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