1956 }
1957 
1958 bool LibraryCallKit::inline_math_negateExactL() {
1959   return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1960 }
1961 
1962 bool LibraryCallKit::inline_math_multiplyExactI() {
1963   return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1964 }
1965 
1966 bool LibraryCallKit::inline_math_multiplyExactL() {
1967   return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1968 }
1969 
1970 bool LibraryCallKit::inline_math_multiplyHigh() {
1971   set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1972   return true;
1973 }
1974 
1975 Node*
1976 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1977   // These are the candidate return value:
1978   Node* xvalue = x0;
1979   Node* yvalue = y0;
1980 
1981   if (xvalue == yvalue) {
1982     return xvalue;
1983   }
1984 
1985   bool want_max = (id == vmIntrinsics::_max);
1986 
1987   const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1988   const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1989   if (txvalue == NULL || tyvalue == NULL)  return top();
1990   // This is not really necessary, but it is consistent with a
1991   // hypothetical MaxINode::Value method:
1992   int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1993 
1994   // %%% This folding logic should (ideally) be in a different place.
1995   // Some should be inside IfNode, and there to be a more reliable
1996   // transformation of ?: style patterns into cmoves.  We also want
1997   // more powerful optimizations around cmove and min/max.
1998 
1999   // Try to find a dominating comparison of these guys.
2000   // It can simplify the index computation for Arrays.copyOf
2001   // and similar uses of System.arraycopy.
2002   // First, compute the normalized version of CmpI(x, y).
2003   int   cmp_op = Op_CmpI;
2004   Node* xkey = xvalue;
2005   Node* ykey = yvalue;
2006   Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2007   if (ideal_cmpxy->is_Cmp()) {
2008     // E.g., if we have CmpI(length - offset, count),
2009     // it might idealize to CmpI(length, count + offset)
2010     cmp_op = ideal_cmpxy->Opcode();
2011     xkey = ideal_cmpxy->in(1);
2012     ykey = ideal_cmpxy->in(2);
2013   }
2014 
2015   // Start by locating any relevant comparisons.
2016   Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2017   Node* cmpxy = NULL;
2018   Node* cmpyx = NULL;
2019   for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2020     Node* cmp = start_from->fast_out(k);
2021     if (cmp->outcnt() > 0 &&            // must have prior uses
2022         cmp->in(0) == NULL &&           // must be context-independent
2023         cmp->Opcode() == cmp_op) {      // right kind of compare
2024       if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
2025       if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
2026     }
2027   }
2028 
2029   const int NCMPS = 2;
2030   Node* cmps[NCMPS] = { cmpxy, cmpyx };
2031   int cmpn;
2032   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2033     if (cmps[cmpn] != NULL)  break;     // find a result
2034   }
2035   if (cmpn < NCMPS) {
2036     // Look for a dominating test that tells us the min and max.
2037     int depth = 0;                // Limit search depth for speed
2038     Node* dom = control();
2039     for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2040       if (++depth >= 100)  break;
2041       Node* ifproj = dom;
2042       if (!ifproj->is_Proj())  continue;
2043       Node* iff = ifproj->in(0);
2044       if (!iff->is_If())  continue;
2045       Node* bol = iff->in(1);
2046       if (!bol->is_Bool())  continue;
2047       Node* cmp = bol->in(1);
2048       if (cmp == NULL)  continue;
2049       for (cmpn = 0; cmpn < NCMPS; cmpn++)
2050         if (cmps[cmpn] == cmp)  break;
2051       if (cmpn == NCMPS)  continue;
2052       BoolTest::mask btest = bol->as_Bool()->_test._test;
2053       if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
2054       if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
2055       // At this point, we know that 'x btest y' is true.
2056       switch (btest) {
2057       case BoolTest::eq:
2058         // They are proven equal, so we can collapse the min/max.
2059         // Either value is the answer.  Choose the simpler.
2060         if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2061           return yvalue;
2062         return xvalue;
2063       case BoolTest::lt:          // x < y
2064       case BoolTest::le:          // x <= y
2065         return (want_max ? yvalue : xvalue);
2066       case BoolTest::gt:          // x > y
2067       case BoolTest::ge:          // x >= y
2068         return (want_max ? xvalue : yvalue);
2069       default:
2070         break;
2071       }
2072     }
2073   }
2074 
2075   // We failed to find a dominating test.
2076   // Let's pick a test that might GVN with prior tests.
2077   Node*          best_bol   = NULL;
2078   BoolTest::mask best_btest = BoolTest::illegal;
2079   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2080     Node* cmp = cmps[cmpn];
2081     if (cmp == NULL)  continue;
2082     for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2083       Node* bol = cmp->fast_out(j);
2084       if (!bol->is_Bool())  continue;
2085       BoolTest::mask btest = bol->as_Bool()->_test._test;
2086       if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
2087       if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
2088       if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2089         best_bol   = bol->as_Bool();
2090         best_btest = btest;
2091       }
2092     }
2093   }
2094 
2095   Node* answer_if_true  = NULL;
2096   Node* answer_if_false = NULL;
2097   switch (best_btest) {
2098   default:
2099     if (cmpxy == NULL)
2100       cmpxy = ideal_cmpxy;
2101     best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2102     // and fall through:
2103   case BoolTest::lt:          // x < y
2104   case BoolTest::le:          // x <= y
2105     answer_if_true  = (want_max ? yvalue : xvalue);
2106     answer_if_false = (want_max ? xvalue : yvalue);
2107     break;
2108   case BoolTest::gt:          // x > y
2109   case BoolTest::ge:          // x >= y
2110     answer_if_true  = (want_max ? xvalue : yvalue);
2111     answer_if_false = (want_max ? yvalue : xvalue);
2112     break;
2113   }
2114 
2115   jint hi, lo;
2116   if (want_max) {
2117     // We can sharpen the minimum.
2118     hi = MAX2(txvalue->_hi, tyvalue->_hi);
2119     lo = MAX2(txvalue->_lo, tyvalue->_lo);
2120   } else {
2121     // We can sharpen the maximum.
2122     hi = MIN2(txvalue->_hi, tyvalue->_hi);
2123     lo = MIN2(txvalue->_lo, tyvalue->_lo);
2124   }
2125 
2126   // Use a flow-free graph structure, to avoid creating excess control edges
2127   // which could hinder other optimizations.
2128   // Since Math.min/max is often used with arraycopy, we want
2129   // tightly_coupled_allocation to be able to see beyond min/max expressions.
2130   Node* cmov = CMoveNode::make(NULL, best_bol,
2131                                answer_if_false, answer_if_true,
2132                                TypeInt::make(lo, hi, widen));
2133 
2134   return _gvn.transform(cmov);
2135 
2136   /*
2137   // This is not as desirable as it may seem, since Min and Max
2138   // nodes do not have a full set of optimizations.
2139   // And they would interfere, anyway, with 'if' optimizations
2140   // and with CMoveI canonical forms.
2141   switch (id) {
2142   case vmIntrinsics::_min:
2143     result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2144   case vmIntrinsics::_max:
2145     result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2146   default:
2147     ShouldNotReachHere();
2148   }
2149   */
2150 }
2151 
2152 inline int
2153 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2154   const TypePtr* base_type = TypePtr::NULL_PTR;
2155   if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
2156   if (base_type == NULL) {
2157     // Unknown type.
2158     return Type::AnyPtr;
2159   } else if (base_type == TypePtr::NULL_PTR) {
2160     // Since this is a NULL+long form, we have to switch to a rawptr.
2161     base   = _gvn.transform(new CastX2PNode(offset));
2162     offset = MakeConX(0);
2163     return Type::RawPtr;
2164   } else if (base_type->base() == Type::RawPtr) {
2165     return Type::RawPtr;
2166   } else if (base_type->isa_oopptr()) {
2167     // Base is never null => always a heap address.
2168     if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2169       return Type::OopPtr;
2170     }
2171     // Offset is small => always a heap address.
2172     const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2173     if (offset_type != NULL &&
2174         base_type->offset() == 0 &&     // (should always be?)
2175         offset_type->_lo >= 0 &&
2176         !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2177       return Type::OopPtr;
2178     } else if (type == T_OBJECT) {
2179       // off heap access to an oop doesn't make any sense. Has to be on
2180       // heap.
2181       return Type::OopPtr;
2182     }
2183     // Otherwise, it might either be oop+off or NULL+addr.
2184     return Type::AnyPtr;
2185   } else {
2186     // No information:
2187     return Type::AnyPtr;
2188   }
2189 }
2190 
2191 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2192   Node* uncasted_base = base;
2193   int kind = classify_unsafe_addr(uncasted_base, offset, type);
2194   if (kind == Type::RawPtr) {
2195     return basic_plus_adr(top(), uncasted_base, offset);
2196   } else if (kind == Type::AnyPtr) {
2197     assert(base == uncasted_base, "unexpected base change");
2198     if (can_cast) {
2199       if (!_gvn.type(base)->speculative_maybe_null() &&
2200           !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2201         // According to profiling, this access is always on
2202         // heap. Casting the base to not null and thus avoiding membars
2203         // around the access should allow better optimizations
2204         Node* null_ctl = top();
2205         base = null_check_oop(base, &null_ctl, true, true, true);
2206         assert(null_ctl->is_top(), "no null control here");
2207         return basic_plus_adr(base, offset);
2208       } else if (_gvn.type(base)->speculative_always_null() &&
2209                  !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2210         // According to profiling, this access is always off
2211         // heap.
2212         base = null_assert(base);
2213         Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2214         offset = MakeConX(0);
2215         return basic_plus_adr(top(), raw_base, offset);
2216       }
2217     }
2218     // We don't know if it's an on heap or off heap access. Fall back
2219     // to raw memory access.
2220     base = access_resolve(base, decorators);
2221     Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2222     return basic_plus_adr(top(), raw, offset);
2223   } else {
2224     assert(base == uncasted_base, "unexpected base change");
2225     // We know it's an on heap access so base can't be null
2226     if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2227       base = must_be_not_null(base, true);
2228     }
2229     return basic_plus_adr(base, offset);
2230   }
2231 }
2232 
2233 //--------------------------inline_number_methods-----------------------------
2234 // inline int     Integer.numberOfLeadingZeros(int)
2235 // inline int        Long.numberOfLeadingZeros(long)
2236 //
2237 // inline int     Integer.numberOfTrailingZeros(int)
2238 // inline int        Long.numberOfTrailingZeros(long)
2239 //
2240 // inline int     Integer.bitCount(int)
2241 // inline int        Long.bitCount(long)
2242 //
2243 // inline char  Character.reverseBytes(char)
2244 // inline short     Short.reverseBytes(short)
2245 // inline int     Integer.reverseBytes(int)
2246 // inline long       Long.reverseBytes(long)
2247 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2248   Node* arg = argument(0);
2249   Node* n = NULL;
2250   switch (id) {
2251   case vmIntrinsics::_numberOfLeadingZeros_i:   n = new CountLeadingZerosINode( arg);  break;
2252   case vmIntrinsics::_numberOfLeadingZeros_l:   n = new CountLeadingZerosLNode( arg);  break;
2253   case vmIntrinsics::_numberOfTrailingZeros_i:  n = new CountTrailingZerosINode(arg);  break;
2254   case vmIntrinsics::_numberOfTrailingZeros_l:  n = new CountTrailingZerosLNode(arg);  break;
2255   case vmIntrinsics::_bitCount_i:               n = new PopCountINode(          arg);  break;
2256   case vmIntrinsics::_bitCount_l:               n = new PopCountLNode(          arg);  break;
2257   case vmIntrinsics::_reverseBytes_c:           n = new ReverseBytesUSNode(0,   arg);  break;
2258   case vmIntrinsics::_reverseBytes_s:           n = new ReverseBytesSNode( 0,   arg);  break;
2259   case vmIntrinsics::_reverseBytes_i:           n = new ReverseBytesINode( 0,   arg);  break;
2260   case vmIntrinsics::_reverseBytes_l:           n = new ReverseBytesLNode( 0,   arg);  break;
2261   default:  fatal_unexpected_iid(id);  break;
2262   }
2263   set_result(_gvn.transform(n));
2264   return true;
2265 }
2266 
2267 //----------------------------inline_unsafe_access----------------------------
2268 
2269 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2270   // Attempt to infer a sharper value type from the offset and base type.
2271   ciKlass* sharpened_klass = NULL;
2272 
2273   // See if it is an instance field, with an object type.
2274   if (alias_type->field() != NULL) {
2275     if (alias_type->field()->type()->is_klass()) {
2276       sharpened_klass = alias_type->field()->type()->as_klass();
2277     }
2278   }
2279 
2280   // See if it is a narrow oop array.
2281   if (adr_type->isa_aryptr()) {
2282     if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2283       const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2284       if (elem_type != NULL) {
2285         sharpened_klass = elem_type->klass();
2286       }
2287     }
2288   }
2289 
2290   // The sharpened class might be unloaded if there is no class loader
2291   // contraint in place.
2292   if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2293     const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2294 
2295 #ifndef PRODUCT
2296     if (C->print_intrinsics() || C->print_inlining()) {
2297       tty->print("  from base type:  ");  adr_type->dump(); tty->cr();
2298       tty->print("  sharpened value: ");  tjp->dump();      tty->cr();
2299     }
2300 #endif
2301     // Sharpen the value type.
2302     return tjp;
2303   }
2304   return NULL;
2305 }
2306 
2307 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2308   switch (kind) {
2309       case Relaxed:
2310         return MO_UNORDERED;
2311       case Opaque:
2312         return MO_RELAXED;
2313       case Acquire:
2314         return MO_ACQUIRE;
2315       case Release:
2316         return MO_RELEASE;
2317       case Volatile:
2318         return MO_SEQ_CST;
2319       default:
2320         ShouldNotReachHere();
2321         return 0;
2322   }
2323 }
2324 
2325 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2326   if (callee()->is_static())  return false;  // caller must have the capability!
2327   DecoratorSet decorators = C2_UNSAFE_ACCESS;
2328   guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2329   guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2330   assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2331 
2332   if (type == T_OBJECT || type == T_ARRAY) {
2333     decorators |= ON_UNKNOWN_OOP_REF;
2334   }
2335 
2336   if (unaligned) {
2337     decorators |= C2_UNALIGNED;
2338   }
2339 
2340 #ifndef PRODUCT
2341   {
2342     ResourceMark rm;
2343     // Check the signatures.
2344     ciSignature* sig = callee()->signature();
2345 #ifdef ASSERT
2346     if (!is_store) {
2347       // Object getReference(Object base, int/long offset), etc.
2348       BasicType rtype = sig->return_type()->basic_type();
2349       assert(rtype == type, "getter must return the expected value");
2350       assert(sig->count() == 2, "oop getter has 2 arguments");
2351       assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2352       assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2353     } else {
2354       // void putReference(Object base, int/long offset, Object x), etc.
2355       assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2356       assert(sig->count() == 3, "oop putter has 3 arguments");
2357       assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2358       assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2359       BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2360       assert(vtype == type, "putter must accept the expected value");
2361     }
2362 #endif // ASSERT
2363  }
2364 #endif //PRODUCT
2365 
2366   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2367 
2368   Node* receiver = argument(0);  // type: oop
2369 
2370   // Build address expression.
2371   Node* adr;
2372   Node* heap_base_oop = top();
2373   Node* offset = top();
2374   Node* val;
2375 
2376   // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2377   Node* base = argument(1);  // type: oop
2378   // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2379   offset = argument(2);  // type: long
2380   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2381   // to be plain byte offsets, which are also the same as those accepted
2382   // by oopDesc::field_addr.
2383   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2384          "fieldOffset must be byte-scaled");
2385   // 32-bit machines ignore the high half!
2386   offset = ConvL2X(offset);
2387   adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2388 
2389   if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2390     heap_base_oop = base;
2391   } else if (type == T_OBJECT) {
2392     return false; // off-heap oop accesses are not supported
2393   }
2394 
2395   // Can base be NULL? Otherwise, always on-heap access.
2396   bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2397 
2398   if (!can_access_non_heap) {
2399     decorators |= IN_HEAP;
2400   }
2401 
2402   val = is_store ? argument(4) : NULL;
2403 
2404   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2405 
2406   // Try to categorize the address.
2407   Compile::AliasType* alias_type = C->alias_type(adr_type);
2408   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2409 
2410   if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2411       alias_type->adr_type() == TypeAryPtr::RANGE) {
2412     return false; // not supported
2413   }
2414 
2415   bool mismatched = false;
2416   BasicType bt = alias_type->basic_type();
2417   if (bt != T_ILLEGAL) {
2418     assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2419     if (bt == T_BYTE && adr_type->isa_aryptr()) {
2420       // Alias type doesn't differentiate between byte[] and boolean[]).
2421       // Use address type to get the element type.
2422       bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2423     }
2424     if (bt == T_ARRAY || bt == T_NARROWOOP) {
2425       // accessing an array field with getReference is not a mismatch
2426       bt = T_OBJECT;
2427     }
2428     if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2429       // Don't intrinsify mismatched object accesses
2430       return false;
2431     }
2432     mismatched = (bt != type);
2433   } else if (alias_type->adr_type()->isa_oopptr()) {
2434     mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2435   }
2436 
2437   assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2438 
2439   if (mismatched) {
2440     decorators |= C2_MISMATCHED;
2441   }
2442 
2443   // First guess at the value type.
2444   const Type *value_type = Type::get_const_basic_type(type);
2445 
2446   // Figure out the memory ordering.
2447   decorators |= mo_decorator_for_access_kind(kind);
2448 
2449   if (!is_store && type == T_OBJECT) {
2450     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2451     if (tjp != NULL) {
2452       value_type = tjp;
2453     }
2454   }
2455 
2456   receiver = null_check(receiver);
2457   if (stopped()) {
2458     return true;
2459   }
2460   // Heap pointers get a null-check from the interpreter,
2461   // as a courtesy.  However, this is not guaranteed by Unsafe,
2462   // and it is not possible to fully distinguish unintended nulls
2463   // from intended ones in this API.
2464 
2465   if (!is_store) {
2466     Node* p = NULL;
2467     // Try to constant fold a load from a constant field
2468     ciField* field = alias_type->field();
2469     if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2470       // final or stable field
2471       p = make_constant_from_field(field, heap_base_oop);
2472     }
2473 
2474     if (p == NULL) { // Could not constant fold the load
2475       p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2476       // Normalize the value returned by getBoolean in the following cases
2477       if (type == T_BOOLEAN &&
2478           (mismatched ||
2479            heap_base_oop == top() ||                  // - heap_base_oop is NULL or
2480            (can_access_non_heap && field == NULL))    // - heap_base_oop is potentially NULL
2481                                                       //   and the unsafe access is made to large offset
2482                                                       //   (i.e., larger than the maximum offset necessary for any
2483                                                       //   field access)
2484             ) {
2485           IdealKit ideal = IdealKit(this);
2486 #define __ ideal.
2487           IdealVariable normalized_result(ideal);
2488           __ declarations_done();
2489           __ set(normalized_result, p);
2490           __ if_then(p, BoolTest::ne, ideal.ConI(0));
2491           __ set(normalized_result, ideal.ConI(1));
2492           ideal.end_if();
2493           final_sync(ideal);
2494           p = __ value(normalized_result);
2495 #undef __
2496       }
2497     }
2498     if (type == T_ADDRESS) {
2499       p = gvn().transform(new CastP2XNode(NULL, p));
2500       p = ConvX2UL(p);
2501     }
2502     // The load node has the control of the preceding MemBarCPUOrder.  All
2503     // following nodes will have the control of the MemBarCPUOrder inserted at
2504     // the end of this method.  So, pushing the load onto the stack at a later
2505     // point is fine.
2506     set_result(p);
2507   } else {
2508     if (bt == T_ADDRESS) {
2509       // Repackage the long as a pointer.
2510       val = ConvL2X(val);
2511       val = gvn().transform(new CastX2PNode(val));
2512     }
2513     access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2514   }
2515 
2516   return true;
2517 }
2518 
2519 //----------------------------inline_unsafe_load_store----------------------------
2520 // This method serves a couple of different customers (depending on LoadStoreKind):
2521 //
2522 // LS_cmp_swap:
2523 //
2524 //   boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2525 //   boolean compareAndSetInt(   Object o, long offset, int    expected, int    x);
2526 //   boolean compareAndSetLong(  Object o, long offset, long   expected, long   x);
2527 //
2528 // LS_cmp_swap_weak:
2529 //
2530 //   boolean weakCompareAndSetReference(       Object o, long offset, Object expected, Object x);
2531 //   boolean weakCompareAndSetReferencePlain(  Object o, long offset, Object expected, Object x);
2532 //   boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2533 //   boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2534 //
2535 //   boolean weakCompareAndSetInt(          Object o, long offset, int    expected, int    x);
2536 //   boolean weakCompareAndSetIntPlain(     Object o, long offset, int    expected, int    x);
2537 //   boolean weakCompareAndSetIntAcquire(   Object o, long offset, int    expected, int    x);
2538 //   boolean weakCompareAndSetIntRelease(   Object o, long offset, int    expected, int    x);
2539 //
2540 //   boolean weakCompareAndSetLong(         Object o, long offset, long   expected, long   x);
2541 //   boolean weakCompareAndSetLongPlain(    Object o, long offset, long   expected, long   x);
2542 //   boolean weakCompareAndSetLongAcquire(  Object o, long offset, long   expected, long   x);
2543 //   boolean weakCompareAndSetLongRelease(  Object o, long offset, long   expected, long   x);
2544 //
2545 // LS_cmp_exchange:
2546 //
2547 //   Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2548 //   Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2549 //   Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2550 //
2551 //   Object compareAndExchangeIntVolatile(   Object o, long offset, Object expected, Object x);
2552 //   Object compareAndExchangeIntAcquire(    Object o, long offset, Object expected, Object x);
2553 //   Object compareAndExchangeIntRelease(    Object o, long offset, Object expected, Object x);
2554 //
2555 //   Object compareAndExchangeLongVolatile(  Object o, long offset, Object expected, Object x);
2556 //   Object compareAndExchangeLongAcquire(   Object o, long offset, Object expected, Object x);
2557 //   Object compareAndExchangeLongRelease(   Object o, long offset, Object expected, Object x);
2558 //
2559 // LS_get_add:
2560 //
2561 //   int  getAndAddInt( Object o, long offset, int  delta)
2562 //   long getAndAddLong(Object o, long offset, long delta)
2563 //
2564 // LS_get_set:
2565 //
2566 //   int    getAndSet(Object o, long offset, int    newValue)
2567 //   long   getAndSet(Object o, long offset, long   newValue)
2568 //   Object getAndSet(Object o, long offset, Object newValue)
2569 //
2570 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2571   // This basic scheme here is the same as inline_unsafe_access, but
2572   // differs in enough details that combining them would make the code
2573   // overly confusing.  (This is a true fact! I originally combined
2574   // them, but even I was confused by it!) As much code/comments as
2575   // possible are retained from inline_unsafe_access though to make
2576   // the correspondences clearer. - dl
2577 
2578   if (callee()->is_static())  return false;  // caller must have the capability!
2579 
2580   DecoratorSet decorators = C2_UNSAFE_ACCESS;
2581   decorators |= mo_decorator_for_access_kind(access_kind);
2582 
2583 #ifndef PRODUCT
2584   BasicType rtype;
2585   {
2586     ResourceMark rm;
2587     // Check the signatures.
2588     ciSignature* sig = callee()->signature();
2589     rtype = sig->return_type()->basic_type();
2590     switch(kind) {
2591       case LS_get_add:
2592       case LS_get_set: {
2593       // Check the signatures.
2594 #ifdef ASSERT
2595       assert(rtype == type, "get and set must return the expected type");
2596       assert(sig->count() == 3, "get and set has 3 arguments");
2597       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2598       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2599       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2600       assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2601 #endif // ASSERT
2602         break;
2603       }
2604       case LS_cmp_swap:
2605       case LS_cmp_swap_weak: {
2606       // Check the signatures.
2607 #ifdef ASSERT
2608       assert(rtype == T_BOOLEAN, "CAS must return boolean");
2609       assert(sig->count() == 4, "CAS has 4 arguments");
2610       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2611       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2612 #endif // ASSERT
2613         break;
2614       }
2615       case LS_cmp_exchange: {
2616       // Check the signatures.
2617 #ifdef ASSERT
2618       assert(rtype == type, "CAS must return the expected type");
2619       assert(sig->count() == 4, "CAS has 4 arguments");
2620       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2621       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2622 #endif // ASSERT
2623         break;
2624       }
2625       default:
2626         ShouldNotReachHere();
2627     }
2628   }
2629 #endif //PRODUCT
2630 
2631   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2632 
2633   // Get arguments:
2634   Node* receiver = NULL;
2635   Node* base     = NULL;
2636   Node* offset   = NULL;
2637   Node* oldval   = NULL;
2638   Node* newval   = NULL;
2639   switch(kind) {
2640     case LS_cmp_swap:
2641     case LS_cmp_swap_weak:
2642     case LS_cmp_exchange: {
2643       const bool two_slot_type = type2size[type] == 2;
2644       receiver = argument(0);  // type: oop
2645       base     = argument(1);  // type: oop
2646       offset   = argument(2);  // type: long
2647       oldval   = argument(4);  // type: oop, int, or long
2648       newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2649       break;
2650     }
2651     case LS_get_add:
2652     case LS_get_set: {
2653       receiver = argument(0);  // type: oop
2654       base     = argument(1);  // type: oop
2655       offset   = argument(2);  // type: long
2656       oldval   = NULL;
2657       newval   = argument(4);  // type: oop, int, or long
2658       break;
2659     }
2660     default:
2661       ShouldNotReachHere();
2662   }
2663 
2664   // Build field offset expression.
2665   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2666   // to be plain byte offsets, which are also the same as those accepted
2667   // by oopDesc::field_addr.
2668   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2669   // 32-bit machines ignore the high half of long offsets
2670   offset = ConvL2X(offset);
2671   Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2672   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2673 
2674   Compile::AliasType* alias_type = C->alias_type(adr_type);
2675   BasicType bt = alias_type->basic_type();
2676   if (bt != T_ILLEGAL &&
2677       ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2678     // Don't intrinsify mismatched object accesses.
2679     return false;
2680   }
2681 
2682   // For CAS, unlike inline_unsafe_access, there seems no point in
2683   // trying to refine types. Just use the coarse types here.
2684   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2685   const Type *value_type = Type::get_const_basic_type(type);
2686 
2687   switch (kind) {
2688     case LS_get_set:
2689     case LS_cmp_exchange: {
2690       if (type == T_OBJECT) {
2691         const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2692         if (tjp != NULL) {
2693           value_type = tjp;
2694         }
2695       }
2696       break;
2697     }
2698     case LS_cmp_swap:
2699     case LS_cmp_swap_weak:
2700     case LS_get_add:
2701       break;
2702     default:
2703       ShouldNotReachHere();
2704   }
2705 
2706   // Null check receiver.
2707   receiver = null_check(receiver);
2708   if (stopped()) {
2709     return true;
2710   }
2711 
2712   int alias_idx = C->get_alias_index(adr_type);
2713 
2714   if (type == T_OBJECT || type == T_ARRAY) {
2715     decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2716 
2717     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2718     // could be delayed during Parse (for example, in adjust_map_after_if()).
2719     // Execute transformation here to avoid barrier generation in such case.
2720     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2721       newval = _gvn.makecon(TypePtr::NULL_PTR);
2722 
2723     if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2724       // Refine the value to a null constant, when it is known to be null
2725       oldval = _gvn.makecon(TypePtr::NULL_PTR);
2726     }
2727   }
2728 
2729   Node* result = NULL;
2730   switch (kind) {
2731     case LS_cmp_exchange: {
2732       result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2733                                             oldval, newval, value_type, type, decorators);
2734       break;
2735     }
2736     case LS_cmp_swap_weak:
2737       decorators |= C2_WEAK_CMPXCHG;
2738     case LS_cmp_swap: {
2739       result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2740                                              oldval, newval, value_type, type, decorators);
2741       break;
2742     }
2743     case LS_get_set: {
2744       result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2745                                      newval, value_type, type, decorators);
2746       break;
2747     }
2748     case LS_get_add: {
2749       result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2750                                     newval, value_type, type, decorators);
2751       break;
2752     }
2753     default:
2754       ShouldNotReachHere();
2755   }
2756 
2757   assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2758   set_result(result);
2759   return true;
2760 }
2761 
2762 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2763   // Regardless of form, don't allow previous ld/st to move down,
2764   // then issue acquire, release, or volatile mem_bar.
2765   insert_mem_bar(Op_MemBarCPUOrder);
2766   switch(id) {
2767     case vmIntrinsics::_loadFence:
2768       insert_mem_bar(Op_LoadFence);
2769       return true;
2770     case vmIntrinsics::_storeFence:
2771       insert_mem_bar(Op_StoreFence);
2772       return true;
2773     case vmIntrinsics::_fullFence:
2774       insert_mem_bar(Op_MemBarVolatile);
2775       return true;
2776     default:
2777       fatal_unexpected_iid(id);
2778       return false;
2779   }
2780 }
2781 
2782 bool LibraryCallKit::inline_onspinwait() {
2783   insert_mem_bar(Op_OnSpinWait);
2784   return true;
2785 }
2786 
2787 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2788   if (!kls->is_Con()) {
2789     return true;
2790   }
2791   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2792   if (klsptr == NULL) {
2793     return true;
2794   }
2795   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2796   // don't need a guard for a klass that is already initialized
2797   return !ik->is_initialized();
2798 }
2799 
2800 //----------------------------inline_unsafe_allocate---------------------------
2801 // public native Object Unsafe.allocateInstance(Class<?> cls);
2802 bool LibraryCallKit::inline_unsafe_allocate() {
2803   if (callee()->is_static())  return false;  // caller must have the capability!
2804 
2805   null_check_receiver();  // null-check, then ignore
2806   Node* cls = null_check(argument(1));
2807   if (stopped())  return true;
2808 
2809   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2810   kls = null_check(kls);
2811   if (stopped())  return true;  // argument was like int.class
2812 
2813   Node* test = NULL;
2814   if (LibraryCallKit::klass_needs_init_guard(kls)) {
2815     // Note:  The argument might still be an illegal value like
2816     // Serializable.class or Object[].class.   The runtime will handle it.
2817     // But we must make an explicit check for initialization.
2818     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2819     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2820     // can generate code to load it as unsigned byte.
2821     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2822     Node* bits = intcon(InstanceKlass::fully_initialized);
2823     test = _gvn.transform(new SubINode(inst, bits));
2824     // The 'test' is non-zero if we need to take a slow path.
2825   }
2826 
2827   Node* obj = new_instance(kls, test);
2828   set_result(obj);
2829   return true;
2830 }
2831 
2832 //------------------------inline_native_time_funcs--------------
2833 // inline code for System.currentTimeMillis() and System.nanoTime()
2834 // these have the same type and signature
2835 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2836   const TypeFunc* tf = OptoRuntime::void_long_Type();
2837   const TypePtr* no_memory_effects = NULL;
2838   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2839   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2840 #ifdef ASSERT
2841   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2842   assert(value_top == top(), "second value must be top");
2843 #endif
2844   set_result(value);
2845   return true;
2846 }
2847 
2848 #ifdef JFR_HAVE_INTRINSICS
2849 
2850 /*
2851 * oop -> myklass
2852 * myklass->trace_id |= USED
2853 * return myklass->trace_id & ~0x3
2854 */
2855 bool LibraryCallKit::inline_native_classID() {
2856   Node* cls = null_check(argument(0), T_OBJECT);
2857   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2858   kls = null_check(kls, T_OBJECT);
2859 
2860   ByteSize offset = KLASS_TRACE_ID_OFFSET;
2861   Node* insp = basic_plus_adr(kls, in_bytes(offset));
2862   Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2863 
2864   Node* clsused = longcon(0x01l); // set the class bit
2865   Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2866   const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2867   store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2868 
2869 #ifdef TRACE_ID_META_BITS
2870   Node* mbits = longcon(~TRACE_ID_META_BITS);
2871   tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2872 #endif
2873 #ifdef TRACE_ID_SHIFT
2874   Node* cbits = intcon(TRACE_ID_SHIFT);
2875   tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2876 #endif
2877 
2878   set_result(tvalue);
2879   return true;
2880 
2881 }
2882 
2883 bool LibraryCallKit::inline_native_getEventWriter() {
2884   Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2885 
2886   Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2887                                   in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2888 
2889   Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2890 
2891   Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2892   Node* test_jobj_eq_null  = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2893 
2894   IfNode* iff_jobj_null =
2895     create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2896 
2897   enum { _normal_path = 1,
2898          _null_path = 2,
2899          PATH_LIMIT };
2900 
2901   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2902   PhiNode*    result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2903 
2904   Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2905   result_rgn->init_req(_null_path, jobj_is_null);
2906   result_val->init_req(_null_path, null());
2907 
2908   Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2909   set_control(jobj_is_not_null);
2910   Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2911                           IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2912   result_rgn->init_req(_normal_path, control());
2913   result_val->init_req(_normal_path, res);
2914 
2915   set_result(result_rgn, result_val);
2916 
2917   return true;
2918 }
2919 
2920 #endif // JFR_HAVE_INTRINSICS
2921 
2922 //------------------------inline_native_currentThread------------------
2923 bool LibraryCallKit::inline_native_currentThread() {
2924   Node* junk = NULL;
2925   set_result(generate_current_thread(junk));
2926   return true;
2927 }
2928 
2929 //------------------------inline_native_isInterrupted------------------
2930 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2931 bool LibraryCallKit::inline_native_isInterrupted() {
2932   // Add a fast path to t.isInterrupted(clear_int):
2933   //   (t == Thread.current() &&
2934   //    (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2935   //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2936   // So, in the common case that the interrupt bit is false,
2937   // we avoid making a call into the VM.  Even if the interrupt bit
2938   // is true, if the clear_int argument is false, we avoid the VM call.
2939   // However, if the receiver is not currentThread, we must call the VM,
2940   // because there must be some locking done around the operation.
2941 
2942   // We only go to the fast case code if we pass two guards.
2943   // Paths which do not pass are accumulated in the slow_region.
2944 
2945   enum {
2946     no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
2947     no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
2948     slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
2949     PATH_LIMIT
2950   };
2951 
2952   // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
2953   // out of the function.
2954   insert_mem_bar(Op_MemBarCPUOrder);
2955 
2956   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2957   PhiNode*    result_val = new PhiNode(result_rgn, TypeInt::BOOL);
2958 
2959   RegionNode* slow_region = new RegionNode(1);
2960   record_for_igvn(slow_region);
2961 
2962   // (a) Receiving thread must be the current thread.
2963   Node* rec_thr = argument(0);
2964   Node* tls_ptr = NULL;
2965   Node* cur_thr = generate_current_thread(tls_ptr);
2966 
2967   // Resolve oops to stable for CmpP below.
2968   cur_thr = access_resolve(cur_thr, 0);
2969   rec_thr = access_resolve(rec_thr, 0);
2970 
2971   Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
2972   Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
2973 
2974   generate_slow_guard(bol_thr, slow_region);
2975 
2976   // (b) Interrupt bit on TLS must be false.
2977   Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
2978   Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2979   p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
2980 
2981   // Set the control input on the field _interrupted read to prevent it floating up.
2982   Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
2983   Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
2984   Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
2985 
2986   IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2987 
2988   // First fast path:  if (!TLS._interrupted) return false;
2989   Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
2990   result_rgn->init_req(no_int_result_path, false_bit);
2991   result_val->init_req(no_int_result_path, intcon(0));
2992 
2993   // drop through to next case
2994   set_control( _gvn.transform(new IfTrueNode(iff_bit)));
2995 
2996 #ifndef _WINDOWS
2997   // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
2998   Node* clr_arg = argument(1);
2999   Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3000   Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3001   IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3002 
3003   // Second fast path:  ... else if (!clear_int) return true;
3004   Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3005   result_rgn->init_req(no_clear_result_path, false_arg);
3006   result_val->init_req(no_clear_result_path, intcon(1));
3007 
3008   // drop through to next case
3009   set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3010 #else
3011   // To return true on Windows you must read the _interrupted field
3012   // and check the event state i.e. take the slow path.
3013 #endif // _WINDOWS
3014 
3015   // (d) Otherwise, go to the slow path.
3016   slow_region->add_req(control());
3017   set_control( _gvn.transform(slow_region));
3018 
3019   if (stopped()) {
3020     // There is no slow path.
3021     result_rgn->init_req(slow_result_path, top());
3022     result_val->init_req(slow_result_path, top());
3023   } else {
3024     // non-virtual because it is a private non-static
3025     CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3026 
3027     Node* slow_val = set_results_for_java_call(slow_call);
3028     // this->control() comes from set_results_for_java_call
3029 
3030     Node* fast_io  = slow_call->in(TypeFunc::I_O);
3031     Node* fast_mem = slow_call->in(TypeFunc::Memory);
3032 
3033     // These two phis are pre-filled with copies of of the fast IO and Memory
3034     PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3035     PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
3036 
3037     result_rgn->init_req(slow_result_path, control());
3038     result_io ->init_req(slow_result_path, i_o());
3039     result_mem->init_req(slow_result_path, reset_memory());
3040     result_val->init_req(slow_result_path, slow_val);
3041 
3042     set_all_memory(_gvn.transform(result_mem));
3043     set_i_o(       _gvn.transform(result_io));
3044   }
3045 
3046   C->set_has_split_ifs(true); // Has chance for split-if optimization
3047   set_result(result_rgn, result_val);
3048   return true;
3049 }
3050 
3051 //---------------------------load_mirror_from_klass----------------------------
3052 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3053 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3054   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3055   Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3056   // mirror = ((OopHandle)mirror)->resolve();
3057   return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3058 }
3059 
3060 //-----------------------load_klass_from_mirror_common-------------------------
3061 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3062 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3063 // and branch to the given path on the region.
3064 // If never_see_null, take an uncommon trap on null, so we can optimistically
3065 // compile for the non-null case.
3066 // If the region is NULL, force never_see_null = true.
3067 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3068                                                     bool never_see_null,
3069                                                     RegionNode* region,
3070                                                     int null_path,
3071                                                     int offset) {
3072   if (region == NULL)  never_see_null = true;
3073   Node* p = basic_plus_adr(mirror, offset);
3074   const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3075   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3076   Node* null_ctl = top();
3077   kls = null_check_oop(kls, &null_ctl, never_see_null);
3078   if (region != NULL) {
3079     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3080     region->init_req(null_path, null_ctl);
3081   } else {
3082     assert(null_ctl == top(), "no loose ends");
3083   }
3084   return kls;
3085 }
3086 
3087 //--------------------(inline_native_Class_query helpers)---------------------
3088 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3089 // Fall through if (mods & mask) == bits, take the guard otherwise.
3090 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3091   // Branch around if the given klass has the given modifier bit set.
3092   // Like generate_guard, adds a new path onto the region.
3093   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3094   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3095   Node* mask = intcon(modifier_mask);
3096   Node* bits = intcon(modifier_bits);
3097   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3098   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3099   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3100   return generate_fair_guard(bol, region);
3101 }
3102 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3103   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3104 }
3105 
3106 //-------------------------inline_native_Class_query-------------------
3107 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3108   const Type* return_type = TypeInt::BOOL;
3109   Node* prim_return_value = top();  // what happens if it's a primitive class?
3110   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3111   bool expect_prim = false;     // most of these guys expect to work on refs
3112 
3113   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3114 
3115   Node* mirror = argument(0);
3116   Node* obj    = top();
3117 
3118   switch (id) {
3119   case vmIntrinsics::_isInstance:
3120     // nothing is an instance of a primitive type
3121     prim_return_value = intcon(0);
3122     obj = argument(1);
3123     break;
3124   case vmIntrinsics::_getModifiers:
3125     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3126     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3127     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3128     break;
3129   case vmIntrinsics::_isInterface:
3130     prim_return_value = intcon(0);
3131     break;
3132   case vmIntrinsics::_isArray:
3133     prim_return_value = intcon(0);
3134     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3135     break;
3136   case vmIntrinsics::_isPrimitive:
3137     prim_return_value = intcon(1);
3138     expect_prim = true;  // obviously
3139     break;
3140   case vmIntrinsics::_getSuperclass:
3141     prim_return_value = null();
3142     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3143     break;
3144   case vmIntrinsics::_getClassAccessFlags:
3145     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3146     return_type = TypeInt::INT;  // not bool!  6297094
3147     break;
3148   default:
3149     fatal_unexpected_iid(id);
3150     break;
3151   }
3152 
3153   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3154   if (mirror_con == NULL)  return false;  // cannot happen?
3155 
3156 #ifndef PRODUCT
3157   if (C->print_intrinsics() || C->print_inlining()) {
3158     ciType* k = mirror_con->java_mirror_type();
3159     if (k) {
3160       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3161       k->print_name();
3162       tty->cr();
3163     }
3164   }
3165 #endif
3166 
3167   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3168   RegionNode* region = new RegionNode(PATH_LIMIT);
3169   record_for_igvn(region);
3170   PhiNode* phi = new PhiNode(region, return_type);
3171 
3172   // The mirror will never be null of Reflection.getClassAccessFlags, however
3173   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3174   // if it is. See bug 4774291.
3175 
3176   // For Reflection.getClassAccessFlags(), the null check occurs in
3177   // the wrong place; see inline_unsafe_access(), above, for a similar
3178   // situation.
3179   mirror = null_check(mirror);
3180   // If mirror or obj is dead, only null-path is taken.
3181   if (stopped())  return true;
3182 
3183   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3184 
3185   // Now load the mirror's klass metaobject, and null-check it.
3186   // Side-effects region with the control path if the klass is null.
3187   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3188   // If kls is null, we have a primitive mirror.
3189   phi->init_req(_prim_path, prim_return_value);
3190   if (stopped()) { set_result(region, phi); return true; }
3191   bool safe_for_replace = (region->in(_prim_path) == top());
3192 
3193   Node* p;  // handy temp
3194   Node* null_ctl;
3195 
3196   // Now that we have the non-null klass, we can perform the real query.
3197   // For constant classes, the query will constant-fold in LoadNode::Value.
3198   Node* query_value = top();
3199   switch (id) {
3200   case vmIntrinsics::_isInstance:
3201     // nothing is an instance of a primitive type
3202     query_value = gen_instanceof(obj, kls, safe_for_replace);
3203     break;
3204 
3205   case vmIntrinsics::_getModifiers:
3206     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3207     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3208     break;
3209 
3210   case vmIntrinsics::_isInterface:
3211     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3212     if (generate_interface_guard(kls, region) != NULL)
3213       // A guard was added.  If the guard is taken, it was an interface.
3214       phi->add_req(intcon(1));
3215     // If we fall through, it's a plain class.
3216     query_value = intcon(0);
3217     break;
3218 
3219   case vmIntrinsics::_isArray:
3220     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3221     if (generate_array_guard(kls, region) != NULL)
3222       // A guard was added.  If the guard is taken, it was an array.
3223       phi->add_req(intcon(1));
3224     // If we fall through, it's a plain class.
3225     query_value = intcon(0);
3226     break;
3227 
3228   case vmIntrinsics::_isPrimitive:
3229     query_value = intcon(0); // "normal" path produces false
3230     break;
3231 
3232   case vmIntrinsics::_getSuperclass:
3233     // The rules here are somewhat unfortunate, but we can still do better
3234     // with random logic than with a JNI call.
3235     // Interfaces store null or Object as _super, but must report null.
3236     // Arrays store an intermediate super as _super, but must report Object.
3237     // Other types can report the actual _super.
3238     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3239     if (generate_interface_guard(kls, region) != NULL)
3240       // A guard was added.  If the guard is taken, it was an interface.
3241       phi->add_req(null());
3242     if (generate_array_guard(kls, region) != NULL)
3243       // A guard was added.  If the guard is taken, it was an array.
3244       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3245     // If we fall through, it's a plain class.  Get its _super.
3246     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3247     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3248     null_ctl = top();
3249     kls = null_check_oop(kls, &null_ctl);
3250     if (null_ctl != top()) {
3251       // If the guard is taken, Object.superClass is null (both klass and mirror).
3252       region->add_req(null_ctl);
3253       phi   ->add_req(null());
3254     }
3255     if (!stopped()) {
3256       query_value = load_mirror_from_klass(kls);
3257     }
3258     break;
3259 
3260   case vmIntrinsics::_getClassAccessFlags:
3261     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3262     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3263     break;
3264 
3265   default:
3266     fatal_unexpected_iid(id);
3267     break;
3268   }
3269 
3270   // Fall-through is the normal case of a query to a real class.
3271   phi->init_req(1, query_value);
3272   region->init_req(1, control());
3273 
3274   C->set_has_split_ifs(true); // Has chance for split-if optimization
3275   set_result(region, phi);
3276   return true;
3277 }
3278 
3279 //-------------------------inline_Class_cast-------------------
3280 bool LibraryCallKit::inline_Class_cast() {
3281   Node* mirror = argument(0); // Class
3282   Node* obj    = argument(1);
3283   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3284   if (mirror_con == NULL) {
3285     return false;  // dead path (mirror->is_top()).
3286   }
3287   if (obj == NULL || obj->is_top()) {
3288     return false;  // dead path
3289   }
3290   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3291 
3292   // First, see if Class.cast() can be folded statically.
3293   // java_mirror_type() returns non-null for compile-time Class constants.
3294   ciType* tm = mirror_con->java_mirror_type();
3295   if (tm != NULL && tm->is_klass() &&
3296       tp != NULL && tp->klass() != NULL) {
3297     if (!tp->klass()->is_loaded()) {
3298       // Don't use intrinsic when class is not loaded.
3299       return false;
3300     } else {
3301       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3302       if (static_res == Compile::SSC_always_true) {
3303         // isInstance() is true - fold the code.
3304         set_result(obj);
3305         return true;
3306       } else if (static_res == Compile::SSC_always_false) {
3307         // Don't use intrinsic, have to throw ClassCastException.
3308         // If the reference is null, the non-intrinsic bytecode will
3309         // be optimized appropriately.
3310         return false;
3311       }
3312     }
3313   }
3314 
3315   // Bailout intrinsic and do normal inlining if exception path is frequent.
3316   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3317     return false;
3318   }
3319 
3320   // Generate dynamic checks.
3321   // Class.cast() is java implementation of _checkcast bytecode.
3322   // Do checkcast (Parse::do_checkcast()) optimizations here.
3323 
3324   mirror = null_check(mirror);
3325   // If mirror is dead, only null-path is taken.
3326   if (stopped()) {
3327     return true;
3328   }
3329 
3330   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3331   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3332   RegionNode* region = new RegionNode(PATH_LIMIT);
3333   record_for_igvn(region);
3334 
3335   // Now load the mirror's klass metaobject, and null-check it.
3336   // If kls is null, we have a primitive mirror and
3337   // nothing is an instance of a primitive type.
3338   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3339 
3340   Node* res = top();
3341   if (!stopped()) {
3342     Node* bad_type_ctrl = top();
3343     // Do checkcast optimizations.
3344     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3345     region->init_req(_bad_type_path, bad_type_ctrl);
3346   }
3347   if (region->in(_prim_path) != top() ||
3348       region->in(_bad_type_path) != top()) {
3349     // Let Interpreter throw ClassCastException.
3350     PreserveJVMState pjvms(this);
3351     set_control(_gvn.transform(region));
3352     uncommon_trap(Deoptimization::Reason_intrinsic,
3353                   Deoptimization::Action_maybe_recompile);
3354   }
3355   if (!stopped()) {
3356     set_result(res);
3357   }
3358   return true;
3359 }
3360 
3361 
3362 //--------------------------inline_native_subtype_check------------------------
3363 // This intrinsic takes the JNI calls out of the heart of
3364 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3365 bool LibraryCallKit::inline_native_subtype_check() {
3366   // Pull both arguments off the stack.
3367   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3368   args[0] = argument(0);
3369   args[1] = argument(1);
3370   Node* klasses[2];             // corresponding Klasses: superk, subk
3371   klasses[0] = klasses[1] = top();
3372 
3373   enum {
3374     // A full decision tree on {superc is prim, subc is prim}:
3375     _prim_0_path = 1,           // {P,N} => false
3376                                 // {P,P} & superc!=subc => false
3377     _prim_same_path,            // {P,P} & superc==subc => true
3378     _prim_1_path,               // {N,P} => false
3379     _ref_subtype_path,          // {N,N} & subtype check wins => true
3380     _both_ref_path,             // {N,N} & subtype check loses => false
3381     PATH_LIMIT
3382   };
3383 
3384   RegionNode* region = new RegionNode(PATH_LIMIT);
3385   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3386   record_for_igvn(region);
3387 
3388   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3389   const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3390   int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3391 
3392   // First null-check both mirrors and load each mirror's klass metaobject.
3393   int which_arg;
3394   for (which_arg = 0; which_arg <= 1; which_arg++) {
3395     Node* arg = args[which_arg];
3396     arg = null_check(arg);
3397     if (stopped())  break;
3398     args[which_arg] = arg;
3399 
3400     Node* p = basic_plus_adr(arg, class_klass_offset);
3401     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3402     klasses[which_arg] = _gvn.transform(kls);
3403   }
3404 
3405   // Resolve oops to stable for CmpP below.
3406   args[0] = access_resolve(args[0], 0);
3407   args[1] = access_resolve(args[1], 0);
3408 
3409   // Having loaded both klasses, test each for null.
3410   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3411   for (which_arg = 0; which_arg <= 1; which_arg++) {
3412     Node* kls = klasses[which_arg];
3413     Node* null_ctl = top();
3414     kls = null_check_oop(kls, &null_ctl, never_see_null);
3415     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3416     region->init_req(prim_path, null_ctl);
3417     if (stopped())  break;
3418     klasses[which_arg] = kls;
3419   }
3420 
3421   if (!stopped()) {
3422     // now we have two reference types, in klasses[0..1]
3423     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3424     Node* superk = klasses[0];  // the receiver
3425     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3426     // now we have a successful reference subtype check
3427     region->set_req(_ref_subtype_path, control());
3428   }
3429 
3430   // If both operands are primitive (both klasses null), then
3431   // we must return true when they are identical primitives.
3432   // It is convenient to test this after the first null klass check.
3433   set_control(region->in(_prim_0_path)); // go back to first null check
3434   if (!stopped()) {
3435     // Since superc is primitive, make a guard for the superc==subc case.
3436     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3437     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3438     generate_guard(bol_eq, region, PROB_FAIR);
3439     if (region->req() == PATH_LIMIT+1) {
3440       // A guard was added.  If the added guard is taken, superc==subc.
3441       region->swap_edges(PATH_LIMIT, _prim_same_path);
3442       region->del_req(PATH_LIMIT);
3443     }
3444     region->set_req(_prim_0_path, control()); // Not equal after all.
3445   }
3446 
3447   // these are the only paths that produce 'true':
3448   phi->set_req(_prim_same_path,   intcon(1));
3449   phi->set_req(_ref_subtype_path, intcon(1));
3450 
3451   // pull together the cases:
3452   assert(region->req() == PATH_LIMIT, "sane region");
3453   for (uint i = 1; i < region->req(); i++) {
3454     Node* ctl = region->in(i);
3455     if (ctl == NULL || ctl == top()) {
3456       region->set_req(i, top());
3457       phi   ->set_req(i, top());
3458     } else if (phi->in(i) == NULL) {
3459       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3460     }
3461   }
3462 
3463   set_control(_gvn.transform(region));
3464   set_result(_gvn.transform(phi));
3465   return true;
3466 }
3467 
3468 //---------------------generate_array_guard_common------------------------
3469 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3470                                                   bool obj_array, bool not_array) {
3471 
3472   if (stopped()) {
3473     return NULL;
3474   }
3475 
3476   // If obj_array/non_array==false/false:
3477   // Branch around if the given klass is in fact an array (either obj or prim).
3478   // If obj_array/non_array==false/true:
3479   // Branch around if the given klass is not an array klass of any kind.
3480   // If obj_array/non_array==true/true:
3481   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3482   // If obj_array/non_array==true/false:
3483   // Branch around if the kls is an oop array (Object[] or subtype)
3484   //
3485   // Like generate_guard, adds a new path onto the region.
3486   jint  layout_con = 0;
3487   Node* layout_val = get_layout_helper(kls, layout_con);
3488   if (layout_val == NULL) {
3489     bool query = (obj_array
3490                   ? Klass::layout_helper_is_objArray(layout_con)
3491                   : Klass::layout_helper_is_array(layout_con));
3492     if (query == not_array) {
3493       return NULL;                       // never a branch
3494     } else {                             // always a branch
3495       Node* always_branch = control();
3496       if (region != NULL)
3497         region->add_req(always_branch);
3498       set_control(top());
3499       return always_branch;
3500     }
3501   }
3502   // Now test the correct condition.
3503   jint  nval = (obj_array
3504                 ? (jint)(Klass::_lh_array_tag_type_value
3505                    <<    Klass::_lh_array_tag_shift)
3506                 : Klass::_lh_neutral_value);
3507   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3508   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3509   // invert the test if we are looking for a non-array
3510   if (not_array)  btest = BoolTest(btest).negate();
3511   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3512   return generate_fair_guard(bol, region);
3513 }
3514 
3515 
3516 //-----------------------inline_native_newArray--------------------------
3517 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3518 // private        native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3519 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3520   Node* mirror;
3521   Node* count_val;
3522   if (uninitialized) {
3523     mirror    = argument(1);
3524     count_val = argument(2);
3525   } else {
3526     mirror    = argument(0);
3527     count_val = argument(1);
3528   }
3529 
3530   mirror = null_check(mirror);
3531   // If mirror or obj is dead, only null-path is taken.
3532   if (stopped())  return true;
3533 
3534   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3535   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3536   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3537   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3538   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3539 
3540   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3541   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3542                                                   result_reg, _slow_path);
3543   Node* normal_ctl   = control();
3544   Node* no_array_ctl = result_reg->in(_slow_path);
3545 
3546   // Generate code for the slow case.  We make a call to newArray().
3547   set_control(no_array_ctl);
3548   if (!stopped()) {
3549     // Either the input type is void.class, or else the
3550     // array klass has not yet been cached.  Either the
3551     // ensuing call will throw an exception, or else it
3552     // will cache the array klass for next time.
3553     PreserveJVMState pjvms(this);
3554     CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3555     Node* slow_result = set_results_for_java_call(slow_call);
3556     // this->control() comes from set_results_for_java_call
3557     result_reg->set_req(_slow_path, control());
3558     result_val->set_req(_slow_path, slow_result);
3559     result_io ->set_req(_slow_path, i_o());
3560     result_mem->set_req(_slow_path, reset_memory());
3561   }
3562 
3563   set_control(normal_ctl);
3564   if (!stopped()) {
3565     // Normal case:  The array type has been cached in the java.lang.Class.
3566     // The following call works fine even if the array type is polymorphic.
3567     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3568     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3569     result_reg->init_req(_normal_path, control());
3570     result_val->init_req(_normal_path, obj);
3571     result_io ->init_req(_normal_path, i_o());
3572     result_mem->init_req(_normal_path, reset_memory());
3573 
3574     if (uninitialized) {
3575       // Mark the allocation so that zeroing is skipped
3576       AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3577       alloc->maybe_set_complete(&_gvn);
3578     }
3579   }
3580 
3581   // Return the combined state.
3582   set_i_o(        _gvn.transform(result_io)  );
3583   set_all_memory( _gvn.transform(result_mem));
3584 
3585   C->set_has_split_ifs(true); // Has chance for split-if optimization
3586   set_result(result_reg, result_val);
3587   return true;
3588 }
3589 
3590 //----------------------inline_native_getLength--------------------------
3591 // public static native int java.lang.reflect.Array.getLength(Object array);
3592 bool LibraryCallKit::inline_native_getLength() {
3593   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3594 
3595   Node* array = null_check(argument(0));
3596   // If array is dead, only null-path is taken.
3597   if (stopped())  return true;
3598 
3599   // Deoptimize if it is a non-array.
3600   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3601 
3602   if (non_array != NULL) {
3603     PreserveJVMState pjvms(this);
3604     set_control(non_array);
3605     uncommon_trap(Deoptimization::Reason_intrinsic,
3606                   Deoptimization::Action_maybe_recompile);
3607   }
3608 
3609   // If control is dead, only non-array-path is taken.
3610   if (stopped())  return true;
3611 
3612   // The works fine even if the array type is polymorphic.
3613   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3614   Node* result = load_array_length(array);
3615 
3616   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3617   set_result(result);
3618   return true;
3619 }
3620 
3621 //------------------------inline_array_copyOf----------------------------
3622 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3623 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3624 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3625   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3626 
3627   // Get the arguments.
3628   Node* original          = argument(0);
3629   Node* start             = is_copyOfRange? argument(1): intcon(0);
3630   Node* end               = is_copyOfRange? argument(2): argument(1);
3631   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3632 
3633   Node* newcopy = NULL;
3634 
3635   // Set the original stack and the reexecute bit for the interpreter to reexecute
3636   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3637   { PreserveReexecuteState preexecs(this);
3638     jvms()->set_should_reexecute(true);
3639 
3640     array_type_mirror = null_check(array_type_mirror);
3641     original          = null_check(original);
3642 
3643     // Check if a null path was taken unconditionally.
3644     if (stopped())  return true;
3645 
3646     Node* orig_length = load_array_length(original);
3647 
3648     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3649     klass_node = null_check(klass_node);
3650 
3651     RegionNode* bailout = new RegionNode(1);
3652     record_for_igvn(bailout);
3653 
3654     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3655     // Bail out if that is so.
3656     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3657     if (not_objArray != NULL) {
3658       // Improve the klass node's type from the new optimistic assumption:
3659       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3660       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3661       Node* cast = new CastPPNode(klass_node, akls);
3662       cast->init_req(0, control());
3663       klass_node = _gvn.transform(cast);
3664     }
3665 
3666     // Bail out if either start or end is negative.
3667     generate_negative_guard(start, bailout, &start);
3668     generate_negative_guard(end,   bailout, &end);
3669 
3670     Node* length = end;
3671     if (_gvn.type(start) != TypeInt::ZERO) {
3672       length = _gvn.transform(new SubINode(end, start));
3673     }
3674 
3675     // Bail out if length is negative.
3676     // Without this the new_array would throw
3677     // NegativeArraySizeException but IllegalArgumentException is what
3678     // should be thrown
3679     generate_negative_guard(length, bailout, &length);
3680 
3681     if (bailout->req() > 1) {
3682       PreserveJVMState pjvms(this);
3683       set_control(_gvn.transform(bailout));
3684       uncommon_trap(Deoptimization::Reason_intrinsic,
3685                     Deoptimization::Action_maybe_recompile);
3686     }
3687 
3688     if (!stopped()) {
3689       // How many elements will we copy from the original?
3690       // The answer is MinI(orig_length - start, length).
3691       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3692       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3693 
3694       original = access_resolve(original, ACCESS_READ);
3695 
3696       // Generate a direct call to the right arraycopy function(s).
3697       // We know the copy is disjoint but we might not know if the
3698       // oop stores need checking.
3699       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3700       // This will fail a store-check if x contains any non-nulls.
3701 
3702       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3703       // loads/stores but it is legal only if we're sure the
3704       // Arrays.copyOf would succeed. So we need all input arguments
3705       // to the copyOf to be validated, including that the copy to the
3706       // new array won't trigger an ArrayStoreException. That subtype
3707       // check can be optimized if we know something on the type of
3708       // the input array from type speculation.
3709       if (_gvn.type(klass_node)->singleton()) {
3710         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3711         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3712 
3713         int test = C->static_subtype_check(superk, subk);
3714         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3715           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3716           if (t_original->speculative_type() != NULL) {
3717             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3718           }
3719         }
3720       }
3721 
3722       bool validated = false;
3723       // Reason_class_check rather than Reason_intrinsic because we
3724       // want to intrinsify even if this traps.
3725       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3726         Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3727                                                    klass_node);
3728 
3729         if (not_subtype_ctrl != top()) {
3730           PreserveJVMState pjvms(this);
3731           set_control(not_subtype_ctrl);
3732           uncommon_trap(Deoptimization::Reason_class_check,
3733                         Deoptimization::Action_make_not_entrant);
3734           assert(stopped(), "Should be stopped");
3735         }
3736         validated = true;
3737       }
3738 
3739       if (!stopped()) {
3740         newcopy = new_array(klass_node, length, 0);  // no arguments to push
3741 
3742         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3743                                                 load_object_klass(original), klass_node);
3744         if (!is_copyOfRange) {
3745           ac->set_copyof(validated);
3746         } else {
3747           ac->set_copyofrange(validated);
3748         }
3749         Node* n = _gvn.transform(ac);
3750         if (n == ac) {
3751           ac->connect_outputs(this);
3752         } else {
3753           assert(validated, "shouldn't transform if all arguments not validated");
3754           set_all_memory(n);
3755         }
3756       }
3757     }
3758   } // original reexecute is set back here
3759 
3760   C->set_has_split_ifs(true); // Has chance for split-if optimization
3761   if (!stopped()) {
3762     set_result(newcopy);
3763   }
3764   return true;
3765 }
3766 
3767 
3768 //----------------------generate_virtual_guard---------------------------
3769 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
3770 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3771                                              RegionNode* slow_region) {
3772   ciMethod* method = callee();
3773   int vtable_index = method->vtable_index();
3774   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3775          "bad index %d", vtable_index);
3776   // Get the Method* out of the appropriate vtable entry.
3777   int entry_offset  = in_bytes(Klass::vtable_start_offset()) +
3778                      vtable_index*vtableEntry::size_in_bytes() +
3779                      vtableEntry::method_offset_in_bytes();
3780   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
3781   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3782 
3783   // Compare the target method with the expected method (e.g., Object.hashCode).
3784   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3785 
3786   Node* native_call = makecon(native_call_addr);
3787   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
3788   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3789 
3790   return generate_slow_guard(test_native, slow_region);
3791 }
3792 
3793 //-----------------------generate_method_call----------------------------
3794 // Use generate_method_call to make a slow-call to the real
3795 // method if the fast path fails.  An alternative would be to
3796 // use a stub like OptoRuntime::slow_arraycopy_Java.
3797 // This only works for expanding the current library call,
3798 // not another intrinsic.  (E.g., don't use this for making an
3799 // arraycopy call inside of the copyOf intrinsic.)
3800 CallJavaNode*
3801 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3802   // When compiling the intrinsic method itself, do not use this technique.
3803   guarantee(callee() != C->method(), "cannot make slow-call to self");
3804 
3805   ciMethod* method = callee();
3806   // ensure the JVMS we have will be correct for this call
3807   guarantee(method_id == method->intrinsic_id(), "must match");
3808 
3809   const TypeFunc* tf = TypeFunc::make(method);
3810   CallJavaNode* slow_call;
3811   if (is_static) {
3812     assert(!is_virtual, "");
3813     slow_call = new CallStaticJavaNode(C, tf,
3814                            SharedRuntime::get_resolve_static_call_stub(),
3815                            method, bci());
3816   } else if (is_virtual) {
3817     null_check_receiver();
3818     int vtable_index = Method::invalid_vtable_index;
3819     if (UseInlineCaches) {
3820       // Suppress the vtable call
3821     } else {
3822       // hashCode and clone are not a miranda methods,
3823       // so the vtable index is fixed.
3824       // No need to use the linkResolver to get it.
3825        vtable_index = method->vtable_index();
3826        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3827               "bad index %d", vtable_index);
3828     }
3829     slow_call = new CallDynamicJavaNode(tf,
3830                           SharedRuntime::get_resolve_virtual_call_stub(),
3831                           method, vtable_index, bci());
3832   } else {  // neither virtual nor static:  opt_virtual
3833     null_check_receiver();
3834     slow_call = new CallStaticJavaNode(C, tf,
3835                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3836                                 method, bci());
3837     slow_call->set_optimized_virtual(true);
3838   }
3839   set_arguments_for_java_call(slow_call);
3840   set_edges_for_java_call(slow_call);
3841   return slow_call;
3842 }
3843 
3844 
3845 /**
3846  * Build special case code for calls to hashCode on an object. This call may
3847  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3848  * slightly different code.
3849  */
3850 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3851   assert(is_static == callee()->is_static(), "correct intrinsic selection");
3852   assert(!(is_virtual && is_static), "either virtual, special, or static");
3853 
3854   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3855 
3856   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3857   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
3858   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3859   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3860   Node* obj = NULL;
3861   if (!is_static) {
3862     // Check for hashing null object
3863     obj = null_check_receiver();
3864     if (stopped())  return true;        // unconditionally null
3865     result_reg->init_req(_null_path, top());
3866     result_val->init_req(_null_path, top());
3867   } else {
3868     // Do a null check, and return zero if null.
3869     // System.identityHashCode(null) == 0
3870     obj = argument(0);
3871     Node* null_ctl = top();
3872     obj = null_check_oop(obj, &null_ctl);
3873     result_reg->init_req(_null_path, null_ctl);
3874     result_val->init_req(_null_path, _gvn.intcon(0));
3875   }
3876 
3877   // Unconditionally null?  Then return right away.
3878   if (stopped()) {
3879     set_control( result_reg->in(_null_path));
3880     if (!stopped())
3881       set_result(result_val->in(_null_path));
3882     return true;
3883   }
3884 
3885   // We only go to the fast case code if we pass a number of guards.  The
3886   // paths which do not pass are accumulated in the slow_region.
3887   RegionNode* slow_region = new RegionNode(1);
3888   record_for_igvn(slow_region);
3889 
3890   // If this is a virtual call, we generate a funny guard.  We pull out
3891   // the vtable entry corresponding to hashCode() from the target object.
3892   // If the target method which we are calling happens to be the native
3893   // Object hashCode() method, we pass the guard.  We do not need this
3894   // guard for non-virtual calls -- the caller is known to be the native
3895   // Object hashCode().
3896   if (is_virtual) {
3897     // After null check, get the object's klass.
3898     Node* obj_klass = load_object_klass(obj);
3899     generate_virtual_guard(obj_klass, slow_region);
3900   }
3901 
3902   // Get the header out of the object, use LoadMarkNode when available
3903   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3904   // The control of the load must be NULL. Otherwise, the load can move before
3905   // the null check after castPP removal.
3906   Node* no_ctrl = NULL;
3907   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3908 
3909   // Test the header to see if it is unlocked.
3910   Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3911   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3912   Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
3913   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3914   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3915 
3916   generate_slow_guard(test_unlocked, slow_region);
3917 
3918   // Get the hash value and check to see that it has been properly assigned.
3919   // We depend on hash_mask being at most 32 bits and avoid the use of
3920   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3921   // vm: see markOop.hpp.
3922   Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
3923   Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
3924   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3925   // This hack lets the hash bits live anywhere in the mark object now, as long
3926   // as the shift drops the relevant bits into the low 32 bits.  Note that
3927   // Java spec says that HashCode is an int so there's no point in capturing
3928   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3929   hshifted_header      = ConvX2I(hshifted_header);
3930   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3931 
3932   Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
3933   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3934   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3935 
3936   generate_slow_guard(test_assigned, slow_region);
3937 
3938   Node* init_mem = reset_memory();
3939   // fill in the rest of the null path:
3940   result_io ->init_req(_null_path, i_o());
3941   result_mem->init_req(_null_path, init_mem);
3942 
3943   result_val->init_req(_fast_path, hash_val);
3944   result_reg->init_req(_fast_path, control());
3945   result_io ->init_req(_fast_path, i_o());
3946   result_mem->init_req(_fast_path, init_mem);
3947 
3948   // Generate code for the slow case.  We make a call to hashCode().
3949   set_control(_gvn.transform(slow_region));
3950   if (!stopped()) {
3951     // No need for PreserveJVMState, because we're using up the present state.
3952     set_all_memory(init_mem);
3953     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3954     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3955     Node* slow_result = set_results_for_java_call(slow_call);
3956     // this->control() comes from set_results_for_java_call
3957     result_reg->init_req(_slow_path, control());
3958     result_val->init_req(_slow_path, slow_result);
3959     result_io  ->set_req(_slow_path, i_o());
3960     result_mem ->set_req(_slow_path, reset_memory());
3961   }
3962 
3963   // Return the combined state.
3964   set_i_o(        _gvn.transform(result_io)  );
3965   set_all_memory( _gvn.transform(result_mem));
3966 
3967   set_result(result_reg, result_val);
3968   return true;
3969 }
3970 
3971 //---------------------------inline_native_getClass----------------------------
3972 // public final native Class<?> java.lang.Object.getClass();
3973 //
3974 // Build special case code for calls to getClass on an object.
3975 bool LibraryCallKit::inline_native_getClass() {
3976   Node* obj = null_check_receiver();
3977   if (stopped())  return true;
3978   set_result(load_mirror_from_klass(load_object_klass(obj)));
3979   return true;
3980 }
3981 
3982 //-----------------inline_native_Reflection_getCallerClass---------------------
3983 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3984 //
3985 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3986 //
3987 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3988 // in that it must skip particular security frames and checks for
3989 // caller sensitive methods.
3990 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3991 #ifndef PRODUCT
3992   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3993     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3994   }
3995 #endif
3996 
3997   if (!jvms()->has_method()) {
3998 #ifndef PRODUCT
3999     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4000       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
4001     }
4002 #endif
4003     return false;
4004   }
4005 
4006   // Walk back up the JVM state to find the caller at the required
4007   // depth.
4008   JVMState* caller_jvms = jvms();
4009 
4010   // Cf. JVM_GetCallerClass
4011   // NOTE: Start the loop at depth 1 because the current JVM state does
4012   // not include the Reflection.getCallerClass() frame.
4013   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4014     ciMethod* m = caller_jvms->method();
4015     switch (n) {
4016     case 0:
4017       fatal("current JVM state does not include the Reflection.getCallerClass frame");
4018       break;
4019     case 1:
4020       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4021       if (!m->caller_sensitive()) {
4022 #ifndef PRODUCT
4023         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4024           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
4025         }
4026 #endif
4027         return false;  // bail-out; let JVM_GetCallerClass do the work
4028       }
4029       break;
4030     default:
4031       if (!m->is_ignored_by_security_stack_walk()) {
4032         // We have reached the desired frame; return the holder class.
4033         // Acquire method holder as java.lang.Class and push as constant.
4034         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4035         ciInstance* caller_mirror = caller_klass->java_mirror();
4036         set_result(makecon(TypeInstPtr::make(caller_mirror)));
4037 
4038 #ifndef PRODUCT
4039         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4040           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());
4041           tty->print_cr("  JVM state at this point:");
4042           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4043             ciMethod* m = jvms()->of_depth(i)->method();
4044             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4045           }
4046         }
4047 #endif
4048         return true;
4049       }
4050       break;
4051     }
4052   }
4053 
4054 #ifndef PRODUCT
4055   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4056     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4057     tty->print_cr("  JVM state at this point:");
4058     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4059       ciMethod* m = jvms()->of_depth(i)->method();
4060       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4061     }
4062   }
4063 #endif
4064 
4065   return false;  // bail-out; let JVM_GetCallerClass do the work
4066 }
4067 
4068 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4069   Node* arg = argument(0);
4070   Node* result = NULL;
4071 
4072   switch (id) {
4073   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4074   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4075   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4076   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4077 
4078   case vmIntrinsics::_doubleToLongBits: {
4079     // two paths (plus control) merge in a wood
4080     RegionNode *r = new RegionNode(3);
4081     Node *phi = new PhiNode(r, TypeLong::LONG);
4082 
4083     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4084     // Build the boolean node
4085     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4086 
4087     // Branch either way.
4088     // NaN case is less traveled, which makes all the difference.
4089     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4090     Node *opt_isnan = _gvn.transform(ifisnan);
4091     assert( opt_isnan->is_If(), "Expect an IfNode");
4092     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4093     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4094 
4095     set_control(iftrue);
4096 
4097     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4098     Node *slow_result = longcon(nan_bits); // return NaN
4099     phi->init_req(1, _gvn.transform( slow_result ));
4100     r->init_req(1, iftrue);
4101 
4102     // Else fall through
4103     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4104     set_control(iffalse);
4105 
4106     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4107     r->init_req(2, iffalse);
4108 
4109     // Post merge
4110     set_control(_gvn.transform(r));
4111     record_for_igvn(r);
4112 
4113     C->set_has_split_ifs(true); // Has chance for split-if optimization
4114     result = phi;
4115     assert(result->bottom_type()->isa_long(), "must be");
4116     break;
4117   }
4118 
4119   case vmIntrinsics::_floatToIntBits: {
4120     // two paths (plus control) merge in a wood
4121     RegionNode *r = new RegionNode(3);
4122     Node *phi = new PhiNode(r, TypeInt::INT);
4123 
4124     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4125     // Build the boolean node
4126     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4127 
4128     // Branch either way.
4129     // NaN case is less traveled, which makes all the difference.
4130     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4131     Node *opt_isnan = _gvn.transform(ifisnan);
4132     assert( opt_isnan->is_If(), "Expect an IfNode");
4133     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4134     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4135 
4136     set_control(iftrue);
4137 
4138     static const jint nan_bits = 0x7fc00000;
4139     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4140     phi->init_req(1, _gvn.transform( slow_result ));
4141     r->init_req(1, iftrue);
4142 
4143     // Else fall through
4144     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4145     set_control(iffalse);
4146 
4147     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4148     r->init_req(2, iffalse);
4149 
4150     // Post merge
4151     set_control(_gvn.transform(r));
4152     record_for_igvn(r);
4153 
4154     C->set_has_split_ifs(true); // Has chance for split-if optimization
4155     result = phi;
4156     assert(result->bottom_type()->isa_int(), "must be");
4157     break;
4158   }
4159 
4160   default:
4161     fatal_unexpected_iid(id);
4162     break;
4163   }
4164   set_result(_gvn.transform(result));
4165   return true;
4166 }
4167 
4168 //----------------------inline_unsafe_copyMemory-------------------------
4169 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4170 bool LibraryCallKit::inline_unsafe_copyMemory() {
4171   if (callee()->is_static())  return false;  // caller must have the capability!
4172   null_check_receiver();  // null-check receiver
4173   if (stopped())  return true;
4174 
4175   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4176 
4177   Node* src_ptr =         argument(1);   // type: oop
4178   Node* src_off = ConvL2X(argument(2));  // type: long
4179   Node* dst_ptr =         argument(4);   // type: oop
4180   Node* dst_off = ConvL2X(argument(5));  // type: long
4181   Node* size    = ConvL2X(argument(7));  // type: long
4182 
4183   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4184          "fieldOffset must be byte-scaled");
4185 
4186   src_ptr = access_resolve(src_ptr, ACCESS_READ);
4187   dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4188   Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4189   Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4190 
4191   // Conservatively insert a memory barrier on all memory slices.
4192   // Do not let writes of the copy source or destination float below the copy.
4193   insert_mem_bar(Op_MemBarCPUOrder);
4194 
4195   // Call it.  Note that the length argument is not scaled.
4196   make_runtime_call(RC_LEAF|RC_NO_FP,
4197                     OptoRuntime::fast_arraycopy_Type(),
4198                     StubRoutines::unsafe_arraycopy(),
4199                     "unsafe_arraycopy",
4200                     TypeRawPtr::BOTTOM,
4201                     src, dst, size XTOP);
4202 
4203   // Do not let reads of the copy destination float above the copy.
4204   insert_mem_bar(Op_MemBarCPUOrder);
4205 
4206   return true;
4207 }
4208 
4209 //------------------------clone_coping-----------------------------------
4210 // Helper function for inline_native_clone.
4211 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4212   assert(obj_size != NULL, "");
4213   Node* raw_obj = alloc_obj->in(1);
4214   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4215 
4216   AllocateNode* alloc = NULL;
4217   if (ReduceBulkZeroing) {
4218     // We will be completely responsible for initializing this object -
4219     // mark Initialize node as complete.
4220     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4221     // The object was just allocated - there should be no any stores!
4222     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4223     // Mark as complete_with_arraycopy so that on AllocateNode
4224     // expansion, we know this AllocateNode is initialized by an array
4225     // copy and a StoreStore barrier exists after the array copy.
4226     alloc->initialization()->set_complete_with_arraycopy();
4227   }
4228 
4229   // Copy the fastest available way.
4230   // TODO: generate fields copies for small objects instead.
4231   Node* size = _gvn.transform(obj_size);
4232 
4233   access_clone(obj, alloc_obj, size, is_array);
4234 
4235   // Do not let reads from the cloned object float above the arraycopy.
4236   if (alloc != NULL) {
4237     // Do not let stores that initialize this object be reordered with
4238     // a subsequent store that would make this object accessible by
4239     // other threads.
4240     // Record what AllocateNode this StoreStore protects so that
4241     // escape analysis can go from the MemBarStoreStoreNode to the
4242     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4243     // based on the escape status of the AllocateNode.
4244     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4245   } else {
4246     insert_mem_bar(Op_MemBarCPUOrder);
4247   }
4248 }
4249 
4250 //------------------------inline_native_clone----------------------------
4251 // protected native Object java.lang.Object.clone();
4252 //
4253 // Here are the simple edge cases:
4254 //  null receiver => normal trap
4255 //  virtual and clone was overridden => slow path to out-of-line clone
4256 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4257 //
4258 // The general case has two steps, allocation and copying.
4259 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4260 //
4261 // Copying also has two cases, oop arrays and everything else.
4262 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4263 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4264 //
4265 // These steps fold up nicely if and when the cloned object's klass
4266 // can be sharply typed as an object array, a type array, or an instance.
4267 //
4268 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4269   PhiNode* result_val;
4270 
4271   // Set the reexecute bit for the interpreter to reexecute
4272   // the bytecode that invokes Object.clone if deoptimization happens.
4273   { PreserveReexecuteState preexecs(this);
4274     jvms()->set_should_reexecute(true);
4275 
4276     Node* obj = null_check_receiver();
4277     if (stopped())  return true;
4278 
4279     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4280 
4281     // If we are going to clone an instance, we need its exact type to
4282     // know the number and types of fields to convert the clone to
4283     // loads/stores. Maybe a speculative type can help us.
4284     if (!obj_type->klass_is_exact() &&
4285         obj_type->speculative_type() != NULL &&
4286         obj_type->speculative_type()->is_instance_klass()) {
4287       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4288       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4289           !spec_ik->has_injected_fields()) {
4290         ciKlass* k = obj_type->klass();
4291         if (!k->is_instance_klass() ||
4292             k->as_instance_klass()->is_interface() ||
4293             k->as_instance_klass()->has_subklass()) {
4294           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4295         }
4296       }
4297     }
4298 
4299     Node* obj_klass = load_object_klass(obj);
4300     const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4301     const TypeOopPtr*   toop   = ((tklass != NULL)
4302                                 ? tklass->as_instance_type()
4303                                 : TypeInstPtr::NOTNULL);
4304 
4305     // Conservatively insert a memory barrier on all memory slices.
4306     // Do not let writes into the original float below the clone.
4307     insert_mem_bar(Op_MemBarCPUOrder);
4308 
4309     // paths into result_reg:
4310     enum {
4311       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4312       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4313       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4314       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4315       PATH_LIMIT
4316     };
4317     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4318     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4319     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4320     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4321     record_for_igvn(result_reg);
4322 
4323     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4324     if (array_ctl != NULL) {
4325       // It's an array.
4326       PreserveJVMState pjvms(this);
4327       set_control(array_ctl);
4328       Node* obj_length = load_array_length(obj);
4329       Node* obj_size  = NULL;
4330       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4331 
4332       BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4333       if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4334         // If it is an oop array, it requires very special treatment,
4335         // because gc barriers are required when accessing the array.
4336         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4337         if (is_obja != NULL) {
4338           PreserveJVMState pjvms2(this);
4339           set_control(is_obja);
4340           obj = access_resolve(obj, ACCESS_READ);
4341           // Generate a direct call to the right arraycopy function(s).
4342           Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4343           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4344           ac->set_cloneoop();
4345           Node* n = _gvn.transform(ac);
4346           assert(n == ac, "cannot disappear");
4347           ac->connect_outputs(this);
4348 
4349           result_reg->init_req(_objArray_path, control());
4350           result_val->init_req(_objArray_path, alloc_obj);
4351           result_i_o ->set_req(_objArray_path, i_o());
4352           result_mem ->set_req(_objArray_path, reset_memory());
4353         }
4354       }
4355       // Otherwise, there are no barriers to worry about.
4356       // (We can dispense with card marks if we know the allocation
4357       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4358       //  causes the non-eden paths to take compensating steps to
4359       //  simulate a fresh allocation, so that no further
4360       //  card marks are required in compiled code to initialize
4361       //  the object.)
4362 
4363       if (!stopped()) {
4364         copy_to_clone(obj, alloc_obj, obj_size, true);
4365 
4366         // Present the results of the copy.
4367         result_reg->init_req(_array_path, control());
4368         result_val->init_req(_array_path, alloc_obj);
4369         result_i_o ->set_req(_array_path, i_o());
4370         result_mem ->set_req(_array_path, reset_memory());
4371       }
4372     }
4373 
4374     // We only go to the instance fast case code if we pass a number of guards.
4375     // The paths which do not pass are accumulated in the slow_region.
4376     RegionNode* slow_region = new RegionNode(1);
4377     record_for_igvn(slow_region);
4378     if (!stopped()) {
4379       // It's an instance (we did array above).  Make the slow-path tests.
4380       // If this is a virtual call, we generate a funny guard.  We grab
4381       // the vtable entry corresponding to clone() from the target object.
4382       // If the target method which we are calling happens to be the
4383       // Object clone() method, we pass the guard.  We do not need this
4384       // guard for non-virtual calls; the caller is known to be the native
4385       // Object clone().
4386       if (is_virtual) {
4387         generate_virtual_guard(obj_klass, slow_region);
4388       }
4389 
4390       // The object must be easily cloneable and must not have a finalizer.
4391       // Both of these conditions may be checked in a single test.
4392       // We could optimize the test further, but we don't care.
4393       generate_access_flags_guard(obj_klass,
4394                                   // Test both conditions:
4395                                   JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4396                                   // Must be cloneable but not finalizer:
4397                                   JVM_ACC_IS_CLONEABLE_FAST,
4398                                   slow_region);
4399     }
4400 
4401     if (!stopped()) {
4402       // It's an instance, and it passed the slow-path tests.
4403       PreserveJVMState pjvms(this);
4404       Node* obj_size  = NULL;
4405       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4406       // is reexecuted if deoptimization occurs and there could be problems when merging
4407       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4408       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4409 
4410       copy_to_clone(obj, alloc_obj, obj_size, false);
4411 
4412       // Present the results of the slow call.
4413       result_reg->init_req(_instance_path, control());
4414       result_val->init_req(_instance_path, alloc_obj);
4415       result_i_o ->set_req(_instance_path, i_o());
4416       result_mem ->set_req(_instance_path, reset_memory());
4417     }
4418 
4419     // Generate code for the slow case.  We make a call to clone().
4420     set_control(_gvn.transform(slow_region));
4421     if (!stopped()) {
4422       PreserveJVMState pjvms(this);
4423       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4424       // We need to deoptimize on exception (see comment above)
4425       Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4426       // this->control() comes from set_results_for_java_call
4427       result_reg->init_req(_slow_path, control());
4428       result_val->init_req(_slow_path, slow_result);
4429       result_i_o ->set_req(_slow_path, i_o());
4430       result_mem ->set_req(_slow_path, reset_memory());
4431     }
4432 
4433     // Return the combined state.
4434     set_control(    _gvn.transform(result_reg));
4435     set_i_o(        _gvn.transform(result_i_o));
4436     set_all_memory( _gvn.transform(result_mem));
4437   } // original reexecute is set back here
4438 
4439   set_result(_gvn.transform(result_val));
4440   return true;
4441 }
4442 
4443 // If we have a tighly coupled allocation, the arraycopy may take care
4444 // of the array initialization. If one of the guards we insert between
4445 // the allocation and the arraycopy causes a deoptimization, an
4446 // unitialized array will escape the compiled method. To prevent that
4447 // we set the JVM state for uncommon traps between the allocation and
4448 // the arraycopy to the state before the allocation so, in case of
4449 // deoptimization, we'll reexecute the allocation and the
4450 // initialization.
4451 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4452   if (alloc != NULL) {
4453     ciMethod* trap_method = alloc->jvms()->method();
4454     int trap_bci = alloc->jvms()->bci();
4455 
4456     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4457           !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4458       // Make sure there's no store between the allocation and the
4459       // arraycopy otherwise visible side effects could be rexecuted
4460       // in case of deoptimization and cause incorrect execution.
4461       bool no_interfering_store = true;
4462       Node* mem = alloc->in(TypeFunc::Memory);
4463       if (mem->is_MergeMem()) {
4464         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4465           Node* n = mms.memory();
4466           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4467             assert(n->is_Store() || n->Opcode() == Op_ShenandoahWBMemProj, "what else?");
4468             no_interfering_store = false;
4469             break;
4470           }
4471         }
4472       } else {
4473         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4474           Node* n = mms.memory();
4475           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4476             assert(n->is_Store() || n->Opcode() == Op_ShenandoahWBMemProj, "what else?");
4477             no_interfering_store = false;
4478             break;
4479           }
4480         }
4481       }
4482 
4483       if (no_interfering_store) {
4484         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4485         uint size = alloc->req();
4486         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4487         old_jvms->set_map(sfpt);
4488         for (uint i = 0; i < size; i++) {
4489           sfpt->init_req(i, alloc->in(i));
4490         }
4491         // re-push array length for deoptimization
4492         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4493         old_jvms->set_sp(old_jvms->sp()+1);
4494         old_jvms->set_monoff(old_jvms->monoff()+1);
4495         old_jvms->set_scloff(old_jvms->scloff()+1);
4496         old_jvms->set_endoff(old_jvms->endoff()+1);
4497         old_jvms->set_should_reexecute(true);
4498 
4499         sfpt->set_i_o(map()->i_o());
4500         sfpt->set_memory(map()->memory());
4501         sfpt->set_control(map()->control());
4502 
4503         JVMState* saved_jvms = jvms();
4504         saved_reexecute_sp = _reexecute_sp;
4505 
4506         set_jvms(sfpt->jvms());
4507         _reexecute_sp = jvms()->sp();
4508 
4509         return saved_jvms;
4510       }
4511     }
4512   }
4513   return NULL;
4514 }
4515 
4516 // In case of a deoptimization, we restart execution at the
4517 // allocation, allocating a new array. We would leave an uninitialized
4518 // array in the heap that GCs wouldn't expect. Move the allocation
4519 // after the traps so we don't allocate the array if we
4520 // deoptimize. This is possible because tightly_coupled_allocation()
4521 // guarantees there's no observer of the allocated array at this point
4522 // and the control flow is simple enough.
4523 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4524                                                     int saved_reexecute_sp, uint new_idx) {
4525   if (saved_jvms != NULL && !stopped()) {
4526     assert(alloc != NULL, "only with a tightly coupled allocation");
4527     // restore JVM state to the state at the arraycopy
4528     saved_jvms->map()->set_control(map()->control());
4529     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4530     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4531     // If we've improved the types of some nodes (null check) while
4532     // emitting the guards, propagate them to the current state
4533     map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4534     set_jvms(saved_jvms);
4535     _reexecute_sp = saved_reexecute_sp;
4536 
4537     // Remove the allocation from above the guards
4538     CallProjections callprojs;
4539     alloc->extract_projections(&callprojs, true);
4540     InitializeNode* init = alloc->initialization();
4541     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4542     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4543     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4544     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4545 
4546     // move the allocation here (after the guards)
4547     _gvn.hash_delete(alloc);
4548     alloc->set_req(TypeFunc::Control, control());
4549     alloc->set_req(TypeFunc::I_O, i_o());
4550     Node *mem = reset_memory();
4551     set_all_memory(mem);
4552     alloc->set_req(TypeFunc::Memory, mem);
4553     set_control(init->proj_out_or_null(TypeFunc::Control));
4554     set_i_o(callprojs.fallthrough_ioproj);
4555 
4556     // Update memory as done in GraphKit::set_output_for_allocation()
4557     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4558     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4559     if (ary_type->isa_aryptr() && length_type != NULL) {
4560       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4561     }
4562     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4563     int            elemidx  = C->get_alias_index(telemref);
4564     set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4565     set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4566 
4567     Node* allocx = _gvn.transform(alloc);
4568     assert(allocx == alloc, "where has the allocation gone?");
4569     assert(dest->is_CheckCastPP(), "not an allocation result?");
4570 
4571     _gvn.hash_delete(dest);
4572     dest->set_req(0, control());
4573     Node* destx = _gvn.transform(dest);
4574     assert(destx == dest, "where has the allocation result gone?");
4575   }
4576 }
4577 
4578 
4579 //------------------------------inline_arraycopy-----------------------
4580 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4581 //                                                      Object dest, int destPos,
4582 //                                                      int length);
4583 bool LibraryCallKit::inline_arraycopy() {
4584   // Get the arguments.
4585   Node* src         = argument(0);  // type: oop
4586   Node* src_offset  = argument(1);  // type: int
4587   Node* dest        = argument(2);  // type: oop
4588   Node* dest_offset = argument(3);  // type: int
4589   Node* length      = argument(4);  // type: int
4590 
4591   uint new_idx = C->unique();
4592 
4593   // Check for allocation before we add nodes that would confuse
4594   // tightly_coupled_allocation()
4595   AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4596 
4597   int saved_reexecute_sp = -1;
4598   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4599   // See arraycopy_restore_alloc_state() comment
4600   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4601   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4602   // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4603   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4604 
4605   // The following tests must be performed
4606   // (1) src and dest are arrays.
4607   // (2) src and dest arrays must have elements of the same BasicType
4608   // (3) src and dest must not be null.
4609   // (4) src_offset must not be negative.
4610   // (5) dest_offset must not be negative.
4611   // (6) length must not be negative.
4612   // (7) src_offset + length must not exceed length of src.
4613   // (8) dest_offset + length must not exceed length of dest.
4614   // (9) each element of an oop array must be assignable
4615 
4616   // (3) src and dest must not be null.
4617   // always do this here because we need the JVM state for uncommon traps
4618   Node* null_ctl = top();
4619   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4620   assert(null_ctl->is_top(), "no null control here");
4621   dest = null_check(dest, T_ARRAY);
4622 
4623   if (!can_emit_guards) {
4624     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4625     // guards but the arraycopy node could still take advantage of a
4626     // tightly allocated allocation. tightly_coupled_allocation() is
4627     // called again to make sure it takes the null check above into
4628     // account: the null check is mandatory and if it caused an
4629     // uncommon trap to be emitted then the allocation can't be
4630     // considered tightly coupled in this context.
4631     alloc = tightly_coupled_allocation(dest, NULL);
4632   }
4633 
4634   bool validated = false;
4635 
4636   const Type* src_type  = _gvn.type(src);
4637   const Type* dest_type = _gvn.type(dest);
4638   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4639   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4640 
4641   // Do we have the type of src?
4642   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4643   // Do we have the type of dest?
4644   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4645   // Is the type for src from speculation?
4646   bool src_spec = false;
4647   // Is the type for dest from speculation?
4648   bool dest_spec = false;
4649 
4650   if ((!has_src || !has_dest) && can_emit_guards) {
4651     // We don't have sufficient type information, let's see if
4652     // speculative types can help. We need to have types for both src
4653     // and dest so that it pays off.
4654 
4655     // Do we already have or could we have type information for src
4656     bool could_have_src = has_src;
4657     // Do we already have or could we have type information for dest
4658     bool could_have_dest = has_dest;
4659 
4660     ciKlass* src_k = NULL;
4661     if (!has_src) {
4662       src_k = src_type->speculative_type_not_null();
4663       if (src_k != NULL && src_k->is_array_klass()) {
4664         could_have_src = true;
4665       }
4666     }
4667 
4668     ciKlass* dest_k = NULL;
4669     if (!has_dest) {
4670       dest_k = dest_type->speculative_type_not_null();
4671       if (dest_k != NULL && dest_k->is_array_klass()) {
4672         could_have_dest = true;
4673       }
4674     }
4675 
4676     if (could_have_src && could_have_dest) {
4677       // This is going to pay off so emit the required guards
4678       if (!has_src) {
4679         src = maybe_cast_profiled_obj(src, src_k, true);
4680         src_type  = _gvn.type(src);
4681         top_src  = src_type->isa_aryptr();
4682         has_src = (top_src != NULL && top_src->klass() != NULL);
4683         src_spec = true;
4684       }
4685       if (!has_dest) {
4686         dest = maybe_cast_profiled_obj(dest, dest_k, true);
4687         dest_type  = _gvn.type(dest);
4688         top_dest  = dest_type->isa_aryptr();
4689         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4690         dest_spec = true;
4691       }
4692     }
4693   }
4694 
4695   if (has_src && has_dest && can_emit_guards) {
4696     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
4697     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4698     if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
4699     if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;
4700 
4701     if (src_elem == dest_elem && src_elem == T_OBJECT) {
4702       // If both arrays are object arrays then having the exact types
4703       // for both will remove the need for a subtype check at runtime
4704       // before the call and may make it possible to pick a faster copy
4705       // routine (without a subtype check on every element)
4706       // Do we have the exact type of src?
4707       bool could_have_src = src_spec;
4708       // Do we have the exact type of dest?
4709       bool could_have_dest = dest_spec;
4710       ciKlass* src_k = top_src->klass();
4711       ciKlass* dest_k = top_dest->klass();
4712       if (!src_spec) {
4713         src_k = src_type->speculative_type_not_null();
4714         if (src_k != NULL && src_k->is_array_klass()) {
4715           could_have_src = true;
4716         }
4717       }
4718       if (!dest_spec) {
4719         dest_k = dest_type->speculative_type_not_null();
4720         if (dest_k != NULL && dest_k->is_array_klass()) {
4721           could_have_dest = true;
4722         }
4723       }
4724       if (could_have_src && could_have_dest) {
4725         // If we can have both exact types, emit the missing guards
4726         if (could_have_src && !src_spec) {
4727           src = maybe_cast_profiled_obj(src, src_k, true);
4728         }
4729         if (could_have_dest && !dest_spec) {
4730           dest = maybe_cast_profiled_obj(dest, dest_k, true);
4731         }
4732       }
4733     }
4734   }
4735 
4736   ciMethod* trap_method = method();
4737   int trap_bci = bci();
4738   if (saved_jvms != NULL) {
4739     trap_method = alloc->jvms()->method();
4740     trap_bci = alloc->jvms()->bci();
4741   }
4742 
4743   bool negative_length_guard_generated = false;
4744 
4745   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4746       can_emit_guards &&
4747       !src->is_top() && !dest->is_top()) {
4748     // validate arguments: enables transformation the ArrayCopyNode
4749     validated = true;
4750 
4751     RegionNode* slow_region = new RegionNode(1);
4752     record_for_igvn(slow_region);
4753 
4754     // (1) src and dest are arrays.
4755     generate_non_array_guard(load_object_klass(src), slow_region);
4756     generate_non_array_guard(load_object_klass(dest), slow_region);
4757 
4758     // (2) src and dest arrays must have elements of the same BasicType
4759     // done at macro expansion or at Ideal transformation time
4760 
4761     // (4) src_offset must not be negative.
4762     generate_negative_guard(src_offset, slow_region);
4763 
4764     // (5) dest_offset must not be negative.
4765     generate_negative_guard(dest_offset, slow_region);
4766 
4767     // (7) src_offset + length must not exceed length of src.
4768     generate_limit_guard(src_offset, length,
4769                          load_array_length(src),
4770                          slow_region);
4771 
4772     // (8) dest_offset + length must not exceed length of dest.
4773     generate_limit_guard(dest_offset, length,
4774                          load_array_length(dest),
4775                          slow_region);
4776 
4777     // (6) length must not be negative.
4778     // This is also checked in generate_arraycopy() during macro expansion, but
4779     // we also have to check it here for the case where the ArrayCopyNode will
4780     // be eliminated by Escape Analysis.
4781     if (EliminateAllocations) {
4782       generate_negative_guard(length, slow_region);
4783       negative_length_guard_generated = true;
4784     }
4785 
4786     // (9) each element of an oop array must be assignable
4787     Node* src_klass  = load_object_klass(src);
4788     Node* dest_klass = load_object_klass(dest);
4789     Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4790 
4791     if (not_subtype_ctrl != top()) {
4792       PreserveJVMState pjvms(this);
4793       set_control(not_subtype_ctrl);
4794       uncommon_trap(Deoptimization::Reason_intrinsic,
4795                     Deoptimization::Action_make_not_entrant);
4796       assert(stopped(), "Should be stopped");
4797     }
4798     {
4799       PreserveJVMState pjvms(this);
4800       set_control(_gvn.transform(slow_region));
4801       uncommon_trap(Deoptimization::Reason_intrinsic,
4802                     Deoptimization::Action_make_not_entrant);
4803       assert(stopped(), "Should be stopped");
4804     }
4805 
4806     const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4807     const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4808     src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4809   }
4810 
4811   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4812 
4813   if (stopped()) {
4814     return true;
4815   }
4816 
4817   Node* new_src = access_resolve(src, ACCESS_READ);
4818   Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4819 
4820   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4821                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
4822                                           // so the compiler has a chance to eliminate them: during macro expansion,
4823                                           // we have to set their control (CastPP nodes are eliminated).
4824                                           load_object_klass(src), load_object_klass(dest),
4825                                           load_array_length(src), load_array_length(dest));
4826 
4827   ac->set_arraycopy(validated);
4828 
4829   Node* n = _gvn.transform(ac);
4830   if (n == ac) {
4831     ac->connect_outputs(this);
4832   } else {
4833     assert(validated, "shouldn't transform if all arguments not validated");
4834     set_all_memory(n);
4835   }
4836   clear_upper_avx();
4837 
4838 
4839   return true;
4840 }
4841 
4842 
4843 // Helper function which determines if an arraycopy immediately follows
4844 // an allocation, with no intervening tests or other escapes for the object.
4845 AllocateArrayNode*
4846 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4847                                            RegionNode* slow_region) {
4848   if (stopped())             return NULL;  // no fast path
4849   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
4850 
4851   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4852   if (alloc == NULL)  return NULL;
4853 
4854   Node* rawmem = memory(Compile::AliasIdxRaw);
4855   // Is the allocation's memory state untouched?
4856   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4857     // Bail out if there have been raw-memory effects since the allocation.
4858     // (Example:  There might have been a call or safepoint.)
4859     return NULL;
4860   }
4861   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4862   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4863     return NULL;
4864   }
4865 
4866   // There must be no unexpected observers of this allocation.
4867   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4868     Node* obs = ptr->fast_out(i);
4869     if (obs != this->map()) {
4870       return NULL;
4871     }
4872   }
4873 
4874   // This arraycopy must unconditionally follow the allocation of the ptr.
4875   Node* alloc_ctl = ptr->in(0);
4876   assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4877 
4878   Node* ctl = control();
4879   while (ctl != alloc_ctl) {
4880     // There may be guards which feed into the slow_region.
4881     // Any other control flow means that we might not get a chance
4882     // to finish initializing the allocated object.
4883     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4884       IfNode* iff = ctl->in(0)->as_If();
4885       Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4886       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4887       if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4888         ctl = iff->in(0);       // This test feeds the known slow_region.
4889         continue;
4890       }
4891       // One more try:  Various low-level checks bottom out in
4892       // uncommon traps.  If the debug-info of the trap omits
4893       // any reference to the allocation, as we've already
4894       // observed, then there can be no objection to the trap.
4895       bool found_trap = false;
4896       for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4897         Node* obs = not_ctl->fast_out(j);
4898         if (obs->in(0) == not_ctl && obs->is_Call() &&
4899             (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4900           found_trap = true; break;
4901         }
4902       }
4903       if (found_trap) {
4904         ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
4905         continue;
4906       }
4907     }
4908     return NULL;
4909   }
4910 
4911   // If we get this far, we have an allocation which immediately
4912   // precedes the arraycopy, and we can take over zeroing the new object.
4913   // The arraycopy will finish the initialization, and provide
4914   // a new control state to which we will anchor the destination pointer.
4915 
4916   return alloc;
4917 }
4918 
4919 //-------------inline_encodeISOArray-----------------------------------
4920 // encode char[] to byte[] in ISO_8859_1
4921 bool LibraryCallKit::inline_encodeISOArray() {
4922   assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4923   // no receiver since it is static method
4924   Node *src         = argument(0);
4925   Node *src_offset  = argument(1);
4926   Node *dst         = argument(2);
4927   Node *dst_offset  = argument(3);
4928   Node *length      = argument(4);
4929 
4930   src = must_be_not_null(src, true);
4931   dst = must_be_not_null(dst, true);
4932 
4933   src = access_resolve(src, ACCESS_READ);
4934   dst = access_resolve(dst, ACCESS_WRITE);
4935 
4936   const Type* src_type = src->Value(&_gvn);
4937   const Type* dst_type = dst->Value(&_gvn);
4938   const TypeAryPtr* top_src = src_type->isa_aryptr();
4939   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4940   if (top_src  == NULL || top_src->klass()  == NULL ||
4941       top_dest == NULL || top_dest->klass() == NULL) {
4942     // failed array check
4943     return false;
4944   }
4945 
4946   // Figure out the size and type of the elements we will be copying.
4947   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4948   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4949   if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4950     return false;
4951   }
4952 
4953   Node* src_start = array_element_address(src, src_offset, T_CHAR);
4954   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4955   // 'src_start' points to src array + scaled offset
4956   // 'dst_start' points to dst array + scaled offset
4957 
4958   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4959   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4960   enc = _gvn.transform(enc);
4961   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4962   set_memory(res_mem, mtype);
4963   set_result(enc);
4964   clear_upper_avx();
4965 
4966   return true;
4967 }
4968 
4969 //-------------inline_multiplyToLen-----------------------------------
4970 bool LibraryCallKit::inline_multiplyToLen() {
4971   assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4972 
4973   address stubAddr = StubRoutines::multiplyToLen();
4974   if (stubAddr == NULL) {
4975     return false; // Intrinsic's stub is not implemented on this platform
4976   }
4977   const char* stubName = "multiplyToLen";
4978 
4979   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4980 
4981   // no receiver because it is a static method
4982   Node* x    = argument(0);
4983   Node* xlen = argument(1);
4984   Node* y    = argument(2);
4985   Node* ylen = argument(3);
4986   Node* z    = argument(4);
4987 
4988   x = must_be_not_null(x, true);
4989   y = must_be_not_null(y, true);
4990 
4991   x = access_resolve(x, ACCESS_READ);
4992   y = access_resolve(y, ACCESS_READ);
4993   z = access_resolve(z, ACCESS_WRITE);
4994 
4995   const Type* x_type = x->Value(&_gvn);
4996   const Type* y_type = y->Value(&_gvn);
4997   const TypeAryPtr* top_x = x_type->isa_aryptr();
4998   const TypeAryPtr* top_y = y_type->isa_aryptr();
4999   if (top_x  == NULL || top_x->klass()  == NULL ||
5000       top_y == NULL || top_y->klass() == NULL) {
5001     // failed array check
5002     return false;
5003   }
5004 
5005   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5006   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5007   if (x_elem != T_INT || y_elem != T_INT) {
5008     return false;
5009   }
5010 
5011   // Set the original stack and the reexecute bit for the interpreter to reexecute
5012   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5013   // on the return from z array allocation in runtime.
5014   { PreserveReexecuteState preexecs(this);
5015     jvms()->set_should_reexecute(true);
5016 
5017     Node* x_start = array_element_address(x, intcon(0), x_elem);
5018     Node* y_start = array_element_address(y, intcon(0), y_elem);
5019     // 'x_start' points to x array + scaled xlen
5020     // 'y_start' points to y array + scaled ylen
5021 
5022     // Allocate the result array
5023     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5024     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5025     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5026 
5027     IdealKit ideal(this);
5028 
5029 #define __ ideal.
5030      Node* one = __ ConI(1);
5031      Node* zero = __ ConI(0);
5032      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5033      __ set(need_alloc, zero);
5034      __ set(z_alloc, z);
5035      __ if_then(z, BoolTest::eq, null()); {
5036        __ increment (need_alloc, one);
5037      } __ else_(); {
5038        // Update graphKit memory and control from IdealKit.
5039        sync_kit(ideal);
5040        Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5041        cast->init_req(0, control());
5042        _gvn.set_type(cast, cast->bottom_type());
5043        C->record_for_igvn(cast);
5044 
5045        Node* zlen_arg = load_array_length(cast);
5046        // Update IdealKit memory and control from graphKit.
5047        __ sync_kit(this);
5048        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5049          __ increment (need_alloc, one);
5050        } __ end_if();
5051      } __ end_if();
5052 
5053      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5054        // Update graphKit memory and control from IdealKit.
5055        sync_kit(ideal);
5056        Node * narr = new_array(klass_node, zlen, 1);
5057        // Update IdealKit memory and control from graphKit.
5058        __ sync_kit(this);
5059        __ set(z_alloc, narr);
5060      } __ end_if();
5061 
5062      sync_kit(ideal);
5063      z = __ value(z_alloc);
5064      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5065      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5066      // Final sync IdealKit and GraphKit.
5067      final_sync(ideal);
5068 #undef __
5069 
5070     Node* z_start = array_element_address(z, intcon(0), T_INT);
5071 
5072     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5073                                    OptoRuntime::multiplyToLen_Type(),
5074                                    stubAddr, stubName, TypePtr::BOTTOM,
5075                                    x_start, xlen, y_start, ylen, z_start, zlen);
5076   } // original reexecute is set back here
5077 
5078   C->set_has_split_ifs(true); // Has chance for split-if optimization
5079   set_result(z);
5080   return true;
5081 }
5082 
5083 //-------------inline_squareToLen------------------------------------
5084 bool LibraryCallKit::inline_squareToLen() {
5085   assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5086 
5087   address stubAddr = StubRoutines::squareToLen();
5088   if (stubAddr == NULL) {
5089     return false; // Intrinsic's stub is not implemented on this platform
5090   }
5091   const char* stubName = "squareToLen";
5092 
5093   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5094 
5095   Node* x    = argument(0);
5096   Node* len  = argument(1);
5097   Node* z    = argument(2);
5098   Node* zlen = argument(3);
5099 
5100   x = must_be_not_null(x, true);
5101   z = must_be_not_null(z, true);
5102 
5103   x = access_resolve(x, ACCESS_READ);
5104   z = access_resolve(z, ACCESS_WRITE);
5105 
5106   const Type* x_type = x->Value(&_gvn);
5107   const Type* z_type = z->Value(&_gvn);
5108   const TypeAryPtr* top_x = x_type->isa_aryptr();
5109   const TypeAryPtr* top_z = z_type->isa_aryptr();
5110   if (top_x  == NULL || top_x->klass()  == NULL ||
5111       top_z  == NULL || top_z->klass()  == NULL) {
5112     // failed array check
5113     return false;
5114   }
5115 
5116   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5117   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5118   if (x_elem != T_INT || z_elem != T_INT) {
5119     return false;
5120   }
5121 
5122 
5123   Node* x_start = array_element_address(x, intcon(0), x_elem);
5124   Node* z_start = array_element_address(z, intcon(0), z_elem);
5125 
5126   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5127                                   OptoRuntime::squareToLen_Type(),
5128                                   stubAddr, stubName, TypePtr::BOTTOM,
5129                                   x_start, len, z_start, zlen);
5130 
5131   set_result(z);
5132   return true;
5133 }
5134 
5135 //-------------inline_mulAdd------------------------------------------
5136 bool LibraryCallKit::inline_mulAdd() {
5137   assert(UseMulAddIntrinsic, "not implemented on this platform");
5138 
5139   address stubAddr = StubRoutines::mulAdd();
5140   if (stubAddr == NULL) {
5141     return false; // Intrinsic's stub is not implemented on this platform
5142   }
5143   const char* stubName = "mulAdd";
5144 
5145   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5146 
5147   Node* out      = argument(0);
5148   Node* in       = argument(1);
5149   Node* offset   = argument(2);
5150   Node* len      = argument(3);
5151   Node* k        = argument(4);
5152 
5153   out = must_be_not_null(out, true);
5154 
5155   in = access_resolve(in, ACCESS_READ);
5156   out = access_resolve(out, ACCESS_WRITE);
5157 
5158   const Type* out_type = out->Value(&_gvn);
5159   const Type* in_type = in->Value(&_gvn);
5160   const TypeAryPtr* top_out = out_type->isa_aryptr();
5161   const TypeAryPtr* top_in = in_type->isa_aryptr();
5162   if (top_out  == NULL || top_out->klass()  == NULL ||
5163       top_in == NULL || top_in->klass() == NULL) {
5164     // failed array check
5165     return false;
5166   }
5167 
5168   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5169   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5170   if (out_elem != T_INT || in_elem != T_INT) {
5171     return false;
5172   }
5173 
5174   Node* outlen = load_array_length(out);
5175   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5176   Node* out_start = array_element_address(out, intcon(0), out_elem);
5177   Node* in_start = array_element_address(in, intcon(0), in_elem);
5178 
5179   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5180                                   OptoRuntime::mulAdd_Type(),
5181                                   stubAddr, stubName, TypePtr::BOTTOM,
5182                                   out_start,in_start, new_offset, len, k);
5183   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5184   set_result(result);
5185   return true;
5186 }
5187 
5188 //-------------inline_montgomeryMultiply-----------------------------------
5189 bool LibraryCallKit::inline_montgomeryMultiply() {
5190   address stubAddr = StubRoutines::montgomeryMultiply();
5191   if (stubAddr == NULL) {
5192     return false; // Intrinsic's stub is not implemented on this platform
5193   }
5194 
5195   assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5196   const char* stubName = "montgomery_multiply";
5197 
5198   assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5199 
5200   Node* a    = argument(0);
5201   Node* b    = argument(1);
5202   Node* n    = argument(2);
5203   Node* len  = argument(3);
5204   Node* inv  = argument(4);
5205   Node* m    = argument(6);
5206 
5207   a = access_resolve(a, ACCESS_READ);
5208   b = access_resolve(b, ACCESS_READ);
5209   n = access_resolve(n, ACCESS_READ);
5210   m = access_resolve(m, ACCESS_WRITE);
5211 
5212   const Type* a_type = a->Value(&_gvn);
5213   const TypeAryPtr* top_a = a_type->isa_aryptr();
5214   const Type* b_type = b->Value(&_gvn);
5215   const TypeAryPtr* top_b = b_type->isa_aryptr();
5216   const Type* n_type = a->Value(&_gvn);
5217   const TypeAryPtr* top_n = n_type->isa_aryptr();
5218   const Type* m_type = a->Value(&_gvn);
5219   const TypeAryPtr* top_m = m_type->isa_aryptr();
5220   if (top_a  == NULL || top_a->klass()  == NULL ||
5221       top_b == NULL || top_b->klass()  == NULL ||
5222       top_n == NULL || top_n->klass()  == NULL ||
5223       top_m == NULL || top_m->klass()  == NULL) {
5224     // failed array check
5225     return false;
5226   }
5227 
5228   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5229   BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5230   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5231   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5232   if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5233     return false;
5234   }
5235 
5236   // Make the call
5237   {
5238     Node* a_start = array_element_address(a, intcon(0), a_elem);
5239     Node* b_start = array_element_address(b, intcon(0), b_elem);
5240     Node* n_start = array_element_address(n, intcon(0), n_elem);
5241     Node* m_start = array_element_address(m, intcon(0), m_elem);
5242 
5243     Node* call = make_runtime_call(RC_LEAF,
5244                                    OptoRuntime::montgomeryMultiply_Type(),
5245                                    stubAddr, stubName, TypePtr::BOTTOM,
5246                                    a_start, b_start, n_start, len, inv, top(),
5247                                    m_start);
5248     set_result(m);
5249   }
5250 
5251   return true;
5252 }
5253 
5254 bool LibraryCallKit::inline_montgomerySquare() {
5255   address stubAddr = StubRoutines::montgomerySquare();
5256   if (stubAddr == NULL) {
5257     return false; // Intrinsic's stub is not implemented on this platform
5258   }
5259 
5260   assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5261   const char* stubName = "montgomery_square";
5262 
5263   assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5264 
5265   Node* a    = argument(0);
5266   Node* n    = argument(1);
5267   Node* len  = argument(2);
5268   Node* inv  = argument(3);
5269   Node* m    = argument(5);
5270 
5271   a = access_resolve(a, ACCESS_READ);
5272   n = access_resolve(n, ACCESS_READ);
5273   m = access_resolve(m, ACCESS_WRITE);
5274 
5275   const Type* a_type = a->Value(&_gvn);
5276   const TypeAryPtr* top_a = a_type->isa_aryptr();
5277   const Type* n_type = a->Value(&_gvn);
5278   const TypeAryPtr* top_n = n_type->isa_aryptr();
5279   const Type* m_type = a->Value(&_gvn);
5280   const TypeAryPtr* top_m = m_type->isa_aryptr();
5281   if (top_a  == NULL || top_a->klass()  == NULL ||
5282       top_n == NULL || top_n->klass()  == NULL ||
5283       top_m == NULL || top_m->klass()  == NULL) {
5284     // failed array check
5285     return false;
5286   }
5287 
5288   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5289   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5290   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5291   if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5292     return false;
5293   }
5294 
5295   // Make the call
5296   {
5297     Node* a_start = array_element_address(a, intcon(0), a_elem);
5298     Node* n_start = array_element_address(n, intcon(0), n_elem);
5299     Node* m_start = array_element_address(m, intcon(0), m_elem);
5300 
5301     Node* call = make_runtime_call(RC_LEAF,
5302                                    OptoRuntime::montgomerySquare_Type(),
5303                                    stubAddr, stubName, TypePtr::BOTTOM,
5304                                    a_start, n_start, len, inv, top(),
5305                                    m_start);
5306     set_result(m);
5307   }
5308 
5309   return true;
5310 }
5311 
5312 //-------------inline_vectorizedMismatch------------------------------
5313 bool LibraryCallKit::inline_vectorizedMismatch() {
5314   assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5315 
5316   address stubAddr = StubRoutines::vectorizedMismatch();
5317   if (stubAddr == NULL) {
5318     return false; // Intrinsic's stub is not implemented on this platform
5319   }
5320   const char* stubName = "vectorizedMismatch";
5321   int size_l = callee()->signature()->size();
5322   assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5323 
5324   Node* obja = argument(0);
5325   Node* aoffset = argument(1);
5326   Node* objb = argument(3);
5327   Node* boffset = argument(4);
5328   Node* length = argument(6);
5329   Node* scale = argument(7);
5330 
5331   const Type* a_type = obja->Value(&_gvn);
5332   const Type* b_type = objb->Value(&_gvn);
5333   const TypeAryPtr* top_a = a_type->isa_aryptr();
5334   const TypeAryPtr* top_b = b_type->isa_aryptr();
5335   if (top_a == NULL || top_a->klass() == NULL ||
5336     top_b == NULL || top_b->klass() == NULL) {
5337     // failed array check
5338     return false;
5339   }
5340 
5341   Node* call;
5342   jvms()->set_should_reexecute(true);
5343 
5344   obja = access_resolve(obja, ACCESS_READ);
5345   objb = access_resolve(objb, ACCESS_READ);
5346   Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5347   Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5348 
5349   call = make_runtime_call(RC_LEAF,
5350     OptoRuntime::vectorizedMismatch_Type(),
5351     stubAddr, stubName, TypePtr::BOTTOM,
5352     obja_adr, objb_adr, length, scale);
5353 
5354   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5355   set_result(result);
5356   return true;
5357 }
5358 
5359 /**
5360  * Calculate CRC32 for byte.
5361  * int java.util.zip.CRC32.update(int crc, int b)
5362  */
5363 bool LibraryCallKit::inline_updateCRC32() {
5364   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5365   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5366   // no receiver since it is static method
5367   Node* crc  = argument(0); // type: int
5368   Node* b    = argument(1); // type: int
5369 
5370   /*
5371    *    int c = ~ crc;
5372    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5373    *    b = b ^ (c >>> 8);
5374    *    crc = ~b;
5375    */
5376 
5377   Node* M1 = intcon(-1);
5378   crc = _gvn.transform(new XorINode(crc, M1));
5379   Node* result = _gvn.transform(new XorINode(crc, b));
5380   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5381 
5382   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5383   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5384   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5385   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5386 
5387   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5388   result = _gvn.transform(new XorINode(crc, result));
5389   result = _gvn.transform(new XorINode(result, M1));
5390   set_result(result);
5391   return true;
5392 }
5393 
5394 /**
5395  * Calculate CRC32 for byte[] array.
5396  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5397  */
5398 bool LibraryCallKit::inline_updateBytesCRC32() {
5399   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5400   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5401   // no receiver since it is static method
5402   Node* crc     = argument(0); // type: int
5403   Node* src     = argument(1); // type: oop
5404   Node* offset  = argument(2); // type: int
5405   Node* length  = argument(3); // type: int
5406 
5407   const Type* src_type = src->Value(&_gvn);
5408   const TypeAryPtr* top_src = src_type->isa_aryptr();
5409   if (top_src  == NULL || top_src->klass()  == NULL) {
5410     // failed array check
5411     return false;
5412   }
5413 
5414   // Figure out the size and type of the elements we will be copying.
5415   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5416   if (src_elem != T_BYTE) {
5417     return false;
5418   }
5419 
5420   // 'src_start' points to src array + scaled offset
5421   src = must_be_not_null(src, true);
5422   src = access_resolve(src, ACCESS_READ);
5423   Node* src_start = array_element_address(src, offset, src_elem);
5424 
5425   // We assume that range check is done by caller.
5426   // TODO: generate range check (offset+length < src.length) in debug VM.
5427 
5428   // Call the stub.
5429   address stubAddr = StubRoutines::updateBytesCRC32();
5430   const char *stubName = "updateBytesCRC32";
5431 
5432   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5433                                  stubAddr, stubName, TypePtr::BOTTOM,
5434                                  crc, src_start, length);
5435   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5436   set_result(result);
5437   return true;
5438 }
5439 
5440 /**
5441  * Calculate CRC32 for ByteBuffer.
5442  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5443  */
5444 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5445   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5446   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5447   // no receiver since it is static method
5448   Node* crc     = argument(0); // type: int
5449   Node* src     = argument(1); // type: long
5450   Node* offset  = argument(3); // type: int
5451   Node* length  = argument(4); // type: int
5452 
5453   src = ConvL2X(src);  // adjust Java long to machine word
5454   Node* base = _gvn.transform(new CastX2PNode(src));
5455   offset = ConvI2X(offset);
5456 
5457   // 'src_start' points to src array + scaled offset
5458   Node* src_start = basic_plus_adr(top(), base, offset);
5459 
5460   // Call the stub.
5461   address stubAddr = StubRoutines::updateBytesCRC32();
5462   const char *stubName = "updateBytesCRC32";
5463 
5464   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5465                                  stubAddr, stubName, TypePtr::BOTTOM,
5466                                  crc, src_start, length);
5467   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5468   set_result(result);
5469   return true;
5470 }
5471 
5472 //------------------------------get_table_from_crc32c_class-----------------------
5473 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5474   Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5475   assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5476 
5477   return table;
5478 }
5479 
5480 //------------------------------inline_updateBytesCRC32C-----------------------
5481 //
5482 // Calculate CRC32C for byte[] array.
5483 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5484 //
5485 bool LibraryCallKit::inline_updateBytesCRC32C() {
5486   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5487   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5488   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5489   // no receiver since it is a static method
5490   Node* crc     = argument(0); // type: int
5491   Node* src     = argument(1); // type: oop
5492   Node* offset  = argument(2); // type: int
5493   Node* end     = argument(3); // type: int
5494 
5495   Node* length = _gvn.transform(new SubINode(end, offset));
5496 
5497   const Type* src_type = src->Value(&_gvn);
5498   const TypeAryPtr* top_src = src_type->isa_aryptr();
5499   if (top_src  == NULL || top_src->klass()  == NULL) {
5500     // failed array check
5501     return false;
5502   }
5503 
5504   // Figure out the size and type of the elements we will be copying.
5505   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5506   if (src_elem != T_BYTE) {
5507     return false;
5508   }
5509 
5510   // 'src_start' points to src array + scaled offset
5511   src = must_be_not_null(src, true);
5512   src = access_resolve(src, ACCESS_READ);
5513   Node* src_start = array_element_address(src, offset, src_elem);
5514 
5515   // static final int[] byteTable in class CRC32C
5516   Node* table = get_table_from_crc32c_class(callee()->holder());
5517   table = must_be_not_null(table, true);
5518   table = access_resolve(table, ACCESS_READ);
5519   Node* table_start = array_element_address(table, intcon(0), T_INT);
5520 
5521   // We assume that range check is done by caller.
5522   // TODO: generate range check (offset+length < src.length) in debug VM.
5523 
5524   // Call the stub.
5525   address stubAddr = StubRoutines::updateBytesCRC32C();
5526   const char *stubName = "updateBytesCRC32C";
5527 
5528   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5529                                  stubAddr, stubName, TypePtr::BOTTOM,
5530                                  crc, src_start, length, table_start);
5531   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5532   set_result(result);
5533   return true;
5534 }
5535 
5536 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5537 //
5538 // Calculate CRC32C for DirectByteBuffer.
5539 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5540 //
5541 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5542   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5543   assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5544   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5545   // no receiver since it is a static method
5546   Node* crc     = argument(0); // type: int
5547   Node* src     = argument(1); // type: long
5548   Node* offset  = argument(3); // type: int
5549   Node* end     = argument(4); // type: int
5550 
5551   Node* length = _gvn.transform(new SubINode(end, offset));
5552 
5553   src = ConvL2X(src);  // adjust Java long to machine word
5554   Node* base = _gvn.transform(new CastX2PNode(src));
5555   offset = ConvI2X(offset);
5556 
5557   // 'src_start' points to src array + scaled offset
5558   Node* src_start = basic_plus_adr(top(), base, offset);
5559 
5560   // static final int[] byteTable in class CRC32C
5561   Node* table = get_table_from_crc32c_class(callee()->holder());
5562   table = must_be_not_null(table, true);
5563   table = access_resolve(table, ACCESS_READ);
5564   Node* table_start = array_element_address(table, intcon(0), T_INT);
5565 
5566   // Call the stub.
5567   address stubAddr = StubRoutines::updateBytesCRC32C();
5568   const char *stubName = "updateBytesCRC32C";
5569 
5570   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5571                                  stubAddr, stubName, TypePtr::BOTTOM,
5572                                  crc, src_start, length, table_start);
5573   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5574   set_result(result);
5575   return true;
5576 }
5577 
5578 //------------------------------inline_updateBytesAdler32----------------------
5579 //
5580 // Calculate Adler32 checksum for byte[] array.
5581 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5582 //
5583 bool LibraryCallKit::inline_updateBytesAdler32() {
5584   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5585   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5586   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5587   // no receiver since it is static method
5588   Node* crc     = argument(0); // type: int
5589   Node* src     = argument(1); // type: oop
5590   Node* offset  = argument(2); // type: int
5591   Node* length  = argument(3); // type: int
5592 
5593   const Type* src_type = src->Value(&_gvn);
5594   const TypeAryPtr* top_src = src_type->isa_aryptr();
5595   if (top_src  == NULL || top_src->klass()  == NULL) {
5596     // failed array check
5597     return false;
5598   }
5599 
5600   // Figure out the size and type of the elements we will be copying.
5601   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5602   if (src_elem != T_BYTE) {
5603     return false;
5604   }
5605 
5606   // 'src_start' points to src array + scaled offset
5607   src = access_resolve(src, ACCESS_READ);
5608   Node* src_start = array_element_address(src, offset, src_elem);
5609 
5610   // We assume that range check is done by caller.
5611   // TODO: generate range check (offset+length < src.length) in debug VM.
5612 
5613   // Call the stub.
5614   address stubAddr = StubRoutines::updateBytesAdler32();
5615   const char *stubName = "updateBytesAdler32";
5616 
5617   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5618                                  stubAddr, stubName, TypePtr::BOTTOM,
5619                                  crc, src_start, length);
5620   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5621   set_result(result);
5622   return true;
5623 }
5624 
5625 //------------------------------inline_updateByteBufferAdler32---------------
5626 //
5627 // Calculate Adler32 checksum for DirectByteBuffer.
5628 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5629 //
5630 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5631   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5632   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5633   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5634   // no receiver since it is static method
5635   Node* crc     = argument(0); // type: int
5636   Node* src     = argument(1); // type: long
5637   Node* offset  = argument(3); // type: int
5638   Node* length  = argument(4); // type: int
5639 
5640   src = ConvL2X(src);  // adjust Java long to machine word
5641   Node* base = _gvn.transform(new CastX2PNode(src));
5642   offset = ConvI2X(offset);
5643 
5644   // 'src_start' points to src array + scaled offset
5645   Node* src_start = basic_plus_adr(top(), base, offset);
5646 
5647   // Call the stub.
5648   address stubAddr = StubRoutines::updateBytesAdler32();
5649   const char *stubName = "updateBytesAdler32";
5650 
5651   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5652                                  stubAddr, stubName, TypePtr::BOTTOM,
5653                                  crc, src_start, length);
5654 
5655   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5656   set_result(result);
5657   return true;
5658 }
5659 
5660 //----------------------------inline_reference_get----------------------------
5661 // public T java.lang.ref.Reference.get();
5662 bool LibraryCallKit::inline_reference_get() {
5663   const int referent_offset = java_lang_ref_Reference::referent_offset;
5664   guarantee(referent_offset > 0, "should have already been set");
5665 
5666   // Get the argument:
5667   Node* reference_obj = null_check_receiver();
5668   if (stopped()) return true;
5669 
5670   const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5671   assert(tinst != NULL, "obj is null");
5672   assert(tinst->klass()->is_loaded(), "obj is not loaded");
5673   ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5674   ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5675                                                      ciSymbol::make("Ljava/lang/Object;"),
5676                                                      false);
5677   assert (field != NULL, "undefined field");
5678 
5679   Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5680   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5681 
5682   ciInstanceKlass* klass = env()->Object_klass();
5683   const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5684 
5685   DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5686   Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5687   // Add memory barrier to prevent commoning reads from this field
5688   // across safepoint since GC can change its value.
5689   insert_mem_bar(Op_MemBarCPUOrder);
5690 
5691   set_result(result);
5692   return true;
5693 }
5694 
5695 
5696 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5697                                               bool is_exact=true, bool is_static=false,
5698                                               ciInstanceKlass * fromKls=NULL) {
5699   if (fromKls == NULL) {
5700     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5701     assert(tinst != NULL, "obj is null");
5702     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5703     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5704     fromKls = tinst->klass()->as_instance_klass();
5705   } else {
5706     assert(is_static, "only for static field access");
5707   }
5708   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5709                                               ciSymbol::make(fieldTypeString),
5710                                               is_static);
5711 
5712   assert (field != NULL, "undefined field");
5713   if (field == NULL) return (Node *) NULL;
5714 
5715   if (is_static) {
5716     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5717     fromObj = makecon(tip);
5718   }
5719 
5720   // Next code  copied from Parse::do_get_xxx():
5721 
5722   // Compute address and memory type.
5723   int offset  = field->offset_in_bytes();
5724   bool is_vol = field->is_volatile();
5725   ciType* field_klass = field->type();
5726   assert(field_klass->is_loaded(), "should be loaded");
5727   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5728   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5729   BasicType bt = field->layout_type();
5730 
5731   // Build the resultant type of the load
5732   const Type *type;
5733   if (bt == T_OBJECT) {
5734     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5735   } else {
5736     type = Type::get_const_basic_type(bt);
5737   }
5738 
5739   DecoratorSet decorators = IN_HEAP;
5740 
5741   if (is_vol) {
5742     decorators |= MO_SEQ_CST;
5743   }
5744 
5745   return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5746 }
5747 
5748 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5749                                                  bool is_exact = true, bool is_static = false,
5750                                                  ciInstanceKlass * fromKls = NULL) {
5751   if (fromKls == NULL) {
5752     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5753     assert(tinst != NULL, "obj is null");
5754     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5755     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5756     fromKls = tinst->klass()->as_instance_klass();
5757   }
5758   else {
5759     assert(is_static, "only for static field access");
5760   }
5761   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5762     ciSymbol::make(fieldTypeString),
5763     is_static);
5764 
5765   assert(field != NULL, "undefined field");
5766   assert(!field->is_volatile(), "not defined for volatile fields");
5767 
5768   if (is_static) {
5769     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5770     fromObj = makecon(tip);
5771   }
5772 
5773   // Next code  copied from Parse::do_get_xxx():
5774 
5775   // Compute address and memory type.
5776   int offset = field->offset_in_bytes();
5777   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5778 
5779   return adr;
5780 }
5781 
5782 //------------------------------inline_aescrypt_Block-----------------------
5783 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5784   address stubAddr = NULL;
5785   const char *stubName;
5786   assert(UseAES, "need AES instruction support");
5787 
5788   switch(id) {
5789   case vmIntrinsics::_aescrypt_encryptBlock:
5790     stubAddr = StubRoutines::aescrypt_encryptBlock();
5791     stubName = "aescrypt_encryptBlock";
5792     break;
5793   case vmIntrinsics::_aescrypt_decryptBlock:
5794     stubAddr = StubRoutines::aescrypt_decryptBlock();
5795     stubName = "aescrypt_decryptBlock";
5796     break;
5797   default:
5798     break;
5799   }
5800   if (stubAddr == NULL) return false;
5801 
5802   Node* aescrypt_object = argument(0);
5803   Node* src             = argument(1);
5804   Node* src_offset      = argument(2);
5805   Node* dest            = argument(3);
5806   Node* dest_offset     = argument(4);
5807 
5808   src = must_be_not_null(src, true);
5809   dest = must_be_not_null(dest, true);
5810 
5811   src = access_resolve(src, ACCESS_READ);
5812   dest = access_resolve(dest, ACCESS_WRITE);
5813 
5814   // (1) src and dest are arrays.
5815   const Type* src_type = src->Value(&_gvn);
5816   const Type* dest_type = dest->Value(&_gvn);
5817   const TypeAryPtr* top_src = src_type->isa_aryptr();
5818   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5819   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5820 
5821   // for the quick and dirty code we will skip all the checks.
5822   // we are just trying to get the call to be generated.
5823   Node* src_start  = src;
5824   Node* dest_start = dest;
5825   if (src_offset != NULL || dest_offset != NULL) {
5826     assert(src_offset != NULL && dest_offset != NULL, "");
5827     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5828     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5829   }
5830 
5831   // now need to get the start of its expanded key array
5832   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5833   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5834   if (k_start == NULL) return false;
5835 
5836   if (Matcher::pass_original_key_for_aes()) {
5837     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5838     // compatibility issues between Java key expansion and SPARC crypto instructions
5839     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5840     if (original_k_start == NULL) return false;
5841 
5842     // Call the stub.
5843     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5844                       stubAddr, stubName, TypePtr::BOTTOM,
5845                       src_start, dest_start, k_start, original_k_start);
5846   } else {
5847     // Call the stub.
5848     make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5849                       stubAddr, stubName, TypePtr::BOTTOM,
5850                       src_start, dest_start, k_start);
5851   }
5852 
5853   return true;
5854 }
5855 
5856 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5857 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5858   address stubAddr = NULL;
5859   const char *stubName = NULL;
5860 
5861   assert(UseAES, "need AES instruction support");
5862 
5863   switch(id) {
5864   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5865     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5866     stubName = "cipherBlockChaining_encryptAESCrypt";
5867     break;
5868   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5869     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5870     stubName = "cipherBlockChaining_decryptAESCrypt";
5871     break;
5872   default:
5873     break;
5874   }
5875   if (stubAddr == NULL) return false;
5876 
5877   Node* cipherBlockChaining_object = argument(0);
5878   Node* src                        = argument(1);
5879   Node* src_offset                 = argument(2);
5880   Node* len                        = argument(3);
5881   Node* dest                       = argument(4);
5882   Node* dest_offset                = argument(5);
5883 
5884   src = must_be_not_null(src, false);
5885   dest = must_be_not_null(dest, false);
5886 
5887   src = access_resolve(src, ACCESS_READ);
5888   dest = access_resolve(dest, ACCESS_WRITE);
5889 
5890   // (1) src and dest are arrays.
5891   const Type* src_type = src->Value(&_gvn);
5892   const Type* dest_type = dest->Value(&_gvn);
5893   const TypeAryPtr* top_src = src_type->isa_aryptr();
5894   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5895   assert (top_src  != NULL && top_src->klass()  != NULL
5896           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5897 
5898   // checks are the responsibility of the caller
5899   Node* src_start  = src;
5900   Node* dest_start = dest;
5901   if (src_offset != NULL || dest_offset != NULL) {
5902     assert(src_offset != NULL && dest_offset != NULL, "");
5903     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5904     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5905   }
5906 
5907   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5908   // (because of the predicated logic executed earlier).
5909   // so we cast it here safely.
5910   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5911 
5912   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5913   if (embeddedCipherObj == NULL) return false;
5914 
5915   // cast it to what we know it will be at runtime
5916   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5917   assert(tinst != NULL, "CBC obj is null");
5918   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5919   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5920   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5921 
5922   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5923   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5924   const TypeOopPtr* xtype = aklass->as_instance_type();
5925   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5926   aescrypt_object = _gvn.transform(aescrypt_object);
5927 
5928   // we need to get the start of the aescrypt_object's expanded key array
5929   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5930   if (k_start == NULL) return false;
5931 
5932   // similarly, get the start address of the r vector
5933   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5934   if (objRvec == NULL) return false;
5935   objRvec = access_resolve(objRvec, ACCESS_WRITE);
5936   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5937 
5938   Node* cbcCrypt;
5939   if (Matcher::pass_original_key_for_aes()) {
5940     // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5941     // compatibility issues between Java key expansion and SPARC crypto instructions
5942     Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5943     if (original_k_start == NULL) return false;
5944 
5945     // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5946     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5947                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5948                                  stubAddr, stubName, TypePtr::BOTTOM,
5949                                  src_start, dest_start, k_start, r_start, len, original_k_start);
5950   } else {
5951     // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5952     cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5953                                  OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5954                                  stubAddr, stubName, TypePtr::BOTTOM,
5955                                  src_start, dest_start, k_start, r_start, len);
5956   }
5957 
5958   // return cipher length (int)
5959   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5960   set_result(retvalue);
5961   return true;
5962 }
5963 
5964 //------------------------------inline_counterMode_AESCrypt-----------------------
5965 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
5966   assert(UseAES, "need AES instruction support");
5967   if (!UseAESCTRIntrinsics) return false;
5968 
5969   address stubAddr = NULL;
5970   const char *stubName = NULL;
5971   if (id == vmIntrinsics::_counterMode_AESCrypt) {
5972     stubAddr = StubRoutines::counterMode_AESCrypt();
5973     stubName = "counterMode_AESCrypt";
5974   }
5975   if (stubAddr == NULL) return false;
5976 
5977   Node* counterMode_object = argument(0);
5978   Node* src = argument(1);
5979   Node* src_offset = argument(2);
5980   Node* len = argument(3);
5981   Node* dest = argument(4);
5982   Node* dest_offset = argument(5);
5983 
5984   src = access_resolve(src, ACCESS_READ);
5985   dest = access_resolve(dest, ACCESS_WRITE);
5986   counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
5987 
5988   // (1) src and dest are arrays.
5989   const Type* src_type = src->Value(&_gvn);
5990   const Type* dest_type = dest->Value(&_gvn);
5991   const TypeAryPtr* top_src = src_type->isa_aryptr();
5992   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5993   assert(top_src != NULL && top_src->klass() != NULL &&
5994          top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5995 
5996   // checks are the responsibility of the caller
5997   Node* src_start = src;
5998   Node* dest_start = dest;
5999   if (src_offset != NULL || dest_offset != NULL) {
6000     assert(src_offset != NULL && dest_offset != NULL, "");
6001     src_start = array_element_address(src, src_offset, T_BYTE);
6002     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6003   }
6004 
6005   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6006   // (because of the predicated logic executed earlier).
6007   // so we cast it here safely.
6008   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6009   Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6010   if (embeddedCipherObj == NULL) return false;
6011   // cast it to what we know it will be at runtime
6012   const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6013   assert(tinst != NULL, "CTR obj is null");
6014   assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6015   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6016   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6017   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6018   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6019   const TypeOopPtr* xtype = aklass->as_instance_type();
6020   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6021   aescrypt_object = _gvn.transform(aescrypt_object);
6022   // we need to get the start of the aescrypt_object's expanded key array
6023   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6024   if (k_start == NULL) return false;
6025   // similarly, get the start address of the r vector
6026   Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6027   if (obj_counter == NULL) return false;
6028   obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6029   Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6030 
6031   Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6032   if (saved_encCounter == NULL) return false;
6033   saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6034   Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6035   Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6036 
6037   Node* ctrCrypt;
6038   if (Matcher::pass_original_key_for_aes()) {
6039     // no SPARC version for AES/CTR intrinsics now.
6040     return false;
6041   }
6042   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6043   ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6044                                OptoRuntime::counterMode_aescrypt_Type(),
6045                                stubAddr, stubName, TypePtr::BOTTOM,
6046                                src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6047 
6048   // return cipher length (int)
6049   Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6050   set_result(retvalue);
6051   return true;
6052 }
6053 
6054 //------------------------------get_key_start_from_aescrypt_object-----------------------
6055 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6056 #if defined(PPC64) || defined(S390)
6057   // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6058   // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6059   // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6060   // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6061   Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6062   assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6063   if (objSessionK == NULL) {
6064     return (Node *) NULL;
6065   }
6066   Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6067 #else
6068   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6069 #endif // PPC64
6070   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6071   if (objAESCryptKey == NULL) return (Node *) NULL;
6072 
6073   // now have the array, need to get the start address of the K array
6074   objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6075   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6076   return k_start;
6077 }
6078 
6079 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6080 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6081   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6082   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6083   if (objAESCryptKey == NULL) return (Node *) NULL;
6084 
6085   // now have the array, need to get the start address of the lastKey array
6086   objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6087   Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6088   return original_k_start;
6089 }
6090 
6091 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6092 // Return node representing slow path of predicate check.
6093 // the pseudo code we want to emulate with this predicate is:
6094 // for encryption:
6095 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6096 // for decryption:
6097 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6098 //    note cipher==plain is more conservative than the original java code but that's OK
6099 //
6100 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6101   // The receiver was checked for NULL already.
6102   Node* objCBC = argument(0);
6103 
6104   Node* src = argument(1);
6105   Node* dest = argument(4);
6106 
6107   // Load embeddedCipher field of CipherBlockChaining object.
6108   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6109 
6110   // get AESCrypt klass for instanceOf check
6111   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6112   // will have same classloader as CipherBlockChaining object
6113   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6114   assert(tinst != NULL, "CBCobj is null");
6115   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6116 
6117   // we want to do an instanceof comparison against the AESCrypt class
6118   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6119   if (!klass_AESCrypt->is_loaded()) {
6120     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6121     Node* ctrl = control();
6122     set_control(top()); // no regular fast path
6123     return ctrl;
6124   }
6125 
6126   src = must_be_not_null(src, true);
6127   dest = must_be_not_null(dest, true);
6128 
6129   // Resolve oops to stable for CmpP below.
6130   src = access_resolve(src, 0);
6131   dest = access_resolve(dest, 0);
6132 
6133   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6134 
6135   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6136   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
6137   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6138 
6139   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6140 
6141   // for encryption, we are done
6142   if (!decrypting)
6143     return instof_false;  // even if it is NULL
6144 
6145   // for decryption, we need to add a further check to avoid
6146   // taking the intrinsic path when cipher and plain are the same
6147   // see the original java code for why.
6148   RegionNode* region = new RegionNode(3);
6149   region->init_req(1, instof_false);
6150 
6151   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6152   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6153   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6154   region->init_req(2, src_dest_conjoint);
6155 
6156   record_for_igvn(region);
6157   return _gvn.transform(region);
6158 }
6159 
6160 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6161 // Return node representing slow path of predicate check.
6162 // the pseudo code we want to emulate with this predicate is:
6163 // for encryption:
6164 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6165 // for decryption:
6166 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6167 //    note cipher==plain is more conservative than the original java code but that's OK
6168 //
6169 
6170 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6171   // The receiver was checked for NULL already.
6172   Node* objCTR = argument(0);
6173 
6174   // Load embeddedCipher field of CipherBlockChaining object.
6175   Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6176 
6177   // get AESCrypt klass for instanceOf check
6178   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6179   // will have same classloader as CipherBlockChaining object
6180   const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6181   assert(tinst != NULL, "CTRobj is null");
6182   assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6183 
6184   // we want to do an instanceof comparison against the AESCrypt class
6185   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6186   if (!klass_AESCrypt->is_loaded()) {
6187     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6188     Node* ctrl = control();
6189     set_control(top()); // no regular fast path
6190     return ctrl;
6191   }
6192 
6193   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6194   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6195   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6196   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6197   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6198 
6199   return instof_false; // even if it is NULL
6200 }
6201 
6202 //------------------------------inline_ghash_processBlocks
6203 bool LibraryCallKit::inline_ghash_processBlocks() {
6204   address stubAddr;
6205   const char *stubName;
6206   assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6207 
6208   stubAddr = StubRoutines::ghash_processBlocks();
6209   stubName = "ghash_processBlocks";
6210 
6211   Node* data           = argument(0);
6212   Node* offset         = argument(1);
6213   Node* len            = argument(2);
6214   Node* state          = argument(3);
6215   Node* subkeyH        = argument(4);
6216 
6217   state = must_be_not_null(state, true);
6218   subkeyH = must_be_not_null(subkeyH, true);
6219   data = must_be_not_null(data, true);
6220 
6221   state = access_resolve(state, ACCESS_WRITE);
6222   subkeyH = access_resolve(subkeyH, ACCESS_READ);
6223   data = access_resolve(data, ACCESS_READ);
6224 
6225   Node* state_start  = array_element_address(state, intcon(0), T_LONG);
6226   assert(state_start, "state is NULL");
6227   Node* subkeyH_start  = array_element_address(subkeyH, intcon(0), T_LONG);
6228   assert(subkeyH_start, "subkeyH is NULL");
6229   Node* data_start  = array_element_address(data, offset, T_BYTE);
6230   assert(data_start, "data is NULL");
6231 
6232   Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6233                                   OptoRuntime::ghash_processBlocks_Type(),
6234                                   stubAddr, stubName, TypePtr::BOTTOM,
6235                                   state_start, subkeyH_start, data_start, len);
6236   return true;
6237 }
6238 
6239 bool LibraryCallKit::inline_base64_encodeBlock() {
6240   address stubAddr;
6241   const char *stubName;
6242   assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6243   assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6244   stubAddr = StubRoutines::base64_encodeBlock();
6245   stubName = "encodeBlock";
6246 
6247   if (!stubAddr) return false;
6248   Node* base64obj = argument(0);
6249   Node* src = argument(1);
6250   Node* offset = argument(2);
6251   Node* len = argument(3);
6252   Node* dest = argument(4);
6253   Node* dp = argument(5);
6254   Node* isURL = argument(6);
6255 
6256   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6257   assert(src_start, "source array is NULL");
6258   Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6259   assert(dest_start, "destination array is NULL");
6260 
6261   Node* base64 = make_runtime_call(RC_LEAF,
6262                                    OptoRuntime::base64_encodeBlock_Type(),
6263                                    stubAddr, stubName, TypePtr::BOTTOM,
6264                                    src_start, offset, len, dest_start, dp, isURL);
6265   return true;
6266 }
6267 
6268 //------------------------------inline_sha_implCompress-----------------------
6269 //
6270 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6271 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6272 //
6273 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6274 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6275 //
6276 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6277 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6278 //
6279 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6280   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6281 
6282   Node* sha_obj = argument(0);
6283   Node* src     = argument(1); // type oop
6284   Node* ofs     = argument(2); // type int
6285 
6286   const Type* src_type = src->Value(&_gvn);
6287   const TypeAryPtr* top_src = src_type->isa_aryptr();
6288   if (top_src  == NULL || top_src->klass()  == NULL) {
6289     // failed array check
6290     return false;
6291   }
6292   // Figure out the size and type of the elements we will be copying.
6293   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6294   if (src_elem != T_BYTE) {
6295     return false;
6296   }
6297   // 'src_start' points to src array + offset
6298   src = must_be_not_null(src, true);
6299   src = access_resolve(src, ACCESS_READ);
6300   Node* src_start = array_element_address(src, ofs, src_elem);
6301   Node* state = NULL;
6302   address stubAddr;
6303   const char *stubName;
6304 
6305   switch(id) {
6306   case vmIntrinsics::_sha_implCompress:
6307     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6308     state = get_state_from_sha_object(sha_obj);
6309     stubAddr = StubRoutines::sha1_implCompress();
6310     stubName = "sha1_implCompress";
6311     break;
6312   case vmIntrinsics::_sha2_implCompress:
6313     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6314     state = get_state_from_sha_object(sha_obj);
6315     stubAddr = StubRoutines::sha256_implCompress();
6316     stubName = "sha256_implCompress";
6317     break;
6318   case vmIntrinsics::_sha5_implCompress:
6319     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6320     state = get_state_from_sha5_object(sha_obj);
6321     stubAddr = StubRoutines::sha512_implCompress();
6322     stubName = "sha512_implCompress";
6323     break;
6324   default:
6325     fatal_unexpected_iid(id);
6326     return false;
6327   }
6328   if (state == NULL) return false;
6329 
6330   // Call the stub.
6331   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6332                                  stubAddr, stubName, TypePtr::BOTTOM,
6333                                  src_start, state);
6334 
6335   return true;
6336 }
6337 
6338 //------------------------------inline_digestBase_implCompressMB-----------------------
6339 //
6340 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6341 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6342 //
6343 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6344   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6345          "need SHA1/SHA256/SHA512 instruction support");
6346   assert((uint)predicate < 3, "sanity");
6347   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6348 
6349   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6350   Node* src            = argument(1); // byte[] array
6351   Node* ofs            = argument(2); // type int
6352   Node* limit          = argument(3); // type int
6353 
6354   const Type* src_type = src->Value(&_gvn);
6355   const TypeAryPtr* top_src = src_type->isa_aryptr();
6356   if (top_src  == NULL || top_src->klass()  == NULL) {
6357     // failed array check
6358     return false;
6359   }
6360   // Figure out the size and type of the elements we will be copying.
6361   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6362   if (src_elem != T_BYTE) {
6363     return false;
6364   }
6365   // 'src_start' points to src array + offset
6366   src = must_be_not_null(src, false);
6367   src = access_resolve(src, ACCESS_READ);
6368   Node* src_start = array_element_address(src, ofs, src_elem);
6369 
6370   const char* klass_SHA_name = NULL;
6371   const char* stub_name = NULL;
6372   address     stub_addr = NULL;
6373   bool        long_state = false;
6374 
6375   switch (predicate) {
6376   case 0:
6377     if (UseSHA1Intrinsics) {
6378       klass_SHA_name = "sun/security/provider/SHA";
6379       stub_name = "sha1_implCompressMB";
6380       stub_addr = StubRoutines::sha1_implCompressMB();
6381     }
6382     break;
6383   case 1:
6384     if (UseSHA256Intrinsics) {
6385       klass_SHA_name = "sun/security/provider/SHA2";
6386       stub_name = "sha256_implCompressMB";
6387       stub_addr = StubRoutines::sha256_implCompressMB();
6388     }
6389     break;
6390   case 2:
6391     if (UseSHA512Intrinsics) {
6392       klass_SHA_name = "sun/security/provider/SHA5";
6393       stub_name = "sha512_implCompressMB";
6394       stub_addr = StubRoutines::sha512_implCompressMB();
6395       long_state = true;
6396     }
6397     break;
6398   default:
6399     fatal("unknown SHA intrinsic predicate: %d", predicate);
6400   }
6401   if (klass_SHA_name != NULL) {
6402     // get DigestBase klass to lookup for SHA klass
6403     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6404     assert(tinst != NULL, "digestBase_obj is not instance???");
6405     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6406 
6407     ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6408     assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6409     ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6410     return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6411   }
6412   return false;
6413 }
6414 //------------------------------inline_sha_implCompressMB-----------------------
6415 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6416                                                bool long_state, address stubAddr, const char *stubName,
6417                                                Node* src_start, Node* ofs, Node* limit) {
6418   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6419   const TypeOopPtr* xtype = aklass->as_instance_type();
6420   Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6421   sha_obj = _gvn.transform(sha_obj);
6422 
6423   Node* state;
6424   if (long_state) {
6425     state = get_state_from_sha5_object(sha_obj);
6426   } else {
6427     state = get_state_from_sha_object(sha_obj);
6428   }
6429   if (state == NULL) return false;
6430 
6431   // Call the stub.
6432   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6433                                  OptoRuntime::digestBase_implCompressMB_Type(),
6434                                  stubAddr, stubName, TypePtr::BOTTOM,
6435                                  src_start, state, ofs, limit);
6436   // return ofs (int)
6437   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6438   set_result(result);
6439 
6440   return true;
6441 }
6442 
6443 //------------------------------get_state_from_sha_object-----------------------
6444 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6445   Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6446   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6447   if (sha_state == NULL) return (Node *) NULL;
6448 
6449   // now have the array, need to get the start address of the state array
6450   sha_state = access_resolve(sha_state, ACCESS_WRITE);
6451   Node* state = array_element_address(sha_state, intcon(0), T_INT);
6452   return state;
6453 }
6454 
6455 //------------------------------get_state_from_sha5_object-----------------------
6456 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6457   Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6458   assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6459   if (sha_state == NULL) return (Node *) NULL;
6460 
6461   // now have the array, need to get the start address of the state array
6462   sha_state = access_resolve(sha_state, ACCESS_WRITE);
6463   Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6464   return state;
6465 }
6466 
6467 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6468 // Return node representing slow path of predicate check.
6469 // the pseudo code we want to emulate with this predicate is:
6470 //    if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6471 //
6472 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6473   assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6474          "need SHA1/SHA256/SHA512 instruction support");
6475   assert((uint)predicate < 3, "sanity");
6476 
6477   // The receiver was checked for NULL already.
6478   Node* digestBaseObj = argument(0);
6479 
6480   // get DigestBase klass for instanceOf check
6481   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6482   assert(tinst != NULL, "digestBaseObj is null");
6483   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6484 
6485   const char* klass_SHA_name = NULL;
6486   switch (predicate) {
6487   case 0:
6488     if (UseSHA1Intrinsics) {
6489       // we want to do an instanceof comparison against the SHA class
6490       klass_SHA_name = "sun/security/provider/SHA";
6491     }
6492     break;
6493   case 1:
6494     if (UseSHA256Intrinsics) {
6495       // we want to do an instanceof comparison against the SHA2 class
6496       klass_SHA_name = "sun/security/provider/SHA2";
6497     }
6498     break;
6499   case 2:
6500     if (UseSHA512Intrinsics) {
6501       // we want to do an instanceof comparison against the SHA5 class
6502       klass_SHA_name = "sun/security/provider/SHA5";
6503     }
6504     break;
6505   default:
6506     fatal("unknown SHA intrinsic predicate: %d", predicate);
6507   }
6508 
6509   ciKlass* klass_SHA = NULL;
6510   if (klass_SHA_name != NULL) {
6511     klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6512   }
6513   if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6514     // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6515     Node* ctrl = control();
6516     set_control(top()); // no intrinsic path
6517     return ctrl;
6518   }
6519   ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6520 
6521   Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6522   Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6523   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6524   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6525 
6526   return instof_false;  // even if it is NULL
6527 }
6528 
6529 //-------------inline_fma-----------------------------------
6530 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6531   Node *a = NULL;
6532   Node *b = NULL;
6533   Node *c = NULL;
6534   Node* result = NULL;
6535   switch (id) {
6536   case vmIntrinsics::_fmaD:
6537     assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6538     // no receiver since it is static method
6539     a = round_double_node(argument(0));
6540     b = round_double_node(argument(2));
6541     c = round_double_node(argument(4));
6542     result = _gvn.transform(new FmaDNode(control(), a, b, c));
6543     break;
6544   case vmIntrinsics::_fmaF:
6545     assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6546     a = argument(0);
6547     b = argument(1);
6548     c = argument(2);
6549     result = _gvn.transform(new FmaFNode(control(), a, b, c));
6550     break;
6551   default:
6552     fatal_unexpected_iid(id);  break;
6553   }
6554   set_result(result);
6555   return true;
6556 }
6557 
6558 bool LibraryCallKit::inline_profileBoolean() {
6559   Node* counts = argument(1);
6560   const TypeAryPtr* ary = NULL;
6561   ciArray* aobj = NULL;
6562   if (counts->is_Con()
6563       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6564       && (aobj = ary->const_oop()->as_array()) != NULL
6565       && (aobj->length() == 2)) {
6566     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6567     jint false_cnt = aobj->element_value(0).as_int();
6568     jint  true_cnt = aobj->element_value(1).as_int();
6569 
6570     if (C->log() != NULL) {
6571       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6572                      false_cnt, true_cnt);
6573     }
6574 
6575     if (false_cnt + true_cnt == 0) {
6576       // According to profile, never executed.
6577       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6578                           Deoptimization::Action_reinterpret);
6579       return true;
6580     }
6581 
6582     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6583     // is a number of each value occurrences.
6584     Node* result = argument(0);
6585     if (false_cnt == 0 || true_cnt == 0) {
6586       // According to profile, one value has been never seen.
6587       int expected_val = (false_cnt == 0) ? 1 : 0;
6588 
6589       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6590       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6591 
6592       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6593       Node* fast_path = _gvn.transform(new IfTrueNode(check));
6594       Node* slow_path = _gvn.transform(new IfFalseNode(check));
6595 
6596       { // Slow path: uncommon trap for never seen value and then reexecute
6597         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6598         // the value has been seen at least once.
6599         PreserveJVMState pjvms(this);
6600         PreserveReexecuteState preexecs(this);
6601         jvms()->set_should_reexecute(true);
6602 
6603         set_control(slow_path);
6604         set_i_o(i_o());
6605 
6606         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6607                             Deoptimization::Action_reinterpret);
6608       }
6609       // The guard for never seen value enables sharpening of the result and
6610       // returning a constant. It allows to eliminate branches on the same value
6611       // later on.
6612       set_control(fast_path);
6613       result = intcon(expected_val);
6614     }
6615     // Stop profiling.
6616     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6617     // By replacing method body with profile data (represented as ProfileBooleanNode
6618     // on IR level) we effectively disable profiling.
6619     // It enables full speed execution once optimized code is generated.
6620     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6621     C->record_for_igvn(profile);
6622     set_result(profile);
6623     return true;
6624   } else {
6625     // Continue profiling.
6626     // Profile data isn't available at the moment. So, execute method's bytecode version.
6627     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6628     // is compiled and counters aren't available since corresponding MethodHandle
6629     // isn't a compile-time constant.
6630     return false;
6631   }
6632 }
6633 
6634 bool LibraryCallKit::inline_isCompileConstant() {
6635   Node* n = argument(0);
6636   set_result(n->is_Con() ? intcon(1) : intcon(0));
6637   return true;
6638 }